fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         struct btrfs_key *location;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 key.type = BTRFS_EXTENT_DATA_KEY;
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end > PAGE_CACHE_SIZE ||
 253             data_len > BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 static inline int inode_need_compress(struct inode *inode)
 352 {
 353         struct btrfs_root *root = BTRFS_I(inode)->root;
 354
 355         /* force compress */
 356         if (btrfs_test_opt(root, FORCE_COMPRESS))
 357                 return 1;
 358         /* bad compression ratios */
 359         if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 360                 return 0;
 361         if (btrfs_test_opt(root, COMPRESS) ||
 362             BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS ||
 363             BTRFS_I(inode)->force_compress)
 364                 return 1;
 365         return 0;
 366 }
 367
 368 /*
 369  * we create compressed extents in two phases.  The first
 370  * phase compresses a range of pages that have already been
 371  * locked (both pages and state bits are locked).
 372  *
 373  * This is done inside an ordered work queue, and the compression
 374  * is spread across many cpus.  The actual IO submission is step
 375  * two, and the ordered work queue takes care of making sure that
 376  * happens in the same order things were put onto the queue by
 377  * writepages and friends.
 378  *
 379  * If this code finds it can't get good compression, it puts an
 380  * entry onto the work queue to write the uncompressed bytes.  This
 381  * makes sure that both compressed inodes and uncompressed inodes
 382  * are written in the same order that the flusher thread sent them
 383  * down.
 384  */
 385 static noinline void compress_file_range(struct inode *inode,
 386                                         struct page *locked_page,
 387                                         u64 start, u64 end,
 388                                         struct async_cow *async_cow,
 389                                         int *num_added)
 390 {
 391         struct btrfs_root *root = BTRFS_I(inode)->root;
 392         u64 num_bytes;
 393         u64 blocksize = root->sectorsize;
 394         u64 actual_end;
 395         u64 isize = i_size_read(inode);
 396         int ret = 0;
 397         struct page **pages = NULL;
 398         unsigned long nr_pages;
 399         unsigned long nr_pages_ret = 0;
 400         unsigned long total_compressed = 0;
 401         unsigned long total_in = 0;
 402         unsigned long max_compressed = 128 * 1024;
 403         unsigned long max_uncompressed = 128 * 1024;
 404         int i;
 405         int will_compress;
 406         int compress_type = root->fs_info->compress_type;
 407         int redirty = 0;
 408
 409         /* if this is a small write inside eof, kick off a defrag */
 410         if ((end - start + 1) < 16 * 1024 &&
 411             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 412                 btrfs_add_inode_defrag(NULL, inode);
 413
 414         actual_end = min_t(u64, isize, end + 1);
 415 again:
 416         will_compress = 0;
 417         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 418         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 419
 420         /*
 421          * we don't want to send crud past the end of i_size through
 422          * compression, that's just a waste of CPU time.  So, if the
 423          * end of the file is before the start of our current
 424          * requested range of bytes, we bail out to the uncompressed
 425          * cleanup code that can deal with all of this.
 426          *
 427          * It isn't really the fastest way to fix things, but this is a
 428          * very uncommon corner.
 429          */
 430         if (actual_end <= start)
 431                 goto cleanup_and_bail_uncompressed;
 432
 433         total_compressed = actual_end - start;
 434
 435         /*
 436          * skip compression for a small file range(<=blocksize) that
 437          * isn't an inline extent, since it dosen't save disk space at all.
 438          */
 439         if (total_compressed <= blocksize &&
 440            (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 441                 goto cleanup_and_bail_uncompressed;
 442
 443         /* we want to make sure that amount of ram required to uncompress
 444          * an extent is reasonable, so we limit the total size in ram
 445          * of a compressed extent to 128k.  This is a crucial number
 446          * because it also controls how easily we can spread reads across
 447          * cpus for decompression.
 448          *
 449          * We also want to make sure the amount of IO required to do
 450          * a random read is reasonably small, so we limit the size of
 451          * a compressed extent to 128k.
 452          */
 453         total_compressed = min(total_compressed, max_uncompressed);
 454         num_bytes = ALIGN(end - start + 1, blocksize);
 455         num_bytes = max(blocksize,  num_bytes);
 456         total_in = 0;
 457         ret = 0;
 458
 459         /*
 460          * we do compression for mount -o compress and when the
 461          * inode has not been flagged as nocompress.  This flag can
 462          * change at any time if we discover bad compression ratios.
 463          */
 464         if (inode_need_compress(inode)) {
 465                 WARN_ON(pages);
 466                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 467                 if (!pages) {
 468                         /* just bail out to the uncompressed code */
 469                         goto cont;
 470                 }
 471
 472                 if (BTRFS_I(inode)->force_compress)
 473                         compress_type = BTRFS_I(inode)->force_compress;
 474
 475                 /*
 476                  * we need to call clear_page_dirty_for_io on each
 477                  * page in the range.  Otherwise applications with the file
 478                  * mmap'd can wander in and change the page contents while
 479                  * we are compressing them.
 480                  *
 481                  * If the compression fails for any reason, we set the pages
 482                  * dirty again later on.
 483                  */
 484                 extent_range_clear_dirty_for_io(inode, start, end);
 485                 redirty = 1;
 486                 ret = btrfs_compress_pages(compress_type,
 487                                            inode->i_mapping, start,
 488                                            total_compressed, pages,
 489                                            nr_pages, &nr_pages_ret,
 490                                            &total_in,
 491                                            &total_compressed,
 492                                            max_compressed);
 493
 494                 if (!ret) {
 495                         unsigned long offset = total_compressed &
 496                                 (PAGE_CACHE_SIZE - 1);
 497                         struct page *page = pages[nr_pages_ret - 1];
 498                         char *kaddr;
 499
 500                         /* zero the tail end of the last page, we might be
 501                          * sending it down to disk
 502                          */
 503                         if (offset) {
 504                                 kaddr = kmap_atomic(page);
 505                                 memset(kaddr + offset, 0,
 506                                        PAGE_CACHE_SIZE - offset);
 507                                 kunmap_atomic(kaddr);
 508                         }
 509                         will_compress = 1;
 510                 }
 511         }
 512 cont:
 513         if (start == 0) {
 514                 /* lets try to make an inline extent */
 515                 if (ret || total_in < (actual_end - start)) {
 516                         /* we didn't compress the entire range, try
 517                          * to make an uncompressed inline extent.
 518                          */
 519                         ret = cow_file_range_inline(root, inode, start, end,
 520                                                     0, 0, NULL);
 521                 } else {
 522                         /* try making a compressed inline extent */
 523                         ret = cow_file_range_inline(root, inode, start, end,
 524                                                     total_compressed,
 525                                                     compress_type, pages);
 526                 }
 527                 if (ret <= 0) {
 528                         unsigned long clear_flags = EXTENT_DELALLOC |
 529                                 EXTENT_DEFRAG;
 530                         unsigned long page_error_op;
 531
 532                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 533                         page_error_op = ret < 0 ? PAGE_SET_ERROR : 0;
 534
 535                         /*
 536                          * inline extent creation worked or returned error,
 537                          * we don't need to create any more async work items.
 538                          * Unlock and free up our temp pages.
 539                          */
 540                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 541                                                      clear_flags, PAGE_UNLOCK |
 542                                                      PAGE_CLEAR_DIRTY |
 543                                                      PAGE_SET_WRITEBACK |
 544                                                      page_error_op |
 545                                                      PAGE_END_WRITEBACK);
 546                         goto free_pages_out;
 547                 }
 548         }
 549
 550         if (will_compress) {
 551                 /*
 552                  * we aren't doing an inline extent round the compressed size
 553                  * up to a block size boundary so the allocator does sane
 554                  * things
 555                  */
 556                 total_compressed = ALIGN(total_compressed, blocksize);
 557
 558                 /*
 559                  * one last check to make sure the compression is really a
 560                  * win, compare the page count read with the blocks on disk
 561                  */
 562                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 563                 if (total_compressed >= total_in) {
 564                         will_compress = 0;
 565                 } else {
 566                         num_bytes = total_in;
 567                 }
 568         }
 569         if (!will_compress && pages) {
 570                 /*
 571                  * the compression code ran but failed to make things smaller,
 572                  * free any pages it allocated and our page pointer array
 573                  */
 574                 for (i = 0; i < nr_pages_ret; i++) {
 575                         WARN_ON(pages[i]->mapping);
 576                         page_cache_release(pages[i]);
 577                 }
 578                 kfree(pages);
 579                 pages = NULL;
 580                 total_compressed = 0;
 581                 nr_pages_ret = 0;
 582
 583                 /* flag the file so we don't compress in the future */
 584                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 585                     !(BTRFS_I(inode)->force_compress)) {
 586                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 587                 }
 588         }
 589         if (will_compress) {
 590                 *num_added += 1;
 591
 592                 /* the async work queues will take care of doing actual
 593                  * allocation on disk for these compressed pages,
 594                  * and will submit them to the elevator.
 595                  */
 596                 add_async_extent(async_cow, start, num_bytes,
 597                                  total_compressed, pages, nr_pages_ret,
 598                                  compress_type);
 599
 600                 if (start + num_bytes < end) {
 601                         start += num_bytes;
 602                         pages = NULL;
 603                         cond_resched();
 604                         goto again;
 605                 }
 606         } else {
 607 cleanup_and_bail_uncompressed:
 608                 /*
 609                  * No compression, but we still need to write the pages in
 610                  * the file we've been given so far.  redirty the locked
 611                  * page if it corresponds to our extent and set things up
 612                  * for the async work queue to run cow_file_range to do
 613                  * the normal delalloc dance
 614                  */
 615                 if (page_offset(locked_page) >= start &&
 616                     page_offset(locked_page) <= end) {
 617                         __set_page_dirty_nobuffers(locked_page);
 618                         /* unlocked later on in the async handlers */
 619                 }
 620                 if (redirty)
 621                         extent_range_redirty_for_io(inode, start, end);
 622                 add_async_extent(async_cow, start, end - start + 1,
 623                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 624                 *num_added += 1;
 625         }
 626
 627         return;
 628
 629 free_pages_out:
 630         for (i = 0; i < nr_pages_ret; i++) {
 631                 WARN_ON(pages[i]->mapping);
 632                 page_cache_release(pages[i]);
 633         }
 634         kfree(pages);
 635 }
 636
 637 static void free_async_extent_pages(struct async_extent *async_extent)
 638 {
 639         int i;
 640
 641         if (!async_extent->pages)
 642                 return;
 643
 644         for (i = 0; i < async_extent->nr_pages; i++) {
 645                 WARN_ON(async_extent->pages[i]->mapping);
 646                 page_cache_release(async_extent->pages[i]);
 647         }
 648         kfree(async_extent->pages);
 649         async_extent->nr_pages = 0;
 650         async_extent->pages = NULL;
 651 }
 652
 653 /*
 654  * phase two of compressed writeback.  This is the ordered portion
 655  * of the code, which only gets called in the order the work was
 656  * queued.  We walk all the async extents created by compress_file_range
 657  * and send them down to the disk.
 658  */
 659 static noinline void submit_compressed_extents(struct inode *inode,
 660                                               struct async_cow *async_cow)
 661 {
 662         struct async_extent *async_extent;
 663         u64 alloc_hint = 0;
 664         struct btrfs_key ins;
 665         struct extent_map *em;
 666         struct btrfs_root *root = BTRFS_I(inode)->root;
 667         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 668         struct extent_io_tree *io_tree;
 669         int ret = 0;
 670
 671 again:
 672         while (!list_empty(&async_cow->extents)) {
 673                 async_extent = list_entry(async_cow->extents.next,
 674                                           struct async_extent, list);
 675                 list_del(&async_extent->list);
 676
 677                 io_tree = &BTRFS_I(inode)->io_tree;
 678
 679 retry:
 680                 /* did the compression code fall back to uncompressed IO? */
 681                 if (!async_extent->pages) {
 682                         int page_started = 0;
 683                         unsigned long nr_written = 0;
 684
 685                         lock_extent(io_tree, async_extent->start,
 686                                          async_extent->start +
 687                                          async_extent->ram_size - 1);
 688
 689                         /* allocate blocks */
 690                         ret = cow_file_range(inode, async_cow->locked_page,
 691                                              async_extent->start,
 692                                              async_extent->start +
 693                                              async_extent->ram_size - 1,
 694                                              &page_started, &nr_written, 0);
 695
 696                         /* JDM XXX */
 697
 698                         /*
 699                          * if page_started, cow_file_range inserted an
 700                          * inline extent and took care of all the unlocking
 701                          * and IO for us.  Otherwise, we need to submit
 702                          * all those pages down to the drive.
 703                          */
 704                         if (!page_started && !ret)
 705                                 extent_write_locked_range(io_tree,
 706                                                   inode, async_extent->start,
 707                                                   async_extent->start +
 708                                                   async_extent->ram_size - 1,
 709                                                   btrfs_get_extent,
 710                                                   WB_SYNC_ALL);
 711                         else if (ret)
 712                                 unlock_page(async_cow->locked_page);
 713                         kfree(async_extent);
 714                         cond_resched();
 715                         continue;
 716                 }
 717
 718                 lock_extent(io_tree, async_extent->start,
 719                             async_extent->start + async_extent->ram_size - 1);
 720
 721                 ret = btrfs_reserve_extent(root,
 722                                            async_extent->compressed_size,
 723                                            async_extent->compressed_size,
 724                                            0, alloc_hint, &ins, 1, 1);
 725                 if (ret) {
 726                         free_async_extent_pages(async_extent);
 727
 728                         if (ret == -ENOSPC) {
 729                                 unlock_extent(io_tree, async_extent->start,
 730                                               async_extent->start +
 731                                               async_extent->ram_size - 1);
 732
 733                                 /*
 734                                  * we need to redirty the pages if we decide to
 735                                  * fallback to uncompressed IO, otherwise we
 736                                  * will not submit these pages down to lower
 737                                  * layers.
 738                                  */
 739                                 extent_range_redirty_for_io(inode,
 740                                                 async_extent->start,
 741                                                 async_extent->start +
 742                                                 async_extent->ram_size - 1);
 743
 744                                 goto retry;
 745                         }
 746                         goto out_free;
 747                 }
 748
 749                 /*
 750                  * here we're doing allocation and writeback of the
 751                  * compressed pages
 752                  */
 753                 btrfs_drop_extent_cache(inode, async_extent->start,
 754                                         async_extent->start +
 755                                         async_extent->ram_size - 1, 0);
 756
 757                 em = alloc_extent_map();
 758                 if (!em) {
 759                         ret = -ENOMEM;
 760                         goto out_free_reserve;
 761                 }
 762                 em->start = async_extent->start;
 763                 em->len = async_extent->ram_size;
 764                 em->orig_start = em->start;
 765                 em->mod_start = em->start;
 766                 em->mod_len = em->len;
 767
 768                 em->block_start = ins.objectid;
 769                 em->block_len = ins.offset;
 770                 em->orig_block_len = ins.offset;
 771                 em->ram_bytes = async_extent->ram_size;
 772                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 773                 em->compress_type = async_extent->compress_type;
 774                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 775                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 776                 em->generation = -1;
 777
 778                 while (1) {
 779                         write_lock(&em_tree->lock);
 780                         ret = add_extent_mapping(em_tree, em, 1);
 781                         write_unlock(&em_tree->lock);
 782                         if (ret != -EEXIST) {
 783                                 free_extent_map(em);
 784                                 break;
 785                         }
 786                         btrfs_drop_extent_cache(inode, async_extent->start,
 787                                                 async_extent->start +
 788                                                 async_extent->ram_size - 1, 0);
 789                 }
 790
 791                 if (ret)
 792                         goto out_free_reserve;
 793
 794                 ret = btrfs_add_ordered_extent_compress(inode,
 795                                                 async_extent->start,
 796                                                 ins.objectid,
 797                                                 async_extent->ram_size,
 798                                                 ins.offset,
 799                                                 BTRFS_ORDERED_COMPRESSED,
 800                                                 async_extent->compress_type);
 801                 if (ret) {
 802                         btrfs_drop_extent_cache(inode, async_extent->start,
 803                                                 async_extent->start +
 804                                                 async_extent->ram_size - 1, 0);
 805                         goto out_free_reserve;
 806                 }
 807
 808                 /*
 809                  * clear dirty, set writeback and unlock the pages.
 810                  */
 811                 extent_clear_unlock_delalloc(inode, async_extent->start,
 812                                 async_extent->start +
 813                                 async_extent->ram_size - 1,
 814                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 815                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 816                                 PAGE_SET_WRITEBACK);
 817                 ret = btrfs_submit_compressed_write(inode,
 818                                     async_extent->start,
 819                                     async_extent->ram_size,
 820                                     ins.objectid,
 821                                     ins.offset, async_extent->pages,
 822                                     async_extent->nr_pages);
 823                 if (ret) {
 824                         struct extent_io_tree *tree = &BTRFS_I(inode)->io_tree;
 825                         struct page *p = async_extent->pages[0];
 826                         const u64 start = async_extent->start;
 827                         const u64 end = start + async_extent->ram_size - 1;
 828
 829                         p->mapping = inode->i_mapping;
 830                         tree->ops->writepage_end_io_hook(p, start, end,
 831                                                          NULL, 0);
 832                         p->mapping = NULL;
 833                         extent_clear_unlock_delalloc(inode, start, end, NULL, 0,
 834                                                      PAGE_END_WRITEBACK |
 835                                                      PAGE_SET_ERROR);
 836                         free_async_extent_pages(async_extent);
 837                 }
 838                 alloc_hint = ins.objectid + ins.offset;
 839                 kfree(async_extent);
 840                 cond_resched();
 841         }
 842         return;
 843 out_free_reserve:
 844         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
 845 out_free:
 846         extent_clear_unlock_delalloc(inode, async_extent->start,
 847                                      async_extent->start +
 848                                      async_extent->ram_size - 1,
 849                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 850                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 851                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 852                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK |
 853                                      PAGE_SET_ERROR);
 854         free_async_extent_pages(async_extent);
 855         kfree(async_extent);
 856         goto again;
 857 }
 858
 859 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 860                                       u64 num_bytes)
 861 {
 862         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 863         struct extent_map *em;
 864         u64 alloc_hint = 0;
 865
 866         read_lock(&em_tree->lock);
 867         em = search_extent_mapping(em_tree, start, num_bytes);
 868         if (em) {
 869                 /*
 870                  * if block start isn't an actual block number then find the
 871                  * first block in this inode and use that as a hint.  If that
 872                  * block is also bogus then just don't worry about it.
 873                  */
 874                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 875                         free_extent_map(em);
 876                         em = search_extent_mapping(em_tree, 0, 0);
 877                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 878                                 alloc_hint = em->block_start;
 879                         if (em)
 880                                 free_extent_map(em);
 881                 } else {
 882                         alloc_hint = em->block_start;
 883                         free_extent_map(em);
 884                 }
 885         }
 886         read_unlock(&em_tree->lock);
 887
 888         return alloc_hint;
 889 }
 890
 891 /*
 892  * when extent_io.c finds a delayed allocation range in the file,
 893  * the call backs end up in this code.  The basic idea is to
 894  * allocate extents on disk for the range, and create ordered data structs
 895  * in ram to track those extents.
 896  *
 897  * locked_page is the page that writepage had locked already.  We use
 898  * it to make sure we don't do extra locks or unlocks.
 899  *
 900  * *page_started is set to one if we unlock locked_page and do everything
 901  * required to start IO on it.  It may be clean and already done with
 902  * IO when we return.
 903  */
 904 static noinline int cow_file_range(struct inode *inode,
 905                                    struct page *locked_page,
 906                                    u64 start, u64 end, int *page_started,
 907                                    unsigned long *nr_written,
 908                                    int unlock)
 909 {
 910         struct btrfs_root *root = BTRFS_I(inode)->root;
 911         u64 alloc_hint = 0;
 912         u64 num_bytes;
 913         unsigned long ram_size;
 914         u64 disk_num_bytes;
 915         u64 cur_alloc_size;
 916         u64 blocksize = root->sectorsize;
 917         struct btrfs_key ins;
 918         struct extent_map *em;
 919         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 920         int ret = 0;
 921
 922         if (btrfs_is_free_space_inode(inode)) {
 923                 WARN_ON_ONCE(1);
 924                 ret = -EINVAL;
 925                 goto out_unlock;
 926         }
 927
 928         num_bytes = ALIGN(end - start + 1, blocksize);
 929         num_bytes = max(blocksize,  num_bytes);
 930         disk_num_bytes = num_bytes;
 931
 932         /* if this is a small write inside eof, kick off defrag */
 933         if (num_bytes < 64 * 1024 &&
 934             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 935                 btrfs_add_inode_defrag(NULL, inode);
 936
 937         if (start == 0) {
 938                 /* lets try to make an inline extent */
 939                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 940                                             NULL);
 941                 if (ret == 0) {
 942                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 943                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 944                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 945                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 946                                      PAGE_END_WRITEBACK);
 947
 948                         *nr_written = *nr_written +
 949                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 950                         *page_started = 1;
 951                         goto out;
 952                 } else if (ret < 0) {
 953                         goto out_unlock;
 954                 }
 955         }
 956
 957         BUG_ON(disk_num_bytes >
 958                btrfs_super_total_bytes(root->fs_info->super_copy));
 959
 960         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 961         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 962
 963         while (disk_num_bytes > 0) {
 964                 unsigned long op;
 965
 966                 cur_alloc_size = disk_num_bytes;
 967                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 968                                            root->sectorsize, 0, alloc_hint,
 969                                            &ins, 1, 1);
 970                 if (ret < 0)
 971                         goto out_unlock;
 972
 973                 em = alloc_extent_map();
 974                 if (!em) {
 975                         ret = -ENOMEM;
 976                         goto out_reserve;
 977                 }
 978                 em->start = start;
 979                 em->orig_start = em->start;
 980                 ram_size = ins.offset;
 981                 em->len = ins.offset;
 982                 em->mod_start = em->start;
 983                 em->mod_len = em->len;
 984
 985                 em->block_start = ins.objectid;
 986                 em->block_len = ins.offset;
 987                 em->orig_block_len = ins.offset;
 988                 em->ram_bytes = ram_size;
 989                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 990                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 991                 em->generation = -1;
 992
 993                 while (1) {
 994                         write_lock(&em_tree->lock);
 995                         ret = add_extent_mapping(em_tree, em, 1);
 996                         write_unlock(&em_tree->lock);
 997                         if (ret != -EEXIST) {
 998                                 free_extent_map(em);
 999                                 break;
1000                         }
1001                         btrfs_drop_extent_cache(inode, start,
1002                                                 start + ram_size - 1, 0);
1003                 }
1004                 if (ret)
1005                         goto out_reserve;
1006
1007                 cur_alloc_size = ins.offset;
1008                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
1009                                                ram_size, cur_alloc_size, 0);
1010                 if (ret)
1011                         goto out_drop_extent_cache;
1012
1013                 if (root->root_key.objectid ==
1014                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1015                         ret = btrfs_reloc_clone_csums(inode, start,
1016                                                       cur_alloc_size);
1017                         if (ret)
1018                                 goto out_drop_extent_cache;
1019                 }
1020
1021                 if (disk_num_bytes < cur_alloc_size)
1022                         break;
1023
1024                 /* we're not doing compressed IO, don't unlock the first
1025                  * page (which the caller expects to stay locked), don't
1026                  * clear any dirty bits and don't set any writeback bits
1027                  *
1028                  * Do set the Private2 bit so we know this page was properly
1029                  * setup for writepage
1030                  */
1031                 op = unlock ? PAGE_UNLOCK : 0;
1032                 op |= PAGE_SET_PRIVATE2;
1033
1034                 extent_clear_unlock_delalloc(inode, start,
1035                                              start + ram_size - 1, locked_page,
1036                                              EXTENT_LOCKED | EXTENT_DELALLOC,
1037                                              op);
1038                 disk_num_bytes -= cur_alloc_size;
1039                 num_bytes -= cur_alloc_size;
1040                 alloc_hint = ins.objectid + ins.offset;
1041                 start += cur_alloc_size;
1042         }
1043 out:
1044         return ret;
1045
1046 out_drop_extent_cache:
1047         btrfs_drop_extent_cache(inode, start, start + ram_size - 1, 0);
1048 out_reserve:
1049         btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
1050 out_unlock:
1051         extent_clear_unlock_delalloc(inode, start, end, locked_page,
1052                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
1053                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
1054                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
1055                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
1056         goto out;
1057 }
1058
1059 /*
1060  * work queue call back to started compression on a file and pages
1061  */
1062 static noinline void async_cow_start(struct btrfs_work *work)
1063 {
1064         struct async_cow *async_cow;
1065         int num_added = 0;
1066         async_cow = container_of(work, struct async_cow, work);
1067
1068         compress_file_range(async_cow->inode, async_cow->locked_page,
1069                             async_cow->start, async_cow->end, async_cow,
1070                             &num_added);
1071         if (num_added == 0) {
1072                 btrfs_add_delayed_iput(async_cow->inode);
1073                 async_cow->inode = NULL;
1074         }
1075 }
1076
1077 /*
1078  * work queue call back to submit previously compressed pages
1079  */
1080 static noinline void async_cow_submit(struct btrfs_work *work)
1081 {
1082         struct async_cow *async_cow;
1083         struct btrfs_root *root;
1084         unsigned long nr_pages;
1085
1086         async_cow = container_of(work, struct async_cow, work);
1087
1088         root = async_cow->root;
1089         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1090                 PAGE_CACHE_SHIFT;
1091
1092         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1093             5 * 1024 * 1024 &&
1094             waitqueue_active(&root->fs_info->async_submit_wait))
1095                 wake_up(&root->fs_info->async_submit_wait);
1096
1097         if (async_cow->inode)
1098                 submit_compressed_extents(async_cow->inode, async_cow);
1099 }
1100
1101 static noinline void async_cow_free(struct btrfs_work *work)
1102 {
1103         struct async_cow *async_cow;
1104         async_cow = container_of(work, struct async_cow, work);
1105         if (async_cow->inode)
1106                 btrfs_add_delayed_iput(async_cow->inode);
1107         kfree(async_cow);
1108 }
1109
1110 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1111                                 u64 start, u64 end, int *page_started,
1112                                 unsigned long *nr_written)
1113 {
1114         struct async_cow *async_cow;
1115         struct btrfs_root *root = BTRFS_I(inode)->root;
1116         unsigned long nr_pages;
1117         u64 cur_end;
1118         int limit = 10 * 1024 * 1024;
1119
1120         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1121                          1, 0, NULL, GFP_NOFS);
1122         while (start < end) {
1123                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1124                 BUG_ON(!async_cow); /* -ENOMEM */
1125                 async_cow->inode = igrab(inode);
1126                 async_cow->root = root;
1127                 async_cow->locked_page = locked_page;
1128                 async_cow->start = start;
1129
1130                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS &&
1131                     !btrfs_test_opt(root, FORCE_COMPRESS))
1132                         cur_end = end;
1133                 else
1134                         cur_end = min(end, start + 512 * 1024 - 1);
1135
1136                 async_cow->end = cur_end;
1137                 INIT_LIST_HEAD(&async_cow->extents);
1138
1139                 btrfs_init_work(&async_cow->work,
1140                                 btrfs_delalloc_helper,
1141                                 async_cow_start, async_cow_submit,
1142                                 async_cow_free);
1143
1144                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1145                         PAGE_CACHE_SHIFT;
1146                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1147
1148                 btrfs_queue_work(root->fs_info->delalloc_workers,
1149                                  &async_cow->work);
1150
1151                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1152                         wait_event(root->fs_info->async_submit_wait,
1153                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1154                             limit));
1155                 }
1156
1157                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1158                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1159                         wait_event(root->fs_info->async_submit_wait,
1160                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1161                            0));
1162                 }
1163
1164                 *nr_written += nr_pages;
1165                 start = cur_end + 1;
1166         }
1167         *page_started = 1;
1168         return 0;
1169 }
1170
1171 static noinline int csum_exist_in_range(struct btrfs_root *root,
1172                                         u64 bytenr, u64 num_bytes)
1173 {
1174         int ret;
1175         struct btrfs_ordered_sum *sums;
1176         LIST_HEAD(list);
1177
1178         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1179                                        bytenr + num_bytes - 1, &list, 0);
1180         if (ret == 0 && list_empty(&list))
1181                 return 0;
1182
1183         while (!list_empty(&list)) {
1184                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1185                 list_del(&sums->list);
1186                 kfree(sums);
1187         }
1188         return 1;
1189 }
1190
1191 /*
1192  * when nowcow writeback call back.  This checks for snapshots or COW copies
1193  * of the extents that exist in the file, and COWs the file as required.
1194  *
1195  * If no cow copies or snapshots exist, we write directly to the existing
1196  * blocks on disk
1197  */
1198 static noinline int run_delalloc_nocow(struct inode *inode,
1199                                        struct page *locked_page,
1200                               u64 start, u64 end, int *page_started, int force,
1201                               unsigned long *nr_written)
1202 {
1203         struct btrfs_root *root = BTRFS_I(inode)->root;
1204         struct btrfs_trans_handle *trans;
1205         struct extent_buffer *leaf;
1206         struct btrfs_path *path;
1207         struct btrfs_file_extent_item *fi;
1208         struct btrfs_key found_key;
1209         u64 cow_start;
1210         u64 cur_offset;
1211         u64 extent_end;
1212         u64 extent_offset;
1213         u64 disk_bytenr;
1214         u64 num_bytes;
1215         u64 disk_num_bytes;
1216         u64 ram_bytes;
1217         int extent_type;
1218         int ret, err;
1219         int type;
1220         int nocow;
1221         int check_prev = 1;
1222         bool nolock;
1223         u64 ino = btrfs_ino(inode);
1224
1225         path = btrfs_alloc_path();
1226         if (!path) {
1227                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1228                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1229                                              EXTENT_DO_ACCOUNTING |
1230                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1231                                              PAGE_CLEAR_DIRTY |
1232                                              PAGE_SET_WRITEBACK |
1233                                              PAGE_END_WRITEBACK);
1234                 return -ENOMEM;
1235         }
1236
1237         nolock = btrfs_is_free_space_inode(inode);
1238
1239         if (nolock)
1240                 trans = btrfs_join_transaction_nolock(root);
1241         else
1242                 trans = btrfs_join_transaction(root);
1243
1244         if (IS_ERR(trans)) {
1245                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1246                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1247                                              EXTENT_DO_ACCOUNTING |
1248                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1249                                              PAGE_CLEAR_DIRTY |
1250                                              PAGE_SET_WRITEBACK |
1251                                              PAGE_END_WRITEBACK);
1252                 btrfs_free_path(path);
1253                 return PTR_ERR(trans);
1254         }
1255
1256         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1257
1258         cow_start = (u64)-1;
1259         cur_offset = start;
1260         while (1) {
1261                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1262                                                cur_offset, 0);
1263                 if (ret < 0)
1264                         goto error;
1265                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1266                         leaf = path->nodes[0];
1267                         btrfs_item_key_to_cpu(leaf, &found_key,
1268                                               path->slots[0] - 1);
1269                         if (found_key.objectid == ino &&
1270                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1271                                 path->slots[0]--;
1272                 }
1273                 check_prev = 0;
1274 next_slot:
1275                 leaf = path->nodes[0];
1276                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1277                         ret = btrfs_next_leaf(root, path);
1278                         if (ret < 0)
1279                                 goto error;
1280                         if (ret > 0)
1281                                 break;
1282                         leaf = path->nodes[0];
1283                 }
1284
1285                 nocow = 0;
1286                 disk_bytenr = 0;
1287                 num_bytes = 0;
1288                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1289
1290                 if (found_key.objectid > ino ||
1291                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1292                     found_key.offset > end)
1293                         break;
1294
1295                 if (found_key.offset > cur_offset) {
1296                         extent_end = found_key.offset;
1297                         extent_type = 0;
1298                         goto out_check;
1299                 }
1300
1301                 fi = btrfs_item_ptr(leaf, path->slots[0],
1302                                     struct btrfs_file_extent_item);
1303                 extent_type = btrfs_file_extent_type(leaf, fi);
1304
1305                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1306                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1307                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1308                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1309                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1310                         extent_end = found_key.offset +
1311                                 btrfs_file_extent_num_bytes(leaf, fi);
1312                         disk_num_bytes =
1313                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1314                         if (extent_end <= start) {
1315                                 path->slots[0]++;
1316                                 goto next_slot;
1317                         }
1318                         if (disk_bytenr == 0)
1319                                 goto out_check;
1320                         if (btrfs_file_extent_compression(leaf, fi) ||
1321                             btrfs_file_extent_encryption(leaf, fi) ||
1322                             btrfs_file_extent_other_encoding(leaf, fi))
1323                                 goto out_check;
1324                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1325                                 goto out_check;
1326                         if (btrfs_extent_readonly(root, disk_bytenr))
1327                                 goto out_check;
1328                         if (btrfs_cross_ref_exist(trans, root, ino,
1329                                                   found_key.offset -
1330                                                   extent_offset, disk_bytenr))
1331                                 goto out_check;
1332                         disk_bytenr += extent_offset;
1333                         disk_bytenr += cur_offset - found_key.offset;
1334                         num_bytes = min(end + 1, extent_end) - cur_offset;
1335                         /*
1336                          * if there are pending snapshots for this root,
1337                          * we fall into common COW way.
1338                          */
1339                         if (!nolock) {
1340                                 err = btrfs_start_write_no_snapshoting(root);
1341                                 if (!err)
1342                                         goto out_check;
1343                         }
1344                         /*
1345                          * force cow if csum exists in the range.
1346                          * this ensure that csum for a given extent are
1347                          * either valid or do not exist.
1348                          */
1349                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1350                                 goto out_check;
1351                         nocow = 1;
1352                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1353                         extent_end = found_key.offset +
1354                                 btrfs_file_extent_inline_len(leaf,
1355                                                      path->slots[0], fi);
1356                         extent_end = ALIGN(extent_end, root->sectorsize);
1357                 } else {
1358                         BUG_ON(1);
1359                 }
1360 out_check:
1361                 if (extent_end <= start) {
1362                         path->slots[0]++;
1363                         if (!nolock && nocow)
1364                                 btrfs_end_write_no_snapshoting(root);
1365                         goto next_slot;
1366                 }
1367                 if (!nocow) {
1368                         if (cow_start == (u64)-1)
1369                                 cow_start = cur_offset;
1370                         cur_offset = extent_end;
1371                         if (cur_offset > end)
1372                                 break;
1373                         path->slots[0]++;
1374                         goto next_slot;
1375                 }
1376
1377                 btrfs_release_path(path);
1378                 if (cow_start != (u64)-1) {
1379                         ret = cow_file_range(inode, locked_page,
1380                                              cow_start, found_key.offset - 1,
1381                                              page_started, nr_written, 1);
1382                         if (ret) {
1383                                 if (!nolock && nocow)
1384                                         btrfs_end_write_no_snapshoting(root);
1385                                 goto error;
1386                         }
1387                         cow_start = (u64)-1;
1388                 }
1389
1390                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1391                         struct extent_map *em;
1392                         struct extent_map_tree *em_tree;
1393                         em_tree = &BTRFS_I(inode)->extent_tree;
1394                         em = alloc_extent_map();
1395                         BUG_ON(!em); /* -ENOMEM */
1396                         em->start = cur_offset;
1397                         em->orig_start = found_key.offset - extent_offset;
1398                         em->len = num_bytes;
1399                         em->block_len = num_bytes;
1400                         em->block_start = disk_bytenr;
1401                         em->orig_block_len = disk_num_bytes;
1402                         em->ram_bytes = ram_bytes;
1403                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1404                         em->mod_start = em->start;
1405                         em->mod_len = em->len;
1406                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1407                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1408                         em->generation = -1;
1409                         while (1) {
1410                                 write_lock(&em_tree->lock);
1411                                 ret = add_extent_mapping(em_tree, em, 1);
1412                                 write_unlock(&em_tree->lock);
1413                                 if (ret != -EEXIST) {
1414                                         free_extent_map(em);
1415                                         break;
1416                                 }
1417                                 btrfs_drop_extent_cache(inode, em->start,
1418                                                 em->start + em->len - 1, 0);
1419                         }
1420                         type = BTRFS_ORDERED_PREALLOC;
1421                 } else {
1422                         type = BTRFS_ORDERED_NOCOW;
1423                 }
1424
1425                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1426                                                num_bytes, num_bytes, type);
1427                 BUG_ON(ret); /* -ENOMEM */
1428
1429                 if (root->root_key.objectid ==
1430                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1431                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1432                                                       num_bytes);
1433                         if (ret) {
1434                                 if (!nolock && nocow)
1435                                         btrfs_end_write_no_snapshoting(root);
1436                                 goto error;
1437                         }
1438                 }
1439
1440                 extent_clear_unlock_delalloc(inode, cur_offset,
1441                                              cur_offset + num_bytes - 1,
1442                                              locked_page, EXTENT_LOCKED |
1443                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1444                                              PAGE_SET_PRIVATE2);
1445                 if (!nolock && nocow)
1446                         btrfs_end_write_no_snapshoting(root);
1447                 cur_offset = extent_end;
1448                 if (cur_offset > end)
1449                         break;
1450         }
1451         btrfs_release_path(path);
1452
1453         if (cur_offset <= end && cow_start == (u64)-1) {
1454                 cow_start = cur_offset;
1455                 cur_offset = end;
1456         }
1457
1458         if (cow_start != (u64)-1) {
1459                 ret = cow_file_range(inode, locked_page, cow_start, end,
1460                                      page_started, nr_written, 1);
1461                 if (ret)
1462                         goto error;
1463         }
1464
1465 error:
1466         err = btrfs_end_transaction(trans, root);
1467         if (!ret)
1468                 ret = err;
1469
1470         if (ret && cur_offset < end)
1471                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1472                                              locked_page, EXTENT_LOCKED |
1473                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1474                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1475                                              PAGE_CLEAR_DIRTY |
1476                                              PAGE_SET_WRITEBACK |
1477                                              PAGE_END_WRITEBACK);
1478         btrfs_free_path(path);
1479         return ret;
1480 }
1481
1482 static inline int need_force_cow(struct inode *inode, u64 start, u64 end)
1483 {
1484
1485         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
1486             !(BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC))
1487                 return 0;
1488
1489         /*
1490          * @defrag_bytes is a hint value, no spinlock held here,
1491          * if is not zero, it means the file is defragging.
1492          * Force cow if given extent needs to be defragged.
1493          */
1494         if (BTRFS_I(inode)->defrag_bytes &&
1495             test_range_bit(&BTRFS_I(inode)->io_tree, start, end,
1496                            EXTENT_DEFRAG, 0, NULL))
1497                 return 1;
1498
1499         return 0;
1500 }
1501
1502 /*
1503  * extent_io.c call back to do delayed allocation processing
1504  */
1505 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1506                               u64 start, u64 end, int *page_started,
1507                               unsigned long *nr_written)
1508 {
1509         int ret;
1510         int force_cow = need_force_cow(inode, start, end);
1511
1512         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW && !force_cow) {
1513                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1514                                          page_started, 1, nr_written);
1515         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC && !force_cow) {
1516                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1517                                          page_started, 0, nr_written);
1518         } else if (!inode_need_compress(inode)) {
1519                 ret = cow_file_range(inode, locked_page, start, end,
1520                                       page_started, nr_written, 1);
1521         } else {
1522                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1523                         &BTRFS_I(inode)->runtime_flags);
1524                 ret = cow_file_range_async(inode, locked_page, start, end,
1525                                            page_started, nr_written);
1526         }
1527         return ret;
1528 }
1529
1530 static void btrfs_split_extent_hook(struct inode *inode,
1531                                     struct extent_state *orig, u64 split)
1532 {
1533         u64 size;
1534
1535         /* not delalloc, ignore it */
1536         if (!(orig->state & EXTENT_DELALLOC))
1537                 return;
1538
1539         size = orig->end - orig->start + 1;
1540         if (size > BTRFS_MAX_EXTENT_SIZE) {
1541                 u64 num_extents;
1542                 u64 new_size;
1543
1544                 /*
1545                  * We need the largest size of the remaining extent to see if we
1546                  * need to add a new outstanding extent.  Think of the following
1547                  * case
1548                  *
1549                  * [MEAX_EXTENT_SIZEx2 - 4k][4k]
1550                  *
1551                  * The new_size would just be 4k and we'd think we had enough
1552                  * outstanding extents for this if we only took one side of the
1553                  * split, same goes for the other direction.  We need to see if
1554                  * the larger size still is the same amount of extents as the
1555                  * original size, because if it is we need to add a new
1556                  * outstanding extent.  But if we split up and the larger size
1557                  * is less than the original then we are good to go since we've
1558                  * already accounted for the extra extent in our original
1559                  * accounting.
1560                  */
1561                 new_size = orig->end - split + 1;
1562                 if ((split - orig->start) > new_size)
1563                         new_size = split - orig->start;
1564
1565                 num_extents = div64_u64(size + BTRFS_MAX_EXTENT_SIZE - 1,
1566                                         BTRFS_MAX_EXTENT_SIZE);
1567                 if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1568                               BTRFS_MAX_EXTENT_SIZE) < num_extents)
1569                         return;
1570         }
1571
1572         spin_lock(&BTRFS_I(inode)->lock);
1573         BTRFS_I(inode)->outstanding_extents++;
1574         spin_unlock(&BTRFS_I(inode)->lock);
1575 }
1576
1577 /*
1578  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1579  * extents so we can keep track of new extents that are just merged onto old
1580  * extents, such as when we are doing sequential writes, so we can properly
1581  * account for the metadata space we'll need.
1582  */
1583 static void btrfs_merge_extent_hook(struct inode *inode,
1584                                     struct extent_state *new,
1585                                     struct extent_state *other)
1586 {
1587         u64 new_size, old_size;
1588         u64 num_extents;
1589
1590         /* not delalloc, ignore it */
1591         if (!(other->state & EXTENT_DELALLOC))
1592                 return;
1593
1594         old_size = other->end - other->start + 1;
1595         new_size = old_size + (new->end - new->start + 1);
1596
1597         /* we're not bigger than the max, unreserve the space and go */
1598         if (new_size <= BTRFS_MAX_EXTENT_SIZE) {
1599                 spin_lock(&BTRFS_I(inode)->lock);
1600                 BTRFS_I(inode)->outstanding_extents--;
1601                 spin_unlock(&BTRFS_I(inode)->lock);
1602                 return;
1603         }
1604
1605         /*
1606          * If we grew by another max_extent, just return, we want to keep that
1607          * reserved amount.
1608          */
1609         num_extents = div64_u64(old_size + BTRFS_MAX_EXTENT_SIZE - 1,
1610                                 BTRFS_MAX_EXTENT_SIZE);
1611         if (div64_u64(new_size + BTRFS_MAX_EXTENT_SIZE - 1,
1612                       BTRFS_MAX_EXTENT_SIZE) > num_extents)
1613                 return;
1614
1615         spin_lock(&BTRFS_I(inode)->lock);
1616         BTRFS_I(inode)->outstanding_extents--;
1617         spin_unlock(&BTRFS_I(inode)->lock);
1618 }
1619
1620 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1621                                       struct inode *inode)
1622 {
1623         spin_lock(&root->delalloc_lock);
1624         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1625                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1626                               &root->delalloc_inodes);
1627                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1628                         &BTRFS_I(inode)->runtime_flags);
1629                 root->nr_delalloc_inodes++;
1630                 if (root->nr_delalloc_inodes == 1) {
1631                         spin_lock(&root->fs_info->delalloc_root_lock);
1632                         BUG_ON(!list_empty(&root->delalloc_root));
1633                         list_add_tail(&root->delalloc_root,
1634                                       &root->fs_info->delalloc_roots);
1635                         spin_unlock(&root->fs_info->delalloc_root_lock);
1636                 }
1637         }
1638         spin_unlock(&root->delalloc_lock);
1639 }
1640
1641 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1642                                      struct inode *inode)
1643 {
1644         spin_lock(&root->delalloc_lock);
1645         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1646                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1647                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1648                           &BTRFS_I(inode)->runtime_flags);
1649                 root->nr_delalloc_inodes--;
1650                 if (!root->nr_delalloc_inodes) {
1651                         spin_lock(&root->fs_info->delalloc_root_lock);
1652                         BUG_ON(list_empty(&root->delalloc_root));
1653                         list_del_init(&root->delalloc_root);
1654                         spin_unlock(&root->fs_info->delalloc_root_lock);
1655                 }
1656         }
1657         spin_unlock(&root->delalloc_lock);
1658 }
1659
1660 /*
1661  * extent_io.c set_bit_hook, used to track delayed allocation
1662  * bytes in this file, and to maintain the list of inodes that
1663  * have pending delalloc work to be done.
1664  */
1665 static void btrfs_set_bit_hook(struct inode *inode,
1666                                struct extent_state *state, unsigned *bits)
1667 {
1668
1669         if ((*bits & EXTENT_DEFRAG) && !(*bits & EXTENT_DELALLOC))
1670                 WARN_ON(1);
1671         /*
1672          * set_bit and clear bit hooks normally require _irqsave/restore
1673          * but in this case, we are only testing for the DELALLOC
1674          * bit, which is only set or cleared with irqs on
1675          */
1676         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1677                 struct btrfs_root *root = BTRFS_I(inode)->root;
1678                 u64 len = state->end + 1 - state->start;
1679                 bool do_list = !btrfs_is_free_space_inode(inode);
1680
1681                 if (*bits & EXTENT_FIRST_DELALLOC) {
1682                         *bits &= ~EXTENT_FIRST_DELALLOC;
1683                 } else {
1684                         spin_lock(&BTRFS_I(inode)->lock);
1685                         BTRFS_I(inode)->outstanding_extents++;
1686                         spin_unlock(&BTRFS_I(inode)->lock);
1687                 }
1688
1689                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1690                                      root->fs_info->delalloc_batch);
1691                 spin_lock(&BTRFS_I(inode)->lock);
1692                 BTRFS_I(inode)->delalloc_bytes += len;
1693                 if (*bits & EXTENT_DEFRAG)
1694                         BTRFS_I(inode)->defrag_bytes += len;
1695                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1696                                          &BTRFS_I(inode)->runtime_flags))
1697                         btrfs_add_delalloc_inodes(root, inode);
1698                 spin_unlock(&BTRFS_I(inode)->lock);
1699         }
1700 }
1701
1702 /*
1703  * extent_io.c clear_bit_hook, see set_bit_hook for why
1704  */
1705 static void btrfs_clear_bit_hook(struct inode *inode,
1706                                  struct extent_state *state,
1707                                  unsigned *bits)
1708 {
1709         u64 len = state->end + 1 - state->start;
1710         u64 num_extents = div64_u64(len + BTRFS_MAX_EXTENT_SIZE -1,
1711                                     BTRFS_MAX_EXTENT_SIZE);
1712
1713         spin_lock(&BTRFS_I(inode)->lock);
1714         if ((state->state & EXTENT_DEFRAG) && (*bits & EXTENT_DEFRAG))
1715                 BTRFS_I(inode)->defrag_bytes -= len;
1716         spin_unlock(&BTRFS_I(inode)->lock);
1717
1718         /*
1719          * set_bit and clear bit hooks normally require _irqsave/restore
1720          * but in this case, we are only testing for the DELALLOC
1721          * bit, which is only set or cleared with irqs on
1722          */
1723         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1724                 struct btrfs_root *root = BTRFS_I(inode)->root;
1725                 bool do_list = !btrfs_is_free_space_inode(inode);
1726
1727                 if (*bits & EXTENT_FIRST_DELALLOC) {
1728                         *bits &= ~EXTENT_FIRST_DELALLOC;
1729                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1730                         spin_lock(&BTRFS_I(inode)->lock);
1731                         BTRFS_I(inode)->outstanding_extents -= num_extents;
1732                         spin_unlock(&BTRFS_I(inode)->lock);
1733                 }
1734
1735                 /*
1736                  * We don't reserve metadata space for space cache inodes so we
1737                  * don't need to call dellalloc_release_metadata if there is an
1738                  * error.
1739                  */
1740                 if (*bits & EXTENT_DO_ACCOUNTING &&
1741                     root != root->fs_info->tree_root)
1742                         btrfs_delalloc_release_metadata(inode, len);
1743
1744                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1745                     && do_list && !(state->state & EXTENT_NORESERVE))
1746                         btrfs_free_reserved_data_space(inode, len);
1747
1748                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1749                                      root->fs_info->delalloc_batch);
1750                 spin_lock(&BTRFS_I(inode)->lock);
1751                 BTRFS_I(inode)->delalloc_bytes -= len;
1752                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1753                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1754                              &BTRFS_I(inode)->runtime_flags))
1755                         btrfs_del_delalloc_inode(root, inode);
1756                 spin_unlock(&BTRFS_I(inode)->lock);
1757         }
1758 }
1759
1760 /*
1761  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1762  * we don't create bios that span stripes or chunks
1763  */
1764 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1765                          size_t size, struct bio *bio,
1766                          unsigned long bio_flags)
1767 {
1768         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1769         u64 logical = (u64)bio->bi_iter.bi_sector << 9;
1770         u64 length = 0;
1771         u64 map_length;
1772         int ret;
1773
1774         if (bio_flags & EXTENT_BIO_COMPRESSED)
1775                 return 0;
1776
1777         length = bio->bi_iter.bi_size;
1778         map_length = length;
1779         ret = btrfs_map_block(root->fs_info, rw, logical,
1780                               &map_length, NULL, 0);
1781         /* Will always return 0 with map_multi == NULL */
1782         BUG_ON(ret < 0);
1783         if (map_length < length + size)
1784                 return 1;
1785         return 0;
1786 }
1787
1788 /*
1789  * in order to insert checksums into the metadata in large chunks,
1790  * we wait until bio submission time.   All the pages in the bio are
1791  * checksummed and sums are attached onto the ordered extent record.
1792  *
1793  * At IO completion time the cums attached on the ordered extent record
1794  * are inserted into the btree
1795  */
1796 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1797                                     struct bio *bio, int mirror_num,
1798                                     unsigned long bio_flags,
1799                                     u64 bio_offset)
1800 {
1801         struct btrfs_root *root = BTRFS_I(inode)->root;
1802         int ret = 0;
1803
1804         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1805         BUG_ON(ret); /* -ENOMEM */
1806         return 0;
1807 }
1808
1809 /*
1810  * in order to insert checksums into the metadata in large chunks,
1811  * we wait until bio submission time.   All the pages in the bio are
1812  * checksummed and sums are attached onto the ordered extent record.
1813  *
1814  * At IO completion time the cums attached on the ordered extent record
1815  * are inserted into the btree
1816  */
1817 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1818                           int mirror_num, unsigned long bio_flags,
1819                           u64 bio_offset)
1820 {
1821         struct btrfs_root *root = BTRFS_I(inode)->root;
1822         int ret;
1823
1824         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1825         if (ret)
1826                 bio_endio(bio, ret);
1827         return ret;
1828 }
1829
1830 /*
1831  * extent_io.c submission hook. This does the right thing for csum calculation
1832  * on write, or reading the csums from the tree before a read
1833  */
1834 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1835                           int mirror_num, unsigned long bio_flags,
1836                           u64 bio_offset)
1837 {
1838         struct btrfs_root *root = BTRFS_I(inode)->root;
1839         int ret = 0;
1840         int skip_sum;
1841         int metadata = 0;
1842         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1843
1844         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1845
1846         if (btrfs_is_free_space_inode(inode))
1847                 metadata = 2;
1848
1849         if (!(rw & REQ_WRITE)) {
1850                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1851                 if (ret)
1852                         goto out;
1853
1854                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1855                         ret = btrfs_submit_compressed_read(inode, bio,
1856                                                            mirror_num,
1857                                                            bio_flags);
1858                         goto out;
1859                 } else if (!skip_sum) {
1860                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1861                         if (ret)
1862                                 goto out;
1863                 }
1864                 goto mapit;
1865         } else if (async && !skip_sum) {
1866                 /* csum items have already been cloned */
1867                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1868                         goto mapit;
1869                 /* we're doing a write, do the async checksumming */
1870                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1871                                    inode, rw, bio, mirror_num,
1872                                    bio_flags, bio_offset,
1873                                    __btrfs_submit_bio_start,
1874                                    __btrfs_submit_bio_done);
1875                 goto out;
1876         } else if (!skip_sum) {
1877                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1878                 if (ret)
1879                         goto out;
1880         }
1881
1882 mapit:
1883         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1884
1885 out:
1886         if (ret < 0)
1887                 bio_endio(bio, ret);
1888         return ret;
1889 }
1890
1891 /*
1892  * given a list of ordered sums record them in the inode.  This happens
1893  * at IO completion time based on sums calculated at bio submission time.
1894  */
1895 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1896                              struct inode *inode, u64 file_offset,
1897                              struct list_head *list)
1898 {
1899         struct btrfs_ordered_sum *sum;
1900
1901         list_for_each_entry(sum, list, list) {
1902                 trans->adding_csums = 1;
1903                 btrfs_csum_file_blocks(trans,
1904                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1905                 trans->adding_csums = 0;
1906         }
1907         return 0;
1908 }
1909
1910 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1911                               struct extent_state **cached_state)
1912 {
1913         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1914         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1915                                    cached_state, GFP_NOFS);
1916 }
1917
1918 /* see btrfs_writepage_start_hook for details on why this is required */
1919 struct btrfs_writepage_fixup {
1920         struct page *page;
1921         struct btrfs_work work;
1922 };
1923
1924 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1925 {
1926         struct btrfs_writepage_fixup *fixup;
1927         struct btrfs_ordered_extent *ordered;
1928         struct extent_state *cached_state = NULL;
1929         struct page *page;
1930         struct inode *inode;
1931         u64 page_start;
1932         u64 page_end;
1933         int ret;
1934
1935         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1936         page = fixup->page;
1937 again:
1938         lock_page(page);
1939         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1940                 ClearPageChecked(page);
1941                 goto out_page;
1942         }
1943
1944         inode = page->mapping->host;
1945         page_start = page_offset(page);
1946         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1947
1948         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1949                          &cached_state);
1950
1951         /* already ordered? We're done */
1952         if (PagePrivate2(page))
1953                 goto out;
1954
1955         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1956         if (ordered) {
1957                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1958                                      page_end, &cached_state, GFP_NOFS);
1959                 unlock_page(page);
1960                 btrfs_start_ordered_extent(inode, ordered, 1);
1961                 btrfs_put_ordered_extent(ordered);
1962                 goto again;
1963         }
1964
1965         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1966         if (ret) {
1967                 mapping_set_error(page->mapping, ret);
1968                 end_extent_writepage(page, ret, page_start, page_end);
1969                 ClearPageChecked(page);
1970                 goto out;
1971          }
1972
1973         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1974         ClearPageChecked(page);
1975         set_page_dirty(page);
1976 out:
1977         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1978                              &cached_state, GFP_NOFS);
1979 out_page:
1980         unlock_page(page);
1981         page_cache_release(page);
1982         kfree(fixup);
1983 }
1984
1985 /*
1986  * There are a few paths in the higher layers of the kernel that directly
1987  * set the page dirty bit without asking the filesystem if it is a
1988  * good idea.  This causes problems because we want to make sure COW
1989  * properly happens and the data=ordered rules are followed.
1990  *
1991  * In our case any range that doesn't have the ORDERED bit set
1992  * hasn't been properly setup for IO.  We kick off an async process
1993  * to fix it up.  The async helper will wait for ordered extents, set
1994  * the delalloc bit and make it safe to write the page.
1995  */
1996 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1997 {
1998         struct inode *inode = page->mapping->host;
1999         struct btrfs_writepage_fixup *fixup;
2000         struct btrfs_root *root = BTRFS_I(inode)->root;
2001
2002         /* this page is properly in the ordered list */
2003         if (TestClearPagePrivate2(page))
2004                 return 0;
2005
2006         if (PageChecked(page))
2007                 return -EAGAIN;
2008
2009         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
2010         if (!fixup)
2011                 return -EAGAIN;
2012
2013         SetPageChecked(page);
2014         page_cache_get(page);
2015         btrfs_init_work(&fixup->work, btrfs_fixup_helper,
2016                         btrfs_writepage_fixup_worker, NULL, NULL);
2017         fixup->page = page;
2018         btrfs_queue_work(root->fs_info->fixup_workers, &fixup->work);
2019         return -EBUSY;
2020 }
2021
2022 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
2023                                        struct inode *inode, u64 file_pos,
2024                                        u64 disk_bytenr, u64 disk_num_bytes,
2025                                        u64 num_bytes, u64 ram_bytes,
2026                                        u8 compression, u8 encryption,
2027                                        u16 other_encoding, int extent_type)
2028 {
2029         struct btrfs_root *root = BTRFS_I(inode)->root;
2030         struct btrfs_file_extent_item *fi;
2031         struct btrfs_path *path;
2032         struct extent_buffer *leaf;
2033         struct btrfs_key ins;
2034         int extent_inserted = 0;
2035         int ret;
2036
2037         path = btrfs_alloc_path();
2038         if (!path)
2039                 return -ENOMEM;
2040
2041         /*
2042          * we may be replacing one extent in the tree with another.
2043          * The new extent is pinned in the extent map, and we don't want
2044          * to drop it from the cache until it is completely in the btree.
2045          *
2046          * So, tell btrfs_drop_extents to leave this extent in the cache.
2047          * the caller is expected to unpin it and allow it to be merged
2048          * with the others.
2049          */
2050         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
2051                                    file_pos + num_bytes, NULL, 0,
2052                                    1, sizeof(*fi), &extent_inserted);
2053         if (ret)
2054                 goto out;
2055
2056         if (!extent_inserted) {
2057                 ins.objectid = btrfs_ino(inode);
2058                 ins.offset = file_pos;
2059                 ins.type = BTRFS_EXTENT_DATA_KEY;
2060
2061                 path->leave_spinning = 1;
2062                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
2063                                               sizeof(*fi));
2064                 if (ret)
2065                         goto out;
2066         }
2067         leaf = path->nodes[0];
2068         fi = btrfs_item_ptr(leaf, path->slots[0],
2069                             struct btrfs_file_extent_item);
2070         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
2071         btrfs_set_file_extent_type(leaf, fi, extent_type);
2072         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
2073         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
2074         btrfs_set_file_extent_offset(leaf, fi, 0);
2075         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
2076         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
2077         btrfs_set_file_extent_compression(leaf, fi, compression);
2078         btrfs_set_file_extent_encryption(leaf, fi, encryption);
2079         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
2080
2081         btrfs_mark_buffer_dirty(leaf);
2082         btrfs_release_path(path);
2083
2084         inode_add_bytes(inode, num_bytes);
2085
2086         ins.objectid = disk_bytenr;
2087         ins.offset = disk_num_bytes;
2088         ins.type = BTRFS_EXTENT_ITEM_KEY;
2089         ret = btrfs_alloc_reserved_file_extent(trans, root,
2090                                         root->root_key.objectid,
2091                                         btrfs_ino(inode), file_pos, &ins);
2092 out:
2093         btrfs_free_path(path);
2094
2095         return ret;
2096 }
2097
2098 /* snapshot-aware defrag */
2099 struct sa_defrag_extent_backref {
2100         struct rb_node node;
2101         struct old_sa_defrag_extent *old;
2102         u64 root_id;
2103         u64 inum;
2104         u64 file_pos;
2105         u64 extent_offset;
2106         u64 num_bytes;
2107         u64 generation;
2108 };
2109
2110 struct old_sa_defrag_extent {
2111         struct list_head list;
2112         struct new_sa_defrag_extent *new;
2113
2114         u64 extent_offset;
2115         u64 bytenr;
2116         u64 offset;
2117         u64 len;
2118         int count;
2119 };
2120
2121 struct new_sa_defrag_extent {
2122         struct rb_root root;
2123         struct list_head head;
2124         struct btrfs_path *path;
2125         struct inode *inode;
2126         u64 file_pos;
2127         u64 len;
2128         u64 bytenr;
2129         u64 disk_len;
2130         u8 compress_type;
2131 };
2132
2133 static int backref_comp(struct sa_defrag_extent_backref *b1,
2134                         struct sa_defrag_extent_backref *b2)
2135 {
2136         if (b1->root_id < b2->root_id)
2137                 return -1;
2138         else if (b1->root_id > b2->root_id)
2139                 return 1;
2140
2141         if (b1->inum < b2->inum)
2142                 return -1;
2143         else if (b1->inum > b2->inum)
2144                 return 1;
2145
2146         if (b1->file_pos < b2->file_pos)
2147                 return -1;
2148         else if (b1->file_pos > b2->file_pos)
2149                 return 1;
2150
2151         /*
2152          * [------------------------------] ===> (a range of space)
2153          *     |<--->|   |<---->| =============> (fs/file tree A)
2154          * |<---------------------------->| ===> (fs/file tree B)
2155          *
2156          * A range of space can refer to two file extents in one tree while
2157          * refer to only one file extent in another tree.
2158          *
2159          * So we may process a disk offset more than one time(two extents in A)
2160          * and locate at the same extent(one extent in B), then insert two same
2161          * backrefs(both refer to the extent in B).
2162          */
2163         return 0;
2164 }
2165
2166 static void backref_insert(struct rb_root *root,
2167                            struct sa_defrag_extent_backref *backref)
2168 {
2169         struct rb_node **p = &root->rb_node;
2170         struct rb_node *parent = NULL;
2171         struct sa_defrag_extent_backref *entry;
2172         int ret;
2173
2174         while (*p) {
2175                 parent = *p;
2176                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2177
2178                 ret = backref_comp(backref, entry);
2179                 if (ret < 0)
2180                         p = &(*p)->rb_left;
2181                 else
2182                         p = &(*p)->rb_right;
2183         }
2184
2185         rb_link_node(&backref->node, parent, p);
2186         rb_insert_color(&backref->node, root);
2187 }
2188
2189 /*
2190  * Note the backref might has changed, and in this case we just return 0.
2191  */
2192 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2193                                        void *ctx)
2194 {
2195         struct btrfs_file_extent_item *extent;
2196         struct btrfs_fs_info *fs_info;
2197         struct old_sa_defrag_extent *old = ctx;
2198         struct new_sa_defrag_extent *new = old->new;
2199         struct btrfs_path *path = new->path;
2200         struct btrfs_key key;
2201         struct btrfs_root *root;
2202         struct sa_defrag_extent_backref *backref;
2203         struct extent_buffer *leaf;
2204         struct inode *inode = new->inode;
2205         int slot;
2206         int ret;
2207         u64 extent_offset;
2208         u64 num_bytes;
2209
2210         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2211             inum == btrfs_ino(inode))
2212                 return 0;
2213
2214         key.objectid = root_id;
2215         key.type = BTRFS_ROOT_ITEM_KEY;
2216         key.offset = (u64)-1;
2217
2218         fs_info = BTRFS_I(inode)->root->fs_info;
2219         root = btrfs_read_fs_root_no_name(fs_info, &key);
2220         if (IS_ERR(root)) {
2221                 if (PTR_ERR(root) == -ENOENT)
2222                         return 0;
2223                 WARN_ON(1);
2224                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2225                          inum, offset, root_id);
2226                 return PTR_ERR(root);
2227         }
2228
2229         key.objectid = inum;
2230         key.type = BTRFS_EXTENT_DATA_KEY;
2231         if (offset > (u64)-1 << 32)
2232                 key.offset = 0;
2233         else
2234                 key.offset = offset;
2235
2236         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2237         if (WARN_ON(ret < 0))
2238                 return ret;
2239         ret = 0;
2240
2241         while (1) {
2242                 cond_resched();
2243
2244                 leaf = path->nodes[0];
2245                 slot = path->slots[0];
2246
2247                 if (slot >= btrfs_header_nritems(leaf)) {
2248                         ret = btrfs_next_leaf(root, path);
2249                         if (ret < 0) {
2250                                 goto out;
2251                         } else if (ret > 0) {
2252                                 ret = 0;
2253                                 goto out;
2254                         }
2255                         continue;
2256                 }
2257
2258                 path->slots[0]++;
2259
2260                 btrfs_item_key_to_cpu(leaf, &key, slot);
2261
2262                 if (key.objectid > inum)
2263                         goto out;
2264
2265                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2266                         continue;
2267
2268                 extent = btrfs_item_ptr(leaf, slot,
2269                                         struct btrfs_file_extent_item);
2270
2271                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2272                         continue;
2273
2274                 /*
2275                  * 'offset' refers to the exact key.offset,
2276                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2277                  * (key.offset - extent_offset).
2278                  */
2279                 if (key.offset != offset)
2280                         continue;
2281
2282                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2283                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2284
2285                 if (extent_offset >= old->extent_offset + old->offset +
2286                     old->len || extent_offset + num_bytes <=
2287                     old->extent_offset + old->offset)
2288                         continue;
2289                 break;
2290         }
2291
2292         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2293         if (!backref) {
2294                 ret = -ENOENT;
2295                 goto out;
2296         }
2297
2298         backref->root_id = root_id;
2299         backref->inum = inum;
2300         backref->file_pos = offset;
2301         backref->num_bytes = num_bytes;
2302         backref->extent_offset = extent_offset;
2303         backref->generation = btrfs_file_extent_generation(leaf, extent);
2304         backref->old = old;
2305         backref_insert(&new->root, backref);
2306         old->count++;
2307 out:
2308         btrfs_release_path(path);
2309         WARN_ON(ret);
2310         return ret;
2311 }
2312
2313 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2314                                    struct new_sa_defrag_extent *new)
2315 {
2316         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2317         struct old_sa_defrag_extent *old, *tmp;
2318         int ret;
2319
2320         new->path = path;
2321
2322         list_for_each_entry_safe(old, tmp, &new->head, list) {
2323                 ret = iterate_inodes_from_logical(old->bytenr +
2324                                                   old->extent_offset, fs_info,
2325                                                   path, record_one_backref,
2326                                                   old);
2327                 if (ret < 0 && ret != -ENOENT)
2328                         return false;
2329
2330                 /* no backref to be processed for this extent */
2331                 if (!old->count) {
2332                         list_del(&old->list);
2333                         kfree(old);
2334                 }
2335         }
2336
2337         if (list_empty(&new->head))
2338                 return false;
2339
2340         return true;
2341 }
2342
2343 static int relink_is_mergable(struct extent_buffer *leaf,
2344                               struct btrfs_file_extent_item *fi,
2345                               struct new_sa_defrag_extent *new)
2346 {
2347         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2348                 return 0;
2349
2350         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2351                 return 0;
2352
2353         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2354                 return 0;
2355
2356         if (btrfs_file_extent_encryption(leaf, fi) ||
2357             btrfs_file_extent_other_encoding(leaf, fi))
2358                 return 0;
2359
2360         return 1;
2361 }
2362
2363 /*
2364  * Note the backref might has changed, and in this case we just return 0.
2365  */
2366 static noinline int relink_extent_backref(struct btrfs_path *path,
2367                                  struct sa_defrag_extent_backref *prev,
2368                                  struct sa_defrag_extent_backref *backref)
2369 {
2370         struct btrfs_file_extent_item *extent;
2371         struct btrfs_file_extent_item *item;
2372         struct btrfs_ordered_extent *ordered;
2373         struct btrfs_trans_handle *trans;
2374         struct btrfs_fs_info *fs_info;
2375         struct btrfs_root *root;
2376         struct btrfs_key key;
2377         struct extent_buffer *leaf;
2378         struct old_sa_defrag_extent *old = backref->old;
2379         struct new_sa_defrag_extent *new = old->new;
2380         struct inode *src_inode = new->inode;
2381         struct inode *inode;
2382         struct extent_state *cached = NULL;
2383         int ret = 0;
2384         u64 start;
2385         u64 len;
2386         u64 lock_start;
2387         u64 lock_end;
2388         bool merge = false;
2389         int index;
2390
2391         if (prev && prev->root_id == backref->root_id &&
2392             prev->inum == backref->inum &&
2393             prev->file_pos + prev->num_bytes == backref->file_pos)
2394                 merge = true;
2395
2396         /* step 1: get root */
2397         key.objectid = backref->root_id;
2398         key.type = BTRFS_ROOT_ITEM_KEY;
2399         key.offset = (u64)-1;
2400
2401         fs_info = BTRFS_I(src_inode)->root->fs_info;
2402         index = srcu_read_lock(&fs_info->subvol_srcu);
2403
2404         root = btrfs_read_fs_root_no_name(fs_info, &key);
2405         if (IS_ERR(root)) {
2406                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2407                 if (PTR_ERR(root) == -ENOENT)
2408                         return 0;
2409                 return PTR_ERR(root);
2410         }
2411
2412         if (btrfs_root_readonly(root)) {
2413                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2414                 return 0;
2415         }
2416
2417         /* step 2: get inode */
2418         key.objectid = backref->inum;
2419         key.type = BTRFS_INODE_ITEM_KEY;
2420         key.offset = 0;
2421
2422         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2423         if (IS_ERR(inode)) {
2424                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2425                 return 0;
2426         }
2427
2428         srcu_read_unlock(&fs_info->subvol_srcu, index);
2429
2430         /* step 3: relink backref */
2431         lock_start = backref->file_pos;
2432         lock_end = backref->file_pos + backref->num_bytes - 1;
2433         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2434                          0, &cached);
2435
2436         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2437         if (ordered) {
2438                 btrfs_put_ordered_extent(ordered);
2439                 goto out_unlock;
2440         }
2441
2442         trans = btrfs_join_transaction(root);
2443         if (IS_ERR(trans)) {
2444                 ret = PTR_ERR(trans);
2445                 goto out_unlock;
2446         }
2447
2448         key.objectid = backref->inum;
2449         key.type = BTRFS_EXTENT_DATA_KEY;
2450         key.offset = backref->file_pos;
2451
2452         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2453         if (ret < 0) {
2454                 goto out_free_path;
2455         } else if (ret > 0) {
2456                 ret = 0;
2457                 goto out_free_path;
2458         }
2459
2460         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2461                                 struct btrfs_file_extent_item);
2462
2463         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2464             backref->generation)
2465                 goto out_free_path;
2466
2467         btrfs_release_path(path);
2468
2469         start = backref->file_pos;
2470         if (backref->extent_offset < old->extent_offset + old->offset)
2471                 start += old->extent_offset + old->offset -
2472                          backref->extent_offset;
2473
2474         len = min(backref->extent_offset + backref->num_bytes,
2475                   old->extent_offset + old->offset + old->len);
2476         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2477
2478         ret = btrfs_drop_extents(trans, root, inode, start,
2479                                  start + len, 1);
2480         if (ret)
2481                 goto out_free_path;
2482 again:
2483         key.objectid = btrfs_ino(inode);
2484         key.type = BTRFS_EXTENT_DATA_KEY;
2485         key.offset = start;
2486
2487         path->leave_spinning = 1;
2488         if (merge) {
2489                 struct btrfs_file_extent_item *fi;
2490                 u64 extent_len;
2491                 struct btrfs_key found_key;
2492
2493                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2494                 if (ret < 0)
2495                         goto out_free_path;
2496
2497                 path->slots[0]--;
2498                 leaf = path->nodes[0];
2499                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2500
2501                 fi = btrfs_item_ptr(leaf, path->slots[0],
2502                                     struct btrfs_file_extent_item);
2503                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2504
2505                 if (extent_len + found_key.offset == start &&
2506                     relink_is_mergable(leaf, fi, new)) {
2507                         btrfs_set_file_extent_num_bytes(leaf, fi,
2508                                                         extent_len + len);
2509                         btrfs_mark_buffer_dirty(leaf);
2510                         inode_add_bytes(inode, len);
2511
2512                         ret = 1;
2513                         goto out_free_path;
2514                 } else {
2515                         merge = false;
2516                         btrfs_release_path(path);
2517                         goto again;
2518                 }
2519         }
2520
2521         ret = btrfs_insert_empty_item(trans, root, path, &key,
2522                                         sizeof(*extent));
2523         if (ret) {
2524                 btrfs_abort_transaction(trans, root, ret);
2525                 goto out_free_path;
2526         }
2527
2528         leaf = path->nodes[0];
2529         item = btrfs_item_ptr(leaf, path->slots[0],
2530                                 struct btrfs_file_extent_item);
2531         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2532         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2533         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2534         btrfs_set_file_extent_num_bytes(leaf, item, len);
2535         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2536         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2537         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2538         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2539         btrfs_set_file_extent_encryption(leaf, item, 0);
2540         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2541
2542         btrfs_mark_buffer_dirty(leaf);
2543         inode_add_bytes(inode, len);
2544         btrfs_release_path(path);
2545
2546         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2547                         new->disk_len, 0,
2548                         backref->root_id, backref->inum,
2549                         new->file_pos, 0);      /* start - extent_offset */
2550         if (ret) {
2551                 btrfs_abort_transaction(trans, root, ret);
2552                 goto out_free_path;
2553         }
2554
2555         ret = 1;
2556 out_free_path:
2557         btrfs_release_path(path);
2558         path->leave_spinning = 0;
2559         btrfs_end_transaction(trans, root);
2560 out_unlock:
2561         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2562                              &cached, GFP_NOFS);
2563         iput(inode);
2564         return ret;
2565 }
2566
2567 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2568 {
2569         struct old_sa_defrag_extent *old, *tmp;
2570
2571         if (!new)
2572                 return;
2573
2574         list_for_each_entry_safe(old, tmp, &new->head, list) {
2575                 list_del(&old->list);
2576                 kfree(old);
2577         }
2578         kfree(new);
2579 }
2580
2581 static void relink_file_extents(struct new_sa_defrag_extent *new)
2582 {
2583         struct btrfs_path *path;
2584         struct sa_defrag_extent_backref *backref;
2585         struct sa_defrag_extent_backref *prev = NULL;
2586         struct inode *inode;
2587         struct btrfs_root *root;
2588         struct rb_node *node;
2589         int ret;
2590
2591         inode = new->inode;
2592         root = BTRFS_I(inode)->root;
2593
2594         path = btrfs_alloc_path();
2595         if (!path)
2596                 return;
2597
2598         if (!record_extent_backrefs(path, new)) {
2599                 btrfs_free_path(path);
2600                 goto out;
2601         }
2602         btrfs_release_path(path);
2603
2604         while (1) {
2605                 node = rb_first(&new->root);
2606                 if (!node)
2607                         break;
2608                 rb_erase(node, &new->root);
2609
2610                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2611
2612                 ret = relink_extent_backref(path, prev, backref);
2613                 WARN_ON(ret < 0);
2614
2615                 kfree(prev);
2616
2617                 if (ret == 1)
2618                         prev = backref;
2619                 else
2620                         prev = NULL;
2621                 cond_resched();
2622         }
2623         kfree(prev);
2624
2625         btrfs_free_path(path);
2626 out:
2627         free_sa_defrag_extent(new);
2628
2629         atomic_dec(&root->fs_info->defrag_running);
2630         wake_up(&root->fs_info->transaction_wait);
2631 }
2632
2633 static struct new_sa_defrag_extent *
2634 record_old_file_extents(struct inode *inode,
2635                         struct btrfs_ordered_extent *ordered)
2636 {
2637         struct btrfs_root *root = BTRFS_I(inode)->root;
2638         struct btrfs_path *path;
2639         struct btrfs_key key;
2640         struct old_sa_defrag_extent *old;
2641         struct new_sa_defrag_extent *new;
2642         int ret;
2643
2644         new = kmalloc(sizeof(*new), GFP_NOFS);
2645         if (!new)
2646                 return NULL;
2647
2648         new->inode = inode;
2649         new->file_pos = ordered->file_offset;
2650         new->len = ordered->len;
2651         new->bytenr = ordered->start;
2652         new->disk_len = ordered->disk_len;
2653         new->compress_type = ordered->compress_type;
2654         new->root = RB_ROOT;
2655         INIT_LIST_HEAD(&new->head);
2656
2657         path = btrfs_alloc_path();
2658         if (!path)
2659                 goto out_kfree;
2660
2661         key.objectid = btrfs_ino(inode);
2662         key.type = BTRFS_EXTENT_DATA_KEY;
2663         key.offset = new->file_pos;
2664
2665         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2666         if (ret < 0)
2667                 goto out_free_path;
2668         if (ret > 0 && path->slots[0] > 0)
2669                 path->slots[0]--;
2670
2671         /* find out all the old extents for the file range */
2672         while (1) {
2673                 struct btrfs_file_extent_item *extent;
2674                 struct extent_buffer *l;
2675                 int slot;
2676                 u64 num_bytes;
2677                 u64 offset;
2678                 u64 end;
2679                 u64 disk_bytenr;
2680                 u64 extent_offset;
2681
2682                 l = path->nodes[0];
2683                 slot = path->slots[0];
2684
2685                 if (slot >= btrfs_header_nritems(l)) {
2686                         ret = btrfs_next_leaf(root, path);
2687                         if (ret < 0)
2688                                 goto out_free_path;
2689                         else if (ret > 0)
2690                                 break;
2691                         continue;
2692                 }
2693
2694                 btrfs_item_key_to_cpu(l, &key, slot);
2695
2696                 if (key.objectid != btrfs_ino(inode))
2697                         break;
2698                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2699                         break;
2700                 if (key.offset >= new->file_pos + new->len)
2701                         break;
2702
2703                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2704
2705                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2706                 if (key.offset + num_bytes < new->file_pos)
2707                         goto next;
2708
2709                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2710                 if (!disk_bytenr)
2711                         goto next;
2712
2713                 extent_offset = btrfs_file_extent_offset(l, extent);
2714
2715                 old = kmalloc(sizeof(*old), GFP_NOFS);
2716                 if (!old)
2717                         goto out_free_path;
2718
2719                 offset = max(new->file_pos, key.offset);
2720                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2721
2722                 old->bytenr = disk_bytenr;
2723                 old->extent_offset = extent_offset;
2724                 old->offset = offset - key.offset;
2725                 old->len = end - offset;
2726                 old->new = new;
2727                 old->count = 0;
2728                 list_add_tail(&old->list, &new->head);
2729 next:
2730                 path->slots[0]++;
2731                 cond_resched();
2732         }
2733
2734         btrfs_free_path(path);
2735         atomic_inc(&root->fs_info->defrag_running);
2736
2737         return new;
2738
2739 out_free_path:
2740         btrfs_free_path(path);
2741 out_kfree:
2742         free_sa_defrag_extent(new);
2743         return NULL;
2744 }
2745
2746 static void btrfs_release_delalloc_bytes(struct btrfs_root *root,
2747                                          u64 start, u64 len)
2748 {
2749         struct btrfs_block_group_cache *cache;
2750
2751         cache = btrfs_lookup_block_group(root->fs_info, start);
2752         ASSERT(cache);
2753
2754         spin_lock(&cache->lock);
2755         cache->delalloc_bytes -= len;
2756         spin_unlock(&cache->lock);
2757
2758         btrfs_put_block_group(cache);
2759 }
2760
2761 /* as ordered data IO finishes, this gets called so we can finish
2762  * an ordered extent if the range of bytes in the file it covers are
2763  * fully written.
2764  */
2765 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2766 {
2767         struct inode *inode = ordered_extent->inode;
2768         struct btrfs_root *root = BTRFS_I(inode)->root;
2769         struct btrfs_trans_handle *trans = NULL;
2770         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2771         struct extent_state *cached_state = NULL;
2772         struct new_sa_defrag_extent *new = NULL;
2773         int compress_type = 0;
2774         int ret = 0;
2775         u64 logical_len = ordered_extent->len;
2776         bool nolock;
2777         bool truncated = false;
2778
2779         nolock = btrfs_is_free_space_inode(inode);
2780
2781         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2782                 ret = -EIO;
2783                 goto out;
2784         }
2785
2786         btrfs_free_io_failure_record(inode, ordered_extent->file_offset,
2787                                      ordered_extent->file_offset +
2788                                      ordered_extent->len - 1);
2789
2790         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2791                 truncated = true;
2792                 logical_len = ordered_extent->truncated_len;
2793                 /* Truncated the entire extent, don't bother adding */
2794                 if (!logical_len)
2795                         goto out;
2796         }
2797
2798         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2799                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2800                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2801                 if (nolock)
2802                         trans = btrfs_join_transaction_nolock(root);
2803                 else
2804                         trans = btrfs_join_transaction(root);
2805                 if (IS_ERR(trans)) {
2806                         ret = PTR_ERR(trans);
2807                         trans = NULL;
2808                         goto out;
2809                 }
2810                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2811                 ret = btrfs_update_inode_fallback(trans, root, inode);
2812                 if (ret) /* -ENOMEM or corruption */
2813                         btrfs_abort_transaction(trans, root, ret);
2814                 goto out;
2815         }
2816
2817         lock_extent_bits(io_tree, ordered_extent->file_offset,
2818                          ordered_extent->file_offset + ordered_extent->len - 1,
2819                          0, &cached_state);
2820
2821         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2822                         ordered_extent->file_offset + ordered_extent->len - 1,
2823                         EXTENT_DEFRAG, 1, cached_state);
2824         if (ret) {
2825                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2826                 if (0 && last_snapshot >= BTRFS_I(inode)->generation)
2827                         /* the inode is shared */
2828                         new = record_old_file_extents(inode, ordered_extent);
2829
2830                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2831                         ordered_extent->file_offset + ordered_extent->len - 1,
2832                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2833         }
2834
2835         if (nolock)
2836                 trans = btrfs_join_transaction_nolock(root);
2837         else
2838                 trans = btrfs_join_transaction(root);
2839         if (IS_ERR(trans)) {
2840                 ret = PTR_ERR(trans);
2841                 trans = NULL;
2842                 goto out_unlock;
2843         }
2844
2845         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2846
2847         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2848                 compress_type = ordered_extent->compress_type;
2849         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2850                 BUG_ON(compress_type);
2851                 ret = btrfs_mark_extent_written(trans, inode,
2852                                                 ordered_extent->file_offset,
2853                                                 ordered_extent->file_offset +
2854                                                 logical_len);
2855         } else {
2856                 BUG_ON(root == root->fs_info->tree_root);
2857                 ret = insert_reserved_file_extent(trans, inode,
2858                                                 ordered_extent->file_offset,
2859                                                 ordered_extent->start,
2860                                                 ordered_extent->disk_len,
2861                                                 logical_len, logical_len,
2862                                                 compress_type, 0, 0,
2863                                                 BTRFS_FILE_EXTENT_REG);
2864                 if (!ret)
2865                         btrfs_release_delalloc_bytes(root,
2866                                                      ordered_extent->start,
2867                                                      ordered_extent->disk_len);
2868         }
2869         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2870                            ordered_extent->file_offset, ordered_extent->len,
2871                            trans->transid);
2872         if (ret < 0) {
2873                 btrfs_abort_transaction(trans, root, ret);
2874                 goto out_unlock;
2875         }
2876
2877         add_pending_csums(trans, inode, ordered_extent->file_offset,
2878                           &ordered_extent->list);
2879
2880         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2881         ret = btrfs_update_inode_fallback(trans, root, inode);
2882         if (ret) { /* -ENOMEM or corruption */
2883                 btrfs_abort_transaction(trans, root, ret);
2884                 goto out_unlock;
2885         }
2886         ret = 0;
2887 out_unlock:
2888         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2889                              ordered_extent->file_offset +
2890                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2891 out:
2892         if (root != root->fs_info->tree_root)
2893                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2894         if (trans)
2895                 btrfs_end_transaction(trans, root);
2896
2897         if (ret || truncated) {
2898                 u64 start, end;
2899
2900                 if (truncated)
2901                         start = ordered_extent->file_offset + logical_len;
2902                 else
2903                         start = ordered_extent->file_offset;
2904                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2905                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2906
2907                 /* Drop the cache for the part of the extent we didn't write. */
2908                 btrfs_drop_extent_cache(inode, start, end, 0);
2909
2910                 /*
2911                  * If the ordered extent had an IOERR or something else went
2912                  * wrong we need to return the space for this ordered extent
2913                  * back to the allocator.  We only free the extent in the
2914                  * truncated case if we didn't write out the extent at all.
2915                  */
2916                 if ((ret || !logical_len) &&
2917                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2918                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2919                         btrfs_free_reserved_extent(root, ordered_extent->start,
2920                                                    ordered_extent->disk_len, 1);
2921         }
2922
2923
2924         /*
2925          * This needs to be done to make sure anybody waiting knows we are done
2926          * updating everything for this ordered extent.
2927          */
2928         btrfs_remove_ordered_extent(inode, ordered_extent);
2929
2930         /* for snapshot-aware defrag */
2931         if (new) {
2932                 if (ret) {
2933                         free_sa_defrag_extent(new);
2934                         atomic_dec(&root->fs_info->defrag_running);
2935                 } else {
2936                         relink_file_extents(new);
2937                 }
2938         }
2939
2940         /* once for us */
2941         btrfs_put_ordered_extent(ordered_extent);
2942         /* once for the tree */
2943         btrfs_put_ordered_extent(ordered_extent);
2944
2945         return ret;
2946 }
2947
2948 static void finish_ordered_fn(struct btrfs_work *work)
2949 {
2950         struct btrfs_ordered_extent *ordered_extent;
2951         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2952         btrfs_finish_ordered_io(ordered_extent);
2953 }
2954
2955 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2956                                 struct extent_state *state, int uptodate)
2957 {
2958         struct inode *inode = page->mapping->host;
2959         struct btrfs_root *root = BTRFS_I(inode)->root;
2960         struct btrfs_ordered_extent *ordered_extent = NULL;
2961         struct btrfs_workqueue *wq;
2962         btrfs_work_func_t func;
2963
2964         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2965
2966         ClearPagePrivate2(page);
2967         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2968                                             end - start + 1, uptodate))
2969                 return 0;
2970
2971         if (btrfs_is_free_space_inode(inode)) {
2972                 wq = root->fs_info->endio_freespace_worker;
2973                 func = btrfs_freespace_write_helper;
2974         } else {
2975                 wq = root->fs_info->endio_write_workers;
2976                 func = btrfs_endio_write_helper;
2977         }
2978
2979         btrfs_init_work(&ordered_extent->work, func, finish_ordered_fn, NULL,
2980                         NULL);
2981         btrfs_queue_work(wq, &ordered_extent->work);
2982
2983         return 0;
2984 }
2985
2986 static int __readpage_endio_check(struct inode *inode,
2987                                   struct btrfs_io_bio *io_bio,
2988                                   int icsum, struct page *page,
2989                                   int pgoff, u64 start, size_t len)
2990 {
2991         char *kaddr;
2992         u32 csum_expected;
2993         u32 csum = ~(u32)0;
2994         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2995                                       DEFAULT_RATELIMIT_BURST);
2996
2997         csum_expected = *(((u32 *)io_bio->csum) + icsum);
2998
2999         kaddr = kmap_atomic(page);
3000         csum = btrfs_csum_data(kaddr + pgoff, csum,  len);
3001         btrfs_csum_final(csum, (char *)&csum);
3002         if (csum != csum_expected)
3003                 goto zeroit;
3004
3005         kunmap_atomic(kaddr);
3006         return 0;
3007 zeroit:
3008         if (__ratelimit(&_rs))
3009                 btrfs_warn(BTRFS_I(inode)->root->fs_info,
3010                            "csum failed ino %llu off %llu csum %u expected csum %u",
3011                            btrfs_ino(inode), start, csum, csum_expected);
3012         memset(kaddr + pgoff, 1, len);
3013         flush_dcache_page(page);
3014         kunmap_atomic(kaddr);
3015         if (csum_expected == 0)
3016                 return 0;
3017         return -EIO;
3018 }
3019
3020 /*
3021  * when reads are done, we need to check csums to verify the data is correct
3022  * if there's a match, we allow the bio to finish.  If not, the code in
3023  * extent_io.c will try to find good copies for us.
3024  */
3025 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
3026                                       u64 phy_offset, struct page *page,
3027                                       u64 start, u64 end, int mirror)
3028 {
3029         size_t offset = start - page_offset(page);
3030         struct inode *inode = page->mapping->host;
3031         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3032         struct btrfs_root *root = BTRFS_I(inode)->root;
3033
3034         if (PageChecked(page)) {
3035                 ClearPageChecked(page);
3036                 return 0;
3037         }
3038
3039         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
3040                 return 0;
3041
3042         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
3043             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
3044                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
3045                                   GFP_NOFS);
3046                 return 0;
3047         }
3048
3049         phy_offset >>= inode->i_sb->s_blocksize_bits;
3050         return __readpage_endio_check(inode, io_bio, phy_offset, page, offset,
3051                                       start, (size_t)(end - start + 1));
3052 }
3053
3054 struct delayed_iput {
3055         struct list_head list;
3056         struct inode *inode;
3057 };
3058
3059 /* JDM: If this is fs-wide, why can't we add a pointer to
3060  * btrfs_inode instead and avoid the allocation? */
3061 void btrfs_add_delayed_iput(struct inode *inode)
3062 {
3063         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
3064         struct delayed_iput *delayed;
3065
3066         if (atomic_add_unless(&inode->i_count, -1, 1))
3067                 return;
3068
3069         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
3070         delayed->inode = inode;
3071
3072         spin_lock(&fs_info->delayed_iput_lock);
3073         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
3074         spin_unlock(&fs_info->delayed_iput_lock);
3075 }
3076
3077 void btrfs_run_delayed_iputs(struct btrfs_root *root)
3078 {
3079         LIST_HEAD(list);
3080         struct btrfs_fs_info *fs_info = root->fs_info;
3081         struct delayed_iput *delayed;
3082         int empty;
3083
3084         spin_lock(&fs_info->delayed_iput_lock);
3085         empty = list_empty(&fs_info->delayed_iputs);
3086         spin_unlock(&fs_info->delayed_iput_lock);
3087         if (empty)
3088                 return;
3089
3090         spin_lock(&fs_info->delayed_iput_lock);
3091         list_splice_init(&fs_info->delayed_iputs, &list);
3092         spin_unlock(&fs_info->delayed_iput_lock);
3093
3094         while (!list_empty(&list)) {
3095                 delayed = list_entry(list.next, struct delayed_iput, list);
3096                 list_del(&delayed->list);
3097                 iput(delayed->inode);
3098                 kfree(delayed);
3099         }
3100 }
3101
3102 /*
3103  * This is called in transaction commit time. If there are no orphan
3104  * files in the subvolume, it removes orphan item and frees block_rsv
3105  * structure.
3106  */
3107 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
3108                               struct btrfs_root *root)
3109 {
3110         struct btrfs_block_rsv *block_rsv;
3111         int ret;
3112
3113         if (atomic_read(&root->orphan_inodes) ||
3114             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
3115                 return;
3116
3117         spin_lock(&root->orphan_lock);
3118         if (atomic_read(&root->orphan_inodes)) {
3119                 spin_unlock(&root->orphan_lock);
3120                 return;
3121         }
3122
3123         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
3124                 spin_unlock(&root->orphan_lock);
3125                 return;
3126         }
3127
3128         block_rsv = root->orphan_block_rsv;
3129         root->orphan_block_rsv = NULL;
3130         spin_unlock(&root->orphan_lock);
3131
3132         if (test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state) &&
3133             btrfs_root_refs(&root->root_item) > 0) {
3134                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
3135                                             root->root_key.objectid);
3136                 if (ret)
3137                         btrfs_abort_transaction(trans, root, ret);
3138                 else
3139                         clear_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED,
3140                                   &root->state);
3141         }
3142
3143         if (block_rsv) {
3144                 WARN_ON(block_rsv->size > 0);
3145                 btrfs_free_block_rsv(root, block_rsv);
3146         }
3147 }
3148
3149 /*
3150  * This creates an orphan entry for the given inode in case something goes
3151  * wrong in the middle of an unlink/truncate.
3152  *
3153  * NOTE: caller of this function should reserve 5 units of metadata for
3154  *       this function.
3155  */
3156 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
3157 {
3158         struct btrfs_root *root = BTRFS_I(inode)->root;
3159         struct btrfs_block_rsv *block_rsv = NULL;
3160         int reserve = 0;
3161         int insert = 0;
3162         int ret;
3163
3164         if (!root->orphan_block_rsv) {
3165                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3166                 if (!block_rsv)
3167                         return -ENOMEM;
3168         }
3169
3170         spin_lock(&root->orphan_lock);
3171         if (!root->orphan_block_rsv) {
3172                 root->orphan_block_rsv = block_rsv;
3173         } else if (block_rsv) {
3174                 btrfs_free_block_rsv(root, block_rsv);
3175                 block_rsv = NULL;
3176         }
3177
3178         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3179                               &BTRFS_I(inode)->runtime_flags)) {
3180 #if 0
3181                 /*
3182                  * For proper ENOSPC handling, we should do orphan
3183                  * cleanup when mounting. But this introduces backward
3184                  * compatibility issue.
3185                  */
3186                 if (!xchg(&root->orphan_item_inserted, 1))
3187                         insert = 2;
3188                 else
3189                         insert = 1;
3190 #endif
3191                 insert = 1;
3192                 atomic_inc(&root->orphan_inodes);
3193         }
3194
3195         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3196                               &BTRFS_I(inode)->runtime_flags))
3197                 reserve = 1;
3198         spin_unlock(&root->orphan_lock);
3199
3200         /* grab metadata reservation from transaction handle */
3201         if (reserve) {
3202                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3203                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3204         }
3205
3206         /* insert an orphan item to track this unlinked/truncated file */
3207         if (insert >= 1) {
3208                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3209                 if (ret) {
3210                         atomic_dec(&root->orphan_inodes);
3211                         if (reserve) {
3212                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3213                                           &BTRFS_I(inode)->runtime_flags);
3214                                 btrfs_orphan_release_metadata(inode);
3215                         }
3216                         if (ret != -EEXIST) {
3217                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3218                                           &BTRFS_I(inode)->runtime_flags);
3219                                 btrfs_abort_transaction(trans, root, ret);
3220                                 return ret;
3221                         }
3222                 }
3223                 ret = 0;
3224         }
3225
3226         /* insert an orphan item to track subvolume contains orphan files */
3227         if (insert >= 2) {
3228                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3229                                                root->root_key.objectid);
3230                 if (ret && ret != -EEXIST) {
3231                         btrfs_abort_transaction(trans, root, ret);
3232                         return ret;
3233                 }
3234         }
3235         return 0;
3236 }
3237
3238 /*
3239  * We have done the truncate/delete so we can go ahead and remove the orphan
3240  * item for this particular inode.
3241  */
3242 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3243                             struct inode *inode)
3244 {
3245         struct btrfs_root *root = BTRFS_I(inode)->root;
3246         int delete_item = 0;
3247         int release_rsv = 0;
3248         int ret = 0;
3249
3250         spin_lock(&root->orphan_lock);
3251         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3252                                &BTRFS_I(inode)->runtime_flags))
3253                 delete_item = 1;
3254
3255         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3256                                &BTRFS_I(inode)->runtime_flags))
3257                 release_rsv = 1;
3258         spin_unlock(&root->orphan_lock);
3259
3260         if (delete_item) {
3261                 atomic_dec(&root->orphan_inodes);
3262                 if (trans)
3263                         ret = btrfs_del_orphan_item(trans, root,
3264                                                     btrfs_ino(inode));
3265         }
3266
3267         if (release_rsv)
3268                 btrfs_orphan_release_metadata(inode);
3269
3270         return ret;
3271 }
3272
3273 /*
3274  * this cleans up any orphans that may be left on the list from the last use
3275  * of this root.
3276  */
3277 int btrfs_orphan_cleanup(struct btrfs_root *root)
3278 {
3279         struct btrfs_path *path;
3280         struct extent_buffer *leaf;
3281         struct btrfs_key key, found_key;
3282         struct btrfs_trans_handle *trans;
3283         struct inode *inode;
3284         u64 last_objectid = 0;
3285         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3286
3287         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3288                 return 0;
3289
3290         path = btrfs_alloc_path();
3291         if (!path) {
3292                 ret = -ENOMEM;
3293                 goto out;
3294         }
3295         path->reada = -1;
3296
3297         key.objectid = BTRFS_ORPHAN_OBJECTID;
3298         key.type = BTRFS_ORPHAN_ITEM_KEY;
3299         key.offset = (u64)-1;
3300
3301         while (1) {
3302                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3303                 if (ret < 0)
3304                         goto out;
3305
3306                 /*
3307                  * if ret == 0 means we found what we were searching for, which
3308                  * is weird, but possible, so only screw with path if we didn't
3309                  * find the key and see if we have stuff that matches
3310                  */
3311                 if (ret > 0) {
3312                         ret = 0;
3313                         if (path->slots[0] == 0)
3314                                 break;
3315                         path->slots[0]--;
3316                 }
3317
3318                 /* pull out the item */
3319                 leaf = path->nodes[0];
3320                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3321
3322                 /* make sure the item matches what we want */
3323                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3324                         break;
3325                 if (found_key.type != BTRFS_ORPHAN_ITEM_KEY)
3326                         break;
3327
3328                 /* release the path since we're done with it */
3329                 btrfs_release_path(path);
3330
3331                 /*
3332                  * this is where we are basically btrfs_lookup, without the
3333                  * crossing root thing.  we store the inode number in the
3334                  * offset of the orphan item.
3335                  */
3336
3337                 if (found_key.offset == last_objectid) {
3338                         btrfs_err(root->fs_info,
3339                                 "Error removing orphan entry, stopping orphan cleanup");
3340                         ret = -EINVAL;
3341                         goto out;
3342                 }
3343
3344                 last_objectid = found_key.offset;
3345
3346                 found_key.objectid = found_key.offset;
3347                 found_key.type = BTRFS_INODE_ITEM_KEY;
3348                 found_key.offset = 0;
3349                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3350                 ret = PTR_ERR_OR_ZERO(inode);
3351                 if (ret && ret != -ESTALE)
3352                         goto out;
3353
3354                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3355                         struct btrfs_root *dead_root;
3356                         struct btrfs_fs_info *fs_info = root->fs_info;
3357                         int is_dead_root = 0;
3358
3359                         /*
3360                          * this is an orphan in the tree root. Currently these
3361                          * could come from 2 sources:
3362                          *  a) a snapshot deletion in progress
3363                          *  b) a free space cache inode
3364                          * We need to distinguish those two, as the snapshot
3365                          * orphan must not get deleted.
3366                          * find_dead_roots already ran before us, so if this
3367                          * is a snapshot deletion, we should find the root
3368                          * in the dead_roots list
3369                          */
3370                         spin_lock(&fs_info->trans_lock);
3371                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3372                                             root_list) {
3373                                 if (dead_root->root_key.objectid ==
3374                                     found_key.objectid) {
3375                                         is_dead_root = 1;
3376                                         break;
3377                                 }
3378                         }
3379                         spin_unlock(&fs_info->trans_lock);
3380                         if (is_dead_root) {
3381                                 /* prevent this orphan from being found again */
3382                                 key.offset = found_key.objectid - 1;
3383                                 continue;
3384                         }
3385                 }
3386                 /*
3387                  * Inode is already gone but the orphan item is still there,
3388                  * kill the orphan item.
3389                  */
3390                 if (ret == -ESTALE) {
3391                         trans = btrfs_start_transaction(root, 1);
3392                         if (IS_ERR(trans)) {
3393                                 ret = PTR_ERR(trans);
3394                                 goto out;
3395                         }
3396                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3397                                 found_key.objectid);
3398                         ret = btrfs_del_orphan_item(trans, root,
3399                                                     found_key.objectid);
3400                         btrfs_end_transaction(trans, root);
3401                         if (ret)
3402                                 goto out;
3403                         continue;
3404                 }
3405
3406                 /*
3407                  * add this inode to the orphan list so btrfs_orphan_del does
3408                  * the proper thing when we hit it
3409                  */
3410                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3411                         &BTRFS_I(inode)->runtime_flags);
3412                 atomic_inc(&root->orphan_inodes);
3413
3414                 /* if we have links, this was a truncate, lets do that */
3415                 if (inode->i_nlink) {
3416                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3417                                 iput(inode);
3418                                 continue;
3419                         }
3420                         nr_truncate++;
3421
3422                         /* 1 for the orphan item deletion. */
3423                         trans = btrfs_start_transaction(root, 1);
3424                         if (IS_ERR(trans)) {
3425                                 iput(inode);
3426                                 ret = PTR_ERR(trans);
3427                                 goto out;
3428                         }
3429                         ret = btrfs_orphan_add(trans, inode);
3430                         btrfs_end_transaction(trans, root);
3431                         if (ret) {
3432                                 iput(inode);
3433                                 goto out;
3434                         }
3435
3436                         ret = btrfs_truncate(inode);
3437                         if (ret)
3438                                 btrfs_orphan_del(NULL, inode);
3439                 } else {
3440                         nr_unlink++;
3441                 }
3442
3443                 /* this will do delete_inode and everything for us */
3444                 iput(inode);
3445                 if (ret)
3446                         goto out;
3447         }
3448         /* release the path since we're done with it */
3449         btrfs_release_path(path);
3450
3451         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3452
3453         if (root->orphan_block_rsv)
3454                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3455                                         (u64)-1);
3456
3457         if (root->orphan_block_rsv ||
3458             test_bit(BTRFS_ROOT_ORPHAN_ITEM_INSERTED, &root->state)) {
3459                 trans = btrfs_join_transaction(root);
3460                 if (!IS_ERR(trans))
3461                         btrfs_end_transaction(trans, root);
3462         }
3463
3464         if (nr_unlink)
3465                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3466         if (nr_truncate)
3467                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3468
3469 out:
3470         if (ret)
3471                 btrfs_err(root->fs_info,
3472                         "could not do orphan cleanup %d", ret);
3473         btrfs_free_path(path);
3474         return ret;
3475 }
3476
3477 /*
3478  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3479  * don't find any xattrs, we know there can't be any acls.
3480  *
3481  * slot is the slot the inode is in, objectid is the objectid of the inode
3482  */
3483 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3484                                           int slot, u64 objectid,
3485                                           int *first_xattr_slot)
3486 {
3487         u32 nritems = btrfs_header_nritems(leaf);
3488         struct btrfs_key found_key;
3489         static u64 xattr_access = 0;
3490         static u64 xattr_default = 0;
3491         int scanned = 0;
3492
3493         if (!xattr_access) {
3494                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3495                                         strlen(POSIX_ACL_XATTR_ACCESS));
3496                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3497                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3498         }
3499
3500         slot++;
3501         *first_xattr_slot = -1;
3502         while (slot < nritems) {
3503                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3504
3505                 /* we found a different objectid, there must not be acls */
3506                 if (found_key.objectid != objectid)
3507                         return 0;
3508
3509                 /* we found an xattr, assume we've got an acl */
3510                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3511                         if (*first_xattr_slot == -1)
3512                                 *first_xattr_slot = slot;
3513                         if (found_key.offset == xattr_access ||
3514                             found_key.offset == xattr_default)
3515                                 return 1;
3516                 }
3517
3518                 /*
3519                  * we found a key greater than an xattr key, there can't
3520                  * be any acls later on
3521                  */
3522                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3523                         return 0;
3524
3525                 slot++;
3526                 scanned++;
3527
3528                 /*
3529                  * it goes inode, inode backrefs, xattrs, extents,
3530                  * so if there are a ton of hard links to an inode there can
3531                  * be a lot of backrefs.  Don't waste time searching too hard,
3532                  * this is just an optimization
3533                  */
3534                 if (scanned >= 8)
3535                         break;
3536         }
3537         /* we hit the end of the leaf before we found an xattr or
3538          * something larger than an xattr.  We have to assume the inode
3539          * has acls
3540          */
3541         if (*first_xattr_slot == -1)
3542                 *first_xattr_slot = slot;
3543         return 1;
3544 }
3545
3546 /*
3547  * read an inode from the btree into the in-memory inode
3548  */
3549 static void btrfs_read_locked_inode(struct inode *inode)
3550 {
3551         struct btrfs_path *path;
3552         struct extent_buffer *leaf;
3553         struct btrfs_inode_item *inode_item;
3554         struct btrfs_root *root = BTRFS_I(inode)->root;
3555         struct btrfs_key location;
3556         unsigned long ptr;
3557         int maybe_acls;
3558         u32 rdev;
3559         int ret;
3560         bool filled = false;
3561         int first_xattr_slot;
3562
3563         ret = btrfs_fill_inode(inode, &rdev);
3564         if (!ret)
3565                 filled = true;
3566
3567         path = btrfs_alloc_path();
3568         if (!path)
3569                 goto make_bad;
3570
3571         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3572
3573         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3574         if (ret)
3575                 goto make_bad;
3576
3577         leaf = path->nodes[0];
3578
3579         if (filled)
3580                 goto cache_index;
3581
3582         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3583                                     struct btrfs_inode_item);
3584         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3585         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3586         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3587         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3588         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3589
3590         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->atime);
3591         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->atime);
3592
3593         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->mtime);
3594         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->mtime);
3595
3596         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, &inode_item->ctime);
3597         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, &inode_item->ctime);
3598
3599         BTRFS_I(inode)->i_otime.tv_sec =
3600                 btrfs_timespec_sec(leaf, &inode_item->otime);
3601         BTRFS_I(inode)->i_otime.tv_nsec =
3602                 btrfs_timespec_nsec(leaf, &inode_item->otime);
3603
3604         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3605         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3606         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3607
3608         /*
3609          * If we were modified in the current generation and evicted from memory
3610          * and then re-read we need to do a full sync since we don't have any
3611          * idea about which extents were modified before we were evicted from
3612          * cache.
3613          */
3614         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3615                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3616                         &BTRFS_I(inode)->runtime_flags);
3617
3618         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3619         inode->i_generation = BTRFS_I(inode)->generation;
3620         inode->i_rdev = 0;
3621         rdev = btrfs_inode_rdev(leaf, inode_item);
3622
3623         BTRFS_I(inode)->index_cnt = (u64)-1;
3624         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3625
3626 cache_index:
3627         path->slots[0]++;
3628         if (inode->i_nlink != 1 ||
3629             path->slots[0] >= btrfs_header_nritems(leaf))
3630                 goto cache_acl;
3631
3632         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3633         if (location.objectid != btrfs_ino(inode))
3634                 goto cache_acl;
3635
3636         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3637         if (location.type == BTRFS_INODE_REF_KEY) {
3638                 struct btrfs_inode_ref *ref;
3639
3640                 ref = (struct btrfs_inode_ref *)ptr;
3641                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3642         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3643                 struct btrfs_inode_extref *extref;
3644
3645                 extref = (struct btrfs_inode_extref *)ptr;
3646                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3647                                                                      extref);
3648         }
3649 cache_acl:
3650         /*
3651          * try to precache a NULL acl entry for files that don't have
3652          * any xattrs or acls
3653          */
3654         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3655                                            btrfs_ino(inode), &first_xattr_slot);
3656         if (first_xattr_slot != -1) {
3657                 path->slots[0] = first_xattr_slot;
3658                 ret = btrfs_load_inode_props(inode, path);
3659                 if (ret)
3660                         btrfs_err(root->fs_info,
3661                                   "error loading props for ino %llu (root %llu): %d",
3662                                   btrfs_ino(inode),
3663                                   root->root_key.objectid, ret);
3664         }
3665         btrfs_free_path(path);
3666
3667         if (!maybe_acls)
3668                 cache_no_acl(inode);
3669
3670         switch (inode->i_mode & S_IFMT) {
3671         case S_IFREG:
3672                 inode->i_mapping->a_ops = &btrfs_aops;
3673                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3674                 inode->i_fop = &btrfs_file_operations;
3675                 inode->i_op = &btrfs_file_inode_operations;
3676                 break;
3677         case S_IFDIR:
3678                 inode->i_fop = &btrfs_dir_file_operations;
3679                 if (root == root->fs_info->tree_root)
3680                         inode->i_op = &btrfs_dir_ro_inode_operations;
3681                 else
3682                         inode->i_op = &btrfs_dir_inode_operations;
3683                 break;
3684         case S_IFLNK:
3685                 inode->i_op = &btrfs_symlink_inode_operations;
3686                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3687                 break;
3688         default:
3689                 inode->i_op = &btrfs_special_inode_operations;
3690                 init_special_inode(inode, inode->i_mode, rdev);
3691                 break;
3692         }
3693
3694         btrfs_update_iflags(inode);
3695         return;
3696
3697 make_bad:
3698         btrfs_free_path(path);
3699         make_bad_inode(inode);
3700 }
3701
3702 /*
3703  * given a leaf and an inode, copy the inode fields into the leaf
3704  */
3705 static void fill_inode_item(struct btrfs_trans_handle *trans,
3706                             struct extent_buffer *leaf,
3707                             struct btrfs_inode_item *item,
3708                             struct inode *inode)
3709 {
3710         struct btrfs_map_token token;
3711
3712         btrfs_init_map_token(&token);
3713
3714         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3715         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3716         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3717                                    &token);
3718         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3719         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3720
3721         btrfs_set_token_timespec_sec(leaf, &item->atime,
3722                                      inode->i_atime.tv_sec, &token);
3723         btrfs_set_token_timespec_nsec(leaf, &item->atime,
3724                                       inode->i_atime.tv_nsec, &token);
3725
3726         btrfs_set_token_timespec_sec(leaf, &item->mtime,
3727                                      inode->i_mtime.tv_sec, &token);
3728         btrfs_set_token_timespec_nsec(leaf, &item->mtime,
3729                                       inode->i_mtime.tv_nsec, &token);
3730
3731         btrfs_set_token_timespec_sec(leaf, &item->ctime,
3732                                      inode->i_ctime.tv_sec, &token);
3733         btrfs_set_token_timespec_nsec(leaf, &item->ctime,
3734                                       inode->i_ctime.tv_nsec, &token);
3735
3736         btrfs_set_token_timespec_sec(leaf, &item->otime,
3737                                      BTRFS_I(inode)->i_otime.tv_sec, &token);
3738         btrfs_set_token_timespec_nsec(leaf, &item->otime,
3739                                       BTRFS_I(inode)->i_otime.tv_nsec, &token);
3740
3741         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3742                                      &token);
3743         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3744                                          &token);
3745         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3746         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3747         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3748         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3749         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3750 }
3751
3752 /*
3753  * copy everything in the in-memory inode into the btree.
3754  */
3755 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3756                                 struct btrfs_root *root, struct inode *inode)
3757 {
3758         struct btrfs_inode_item *inode_item;
3759         struct btrfs_path *path;
3760         struct extent_buffer *leaf;
3761         int ret;
3762
3763         path = btrfs_alloc_path();
3764         if (!path)
3765                 return -ENOMEM;
3766
3767         path->leave_spinning = 1;
3768         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3769                                  1);
3770         if (ret) {
3771                 if (ret > 0)
3772                         ret = -ENOENT;
3773                 goto failed;
3774         }
3775
3776         leaf = path->nodes[0];
3777         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3778                                     struct btrfs_inode_item);
3779
3780         fill_inode_item(trans, leaf, inode_item, inode);
3781         btrfs_mark_buffer_dirty(leaf);
3782         btrfs_set_inode_last_trans(trans, inode);
3783         ret = 0;
3784 failed:
3785         btrfs_free_path(path);
3786         return ret;
3787 }
3788
3789 /*
3790  * copy everything in the in-memory inode into the btree.
3791  */
3792 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3793                                 struct btrfs_root *root, struct inode *inode)
3794 {
3795         int ret;
3796
3797         /*
3798          * If the inode is a free space inode, we can deadlock during commit
3799          * if we put it into the delayed code.
3800          *
3801          * The data relocation inode should also be directly updated
3802          * without delay
3803          */
3804         if (!btrfs_is_free_space_inode(inode)
3805             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
3806             && !root->fs_info->log_root_recovering) {
3807                 btrfs_update_root_times(trans, root);
3808
3809                 ret = btrfs_delayed_update_inode(trans, root, inode);
3810                 if (!ret)
3811                         btrfs_set_inode_last_trans(trans, inode);
3812                 return ret;
3813         }
3814
3815         return btrfs_update_inode_item(trans, root, inode);
3816 }
3817
3818 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3819                                          struct btrfs_root *root,
3820                                          struct inode *inode)
3821 {
3822         int ret;
3823
3824         ret = btrfs_update_inode(trans, root, inode);
3825         if (ret == -ENOSPC)
3826                 return btrfs_update_inode_item(trans, root, inode);
3827         return ret;
3828 }
3829
3830 /*
3831  * unlink helper that gets used here in inode.c and in the tree logging
3832  * recovery code.  It remove a link in a directory with a given name, and
3833  * also drops the back refs in the inode to the directory
3834  */
3835 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3836                                 struct btrfs_root *root,
3837                                 struct inode *dir, struct inode *inode,
3838                                 const char *name, int name_len)
3839 {
3840         struct btrfs_path *path;
3841         int ret = 0;
3842         struct extent_buffer *leaf;
3843         struct btrfs_dir_item *di;
3844         struct btrfs_key key;
3845         u64 index;
3846         u64 ino = btrfs_ino(inode);
3847         u64 dir_ino = btrfs_ino(dir);
3848
3849         path = btrfs_alloc_path();
3850         if (!path) {
3851                 ret = -ENOMEM;
3852                 goto out;
3853         }
3854
3855         path->leave_spinning = 1;
3856         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3857                                     name, name_len, -1);
3858         if (IS_ERR(di)) {
3859                 ret = PTR_ERR(di);
3860                 goto err;
3861         }
3862         if (!di) {
3863                 ret = -ENOENT;
3864                 goto err;
3865         }
3866         leaf = path->nodes[0];
3867         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3868         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3869         if (ret)
3870                 goto err;
3871         btrfs_release_path(path);
3872
3873         /*
3874          * If we don't have dir index, we have to get it by looking up
3875          * the inode ref, since we get the inode ref, remove it directly,
3876          * it is unnecessary to do delayed deletion.
3877          *
3878          * But if we have dir index, needn't search inode ref to get it.
3879          * Since the inode ref is close to the inode item, it is better
3880          * that we delay to delete it, and just do this deletion when
3881          * we update the inode item.
3882          */
3883         if (BTRFS_I(inode)->dir_index) {
3884                 ret = btrfs_delayed_delete_inode_ref(inode);
3885                 if (!ret) {
3886                         index = BTRFS_I(inode)->dir_index;
3887                         goto skip_backref;
3888                 }
3889         }
3890
3891         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3892                                   dir_ino, &index);
3893         if (ret) {
3894                 btrfs_info(root->fs_info,
3895                         "failed to delete reference to %.*s, inode %llu parent %llu",
3896                         name_len, name, ino, dir_ino);
3897                 btrfs_abort_transaction(trans, root, ret);
3898                 goto err;
3899         }
3900 skip_backref:
3901         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3902         if (ret) {
3903                 btrfs_abort_transaction(trans, root, ret);
3904                 goto err;
3905         }
3906
3907         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3908                                          inode, dir_ino);
3909         if (ret != 0 && ret != -ENOENT) {
3910                 btrfs_abort_transaction(trans, root, ret);
3911                 goto err;
3912         }
3913
3914         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3915                                            dir, index);
3916         if (ret == -ENOENT)
3917                 ret = 0;
3918         else if (ret)
3919                 btrfs_abort_transaction(trans, root, ret);
3920 err:
3921         btrfs_free_path(path);
3922         if (ret)
3923                 goto out;
3924
3925         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3926         inode_inc_iversion(inode);
3927         inode_inc_iversion(dir);
3928         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3929         ret = btrfs_update_inode(trans, root, dir);
3930 out:
3931         return ret;
3932 }
3933
3934 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3935                        struct btrfs_root *root,
3936                        struct inode *dir, struct inode *inode,
3937                        const char *name, int name_len)
3938 {
3939         int ret;
3940         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3941         if (!ret) {
3942                 drop_nlink(inode);
3943                 ret = btrfs_update_inode(trans, root, inode);
3944         }
3945         return ret;
3946 }
3947
3948 /*
3949  * helper to start transaction for unlink and rmdir.
3950  *
3951  * unlink and rmdir are special in btrfs, they do not always free space, so
3952  * if we cannot make our reservations the normal way try and see if there is
3953  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3954  * allow the unlink to occur.
3955  */
3956 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3957 {
3958         struct btrfs_trans_handle *trans;
3959         struct btrfs_root *root = BTRFS_I(dir)->root;
3960         int ret;
3961
3962         /*
3963          * 1 for the possible orphan item
3964          * 1 for the dir item
3965          * 1 for the dir index
3966          * 1 for the inode ref
3967          * 1 for the inode
3968          */
3969         trans = btrfs_start_transaction(root, 5);
3970         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3971                 return trans;
3972
3973         if (PTR_ERR(trans) == -ENOSPC) {
3974                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3975
3976                 trans = btrfs_start_transaction(root, 0);
3977                 if (IS_ERR(trans))
3978                         return trans;
3979                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3980                                                &root->fs_info->trans_block_rsv,
3981                                                num_bytes, 5);
3982                 if (ret) {
3983                         btrfs_end_transaction(trans, root);
3984                         return ERR_PTR(ret);
3985                 }
3986                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3987                 trans->bytes_reserved = num_bytes;
3988         }
3989         return trans;
3990 }
3991
3992 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3993 {
3994         struct btrfs_root *root = BTRFS_I(dir)->root;
3995         struct btrfs_trans_handle *trans;
3996         struct inode *inode = dentry->d_inode;
3997         int ret;
3998
3999         trans = __unlink_start_trans(dir);
4000         if (IS_ERR(trans))
4001                 return PTR_ERR(trans);
4002
4003         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
4004
4005         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4006                                  dentry->d_name.name, dentry->d_name.len);
4007         if (ret)
4008                 goto out;
4009
4010         if (inode->i_nlink == 0) {
4011                 ret = btrfs_orphan_add(trans, inode);
4012                 if (ret)
4013                         goto out;
4014         }
4015
4016 out:
4017         btrfs_end_transaction(trans, root);
4018         btrfs_btree_balance_dirty(root);
4019         return ret;
4020 }
4021
4022 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
4023                         struct btrfs_root *root,
4024                         struct inode *dir, u64 objectid,
4025                         const char *name, int name_len)
4026 {
4027         struct btrfs_path *path;
4028         struct extent_buffer *leaf;
4029         struct btrfs_dir_item *di;
4030         struct btrfs_key key;
4031         u64 index;
4032         int ret;
4033         u64 dir_ino = btrfs_ino(dir);
4034
4035         path = btrfs_alloc_path();
4036         if (!path)
4037                 return -ENOMEM;
4038
4039         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
4040                                    name, name_len, -1);
4041         if (IS_ERR_OR_NULL(di)) {
4042                 if (!di)
4043                         ret = -ENOENT;
4044                 else
4045                         ret = PTR_ERR(di);
4046                 goto out;
4047         }
4048
4049         leaf = path->nodes[0];
4050         btrfs_dir_item_key_to_cpu(leaf, di, &key);
4051         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
4052         ret = btrfs_delete_one_dir_name(trans, root, path, di);
4053         if (ret) {
4054                 btrfs_abort_transaction(trans, root, ret);
4055                 goto out;
4056         }
4057         btrfs_release_path(path);
4058
4059         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4060                                  objectid, root->root_key.objectid,
4061                                  dir_ino, &index, name, name_len);
4062         if (ret < 0) {
4063                 if (ret != -ENOENT) {
4064                         btrfs_abort_transaction(trans, root, ret);
4065                         goto out;
4066                 }
4067                 di = btrfs_search_dir_index_item(root, path, dir_ino,
4068                                                  name, name_len);
4069                 if (IS_ERR_OR_NULL(di)) {
4070                         if (!di)
4071                                 ret = -ENOENT;
4072                         else
4073                                 ret = PTR_ERR(di);
4074                         btrfs_abort_transaction(trans, root, ret);
4075                         goto out;
4076                 }
4077
4078                 leaf = path->nodes[0];
4079                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4080                 btrfs_release_path(path);
4081                 index = key.offset;
4082         }
4083         btrfs_release_path(path);
4084
4085         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
4086         if (ret) {
4087                 btrfs_abort_transaction(trans, root, ret);
4088                 goto out;
4089         }
4090
4091         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
4092         inode_inc_iversion(dir);
4093         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
4094         ret = btrfs_update_inode_fallback(trans, root, dir);
4095         if (ret)
4096                 btrfs_abort_transaction(trans, root, ret);
4097 out:
4098         btrfs_free_path(path);
4099         return ret;
4100 }
4101
4102 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
4103 {
4104         struct inode *inode = dentry->d_inode;
4105         int err = 0;
4106         struct btrfs_root *root = BTRFS_I(dir)->root;
4107         struct btrfs_trans_handle *trans;
4108
4109         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
4110                 return -ENOTEMPTY;
4111         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
4112                 return -EPERM;
4113
4114         trans = __unlink_start_trans(dir);
4115         if (IS_ERR(trans))
4116                 return PTR_ERR(trans);
4117
4118         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4119                 err = btrfs_unlink_subvol(trans, root, dir,
4120                                           BTRFS_I(inode)->location.objectid,
4121                                           dentry->d_name.name,
4122                                           dentry->d_name.len);
4123                 goto out;
4124         }
4125
4126         err = btrfs_orphan_add(trans, inode);
4127         if (err)
4128                 goto out;
4129
4130         /* now the directory is empty */
4131         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4132                                  dentry->d_name.name, dentry->d_name.len);
4133         if (!err)
4134                 btrfs_i_size_write(inode, 0);
4135 out:
4136         btrfs_end_transaction(trans, root);
4137         btrfs_btree_balance_dirty(root);
4138
4139         return err;
4140 }
4141
4142 /*
4143  * this can truncate away extent items, csum items and directory items.
4144  * It starts at a high offset and removes keys until it can't find
4145  * any higher than new_size
4146  *
4147  * csum items that cross the new i_size are truncated to the new size
4148  * as well.
4149  *
4150  * min_type is the minimum key type to truncate down to.  If set to 0, this
4151  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4152  */
4153 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4154                                struct btrfs_root *root,
4155                                struct inode *inode,
4156                                u64 new_size, u32 min_type)
4157 {
4158         struct btrfs_path *path;
4159         struct extent_buffer *leaf;
4160         struct btrfs_file_extent_item *fi;
4161         struct btrfs_key key;
4162         struct btrfs_key found_key;
4163         u64 extent_start = 0;
4164         u64 extent_num_bytes = 0;
4165         u64 extent_offset = 0;
4166         u64 item_end = 0;
4167         u64 last_size = (u64)-1;
4168         u32 found_type = (u8)-1;
4169         int found_extent;
4170         int del_item;
4171         int pending_del_nr = 0;
4172         int pending_del_slot = 0;
4173         int extent_type = -1;
4174         int ret;
4175         int err = 0;
4176         u64 ino = btrfs_ino(inode);
4177
4178         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4179
4180         path = btrfs_alloc_path();
4181         if (!path)
4182                 return -ENOMEM;
4183         path->reada = -1;
4184
4185         /*
4186          * We want to drop from the next block forward in case this new size is
4187          * not block aligned since we will be keeping the last block of the
4188          * extent just the way it is.
4189          */
4190         if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4191             root == root->fs_info->tree_root)
4192                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4193                                         root->sectorsize), (u64)-1, 0);
4194
4195         /*
4196          * This function is also used to drop the items in the log tree before
4197          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4198          * it is used to drop the loged items. So we shouldn't kill the delayed
4199          * items.
4200          */
4201         if (min_type == 0 && root == BTRFS_I(inode)->root)
4202                 btrfs_kill_delayed_inode_items(inode);
4203
4204         key.objectid = ino;
4205         key.offset = (u64)-1;
4206         key.type = (u8)-1;
4207
4208 search_again:
4209         path->leave_spinning = 1;
4210         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4211         if (ret < 0) {
4212                 err = ret;
4213                 goto out;
4214         }
4215
4216         if (ret > 0) {
4217                 /* there are no items in the tree for us to truncate, we're
4218                  * done
4219                  */
4220                 if (path->slots[0] == 0)
4221                         goto out;
4222                 path->slots[0]--;
4223         }
4224
4225         while (1) {
4226                 fi = NULL;
4227                 leaf = path->nodes[0];
4228                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4229                 found_type = found_key.type;
4230
4231                 if (found_key.objectid != ino)
4232                         break;
4233
4234                 if (found_type < min_type)
4235                         break;
4236
4237                 item_end = found_key.offset;
4238                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4239                         fi = btrfs_item_ptr(leaf, path->slots[0],
4240                                             struct btrfs_file_extent_item);
4241                         extent_type = btrfs_file_extent_type(leaf, fi);
4242                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4243                                 item_end +=
4244                                     btrfs_file_extent_num_bytes(leaf, fi);
4245                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4246                                 item_end += btrfs_file_extent_inline_len(leaf,
4247                                                          path->slots[0], fi);
4248                         }
4249                         item_end--;
4250                 }
4251                 if (found_type > min_type) {
4252                         del_item = 1;
4253                 } else {
4254                         if (item_end < new_size)
4255                                 break;
4256                         if (found_key.offset >= new_size)
4257                                 del_item = 1;
4258                         else
4259                                 del_item = 0;
4260                 }
4261                 found_extent = 0;
4262                 /* FIXME, shrink the extent if the ref count is only 1 */
4263                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4264                         goto delete;
4265
4266                 if (del_item)
4267                         last_size = found_key.offset;
4268                 else
4269                         last_size = new_size;
4270
4271                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4272                         u64 num_dec;
4273                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4274                         if (!del_item) {
4275                                 u64 orig_num_bytes =
4276                                         btrfs_file_extent_num_bytes(leaf, fi);
4277                                 extent_num_bytes = ALIGN(new_size -
4278                                                 found_key.offset,
4279                                                 root->sectorsize);
4280                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4281                                                          extent_num_bytes);
4282                                 num_dec = (orig_num_bytes -
4283                                            extent_num_bytes);
4284                                 if (test_bit(BTRFS_ROOT_REF_COWS,
4285                                              &root->state) &&
4286                                     extent_start != 0)
4287                                         inode_sub_bytes(inode, num_dec);
4288                                 btrfs_mark_buffer_dirty(leaf);
4289                         } else {
4290                                 extent_num_bytes =
4291                                         btrfs_file_extent_disk_num_bytes(leaf,
4292                                                                          fi);
4293                                 extent_offset = found_key.offset -
4294                                         btrfs_file_extent_offset(leaf, fi);
4295
4296                                 /* FIXME blocksize != 4096 */
4297                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4298                                 if (extent_start != 0) {
4299                                         found_extent = 1;
4300                                         if (test_bit(BTRFS_ROOT_REF_COWS,
4301                                                      &root->state))
4302                                                 inode_sub_bytes(inode, num_dec);
4303                                 }
4304                         }
4305                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4306                         /*
4307                          * we can't truncate inline items that have had
4308                          * special encodings
4309                          */
4310                         if (!del_item &&
4311                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4312                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4313                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4314                                 u32 size = new_size - found_key.offset;
4315
4316                                 if (test_bit(BTRFS_ROOT_REF_COWS, &root->state))
4317                                         inode_sub_bytes(inode, item_end + 1 -
4318                                                         new_size);
4319
4320                                 /*
4321                                  * update the ram bytes to properly reflect
4322                                  * the new size of our item
4323                                  */
4324                                 btrfs_set_file_extent_ram_bytes(leaf, fi, size);
4325                                 size =
4326                                     btrfs_file_extent_calc_inline_size(size);
4327                                 btrfs_truncate_item(root, path, size, 1);
4328                         } else if (test_bit(BTRFS_ROOT_REF_COWS,
4329                                             &root->state)) {
4330                                 inode_sub_bytes(inode, item_end + 1 -
4331                                                 found_key.offset);
4332                         }
4333                 }
4334 delete:
4335                 if (del_item) {
4336                         if (!pending_del_nr) {
4337                                 /* no pending yet, add ourselves */
4338                                 pending_del_slot = path->slots[0];
4339                                 pending_del_nr = 1;
4340                         } else if (pending_del_nr &&
4341                                    path->slots[0] + 1 == pending_del_slot) {
4342                                 /* hop on the pending chunk */
4343                                 pending_del_nr++;
4344                                 pending_del_slot = path->slots[0];
4345                         } else {
4346                                 BUG();
4347                         }
4348                 } else {
4349                         break;
4350                 }
4351                 if (found_extent &&
4352                     (test_bit(BTRFS_ROOT_REF_COWS, &root->state) ||
4353                      root == root->fs_info->tree_root)) {
4354                         btrfs_set_path_blocking(path);
4355                         ret = btrfs_free_extent(trans, root, extent_start,
4356                                                 extent_num_bytes, 0,
4357                                                 btrfs_header_owner(leaf),
4358                                                 ino, extent_offset, 0);
4359                         BUG_ON(ret);
4360                 }
4361
4362                 if (found_type == BTRFS_INODE_ITEM_KEY)
4363                         break;
4364
4365                 if (path->slots[0] == 0 ||
4366                     path->slots[0] != pending_del_slot) {
4367                         if (pending_del_nr) {
4368                                 ret = btrfs_del_items(trans, root, path,
4369                                                 pending_del_slot,
4370                                                 pending_del_nr);
4371                                 if (ret) {
4372                                         btrfs_abort_transaction(trans,
4373                                                                 root, ret);
4374                                         goto error;
4375                                 }
4376                                 pending_del_nr = 0;
4377                         }
4378                         btrfs_release_path(path);
4379                         goto search_again;
4380                 } else {
4381                         path->slots[0]--;
4382                 }
4383         }
4384 out:
4385         if (pending_del_nr) {
4386                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4387                                       pending_del_nr);
4388                 if (ret)
4389                         btrfs_abort_transaction(trans, root, ret);
4390         }
4391 error:
4392         if (last_size != (u64)-1 &&
4393             root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID)
4394                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4395         btrfs_free_path(path);
4396         return err;
4397 }
4398
4399 /*
4400  * btrfs_truncate_page - read, zero a chunk and write a page
4401  * @inode - inode that we're zeroing
4402  * @from - the offset to start zeroing
4403  * @len - the length to zero, 0 to zero the entire range respective to the
4404  *      offset
4405  * @front - zero up to the offset instead of from the offset on
4406  *
4407  * This will find the page for the "from" offset and cow the page and zero the
4408  * part we want to zero.  This is used with truncate and hole punching.
4409  */
4410 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4411                         int front)
4412 {
4413         struct address_space *mapping = inode->i_mapping;
4414         struct btrfs_root *root = BTRFS_I(inode)->root;
4415         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4416         struct btrfs_ordered_extent *ordered;
4417         struct extent_state *cached_state = NULL;
4418         char *kaddr;
4419         u32 blocksize = root->sectorsize;
4420         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4421         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4422         struct page *page;
4423         gfp_t mask = btrfs_alloc_write_mask(mapping);
4424         int ret = 0;
4425         u64 page_start;
4426         u64 page_end;
4427
4428         if ((offset & (blocksize - 1)) == 0 &&
4429             (!len || ((len & (blocksize - 1)) == 0)))
4430                 goto out;
4431         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4432         if (ret)
4433                 goto out;
4434
4435 again:
4436         page = find_or_create_page(mapping, index, mask);
4437         if (!page) {
4438                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4439                 ret = -ENOMEM;
4440                 goto out;
4441         }
4442
4443         page_start = page_offset(page);
4444         page_end = page_start + PAGE_CACHE_SIZE - 1;
4445
4446         if (!PageUptodate(page)) {
4447                 ret = btrfs_readpage(NULL, page);
4448                 lock_page(page);
4449                 if (page->mapping != mapping) {
4450                         unlock_page(page);
4451                         page_cache_release(page);
4452                         goto again;
4453                 }
4454                 if (!PageUptodate(page)) {
4455                         ret = -EIO;
4456                         goto out_unlock;
4457                 }
4458         }
4459         wait_on_page_writeback(page);
4460
4461         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4462         set_page_extent_mapped(page);
4463
4464         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4465         if (ordered) {
4466                 unlock_extent_cached(io_tree, page_start, page_end,
4467                                      &cached_state, GFP_NOFS);
4468                 unlock_page(page);
4469                 page_cache_release(page);
4470                 btrfs_start_ordered_extent(inode, ordered, 1);
4471                 btrfs_put_ordered_extent(ordered);
4472                 goto again;
4473         }
4474
4475         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4476                           EXTENT_DIRTY | EXTENT_DELALLOC |
4477                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4478                           0, 0, &cached_state, GFP_NOFS);
4479
4480         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4481                                         &cached_state);
4482         if (ret) {
4483                 unlock_extent_cached(io_tree, page_start, page_end,
4484                                      &cached_state, GFP_NOFS);
4485                 goto out_unlock;
4486         }
4487
4488         if (offset != PAGE_CACHE_SIZE) {
4489                 if (!len)
4490                         len = PAGE_CACHE_SIZE - offset;
4491                 kaddr = kmap(page);
4492                 if (front)
4493                         memset(kaddr, 0, offset);
4494                 else
4495                         memset(kaddr + offset, 0, len);
4496                 flush_dcache_page(page);
4497                 kunmap(page);
4498         }
4499         ClearPageChecked(page);
4500         set_page_dirty(page);
4501         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4502                              GFP_NOFS);
4503
4504 out_unlock:
4505         if (ret)
4506                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4507         unlock_page(page);
4508         page_cache_release(page);
4509 out:
4510         return ret;
4511 }
4512
4513 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4514                              u64 offset, u64 len)
4515 {
4516         struct btrfs_trans_handle *trans;
4517         int ret;
4518
4519         /*
4520          * Still need to make sure the inode looks like it's been updated so
4521          * that any holes get logged if we fsync.
4522          */
4523         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4524                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4525                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4526                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4527                 return 0;
4528         }
4529
4530         /*
4531          * 1 - for the one we're dropping
4532          * 1 - for the one we're adding
4533          * 1 - for updating the inode.
4534          */
4535         trans = btrfs_start_transaction(root, 3);
4536         if (IS_ERR(trans))
4537                 return PTR_ERR(trans);
4538
4539         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4540         if (ret) {
4541                 btrfs_abort_transaction(trans, root, ret);
4542                 btrfs_end_transaction(trans, root);
4543                 return ret;
4544         }
4545
4546         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4547                                        0, 0, len, 0, len, 0, 0, 0);
4548         if (ret)
4549                 btrfs_abort_transaction(trans, root, ret);
4550         else
4551                 btrfs_update_inode(trans, root, inode);
4552         btrfs_end_transaction(trans, root);
4553         return ret;
4554 }
4555
4556 /*
4557  * This function puts in dummy file extents for the area we're creating a hole
4558  * for.  So if we are truncating this file to a larger size we need to insert
4559  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4560  * the range between oldsize and size
4561  */
4562 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4563 {
4564         struct btrfs_root *root = BTRFS_I(inode)->root;
4565         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4566         struct extent_map *em = NULL;
4567         struct extent_state *cached_state = NULL;
4568         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4569         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4570         u64 block_end = ALIGN(size, root->sectorsize);
4571         u64 last_byte;
4572         u64 cur_offset;
4573         u64 hole_size;
4574         int err = 0;
4575
4576         /*
4577          * If our size started in the middle of a page we need to zero out the
4578          * rest of the page before we expand the i_size, otherwise we could
4579          * expose stale data.
4580          */
4581         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4582         if (err)
4583                 return err;
4584
4585         if (size <= hole_start)
4586                 return 0;
4587
4588         while (1) {
4589                 struct btrfs_ordered_extent *ordered;
4590
4591                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4592                                  &cached_state);
4593                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4594                                                      block_end - hole_start);
4595                 if (!ordered)
4596                         break;
4597                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4598                                      &cached_state, GFP_NOFS);
4599                 btrfs_start_ordered_extent(inode, ordered, 1);
4600                 btrfs_put_ordered_extent(ordered);
4601         }
4602
4603         cur_offset = hole_start;
4604         while (1) {
4605                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4606                                 block_end - cur_offset, 0);
4607                 if (IS_ERR(em)) {
4608                         err = PTR_ERR(em);
4609                         em = NULL;
4610                         break;
4611                 }
4612                 last_byte = min(extent_map_end(em), block_end);
4613                 last_byte = ALIGN(last_byte , root->sectorsize);
4614                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4615                         struct extent_map *hole_em;
4616                         hole_size = last_byte - cur_offset;
4617
4618                         err = maybe_insert_hole(root, inode, cur_offset,
4619                                                 hole_size);
4620                         if (err)
4621                                 break;
4622                         btrfs_drop_extent_cache(inode, cur_offset,
4623                                                 cur_offset + hole_size - 1, 0);
4624                         hole_em = alloc_extent_map();
4625                         if (!hole_em) {
4626                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4627                                         &BTRFS_I(inode)->runtime_flags);
4628                                 goto next;
4629                         }
4630                         hole_em->start = cur_offset;
4631                         hole_em->len = hole_size;
4632                         hole_em->orig_start = cur_offset;
4633
4634                         hole_em->block_start = EXTENT_MAP_HOLE;
4635                         hole_em->block_len = 0;
4636                         hole_em->orig_block_len = 0;
4637                         hole_em->ram_bytes = hole_size;
4638                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4639                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4640                         hole_em->generation = root->fs_info->generation;
4641
4642                         while (1) {
4643                                 write_lock(&em_tree->lock);
4644                                 err = add_extent_mapping(em_tree, hole_em, 1);
4645                                 write_unlock(&em_tree->lock);
4646                                 if (err != -EEXIST)
4647                                         break;
4648                                 btrfs_drop_extent_cache(inode, cur_offset,
4649                                                         cur_offset +
4650                                                         hole_size - 1, 0);
4651                         }
4652                         free_extent_map(hole_em);
4653                 }
4654 next:
4655                 free_extent_map(em);
4656                 em = NULL;
4657                 cur_offset = last_byte;
4658                 if (cur_offset >= block_end)
4659                         break;
4660         }
4661         free_extent_map(em);
4662         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4663                              GFP_NOFS);
4664         return err;
4665 }
4666
4667 static int wait_snapshoting_atomic_t(atomic_t *a)
4668 {
4669         schedule();
4670         return 0;
4671 }
4672
4673 static void wait_for_snapshot_creation(struct btrfs_root *root)
4674 {
4675         while (true) {
4676                 int ret;
4677
4678                 ret = btrfs_start_write_no_snapshoting(root);
4679                 if (ret)
4680                         break;
4681                 wait_on_atomic_t(&root->will_be_snapshoted,
4682                                  wait_snapshoting_atomic_t,
4683                                  TASK_UNINTERRUPTIBLE);
4684         }
4685 }
4686
4687 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4688 {
4689         struct btrfs_root *root = BTRFS_I(inode)->root;
4690         struct btrfs_trans_handle *trans;
4691         loff_t oldsize = i_size_read(inode);
4692         loff_t newsize = attr->ia_size;
4693         int mask = attr->ia_valid;
4694         int ret;
4695
4696         /*
4697          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4698          * special case where we need to update the times despite not having
4699          * these flags set.  For all other operations the VFS set these flags
4700          * explicitly if it wants a timestamp update.
4701          */
4702         if (newsize != oldsize) {
4703                 inode_inc_iversion(inode);
4704                 if (!(mask & (ATTR_CTIME | ATTR_MTIME)))
4705                         inode->i_ctime = inode->i_mtime =
4706                                 current_fs_time(inode->i_sb);
4707         }
4708
4709         if (newsize > oldsize) {
4710                 truncate_pagecache(inode, newsize);
4711                 /*
4712                  * Don't do an expanding truncate while snapshoting is ongoing.
4713                  * This is to ensure the snapshot captures a fully consistent
4714                  * state of this file - if the snapshot captures this expanding
4715                  * truncation, it must capture all writes that happened before
4716                  * this truncation.
4717                  */
4718                 wait_for_snapshot_creation(root);
4719                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4720                 if (ret) {
4721                         btrfs_end_write_no_snapshoting(root);
4722                         return ret;
4723                 }
4724
4725                 trans = btrfs_start_transaction(root, 1);
4726                 if (IS_ERR(trans)) {
4727                         btrfs_end_write_no_snapshoting(root);
4728                         return PTR_ERR(trans);
4729                 }
4730
4731                 i_size_write(inode, newsize);
4732                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4733                 ret = btrfs_update_inode(trans, root, inode);
4734                 btrfs_end_write_no_snapshoting(root);
4735                 btrfs_end_transaction(trans, root);
4736         } else {
4737
4738                 /*
4739                  * We're truncating a file that used to have good data down to
4740                  * zero. Make sure it gets into the ordered flush list so that
4741                  * any new writes get down to disk quickly.
4742                  */
4743                 if (newsize == 0)
4744                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4745                                 &BTRFS_I(inode)->runtime_flags);
4746
4747                 /*
4748                  * 1 for the orphan item we're going to add
4749                  * 1 for the orphan item deletion.
4750                  */
4751                 trans = btrfs_start_transaction(root, 2);
4752                 if (IS_ERR(trans))
4753                         return PTR_ERR(trans);
4754
4755                 /*
4756                  * We need to do this in case we fail at _any_ point during the
4757                  * actual truncate.  Once we do the truncate_setsize we could
4758                  * invalidate pages which forces any outstanding ordered io to
4759                  * be instantly completed which will give us extents that need
4760                  * to be truncated.  If we fail to get an orphan inode down we
4761                  * could have left over extents that were never meant to live,
4762                  * so we need to garuntee from this point on that everything
4763                  * will be consistent.
4764                  */
4765                 ret = btrfs_orphan_add(trans, inode);
4766                 btrfs_end_transaction(trans, root);
4767                 if (ret)
4768                         return ret;
4769
4770                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4771                 truncate_setsize(inode, newsize);
4772
4773                 /* Disable nonlocked read DIO to avoid the end less truncate */
4774                 btrfs_inode_block_unlocked_dio(inode);
4775                 inode_dio_wait(inode);
4776                 btrfs_inode_resume_unlocked_dio(inode);
4777
4778                 ret = btrfs_truncate(inode);
4779                 if (ret && inode->i_nlink) {
4780                         int err;
4781
4782                         /*
4783                          * failed to truncate, disk_i_size is only adjusted down
4784                          * as we remove extents, so it should represent the true
4785                          * size of the inode, so reset the in memory size and
4786                          * delete our orphan entry.
4787                          */
4788                         trans = btrfs_join_transaction(root);
4789                         if (IS_ERR(trans)) {
4790                                 btrfs_orphan_del(NULL, inode);
4791                                 return ret;
4792                         }
4793                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4794                         err = btrfs_orphan_del(trans, inode);
4795                         if (err)
4796                                 btrfs_abort_transaction(trans, root, err);
4797                         btrfs_end_transaction(trans, root);
4798                 }
4799         }
4800
4801         return ret;
4802 }
4803
4804 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4805 {
4806         struct inode *inode = dentry->d_inode;
4807         struct btrfs_root *root = BTRFS_I(inode)->root;
4808         int err;
4809
4810         if (btrfs_root_readonly(root))
4811                 return -EROFS;
4812
4813         err = inode_change_ok(inode, attr);
4814         if (err)
4815                 return err;
4816
4817         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4818                 err = btrfs_setsize(inode, attr);
4819                 if (err)
4820                         return err;
4821         }
4822
4823         if (attr->ia_valid) {
4824                 setattr_copy(inode, attr);
4825                 inode_inc_iversion(inode);
4826                 err = btrfs_dirty_inode(inode);
4827
4828                 if (!err && attr->ia_valid & ATTR_MODE)
4829                         err = posix_acl_chmod(inode, inode->i_mode);
4830         }
4831
4832         return err;
4833 }
4834
4835 /*
4836  * While truncating the inode pages during eviction, we get the VFS calling
4837  * btrfs_invalidatepage() against each page of the inode. This is slow because
4838  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4839  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4840  * extent_state structures over and over, wasting lots of time.
4841  *
4842  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4843  * those expensive operations on a per page basis and do only the ordered io
4844  * finishing, while we release here the extent_map and extent_state structures,
4845  * without the excessive merging and splitting.
4846  */
4847 static void evict_inode_truncate_pages(struct inode *inode)
4848 {
4849         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4850         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4851         struct rb_node *node;
4852
4853         ASSERT(inode->i_state & I_FREEING);
4854         truncate_inode_pages_final(&inode->i_data);
4855
4856         write_lock(&map_tree->lock);
4857         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4858                 struct extent_map *em;
4859
4860                 node = rb_first(&map_tree->map);
4861                 em = rb_entry(node, struct extent_map, rb_node);
4862                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4863                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4864                 remove_extent_mapping(map_tree, em);
4865                 free_extent_map(em);
4866                 if (need_resched()) {
4867                         write_unlock(&map_tree->lock);
4868                         cond_resched();
4869                         write_lock(&map_tree->lock);
4870                 }
4871         }
4872         write_unlock(&map_tree->lock);
4873
4874         spin_lock(&io_tree->lock);
4875         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4876                 struct extent_state *state;
4877                 struct extent_state *cached_state = NULL;
4878
4879                 node = rb_first(&io_tree->state);
4880                 state = rb_entry(node, struct extent_state, rb_node);
4881                 atomic_inc(&state->refs);
4882                 spin_unlock(&io_tree->lock);
4883
4884                 lock_extent_bits(io_tree, state->start, state->end,
4885                                  0, &cached_state);
4886                 clear_extent_bit(io_tree, state->start, state->end,
4887                                  EXTENT_LOCKED | EXTENT_DIRTY |
4888                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4889                                  EXTENT_DEFRAG, 1, 1,
4890                                  &cached_state, GFP_NOFS);
4891                 free_extent_state(state);
4892
4893                 cond_resched();
4894                 spin_lock(&io_tree->lock);
4895         }
4896         spin_unlock(&io_tree->lock);
4897 }
4898
4899 void btrfs_evict_inode(struct inode *inode)
4900 {
4901         struct btrfs_trans_handle *trans;
4902         struct btrfs_root *root = BTRFS_I(inode)->root;
4903         struct btrfs_block_rsv *rsv, *global_rsv;
4904         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4905         int ret;
4906
4907         trace_btrfs_inode_evict(inode);
4908
4909         evict_inode_truncate_pages(inode);
4910
4911         if (inode->i_nlink &&
4912             ((btrfs_root_refs(&root->root_item) != 0 &&
4913               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4914              btrfs_is_free_space_inode(inode)))
4915                 goto no_delete;
4916
4917         if (is_bad_inode(inode)) {
4918                 btrfs_orphan_del(NULL, inode);
4919                 goto no_delete;
4920         }
4921         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4922         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4923
4924         btrfs_free_io_failure_record(inode, 0, (u64)-1);
4925
4926         if (root->fs_info->log_root_recovering) {
4927                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4928                                  &BTRFS_I(inode)->runtime_flags));
4929                 goto no_delete;
4930         }
4931
4932         if (inode->i_nlink > 0) {
4933                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4934                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4935                 goto no_delete;
4936         }
4937
4938         ret = btrfs_commit_inode_delayed_inode(inode);
4939         if (ret) {
4940                 btrfs_orphan_del(NULL, inode);
4941                 goto no_delete;
4942         }
4943
4944         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4945         if (!rsv) {
4946                 btrfs_orphan_del(NULL, inode);
4947                 goto no_delete;
4948         }
4949         rsv->size = min_size;
4950         rsv->failfast = 1;
4951         global_rsv = &root->fs_info->global_block_rsv;
4952
4953         btrfs_i_size_write(inode, 0);
4954
4955         /*
4956          * This is a bit simpler than btrfs_truncate since we've already
4957          * reserved our space for our orphan item in the unlink, so we just
4958          * need to reserve some slack space in case we add bytes and update
4959          * inode item when doing the truncate.
4960          */
4961         while (1) {
4962                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4963                                              BTRFS_RESERVE_FLUSH_LIMIT);
4964
4965                 /*
4966                  * Try and steal from the global reserve since we will
4967                  * likely not use this space anyway, we want to try as
4968                  * hard as possible to get this to work.
4969                  */
4970                 if (ret)
4971                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4972
4973                 if (ret) {
4974                         btrfs_warn(root->fs_info,
4975                                 "Could not get space for a delete, will truncate on mount %d",
4976                                 ret);
4977                         btrfs_orphan_del(NULL, inode);
4978                         btrfs_free_block_rsv(root, rsv);
4979                         goto no_delete;
4980                 }
4981
4982                 trans = btrfs_join_transaction(root);
4983                 if (IS_ERR(trans)) {
4984                         btrfs_orphan_del(NULL, inode);
4985                         btrfs_free_block_rsv(root, rsv);
4986                         goto no_delete;
4987                 }
4988
4989                 trans->block_rsv = rsv;
4990
4991                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4992                 if (ret != -ENOSPC)
4993                         break;
4994
4995                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4996                 btrfs_end_transaction(trans, root);
4997                 trans = NULL;
4998                 btrfs_btree_balance_dirty(root);
4999         }
5000
5001         btrfs_free_block_rsv(root, rsv);
5002
5003         /*
5004          * Errors here aren't a big deal, it just means we leave orphan items
5005          * in the tree.  They will be cleaned up on the next mount.
5006          */
5007         if (ret == 0) {
5008                 trans->block_rsv = root->orphan_block_rsv;
5009                 btrfs_orphan_del(trans, inode);
5010         } else {
5011                 btrfs_orphan_del(NULL, inode);
5012         }
5013
5014         trans->block_rsv = &root->fs_info->trans_block_rsv;
5015         if (!(root == root->fs_info->tree_root ||
5016               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
5017                 btrfs_return_ino(root, btrfs_ino(inode));
5018
5019         btrfs_end_transaction(trans, root);
5020         btrfs_btree_balance_dirty(root);
5021 no_delete:
5022         btrfs_remove_delayed_node(inode);
5023         clear_inode(inode);
5024         return;
5025 }
5026
5027 /*
5028  * this returns the key found in the dir entry in the location pointer.
5029  * If no dir entries were found, location->objectid is 0.
5030  */
5031 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
5032                                struct btrfs_key *location)
5033 {
5034         const char *name = dentry->d_name.name;
5035         int namelen = dentry->d_name.len;
5036         struct btrfs_dir_item *di;
5037         struct btrfs_path *path;
5038         struct btrfs_root *root = BTRFS_I(dir)->root;
5039         int ret = 0;
5040
5041         path = btrfs_alloc_path();
5042         if (!path)
5043                 return -ENOMEM;
5044
5045         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
5046                                     namelen, 0);
5047         if (IS_ERR(di))
5048                 ret = PTR_ERR(di);
5049
5050         if (IS_ERR_OR_NULL(di))
5051                 goto out_err;
5052
5053         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
5054 out:
5055         btrfs_free_path(path);
5056         return ret;
5057 out_err:
5058         location->objectid = 0;
5059         goto out;
5060 }
5061
5062 /*
5063  * when we hit a tree root in a directory, the btrfs part of the inode
5064  * needs to be changed to reflect the root directory of the tree root.  This
5065  * is kind of like crossing a mount point.
5066  */
5067 static int fixup_tree_root_location(struct btrfs_root *root,
5068                                     struct inode *dir,
5069                                     struct dentry *dentry,
5070                                     struct btrfs_key *location,
5071                                     struct btrfs_root **sub_root)
5072 {
5073         struct btrfs_path *path;
5074         struct btrfs_root *new_root;
5075         struct btrfs_root_ref *ref;
5076         struct extent_buffer *leaf;
5077         struct btrfs_key key;
5078         int ret;
5079         int err = 0;
5080
5081         path = btrfs_alloc_path();
5082         if (!path) {
5083                 err = -ENOMEM;
5084                 goto out;
5085         }
5086
5087         err = -ENOENT;
5088         key.objectid = BTRFS_I(dir)->root->root_key.objectid;
5089         key.type = BTRFS_ROOT_REF_KEY;
5090         key.offset = location->objectid;
5091
5092         ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, path,
5093                                 0, 0);
5094         if (ret) {
5095                 if (ret < 0)
5096                         err = ret;
5097                 goto out;
5098         }
5099
5100         leaf = path->nodes[0];
5101         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
5102         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
5103             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
5104                 goto out;
5105
5106         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
5107                                    (unsigned long)(ref + 1),
5108                                    dentry->d_name.len);
5109         if (ret)
5110                 goto out;
5111
5112         btrfs_release_path(path);
5113
5114         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
5115         if (IS_ERR(new_root)) {
5116                 err = PTR_ERR(new_root);
5117                 goto out;
5118         }
5119
5120         *sub_root = new_root;
5121         location->objectid = btrfs_root_dirid(&new_root->root_item);
5122         location->type = BTRFS_INODE_ITEM_KEY;
5123         location->offset = 0;
5124         err = 0;
5125 out:
5126         btrfs_free_path(path);
5127         return err;
5128 }
5129
5130 static void inode_tree_add(struct inode *inode)
5131 {
5132         struct btrfs_root *root = BTRFS_I(inode)->root;
5133         struct btrfs_inode *entry;
5134         struct rb_node **p;
5135         struct rb_node *parent;
5136         struct rb_node *new = &BTRFS_I(inode)->rb_node;
5137         u64 ino = btrfs_ino(inode);
5138
5139         if (inode_unhashed(inode))
5140                 return;
5141         parent = NULL;
5142         spin_lock(&root->inode_lock);
5143         p = &root->inode_tree.rb_node;
5144         while (*p) {
5145                 parent = *p;
5146                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
5147
5148                 if (ino < btrfs_ino(&entry->vfs_inode))
5149                         p = &parent->rb_left;
5150                 else if (ino > btrfs_ino(&entry->vfs_inode))
5151                         p = &parent->rb_right;
5152                 else {
5153                         WARN_ON(!(entry->vfs_inode.i_state &
5154                                   (I_WILL_FREE | I_FREEING)));
5155                         rb_replace_node(parent, new, &root->inode_tree);
5156                         RB_CLEAR_NODE(parent);
5157                         spin_unlock(&root->inode_lock);
5158                         return;
5159                 }
5160         }
5161         rb_link_node(new, parent, p);
5162         rb_insert_color(new, &root->inode_tree);
5163         spin_unlock(&root->inode_lock);
5164 }
5165
5166 static void inode_tree_del(struct inode *inode)
5167 {
5168         struct btrfs_root *root = BTRFS_I(inode)->root;
5169         int empty = 0;
5170
5171         spin_lock(&root->inode_lock);
5172         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
5173                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
5174                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
5175                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5176         }
5177         spin_unlock(&root->inode_lock);
5178
5179         if (empty && btrfs_root_refs(&root->root_item) == 0) {
5180                 synchronize_srcu(&root->fs_info->subvol_srcu);
5181                 spin_lock(&root->inode_lock);
5182                 empty = RB_EMPTY_ROOT(&root->inode_tree);
5183                 spin_unlock(&root->inode_lock);
5184                 if (empty)
5185                         btrfs_add_dead_root(root);
5186         }
5187 }
5188
5189 void btrfs_invalidate_inodes(struct btrfs_root *root)
5190 {
5191         struct rb_node *node;
5192         struct rb_node *prev;
5193         struct btrfs_inode *entry;
5194         struct inode *inode;
5195         u64 objectid = 0;
5196
5197         if (!test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
5198                 WARN_ON(btrfs_root_refs(&root->root_item) != 0);
5199
5200         spin_lock(&root->inode_lock);
5201 again:
5202         node = root->inode_tree.rb_node;
5203         prev = NULL;
5204         while (node) {
5205                 prev = node;
5206                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5207
5208                 if (objectid < btrfs_ino(&entry->vfs_inode))
5209                         node = node->rb_left;
5210                 else if (objectid > btrfs_ino(&entry->vfs_inode))
5211                         node = node->rb_right;
5212                 else
5213                         break;
5214         }
5215         if (!node) {
5216                 while (prev) {
5217                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
5218                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
5219                                 node = prev;
5220                                 break;
5221                         }
5222                         prev = rb_next(prev);
5223                 }
5224         }
5225         while (node) {
5226                 entry = rb_entry(node, struct btrfs_inode, rb_node);
5227                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
5228                 inode = igrab(&entry->vfs_inode);
5229                 if (inode) {
5230                         spin_unlock(&root->inode_lock);
5231                         if (atomic_read(&inode->i_count) > 1)
5232                                 d_prune_aliases(inode);
5233                         /*
5234                          * btrfs_drop_inode will have it removed from
5235                          * the inode cache when its usage count
5236                          * hits zero.
5237                          */
5238                         iput(inode);
5239                         cond_resched();
5240                         spin_lock(&root->inode_lock);
5241                         goto again;
5242                 }
5243
5244                 if (cond_resched_lock(&root->inode_lock))
5245                         goto again;
5246
5247                 node = rb_next(node);
5248         }
5249         spin_unlock(&root->inode_lock);
5250 }
5251
5252 static int btrfs_init_locked_inode(struct inode *inode, void *p)
5253 {
5254         struct btrfs_iget_args *args = p;
5255         inode->i_ino = args->location->objectid;
5256         memcpy(&BTRFS_I(inode)->location, args->location,
5257                sizeof(*args->location));
5258         BTRFS_I(inode)->root = args->root;
5259         return 0;
5260 }
5261
5262 static int btrfs_find_actor(struct inode *inode, void *opaque)
5263 {
5264         struct btrfs_iget_args *args = opaque;
5265         return args->location->objectid == BTRFS_I(inode)->location.objectid &&
5266                 args->root == BTRFS_I(inode)->root;
5267 }
5268
5269 static struct inode *btrfs_iget_locked(struct super_block *s,
5270                                        struct btrfs_key *location,
5271                                        struct btrfs_root *root)
5272 {
5273         struct inode *inode;
5274         struct btrfs_iget_args args;
5275         unsigned long hashval = btrfs_inode_hash(location->objectid, root);
5276
5277         args.location = location;
5278         args.root = root;
5279
5280         inode = iget5_locked(s, hashval, btrfs_find_actor,
5281                              btrfs_init_locked_inode,
5282                              (void *)&args);
5283         return inode;
5284 }
5285
5286 /* Get an inode object given its location and corresponding root.
5287  * Returns in *is_new if the inode was read from disk
5288  */
5289 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5290                          struct btrfs_root *root, int *new)
5291 {
5292         struct inode *inode;
5293
5294         inode = btrfs_iget_locked(s, location, root);
5295         if (!inode)
5296                 return ERR_PTR(-ENOMEM);
5297
5298         if (inode->i_state & I_NEW) {
5299                 btrfs_read_locked_inode(inode);
5300                 if (!is_bad_inode(inode)) {
5301                         inode_tree_add(inode);
5302                         unlock_new_inode(inode);
5303                         if (new)
5304                                 *new = 1;
5305                 } else {
5306                         unlock_new_inode(inode);
5307                         iput(inode);
5308                         inode = ERR_PTR(-ESTALE);
5309                 }
5310         }
5311
5312         return inode;
5313 }
5314
5315 static struct inode *new_simple_dir(struct super_block *s,
5316                                     struct btrfs_key *key,
5317                                     struct btrfs_root *root)
5318 {
5319         struct inode *inode = new_inode(s);
5320
5321         if (!inode)
5322                 return ERR_PTR(-ENOMEM);
5323
5324         BTRFS_I(inode)->root = root;
5325         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5326         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5327
5328         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5329         inode->i_op = &btrfs_dir_ro_inode_operations;
5330         inode->i_fop = &simple_dir_operations;
5331         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5332         inode->i_mtime = CURRENT_TIME;
5333         inode->i_atime = inode->i_mtime;
5334         inode->i_ctime = inode->i_mtime;
5335         BTRFS_I(inode)->i_otime = inode->i_mtime;
5336
5337         return inode;
5338 }
5339
5340 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5341 {
5342         struct inode *inode;
5343         struct btrfs_root *root = BTRFS_I(dir)->root;
5344         struct btrfs_root *sub_root = root;
5345         struct btrfs_key location;
5346         int index;
5347         int ret = 0;
5348
5349         if (dentry->d_name.len > BTRFS_NAME_LEN)
5350                 return ERR_PTR(-ENAMETOOLONG);
5351
5352         ret = btrfs_inode_by_name(dir, dentry, &location);
5353         if (ret < 0)
5354                 return ERR_PTR(ret);
5355
5356         if (location.objectid == 0)
5357                 return ERR_PTR(-ENOENT);
5358
5359         if (location.type == BTRFS_INODE_ITEM_KEY) {
5360                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5361                 return inode;
5362         }
5363
5364         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5365
5366         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5367         ret = fixup_tree_root_location(root, dir, dentry,
5368                                        &location, &sub_root);
5369         if (ret < 0) {
5370                 if (ret != -ENOENT)
5371                         inode = ERR_PTR(ret);
5372                 else
5373                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5374         } else {
5375                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5376         }
5377         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5378
5379         if (!IS_ERR(inode) && root != sub_root) {
5380                 down_read(&root->fs_info->cleanup_work_sem);
5381                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5382                         ret = btrfs_orphan_cleanup(sub_root);
5383                 up_read(&root->fs_info->cleanup_work_sem);
5384                 if (ret) {
5385                         iput(inode);
5386                         inode = ERR_PTR(ret);
5387                 }
5388         }
5389
5390         return inode;
5391 }
5392
5393 static int btrfs_dentry_delete(const struct dentry *dentry)
5394 {
5395         struct btrfs_root *root;
5396         struct inode *inode = dentry->d_inode;
5397
5398         if (!inode && !IS_ROOT(dentry))
5399                 inode = dentry->d_parent->d_inode;
5400
5401         if (inode) {
5402                 root = BTRFS_I(inode)->root;
5403                 if (btrfs_root_refs(&root->root_item) == 0)
5404                         return 1;
5405
5406                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5407                         return 1;
5408         }
5409         return 0;
5410 }
5411
5412 static void btrfs_dentry_release(struct dentry *dentry)
5413 {
5414         kfree(dentry->d_fsdata);
5415 }
5416
5417 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5418                                    unsigned int flags)
5419 {
5420         struct inode *inode;
5421
5422         inode = btrfs_lookup_dentry(dir, dentry);
5423         if (IS_ERR(inode)) {
5424                 if (PTR_ERR(inode) == -ENOENT)
5425                         inode = NULL;
5426                 else
5427                         return ERR_CAST(inode);
5428         }
5429
5430         return d_splice_alias(inode, dentry);
5431 }
5432
5433 unsigned char btrfs_filetype_table[] = {
5434         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5435 };
5436
5437 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5438 {
5439         struct inode *inode = file_inode(file);
5440         struct btrfs_root *root = BTRFS_I(inode)->root;
5441         struct btrfs_item *item;
5442         struct btrfs_dir_item *di;
5443         struct btrfs_key key;
5444         struct btrfs_key found_key;
5445         struct btrfs_path *path;
5446         struct list_head ins_list;
5447         struct list_head del_list;
5448         int ret;
5449         struct extent_buffer *leaf;
5450         int slot;
5451         unsigned char d_type;
5452         int over = 0;
5453         u32 di_cur;
5454         u32 di_total;
5455         u32 di_len;
5456         int key_type = BTRFS_DIR_INDEX_KEY;
5457         char tmp_name[32];
5458         char *name_ptr;
5459         int name_len;
5460         int is_curr = 0;        /* ctx->pos points to the current index? */
5461
5462         /* FIXME, use a real flag for deciding about the key type */
5463         if (root->fs_info->tree_root == root)
5464                 key_type = BTRFS_DIR_ITEM_KEY;
5465
5466         if (!dir_emit_dots(file, ctx))
5467                 return 0;
5468
5469         path = btrfs_alloc_path();
5470         if (!path)
5471                 return -ENOMEM;
5472
5473         path->reada = 1;
5474
5475         if (key_type == BTRFS_DIR_INDEX_KEY) {
5476                 INIT_LIST_HEAD(&ins_list);
5477                 INIT_LIST_HEAD(&del_list);
5478                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5479         }
5480
5481         key.type = key_type;
5482         key.offset = ctx->pos;
5483         key.objectid = btrfs_ino(inode);
5484
5485         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5486         if (ret < 0)
5487                 goto err;
5488
5489         while (1) {
5490                 leaf = path->nodes[0];
5491                 slot = path->slots[0];
5492                 if (slot >= btrfs_header_nritems(leaf)) {
5493                         ret = btrfs_next_leaf(root, path);
5494                         if (ret < 0)
5495                                 goto err;
5496                         else if (ret > 0)
5497                                 break;
5498                         continue;
5499                 }
5500
5501                 item = btrfs_item_nr(slot);
5502                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5503
5504                 if (found_key.objectid != key.objectid)
5505                         break;
5506                 if (found_key.type != key_type)
5507                         break;
5508                 if (found_key.offset < ctx->pos)
5509                         goto next;
5510                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5511                     btrfs_should_delete_dir_index(&del_list,
5512                                                   found_key.offset))
5513                         goto next;
5514
5515                 ctx->pos = found_key.offset;
5516                 is_curr = 1;
5517
5518                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5519                 di_cur = 0;
5520                 di_total = btrfs_item_size(leaf, item);
5521
5522                 while (di_cur < di_total) {
5523                         struct btrfs_key location;
5524
5525                         if (verify_dir_item(root, leaf, di))
5526                                 break;
5527
5528                         name_len = btrfs_dir_name_len(leaf, di);
5529                         if (name_len <= sizeof(tmp_name)) {
5530                                 name_ptr = tmp_name;
5531                         } else {
5532                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5533                                 if (!name_ptr) {
5534                                         ret = -ENOMEM;
5535                                         goto err;
5536                                 }
5537                         }
5538                         read_extent_buffer(leaf, name_ptr,
5539                                            (unsigned long)(di + 1), name_len);
5540
5541                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5542                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5543
5544
5545                         /* is this a reference to our own snapshot? If so
5546                          * skip it.
5547                          *
5548                          * In contrast to old kernels, we insert the snapshot's
5549                          * dir item and dir index after it has been created, so
5550                          * we won't find a reference to our own snapshot. We
5551                          * still keep the following code for backward
5552                          * compatibility.
5553                          */
5554                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5555                             location.objectid == root->root_key.objectid) {
5556                                 over = 0;
5557                                 goto skip;
5558                         }
5559                         over = !dir_emit(ctx, name_ptr, name_len,
5560                                        location.objectid, d_type);
5561
5562 skip:
5563                         if (name_ptr != tmp_name)
5564                                 kfree(name_ptr);
5565
5566                         if (over)
5567                                 goto nopos;
5568                         di_len = btrfs_dir_name_len(leaf, di) +
5569                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5570                         di_cur += di_len;
5571                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5572                 }
5573 next:
5574                 path->slots[0]++;
5575         }
5576
5577         if (key_type == BTRFS_DIR_INDEX_KEY) {
5578                 if (is_curr)
5579                         ctx->pos++;
5580                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5581                 if (ret)
5582                         goto nopos;
5583         }
5584
5585         /* Reached end of directory/root. Bump pos past the last item. */
5586         ctx->pos++;
5587
5588         /*
5589          * Stop new entries from being returned after we return the last
5590          * entry.
5591          *
5592          * New directory entries are assigned a strictly increasing
5593          * offset.  This means that new entries created during readdir
5594          * are *guaranteed* to be seen in the future by that readdir.
5595          * This has broken buggy programs which operate on names as
5596          * they're returned by readdir.  Until we re-use freed offsets
5597          * we have this hack to stop new entries from being returned
5598          * under the assumption that they'll never reach this huge
5599          * offset.
5600          *
5601          * This is being careful not to overflow 32bit loff_t unless the
5602          * last entry requires it because doing so has broken 32bit apps
5603          * in the past.
5604          */
5605         if (key_type == BTRFS_DIR_INDEX_KEY) {
5606                 if (ctx->pos >= INT_MAX)
5607                         ctx->pos = LLONG_MAX;
5608                 else
5609                         ctx->pos = INT_MAX;
5610         }
5611 nopos:
5612         ret = 0;
5613 err:
5614         if (key_type == BTRFS_DIR_INDEX_KEY)
5615                 btrfs_put_delayed_items(&ins_list, &del_list);
5616         btrfs_free_path(path);
5617         return ret;
5618 }
5619
5620 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5621 {
5622         struct btrfs_root *root = BTRFS_I(inode)->root;
5623         struct btrfs_trans_handle *trans;
5624         int ret = 0;
5625         bool nolock = false;
5626
5627         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5628                 return 0;
5629
5630         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5631                 nolock = true;
5632
5633         if (wbc->sync_mode == WB_SYNC_ALL) {
5634                 if (nolock)
5635                         trans = btrfs_join_transaction_nolock(root);
5636                 else
5637                         trans = btrfs_join_transaction(root);
5638                 if (IS_ERR(trans))
5639                         return PTR_ERR(trans);
5640                 ret = btrfs_commit_transaction(trans, root);
5641         }
5642         return ret;
5643 }
5644
5645 /*
5646  * This is somewhat expensive, updating the tree every time the
5647  * inode changes.  But, it is most likely to find the inode in cache.
5648  * FIXME, needs more benchmarking...there are no reasons other than performance
5649  * to keep or drop this code.
5650  */
5651 static int btrfs_dirty_inode(struct inode *inode)
5652 {
5653         struct btrfs_root *root = BTRFS_I(inode)->root;
5654         struct btrfs_trans_handle *trans;
5655         int ret;
5656
5657         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5658                 return 0;
5659
5660         trans = btrfs_join_transaction(root);
5661         if (IS_ERR(trans))
5662                 return PTR_ERR(trans);
5663
5664         ret = btrfs_update_inode(trans, root, inode);
5665         if (ret && ret == -ENOSPC) {
5666                 /* whoops, lets try again with the full transaction */
5667                 btrfs_end_transaction(trans, root);
5668                 trans = btrfs_start_transaction(root, 1);
5669                 if (IS_ERR(trans))
5670                         return PTR_ERR(trans);
5671
5672                 ret = btrfs_update_inode(trans, root, inode);
5673         }
5674         btrfs_end_transaction(trans, root);
5675         if (BTRFS_I(inode)->delayed_node)
5676                 btrfs_balance_delayed_items(root);
5677
5678         return ret;
5679 }
5680
5681 /*
5682  * This is a copy of file_update_time.  We need this so we can return error on
5683  * ENOSPC for updating the inode in the case of file write and mmap writes.
5684  */
5685 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5686                              int flags)
5687 {
5688         struct btrfs_root *root = BTRFS_I(inode)->root;
5689
5690         if (btrfs_root_readonly(root))
5691                 return -EROFS;
5692
5693         if (flags & S_VERSION)
5694                 inode_inc_iversion(inode);
5695         if (flags & S_CTIME)
5696                 inode->i_ctime = *now;
5697         if (flags & S_MTIME)
5698                 inode->i_mtime = *now;
5699         if (flags & S_ATIME)
5700                 inode->i_atime = *now;
5701         return btrfs_dirty_inode(inode);
5702 }
5703
5704 /*
5705  * find the highest existing sequence number in a directory
5706  * and then set the in-memory index_cnt variable to reflect
5707  * free sequence numbers
5708  */
5709 static int btrfs_set_inode_index_count(struct inode *inode)
5710 {
5711         struct btrfs_root *root = BTRFS_I(inode)->root;
5712         struct btrfs_key key, found_key;
5713         struct btrfs_path *path;
5714         struct extent_buffer *leaf;
5715         int ret;
5716
5717         key.objectid = btrfs_ino(inode);
5718         key.type = BTRFS_DIR_INDEX_KEY;
5719         key.offset = (u64)-1;
5720
5721         path = btrfs_alloc_path();
5722         if (!path)
5723                 return -ENOMEM;
5724
5725         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5726         if (ret < 0)
5727                 goto out;
5728         /* FIXME: we should be able to handle this */
5729         if (ret == 0)
5730                 goto out;
5731         ret = 0;
5732
5733         /*
5734          * MAGIC NUMBER EXPLANATION:
5735          * since we search a directory based on f_pos we have to start at 2
5736          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5737          * else has to start at 2
5738          */
5739         if (path->slots[0] == 0) {
5740                 BTRFS_I(inode)->index_cnt = 2;
5741                 goto out;
5742         }
5743
5744         path->slots[0]--;
5745
5746         leaf = path->nodes[0];
5747         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5748
5749         if (found_key.objectid != btrfs_ino(inode) ||
5750             found_key.type != BTRFS_DIR_INDEX_KEY) {
5751                 BTRFS_I(inode)->index_cnt = 2;
5752                 goto out;
5753         }
5754
5755         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5756 out:
5757         btrfs_free_path(path);
5758         return ret;
5759 }
5760
5761 /*
5762  * helper to find a free sequence number in a given directory.  This current
5763  * code is very simple, later versions will do smarter things in the btree
5764  */
5765 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5766 {
5767         int ret = 0;
5768
5769         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5770                 ret = btrfs_inode_delayed_dir_index_count(dir);
5771                 if (ret) {
5772                         ret = btrfs_set_inode_index_count(dir);
5773                         if (ret)
5774                                 return ret;
5775                 }
5776         }
5777
5778         *index = BTRFS_I(dir)->index_cnt;
5779         BTRFS_I(dir)->index_cnt++;
5780
5781         return ret;
5782 }
5783
5784 static int btrfs_insert_inode_locked(struct inode *inode)
5785 {
5786         struct btrfs_iget_args args;
5787         args.location = &BTRFS_I(inode)->location;
5788         args.root = BTRFS_I(inode)->root;
5789
5790         return insert_inode_locked4(inode,
5791                    btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root),
5792                    btrfs_find_actor, &args);
5793 }
5794
5795 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5796                                      struct btrfs_root *root,
5797                                      struct inode *dir,
5798                                      const char *name, int name_len,
5799                                      u64 ref_objectid, u64 objectid,
5800                                      umode_t mode, u64 *index)
5801 {
5802         struct inode *inode;
5803         struct btrfs_inode_item *inode_item;
5804         struct btrfs_key *location;
5805         struct btrfs_path *path;
5806         struct btrfs_inode_ref *ref;
5807         struct btrfs_key key[2];
5808         u32 sizes[2];
5809         int nitems = name ? 2 : 1;
5810         unsigned long ptr;
5811         int ret;
5812
5813         path = btrfs_alloc_path();
5814         if (!path)
5815                 return ERR_PTR(-ENOMEM);
5816
5817         inode = new_inode(root->fs_info->sb);
5818         if (!inode) {
5819                 btrfs_free_path(path);
5820                 return ERR_PTR(-ENOMEM);
5821         }
5822
5823         /*
5824          * O_TMPFILE, set link count to 0, so that after this point,
5825          * we fill in an inode item with the correct link count.
5826          */
5827         if (!name)
5828                 set_nlink(inode, 0);
5829
5830         /*
5831          * we have to initialize this early, so we can reclaim the inode
5832          * number if we fail afterwards in this function.
5833          */
5834         inode->i_ino = objectid;
5835
5836         if (dir && name) {
5837                 trace_btrfs_inode_request(dir);
5838
5839                 ret = btrfs_set_inode_index(dir, index);
5840                 if (ret) {
5841                         btrfs_free_path(path);
5842                         iput(inode);
5843                         return ERR_PTR(ret);
5844                 }
5845         } else if (dir) {
5846                 *index = 0;
5847         }
5848         /*
5849          * index_cnt is ignored for everything but a dir,
5850          * btrfs_get_inode_index_count has an explanation for the magic
5851          * number
5852          */
5853         BTRFS_I(inode)->index_cnt = 2;
5854         BTRFS_I(inode)->dir_index = *index;
5855         BTRFS_I(inode)->root = root;
5856         BTRFS_I(inode)->generation = trans->transid;
5857         inode->i_generation = BTRFS_I(inode)->generation;
5858
5859         /*
5860          * We could have gotten an inode number from somebody who was fsynced
5861          * and then removed in this same transaction, so let's just set full
5862          * sync since it will be a full sync anyway and this will blow away the
5863          * old info in the log.
5864          */
5865         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5866
5867         key[0].objectid = objectid;
5868         key[0].type = BTRFS_INODE_ITEM_KEY;
5869         key[0].offset = 0;
5870
5871         sizes[0] = sizeof(struct btrfs_inode_item);
5872
5873         if (name) {
5874                 /*
5875                  * Start new inodes with an inode_ref. This is slightly more
5876                  * efficient for small numbers of hard links since they will
5877                  * be packed into one item. Extended refs will kick in if we
5878                  * add more hard links than can fit in the ref item.
5879                  */
5880                 key[1].objectid = objectid;
5881                 key[1].type = BTRFS_INODE_REF_KEY;
5882                 key[1].offset = ref_objectid;
5883
5884                 sizes[1] = name_len + sizeof(*ref);
5885         }
5886
5887         location = &BTRFS_I(inode)->location;
5888         location->objectid = objectid;
5889         location->offset = 0;
5890         location->type = BTRFS_INODE_ITEM_KEY;
5891
5892         ret = btrfs_insert_inode_locked(inode);
5893         if (ret < 0)
5894                 goto fail;
5895
5896         path->leave_spinning = 1;
5897         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, nitems);
5898         if (ret != 0)
5899                 goto fail_unlock;
5900
5901         inode_init_owner(inode, dir, mode);
5902         inode_set_bytes(inode, 0);
5903
5904         inode->i_mtime = CURRENT_TIME;
5905         inode->i_atime = inode->i_mtime;
5906         inode->i_ctime = inode->i_mtime;
5907         BTRFS_I(inode)->i_otime = inode->i_mtime;
5908
5909         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5910                                   struct btrfs_inode_item);
5911         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5912                              sizeof(*inode_item));
5913         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5914
5915         if (name) {
5916                 ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5917                                      struct btrfs_inode_ref);
5918                 btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5919                 btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5920                 ptr = (unsigned long)(ref + 1);
5921                 write_extent_buffer(path->nodes[0], name, ptr, name_len);
5922         }
5923
5924         btrfs_mark_buffer_dirty(path->nodes[0]);
5925         btrfs_free_path(path);
5926
5927         btrfs_inherit_iflags(inode, dir);
5928
5929         if (S_ISREG(mode)) {
5930                 if (btrfs_test_opt(root, NODATASUM))
5931                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5932                 if (btrfs_test_opt(root, NODATACOW))
5933                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5934                                 BTRFS_INODE_NODATASUM;
5935         }
5936
5937         inode_tree_add(inode);
5938
5939         trace_btrfs_inode_new(inode);
5940         btrfs_set_inode_last_trans(trans, inode);
5941
5942         btrfs_update_root_times(trans, root);
5943
5944         ret = btrfs_inode_inherit_props(trans, inode, dir);
5945         if (ret)
5946                 btrfs_err(root->fs_info,
5947                           "error inheriting props for ino %llu (root %llu): %d",
5948                           btrfs_ino(inode), root->root_key.objectid, ret);
5949
5950         return inode;
5951
5952 fail_unlock:
5953         unlock_new_inode(inode);
5954 fail:
5955         if (dir && name)
5956                 BTRFS_I(dir)->index_cnt--;
5957         btrfs_free_path(path);
5958         iput(inode);
5959         return ERR_PTR(ret);
5960 }
5961
5962 static inline u8 btrfs_inode_type(struct inode *inode)
5963 {
5964         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5965 }
5966
5967 /*
5968  * utility function to add 'inode' into 'parent_inode' with
5969  * a give name and a given sequence number.
5970  * if 'add_backref' is true, also insert a backref from the
5971  * inode to the parent directory.
5972  */
5973 int btrfs_add_link(struct btrfs_trans_handle *trans,
5974                    struct inode *parent_inode, struct inode *inode,
5975                    const char *name, int name_len, int add_backref, u64 index)
5976 {
5977         int ret = 0;
5978         struct btrfs_key key;
5979         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5980         u64 ino = btrfs_ino(inode);
5981         u64 parent_ino = btrfs_ino(parent_inode);
5982
5983         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5984                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5985         } else {
5986                 key.objectid = ino;
5987                 key.type = BTRFS_INODE_ITEM_KEY;
5988                 key.offset = 0;
5989         }
5990
5991         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5992                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5993                                          key.objectid, root->root_key.objectid,
5994                                          parent_ino, index, name, name_len);
5995         } else if (add_backref) {
5996                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5997                                              parent_ino, index);
5998         }
5999
6000         /* Nothing to clean up yet */
6001         if (ret)
6002                 return ret;
6003
6004         ret = btrfs_insert_dir_item(trans, root, name, name_len,
6005                                     parent_inode, &key,
6006                                     btrfs_inode_type(inode), index);
6007         if (ret == -EEXIST || ret == -EOVERFLOW)
6008                 goto fail_dir_item;
6009         else if (ret) {
6010                 btrfs_abort_transaction(trans, root, ret);
6011                 return ret;
6012         }
6013
6014         btrfs_i_size_write(parent_inode, parent_inode->i_size +
6015                            name_len * 2);
6016         inode_inc_iversion(parent_inode);
6017         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
6018         ret = btrfs_update_inode(trans, root, parent_inode);
6019         if (ret)
6020                 btrfs_abort_transaction(trans, root, ret);
6021         return ret;
6022
6023 fail_dir_item:
6024         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
6025                 u64 local_index;
6026                 int err;
6027                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
6028                                  key.objectid, root->root_key.objectid,
6029                                  parent_ino, &local_index, name, name_len);
6030
6031         } else if (add_backref) {
6032                 u64 local_index;
6033                 int err;
6034
6035                 err = btrfs_del_inode_ref(trans, root, name, name_len,
6036                                           ino, parent_ino, &local_index);
6037         }
6038         return ret;
6039 }
6040
6041 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
6042                             struct inode *dir, struct dentry *dentry,
6043                             struct inode *inode, int backref, u64 index)
6044 {
6045         int err = btrfs_add_link(trans, dir, inode,
6046                                  dentry->d_name.name, dentry->d_name.len,
6047                                  backref, index);
6048         if (err > 0)
6049                 err = -EEXIST;
6050         return err;
6051 }
6052
6053 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
6054                         umode_t mode, dev_t rdev)
6055 {
6056         struct btrfs_trans_handle *trans;
6057         struct btrfs_root *root = BTRFS_I(dir)->root;
6058         struct inode *inode = NULL;
6059         int err;
6060         int drop_inode = 0;
6061         u64 objectid;
6062         u64 index = 0;
6063
6064         if (!new_valid_dev(rdev))
6065                 return -EINVAL;
6066
6067         /*
6068          * 2 for inode item and ref
6069          * 2 for dir items
6070          * 1 for xattr if selinux is on
6071          */
6072         trans = btrfs_start_transaction(root, 5);
6073         if (IS_ERR(trans))
6074                 return PTR_ERR(trans);
6075
6076         err = btrfs_find_free_ino(root, &objectid);
6077         if (err)
6078                 goto out_unlock;
6079
6080         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6081                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6082                                 mode, &index);
6083         if (IS_ERR(inode)) {
6084                 err = PTR_ERR(inode);
6085                 goto out_unlock;
6086         }
6087
6088         /*
6089         * If the active LSM wants to access the inode during
6090         * d_instantiate it needs these. Smack checks to see
6091         * if the filesystem supports xattrs by looking at the
6092         * ops vector.
6093         */
6094         inode->i_op = &btrfs_special_inode_operations;
6095         init_special_inode(inode, inode->i_mode, rdev);
6096
6097         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6098         if (err)
6099                 goto out_unlock_inode;
6100
6101         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6102         if (err) {
6103                 goto out_unlock_inode;
6104         } else {
6105                 btrfs_update_inode(trans, root, inode);
6106                 unlock_new_inode(inode);
6107                 d_instantiate(dentry, inode);
6108         }
6109
6110 out_unlock:
6111         btrfs_end_transaction(trans, root);
6112         btrfs_balance_delayed_items(root);
6113         btrfs_btree_balance_dirty(root);
6114         if (drop_inode) {
6115                 inode_dec_link_count(inode);
6116                 iput(inode);
6117         }
6118         return err;
6119
6120 out_unlock_inode:
6121         drop_inode = 1;
6122         unlock_new_inode(inode);
6123         goto out_unlock;
6124
6125 }
6126
6127 static int btrfs_create(struct inode *dir, struct dentry *dentry,
6128                         umode_t mode, bool excl)
6129 {
6130         struct btrfs_trans_handle *trans;
6131         struct btrfs_root *root = BTRFS_I(dir)->root;
6132         struct inode *inode = NULL;
6133         int drop_inode_on_err = 0;
6134         int err;
6135         u64 objectid;
6136         u64 index = 0;
6137
6138         /*
6139          * 2 for inode item and ref
6140          * 2 for dir items
6141          * 1 for xattr if selinux is on
6142          */
6143         trans = btrfs_start_transaction(root, 5);
6144         if (IS_ERR(trans))
6145                 return PTR_ERR(trans);
6146
6147         err = btrfs_find_free_ino(root, &objectid);
6148         if (err)
6149                 goto out_unlock;
6150
6151         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6152                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6153                                 mode, &index);
6154         if (IS_ERR(inode)) {
6155                 err = PTR_ERR(inode);
6156                 goto out_unlock;
6157         }
6158         drop_inode_on_err = 1;
6159         /*
6160         * If the active LSM wants to access the inode during
6161         * d_instantiate it needs these. Smack checks to see
6162         * if the filesystem supports xattrs by looking at the
6163         * ops vector.
6164         */
6165         inode->i_fop = &btrfs_file_operations;
6166         inode->i_op = &btrfs_file_inode_operations;
6167         inode->i_mapping->a_ops = &btrfs_aops;
6168
6169         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6170         if (err)
6171                 goto out_unlock_inode;
6172
6173         err = btrfs_update_inode(trans, root, inode);
6174         if (err)
6175                 goto out_unlock_inode;
6176
6177         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6178         if (err)
6179                 goto out_unlock_inode;
6180
6181         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6182         unlock_new_inode(inode);
6183         d_instantiate(dentry, inode);
6184
6185 out_unlock:
6186         btrfs_end_transaction(trans, root);
6187         if (err && drop_inode_on_err) {
6188                 inode_dec_link_count(inode);
6189                 iput(inode);
6190         }
6191         btrfs_balance_delayed_items(root);
6192         btrfs_btree_balance_dirty(root);
6193         return err;
6194
6195 out_unlock_inode:
6196         unlock_new_inode(inode);
6197         goto out_unlock;
6198
6199 }
6200
6201 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
6202                       struct dentry *dentry)
6203 {
6204         struct btrfs_trans_handle *trans;
6205         struct btrfs_root *root = BTRFS_I(dir)->root;
6206         struct inode *inode = old_dentry->d_inode;
6207         u64 index;
6208         int err;
6209         int drop_inode = 0;
6210
6211         /* do not allow sys_link's with other subvols of the same device */
6212         if (root->objectid != BTRFS_I(inode)->root->objectid)
6213                 return -EXDEV;
6214
6215         if (inode->i_nlink >= BTRFS_LINK_MAX)
6216                 return -EMLINK;
6217
6218         err = btrfs_set_inode_index(dir, &index);
6219         if (err)
6220                 goto fail;
6221
6222         /*
6223          * 2 items for inode and inode ref
6224          * 2 items for dir items
6225          * 1 item for parent inode
6226          */
6227         trans = btrfs_start_transaction(root, 5);
6228         if (IS_ERR(trans)) {
6229                 err = PTR_ERR(trans);
6230                 goto fail;
6231         }
6232
6233         /* There are several dir indexes for this inode, clear the cache. */
6234         BTRFS_I(inode)->dir_index = 0ULL;
6235         inc_nlink(inode);
6236         inode_inc_iversion(inode);
6237         inode->i_ctime = CURRENT_TIME;
6238         ihold(inode);
6239         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
6240
6241         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
6242
6243         if (err) {
6244                 drop_inode = 1;
6245         } else {
6246                 struct dentry *parent = dentry->d_parent;
6247                 err = btrfs_update_inode(trans, root, inode);
6248                 if (err)
6249                         goto fail;
6250                 if (inode->i_nlink == 1) {
6251                         /*
6252                          * If new hard link count is 1, it's a file created
6253                          * with open(2) O_TMPFILE flag.
6254                          */
6255                         err = btrfs_orphan_del(trans, inode);
6256                         if (err)
6257                                 goto fail;
6258                 }
6259                 d_instantiate(dentry, inode);
6260                 btrfs_log_new_name(trans, inode, NULL, parent);
6261         }
6262
6263         btrfs_end_transaction(trans, root);
6264         btrfs_balance_delayed_items(root);
6265 fail:
6266         if (drop_inode) {
6267                 inode_dec_link_count(inode);
6268                 iput(inode);
6269         }
6270         btrfs_btree_balance_dirty(root);
6271         return err;
6272 }
6273
6274 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
6275 {
6276         struct inode *inode = NULL;
6277         struct btrfs_trans_handle *trans;
6278         struct btrfs_root *root = BTRFS_I(dir)->root;
6279         int err = 0;
6280         int drop_on_err = 0;
6281         u64 objectid = 0;
6282         u64 index = 0;
6283
6284         /*
6285          * 2 items for inode and ref
6286          * 2 items for dir items
6287          * 1 for xattr if selinux is on
6288          */
6289         trans = btrfs_start_transaction(root, 5);
6290         if (IS_ERR(trans))
6291                 return PTR_ERR(trans);
6292
6293         err = btrfs_find_free_ino(root, &objectid);
6294         if (err)
6295                 goto out_fail;
6296
6297         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6298                                 dentry->d_name.len, btrfs_ino(dir), objectid,
6299                                 S_IFDIR | mode, &index);
6300         if (IS_ERR(inode)) {
6301                 err = PTR_ERR(inode);
6302                 goto out_fail;
6303         }
6304
6305         drop_on_err = 1;
6306         /* these must be set before we unlock the inode */
6307         inode->i_op = &btrfs_dir_inode_operations;
6308         inode->i_fop = &btrfs_dir_file_operations;
6309
6310         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
6311         if (err)
6312                 goto out_fail_inode;
6313
6314         btrfs_i_size_write(inode, 0);
6315         err = btrfs_update_inode(trans, root, inode);
6316         if (err)
6317                 goto out_fail_inode;
6318
6319         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
6320                              dentry->d_name.len, 0, index);
6321         if (err)
6322                 goto out_fail_inode;
6323
6324         d_instantiate(dentry, inode);
6325         /*
6326          * mkdir is special.  We're unlocking after we call d_instantiate
6327          * to avoid a race with nfsd calling d_instantiate.
6328          */
6329         unlock_new_inode(inode);
6330         drop_on_err = 0;
6331
6332 out_fail:
6333         btrfs_end_transaction(trans, root);
6334         if (drop_on_err) {
6335                 inode_dec_link_count(inode);
6336                 iput(inode);
6337         }
6338         btrfs_balance_delayed_items(root);
6339         btrfs_btree_balance_dirty(root);
6340         return err;
6341
6342 out_fail_inode:
6343         unlock_new_inode(inode);
6344         goto out_fail;
6345 }
6346
6347 /* Find next extent map of a given extent map, caller needs to ensure locks */
6348 static struct extent_map *next_extent_map(struct extent_map *em)
6349 {
6350         struct rb_node *next;
6351
6352         next = rb_next(&em->rb_node);
6353         if (!next)
6354                 return NULL;
6355         return container_of(next, struct extent_map, rb_node);
6356 }
6357
6358 static struct extent_map *prev_extent_map(struct extent_map *em)
6359 {
6360         struct rb_node *prev;
6361
6362         prev = rb_prev(&em->rb_node);
6363         if (!prev)
6364                 return NULL;
6365         return container_of(prev, struct extent_map, rb_node);
6366 }
6367
6368 /* helper for btfs_get_extent.  Given an existing extent in the tree,
6369  * the existing extent is the nearest extent to map_start,
6370  * and an extent that you want to insert, deal with overlap and insert
6371  * the best fitted new extent into the tree.
6372  */
6373 static int merge_extent_mapping(struct extent_map_tree *em_tree,
6374                                 struct extent_map *existing,
6375                                 struct extent_map *em,
6376                                 u64 map_start)
6377 {
6378         struct extent_map *prev;
6379         struct extent_map *next;
6380         u64 start;
6381         u64 end;
6382         u64 start_diff;
6383
6384         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6385
6386         if (existing->start > map_start) {
6387                 next = existing;
6388                 prev = prev_extent_map(next);
6389         } else {
6390                 prev = existing;
6391                 next = next_extent_map(prev);
6392         }
6393
6394         start = prev ? extent_map_end(prev) : em->start;
6395         start = max_t(u64, start, em->start);
6396         end = next ? next->start : extent_map_end(em);
6397         end = min_t(u64, end, extent_map_end(em));
6398         start_diff = start - em->start;
6399         em->start = start;
6400         em->len = end - start;
6401         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6402             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6403                 em->block_start += start_diff;
6404                 em->block_len -= start_diff;
6405         }
6406         return add_extent_mapping(em_tree, em, 0);
6407 }
6408
6409 static noinline int uncompress_inline(struct btrfs_path *path,
6410                                       struct inode *inode, struct page *page,
6411                                       size_t pg_offset, u64 extent_offset,
6412                                       struct btrfs_file_extent_item *item)
6413 {
6414         int ret;
6415         struct extent_buffer *leaf = path->nodes[0];
6416         char *tmp;
6417         size_t max_size;
6418         unsigned long inline_size;
6419         unsigned long ptr;
6420         int compress_type;
6421
6422         WARN_ON(pg_offset != 0);
6423         compress_type = btrfs_file_extent_compression(leaf, item);
6424         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6425         inline_size = btrfs_file_extent_inline_item_len(leaf,
6426                                         btrfs_item_nr(path->slots[0]));
6427         tmp = kmalloc(inline_size, GFP_NOFS);
6428         if (!tmp)
6429                 return -ENOMEM;
6430         ptr = btrfs_file_extent_inline_start(item);
6431
6432         read_extent_buffer(leaf, tmp, ptr, inline_size);
6433
6434         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6435         ret = btrfs_decompress(compress_type, tmp, page,
6436                                extent_offset, inline_size, max_size);
6437         kfree(tmp);
6438         return ret;
6439 }
6440
6441 /*
6442  * a bit scary, this does extent mapping from logical file offset to the disk.
6443  * the ugly parts come from merging extents from the disk with the in-ram
6444  * representation.  This gets more complex because of the data=ordered code,
6445  * where the in-ram extents might be locked pending data=ordered completion.
6446  *
6447  * This also copies inline extents directly into the page.
6448  */
6449
6450 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6451                                     size_t pg_offset, u64 start, u64 len,
6452                                     int create)
6453 {
6454         int ret;
6455         int err = 0;
6456         u64 extent_start = 0;
6457         u64 extent_end = 0;
6458         u64 objectid = btrfs_ino(inode);
6459         u32 found_type;
6460         struct btrfs_path *path = NULL;
6461         struct btrfs_root *root = BTRFS_I(inode)->root;
6462         struct btrfs_file_extent_item *item;
6463         struct extent_buffer *leaf;
6464         struct btrfs_key found_key;
6465         struct extent_map *em = NULL;
6466         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6467         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6468         struct btrfs_trans_handle *trans = NULL;
6469         const bool new_inline = !page || create;
6470
6471 again:
6472         read_lock(&em_tree->lock);
6473         em = lookup_extent_mapping(em_tree, start, len);
6474         if (em)
6475                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6476         read_unlock(&em_tree->lock);
6477
6478         if (em) {
6479                 if (em->start > start || em->start + em->len <= start)
6480                         free_extent_map(em);
6481                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6482                         free_extent_map(em);
6483                 else
6484                         goto out;
6485         }
6486         em = alloc_extent_map();
6487         if (!em) {
6488                 err = -ENOMEM;
6489                 goto out;
6490         }
6491         em->bdev = root->fs_info->fs_devices->latest_bdev;
6492         em->start = EXTENT_MAP_HOLE;
6493         em->orig_start = EXTENT_MAP_HOLE;
6494         em->len = (u64)-1;
6495         em->block_len = (u64)-1;
6496
6497         if (!path) {
6498                 path = btrfs_alloc_path();
6499                 if (!path) {
6500                         err = -ENOMEM;
6501                         goto out;
6502                 }
6503                 /*
6504                  * Chances are we'll be called again, so go ahead and do
6505                  * readahead
6506                  */
6507                 path->reada = 1;
6508         }
6509
6510         ret = btrfs_lookup_file_extent(trans, root, path,
6511                                        objectid, start, trans != NULL);
6512         if (ret < 0) {
6513                 err = ret;
6514                 goto out;
6515         }
6516
6517         if (ret != 0) {
6518                 if (path->slots[0] == 0)
6519                         goto not_found;
6520                 path->slots[0]--;
6521         }
6522
6523         leaf = path->nodes[0];
6524         item = btrfs_item_ptr(leaf, path->slots[0],
6525                               struct btrfs_file_extent_item);
6526         /* are we inside the extent that was found? */
6527         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6528         found_type = found_key.type;
6529         if (found_key.objectid != objectid ||
6530             found_type != BTRFS_EXTENT_DATA_KEY) {
6531                 /*
6532                  * If we backup past the first extent we want to move forward
6533                  * and see if there is an extent in front of us, otherwise we'll
6534                  * say there is a hole for our whole search range which can
6535                  * cause problems.
6536                  */
6537                 extent_end = start;
6538                 goto next;
6539         }
6540
6541         found_type = btrfs_file_extent_type(leaf, item);
6542         extent_start = found_key.offset;
6543         if (found_type == BTRFS_FILE_EXTENT_REG ||
6544             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6545                 extent_end = extent_start +
6546                        btrfs_file_extent_num_bytes(leaf, item);
6547         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6548                 size_t size;
6549                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6550                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6551         }
6552 next:
6553         if (start >= extent_end) {
6554                 path->slots[0]++;
6555                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6556                         ret = btrfs_next_leaf(root, path);
6557                         if (ret < 0) {
6558                                 err = ret;
6559                                 goto out;
6560                         }
6561                         if (ret > 0)
6562                                 goto not_found;
6563                         leaf = path->nodes[0];
6564                 }
6565                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6566                 if (found_key.objectid != objectid ||
6567                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6568                         goto not_found;
6569                 if (start + len <= found_key.offset)
6570                         goto not_found;
6571                 if (start > found_key.offset)
6572                         goto next;
6573                 em->start = start;
6574                 em->orig_start = start;
6575                 em->len = found_key.offset - start;
6576                 goto not_found_em;
6577         }
6578
6579         btrfs_extent_item_to_extent_map(inode, path, item, new_inline, em);
6580
6581         if (found_type == BTRFS_FILE_EXTENT_REG ||
6582             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6583                 goto insert;
6584         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6585                 unsigned long ptr;
6586                 char *map;
6587                 size_t size;
6588                 size_t extent_offset;
6589                 size_t copy_size;
6590
6591                 if (new_inline)
6592                         goto out;
6593
6594                 size = btrfs_file_extent_inline_len(leaf, path->slots[0], item);
6595                 extent_offset = page_offset(page) + pg_offset - extent_start;
6596                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6597                                 size - extent_offset);
6598                 em->start = extent_start + extent_offset;
6599                 em->len = ALIGN(copy_size, root->sectorsize);
6600                 em->orig_block_len = em->len;
6601                 em->orig_start = em->start;
6602                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6603                 if (create == 0 && !PageUptodate(page)) {
6604                         if (btrfs_file_extent_compression(leaf, item) !=
6605                             BTRFS_COMPRESS_NONE) {
6606                                 ret = uncompress_inline(path, inode, page,
6607                                                         pg_offset,
6608                                                         extent_offset, item);
6609                                 if (ret) {
6610                                         err = ret;
6611                                         goto out;
6612                                 }
6613                         } else {
6614                                 map = kmap(page);
6615                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6616                                                    copy_size);
6617                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6618                                         memset(map + pg_offset + copy_size, 0,
6619                                                PAGE_CACHE_SIZE - pg_offset -
6620                                                copy_size);
6621                                 }
6622                                 kunmap(page);
6623                         }
6624                         flush_dcache_page(page);
6625                 } else if (create && PageUptodate(page)) {
6626                         BUG();
6627                         if (!trans) {
6628                                 kunmap(page);
6629                                 free_extent_map(em);
6630                                 em = NULL;
6631
6632                                 btrfs_release_path(path);
6633                                 trans = btrfs_join_transaction(root);
6634
6635                                 if (IS_ERR(trans))
6636                                         return ERR_CAST(trans);
6637                                 goto again;
6638                         }
6639                         map = kmap(page);
6640                         write_extent_buffer(leaf, map + pg_offset, ptr,
6641                                             copy_size);
6642                         kunmap(page);
6643                         btrfs_mark_buffer_dirty(leaf);
6644                 }
6645                 set_extent_uptodate(io_tree, em->start,
6646                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6647                 goto insert;
6648         }
6649 not_found:
6650         em->start = start;
6651         em->orig_start = start;
6652         em->len = len;
6653 not_found_em:
6654         em->block_start = EXTENT_MAP_HOLE;
6655         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6656 insert:
6657         btrfs_release_path(path);
6658         if (em->start > start || extent_map_end(em) <= start) {
6659                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6660                         em->start, em->len, start, len);
6661                 err = -EIO;
6662                 goto out;
6663         }
6664
6665         err = 0;
6666         write_lock(&em_tree->lock);
6667         ret = add_extent_mapping(em_tree, em, 0);
6668         /* it is possible that someone inserted the extent into the tree
6669          * while we had the lock dropped.  It is also possible that
6670          * an overlapping map exists in the tree
6671          */
6672         if (ret == -EEXIST) {
6673                 struct extent_map *existing;
6674
6675                 ret = 0;
6676
6677                 existing = search_extent_mapping(em_tree, start, len);
6678                 /*
6679                  * existing will always be non-NULL, since there must be
6680                  * extent causing the -EEXIST.
6681                  */
6682                 if (start >= extent_map_end(existing) ||
6683                     start <= existing->start) {
6684                         /*
6685                          * The existing extent map is the one nearest to
6686                          * the [start, start + len) range which overlaps
6687                          */
6688                         err = merge_extent_mapping(em_tree, existing,
6689                                                    em, start);
6690                         free_extent_map(existing);
6691                         if (err) {
6692                                 free_extent_map(em);
6693                                 em = NULL;
6694                         }
6695                 } else {
6696                         free_extent_map(em);
6697                         em = existing;
6698                         err = 0;
6699                 }
6700         }
6701         write_unlock(&em_tree->lock);
6702 out:
6703
6704         trace_btrfs_get_extent(root, em);
6705
6706         if (path)
6707                 btrfs_free_path(path);
6708         if (trans) {
6709                 ret = btrfs_end_transaction(trans, root);
6710                 if (!err)
6711                         err = ret;
6712         }
6713         if (err) {
6714                 free_extent_map(em);
6715                 return ERR_PTR(err);
6716         }
6717         BUG_ON(!em); /* Error is always set */
6718         return em;
6719 }
6720
6721 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6722                                            size_t pg_offset, u64 start, u64 len,
6723                                            int create)
6724 {
6725         struct extent_map *em;
6726         struct extent_map *hole_em = NULL;
6727         u64 range_start = start;
6728         u64 end;
6729         u64 found;
6730         u64 found_end;
6731         int err = 0;
6732
6733         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6734         if (IS_ERR(em))
6735                 return em;
6736         if (em) {
6737                 /*
6738                  * if our em maps to
6739                  * -  a hole or
6740                  * -  a pre-alloc extent,
6741                  * there might actually be delalloc bytes behind it.
6742                  */
6743                 if (em->block_start != EXTENT_MAP_HOLE &&
6744                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6745                         return em;
6746                 else
6747                         hole_em = em;
6748         }
6749
6750         /* check to see if we've wrapped (len == -1 or similar) */
6751         end = start + len;
6752         if (end < start)
6753                 end = (u64)-1;
6754         else
6755                 end -= 1;
6756
6757         em = NULL;
6758
6759         /* ok, we didn't find anything, lets look for delalloc */
6760         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6761                                  end, len, EXTENT_DELALLOC, 1);
6762         found_end = range_start + found;
6763         if (found_end < range_start)
6764                 found_end = (u64)-1;
6765
6766         /*
6767          * we didn't find anything useful, return
6768          * the original results from get_extent()
6769          */
6770         if (range_start > end || found_end <= start) {
6771                 em = hole_em;
6772                 hole_em = NULL;
6773                 goto out;
6774         }
6775
6776         /* adjust the range_start to make sure it doesn't
6777          * go backwards from the start they passed in
6778          */
6779         range_start = max(start, range_start);
6780         found = found_end - range_start;
6781
6782         if (found > 0) {
6783                 u64 hole_start = start;
6784                 u64 hole_len = len;
6785
6786                 em = alloc_extent_map();
6787                 if (!em) {
6788                         err = -ENOMEM;
6789                         goto out;
6790                 }
6791                 /*
6792                  * when btrfs_get_extent can't find anything it
6793                  * returns one huge hole
6794                  *
6795                  * make sure what it found really fits our range, and
6796                  * adjust to make sure it is based on the start from
6797                  * the caller
6798                  */
6799                 if (hole_em) {
6800                         u64 calc_end = extent_map_end(hole_em);
6801
6802                         if (calc_end <= start || (hole_em->start > end)) {
6803                                 free_extent_map(hole_em);
6804                                 hole_em = NULL;
6805                         } else {
6806                                 hole_start = max(hole_em->start, start);
6807                                 hole_len = calc_end - hole_start;
6808                         }
6809                 }
6810                 em->bdev = NULL;
6811                 if (hole_em && range_start > hole_start) {
6812                         /* our hole starts before our delalloc, so we
6813                          * have to return just the parts of the hole
6814                          * that go until  the delalloc starts
6815                          */
6816                         em->len = min(hole_len,
6817                                       range_start - hole_start);
6818                         em->start = hole_start;
6819                         em->orig_start = hole_start;
6820                         /*
6821                          * don't adjust block start at all,
6822                          * it is fixed at EXTENT_MAP_HOLE
6823                          */
6824                         em->block_start = hole_em->block_start;
6825                         em->block_len = hole_len;
6826                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6827                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6828                 } else {
6829                         em->start = range_start;
6830                         em->len = found;
6831                         em->orig_start = range_start;
6832                         em->block_start = EXTENT_MAP_DELALLOC;
6833                         em->block_len = found;
6834                 }
6835         } else if (hole_em) {
6836                 return hole_em;
6837         }
6838 out:
6839
6840         free_extent_map(hole_em);
6841         if (err) {
6842                 free_extent_map(em);
6843                 return ERR_PTR(err);
6844         }
6845         return em;
6846 }
6847
6848 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6849                                                   u64 start, u64 len)
6850 {
6851         struct btrfs_root *root = BTRFS_I(inode)->root;
6852         struct extent_map *em;
6853         struct btrfs_key ins;
6854         u64 alloc_hint;
6855         int ret;
6856
6857         alloc_hint = get_extent_allocation_hint(inode, start, len);
6858         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6859                                    alloc_hint, &ins, 1, 1);
6860         if (ret)
6861                 return ERR_PTR(ret);
6862
6863         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6864                               ins.offset, ins.offset, ins.offset, 0);
6865         if (IS_ERR(em)) {
6866                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6867                 return em;
6868         }
6869
6870         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6871                                            ins.offset, ins.offset, 0);
6872         if (ret) {
6873                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset, 1);
6874                 free_extent_map(em);
6875                 return ERR_PTR(ret);
6876         }
6877
6878         return em;
6879 }
6880
6881 /*
6882  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6883  * block must be cow'd
6884  */
6885 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6886                               u64 *orig_start, u64 *orig_block_len,
6887                               u64 *ram_bytes)
6888 {
6889         struct btrfs_trans_handle *trans;
6890         struct btrfs_path *path;
6891         int ret;
6892         struct extent_buffer *leaf;
6893         struct btrfs_root *root = BTRFS_I(inode)->root;
6894         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6895         struct btrfs_file_extent_item *fi;
6896         struct btrfs_key key;
6897         u64 disk_bytenr;
6898         u64 backref_offset;
6899         u64 extent_end;
6900         u64 num_bytes;
6901         int slot;
6902         int found_type;
6903         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6904
6905         path = btrfs_alloc_path();
6906         if (!path)
6907                 return -ENOMEM;
6908
6909         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6910                                        offset, 0);
6911         if (ret < 0)
6912                 goto out;
6913
6914         slot = path->slots[0];
6915         if (ret == 1) {
6916                 if (slot == 0) {
6917                         /* can't find the item, must cow */
6918                         ret = 0;
6919                         goto out;
6920                 }
6921                 slot--;
6922         }
6923         ret = 0;
6924         leaf = path->nodes[0];
6925         btrfs_item_key_to_cpu(leaf, &key, slot);
6926         if (key.objectid != btrfs_ino(inode) ||
6927             key.type != BTRFS_EXTENT_DATA_KEY) {
6928                 /* not our file or wrong item type, must cow */
6929                 goto out;
6930         }
6931
6932         if (key.offset > offset) {
6933                 /* Wrong offset, must cow */
6934                 goto out;
6935         }
6936
6937         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6938         found_type = btrfs_file_extent_type(leaf, fi);
6939         if (found_type != BTRFS_FILE_EXTENT_REG &&
6940             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6941                 /* not a regular extent, must cow */
6942                 goto out;
6943         }
6944
6945         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6946                 goto out;
6947
6948         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6949         if (extent_end <= offset)
6950                 goto out;
6951
6952         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6953         if (disk_bytenr == 0)
6954                 goto out;
6955
6956         if (btrfs_file_extent_compression(leaf, fi) ||
6957             btrfs_file_extent_encryption(leaf, fi) ||
6958             btrfs_file_extent_other_encoding(leaf, fi))
6959                 goto out;
6960
6961         backref_offset = btrfs_file_extent_offset(leaf, fi);
6962
6963         if (orig_start) {
6964                 *orig_start = key.offset - backref_offset;
6965                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6966                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6967         }
6968
6969         if (btrfs_extent_readonly(root, disk_bytenr))
6970                 goto out;
6971
6972         num_bytes = min(offset + *len, extent_end) - offset;
6973         if (!nocow && found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6974                 u64 range_end;
6975
6976                 range_end = round_up(offset + num_bytes, root->sectorsize) - 1;
6977                 ret = test_range_bit(io_tree, offset, range_end,
6978                                      EXTENT_DELALLOC, 0, NULL);
6979                 if (ret) {
6980                         ret = -EAGAIN;
6981                         goto out;
6982                 }
6983         }
6984
6985         btrfs_release_path(path);
6986
6987         /*
6988          * look for other files referencing this extent, if we
6989          * find any we must cow
6990          */
6991         trans = btrfs_join_transaction(root);
6992         if (IS_ERR(trans)) {
6993                 ret = 0;
6994                 goto out;
6995         }
6996
6997         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6998                                     key.offset - backref_offset, disk_bytenr);
6999         btrfs_end_transaction(trans, root);
7000         if (ret) {
7001                 ret = 0;
7002                 goto out;
7003         }
7004
7005         /*
7006          * adjust disk_bytenr and num_bytes to cover just the bytes
7007          * in this extent we are about to write.  If there
7008          * are any csums in that range we have to cow in order
7009          * to keep the csums correct
7010          */
7011         disk_bytenr += backref_offset;
7012         disk_bytenr += offset - key.offset;
7013         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
7014                                 goto out;
7015         /*
7016          * all of the above have passed, it is safe to overwrite this extent
7017          * without cow
7018          */
7019         *len = num_bytes;
7020         ret = 1;
7021 out:
7022         btrfs_free_path(path);
7023         return ret;
7024 }
7025
7026 bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end)
7027 {
7028         struct radix_tree_root *root = &inode->i_mapping->page_tree;
7029         int found = false;
7030         void **pagep = NULL;
7031         struct page *page = NULL;
7032         int start_idx;
7033         int end_idx;
7034
7035         start_idx = start >> PAGE_CACHE_SHIFT;
7036
7037         /*
7038          * end is the last byte in the last page.  end == start is legal
7039          */
7040         end_idx = end >> PAGE_CACHE_SHIFT;
7041
7042         rcu_read_lock();
7043
7044         /* Most of the code in this while loop is lifted from
7045          * find_get_page.  It's been modified to begin searching from a
7046          * page and return just the first page found in that range.  If the
7047          * found idx is less than or equal to the end idx then we know that
7048          * a page exists.  If no pages are found or if those pages are
7049          * outside of the range then we're fine (yay!) */
7050         while (page == NULL &&
7051                radix_tree_gang_lookup_slot(root, &pagep, NULL, start_idx, 1)) {
7052                 page = radix_tree_deref_slot(pagep);
7053                 if (unlikely(!page))
7054                         break;
7055
7056                 if (radix_tree_exception(page)) {
7057                         if (radix_tree_deref_retry(page)) {
7058                                 page = NULL;
7059                                 continue;
7060                         }
7061                         /*
7062                          * Otherwise, shmem/tmpfs must be storing a swap entry
7063                          * here as an exceptional entry: so return it without
7064                          * attempting to raise page count.
7065                          */
7066                         page = NULL;
7067                         break; /* TODO: Is this relevant for this use case? */
7068                 }
7069
7070                 if (!page_cache_get_speculative(page)) {
7071                         page = NULL;
7072                         continue;
7073                 }
7074
7075                 /*
7076                  * Has the page moved?
7077                  * This is part of the lockless pagecache protocol. See
7078                  * include/linux/pagemap.h for details.
7079                  */
7080                 if (unlikely(page != *pagep)) {
7081                         page_cache_release(page);
7082                         page = NULL;
7083                 }
7084         }
7085
7086         if (page) {
7087                 if (page->index <= end_idx)
7088                         found = true;
7089                 page_cache_release(page);
7090         }
7091
7092         rcu_read_unlock();
7093         return found;
7094 }
7095
7096 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
7097                               struct extent_state **cached_state, int writing)
7098 {
7099         struct btrfs_ordered_extent *ordered;
7100         int ret = 0;
7101
7102         while (1) {
7103                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7104                                  0, cached_state);
7105                 /*
7106                  * We're concerned with the entire range that we're going to be
7107                  * doing DIO to, so we need to make sure theres no ordered
7108                  * extents in this range.
7109                  */
7110                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
7111                                                      lockend - lockstart + 1);
7112
7113                 /*
7114                  * We need to make sure there are no buffered pages in this
7115                  * range either, we could have raced between the invalidate in
7116                  * generic_file_direct_write and locking the extent.  The
7117                  * invalidate needs to happen so that reads after a write do not
7118                  * get stale data.
7119                  */
7120                 if (!ordered &&
7121                     (!writing ||
7122                      !btrfs_page_exists_in_range(inode, lockstart, lockend)))
7123                         break;
7124
7125                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7126                                      cached_state, GFP_NOFS);
7127
7128                 if (ordered) {
7129                         btrfs_start_ordered_extent(inode, ordered, 1);
7130                         btrfs_put_ordered_extent(ordered);
7131                 } else {
7132                         /* Screw you mmap */
7133                         ret = btrfs_fdatawrite_range(inode, lockstart, lockend);
7134                         if (ret)
7135                                 break;
7136                         ret = filemap_fdatawait_range(inode->i_mapping,
7137                                                       lockstart,
7138                                                       lockend);
7139                         if (ret)
7140                                 break;
7141
7142                         /*
7143                          * If we found a page that couldn't be invalidated just
7144                          * fall back to buffered.
7145                          */
7146                         ret = invalidate_inode_pages2_range(inode->i_mapping,
7147                                         lockstart >> PAGE_CACHE_SHIFT,
7148                                         lockend >> PAGE_CACHE_SHIFT);
7149                         if (ret)
7150                                 break;
7151                 }
7152
7153                 cond_resched();
7154         }
7155
7156         return ret;
7157 }
7158
7159 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
7160                                            u64 len, u64 orig_start,
7161                                            u64 block_start, u64 block_len,
7162                                            u64 orig_block_len, u64 ram_bytes,
7163                                            int type)
7164 {
7165         struct extent_map_tree *em_tree;
7166         struct extent_map *em;
7167         struct btrfs_root *root = BTRFS_I(inode)->root;
7168         int ret;
7169
7170         em_tree = &BTRFS_I(inode)->extent_tree;
7171         em = alloc_extent_map();
7172         if (!em)
7173                 return ERR_PTR(-ENOMEM);
7174
7175         em->start = start;
7176         em->orig_start = orig_start;
7177         em->mod_start = start;
7178         em->mod_len = len;
7179         em->len = len;
7180         em->block_len = block_len;
7181         em->block_start = block_start;
7182         em->bdev = root->fs_info->fs_devices->latest_bdev;
7183         em->orig_block_len = orig_block_len;
7184         em->ram_bytes = ram_bytes;
7185         em->generation = -1;
7186         set_bit(EXTENT_FLAG_PINNED, &em->flags);
7187         if (type == BTRFS_ORDERED_PREALLOC)
7188                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
7189
7190         do {
7191                 btrfs_drop_extent_cache(inode, em->start,
7192                                 em->start + em->len - 1, 0);
7193                 write_lock(&em_tree->lock);
7194                 ret = add_extent_mapping(em_tree, em, 1);
7195                 write_unlock(&em_tree->lock);
7196         } while (ret == -EEXIST);
7197
7198         if (ret) {
7199                 free_extent_map(em);
7200                 return ERR_PTR(ret);
7201         }
7202
7203         return em;
7204 }
7205
7206
7207 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
7208                                    struct buffer_head *bh_result, int create)
7209 {
7210         struct extent_map *em;
7211         struct btrfs_root *root = BTRFS_I(inode)->root;
7212         struct extent_state *cached_state = NULL;
7213         u64 start = iblock << inode->i_blkbits;
7214         u64 lockstart, lockend;
7215         u64 len = bh_result->b_size;
7216         u64 orig_len = len;
7217         int unlock_bits = EXTENT_LOCKED;
7218         int ret = 0;
7219
7220         if (create)
7221                 unlock_bits |= EXTENT_DIRTY;
7222         else
7223                 len = min_t(u64, len, root->sectorsize);
7224
7225         lockstart = start;
7226         lockend = start + len - 1;
7227
7228         /*
7229          * If this errors out it's because we couldn't invalidate pagecache for
7230          * this range and we need to fallback to buffered.
7231          */
7232         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
7233                 return -ENOTBLK;
7234
7235         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
7236         if (IS_ERR(em)) {
7237                 ret = PTR_ERR(em);
7238                 goto unlock_err;
7239         }
7240
7241         /*
7242          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
7243          * io.  INLINE is special, and we could probably kludge it in here, but
7244          * it's still buffered so for safety lets just fall back to the generic
7245          * buffered path.
7246          *
7247          * For COMPRESSED we _have_ to read the entire extent in so we can
7248          * decompress it, so there will be buffering required no matter what we
7249          * do, so go ahead and fallback to buffered.
7250          *
7251          * We return -ENOTBLK because thats what makes DIO go ahead and go back
7252          * to buffered IO.  Don't blame me, this is the price we pay for using
7253          * the generic code.
7254          */
7255         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
7256             em->block_start == EXTENT_MAP_INLINE) {
7257                 free_extent_map(em);
7258                 ret = -ENOTBLK;
7259                 goto unlock_err;
7260         }
7261
7262         /* Just a good old fashioned hole, return */
7263         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
7264                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7265                 free_extent_map(em);
7266                 goto unlock_err;
7267         }
7268
7269         /*
7270          * We don't allocate a new extent in the following cases
7271          *
7272          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
7273          * existing extent.
7274          * 2) The extent is marked as PREALLOC.  We're good to go here and can
7275          * just use the extent.
7276          *
7277          */
7278         if (!create) {
7279                 len = min(len, em->len - (start - em->start));
7280                 lockstart = start + len;
7281                 goto unlock;
7282         }
7283
7284         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
7285             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
7286              em->block_start != EXTENT_MAP_HOLE)) {
7287                 int type;
7288                 u64 block_start, orig_start, orig_block_len, ram_bytes;
7289
7290                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7291                         type = BTRFS_ORDERED_PREALLOC;
7292                 else
7293                         type = BTRFS_ORDERED_NOCOW;
7294                 len = min(len, em->len - (start - em->start));
7295                 block_start = em->block_start + (start - em->start);
7296
7297                 if (can_nocow_extent(inode, start, &len, &orig_start,
7298                                      &orig_block_len, &ram_bytes) == 1) {
7299                         if (type == BTRFS_ORDERED_PREALLOC) {
7300                                 free_extent_map(em);
7301                                 em = create_pinned_em(inode, start, len,
7302                                                        orig_start,
7303                                                        block_start, len,
7304                                                        orig_block_len,
7305                                                        ram_bytes, type);
7306                                 if (IS_ERR(em)) {
7307                                         ret = PTR_ERR(em);
7308                                         goto unlock_err;
7309                                 }
7310                         }
7311
7312                         ret = btrfs_add_ordered_extent_dio(inode, start,
7313                                            block_start, len, len, type);
7314                         if (ret) {
7315                                 free_extent_map(em);
7316                                 goto unlock_err;
7317                         }
7318                         goto unlock;
7319                 }
7320         }
7321
7322         /*
7323          * this will cow the extent, reset the len in case we changed
7324          * it above
7325          */
7326         len = bh_result->b_size;
7327         free_extent_map(em);
7328         em = btrfs_new_extent_direct(inode, start, len);
7329         if (IS_ERR(em)) {
7330                 ret = PTR_ERR(em);
7331                 goto unlock_err;
7332         }
7333         len = min(len, em->len - (start - em->start));
7334 unlock:
7335         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
7336                 inode->i_blkbits;
7337         bh_result->b_size = len;
7338         bh_result->b_bdev = em->bdev;
7339         set_buffer_mapped(bh_result);
7340         if (create) {
7341                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
7342                         set_buffer_new(bh_result);
7343
7344                 /*
7345                  * Need to update the i_size under the extent lock so buffered
7346                  * readers will get the updated i_size when we unlock.
7347                  */
7348                 if (start + len > i_size_read(inode))
7349                         i_size_write(inode, start + len);
7350
7351                 if (len < orig_len) {
7352                         spin_lock(&BTRFS_I(inode)->lock);
7353                         BTRFS_I(inode)->outstanding_extents++;
7354                         spin_unlock(&BTRFS_I(inode)->lock);
7355                 }
7356                 btrfs_free_reserved_data_space(inode, len);
7357         }
7358
7359         /*
7360          * In the case of write we need to clear and unlock the entire range,
7361          * in the case of read we need to unlock only the end area that we
7362          * aren't using if there is any left over space.
7363          */
7364         if (lockstart < lockend) {
7365                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
7366                                  lockend, unlock_bits, 1, 0,
7367                                  &cached_state, GFP_NOFS);
7368         } else {
7369                 free_extent_state(cached_state);
7370         }
7371
7372         free_extent_map(em);
7373
7374         return 0;
7375
7376 unlock_err:
7377         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
7378                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
7379         return ret;
7380 }
7381
7382 static inline int submit_dio_repair_bio(struct inode *inode, struct bio *bio,
7383                                         int rw, int mirror_num)
7384 {
7385         struct btrfs_root *root = BTRFS_I(inode)->root;
7386         int ret;
7387
7388         BUG_ON(rw & REQ_WRITE);
7389
7390         bio_get(bio);
7391
7392         ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7393                                   BTRFS_WQ_ENDIO_DIO_REPAIR);
7394         if (ret)
7395                 goto err;
7396
7397         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
7398 err:
7399         bio_put(bio);
7400         return ret;
7401 }
7402
7403 static int btrfs_check_dio_repairable(struct inode *inode,
7404                                       struct bio *failed_bio,
7405                                       struct io_failure_record *failrec,
7406                                       int failed_mirror)
7407 {
7408         int num_copies;
7409
7410         num_copies = btrfs_num_copies(BTRFS_I(inode)->root->fs_info,
7411                                       failrec->logical, failrec->len);
7412         if (num_copies == 1) {
7413                 /*
7414                  * we only have a single copy of the data, so don't bother with
7415                  * all the retry and error correction code that follows. no
7416                  * matter what the error is, it is very likely to persist.
7417                  */
7418                 pr_debug("Check DIO Repairable: cannot repair, num_copies=%d, next_mirror %d, failed_mirror %d\n",
7419                          num_copies, failrec->this_mirror, failed_mirror);
7420                 return 0;
7421         }
7422
7423         failrec->failed_mirror = failed_mirror;
7424         failrec->this_mirror++;
7425         if (failrec->this_mirror == failed_mirror)
7426                 failrec->this_mirror++;
7427
7428         if (failrec->this_mirror > num_copies) {
7429                 pr_debug("Check DIO Repairable: (fail) num_copies=%d, next_mirror %d, failed_mirror %d\n",
7430                          num_copies, failrec->this_mirror, failed_mirror);
7431                 return 0;
7432         }
7433
7434         return 1;
7435 }
7436
7437 static int dio_read_error(struct inode *inode, struct bio *failed_bio,
7438                           struct page *page, u64 start, u64 end,
7439                           int failed_mirror, bio_end_io_t *repair_endio,
7440                           void *repair_arg)
7441 {
7442         struct io_failure_record *failrec;
7443         struct bio *bio;
7444         int isector;
7445         int read_mode;
7446         int ret;
7447
7448         BUG_ON(failed_bio->bi_rw & REQ_WRITE);
7449
7450         ret = btrfs_get_io_failure_record(inode, start, end, &failrec);
7451         if (ret)
7452                 return ret;
7453
7454         ret = btrfs_check_dio_repairable(inode, failed_bio, failrec,
7455                                          failed_mirror);
7456         if (!ret) {
7457                 free_io_failure(inode, failrec);
7458                 return -EIO;
7459         }
7460
7461         if (failed_bio->bi_vcnt > 1)
7462                 read_mode = READ_SYNC | REQ_FAILFAST_DEV;
7463         else
7464                 read_mode = READ_SYNC;
7465
7466         isector = start - btrfs_io_bio(failed_bio)->logical;
7467         isector >>= inode->i_sb->s_blocksize_bits;
7468         bio = btrfs_create_repair_bio(inode, failed_bio, failrec, page,
7469                                       0, isector, repair_endio, repair_arg);
7470         if (!bio) {
7471                 free_io_failure(inode, failrec);
7472                 return -EIO;
7473         }
7474
7475         btrfs_debug(BTRFS_I(inode)->root->fs_info,
7476                     "Repair DIO Read Error: submitting new dio read[%#x] to this_mirror=%d, in_validation=%d\n",
7477                     read_mode, failrec->this_mirror, failrec->in_validation);
7478
7479         ret = submit_dio_repair_bio(inode, bio, read_mode,
7480                                     failrec->this_mirror);
7481         if (ret) {
7482                 free_io_failure(inode, failrec);
7483                 bio_put(bio);
7484         }
7485
7486         return ret;
7487 }
7488
7489 struct btrfs_retry_complete {
7490         struct completion done;
7491         struct inode *inode;
7492         u64 start;
7493         int uptodate;
7494 };
7495
7496 static void btrfs_retry_endio_nocsum(struct bio *bio, int err)
7497 {
7498         struct btrfs_retry_complete *done = bio->bi_private;
7499         struct bio_vec *bvec;
7500         int i;
7501
7502         if (err)
7503                 goto end;
7504
7505         done->uptodate = 1;
7506         bio_for_each_segment_all(bvec, bio, i)
7507                 clean_io_failure(done->inode, done->start, bvec->bv_page, 0);
7508 end:
7509         complete(&done->done);
7510         bio_put(bio);
7511 }
7512
7513 static int __btrfs_correct_data_nocsum(struct inode *inode,
7514                                        struct btrfs_io_bio *io_bio)
7515 {
7516         struct bio_vec *bvec;
7517         struct btrfs_retry_complete done;
7518         u64 start;
7519         int i;
7520         int ret;
7521
7522         start = io_bio->logical;
7523         done.inode = inode;
7524
7525         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7526 try_again:
7527                 done.uptodate = 0;
7528                 done.start = start;
7529                 init_completion(&done.done);
7530
7531                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7532                                      start + bvec->bv_len - 1,
7533                                      io_bio->mirror_num,
7534                                      btrfs_retry_endio_nocsum, &done);
7535                 if (ret)
7536                         return ret;
7537
7538                 wait_for_completion(&done.done);
7539
7540                 if (!done.uptodate) {
7541                         /* We might have another mirror, so try again */
7542                         goto try_again;
7543                 }
7544
7545                 start += bvec->bv_len;
7546         }
7547
7548         return 0;
7549 }
7550
7551 static void btrfs_retry_endio(struct bio *bio, int err)
7552 {
7553         struct btrfs_retry_complete *done = bio->bi_private;
7554         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7555         struct bio_vec *bvec;
7556         int uptodate;
7557         int ret;
7558         int i;
7559
7560         if (err)
7561                 goto end;
7562
7563         uptodate = 1;
7564         bio_for_each_segment_all(bvec, bio, i) {
7565                 ret = __readpage_endio_check(done->inode, io_bio, i,
7566                                              bvec->bv_page, 0,
7567                                              done->start, bvec->bv_len);
7568                 if (!ret)
7569                         clean_io_failure(done->inode, done->start,
7570                                          bvec->bv_page, 0);
7571                 else
7572                         uptodate = 0;
7573         }
7574
7575         done->uptodate = uptodate;
7576 end:
7577         complete(&done->done);
7578         bio_put(bio);
7579 }
7580
7581 static int __btrfs_subio_endio_read(struct inode *inode,
7582                                     struct btrfs_io_bio *io_bio, int err)
7583 {
7584         struct bio_vec *bvec;
7585         struct btrfs_retry_complete done;
7586         u64 start;
7587         u64 offset = 0;
7588         int i;
7589         int ret;
7590
7591         err = 0;
7592         start = io_bio->logical;
7593         done.inode = inode;
7594
7595         bio_for_each_segment_all(bvec, &io_bio->bio, i) {
7596                 ret = __readpage_endio_check(inode, io_bio, i, bvec->bv_page,
7597                                              0, start, bvec->bv_len);
7598                 if (likely(!ret))
7599                         goto next;
7600 try_again:
7601                 done.uptodate = 0;
7602                 done.start = start;
7603                 init_completion(&done.done);
7604
7605                 ret = dio_read_error(inode, &io_bio->bio, bvec->bv_page, start,
7606                                      start + bvec->bv_len - 1,
7607                                      io_bio->mirror_num,
7608                                      btrfs_retry_endio, &done);
7609                 if (ret) {
7610                         err = ret;
7611                         goto next;
7612                 }
7613
7614                 wait_for_completion(&done.done);
7615
7616                 if (!done.uptodate) {
7617                         /* We might have another mirror, so try again */
7618                         goto try_again;
7619                 }
7620 next:
7621                 offset += bvec->bv_len;
7622                 start += bvec->bv_len;
7623         }
7624
7625         return err;
7626 }
7627
7628 static int btrfs_subio_endio_read(struct inode *inode,
7629                                   struct btrfs_io_bio *io_bio, int err)
7630 {
7631         bool skip_csum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7632
7633         if (skip_csum) {
7634                 if (unlikely(err))
7635                         return __btrfs_correct_data_nocsum(inode, io_bio);
7636                 else
7637                         return 0;
7638         } else {
7639                 return __btrfs_subio_endio_read(inode, io_bio, err);
7640         }
7641 }
7642
7643 static void btrfs_endio_direct_read(struct bio *bio, int err)
7644 {
7645         struct btrfs_dio_private *dip = bio->bi_private;
7646         struct inode *inode = dip->inode;
7647         struct bio *dio_bio;
7648         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7649
7650         if (dip->flags & BTRFS_DIO_ORIG_BIO_SUBMITTED)
7651                 err = btrfs_subio_endio_read(inode, io_bio, err);
7652
7653         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
7654                       dip->logical_offset + dip->bytes - 1);
7655         dio_bio = dip->dio_bio;
7656
7657         kfree(dip);
7658
7659         /* If we had a csum failure make sure to clear the uptodate flag */
7660         if (err)
7661                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7662         dio_end_io(dio_bio, err);
7663
7664         if (io_bio->end_io)
7665                 io_bio->end_io(io_bio, err);
7666         bio_put(bio);
7667 }
7668
7669 static void btrfs_endio_direct_write(struct bio *bio, int err)
7670 {
7671         struct btrfs_dio_private *dip = bio->bi_private;
7672         struct inode *inode = dip->inode;
7673         struct btrfs_root *root = BTRFS_I(inode)->root;
7674         struct btrfs_ordered_extent *ordered = NULL;
7675         u64 ordered_offset = dip->logical_offset;
7676         u64 ordered_bytes = dip->bytes;
7677         struct bio *dio_bio;
7678         int ret;
7679
7680         if (err)
7681                 goto out_done;
7682 again:
7683         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7684                                                    &ordered_offset,
7685                                                    ordered_bytes, !err);
7686         if (!ret)
7687                 goto out_test;
7688
7689         btrfs_init_work(&ordered->work, btrfs_endio_write_helper,
7690                         finish_ordered_fn, NULL, NULL);
7691         btrfs_queue_work(root->fs_info->endio_write_workers,
7692                          &ordered->work);
7693 out_test:
7694         /*
7695          * our bio might span multiple ordered extents.  If we haven't
7696          * completed the accounting for the whole dio, go back and try again
7697          */
7698         if (ordered_offset < dip->logical_offset + dip->bytes) {
7699                 ordered_bytes = dip->logical_offset + dip->bytes -
7700                         ordered_offset;
7701                 ordered = NULL;
7702                 goto again;
7703         }
7704 out_done:
7705         dio_bio = dip->dio_bio;
7706
7707         kfree(dip);
7708
7709         /* If we had an error make sure to clear the uptodate flag */
7710         if (err)
7711                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7712         dio_end_io(dio_bio, err);
7713         bio_put(bio);
7714 }
7715
7716 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7717                                     struct bio *bio, int mirror_num,
7718                                     unsigned long bio_flags, u64 offset)
7719 {
7720         int ret;
7721         struct btrfs_root *root = BTRFS_I(inode)->root;
7722         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7723         BUG_ON(ret); /* -ENOMEM */
7724         return 0;
7725 }
7726
7727 static void btrfs_end_dio_bio(struct bio *bio, int err)
7728 {
7729         struct btrfs_dio_private *dip = bio->bi_private;
7730
7731         if (err)
7732                 btrfs_warn(BTRFS_I(dip->inode)->root->fs_info,
7733                            "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7734                            btrfs_ino(dip->inode), bio->bi_rw,
7735                            (unsigned long long)bio->bi_iter.bi_sector,
7736                            bio->bi_iter.bi_size, err);
7737
7738         if (dip->subio_endio)
7739                 err = dip->subio_endio(dip->inode, btrfs_io_bio(bio), err);
7740
7741         if (err) {
7742                 dip->errors = 1;
7743
7744                 /*
7745                  * before atomic variable goto zero, we must make sure
7746                  * dip->errors is perceived to be set.
7747                  */
7748                 smp_mb__before_atomic();
7749         }
7750
7751         /* if there are more bios still pending for this dio, just exit */
7752         if (!atomic_dec_and_test(&dip->pending_bios))
7753                 goto out;
7754
7755         if (dip->errors) {
7756                 bio_io_error(dip->orig_bio);
7757         } else {
7758                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7759                 bio_endio(dip->orig_bio, 0);
7760         }
7761 out:
7762         bio_put(bio);
7763 }
7764
7765 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7766                                        u64 first_sector, gfp_t gfp_flags)
7767 {
7768         int nr_vecs = bio_get_nr_vecs(bdev);
7769         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7770 }
7771
7772 static inline int btrfs_lookup_and_bind_dio_csum(struct btrfs_root *root,
7773                                                  struct inode *inode,
7774                                                  struct btrfs_dio_private *dip,
7775                                                  struct bio *bio,
7776                                                  u64 file_offset)
7777 {
7778         struct btrfs_io_bio *io_bio = btrfs_io_bio(bio);
7779         struct btrfs_io_bio *orig_io_bio = btrfs_io_bio(dip->orig_bio);
7780         int ret;
7781
7782         /*
7783          * We load all the csum data we need when we submit
7784          * the first bio to reduce the csum tree search and
7785          * contention.
7786          */
7787         if (dip->logical_offset == file_offset) {
7788                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip->orig_bio,
7789                                                 file_offset);
7790                 if (ret)
7791                         return ret;
7792         }
7793
7794         if (bio == dip->orig_bio)
7795                 return 0;
7796
7797         file_offset -= dip->logical_offset;
7798         file_offset >>= inode->i_sb->s_blocksize_bits;
7799         io_bio->csum = (u8 *)(((u32 *)orig_io_bio->csum) + file_offset);
7800
7801         return 0;
7802 }
7803
7804 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7805                                          int rw, u64 file_offset, int skip_sum,
7806                                          int async_submit)
7807 {
7808         struct btrfs_dio_private *dip = bio->bi_private;
7809         int write = rw & REQ_WRITE;
7810         struct btrfs_root *root = BTRFS_I(inode)->root;
7811         int ret;
7812
7813         if (async_submit)
7814                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7815
7816         bio_get(bio);
7817
7818         if (!write) {
7819                 ret = btrfs_bio_wq_end_io(root->fs_info, bio,
7820                                 BTRFS_WQ_ENDIO_DATA);
7821                 if (ret)
7822                         goto err;
7823         }
7824
7825         if (skip_sum)
7826                 goto map;
7827
7828         if (write && async_submit) {
7829                 ret = btrfs_wq_submit_bio(root->fs_info,
7830                                    inode, rw, bio, 0, 0,
7831                                    file_offset,
7832                                    __btrfs_submit_bio_start_direct_io,
7833                                    __btrfs_submit_bio_done);
7834                 goto err;
7835         } else if (write) {
7836                 /*
7837                  * If we aren't doing async submit, calculate the csum of the
7838                  * bio now.
7839                  */
7840                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7841                 if (ret)
7842                         goto err;
7843         } else {
7844                 ret = btrfs_lookup_and_bind_dio_csum(root, inode, dip, bio,
7845                                                      file_offset);
7846                 if (ret)
7847                         goto err;
7848         }
7849 map:
7850         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7851 err:
7852         bio_put(bio);
7853         return ret;
7854 }
7855
7856 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7857                                     int skip_sum)
7858 {
7859         struct inode *inode = dip->inode;
7860         struct btrfs_root *root = BTRFS_I(inode)->root;
7861         struct bio *bio;
7862         struct bio *orig_bio = dip->orig_bio;
7863         struct bio_vec *bvec = orig_bio->bi_io_vec;
7864         u64 start_sector = orig_bio->bi_iter.bi_sector;
7865         u64 file_offset = dip->logical_offset;
7866         u64 submit_len = 0;
7867         u64 map_length;
7868         int nr_pages = 0;
7869         int ret;
7870         int async_submit = 0;
7871
7872         map_length = orig_bio->bi_iter.bi_size;
7873         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7874                               &map_length, NULL, 0);
7875         if (ret)
7876                 return -EIO;
7877
7878         if (map_length >= orig_bio->bi_iter.bi_size) {
7879                 bio = orig_bio;
7880                 dip->flags |= BTRFS_DIO_ORIG_BIO_SUBMITTED;
7881                 goto submit;
7882         }
7883
7884         /* async crcs make it difficult to collect full stripe writes. */
7885         if (btrfs_get_alloc_profile(root, 1) & BTRFS_BLOCK_GROUP_RAID56_MASK)
7886                 async_submit = 0;
7887         else
7888                 async_submit = 1;
7889
7890         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7891         if (!bio)
7892                 return -ENOMEM;
7893
7894         bio->bi_private = dip;
7895         bio->bi_end_io = btrfs_end_dio_bio;
7896         btrfs_io_bio(bio)->logical = file_offset;
7897         atomic_inc(&dip->pending_bios);
7898
7899         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7900                 if (map_length < submit_len + bvec->bv_len ||
7901                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7902                                  bvec->bv_offset) < bvec->bv_len) {
7903                         /*
7904                          * inc the count before we submit the bio so
7905                          * we know the end IO handler won't happen before
7906                          * we inc the count. Otherwise, the dip might get freed
7907                          * before we're done setting it up
7908                          */
7909                         atomic_inc(&dip->pending_bios);
7910                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7911                                                      file_offset, skip_sum,
7912                                                      async_submit);
7913                         if (ret) {
7914                                 bio_put(bio);
7915                                 atomic_dec(&dip->pending_bios);
7916                                 goto out_err;
7917                         }
7918
7919                         start_sector += submit_len >> 9;
7920                         file_offset += submit_len;
7921
7922                         submit_len = 0;
7923                         nr_pages = 0;
7924
7925                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7926                                                   start_sector, GFP_NOFS);
7927                         if (!bio)
7928                                 goto out_err;
7929                         bio->bi_private = dip;
7930                         bio->bi_end_io = btrfs_end_dio_bio;
7931                         btrfs_io_bio(bio)->logical = file_offset;
7932
7933                         map_length = orig_bio->bi_iter.bi_size;
7934                         ret = btrfs_map_block(root->fs_info, rw,
7935                                               start_sector << 9,
7936                                               &map_length, NULL, 0);
7937                         if (ret) {
7938                                 bio_put(bio);
7939                                 goto out_err;
7940                         }
7941                 } else {
7942                         submit_len += bvec->bv_len;
7943                         nr_pages++;
7944                         bvec++;
7945                 }
7946         }
7947
7948 submit:
7949         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7950                                      async_submit);
7951         if (!ret)
7952                 return 0;
7953
7954         bio_put(bio);
7955 out_err:
7956         dip->errors = 1;
7957         /*
7958          * before atomic variable goto zero, we must
7959          * make sure dip->errors is perceived to be set.
7960          */
7961         smp_mb__before_atomic();
7962         if (atomic_dec_and_test(&dip->pending_bios))
7963                 bio_io_error(dip->orig_bio);
7964
7965         /* bio_end_io() will handle error, so we needn't return it */
7966         return 0;
7967 }
7968
7969 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7970                                 struct inode *inode, loff_t file_offset)
7971 {
7972         struct btrfs_root *root = BTRFS_I(inode)->root;
7973         struct btrfs_dio_private *dip;
7974         struct bio *io_bio;
7975         struct btrfs_io_bio *btrfs_bio;
7976         int skip_sum;
7977         int write = rw & REQ_WRITE;
7978         int ret = 0;
7979
7980         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7981
7982         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7983         if (!io_bio) {
7984                 ret = -ENOMEM;
7985                 goto free_ordered;
7986         }
7987
7988         dip = kzalloc(sizeof(*dip), GFP_NOFS);
7989         if (!dip) {
7990                 ret = -ENOMEM;
7991                 goto free_io_bio;
7992         }
7993
7994         dip->private = dio_bio->bi_private;
7995         dip->inode = inode;
7996         dip->logical_offset = file_offset;
7997         dip->bytes = dio_bio->bi_iter.bi_size;
7998         dip->disk_bytenr = (u64)dio_bio->bi_iter.bi_sector << 9;
7999         io_bio->bi_private = dip;
8000         dip->orig_bio = io_bio;
8001         dip->dio_bio = dio_bio;
8002         atomic_set(&dip->pending_bios, 0);
8003         btrfs_bio = btrfs_io_bio(io_bio);
8004         btrfs_bio->logical = file_offset;
8005
8006         if (write) {
8007                 io_bio->bi_end_io = btrfs_endio_direct_write;
8008         } else {
8009                 io_bio->bi_end_io = btrfs_endio_direct_read;
8010                 dip->subio_endio = btrfs_subio_endio_read;
8011         }
8012
8013         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
8014         if (!ret)
8015                 return;
8016
8017         if (btrfs_bio->end_io)
8018                 btrfs_bio->end_io(btrfs_bio, ret);
8019 free_io_bio:
8020         bio_put(io_bio);
8021
8022 free_ordered:
8023         /*
8024          * If this is a write, we need to clean up the reserved space and kill
8025          * the ordered extent.
8026          */
8027         if (write) {
8028                 struct btrfs_ordered_extent *ordered;
8029                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
8030                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
8031                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
8032                         btrfs_free_reserved_extent(root, ordered->start,
8033                                                    ordered->disk_len, 1);
8034                 btrfs_put_ordered_extent(ordered);
8035                 btrfs_put_ordered_extent(ordered);
8036         }
8037         bio_endio(dio_bio, ret);
8038 }
8039
8040 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
8041                         const struct iov_iter *iter, loff_t offset)
8042 {
8043         int seg;
8044         int i;
8045         unsigned blocksize_mask = root->sectorsize - 1;
8046         ssize_t retval = -EINVAL;
8047
8048         if (offset & blocksize_mask)
8049                 goto out;
8050
8051         if (iov_iter_alignment(iter) & blocksize_mask)
8052                 goto out;
8053
8054         /* If this is a write we don't need to check anymore */
8055         if (rw & WRITE)
8056                 return 0;
8057         /*
8058          * Check to make sure we don't have duplicate iov_base's in this
8059          * iovec, if so return EINVAL, otherwise we'll get csum errors
8060          * when reading back.
8061          */
8062         for (seg = 0; seg < iter->nr_segs; seg++) {
8063                 for (i = seg + 1; i < iter->nr_segs; i++) {
8064                         if (iter->iov[seg].iov_base == iter->iov[i].iov_base)
8065                                 goto out;
8066                 }
8067         }
8068         retval = 0;
8069 out:
8070         return retval;
8071 }
8072
8073 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
8074                         struct iov_iter *iter, loff_t offset)
8075 {
8076         struct file *file = iocb->ki_filp;
8077         struct inode *inode = file->f_mapping->host;
8078         size_t count = 0;
8079         int flags = 0;
8080         bool wakeup = true;
8081         bool relock = false;
8082         ssize_t ret;
8083
8084         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iter, offset))
8085                 return 0;
8086
8087         atomic_inc(&inode->i_dio_count);
8088         smp_mb__after_atomic();
8089
8090         /*
8091          * The generic stuff only does filemap_write_and_wait_range, which
8092          * isn't enough if we've written compressed pages to this area, so
8093          * we need to flush the dirty pages again to make absolutely sure
8094          * that any outstanding dirty pages are on disk.
8095          */
8096         count = iov_iter_count(iter);
8097         if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8098                      &BTRFS_I(inode)->runtime_flags))
8099                 filemap_fdatawrite_range(inode->i_mapping, offset,
8100                                          offset + count - 1);
8101
8102         if (rw & WRITE) {
8103                 /*
8104                  * If the write DIO is beyond the EOF, we need update
8105                  * the isize, but it is protected by i_mutex. So we can
8106                  * not unlock the i_mutex at this case.
8107                  */
8108                 if (offset + count <= inode->i_size) {
8109                         mutex_unlock(&inode->i_mutex);
8110                         relock = true;
8111                 }
8112                 ret = btrfs_delalloc_reserve_space(inode, count);
8113                 if (ret)
8114                         goto out;
8115         } else if (test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
8116                                      &BTRFS_I(inode)->runtime_flags)) {
8117                 inode_dio_done(inode);
8118                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
8119                 wakeup = false;
8120         }
8121
8122         ret = __blockdev_direct_IO(rw, iocb, inode,
8123                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
8124                         iter, offset, btrfs_get_blocks_direct, NULL,
8125                         btrfs_submit_direct, flags);
8126         if (rw & WRITE) {
8127                 if (ret < 0 && ret != -EIOCBQUEUED)
8128                         btrfs_delalloc_release_space(inode, count);
8129                 else if (ret >= 0 && (size_t)ret < count)
8130                         btrfs_delalloc_release_space(inode,
8131                                                      count - (size_t)ret);
8132         }
8133 out:
8134         if (wakeup)
8135                 inode_dio_done(inode);
8136         if (relock)
8137                 mutex_lock(&inode->i_mutex);
8138
8139         return ret;
8140 }
8141
8142 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
8143
8144 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
8145                 __u64 start, __u64 len)
8146 {
8147         int     ret;
8148
8149         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
8150         if (ret)
8151                 return ret;
8152
8153         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
8154 }
8155
8156 int btrfs_readpage(struct file *file, struct page *page)
8157 {
8158         struct extent_io_tree *tree;
8159         tree = &BTRFS_I(page->mapping->host)->io_tree;
8160         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
8161 }
8162
8163 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
8164 {
8165         struct extent_io_tree *tree;
8166
8167
8168         if (current->flags & PF_MEMALLOC) {
8169                 redirty_page_for_writepage(wbc, page);
8170                 unlock_page(page);
8171                 return 0;
8172         }
8173         tree = &BTRFS_I(page->mapping->host)->io_tree;
8174         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
8175 }
8176
8177 static int btrfs_writepages(struct address_space *mapping,
8178                             struct writeback_control *wbc)
8179 {
8180         struct extent_io_tree *tree;
8181
8182         tree = &BTRFS_I(mapping->host)->io_tree;
8183         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
8184 }
8185
8186 static int
8187 btrfs_readpages(struct file *file, struct address_space *mapping,
8188                 struct list_head *pages, unsigned nr_pages)
8189 {
8190         struct extent_io_tree *tree;
8191         tree = &BTRFS_I(mapping->host)->io_tree;
8192         return extent_readpages(tree, mapping, pages, nr_pages,
8193                                 btrfs_get_extent);
8194 }
8195 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8196 {
8197         struct extent_io_tree *tree;
8198         struct extent_map_tree *map;
8199         int ret;
8200
8201         tree = &BTRFS_I(page->mapping->host)->io_tree;
8202         map = &BTRFS_I(page->mapping->host)->extent_tree;
8203         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
8204         if (ret == 1) {
8205                 ClearPagePrivate(page);
8206                 set_page_private(page, 0);
8207                 page_cache_release(page);
8208         }
8209         return ret;
8210 }
8211
8212 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
8213 {
8214         if (PageWriteback(page) || PageDirty(page))
8215                 return 0;
8216         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
8217 }
8218
8219 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
8220                                  unsigned int length)
8221 {
8222         struct inode *inode = page->mapping->host;
8223         struct extent_io_tree *tree;
8224         struct btrfs_ordered_extent *ordered;
8225         struct extent_state *cached_state = NULL;
8226         u64 page_start = page_offset(page);
8227         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
8228         int inode_evicting = inode->i_state & I_FREEING;
8229
8230         /*
8231          * we have the page locked, so new writeback can't start,
8232          * and the dirty bit won't be cleared while we are here.
8233          *
8234          * Wait for IO on this page so that we can safely clear
8235          * the PagePrivate2 bit and do ordered accounting
8236          */
8237         wait_on_page_writeback(page);
8238
8239         tree = &BTRFS_I(inode)->io_tree;
8240         if (offset) {
8241                 btrfs_releasepage(page, GFP_NOFS);
8242                 return;
8243         }
8244
8245         if (!inode_evicting)
8246                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
8247         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8248         if (ordered) {
8249                 /*
8250                  * IO on this page will never be started, so we need
8251                  * to account for any ordered extents now
8252                  */
8253                 if (!inode_evicting)
8254                         clear_extent_bit(tree, page_start, page_end,
8255                                          EXTENT_DIRTY | EXTENT_DELALLOC |
8256                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
8257                                          EXTENT_DEFRAG, 1, 0, &cached_state,
8258                                          GFP_NOFS);
8259                 /*
8260                  * whoever cleared the private bit is responsible
8261                  * for the finish_ordered_io
8262                  */
8263                 if (TestClearPagePrivate2(page)) {
8264                         struct btrfs_ordered_inode_tree *tree;
8265                         u64 new_len;
8266
8267                         tree = &BTRFS_I(inode)->ordered_tree;
8268
8269                         spin_lock_irq(&tree->lock);
8270                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
8271                         new_len = page_start - ordered->file_offset;
8272                         if (new_len < ordered->truncated_len)
8273                                 ordered->truncated_len = new_len;
8274                         spin_unlock_irq(&tree->lock);
8275
8276                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
8277                                                            page_start,
8278                                                            PAGE_CACHE_SIZE, 1))
8279                                 btrfs_finish_ordered_io(ordered);
8280                 }
8281                 btrfs_put_ordered_extent(ordered);
8282                 if (!inode_evicting) {
8283                         cached_state = NULL;
8284                         lock_extent_bits(tree, page_start, page_end, 0,
8285                                          &cached_state);
8286                 }
8287         }
8288
8289         if (!inode_evicting) {
8290                 clear_extent_bit(tree, page_start, page_end,
8291                                  EXTENT_LOCKED | EXTENT_DIRTY |
8292                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
8293                                  EXTENT_DEFRAG, 1, 1,
8294                                  &cached_state, GFP_NOFS);
8295
8296                 __btrfs_releasepage(page, GFP_NOFS);
8297         }
8298
8299         ClearPageChecked(page);
8300         if (PagePrivate(page)) {
8301                 ClearPagePrivate(page);
8302                 set_page_private(page, 0);
8303                 page_cache_release(page);
8304         }
8305 }
8306
8307 /*
8308  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
8309  * called from a page fault handler when a page is first dirtied. Hence we must
8310  * be careful to check for EOF conditions here. We set the page up correctly
8311  * for a written page which means we get ENOSPC checking when writing into
8312  * holes and correct delalloc and unwritten extent mapping on filesystems that
8313  * support these features.
8314  *
8315  * We are not allowed to take the i_mutex here so we have to play games to
8316  * protect against truncate races as the page could now be beyond EOF.  Because
8317  * vmtruncate() writes the inode size before removing pages, once we have the
8318  * page lock we can determine safely if the page is beyond EOF. If it is not
8319  * beyond EOF, then the page is guaranteed safe against truncation until we
8320  * unlock the page.
8321  */
8322 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
8323 {
8324         struct page *page = vmf->page;
8325         struct inode *inode = file_inode(vma->vm_file);
8326         struct btrfs_root *root = BTRFS_I(inode)->root;
8327         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
8328         struct btrfs_ordered_extent *ordered;
8329         struct extent_state *cached_state = NULL;
8330         char *kaddr;
8331         unsigned long zero_start;
8332         loff_t size;
8333         int ret;
8334         int reserved = 0;
8335         u64 page_start;
8336         u64 page_end;
8337
8338         sb_start_pagefault(inode->i_sb);
8339         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
8340         if (!ret) {
8341                 ret = file_update_time(vma->vm_file);
8342                 reserved = 1;
8343         }
8344         if (ret) {
8345                 if (ret == -ENOMEM)
8346                         ret = VM_FAULT_OOM;
8347                 else /* -ENOSPC, -EIO, etc */
8348                         ret = VM_FAULT_SIGBUS;
8349                 if (reserved)
8350                         goto out;
8351                 goto out_noreserve;
8352         }
8353
8354         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
8355 again:
8356         lock_page(page);
8357         size = i_size_read(inode);
8358         page_start = page_offset(page);
8359         page_end = page_start + PAGE_CACHE_SIZE - 1;
8360
8361         if ((page->mapping != inode->i_mapping) ||
8362             (page_start >= size)) {
8363                 /* page got truncated out from underneath us */
8364                 goto out_unlock;
8365         }
8366         wait_on_page_writeback(page);
8367
8368         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
8369         set_page_extent_mapped(page);
8370
8371         /*
8372          * we can't set the delalloc bits if there are pending ordered
8373          * extents.  Drop our locks and wait for them to finish
8374          */
8375         ordered = btrfs_lookup_ordered_extent(inode, page_start);
8376         if (ordered) {
8377                 unlock_extent_cached(io_tree, page_start, page_end,
8378                                      &cached_state, GFP_NOFS);
8379                 unlock_page(page);
8380                 btrfs_start_ordered_extent(inode, ordered, 1);
8381                 btrfs_put_ordered_extent(ordered);
8382                 goto again;
8383         }
8384
8385         /*
8386          * XXX - page_mkwrite gets called every time the page is dirtied, even
8387          * if it was already dirty, so for space accounting reasons we need to
8388          * clear any delalloc bits for the range we are fixing to save.  There
8389          * is probably a better way to do this, but for now keep consistent with
8390          * prepare_pages in the normal write path.
8391          */
8392         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
8393                           EXTENT_DIRTY | EXTENT_DELALLOC |
8394                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
8395                           0, 0, &cached_state, GFP_NOFS);
8396
8397         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
8398                                         &cached_state);
8399         if (ret) {
8400                 unlock_extent_cached(io_tree, page_start, page_end,
8401                                      &cached_state, GFP_NOFS);
8402                 ret = VM_FAULT_SIGBUS;
8403                 goto out_unlock;
8404         }
8405         ret = 0;
8406
8407         /* page is wholly or partially inside EOF */
8408         if (page_start + PAGE_CACHE_SIZE > size)
8409                 zero_start = size & ~PAGE_CACHE_MASK;
8410         else
8411                 zero_start = PAGE_CACHE_SIZE;
8412
8413         if (zero_start != PAGE_CACHE_SIZE) {
8414                 kaddr = kmap(page);
8415                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
8416                 flush_dcache_page(page);
8417                 kunmap(page);
8418         }
8419         ClearPageChecked(page);
8420         set_page_dirty(page);
8421         SetPageUptodate(page);
8422
8423         BTRFS_I(inode)->last_trans = root->fs_info->generation;
8424         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
8425         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
8426
8427         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
8428
8429 out_unlock:
8430         if (!ret) {
8431                 sb_end_pagefault(inode->i_sb);
8432                 return VM_FAULT_LOCKED;
8433         }
8434         unlock_page(page);
8435 out:
8436         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
8437 out_noreserve:
8438         sb_end_pagefault(inode->i_sb);
8439         return ret;
8440 }
8441
8442 static int btrfs_truncate(struct inode *inode)
8443 {
8444         struct btrfs_root *root = BTRFS_I(inode)->root;
8445         struct btrfs_block_rsv *rsv;
8446         int ret = 0;
8447         int err = 0;
8448         struct btrfs_trans_handle *trans;
8449         u64 mask = root->sectorsize - 1;
8450         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
8451
8452         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
8453                                        (u64)-1);
8454         if (ret)
8455                 return ret;
8456
8457         /*
8458          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
8459          * 3 things going on here
8460          *
8461          * 1) We need to reserve space for our orphan item and the space to
8462          * delete our orphan item.  Lord knows we don't want to have a dangling
8463          * orphan item because we didn't reserve space to remove it.
8464          *
8465          * 2) We need to reserve space to update our inode.
8466          *
8467          * 3) We need to have something to cache all the space that is going to
8468          * be free'd up by the truncate operation, but also have some slack
8469          * space reserved in case it uses space during the truncate (thank you
8470          * very much snapshotting).
8471          *
8472          * And we need these to all be seperate.  The fact is we can use alot of
8473          * space doing the truncate, and we have no earthly idea how much space
8474          * we will use, so we need the truncate reservation to be seperate so it
8475          * doesn't end up using space reserved for updating the inode or
8476          * removing the orphan item.  We also need to be able to stop the
8477          * transaction and start a new one, which means we need to be able to
8478          * update the inode several times, and we have no idea of knowing how
8479          * many times that will be, so we can't just reserve 1 item for the
8480          * entirety of the opration, so that has to be done seperately as well.
8481          * Then there is the orphan item, which does indeed need to be held on
8482          * to for the whole operation, and we need nobody to touch this reserved
8483          * space except the orphan code.
8484          *
8485          * So that leaves us with
8486          *
8487          * 1) root->orphan_block_rsv - for the orphan deletion.
8488          * 2) rsv - for the truncate reservation, which we will steal from the
8489          * transaction reservation.
8490          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
8491          * updating the inode.
8492          */
8493         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
8494         if (!rsv)
8495                 return -ENOMEM;
8496         rsv->size = min_size;
8497         rsv->failfast = 1;
8498
8499         /*
8500          * 1 for the truncate slack space
8501          * 1 for updating the inode.
8502          */
8503         trans = btrfs_start_transaction(root, 2);
8504         if (IS_ERR(trans)) {
8505                 err = PTR_ERR(trans);
8506                 goto out;
8507         }
8508
8509         /* Migrate the slack space for the truncate to our reserve */
8510         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
8511                                       min_size);
8512         BUG_ON(ret);
8513
8514         /*
8515          * So if we truncate and then write and fsync we normally would just
8516          * write the extents that changed, which is a problem if we need to
8517          * first truncate that entire inode.  So set this flag so we write out
8518          * all of the extents in the inode to the sync log so we're completely
8519          * safe.
8520          */
8521         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
8522         trans->block_rsv = rsv;
8523
8524         while (1) {
8525                 ret = btrfs_truncate_inode_items(trans, root, inode,
8526                                                  inode->i_size,
8527                                                  BTRFS_EXTENT_DATA_KEY);
8528                 if (ret != -ENOSPC) {
8529                         err = ret;
8530                         break;
8531                 }
8532
8533                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8534                 ret = btrfs_update_inode(trans, root, inode);
8535                 if (ret) {
8536                         err = ret;
8537                         break;
8538                 }
8539
8540                 btrfs_end_transaction(trans, root);
8541                 btrfs_btree_balance_dirty(root);
8542
8543                 trans = btrfs_start_transaction(root, 2);
8544                 if (IS_ERR(trans)) {
8545                         ret = err = PTR_ERR(trans);
8546                         trans = NULL;
8547                         break;
8548                 }
8549
8550                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
8551                                               rsv, min_size);
8552                 BUG_ON(ret);    /* shouldn't happen */
8553                 trans->block_rsv = rsv;
8554         }
8555
8556         if (ret == 0 && inode->i_nlink > 0) {
8557                 trans->block_rsv = root->orphan_block_rsv;
8558                 ret = btrfs_orphan_del(trans, inode);
8559                 if (ret)
8560                         err = ret;
8561         }
8562
8563         if (trans) {
8564                 trans->block_rsv = &root->fs_info->trans_block_rsv;
8565                 ret = btrfs_update_inode(trans, root, inode);
8566                 if (ret && !err)
8567                         err = ret;
8568
8569                 ret = btrfs_end_transaction(trans, root);
8570                 btrfs_btree_balance_dirty(root);
8571         }
8572
8573 out:
8574         btrfs_free_block_rsv(root, rsv);
8575
8576         if (ret && !err)
8577                 err = ret;
8578
8579         return err;
8580 }
8581
8582 /*
8583  * create a new subvolume directory/inode (helper for the ioctl).
8584  */
8585 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
8586                              struct btrfs_root *new_root,
8587                              struct btrfs_root *parent_root,
8588                              u64 new_dirid)
8589 {
8590         struct inode *inode;
8591         int err;
8592         u64 index = 0;
8593
8594         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
8595                                 new_dirid, new_dirid,
8596                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
8597                                 &index);
8598         if (IS_ERR(inode))
8599                 return PTR_ERR(inode);
8600         inode->i_op = &btrfs_dir_inode_operations;
8601         inode->i_fop = &btrfs_dir_file_operations;
8602
8603         set_nlink(inode, 1);
8604         btrfs_i_size_write(inode, 0);
8605         unlock_new_inode(inode);
8606
8607         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
8608         if (err)
8609                 btrfs_err(new_root->fs_info,
8610                           "error inheriting subvolume %llu properties: %d",
8611                           new_root->root_key.objectid, err);
8612
8613         err = btrfs_update_inode(trans, new_root, inode);
8614
8615         iput(inode);
8616         return err;
8617 }
8618
8619 struct inode *btrfs_alloc_inode(struct super_block *sb)
8620 {
8621         struct btrfs_inode *ei;
8622         struct inode *inode;
8623
8624         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
8625         if (!ei)
8626                 return NULL;
8627
8628         ei->root = NULL;
8629         ei->generation = 0;
8630         ei->last_trans = 0;
8631         ei->last_sub_trans = 0;
8632         ei->logged_trans = 0;
8633         ei->delalloc_bytes = 0;
8634         ei->defrag_bytes = 0;
8635         ei->disk_i_size = 0;
8636         ei->flags = 0;
8637         ei->csum_bytes = 0;
8638         ei->index_cnt = (u64)-1;
8639         ei->dir_index = 0;
8640         ei->last_unlink_trans = 0;
8641         ei->last_log_commit = 0;
8642
8643         spin_lock_init(&ei->lock);
8644         ei->outstanding_extents = 0;
8645         ei->reserved_extents = 0;
8646
8647         ei->runtime_flags = 0;
8648         ei->force_compress = BTRFS_COMPRESS_NONE;
8649
8650         ei->delayed_node = NULL;
8651
8652         ei->i_otime.tv_sec = 0;
8653         ei->i_otime.tv_nsec = 0;
8654
8655         inode = &ei->vfs_inode;
8656         extent_map_tree_init(&ei->extent_tree);
8657         extent_io_tree_init(&ei->io_tree, &inode->i_data);
8658         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
8659         ei->io_tree.track_uptodate = 1;
8660         ei->io_failure_tree.track_uptodate = 1;
8661         atomic_set(&ei->sync_writers, 0);
8662         mutex_init(&ei->log_mutex);
8663         mutex_init(&ei->delalloc_mutex);
8664         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
8665         INIT_LIST_HEAD(&ei->delalloc_inodes);
8666         RB_CLEAR_NODE(&ei->rb_node);
8667
8668         return inode;
8669 }
8670
8671 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
8672 void btrfs_test_destroy_inode(struct inode *inode)
8673 {
8674         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8675         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8676 }
8677 #endif
8678
8679 static void btrfs_i_callback(struct rcu_head *head)
8680 {
8681         struct inode *inode = container_of(head, struct inode, i_rcu);
8682         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8683 }
8684
8685 void btrfs_destroy_inode(struct inode *inode)
8686 {
8687         struct btrfs_ordered_extent *ordered;
8688         struct btrfs_root *root = BTRFS_I(inode)->root;
8689
8690         WARN_ON(!hlist_empty(&inode->i_dentry));
8691         WARN_ON(inode->i_data.nrpages);
8692         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8693         WARN_ON(BTRFS_I(inode)->reserved_extents);
8694         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8695         WARN_ON(BTRFS_I(inode)->csum_bytes);
8696         WARN_ON(BTRFS_I(inode)->defrag_bytes);
8697
8698         /*
8699          * This can happen where we create an inode, but somebody else also
8700          * created the same inode and we need to destroy the one we already
8701          * created.
8702          */
8703         if (!root)
8704                 goto free;
8705
8706         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8707                      &BTRFS_I(inode)->runtime_flags)) {
8708                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8709                         btrfs_ino(inode));
8710                 atomic_dec(&root->orphan_inodes);
8711         }
8712
8713         while (1) {
8714                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8715                 if (!ordered)
8716                         break;
8717                 else {
8718                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8719                                 ordered->file_offset, ordered->len);
8720                         btrfs_remove_ordered_extent(inode, ordered);
8721                         btrfs_put_ordered_extent(ordered);
8722                         btrfs_put_ordered_extent(ordered);
8723                 }
8724         }
8725         inode_tree_del(inode);
8726         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8727 free:
8728         call_rcu(&inode->i_rcu, btrfs_i_callback);
8729 }
8730
8731 int btrfs_drop_inode(struct inode *inode)
8732 {
8733         struct btrfs_root *root = BTRFS_I(inode)->root;
8734
8735         if (root == NULL)
8736                 return 1;
8737
8738         /* the snap/subvol tree is on deleting */
8739         if (btrfs_root_refs(&root->root_item) == 0)
8740                 return 1;
8741         else
8742                 return generic_drop_inode(inode);
8743 }
8744
8745 static void init_once(void *foo)
8746 {
8747         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8748
8749         inode_init_once(&ei->vfs_inode);
8750 }
8751
8752 void btrfs_destroy_cachep(void)
8753 {
8754         /*
8755          * Make sure all delayed rcu free inodes are flushed before we
8756          * destroy cache.
8757          */
8758         rcu_barrier();
8759         if (btrfs_inode_cachep)
8760                 kmem_cache_destroy(btrfs_inode_cachep);
8761         if (btrfs_trans_handle_cachep)
8762                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8763         if (btrfs_transaction_cachep)
8764                 kmem_cache_destroy(btrfs_transaction_cachep);
8765         if (btrfs_path_cachep)
8766                 kmem_cache_destroy(btrfs_path_cachep);
8767         if (btrfs_free_space_cachep)
8768                 kmem_cache_destroy(btrfs_free_space_cachep);
8769         if (btrfs_delalloc_work_cachep)
8770                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8771 }
8772
8773 int btrfs_init_cachep(void)
8774 {
8775         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8776                         sizeof(struct btrfs_inode), 0,
8777                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8778         if (!btrfs_inode_cachep)
8779                 goto fail;
8780
8781         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8782                         sizeof(struct btrfs_trans_handle), 0,
8783                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8784         if (!btrfs_trans_handle_cachep)
8785                 goto fail;
8786
8787         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8788                         sizeof(struct btrfs_transaction), 0,
8789                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8790         if (!btrfs_transaction_cachep)
8791                 goto fail;
8792
8793         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8794                         sizeof(struct btrfs_path), 0,
8795                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8796         if (!btrfs_path_cachep)
8797                 goto fail;
8798
8799         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8800                         sizeof(struct btrfs_free_space), 0,
8801                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8802         if (!btrfs_free_space_cachep)
8803                 goto fail;
8804
8805         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8806                         sizeof(struct btrfs_delalloc_work), 0,
8807                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8808                         NULL);
8809         if (!btrfs_delalloc_work_cachep)
8810                 goto fail;
8811
8812         return 0;
8813 fail:
8814         btrfs_destroy_cachep();
8815         return -ENOMEM;
8816 }
8817
8818 static int btrfs_getattr(struct vfsmount *mnt,
8819                          struct dentry *dentry, struct kstat *stat)
8820 {
8821         u64 delalloc_bytes;
8822         struct inode *inode = dentry->d_inode;
8823         u32 blocksize = inode->i_sb->s_blocksize;
8824
8825         generic_fillattr(inode, stat);
8826         stat->dev = BTRFS_I(inode)->root->anon_dev;
8827         stat->blksize = PAGE_CACHE_SIZE;
8828
8829         spin_lock(&BTRFS_I(inode)->lock);
8830         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8831         spin_unlock(&BTRFS_I(inode)->lock);
8832         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8833                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8834         return 0;
8835 }
8836
8837 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8838                            struct inode *new_dir, struct dentry *new_dentry)
8839 {
8840         struct btrfs_trans_handle *trans;
8841         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8842         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8843         struct inode *new_inode = new_dentry->d_inode;
8844         struct inode *old_inode = old_dentry->d_inode;
8845         struct timespec ctime = CURRENT_TIME;
8846         u64 index = 0;
8847         u64 root_objectid;
8848         int ret;
8849         u64 old_ino = btrfs_ino(old_inode);
8850
8851         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8852                 return -EPERM;
8853
8854         /* we only allow rename subvolume link between subvolumes */
8855         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8856                 return -EXDEV;
8857
8858         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8859             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8860                 return -ENOTEMPTY;
8861
8862         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8863             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8864                 return -ENOTEMPTY;
8865
8866
8867         /* check for collisions, even if the  name isn't there */
8868         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8869                              new_dentry->d_name.name,
8870                              new_dentry->d_name.len);
8871
8872         if (ret) {
8873                 if (ret == -EEXIST) {
8874                         /* we shouldn't get
8875                          * eexist without a new_inode */
8876                         if (WARN_ON(!new_inode)) {
8877                                 return ret;
8878                         }
8879                 } else {
8880                         /* maybe -EOVERFLOW */
8881                         return ret;
8882                 }
8883         }
8884         ret = 0;
8885
8886         /*
8887          * we're using rename to replace one file with another.  Start IO on it
8888          * now so  we don't add too much work to the end of the transaction
8889          */
8890         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size)
8891                 filemap_flush(old_inode->i_mapping);
8892
8893         /* close the racy window with snapshot create/destroy ioctl */
8894         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8895                 down_read(&root->fs_info->subvol_sem);
8896         /*
8897          * We want to reserve the absolute worst case amount of items.  So if
8898          * both inodes are subvols and we need to unlink them then that would
8899          * require 4 item modifications, but if they are both normal inodes it
8900          * would require 5 item modifications, so we'll assume their normal
8901          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8902          * should cover the worst case number of items we'll modify.
8903          */
8904         trans = btrfs_start_transaction(root, 11);
8905         if (IS_ERR(trans)) {
8906                 ret = PTR_ERR(trans);
8907                 goto out_notrans;
8908         }
8909
8910         if (dest != root)
8911                 btrfs_record_root_in_trans(trans, dest);
8912
8913         ret = btrfs_set_inode_index(new_dir, &index);
8914         if (ret)
8915                 goto out_fail;
8916
8917         BTRFS_I(old_inode)->dir_index = 0ULL;
8918         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8919                 /* force full log commit if subvolume involved. */
8920                 btrfs_set_log_full_commit(root->fs_info, trans);
8921         } else {
8922                 ret = btrfs_insert_inode_ref(trans, dest,
8923                                              new_dentry->d_name.name,
8924                                              new_dentry->d_name.len,
8925                                              old_ino,
8926                                              btrfs_ino(new_dir), index);
8927                 if (ret)
8928                         goto out_fail;
8929                 /*
8930                  * this is an ugly little race, but the rename is required
8931                  * to make sure that if we crash, the inode is either at the
8932                  * old name or the new one.  pinning the log transaction lets
8933                  * us make sure we don't allow a log commit to come in after
8934                  * we unlink the name but before we add the new name back in.
8935                  */
8936                 btrfs_pin_log_trans(root);
8937         }
8938
8939         inode_inc_iversion(old_dir);
8940         inode_inc_iversion(new_dir);
8941         inode_inc_iversion(old_inode);
8942         old_dir->i_ctime = old_dir->i_mtime = ctime;
8943         new_dir->i_ctime = new_dir->i_mtime = ctime;
8944         old_inode->i_ctime = ctime;
8945
8946         if (old_dentry->d_parent != new_dentry->d_parent)
8947                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8948
8949         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8950                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8951                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8952                                         old_dentry->d_name.name,
8953                                         old_dentry->d_name.len);
8954         } else {
8955                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8956                                         old_dentry->d_inode,
8957                                         old_dentry->d_name.name,
8958                                         old_dentry->d_name.len);
8959                 if (!ret)
8960                         ret = btrfs_update_inode(trans, root, old_inode);
8961         }
8962         if (ret) {
8963                 btrfs_abort_transaction(trans, root, ret);
8964                 goto out_fail;
8965         }
8966
8967         if (new_inode) {
8968                 inode_inc_iversion(new_inode);
8969                 new_inode->i_ctime = CURRENT_TIME;
8970                 if (unlikely(btrfs_ino(new_inode) ==
8971                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8972                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8973                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8974                                                 root_objectid,
8975                                                 new_dentry->d_name.name,
8976                                                 new_dentry->d_name.len);
8977                         BUG_ON(new_inode->i_nlink == 0);
8978                 } else {
8979                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8980                                                  new_dentry->d_inode,
8981                                                  new_dentry->d_name.name,
8982                                                  new_dentry->d_name.len);
8983                 }
8984                 if (!ret && new_inode->i_nlink == 0)
8985                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8986                 if (ret) {
8987                         btrfs_abort_transaction(trans, root, ret);
8988                         goto out_fail;
8989                 }
8990         }
8991
8992         ret = btrfs_add_link(trans, new_dir, old_inode,
8993                              new_dentry->d_name.name,
8994                              new_dentry->d_name.len, 0, index);
8995         if (ret) {
8996                 btrfs_abort_transaction(trans, root, ret);
8997                 goto out_fail;
8998         }
8999
9000         if (old_inode->i_nlink == 1)
9001                 BTRFS_I(old_inode)->dir_index = index;
9002
9003         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
9004                 struct dentry *parent = new_dentry->d_parent;
9005                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
9006                 btrfs_end_log_trans(root);
9007         }
9008 out_fail:
9009         btrfs_end_transaction(trans, root);
9010 out_notrans:
9011         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
9012                 up_read(&root->fs_info->subvol_sem);
9013
9014         return ret;
9015 }
9016
9017 static int btrfs_rename2(struct inode *old_dir, struct dentry *old_dentry,
9018                          struct inode *new_dir, struct dentry *new_dentry,
9019                          unsigned int flags)
9020 {
9021         if (flags & ~RENAME_NOREPLACE)
9022                 return -EINVAL;
9023
9024         return btrfs_rename(old_dir, old_dentry, new_dir, new_dentry);
9025 }
9026
9027 static void btrfs_run_delalloc_work(struct btrfs_work *work)
9028 {
9029         struct btrfs_delalloc_work *delalloc_work;
9030         struct inode *inode;
9031
9032         delalloc_work = container_of(work, struct btrfs_delalloc_work,
9033                                      work);
9034         inode = delalloc_work->inode;
9035         if (delalloc_work->wait) {
9036                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
9037         } else {
9038                 filemap_flush(inode->i_mapping);
9039                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
9040                              &BTRFS_I(inode)->runtime_flags))
9041                         filemap_flush(inode->i_mapping);
9042         }
9043
9044         if (delalloc_work->delay_iput)
9045                 btrfs_add_delayed_iput(inode);
9046         else
9047                 iput(inode);
9048         complete(&delalloc_work->completion);
9049 }
9050
9051 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
9052                                                     int wait, int delay_iput)
9053 {
9054         struct btrfs_delalloc_work *work;
9055
9056         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
9057         if (!work)
9058                 return NULL;
9059
9060         init_completion(&work->completion);
9061         INIT_LIST_HEAD(&work->list);
9062         work->inode = inode;
9063         work->wait = wait;
9064         work->delay_iput = delay_iput;
9065         WARN_ON_ONCE(!inode);
9066         btrfs_init_work(&work->work, btrfs_flush_delalloc_helper,
9067                         btrfs_run_delalloc_work, NULL, NULL);
9068
9069         return work;
9070 }
9071
9072 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
9073 {
9074         wait_for_completion(&work->completion);
9075         kmem_cache_free(btrfs_delalloc_work_cachep, work);
9076 }
9077
9078 /*
9079  * some fairly slow code that needs optimization. This walks the list
9080  * of all the inodes with pending delalloc and forces them to disk.
9081  */
9082 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput,
9083                                    int nr)
9084 {
9085         struct btrfs_inode *binode;
9086         struct inode *inode;
9087         struct btrfs_delalloc_work *work, *next;
9088         struct list_head works;
9089         struct list_head splice;
9090         int ret = 0;
9091
9092         INIT_LIST_HEAD(&works);
9093         INIT_LIST_HEAD(&splice);
9094
9095         mutex_lock(&root->delalloc_mutex);
9096         spin_lock(&root->delalloc_lock);
9097         list_splice_init(&root->delalloc_inodes, &splice);
9098         while (!list_empty(&splice)) {
9099                 binode = list_entry(splice.next, struct btrfs_inode,
9100                                     delalloc_inodes);
9101
9102                 list_move_tail(&binode->delalloc_inodes,
9103                                &root->delalloc_inodes);
9104                 inode = igrab(&binode->vfs_inode);
9105                 if (!inode) {
9106                         cond_resched_lock(&root->delalloc_lock);
9107                         continue;
9108                 }
9109                 spin_unlock(&root->delalloc_lock);
9110
9111                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
9112                 if (!work) {
9113                         if (delay_iput)
9114                                 btrfs_add_delayed_iput(inode);
9115                         else
9116                                 iput(inode);
9117                         ret = -ENOMEM;
9118                         goto out;
9119                 }
9120                 list_add_tail(&work->list, &works);
9121                 btrfs_queue_work(root->fs_info->flush_workers,
9122                                  &work->work);
9123                 ret++;
9124                 if (nr != -1 && ret >= nr)
9125                         goto out;
9126                 cond_resched();
9127                 spin_lock(&root->delalloc_lock);
9128         }
9129         spin_unlock(&root->delalloc_lock);
9130
9131 out:
9132         list_for_each_entry_safe(work, next, &works, list) {
9133                 list_del_init(&work->list);
9134                 btrfs_wait_and_free_delalloc_work(work);
9135         }
9136
9137         if (!list_empty_careful(&splice)) {
9138                 spin_lock(&root->delalloc_lock);
9139                 list_splice_tail(&splice, &root->delalloc_inodes);
9140                 spin_unlock(&root->delalloc_lock);
9141         }
9142         mutex_unlock(&root->delalloc_mutex);
9143         return ret;
9144 }
9145
9146 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
9147 {
9148         int ret;
9149
9150         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
9151                 return -EROFS;
9152
9153         ret = __start_delalloc_inodes(root, delay_iput, -1);
9154         if (ret > 0)
9155                 ret = 0;
9156         /*
9157          * the filemap_flush will queue IO into the worker threads, but
9158          * we have to make sure the IO is actually started and that
9159          * ordered extents get created before we return
9160          */
9161         atomic_inc(&root->fs_info->async_submit_draining);
9162         while (atomic_read(&root->fs_info->nr_async_submits) ||
9163               atomic_read(&root->fs_info->async_delalloc_pages)) {
9164                 wait_event(root->fs_info->async_submit_wait,
9165                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
9166                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
9167         }
9168         atomic_dec(&root->fs_info->async_submit_draining);
9169         return ret;
9170 }
9171
9172 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput,
9173                                int nr)
9174 {
9175         struct btrfs_root *root;
9176         struct list_head splice;
9177         int ret;
9178
9179         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
9180                 return -EROFS;
9181
9182         INIT_LIST_HEAD(&splice);
9183
9184         mutex_lock(&fs_info->delalloc_root_mutex);
9185         spin_lock(&fs_info->delalloc_root_lock);
9186         list_splice_init(&fs_info->delalloc_roots, &splice);
9187         while (!list_empty(&splice) && nr) {
9188                 root = list_first_entry(&splice, struct btrfs_root,
9189                                         delalloc_root);
9190                 root = btrfs_grab_fs_root(root);
9191                 BUG_ON(!root);
9192                 list_move_tail(&root->delalloc_root,
9193                                &fs_info->delalloc_roots);
9194                 spin_unlock(&fs_info->delalloc_root_lock);
9195
9196                 ret = __start_delalloc_inodes(root, delay_iput, nr);
9197                 btrfs_put_fs_root(root);
9198                 if (ret < 0)
9199                         goto out;
9200
9201                 if (nr != -1) {
9202                         nr -= ret;
9203                         WARN_ON(nr < 0);
9204                 }
9205                 spin_lock(&fs_info->delalloc_root_lock);
9206         }
9207         spin_unlock(&fs_info->delalloc_root_lock);
9208
9209         ret = 0;
9210         atomic_inc(&fs_info->async_submit_draining);
9211         while (atomic_read(&fs_info->nr_async_submits) ||
9212               atomic_read(&fs_info->async_delalloc_pages)) {
9213                 wait_event(fs_info->async_submit_wait,
9214                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
9215                     atomic_read(&fs_info->async_delalloc_pages) == 0));
9216         }
9217         atomic_dec(&fs_info->async_submit_draining);
9218 out:
9219         if (!list_empty_careful(&splice)) {
9220                 spin_lock(&fs_info->delalloc_root_lock);
9221                 list_splice_tail(&splice, &fs_info->delalloc_roots);
9222                 spin_unlock(&fs_info->delalloc_root_lock);
9223         }
9224         mutex_unlock(&fs_info->delalloc_root_mutex);
9225         return ret;
9226 }
9227
9228 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
9229                          const char *symname)
9230 {
9231         struct btrfs_trans_handle *trans;
9232         struct btrfs_root *root = BTRFS_I(dir)->root;
9233         struct btrfs_path *path;
9234         struct btrfs_key key;
9235         struct inode *inode = NULL;
9236         int err;
9237         int drop_inode = 0;
9238         u64 objectid;
9239         u64 index = 0;
9240         int name_len;
9241         int datasize;
9242         unsigned long ptr;
9243         struct btrfs_file_extent_item *ei;
9244         struct extent_buffer *leaf;
9245
9246         name_len = strlen(symname);
9247         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
9248                 return -ENAMETOOLONG;
9249
9250         /*
9251          * 2 items for inode item and ref
9252          * 2 items for dir items
9253          * 1 item for xattr if selinux is on
9254          */
9255         trans = btrfs_start_transaction(root, 5);
9256         if (IS_ERR(trans))
9257                 return PTR_ERR(trans);
9258
9259         err = btrfs_find_free_ino(root, &objectid);
9260         if (err)
9261                 goto out_unlock;
9262
9263         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
9264                                 dentry->d_name.len, btrfs_ino(dir), objectid,
9265                                 S_IFLNK|S_IRWXUGO, &index);
9266         if (IS_ERR(inode)) {
9267                 err = PTR_ERR(inode);
9268                 goto out_unlock;
9269         }
9270
9271         /*
9272         * If the active LSM wants to access the inode during
9273         * d_instantiate it needs these. Smack checks to see
9274         * if the filesystem supports xattrs by looking at the
9275         * ops vector.
9276         */
9277         inode->i_fop = &btrfs_file_operations;
9278         inode->i_op = &btrfs_file_inode_operations;
9279         inode->i_mapping->a_ops = &btrfs_aops;
9280         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9281
9282         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
9283         if (err)
9284                 goto out_unlock_inode;
9285
9286         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
9287         if (err)
9288                 goto out_unlock_inode;
9289
9290         path = btrfs_alloc_path();
9291         if (!path) {
9292                 err = -ENOMEM;
9293                 goto out_unlock_inode;
9294         }
9295         key.objectid = btrfs_ino(inode);
9296         key.offset = 0;
9297         key.type = BTRFS_EXTENT_DATA_KEY;
9298         datasize = btrfs_file_extent_calc_inline_size(name_len);
9299         err = btrfs_insert_empty_item(trans, root, path, &key,
9300                                       datasize);
9301         if (err) {
9302                 btrfs_free_path(path);
9303                 goto out_unlock_inode;
9304         }
9305         leaf = path->nodes[0];
9306         ei = btrfs_item_ptr(leaf, path->slots[0],
9307                             struct btrfs_file_extent_item);
9308         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
9309         btrfs_set_file_extent_type(leaf, ei,
9310                                    BTRFS_FILE_EXTENT_INLINE);
9311         btrfs_set_file_extent_encryption(leaf, ei, 0);
9312         btrfs_set_file_extent_compression(leaf, ei, 0);
9313         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
9314         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
9315
9316         ptr = btrfs_file_extent_inline_start(ei);
9317         write_extent_buffer(leaf, symname, ptr, name_len);
9318         btrfs_mark_buffer_dirty(leaf);
9319         btrfs_free_path(path);
9320
9321         inode->i_op = &btrfs_symlink_inode_operations;
9322         inode->i_mapping->a_ops = &btrfs_symlink_aops;
9323         inode_set_bytes(inode, name_len);
9324         btrfs_i_size_write(inode, name_len);
9325         err = btrfs_update_inode(trans, root, inode);
9326         if (err) {
9327                 drop_inode = 1;
9328                 goto out_unlock_inode;
9329         }
9330
9331         unlock_new_inode(inode);
9332         d_instantiate(dentry, inode);
9333
9334 out_unlock:
9335         btrfs_end_transaction(trans, root);
9336         if (drop_inode) {
9337                 inode_dec_link_count(inode);
9338                 iput(inode);
9339         }
9340         btrfs_btree_balance_dirty(root);
9341         return err;
9342
9343 out_unlock_inode:
9344         drop_inode = 1;
9345         unlock_new_inode(inode);
9346         goto out_unlock;
9347 }
9348
9349 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
9350                                        u64 start, u64 num_bytes, u64 min_size,
9351                                        loff_t actual_len, u64 *alloc_hint,
9352                                        struct btrfs_trans_handle *trans)
9353 {
9354         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
9355         struct extent_map *em;
9356         struct btrfs_root *root = BTRFS_I(inode)->root;
9357         struct btrfs_key ins;
9358         u64 cur_offset = start;
9359         u64 i_size;
9360         u64 cur_bytes;
9361         int ret = 0;
9362         bool own_trans = true;
9363
9364         if (trans)
9365                 own_trans = false;
9366         while (num_bytes > 0) {
9367                 if (own_trans) {
9368                         trans = btrfs_start_transaction(root, 3);
9369                         if (IS_ERR(trans)) {
9370                                 ret = PTR_ERR(trans);
9371                                 break;
9372                         }
9373                 }
9374
9375                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
9376                 cur_bytes = max(cur_bytes, min_size);
9377                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
9378                                            *alloc_hint, &ins, 1, 0);
9379                 if (ret) {
9380                         if (own_trans)
9381                                 btrfs_end_transaction(trans, root);
9382                         break;
9383                 }
9384
9385                 ret = insert_reserved_file_extent(trans, inode,
9386                                                   cur_offset, ins.objectid,
9387                                                   ins.offset, ins.offset,
9388                                                   ins.offset, 0, 0, 0,
9389                                                   BTRFS_FILE_EXTENT_PREALLOC);
9390                 if (ret) {
9391                         btrfs_free_reserved_extent(root, ins.objectid,
9392                                                    ins.offset, 0);
9393                         btrfs_abort_transaction(trans, root, ret);
9394                         if (own_trans)
9395                                 btrfs_end_transaction(trans, root);
9396                         break;
9397                 }
9398                 btrfs_drop_extent_cache(inode, cur_offset,
9399                                         cur_offset + ins.offset -1, 0);
9400
9401                 em = alloc_extent_map();
9402                 if (!em) {
9403                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
9404                                 &BTRFS_I(inode)->runtime_flags);
9405                         goto next;
9406                 }
9407
9408                 em->start = cur_offset;
9409                 em->orig_start = cur_offset;
9410                 em->len = ins.offset;
9411                 em->block_start = ins.objectid;
9412                 em->block_len = ins.offset;
9413                 em->orig_block_len = ins.offset;
9414                 em->ram_bytes = ins.offset;
9415                 em->bdev = root->fs_info->fs_devices->latest_bdev;
9416                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
9417                 em->generation = trans->transid;
9418
9419                 while (1) {
9420                         write_lock(&em_tree->lock);
9421                         ret = add_extent_mapping(em_tree, em, 1);
9422                         write_unlock(&em_tree->lock);
9423                         if (ret != -EEXIST)
9424                                 break;
9425                         btrfs_drop_extent_cache(inode, cur_offset,
9426                                                 cur_offset + ins.offset - 1,
9427                                                 0);
9428                 }
9429                 free_extent_map(em);
9430 next:
9431                 num_bytes -= ins.offset;
9432                 cur_offset += ins.offset;
9433                 *alloc_hint = ins.objectid + ins.offset;
9434
9435                 inode_inc_iversion(inode);
9436                 inode->i_ctime = CURRENT_TIME;
9437                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
9438                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
9439                     (actual_len > inode->i_size) &&
9440                     (cur_offset > inode->i_size)) {
9441                         if (cur_offset > actual_len)
9442                                 i_size = actual_len;
9443                         else
9444                                 i_size = cur_offset;
9445                         i_size_write(inode, i_size);
9446                         btrfs_ordered_update_i_size(inode, i_size, NULL);
9447                 }
9448
9449                 ret = btrfs_update_inode(trans, root, inode);
9450
9451                 if (ret) {
9452                         btrfs_abort_transaction(trans, root, ret);
9453                         if (own_trans)
9454                                 btrfs_end_transaction(trans, root);
9455                         break;
9456                 }
9457
9458                 if (own_trans)
9459                         btrfs_end_transaction(trans, root);
9460         }
9461         return ret;
9462 }
9463
9464 int btrfs_prealloc_file_range(struct inode *inode, int mode,
9465                               u64 start, u64 num_bytes, u64 min_size,
9466                               loff_t actual_len, u64 *alloc_hint)
9467 {
9468         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9469                                            min_size, actual_len, alloc_hint,
9470                                            NULL);
9471 }
9472
9473 int btrfs_prealloc_file_range_trans(struct inode *inode,
9474                                     struct btrfs_trans_handle *trans, int mode,
9475                                     u64 start, u64 num_bytes, u64 min_size,
9476                                     loff_t actual_len, u64 *alloc_hint)
9477 {
9478         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
9479                                            min_size, actual_len, alloc_hint, trans);
9480 }
9481
9482 static int btrfs_set_page_dirty(struct page *page)
9483 {
9484         return __set_page_dirty_nobuffers(page);
9485 }
9486
9487 static int btrfs_permission(struct inode *inode, int mask)
9488 {
9489         struct btrfs_root *root = BTRFS_I(inode)->root;
9490         umode_t mode = inode->i_mode;
9491
9492         if (mask & MAY_WRITE &&
9493             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
9494                 if (btrfs_root_readonly(root))
9495                         return -EROFS;
9496                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
9497                         return -EACCES;
9498         }
9499         return generic_permission(inode, mask);
9500 }
9501
9502 static int btrfs_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
9503 {
9504         struct btrfs_trans_handle *trans;
9505         struct btrfs_root *root = BTRFS_I(dir)->root;
9506         struct inode *inode = NULL;
9507         u64 objectid;
9508         u64 index;
9509         int ret = 0;
9510
9511         /*
9512          * 5 units required for adding orphan entry
9513          */
9514         trans = btrfs_start_transaction(root, 5);
9515         if (IS_ERR(trans))
9516                 return PTR_ERR(trans);
9517
9518         ret = btrfs_find_free_ino(root, &objectid);
9519         if (ret)
9520                 goto out;
9521
9522         inode = btrfs_new_inode(trans, root, dir, NULL, 0,
9523                                 btrfs_ino(dir), objectid, mode, &index);
9524         if (IS_ERR(inode)) {
9525                 ret = PTR_ERR(inode);
9526                 inode = NULL;
9527                 goto out;
9528         }
9529
9530         inode->i_fop = &btrfs_file_operations;
9531         inode->i_op = &btrfs_file_inode_operations;
9532
9533         inode->i_mapping->a_ops = &btrfs_aops;
9534         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
9535
9536         ret = btrfs_init_inode_security(trans, inode, dir, NULL);
9537         if (ret)
9538                 goto out_inode;
9539
9540         ret = btrfs_update_inode(trans, root, inode);
9541         if (ret)
9542                 goto out_inode;
9543         ret = btrfs_orphan_add(trans, inode);
9544         if (ret)
9545                 goto out_inode;
9546
9547         /*
9548          * We set number of links to 0 in btrfs_new_inode(), and here we set
9549          * it to 1 because d_tmpfile() will issue a warning if the count is 0,
9550          * through:
9551          *
9552          *    d_tmpfile() -> inode_dec_link_count() -> drop_nlink()
9553          */
9554         set_nlink(inode, 1);
9555         unlock_new_inode(inode);
9556         d_tmpfile(dentry, inode);
9557         mark_inode_dirty(inode);
9558
9559 out:
9560         btrfs_end_transaction(trans, root);
9561         if (ret)
9562                 iput(inode);
9563         btrfs_balance_delayed_items(root);
9564         btrfs_btree_balance_dirty(root);
9565         return ret;
9566
9567 out_inode:
9568         unlock_new_inode(inode);
9569         goto out;
9570
9571 }
9572
9573 /* Inspired by filemap_check_errors() */
9574 int btrfs_inode_check_errors(struct inode *inode)
9575 {
9576         int ret = 0;
9577
9578         if (test_bit(AS_ENOSPC, &inode->i_mapping->flags) &&
9579             test_and_clear_bit(AS_ENOSPC, &inode->i_mapping->flags))
9580                 ret = -ENOSPC;
9581         if (test_bit(AS_EIO, &inode->i_mapping->flags) &&
9582             test_and_clear_bit(AS_EIO, &inode->i_mapping->flags))
9583                 ret = -EIO;
9584
9585         return ret;
9586 }
9587
9588 static const struct inode_operations btrfs_dir_inode_operations = {
9589         .getattr        = btrfs_getattr,
9590         .lookup         = btrfs_lookup,
9591         .create         = btrfs_create,
9592         .unlink         = btrfs_unlink,
9593         .link           = btrfs_link,
9594         .mkdir          = btrfs_mkdir,
9595         .rmdir          = btrfs_rmdir,
9596         .rename2        = btrfs_rename2,
9597         .symlink        = btrfs_symlink,
9598         .setattr        = btrfs_setattr,
9599         .mknod          = btrfs_mknod,
9600         .setxattr       = btrfs_setxattr,
9601         .getxattr       = btrfs_getxattr,
9602         .listxattr      = btrfs_listxattr,
9603         .removexattr    = btrfs_removexattr,
9604         .permission     = btrfs_permission,
9605         .get_acl        = btrfs_get_acl,
9606         .set_acl        = btrfs_set_acl,
9607         .update_time    = btrfs_update_time,
9608         .tmpfile        = btrfs_tmpfile,
9609 };
9610 static const struct inode_operations btrfs_dir_ro_inode_operations = {
9611         .lookup         = btrfs_lookup,
9612         .permission     = btrfs_permission,
9613         .get_acl        = btrfs_get_acl,
9614         .set_acl        = btrfs_set_acl,
9615         .update_time    = btrfs_update_time,
9616 };
9617
9618 static const struct file_operations btrfs_dir_file_operations = {
9619         .llseek         = generic_file_llseek,
9620         .read           = generic_read_dir,
9621         .iterate        = btrfs_real_readdir,
9622         .unlocked_ioctl = btrfs_ioctl,
9623 #ifdef CONFIG_COMPAT
9624         .compat_ioctl   = btrfs_ioctl,
9625 #endif
9626         .release        = btrfs_release_file,
9627         .fsync          = btrfs_sync_file,
9628 };
9629
9630 static struct extent_io_ops btrfs_extent_io_ops = {
9631         .fill_delalloc = run_delalloc_range,
9632         .submit_bio_hook = btrfs_submit_bio_hook,
9633         .merge_bio_hook = btrfs_merge_bio_hook,
9634         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
9635         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
9636         .writepage_start_hook = btrfs_writepage_start_hook,
9637         .set_bit_hook = btrfs_set_bit_hook,
9638         .clear_bit_hook = btrfs_clear_bit_hook,
9639         .merge_extent_hook = btrfs_merge_extent_hook,
9640         .split_extent_hook = btrfs_split_extent_hook,
9641 };
9642
9643 /*
9644  * btrfs doesn't support the bmap operation because swapfiles
9645  * use bmap to make a mapping of extents in the file.  They assume
9646  * these extents won't change over the life of the file and they
9647  * use the bmap result to do IO directly to the drive.
9648  *
9649  * the btrfs bmap call would return logical addresses that aren't
9650  * suitable for IO and they also will change frequently as COW
9651  * operations happen.  So, swapfile + btrfs == corruption.
9652  *
9653  * For now we're avoiding this by dropping bmap.
9654  */
9655 static const struct address_space_operations btrfs_aops = {
9656         .readpage       = btrfs_readpage,
9657         .writepage      = btrfs_writepage,
9658         .writepages     = btrfs_writepages,
9659         .readpages      = btrfs_readpages,
9660         .direct_IO      = btrfs_direct_IO,
9661         .invalidatepage = btrfs_invalidatepage,
9662         .releasepage    = btrfs_releasepage,
9663         .set_page_dirty = btrfs_set_page_dirty,
9664         .error_remove_page = generic_error_remove_page,
9665 };
9666
9667 static const struct address_space_operations btrfs_symlink_aops = {
9668         .readpage       = btrfs_readpage,
9669         .writepage      = btrfs_writepage,
9670         .invalidatepage = btrfs_invalidatepage,
9671         .releasepage    = btrfs_releasepage,
9672 };
9673
9674 static const struct inode_operations btrfs_file_inode_operations = {
9675         .getattr        = btrfs_getattr,
9676         .setattr        = btrfs_setattr,
9677         .setxattr       = btrfs_setxattr,
9678         .getxattr       = btrfs_getxattr,
9679         .listxattr      = btrfs_listxattr,
9680         .removexattr    = btrfs_removexattr,
9681         .permission     = btrfs_permission,
9682         .fiemap         = btrfs_fiemap,
9683         .get_acl        = btrfs_get_acl,
9684         .set_acl        = btrfs_set_acl,
9685         .update_time    = btrfs_update_time,
9686 };
9687 static const struct inode_operations btrfs_special_inode_operations = {
9688         .getattr        = btrfs_getattr,
9689         .setattr        = btrfs_setattr,
9690         .permission     = btrfs_permission,
9691         .setxattr       = btrfs_setxattr,
9692         .getxattr       = btrfs_getxattr,
9693         .listxattr      = btrfs_listxattr,
9694         .removexattr    = btrfs_removexattr,
9695         .get_acl        = btrfs_get_acl,
9696         .set_acl        = btrfs_set_acl,
9697         .update_time    = btrfs_update_time,
9698 };
9699 static const struct inode_operations btrfs_symlink_inode_operations = {
9700         .readlink       = generic_readlink,
9701         .follow_link    = page_follow_link_light,
9702         .put_link       = page_put_link,
9703         .getattr        = btrfs_getattr,
9704         .setattr        = btrfs_setattr,
9705         .permission     = btrfs_permission,
9706         .setxattr       = btrfs_setxattr,
9707         .getxattr       = btrfs_getxattr,
9708         .listxattr      = btrfs_listxattr,
9709         .removexattr    = btrfs_removexattr,
9710         .update_time    = btrfs_update_time,
9711 };
9712
9713 const struct dentry_operations btrfs_dentry_operations = {
9714         .d_delete       = btrfs_dentry_delete,
9715         .d_release      = btrfs_dentry_release,
9716 };