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/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
  99                                            u64 len, u64 orig_start,
 100                                            u64 block_start, u64 block_len,
 101                                            u64 orig_block_len, int type);
 102
 103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 104                                      struct inode *inode,  struct inode *dir,
 105                                      const struct qstr *qstr)
 106 {
 107         int err;
 108
 109         err = btrfs_init_acl(trans, inode, dir);
 110         if (!err)
 111                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 112         return err;
 113 }
 114
 115 /*
 116  * this does all the hard work for inserting an inline extent into
 117  * the btree.  The caller should have done a btrfs_drop_extents so that
 118  * no overlapping inline items exist in the btree
 119  */
 120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 121                                 struct btrfs_root *root, struct inode *inode,
 122                                 u64 start, size_t size, size_t compressed_size,
 123                                 int compress_type,
 124                                 struct page **compressed_pages)
 125 {
 126         struct btrfs_key key;
 127         struct btrfs_path *path;
 128         struct extent_buffer *leaf;
 129         struct page *page = NULL;
 130         char *kaddr;
 131         unsigned long ptr;
 132         struct btrfs_file_extent_item *ei;
 133         int err = 0;
 134         int ret;
 135         size_t cur_size = size;
 136         size_t datasize;
 137         unsigned long offset;
 138
 139         if (compressed_size && compressed_pages)
 140                 cur_size = compressed_size;
 141
 142         path = btrfs_alloc_path();
 143         if (!path)
 144                 return -ENOMEM;
 145
 146         path->leave_spinning = 1;
 147
 148         key.objectid = btrfs_ino(inode);
 149         key.offset = start;
 150         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 151         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 152
 153         inode_add_bytes(inode, size);
 154         ret = btrfs_insert_empty_item(trans, root, path, &key,
 155                                       datasize);
 156         if (ret) {
 157                 err = ret;
 158                 goto fail;
 159         }
 160         leaf = path->nodes[0];
 161         ei = btrfs_item_ptr(leaf, path->slots[0],
 162                             struct btrfs_file_extent_item);
 163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 164         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 165         btrfs_set_file_extent_encryption(leaf, ei, 0);
 166         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 167         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 168         ptr = btrfs_file_extent_inline_start(ei);
 169
 170         if (compress_type != BTRFS_COMPRESS_NONE) {
 171                 struct page *cpage;
 172                 int i = 0;
 173                 while (compressed_size > 0) {
 174                         cpage = compressed_pages[i];
 175                         cur_size = min_t(unsigned long, compressed_size,
 176                                        PAGE_CACHE_SIZE);
 177
 178                         kaddr = kmap_atomic(cpage);
 179                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 180                         kunmap_atomic(kaddr);
 181
 182                         i++;
 183                         ptr += cur_size;
 184                         compressed_size -= cur_size;
 185                 }
 186                 btrfs_set_file_extent_compression(leaf, ei,
 187                                                   compress_type);
 188         } else {
 189                 page = find_get_page(inode->i_mapping,
 190                                      start >> PAGE_CACHE_SHIFT);
 191                 btrfs_set_file_extent_compression(leaf, ei, 0);
 192                 kaddr = kmap_atomic(page);
 193                 offset = start & (PAGE_CACHE_SIZE - 1);
 194                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 195                 kunmap_atomic(kaddr);
 196                 page_cache_release(page);
 197         }
 198         btrfs_mark_buffer_dirty(leaf);
 199         btrfs_free_path(path);
 200
 201         /*
 202          * we're an inline extent, so nobody can
 203          * extend the file past i_size without locking
 204          * a page we already have locked.
 205          *
 206          * We must do any isize and inode updates
 207          * before we unlock the pages.  Otherwise we
 208          * could end up racing with unlink.
 209          */
 210         BTRFS_I(inode)->disk_i_size = inode->i_size;
 211         ret = btrfs_update_inode(trans, root, inode);
 212
 213         return ret;
 214 fail:
 215         btrfs_free_path(path);
 216         return err;
 217 }
 218
 219
 220 /*
 221  * conditionally insert an inline extent into the file.  This
 222  * does the checks required to make sure the data is small enough
 223  * to fit as an inline extent.
 224  */
 225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 226                                  struct btrfs_root *root,
 227                                  struct inode *inode, u64 start, u64 end,
 228                                  size_t compressed_size, int compress_type,
 229                                  struct page **compressed_pages)
 230 {
 231         u64 isize = i_size_read(inode);
 232         u64 actual_end = min(end + 1, isize);
 233         u64 inline_len = actual_end - start;
 234         u64 aligned_end = (end + root->sectorsize - 1) &
 235                         ~((u64)root->sectorsize - 1);
 236         u64 data_len = inline_len;
 237         int ret;
 238
 239         if (compressed_size)
 240                 data_len = compressed_size;
 241
 242         if (start > 0 ||
 243             actual_end >= PAGE_CACHE_SIZE ||
 244             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 245             (!compressed_size &&
 246             (actual_end & (root->sectorsize - 1)) == 0) ||
 247             end + 1 < isize ||
 248             data_len > root->fs_info->max_inline) {
 249                 return 1;
 250         }
 251
 252         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 253         if (ret)
 254                 return ret;
 255
 256         if (isize > actual_end)
 257                 inline_len = min_t(u64, isize, actual_end);
 258         ret = insert_inline_extent(trans, root, inode, start,
 259                                    inline_len, compressed_size,
 260                                    compress_type, compressed_pages);
 261         if (ret && ret != -ENOSPC) {
 262                 btrfs_abort_transaction(trans, root, ret);
 263                 return ret;
 264         } else if (ret == -ENOSPC) {
 265                 return 1;
 266         }
 267
 268         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 269         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 270         return 0;
 271 }
 272
 273 struct async_extent {
 274         u64 start;
 275         u64 ram_size;
 276         u64 compressed_size;
 277         struct page **pages;
 278         unsigned long nr_pages;
 279         int compress_type;
 280         struct list_head list;
 281 };
 282
 283 struct async_cow {
 284         struct inode *inode;
 285         struct btrfs_root *root;
 286         struct page *locked_page;
 287         u64 start;
 288         u64 end;
 289         struct list_head extents;
 290         struct btrfs_work work;
 291 };
 292
 293 static noinline int add_async_extent(struct async_cow *cow,
 294                                      u64 start, u64 ram_size,
 295                                      u64 compressed_size,
 296                                      struct page **pages,
 297                                      unsigned long nr_pages,
 298                                      int compress_type)
 299 {
 300         struct async_extent *async_extent;
 301
 302         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 303         BUG_ON(!async_extent); /* -ENOMEM */
 304         async_extent->start = start;
 305         async_extent->ram_size = ram_size;
 306         async_extent->compressed_size = compressed_size;
 307         async_extent->pages = pages;
 308         async_extent->nr_pages = nr_pages;
 309         async_extent->compress_type = compress_type;
 310         list_add_tail(&async_extent->list, &cow->extents);
 311         return 0;
 312 }
 313
 314 /*
 315  * we create compressed extents in two phases.  The first
 316  * phase compresses a range of pages that have already been
 317  * locked (both pages and state bits are locked).
 318  *
 319  * This is done inside an ordered work queue, and the compression
 320  * is spread across many cpus.  The actual IO submission is step
 321  * two, and the ordered work queue takes care of making sure that
 322  * happens in the same order things were put onto the queue by
 323  * writepages and friends.
 324  *
 325  * If this code finds it can't get good compression, it puts an
 326  * entry onto the work queue to write the uncompressed bytes.  This
 327  * makes sure that both compressed inodes and uncompressed inodes
 328  * are written in the same order that the flusher thread sent them
 329  * down.
 330  */
 331 static noinline int compress_file_range(struct inode *inode,
 332                                         struct page *locked_page,
 333                                         u64 start, u64 end,
 334                                         struct async_cow *async_cow,
 335                                         int *num_added)
 336 {
 337         struct btrfs_root *root = BTRFS_I(inode)->root;
 338         struct btrfs_trans_handle *trans;
 339         u64 num_bytes;
 340         u64 blocksize = root->sectorsize;
 341         u64 actual_end;
 342         u64 isize = i_size_read(inode);
 343         int ret = 0;
 344         struct page **pages = NULL;
 345         unsigned long nr_pages;
 346         unsigned long nr_pages_ret = 0;
 347         unsigned long total_compressed = 0;
 348         unsigned long total_in = 0;
 349         unsigned long max_compressed = 128 * 1024;
 350         unsigned long max_uncompressed = 128 * 1024;
 351         int i;
 352         int will_compress;
 353         int compress_type = root->fs_info->compress_type;
 354
 355         /* if this is a small write inside eof, kick off a defrag */
 356         if ((end - start + 1) < 16 * 1024 &&
 357             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 358                 btrfs_add_inode_defrag(NULL, inode);
 359
 360         actual_end = min_t(u64, isize, end + 1);
 361 again:
 362         will_compress = 0;
 363         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 364         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 365
 366         /*
 367          * we don't want to send crud past the end of i_size through
 368          * compression, that's just a waste of CPU time.  So, if the
 369          * end of the file is before the start of our current
 370          * requested range of bytes, we bail out to the uncompressed
 371          * cleanup code that can deal with all of this.
 372          *
 373          * It isn't really the fastest way to fix things, but this is a
 374          * very uncommon corner.
 375          */
 376         if (actual_end <= start)
 377                 goto cleanup_and_bail_uncompressed;
 378
 379         total_compressed = actual_end - start;
 380
 381         /* we want to make sure that amount of ram required to uncompress
 382          * an extent is reasonable, so we limit the total size in ram
 383          * of a compressed extent to 128k.  This is a crucial number
 384          * because it also controls how easily we can spread reads across
 385          * cpus for decompression.
 386          *
 387          * We also want to make sure the amount of IO required to do
 388          * a random read is reasonably small, so we limit the size of
 389          * a compressed extent to 128k.
 390          */
 391         total_compressed = min(total_compressed, max_uncompressed);
 392         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 393         num_bytes = max(blocksize,  num_bytes);
 394         total_in = 0;
 395         ret = 0;
 396
 397         /*
 398          * we do compression for mount -o compress and when the
 399          * inode has not been flagged as nocompress.  This flag can
 400          * change at any time if we discover bad compression ratios.
 401          */
 402         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 403             (btrfs_test_opt(root, COMPRESS) ||
 404              (BTRFS_I(inode)->force_compress) ||
 405              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 406                 WARN_ON(pages);
 407                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 408                 if (!pages) {
 409                         /* just bail out to the uncompressed code */
 410                         goto cont;
 411                 }
 412
 413                 if (BTRFS_I(inode)->force_compress)
 414                         compress_type = BTRFS_I(inode)->force_compress;
 415
 416                 ret = btrfs_compress_pages(compress_type,
 417                                            inode->i_mapping, start,
 418                                            total_compressed, pages,
 419                                            nr_pages, &nr_pages_ret,
 420                                            &total_in,
 421                                            &total_compressed,
 422                                            max_compressed);
 423
 424                 if (!ret) {
 425                         unsigned long offset = total_compressed &
 426                                 (PAGE_CACHE_SIZE - 1);
 427                         struct page *page = pages[nr_pages_ret - 1];
 428                         char *kaddr;
 429
 430                         /* zero the tail end of the last page, we might be
 431                          * sending it down to disk
 432                          */
 433                         if (offset) {
 434                                 kaddr = kmap_atomic(page);
 435                                 memset(kaddr + offset, 0,
 436                                        PAGE_CACHE_SIZE - offset);
 437                                 kunmap_atomic(kaddr);
 438                         }
 439                         will_compress = 1;
 440                 }
 441         }
 442 cont:
 443         if (start == 0) {
 444                 trans = btrfs_join_transaction(root);
 445                 if (IS_ERR(trans)) {
 446                         ret = PTR_ERR(trans);
 447                         trans = NULL;
 448                         goto cleanup_and_out;
 449                 }
 450                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 451
 452                 /* lets try to make an inline extent */
 453                 if (ret || total_in < (actual_end - start)) {
 454                         /* we didn't compress the entire range, try
 455                          * to make an uncompressed inline extent.
 456                          */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end, 0, 0, NULL);
 459                 } else {
 460                         /* try making a compressed inline extent */
 461                         ret = cow_file_range_inline(trans, root, inode,
 462                                                     start, end,
 463                                                     total_compressed,
 464                                                     compress_type, pages);
 465                 }
 466                 if (ret <= 0) {
 467                         /*
 468                          * inline extent creation worked or returned error,
 469                          * we don't need to create any more async work items.
 470                          * Unlock and free up our temp pages.
 471                          */
 472                         extent_clear_unlock_delalloc(inode,
 473                              &BTRFS_I(inode)->io_tree,
 474                              start, end, NULL,
 475                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 476                              EXTENT_CLEAR_DELALLOC |
 477                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 478
 479                         btrfs_end_transaction(trans, root);
 480                         goto free_pages_out;
 481                 }
 482                 btrfs_end_transaction(trans, root);
 483         }
 484
 485         if (will_compress) {
 486                 /*
 487                  * we aren't doing an inline extent round the compressed size
 488                  * up to a block size boundary so the allocator does sane
 489                  * things
 490                  */
 491                 total_compressed = (total_compressed + blocksize - 1) &
 492                         ~(blocksize - 1);
 493
 494                 /*
 495                  * one last check to make sure the compression is really a
 496                  * win, compare the page count read with the blocks on disk
 497                  */
 498                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 499                         ~(PAGE_CACHE_SIZE - 1);
 500                 if (total_compressed >= total_in) {
 501                         will_compress = 0;
 502                 } else {
 503                         num_bytes = total_in;
 504                 }
 505         }
 506         if (!will_compress && pages) {
 507                 /*
 508                  * the compression code ran but failed to make things smaller,
 509                  * free any pages it allocated and our page pointer array
 510                  */
 511                 for (i = 0; i < nr_pages_ret; i++) {
 512                         WARN_ON(pages[i]->mapping);
 513                         page_cache_release(pages[i]);
 514                 }
 515                 kfree(pages);
 516                 pages = NULL;
 517                 total_compressed = 0;
 518                 nr_pages_ret = 0;
 519
 520                 /* flag the file so we don't compress in the future */
 521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 522                     !(BTRFS_I(inode)->force_compress)) {
 523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 524                 }
 525         }
 526         if (will_compress) {
 527                 *num_added += 1;
 528
 529                 /* the async work queues will take care of doing actual
 530                  * allocation on disk for these compressed pages,
 531                  * and will submit them to the elevator.
 532                  */
 533                 add_async_extent(async_cow, start, num_bytes,
 534                                  total_compressed, pages, nr_pages_ret,
 535                                  compress_type);
 536
 537                 if (start + num_bytes < end) {
 538                         start += num_bytes;
 539                         pages = NULL;
 540                         cond_resched();
 541                         goto again;
 542                 }
 543         } else {
 544 cleanup_and_bail_uncompressed:
 545                 /*
 546                  * No compression, but we still need to write the pages in
 547                  * the file we've been given so far.  redirty the locked
 548                  * page if it corresponds to our extent and set things up
 549                  * for the async work queue to run cow_file_range to do
 550                  * the normal delalloc dance
 551                  */
 552                 if (page_offset(locked_page) >= start &&
 553                     page_offset(locked_page) <= end) {
 554                         __set_page_dirty_nobuffers(locked_page);
 555                         /* unlocked later on in the async handlers */
 556                 }
 557                 add_async_extent(async_cow, start, end - start + 1,
 558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 559                 *num_added += 1;
 560         }
 561
 562 out:
 563         return ret;
 564
 565 free_pages_out:
 566         for (i = 0; i < nr_pages_ret; i++) {
 567                 WARN_ON(pages[i]->mapping);
 568                 page_cache_release(pages[i]);
 569         }
 570         kfree(pages);
 571
 572         goto out;
 573
 574 cleanup_and_out:
 575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 576                                      start, end, NULL,
 577                                      EXTENT_CLEAR_UNLOCK_PAGE |
 578                                      EXTENT_CLEAR_DIRTY |
 579                                      EXTENT_CLEAR_DELALLOC |
 580                                      EXTENT_SET_WRITEBACK |
 581                                      EXTENT_END_WRITEBACK);
 582         if (!trans || IS_ERR(trans))
 583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 584         else
 585                 btrfs_abort_transaction(trans, root, ret);
 586         goto free_pages_out;
 587 }
 588
 589 /*
 590  * phase two of compressed writeback.  This is the ordered portion
 591  * of the code, which only gets called in the order the work was
 592  * queued.  We walk all the async extents created by compress_file_range
 593  * and send them down to the disk.
 594  */
 595 static noinline int submit_compressed_extents(struct inode *inode,
 596                                               struct async_cow *async_cow)
 597 {
 598         struct async_extent *async_extent;
 599         u64 alloc_hint = 0;
 600         struct btrfs_trans_handle *trans;
 601         struct btrfs_key ins;
 602         struct extent_map *em;
 603         struct btrfs_root *root = BTRFS_I(inode)->root;
 604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 605         struct extent_io_tree *io_tree;
 606         int ret = 0;
 607
 608         if (list_empty(&async_cow->extents))
 609                 return 0;
 610
 611
 612         while (!list_empty(&async_cow->extents)) {
 613                 async_extent = list_entry(async_cow->extents.next,
 614                                           struct async_extent, list);
 615                 list_del(&async_extent->list);
 616
 617                 io_tree = &BTRFS_I(inode)->io_tree;
 618
 619 retry:
 620                 /* did the compression code fall back to uncompressed IO? */
 621                 if (!async_extent->pages) {
 622                         int page_started = 0;
 623                         unsigned long nr_written = 0;
 624
 625                         lock_extent(io_tree, async_extent->start,
 626                                          async_extent->start +
 627                                          async_extent->ram_size - 1);
 628
 629                         /* allocate blocks */
 630                         ret = cow_file_range(inode, async_cow->locked_page,
 631                                              async_extent->start,
 632                                              async_extent->start +
 633                                              async_extent->ram_size - 1,
 634                                              &page_started, &nr_written, 0);
 635
 636                         /* JDM XXX */
 637
 638                         /*
 639                          * if page_started, cow_file_range inserted an
 640                          * inline extent and took care of all the unlocking
 641                          * and IO for us.  Otherwise, we need to submit
 642                          * all those pages down to the drive.
 643                          */
 644                         if (!page_started && !ret)
 645                                 extent_write_locked_range(io_tree,
 646                                                   inode, async_extent->start,
 647                                                   async_extent->start +
 648                                                   async_extent->ram_size - 1,
 649                                                   btrfs_get_extent,
 650                                                   WB_SYNC_ALL);
 651                         kfree(async_extent);
 652                         cond_resched();
 653                         continue;
 654                 }
 655
 656                 lock_extent(io_tree, async_extent->start,
 657                             async_extent->start + async_extent->ram_size - 1);
 658
 659                 trans = btrfs_join_transaction(root);
 660                 if (IS_ERR(trans)) {
 661                         ret = PTR_ERR(trans);
 662                 } else {
 663                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 664                         ret = btrfs_reserve_extent(trans, root,
 665                                            async_extent->compressed_size,
 666                                            async_extent->compressed_size,
 667                                            0, alloc_hint, &ins, 1);
 668                         if (ret && ret != -ENOSPC)
 669                                 btrfs_abort_transaction(trans, root, ret);
 670                         btrfs_end_transaction(trans, root);
 671                 }
 672
 673                 if (ret) {
 674                         int i;
 675                         for (i = 0; i < async_extent->nr_pages; i++) {
 676                                 WARN_ON(async_extent->pages[i]->mapping);
 677                                 page_cache_release(async_extent->pages[i]);
 678                         }
 679                         kfree(async_extent->pages);
 680                         async_extent->nr_pages = 0;
 681                         async_extent->pages = NULL;
 682                         unlock_extent(io_tree, async_extent->start,
 683                                       async_extent->start +
 684                                       async_extent->ram_size - 1);
 685                         if (ret == -ENOSPC)
 686                                 goto retry;
 687                         goto out_free; /* JDM: Requeue? */
 688                 }
 689
 690                 /*
 691                  * here we're doing allocation and writeback of the
 692                  * compressed pages
 693                  */
 694                 btrfs_drop_extent_cache(inode, async_extent->start,
 695                                         async_extent->start +
 696                                         async_extent->ram_size - 1, 0);
 697
 698                 em = alloc_extent_map();
 699                 BUG_ON(!em); /* -ENOMEM */
 700                 em->start = async_extent->start;
 701                 em->len = async_extent->ram_size;
 702                 em->orig_start = em->start;
 703                 em->mod_start = em->start;
 704                 em->mod_len = em->len;
 705
 706                 em->block_start = ins.objectid;
 707                 em->block_len = ins.offset;
 708                 em->orig_block_len = ins.offset;
 709                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 710                 em->compress_type = async_extent->compress_type;
 711                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 712                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 713                 em->generation = -1;
 714
 715                 while (1) {
 716                         write_lock(&em_tree->lock);
 717                         ret = add_extent_mapping(em_tree, em);
 718                         if (!ret)
 719                                 list_move(&em->list,
 720                                           &em_tree->modified_extents);
 721                         write_unlock(&em_tree->lock);
 722                         if (ret != -EEXIST) {
 723                                 free_extent_map(em);
 724                                 break;
 725                         }
 726                         btrfs_drop_extent_cache(inode, async_extent->start,
 727                                                 async_extent->start +
 728                                                 async_extent->ram_size - 1, 0);
 729                 }
 730
 731                 ret = btrfs_add_ordered_extent_compress(inode,
 732                                                 async_extent->start,
 733                                                 ins.objectid,
 734                                                 async_extent->ram_size,
 735                                                 ins.offset,
 736                                                 BTRFS_ORDERED_COMPRESSED,
 737                                                 async_extent->compress_type);
 738                 BUG_ON(ret); /* -ENOMEM */
 739
 740                 /*
 741                  * clear dirty, set writeback and unlock the pages.
 742                  */
 743                 extent_clear_unlock_delalloc(inode,
 744                                 &BTRFS_I(inode)->io_tree,
 745                                 async_extent->start,
 746                                 async_extent->start +
 747                                 async_extent->ram_size - 1,
 748                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 749                                 EXTENT_CLEAR_UNLOCK |
 750                                 EXTENT_CLEAR_DELALLOC |
 751                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 752
 753                 ret = btrfs_submit_compressed_write(inode,
 754                                     async_extent->start,
 755                                     async_extent->ram_size,
 756                                     ins.objectid,
 757                                     ins.offset, async_extent->pages,
 758                                     async_extent->nr_pages);
 759
 760                 BUG_ON(ret); /* -ENOMEM */
 761                 alloc_hint = ins.objectid + ins.offset;
 762                 kfree(async_extent);
 763                 cond_resched();
 764         }
 765         ret = 0;
 766 out:
 767         return ret;
 768 out_free:
 769         kfree(async_extent);
 770         goto out;
 771 }
 772
 773 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 774                                       u64 num_bytes)
 775 {
 776         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 777         struct extent_map *em;
 778         u64 alloc_hint = 0;
 779
 780         read_lock(&em_tree->lock);
 781         em = search_extent_mapping(em_tree, start, num_bytes);
 782         if (em) {
 783                 /*
 784                  * if block start isn't an actual block number then find the
 785                  * first block in this inode and use that as a hint.  If that
 786                  * block is also bogus then just don't worry about it.
 787                  */
 788                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 789                         free_extent_map(em);
 790                         em = search_extent_mapping(em_tree, 0, 0);
 791                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 792                                 alloc_hint = em->block_start;
 793                         if (em)
 794                                 free_extent_map(em);
 795                 } else {
 796                         alloc_hint = em->block_start;
 797                         free_extent_map(em);
 798                 }
 799         }
 800         read_unlock(&em_tree->lock);
 801
 802         return alloc_hint;
 803 }
 804
 805 /*
 806  * when extent_io.c finds a delayed allocation range in the file,
 807  * the call backs end up in this code.  The basic idea is to
 808  * allocate extents on disk for the range, and create ordered data structs
 809  * in ram to track those extents.
 810  *
 811  * locked_page is the page that writepage had locked already.  We use
 812  * it to make sure we don't do extra locks or unlocks.
 813  *
 814  * *page_started is set to one if we unlock locked_page and do everything
 815  * required to start IO on it.  It may be clean and already done with
 816  * IO when we return.
 817  */
 818 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 819                                      struct inode *inode,
 820                                      struct btrfs_root *root,
 821                                      struct page *locked_page,
 822                                      u64 start, u64 end, int *page_started,
 823                                      unsigned long *nr_written,
 824                                      int unlock)
 825 {
 826         u64 alloc_hint = 0;
 827         u64 num_bytes;
 828         unsigned long ram_size;
 829         u64 disk_num_bytes;
 830         u64 cur_alloc_size;
 831         u64 blocksize = root->sectorsize;
 832         struct btrfs_key ins;
 833         struct extent_map *em;
 834         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 835         int ret = 0;
 836
 837         BUG_ON(btrfs_is_free_space_inode(inode));
 838
 839         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 840         num_bytes = max(blocksize,  num_bytes);
 841         disk_num_bytes = num_bytes;
 842
 843         /* if this is a small write inside eof, kick off defrag */
 844         if (num_bytes < 64 * 1024 &&
 845             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 846                 btrfs_add_inode_defrag(trans, inode);
 847
 848         if (start == 0) {
 849                 /* lets try to make an inline extent */
 850                 ret = cow_file_range_inline(trans, root, inode,
 851                                             start, end, 0, 0, NULL);
 852                 if (ret == 0) {
 853                         extent_clear_unlock_delalloc(inode,
 854                                      &BTRFS_I(inode)->io_tree,
 855                                      start, end, NULL,
 856                                      EXTENT_CLEAR_UNLOCK_PAGE |
 857                                      EXTENT_CLEAR_UNLOCK |
 858                                      EXTENT_CLEAR_DELALLOC |
 859                                      EXTENT_CLEAR_DIRTY |
 860                                      EXTENT_SET_WRITEBACK |
 861                                      EXTENT_END_WRITEBACK);
 862
 863                         *nr_written = *nr_written +
 864                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 865                         *page_started = 1;
 866                         goto out;
 867                 } else if (ret < 0) {
 868                         btrfs_abort_transaction(trans, root, ret);
 869                         goto out_unlock;
 870                 }
 871         }
 872
 873         BUG_ON(disk_num_bytes >
 874                btrfs_super_total_bytes(root->fs_info->super_copy));
 875
 876         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 877         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 878
 879         while (disk_num_bytes > 0) {
 880                 unsigned long op;
 881
 882                 cur_alloc_size = disk_num_bytes;
 883                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 884                                            root->sectorsize, 0, alloc_hint,
 885                                            &ins, 1);
 886                 if (ret < 0) {
 887                         btrfs_abort_transaction(trans, root, ret);
 888                         goto out_unlock;
 889                 }
 890
 891                 em = alloc_extent_map();
 892                 BUG_ON(!em); /* -ENOMEM */
 893                 em->start = start;
 894                 em->orig_start = em->start;
 895                 ram_size = ins.offset;
 896                 em->len = ins.offset;
 897                 em->mod_start = em->start;
 898                 em->mod_len = em->len;
 899
 900                 em->block_start = ins.objectid;
 901                 em->block_len = ins.offset;
 902                 em->orig_block_len = ins.offset;
 903                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 904                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 905                 em->generation = -1;
 906
 907                 while (1) {
 908                         write_lock(&em_tree->lock);
 909                         ret = add_extent_mapping(em_tree, em);
 910                         if (!ret)
 911                                 list_move(&em->list,
 912                                           &em_tree->modified_extents);
 913                         write_unlock(&em_tree->lock);
 914                         if (ret != -EEXIST) {
 915                                 free_extent_map(em);
 916                                 break;
 917                         }
 918                         btrfs_drop_extent_cache(inode, start,
 919                                                 start + ram_size - 1, 0);
 920                 }
 921
 922                 cur_alloc_size = ins.offset;
 923                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 924                                                ram_size, cur_alloc_size, 0);
 925                 BUG_ON(ret); /* -ENOMEM */
 926
 927                 if (root->root_key.objectid ==
 928                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 929                         ret = btrfs_reloc_clone_csums(inode, start,
 930                                                       cur_alloc_size);
 931                         if (ret) {
 932                                 btrfs_abort_transaction(trans, root, ret);
 933                                 goto out_unlock;
 934                         }
 935                 }
 936
 937                 if (disk_num_bytes < cur_alloc_size)
 938                         break;
 939
 940                 /* we're not doing compressed IO, don't unlock the first
 941                  * page (which the caller expects to stay locked), don't
 942                  * clear any dirty bits and don't set any writeback bits
 943                  *
 944                  * Do set the Private2 bit so we know this page was properly
 945                  * setup for writepage
 946                  */
 947                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 948                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 949                         EXTENT_SET_PRIVATE2;
 950
 951                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 952                                              start, start + ram_size - 1,
 953                                              locked_page, op);
 954                 disk_num_bytes -= cur_alloc_size;
 955                 num_bytes -= cur_alloc_size;
 956                 alloc_hint = ins.objectid + ins.offset;
 957                 start += cur_alloc_size;
 958         }
 959 out:
 960         return ret;
 961
 962 out_unlock:
 963         extent_clear_unlock_delalloc(inode,
 964                      &BTRFS_I(inode)->io_tree,
 965                      start, end, locked_page,
 966                      EXTENT_CLEAR_UNLOCK_PAGE |
 967                      EXTENT_CLEAR_UNLOCK |
 968                      EXTENT_CLEAR_DELALLOC |
 969                      EXTENT_CLEAR_DIRTY |
 970                      EXTENT_SET_WRITEBACK |
 971                      EXTENT_END_WRITEBACK);
 972
 973         goto out;
 974 }
 975
 976 static noinline int cow_file_range(struct inode *inode,
 977                                    struct page *locked_page,
 978                                    u64 start, u64 end, int *page_started,
 979                                    unsigned long *nr_written,
 980                                    int unlock)
 981 {
 982         struct btrfs_trans_handle *trans;
 983         struct btrfs_root *root = BTRFS_I(inode)->root;
 984         int ret;
 985
 986         trans = btrfs_join_transaction(root);
 987         if (IS_ERR(trans)) {
 988                 extent_clear_unlock_delalloc(inode,
 989                              &BTRFS_I(inode)->io_tree,
 990                              start, end, locked_page,
 991                              EXTENT_CLEAR_UNLOCK_PAGE |
 992                              EXTENT_CLEAR_UNLOCK |
 993                              EXTENT_CLEAR_DELALLOC |
 994                              EXTENT_CLEAR_DIRTY |
 995                              EXTENT_SET_WRITEBACK |
 996                              EXTENT_END_WRITEBACK);
 997                 return PTR_ERR(trans);
 998         }
 999         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1000
1001         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1002                                page_started, nr_written, unlock);
1003
1004         btrfs_end_transaction(trans, root);
1005
1006         return ret;
1007 }
1008
1009 /*
1010  * work queue call back to started compression on a file and pages
1011  */
1012 static noinline void async_cow_start(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         int num_added = 0;
1016         async_cow = container_of(work, struct async_cow, work);
1017
1018         compress_file_range(async_cow->inode, async_cow->locked_page,
1019                             async_cow->start, async_cow->end, async_cow,
1020                             &num_added);
1021         if (num_added == 0) {
1022                 btrfs_add_delayed_iput(async_cow->inode);
1023                 async_cow->inode = NULL;
1024         }
1025 }
1026
1027 /*
1028  * work queue call back to submit previously compressed pages
1029  */
1030 static noinline void async_cow_submit(struct btrfs_work *work)
1031 {
1032         struct async_cow *async_cow;
1033         struct btrfs_root *root;
1034         unsigned long nr_pages;
1035
1036         async_cow = container_of(work, struct async_cow, work);
1037
1038         root = async_cow->root;
1039         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1040                 PAGE_CACHE_SHIFT;
1041
1042         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1043             5 * 1024 * 1024 &&
1044             waitqueue_active(&root->fs_info->async_submit_wait))
1045                 wake_up(&root->fs_info->async_submit_wait);
1046
1047         if (async_cow->inode)
1048                 submit_compressed_extents(async_cow->inode, async_cow);
1049 }
1050
1051 static noinline void async_cow_free(struct btrfs_work *work)
1052 {
1053         struct async_cow *async_cow;
1054         async_cow = container_of(work, struct async_cow, work);
1055         if (async_cow->inode)
1056                 btrfs_add_delayed_iput(async_cow->inode);
1057         kfree(async_cow);
1058 }
1059
1060 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1061                                 u64 start, u64 end, int *page_started,
1062                                 unsigned long *nr_written)
1063 {
1064         struct async_cow *async_cow;
1065         struct btrfs_root *root = BTRFS_I(inode)->root;
1066         unsigned long nr_pages;
1067         u64 cur_end;
1068         int limit = 10 * 1024 * 1024;
1069
1070         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1071                          1, 0, NULL, GFP_NOFS);
1072         while (start < end) {
1073                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1074                 BUG_ON(!async_cow); /* -ENOMEM */
1075                 async_cow->inode = igrab(inode);
1076                 async_cow->root = root;
1077                 async_cow->locked_page = locked_page;
1078                 async_cow->start = start;
1079
1080                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1081                         cur_end = end;
1082                 else
1083                         cur_end = min(end, start + 512 * 1024 - 1);
1084
1085                 async_cow->end = cur_end;
1086                 INIT_LIST_HEAD(&async_cow->extents);
1087
1088                 async_cow->work.func = async_cow_start;
1089                 async_cow->work.ordered_func = async_cow_submit;
1090                 async_cow->work.ordered_free = async_cow_free;
1091                 async_cow->work.flags = 0;
1092
1093                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1094                         PAGE_CACHE_SHIFT;
1095                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1096
1097                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1098                                    &async_cow->work);
1099
1100                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1101                         wait_event(root->fs_info->async_submit_wait,
1102                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1103                             limit));
1104                 }
1105
1106                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1107                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1108                         wait_event(root->fs_info->async_submit_wait,
1109                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1110                            0));
1111                 }
1112
1113                 *nr_written += nr_pages;
1114                 start = cur_end + 1;
1115         }
1116         *page_started = 1;
1117         return 0;
1118 }
1119
1120 static noinline int csum_exist_in_range(struct btrfs_root *root,
1121                                         u64 bytenr, u64 num_bytes)
1122 {
1123         int ret;
1124         struct btrfs_ordered_sum *sums;
1125         LIST_HEAD(list);
1126
1127         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1128                                        bytenr + num_bytes - 1, &list, 0);
1129         if (ret == 0 && list_empty(&list))
1130                 return 0;
1131
1132         while (!list_empty(&list)) {
1133                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1134                 list_del(&sums->list);
1135                 kfree(sums);
1136         }
1137         return 1;
1138 }
1139
1140 /*
1141  * when nowcow writeback call back.  This checks for snapshots or COW copies
1142  * of the extents that exist in the file, and COWs the file as required.
1143  *
1144  * If no cow copies or snapshots exist, we write directly to the existing
1145  * blocks on disk
1146  */
1147 static noinline int run_delalloc_nocow(struct inode *inode,
1148                                        struct page *locked_page,
1149                               u64 start, u64 end, int *page_started, int force,
1150                               unsigned long *nr_written)
1151 {
1152         struct btrfs_root *root = BTRFS_I(inode)->root;
1153         struct btrfs_trans_handle *trans;
1154         struct extent_buffer *leaf;
1155         struct btrfs_path *path;
1156         struct btrfs_file_extent_item *fi;
1157         struct btrfs_key found_key;
1158         u64 cow_start;
1159         u64 cur_offset;
1160         u64 extent_end;
1161         u64 extent_offset;
1162         u64 disk_bytenr;
1163         u64 num_bytes;
1164         u64 disk_num_bytes;
1165         int extent_type;
1166         int ret, err;
1167         int type;
1168         int nocow;
1169         int check_prev = 1;
1170         bool nolock;
1171         u64 ino = btrfs_ino(inode);
1172
1173         path = btrfs_alloc_path();
1174         if (!path) {
1175                 extent_clear_unlock_delalloc(inode,
1176                              &BTRFS_I(inode)->io_tree,
1177                              start, end, locked_page,
1178                              EXTENT_CLEAR_UNLOCK_PAGE |
1179                              EXTENT_CLEAR_UNLOCK |
1180                              EXTENT_CLEAR_DELALLOC |
1181                              EXTENT_CLEAR_DIRTY |
1182                              EXTENT_SET_WRITEBACK |
1183                              EXTENT_END_WRITEBACK);
1184                 return -ENOMEM;
1185         }
1186
1187         nolock = btrfs_is_free_space_inode(inode);
1188
1189         if (nolock)
1190                 trans = btrfs_join_transaction_nolock(root);
1191         else
1192                 trans = btrfs_join_transaction(root);
1193
1194         if (IS_ERR(trans)) {
1195                 extent_clear_unlock_delalloc(inode,
1196                              &BTRFS_I(inode)->io_tree,
1197                              start, end, locked_page,
1198                              EXTENT_CLEAR_UNLOCK_PAGE |
1199                              EXTENT_CLEAR_UNLOCK |
1200                              EXTENT_CLEAR_DELALLOC |
1201                              EXTENT_CLEAR_DIRTY |
1202                              EXTENT_SET_WRITEBACK |
1203                              EXTENT_END_WRITEBACK);
1204                 btrfs_free_path(path);
1205                 return PTR_ERR(trans);
1206         }
1207
1208         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1209
1210         cow_start = (u64)-1;
1211         cur_offset = start;
1212         while (1) {
1213                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1214                                                cur_offset, 0);
1215                 if (ret < 0) {
1216                         btrfs_abort_transaction(trans, root, ret);
1217                         goto error;
1218                 }
1219                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1220                         leaf = path->nodes[0];
1221                         btrfs_item_key_to_cpu(leaf, &found_key,
1222                                               path->slots[0] - 1);
1223                         if (found_key.objectid == ino &&
1224                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1225                                 path->slots[0]--;
1226                 }
1227                 check_prev = 0;
1228 next_slot:
1229                 leaf = path->nodes[0];
1230                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1231                         ret = btrfs_next_leaf(root, path);
1232                         if (ret < 0) {
1233                                 btrfs_abort_transaction(trans, root, ret);
1234                                 goto error;
1235                         }
1236                         if (ret > 0)
1237                                 break;
1238                         leaf = path->nodes[0];
1239                 }
1240
1241                 nocow = 0;
1242                 disk_bytenr = 0;
1243                 num_bytes = 0;
1244                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1245
1246                 if (found_key.objectid > ino ||
1247                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1248                     found_key.offset > end)
1249                         break;
1250
1251                 if (found_key.offset > cur_offset) {
1252                         extent_end = found_key.offset;
1253                         extent_type = 0;
1254                         goto out_check;
1255                 }
1256
1257                 fi = btrfs_item_ptr(leaf, path->slots[0],
1258                                     struct btrfs_file_extent_item);
1259                 extent_type = btrfs_file_extent_type(leaf, fi);
1260
1261                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1262                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1263                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1264                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1265                         extent_end = found_key.offset +
1266                                 btrfs_file_extent_num_bytes(leaf, fi);
1267                         disk_num_bytes =
1268                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1269                         if (extent_end <= start) {
1270                                 path->slots[0]++;
1271                                 goto next_slot;
1272                         }
1273                         if (disk_bytenr == 0)
1274                                 goto out_check;
1275                         if (btrfs_file_extent_compression(leaf, fi) ||
1276                             btrfs_file_extent_encryption(leaf, fi) ||
1277                             btrfs_file_extent_other_encoding(leaf, fi))
1278                                 goto out_check;
1279                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1280                                 goto out_check;
1281                         if (btrfs_extent_readonly(root, disk_bytenr))
1282                                 goto out_check;
1283                         if (btrfs_cross_ref_exist(trans, root, ino,
1284                                                   found_key.offset -
1285                                                   extent_offset, disk_bytenr))
1286                                 goto out_check;
1287                         disk_bytenr += extent_offset;
1288                         disk_bytenr += cur_offset - found_key.offset;
1289                         num_bytes = min(end + 1, extent_end) - cur_offset;
1290                         /*
1291                          * force cow if csum exists in the range.
1292                          * this ensure that csum for a given extent are
1293                          * either valid or do not exist.
1294                          */
1295                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296                                 goto out_check;
1297                         nocow = 1;
1298                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299                         extent_end = found_key.offset +
1300                                 btrfs_file_extent_inline_len(leaf, fi);
1301                         extent_end = ALIGN(extent_end, root->sectorsize);
1302                 } else {
1303                         BUG_ON(1);
1304                 }
1305 out_check:
1306                 if (extent_end <= start) {
1307                         path->slots[0]++;
1308                         goto next_slot;
1309                 }
1310                 if (!nocow) {
1311                         if (cow_start == (u64)-1)
1312                                 cow_start = cur_offset;
1313                         cur_offset = extent_end;
1314                         if (cur_offset > end)
1315                                 break;
1316                         path->slots[0]++;
1317                         goto next_slot;
1318                 }
1319
1320                 btrfs_release_path(path);
1321                 if (cow_start != (u64)-1) {
1322                         ret = __cow_file_range(trans, inode, root, locked_page,
1323                                                cow_start, found_key.offset - 1,
1324                                                page_started, nr_written, 1);
1325                         if (ret) {
1326                                 btrfs_abort_transaction(trans, root, ret);
1327                                 goto error;
1328                         }
1329                         cow_start = (u64)-1;
1330                 }
1331
1332                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1333                         struct extent_map *em;
1334                         struct extent_map_tree *em_tree;
1335                         em_tree = &BTRFS_I(inode)->extent_tree;
1336                         em = alloc_extent_map();
1337                         BUG_ON(!em); /* -ENOMEM */
1338                         em->start = cur_offset;
1339                         em->orig_start = found_key.offset - extent_offset;
1340                         em->len = num_bytes;
1341                         em->block_len = num_bytes;
1342                         em->block_start = disk_bytenr;
1343                         em->orig_block_len = disk_num_bytes;
1344                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1345                         em->mod_start = em->start;
1346                         em->mod_len = em->len;
1347                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1348                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1349                         em->generation = -1;
1350                         while (1) {
1351                                 write_lock(&em_tree->lock);
1352                                 ret = add_extent_mapping(em_tree, em);
1353                                 if (!ret)
1354                                         list_move(&em->list,
1355                                                   &em_tree->modified_extents);
1356                                 write_unlock(&em_tree->lock);
1357                                 if (ret != -EEXIST) {
1358                                         free_extent_map(em);
1359                                         break;
1360                                 }
1361                                 btrfs_drop_extent_cache(inode, em->start,
1362                                                 em->start + em->len - 1, 0);
1363                         }
1364                         type = BTRFS_ORDERED_PREALLOC;
1365                 } else {
1366                         type = BTRFS_ORDERED_NOCOW;
1367                 }
1368
1369                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1370                                                num_bytes, num_bytes, type);
1371                 BUG_ON(ret); /* -ENOMEM */
1372
1373                 if (root->root_key.objectid ==
1374                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1375                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1376                                                       num_bytes);
1377                         if (ret) {
1378                                 btrfs_abort_transaction(trans, root, ret);
1379                                 goto error;
1380                         }
1381                 }
1382
1383                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1384                                 cur_offset, cur_offset + num_bytes - 1,
1385                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1386                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1387                                 EXTENT_SET_PRIVATE2);
1388                 cur_offset = extent_end;
1389                 if (cur_offset > end)
1390                         break;
1391         }
1392         btrfs_release_path(path);
1393
1394         if (cur_offset <= end && cow_start == (u64)-1) {
1395                 cow_start = cur_offset;
1396                 cur_offset = end;
1397         }
1398
1399         if (cow_start != (u64)-1) {
1400                 ret = __cow_file_range(trans, inode, root, locked_page,
1401                                        cow_start, end,
1402                                        page_started, nr_written, 1);
1403                 if (ret) {
1404                         btrfs_abort_transaction(trans, root, ret);
1405                         goto error;
1406                 }
1407         }
1408
1409 error:
1410         err = btrfs_end_transaction(trans, root);
1411         if (!ret)
1412                 ret = err;
1413
1414         if (ret && cur_offset < end)
1415                 extent_clear_unlock_delalloc(inode,
1416                              &BTRFS_I(inode)->io_tree,
1417                              cur_offset, end, locked_page,
1418                              EXTENT_CLEAR_UNLOCK_PAGE |
1419                              EXTENT_CLEAR_UNLOCK |
1420                              EXTENT_CLEAR_DELALLOC |
1421                              EXTENT_CLEAR_DIRTY |
1422                              EXTENT_SET_WRITEBACK |
1423                              EXTENT_END_WRITEBACK);
1424
1425         btrfs_free_path(path);
1426         return ret;
1427 }
1428
1429 /*
1430  * extent_io.c call back to do delayed allocation processing
1431  */
1432 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1433                               u64 start, u64 end, int *page_started,
1434                               unsigned long *nr_written)
1435 {
1436         int ret;
1437         struct btrfs_root *root = BTRFS_I(inode)->root;
1438
1439         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1440                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1441                                          page_started, 1, nr_written);
1442         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1443                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1444                                          page_started, 0, nr_written);
1445         } else if (!btrfs_test_opt(root, COMPRESS) &&
1446                    !(BTRFS_I(inode)->force_compress) &&
1447                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1448                 ret = cow_file_range(inode, locked_page, start, end,
1449                                       page_started, nr_written, 1);
1450         } else {
1451                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1452                         &BTRFS_I(inode)->runtime_flags);
1453                 ret = cow_file_range_async(inode, locked_page, start, end,
1454                                            page_started, nr_written);
1455         }
1456         return ret;
1457 }
1458
1459 static void btrfs_split_extent_hook(struct inode *inode,
1460                                     struct extent_state *orig, u64 split)
1461 {
1462         /* not delalloc, ignore it */
1463         if (!(orig->state & EXTENT_DELALLOC))
1464                 return;
1465
1466         spin_lock(&BTRFS_I(inode)->lock);
1467         BTRFS_I(inode)->outstanding_extents++;
1468         spin_unlock(&BTRFS_I(inode)->lock);
1469 }
1470
1471 /*
1472  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1473  * extents so we can keep track of new extents that are just merged onto old
1474  * extents, such as when we are doing sequential writes, so we can properly
1475  * account for the metadata space we'll need.
1476  */
1477 static void btrfs_merge_extent_hook(struct inode *inode,
1478                                     struct extent_state *new,
1479                                     struct extent_state *other)
1480 {
1481         /* not delalloc, ignore it */
1482         if (!(other->state & EXTENT_DELALLOC))
1483                 return;
1484
1485         spin_lock(&BTRFS_I(inode)->lock);
1486         BTRFS_I(inode)->outstanding_extents--;
1487         spin_unlock(&BTRFS_I(inode)->lock);
1488 }
1489
1490 /*
1491  * extent_io.c set_bit_hook, used to track delayed allocation
1492  * bytes in this file, and to maintain the list of inodes that
1493  * have pending delalloc work to be done.
1494  */
1495 static void btrfs_set_bit_hook(struct inode *inode,
1496                                struct extent_state *state, int *bits)
1497 {
1498
1499         /*
1500          * set_bit and clear bit hooks normally require _irqsave/restore
1501          * but in this case, we are only testing for the DELALLOC
1502          * bit, which is only set or cleared with irqs on
1503          */
1504         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1505                 struct btrfs_root *root = BTRFS_I(inode)->root;
1506                 u64 len = state->end + 1 - state->start;
1507                 bool do_list = !btrfs_is_free_space_inode(inode);
1508
1509                 if (*bits & EXTENT_FIRST_DELALLOC) {
1510                         *bits &= ~EXTENT_FIRST_DELALLOC;
1511                 } else {
1512                         spin_lock(&BTRFS_I(inode)->lock);
1513                         BTRFS_I(inode)->outstanding_extents++;
1514                         spin_unlock(&BTRFS_I(inode)->lock);
1515                 }
1516
1517                 spin_lock(&root->fs_info->delalloc_lock);
1518                 BTRFS_I(inode)->delalloc_bytes += len;
1519                 root->fs_info->delalloc_bytes += len;
1520                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1521                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1522                                       &root->fs_info->delalloc_inodes);
1523                 }
1524                 spin_unlock(&root->fs_info->delalloc_lock);
1525         }
1526 }
1527
1528 /*
1529  * extent_io.c clear_bit_hook, see set_bit_hook for why
1530  */
1531 static void btrfs_clear_bit_hook(struct inode *inode,
1532                                  struct extent_state *state, int *bits)
1533 {
1534         /*
1535          * set_bit and clear bit hooks normally require _irqsave/restore
1536          * but in this case, we are only testing for the DELALLOC
1537          * bit, which is only set or cleared with irqs on
1538          */
1539         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1540                 struct btrfs_root *root = BTRFS_I(inode)->root;
1541                 u64 len = state->end + 1 - state->start;
1542                 bool do_list = !btrfs_is_free_space_inode(inode);
1543
1544                 if (*bits & EXTENT_FIRST_DELALLOC) {
1545                         *bits &= ~EXTENT_FIRST_DELALLOC;
1546                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1547                         spin_lock(&BTRFS_I(inode)->lock);
1548                         BTRFS_I(inode)->outstanding_extents--;
1549                         spin_unlock(&BTRFS_I(inode)->lock);
1550                 }
1551
1552                 if (*bits & EXTENT_DO_ACCOUNTING)
1553                         btrfs_delalloc_release_metadata(inode, len);
1554
1555                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1556                     && do_list)
1557                         btrfs_free_reserved_data_space(inode, len);
1558
1559                 spin_lock(&root->fs_info->delalloc_lock);
1560                 root->fs_info->delalloc_bytes -= len;
1561                 BTRFS_I(inode)->delalloc_bytes -= len;
1562
1563                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1564                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1565                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1566                 }
1567                 spin_unlock(&root->fs_info->delalloc_lock);
1568         }
1569 }
1570
1571 /*
1572  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573  * we don't create bios that span stripes or chunks
1574  */
1575 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1576                          size_t size, struct bio *bio,
1577                          unsigned long bio_flags)
1578 {
1579         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1580         u64 logical = (u64)bio->bi_sector << 9;
1581         u64 length = 0;
1582         u64 map_length;
1583         int ret;
1584
1585         if (bio_flags & EXTENT_BIO_COMPRESSED)
1586                 return 0;
1587
1588         length = bio->bi_size;
1589         map_length = length;
1590         ret = btrfs_map_block(root->fs_info, READ, logical,
1591                               &map_length, NULL, 0);
1592         /* Will always return 0 with map_multi == NULL */
1593         BUG_ON(ret < 0);
1594         if (map_length < length + size)
1595                 return 1;
1596         return 0;
1597 }
1598
1599 /*
1600  * in order to insert checksums into the metadata in large chunks,
1601  * we wait until bio submission time.   All the pages in the bio are
1602  * checksummed and sums are attached onto the ordered extent record.
1603  *
1604  * At IO completion time the cums attached on the ordered extent record
1605  * are inserted into the btree
1606  */
1607 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608                                     struct bio *bio, int mirror_num,
1609                                     unsigned long bio_flags,
1610                                     u64 bio_offset)
1611 {
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613         int ret = 0;
1614
1615         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1616         BUG_ON(ret); /* -ENOMEM */
1617         return 0;
1618 }
1619
1620 /*
1621  * in order to insert checksums into the metadata in large chunks,
1622  * we wait until bio submission time.   All the pages in the bio are
1623  * checksummed and sums are attached onto the ordered extent record.
1624  *
1625  * At IO completion time the cums attached on the ordered extent record
1626  * are inserted into the btree
1627  */
1628 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1629                           int mirror_num, unsigned long bio_flags,
1630                           u64 bio_offset)
1631 {
1632         struct btrfs_root *root = BTRFS_I(inode)->root;
1633         int ret;
1634
1635         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1636         if (ret)
1637                 bio_endio(bio, ret);
1638         return ret;
1639 }
1640
1641 /*
1642  * extent_io.c submission hook. This does the right thing for csum calculation
1643  * on write, or reading the csums from the tree before a read
1644  */
1645 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1646                           int mirror_num, unsigned long bio_flags,
1647                           u64 bio_offset)
1648 {
1649         struct btrfs_root *root = BTRFS_I(inode)->root;
1650         int ret = 0;
1651         int skip_sum;
1652         int metadata = 0;
1653         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1654
1655         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1656
1657         if (btrfs_is_free_space_inode(inode))
1658                 metadata = 2;
1659
1660         if (!(rw & REQ_WRITE)) {
1661                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1662                 if (ret)
1663                         goto out;
1664
1665                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1666                         ret = btrfs_submit_compressed_read(inode, bio,
1667                                                            mirror_num,
1668                                                            bio_flags);
1669                         goto out;
1670                 } else if (!skip_sum) {
1671                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1672                         if (ret)
1673                                 goto out;
1674                 }
1675                 goto mapit;
1676         } else if (async && !skip_sum) {
1677                 /* csum items have already been cloned */
1678                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1679                         goto mapit;
1680                 /* we're doing a write, do the async checksumming */
1681                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1682                                    inode, rw, bio, mirror_num,
1683                                    bio_flags, bio_offset,
1684                                    __btrfs_submit_bio_start,
1685                                    __btrfs_submit_bio_done);
1686                 goto out;
1687         } else if (!skip_sum) {
1688                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1689                 if (ret)
1690                         goto out;
1691         }
1692
1693 mapit:
1694         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1695
1696 out:
1697         if (ret < 0)
1698                 bio_endio(bio, ret);
1699         return ret;
1700 }
1701
1702 /*
1703  * given a list of ordered sums record them in the inode.  This happens
1704  * at IO completion time based on sums calculated at bio submission time.
1705  */
1706 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1707                              struct inode *inode, u64 file_offset,
1708                              struct list_head *list)
1709 {
1710         struct btrfs_ordered_sum *sum;
1711
1712         list_for_each_entry(sum, list, list) {
1713                 btrfs_csum_file_blocks(trans,
1714                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1715         }
1716         return 0;
1717 }
1718
1719 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1720                               struct extent_state **cached_state)
1721 {
1722         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1723         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1724                                    cached_state, GFP_NOFS);
1725 }
1726
1727 /* see btrfs_writepage_start_hook for details on why this is required */
1728 struct btrfs_writepage_fixup {
1729         struct page *page;
1730         struct btrfs_work work;
1731 };
1732
1733 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1734 {
1735         struct btrfs_writepage_fixup *fixup;
1736         struct btrfs_ordered_extent *ordered;
1737         struct extent_state *cached_state = NULL;
1738         struct page *page;
1739         struct inode *inode;
1740         u64 page_start;
1741         u64 page_end;
1742         int ret;
1743
1744         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1745         page = fixup->page;
1746 again:
1747         lock_page(page);
1748         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1749                 ClearPageChecked(page);
1750                 goto out_page;
1751         }
1752
1753         inode = page->mapping->host;
1754         page_start = page_offset(page);
1755         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1756
1757         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1758                          &cached_state);
1759
1760         /* already ordered? We're done */
1761         if (PagePrivate2(page))
1762                 goto out;
1763
1764         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1765         if (ordered) {
1766                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1767                                      page_end, &cached_state, GFP_NOFS);
1768                 unlock_page(page);
1769                 btrfs_start_ordered_extent(inode, ordered, 1);
1770                 btrfs_put_ordered_extent(ordered);
1771                 goto again;
1772         }
1773
1774         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1775         if (ret) {
1776                 mapping_set_error(page->mapping, ret);
1777                 end_extent_writepage(page, ret, page_start, page_end);
1778                 ClearPageChecked(page);
1779                 goto out;
1780          }
1781
1782         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1783         ClearPageChecked(page);
1784         set_page_dirty(page);
1785 out:
1786         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1787                              &cached_state, GFP_NOFS);
1788 out_page:
1789         unlock_page(page);
1790         page_cache_release(page);
1791         kfree(fixup);
1792 }
1793
1794 /*
1795  * There are a few paths in the higher layers of the kernel that directly
1796  * set the page dirty bit without asking the filesystem if it is a
1797  * good idea.  This causes problems because we want to make sure COW
1798  * properly happens and the data=ordered rules are followed.
1799  *
1800  * In our case any range that doesn't have the ORDERED bit set
1801  * hasn't been properly setup for IO.  We kick off an async process
1802  * to fix it up.  The async helper will wait for ordered extents, set
1803  * the delalloc bit and make it safe to write the page.
1804  */
1805 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1806 {
1807         struct inode *inode = page->mapping->host;
1808         struct btrfs_writepage_fixup *fixup;
1809         struct btrfs_root *root = BTRFS_I(inode)->root;
1810
1811         /* this page is properly in the ordered list */
1812         if (TestClearPagePrivate2(page))
1813                 return 0;
1814
1815         if (PageChecked(page))
1816                 return -EAGAIN;
1817
1818         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1819         if (!fixup)
1820                 return -EAGAIN;
1821
1822         SetPageChecked(page);
1823         page_cache_get(page);
1824         fixup->work.func = btrfs_writepage_fixup_worker;
1825         fixup->page = page;
1826         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1827         return -EBUSY;
1828 }
1829
1830 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1831                                        struct inode *inode, u64 file_pos,
1832                                        u64 disk_bytenr, u64 disk_num_bytes,
1833                                        u64 num_bytes, u64 ram_bytes,
1834                                        u8 compression, u8 encryption,
1835                                        u16 other_encoding, int extent_type)
1836 {
1837         struct btrfs_root *root = BTRFS_I(inode)->root;
1838         struct btrfs_file_extent_item *fi;
1839         struct btrfs_path *path;
1840         struct extent_buffer *leaf;
1841         struct btrfs_key ins;
1842         int ret;
1843
1844         path = btrfs_alloc_path();
1845         if (!path)
1846                 return -ENOMEM;
1847
1848         path->leave_spinning = 1;
1849
1850         /*
1851          * we may be replacing one extent in the tree with another.
1852          * The new extent is pinned in the extent map, and we don't want
1853          * to drop it from the cache until it is completely in the btree.
1854          *
1855          * So, tell btrfs_drop_extents to leave this extent in the cache.
1856          * the caller is expected to unpin it and allow it to be merged
1857          * with the others.
1858          */
1859         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1860                                  file_pos + num_bytes, 0);
1861         if (ret)
1862                 goto out;
1863
1864         ins.objectid = btrfs_ino(inode);
1865         ins.offset = file_pos;
1866         ins.type = BTRFS_EXTENT_DATA_KEY;
1867         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1868         if (ret)
1869                 goto out;
1870         leaf = path->nodes[0];
1871         fi = btrfs_item_ptr(leaf, path->slots[0],
1872                             struct btrfs_file_extent_item);
1873         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1874         btrfs_set_file_extent_type(leaf, fi, extent_type);
1875         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1876         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1877         btrfs_set_file_extent_offset(leaf, fi, 0);
1878         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1879         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1880         btrfs_set_file_extent_compression(leaf, fi, compression);
1881         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1882         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1883
1884         btrfs_mark_buffer_dirty(leaf);
1885         btrfs_release_path(path);
1886
1887         inode_add_bytes(inode, num_bytes);
1888
1889         ins.objectid = disk_bytenr;
1890         ins.offset = disk_num_bytes;
1891         ins.type = BTRFS_EXTENT_ITEM_KEY;
1892         ret = btrfs_alloc_reserved_file_extent(trans, root,
1893                                         root->root_key.objectid,
1894                                         btrfs_ino(inode), file_pos, &ins);
1895 out:
1896         btrfs_free_path(path);
1897
1898         return ret;
1899 }
1900
1901 /*
1902  * helper function for btrfs_finish_ordered_io, this
1903  * just reads in some of the csum leaves to prime them into ram
1904  * before we start the transaction.  It limits the amount of btree
1905  * reads required while inside the transaction.
1906  */
1907 /* as ordered data IO finishes, this gets called so we can finish
1908  * an ordered extent if the range of bytes in the file it covers are
1909  * fully written.
1910  */
1911 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1912 {
1913         struct inode *inode = ordered_extent->inode;
1914         struct btrfs_root *root = BTRFS_I(inode)->root;
1915         struct btrfs_trans_handle *trans = NULL;
1916         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1917         struct extent_state *cached_state = NULL;
1918         int compress_type = 0;
1919         int ret;
1920         bool nolock;
1921
1922         nolock = btrfs_is_free_space_inode(inode);
1923
1924         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1925                 ret = -EIO;
1926                 goto out;
1927         }
1928
1929         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1930                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1931                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1932                 if (nolock)
1933                         trans = btrfs_join_transaction_nolock(root);
1934                 else
1935                         trans = btrfs_join_transaction(root);
1936                 if (IS_ERR(trans)) {
1937                         ret = PTR_ERR(trans);
1938                         trans = NULL;
1939                         goto out;
1940                 }
1941                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1942                 ret = btrfs_update_inode_fallback(trans, root, inode);
1943                 if (ret) /* -ENOMEM or corruption */
1944                         btrfs_abort_transaction(trans, root, ret);
1945                 goto out;
1946         }
1947
1948         lock_extent_bits(io_tree, ordered_extent->file_offset,
1949                          ordered_extent->file_offset + ordered_extent->len - 1,
1950                          0, &cached_state);
1951
1952         if (nolock)
1953                 trans = btrfs_join_transaction_nolock(root);
1954         else
1955                 trans = btrfs_join_transaction(root);
1956         if (IS_ERR(trans)) {
1957                 ret = PTR_ERR(trans);
1958                 trans = NULL;
1959                 goto out_unlock;
1960         }
1961         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1962
1963         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1964                 compress_type = ordered_extent->compress_type;
1965         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1966                 BUG_ON(compress_type);
1967                 ret = btrfs_mark_extent_written(trans, inode,
1968                                                 ordered_extent->file_offset,
1969                                                 ordered_extent->file_offset +
1970                                                 ordered_extent->len);
1971         } else {
1972                 BUG_ON(root == root->fs_info->tree_root);
1973                 ret = insert_reserved_file_extent(trans, inode,
1974                                                 ordered_extent->file_offset,
1975                                                 ordered_extent->start,
1976                                                 ordered_extent->disk_len,
1977                                                 ordered_extent->len,
1978                                                 ordered_extent->len,
1979                                                 compress_type, 0, 0,
1980                                                 BTRFS_FILE_EXTENT_REG);
1981         }
1982         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1983                            ordered_extent->file_offset, ordered_extent->len,
1984                            trans->transid);
1985         if (ret < 0) {
1986                 btrfs_abort_transaction(trans, root, ret);
1987                 goto out_unlock;
1988         }
1989
1990         add_pending_csums(trans, inode, ordered_extent->file_offset,
1991                           &ordered_extent->list);
1992
1993         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1994         ret = btrfs_update_inode_fallback(trans, root, inode);
1995         if (ret) { /* -ENOMEM or corruption */
1996                 btrfs_abort_transaction(trans, root, ret);
1997                 goto out_unlock;
1998         }
1999         ret = 0;
2000 out_unlock:
2001         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2002                              ordered_extent->file_offset +
2003                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2004 out:
2005         if (root != root->fs_info->tree_root)
2006                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2007         if (trans)
2008                 btrfs_end_transaction(trans, root);
2009
2010         if (ret)
2011                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2012                                       ordered_extent->file_offset +
2013                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2014
2015         /*
2016          * This needs to be done to make sure anybody waiting knows we are done
2017          * updating everything for this ordered extent.
2018          */
2019         btrfs_remove_ordered_extent(inode, ordered_extent);
2020
2021         /* once for us */
2022         btrfs_put_ordered_extent(ordered_extent);
2023         /* once for the tree */
2024         btrfs_put_ordered_extent(ordered_extent);
2025
2026         return ret;
2027 }
2028
2029 static void finish_ordered_fn(struct btrfs_work *work)
2030 {
2031         struct btrfs_ordered_extent *ordered_extent;
2032         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2033         btrfs_finish_ordered_io(ordered_extent);
2034 }
2035
2036 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2037                                 struct extent_state *state, int uptodate)
2038 {
2039         struct inode *inode = page->mapping->host;
2040         struct btrfs_root *root = BTRFS_I(inode)->root;
2041         struct btrfs_ordered_extent *ordered_extent = NULL;
2042         struct btrfs_workers *workers;
2043
2044         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2045
2046         ClearPagePrivate2(page);
2047         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2048                                             end - start + 1, uptodate))
2049                 return 0;
2050
2051         ordered_extent->work.func = finish_ordered_fn;
2052         ordered_extent->work.flags = 0;
2053
2054         if (btrfs_is_free_space_inode(inode))
2055                 workers = &root->fs_info->endio_freespace_worker;
2056         else
2057                 workers = &root->fs_info->endio_write_workers;
2058         btrfs_queue_worker(workers, &ordered_extent->work);
2059
2060         return 0;
2061 }
2062
2063 /*
2064  * when reads are done, we need to check csums to verify the data is correct
2065  * if there's a match, we allow the bio to finish.  If not, the code in
2066  * extent_io.c will try to find good copies for us.
2067  */
2068 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2069                                struct extent_state *state, int mirror)
2070 {
2071         size_t offset = start - page_offset(page);
2072         struct inode *inode = page->mapping->host;
2073         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2074         char *kaddr;
2075         u64 private = ~(u32)0;
2076         int ret;
2077         struct btrfs_root *root = BTRFS_I(inode)->root;
2078         u32 csum = ~(u32)0;
2079
2080         if (PageChecked(page)) {
2081                 ClearPageChecked(page);
2082                 goto good;
2083         }
2084
2085         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2086                 goto good;
2087
2088         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2089             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2090                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2091                                   GFP_NOFS);
2092                 return 0;
2093         }
2094
2095         if (state && state->start == start) {
2096                 private = state->private;
2097                 ret = 0;
2098         } else {
2099                 ret = get_state_private(io_tree, start, &private);
2100         }
2101         kaddr = kmap_atomic(page);
2102         if (ret)
2103                 goto zeroit;
2104
2105         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2106         btrfs_csum_final(csum, (char *)&csum);
2107         if (csum != private)
2108                 goto zeroit;
2109
2110         kunmap_atomic(kaddr);
2111 good:
2112         return 0;
2113
2114 zeroit:
2115         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2116                        "private %llu\n",
2117                        (unsigned long long)btrfs_ino(page->mapping->host),
2118                        (unsigned long long)start, csum,
2119                        (unsigned long long)private);
2120         memset(kaddr + offset, 1, end - start + 1);
2121         flush_dcache_page(page);
2122         kunmap_atomic(kaddr);
2123         if (private == 0)
2124                 return 0;
2125         return -EIO;
2126 }
2127
2128 struct delayed_iput {
2129         struct list_head list;
2130         struct inode *inode;
2131 };
2132
2133 /* JDM: If this is fs-wide, why can't we add a pointer to
2134  * btrfs_inode instead and avoid the allocation? */
2135 void btrfs_add_delayed_iput(struct inode *inode)
2136 {
2137         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2138         struct delayed_iput *delayed;
2139
2140         if (atomic_add_unless(&inode->i_count, -1, 1))
2141                 return;
2142
2143         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2144         delayed->inode = inode;
2145
2146         spin_lock(&fs_info->delayed_iput_lock);
2147         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2148         spin_unlock(&fs_info->delayed_iput_lock);
2149 }
2150
2151 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2152 {
2153         LIST_HEAD(list);
2154         struct btrfs_fs_info *fs_info = root->fs_info;
2155         struct delayed_iput *delayed;
2156         int empty;
2157
2158         spin_lock(&fs_info->delayed_iput_lock);
2159         empty = list_empty(&fs_info->delayed_iputs);
2160         spin_unlock(&fs_info->delayed_iput_lock);
2161         if (empty)
2162                 return;
2163
2164         spin_lock(&fs_info->delayed_iput_lock);
2165         list_splice_init(&fs_info->delayed_iputs, &list);
2166         spin_unlock(&fs_info->delayed_iput_lock);
2167
2168         while (!list_empty(&list)) {
2169                 delayed = list_entry(list.next, struct delayed_iput, list);
2170                 list_del(&delayed->list);
2171                 iput(delayed->inode);
2172                 kfree(delayed);
2173         }
2174 }
2175
2176 enum btrfs_orphan_cleanup_state {
2177         ORPHAN_CLEANUP_STARTED  = 1,
2178         ORPHAN_CLEANUP_DONE     = 2,
2179 };
2180
2181 /*
2182  * This is called in transaction commit time. If there are no orphan
2183  * files in the subvolume, it removes orphan item and frees block_rsv
2184  * structure.
2185  */
2186 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2187                               struct btrfs_root *root)
2188 {
2189         struct btrfs_block_rsv *block_rsv;
2190         int ret;
2191
2192         if (atomic_read(&root->orphan_inodes) ||
2193             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2194                 return;
2195
2196         spin_lock(&root->orphan_lock);
2197         if (atomic_read(&root->orphan_inodes)) {
2198                 spin_unlock(&root->orphan_lock);
2199                 return;
2200         }
2201
2202         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2203                 spin_unlock(&root->orphan_lock);
2204                 return;
2205         }
2206
2207         block_rsv = root->orphan_block_rsv;
2208         root->orphan_block_rsv = NULL;
2209         spin_unlock(&root->orphan_lock);
2210
2211         if (root->orphan_item_inserted &&
2212             btrfs_root_refs(&root->root_item) > 0) {
2213                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2214                                             root->root_key.objectid);
2215                 BUG_ON(ret);
2216                 root->orphan_item_inserted = 0;
2217         }
2218
2219         if (block_rsv) {
2220                 WARN_ON(block_rsv->size > 0);
2221                 btrfs_free_block_rsv(root, block_rsv);
2222         }
2223 }
2224
2225 /*
2226  * This creates an orphan entry for the given inode in case something goes
2227  * wrong in the middle of an unlink/truncate.
2228  *
2229  * NOTE: caller of this function should reserve 5 units of metadata for
2230  *       this function.
2231  */
2232 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2233 {
2234         struct btrfs_root *root = BTRFS_I(inode)->root;
2235         struct btrfs_block_rsv *block_rsv = NULL;
2236         int reserve = 0;
2237         int insert = 0;
2238         int ret;
2239
2240         if (!root->orphan_block_rsv) {
2241                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2242                 if (!block_rsv)
2243                         return -ENOMEM;
2244         }
2245
2246         spin_lock(&root->orphan_lock);
2247         if (!root->orphan_block_rsv) {
2248                 root->orphan_block_rsv = block_rsv;
2249         } else if (block_rsv) {
2250                 btrfs_free_block_rsv(root, block_rsv);
2251                 block_rsv = NULL;
2252         }
2253
2254         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2255                               &BTRFS_I(inode)->runtime_flags)) {
2256 #if 0
2257                 /*
2258                  * For proper ENOSPC handling, we should do orphan
2259                  * cleanup when mounting. But this introduces backward
2260                  * compatibility issue.
2261                  */
2262                 if (!xchg(&root->orphan_item_inserted, 1))
2263                         insert = 2;
2264                 else
2265                         insert = 1;
2266 #endif
2267                 insert = 1;
2268                 atomic_inc(&root->orphan_inodes);
2269         }
2270
2271         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2272                               &BTRFS_I(inode)->runtime_flags))
2273                 reserve = 1;
2274         spin_unlock(&root->orphan_lock);
2275
2276         /* grab metadata reservation from transaction handle */
2277         if (reserve) {
2278                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2279                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2280         }
2281
2282         /* insert an orphan item to track this unlinked/truncated file */
2283         if (insert >= 1) {
2284                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2285                 if (ret && ret != -EEXIST) {
2286                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2287                                   &BTRFS_I(inode)->runtime_flags);
2288                         btrfs_abort_transaction(trans, root, ret);
2289                         return ret;
2290                 }
2291                 ret = 0;
2292         }
2293
2294         /* insert an orphan item to track subvolume contains orphan files */
2295         if (insert >= 2) {
2296                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2297                                                root->root_key.objectid);
2298                 if (ret && ret != -EEXIST) {
2299                         btrfs_abort_transaction(trans, root, ret);
2300                         return ret;
2301                 }
2302         }
2303         return 0;
2304 }
2305
2306 /*
2307  * We have done the truncate/delete so we can go ahead and remove the orphan
2308  * item for this particular inode.
2309  */
2310 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2311 {
2312         struct btrfs_root *root = BTRFS_I(inode)->root;
2313         int delete_item = 0;
2314         int release_rsv = 0;
2315         int ret = 0;
2316
2317         spin_lock(&root->orphan_lock);
2318         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2319                                &BTRFS_I(inode)->runtime_flags))
2320                 delete_item = 1;
2321
2322         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2323                                &BTRFS_I(inode)->runtime_flags))
2324                 release_rsv = 1;
2325         spin_unlock(&root->orphan_lock);
2326
2327         if (trans && delete_item) {
2328                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2329                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2330         }
2331
2332         if (release_rsv) {
2333                 btrfs_orphan_release_metadata(inode);
2334                 atomic_dec(&root->orphan_inodes);
2335         }
2336
2337         return 0;
2338 }
2339
2340 /*
2341  * this cleans up any orphans that may be left on the list from the last use
2342  * of this root.
2343  */
2344 int btrfs_orphan_cleanup(struct btrfs_root *root)
2345 {
2346         struct btrfs_path *path;
2347         struct extent_buffer *leaf;
2348         struct btrfs_key key, found_key;
2349         struct btrfs_trans_handle *trans;
2350         struct inode *inode;
2351         u64 last_objectid = 0;
2352         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2353
2354         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2355                 return 0;
2356
2357         path = btrfs_alloc_path();
2358         if (!path) {
2359                 ret = -ENOMEM;
2360                 goto out;
2361         }
2362         path->reada = -1;
2363
2364         key.objectid = BTRFS_ORPHAN_OBJECTID;
2365         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2366         key.offset = (u64)-1;
2367
2368         while (1) {
2369                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2370                 if (ret < 0)
2371                         goto out;
2372
2373                 /*
2374                  * if ret == 0 means we found what we were searching for, which
2375                  * is weird, but possible, so only screw with path if we didn't
2376                  * find the key and see if we have stuff that matches
2377                  */
2378                 if (ret > 0) {
2379                         ret = 0;
2380                         if (path->slots[0] == 0)
2381                                 break;
2382                         path->slots[0]--;
2383                 }
2384
2385                 /* pull out the item */
2386                 leaf = path->nodes[0];
2387                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2388
2389                 /* make sure the item matches what we want */
2390                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2391                         break;
2392                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2393                         break;
2394
2395                 /* release the path since we're done with it */
2396                 btrfs_release_path(path);
2397
2398                 /*
2399                  * this is where we are basically btrfs_lookup, without the
2400                  * crossing root thing.  we store the inode number in the
2401                  * offset of the orphan item.
2402                  */
2403
2404                 if (found_key.offset == last_objectid) {
2405                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2406                                "stopping orphan cleanup\n");
2407                         ret = -EINVAL;
2408                         goto out;
2409                 }
2410
2411                 last_objectid = found_key.offset;
2412
2413                 found_key.objectid = found_key.offset;
2414                 found_key.type = BTRFS_INODE_ITEM_KEY;
2415                 found_key.offset = 0;
2416                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2417                 ret = PTR_RET(inode);
2418                 if (ret && ret != -ESTALE)
2419                         goto out;
2420
2421                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2422                         struct btrfs_root *dead_root;
2423                         struct btrfs_fs_info *fs_info = root->fs_info;
2424                         int is_dead_root = 0;
2425
2426                         /*
2427                          * this is an orphan in the tree root. Currently these
2428                          * could come from 2 sources:
2429                          *  a) a snapshot deletion in progress
2430                          *  b) a free space cache inode
2431                          * We need to distinguish those two, as the snapshot
2432                          * orphan must not get deleted.
2433                          * find_dead_roots already ran before us, so if this
2434                          * is a snapshot deletion, we should find the root
2435                          * in the dead_roots list
2436                          */
2437                         spin_lock(&fs_info->trans_lock);
2438                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2439                                             root_list) {
2440                                 if (dead_root->root_key.objectid ==
2441                                     found_key.objectid) {
2442                                         is_dead_root = 1;
2443                                         break;
2444                                 }
2445                         }
2446                         spin_unlock(&fs_info->trans_lock);
2447                         if (is_dead_root) {
2448                                 /* prevent this orphan from being found again */
2449                                 key.offset = found_key.objectid - 1;
2450                                 continue;
2451                         }
2452                 }
2453                 /*
2454                  * Inode is already gone but the orphan item is still there,
2455                  * kill the orphan item.
2456                  */
2457                 if (ret == -ESTALE) {
2458                         trans = btrfs_start_transaction(root, 1);
2459                         if (IS_ERR(trans)) {
2460                                 ret = PTR_ERR(trans);
2461                                 goto out;
2462                         }
2463                         printk(KERN_ERR "auto deleting %Lu\n",
2464                                found_key.objectid);
2465                         ret = btrfs_del_orphan_item(trans, root,
2466                                                     found_key.objectid);
2467                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2468                         btrfs_end_transaction(trans, root);
2469                         continue;
2470                 }
2471
2472                 /*
2473                  * add this inode to the orphan list so btrfs_orphan_del does
2474                  * the proper thing when we hit it
2475                  */
2476                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2477                         &BTRFS_I(inode)->runtime_flags);
2478
2479                 /* if we have links, this was a truncate, lets do that */
2480                 if (inode->i_nlink) {
2481                         if (!S_ISREG(inode->i_mode)) {
2482                                 WARN_ON(1);
2483                                 iput(inode);
2484                                 continue;
2485                         }
2486                         nr_truncate++;
2487
2488                         /* 1 for the orphan item deletion. */
2489                         trans = btrfs_start_transaction(root, 1);
2490                         if (IS_ERR(trans)) {
2491                                 ret = PTR_ERR(trans);
2492                                 goto out;
2493                         }
2494                         ret = btrfs_orphan_add(trans, inode);
2495                         btrfs_end_transaction(trans, root);
2496                         if (ret)
2497                                 goto out;
2498
2499                         ret = btrfs_truncate(inode);
2500                 } else {
2501                         nr_unlink++;
2502                 }
2503
2504                 /* this will do delete_inode and everything for us */
2505                 iput(inode);
2506                 if (ret)
2507                         goto out;
2508         }
2509         /* release the path since we're done with it */
2510         btrfs_release_path(path);
2511
2512         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2513
2514         if (root->orphan_block_rsv)
2515                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2516                                         (u64)-1);
2517
2518         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2519                 trans = btrfs_join_transaction(root);
2520                 if (!IS_ERR(trans))
2521                         btrfs_end_transaction(trans, root);
2522         }
2523
2524         if (nr_unlink)
2525                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2526         if (nr_truncate)
2527                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2528
2529 out:
2530         if (ret)
2531                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2532         btrfs_free_path(path);
2533         return ret;
2534 }
2535
2536 /*
2537  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2538  * don't find any xattrs, we know there can't be any acls.
2539  *
2540  * slot is the slot the inode is in, objectid is the objectid of the inode
2541  */
2542 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2543                                           int slot, u64 objectid)
2544 {
2545         u32 nritems = btrfs_header_nritems(leaf);
2546         struct btrfs_key found_key;
2547         int scanned = 0;
2548
2549         slot++;
2550         while (slot < nritems) {
2551                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2552
2553                 /* we found a different objectid, there must not be acls */
2554                 if (found_key.objectid != objectid)
2555                         return 0;
2556
2557                 /* we found an xattr, assume we've got an acl */
2558                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2559                         return 1;
2560
2561                 /*
2562                  * we found a key greater than an xattr key, there can't
2563                  * be any acls later on
2564                  */
2565                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2566                         return 0;
2567
2568                 slot++;
2569                 scanned++;
2570
2571                 /*
2572                  * it goes inode, inode backrefs, xattrs, extents,
2573                  * so if there are a ton of hard links to an inode there can
2574                  * be a lot of backrefs.  Don't waste time searching too hard,
2575                  * this is just an optimization
2576                  */
2577                 if (scanned >= 8)
2578                         break;
2579         }
2580         /* we hit the end of the leaf before we found an xattr or
2581          * something larger than an xattr.  We have to assume the inode
2582          * has acls
2583          */
2584         return 1;
2585 }
2586
2587 /*
2588  * read an inode from the btree into the in-memory inode
2589  */
2590 static void btrfs_read_locked_inode(struct inode *inode)
2591 {
2592         struct btrfs_path *path;
2593         struct extent_buffer *leaf;
2594         struct btrfs_inode_item *inode_item;
2595         struct btrfs_timespec *tspec;
2596         struct btrfs_root *root = BTRFS_I(inode)->root;
2597         struct btrfs_key location;
2598         int maybe_acls;
2599         u32 rdev;
2600         int ret;
2601         bool filled = false;
2602
2603         ret = btrfs_fill_inode(inode, &rdev);
2604         if (!ret)
2605                 filled = true;
2606
2607         path = btrfs_alloc_path();
2608         if (!path)
2609                 goto make_bad;
2610
2611         path->leave_spinning = 1;
2612         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2613
2614         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2615         if (ret)
2616                 goto make_bad;
2617
2618         leaf = path->nodes[0];
2619
2620         if (filled)
2621                 goto cache_acl;
2622
2623         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2624                                     struct btrfs_inode_item);
2625         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2626         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2627         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2628         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2629         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2630
2631         tspec = btrfs_inode_atime(inode_item);
2632         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2633         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2634
2635         tspec = btrfs_inode_mtime(inode_item);
2636         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2637         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2638
2639         tspec = btrfs_inode_ctime(inode_item);
2640         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2641         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2642
2643         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2644         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2645         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2646
2647         /*
2648          * If we were modified in the current generation and evicted from memory
2649          * and then re-read we need to do a full sync since we don't have any
2650          * idea about which extents were modified before we were evicted from
2651          * cache.
2652          */
2653         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2654                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2655                         &BTRFS_I(inode)->runtime_flags);
2656
2657         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2658         inode->i_generation = BTRFS_I(inode)->generation;
2659         inode->i_rdev = 0;
2660         rdev = btrfs_inode_rdev(leaf, inode_item);
2661
2662         BTRFS_I(inode)->index_cnt = (u64)-1;
2663         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2664 cache_acl:
2665         /*
2666          * try to precache a NULL acl entry for files that don't have
2667          * any xattrs or acls
2668          */
2669         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2670                                            btrfs_ino(inode));
2671         if (!maybe_acls)
2672                 cache_no_acl(inode);
2673
2674         btrfs_free_path(path);
2675
2676         switch (inode->i_mode & S_IFMT) {
2677         case S_IFREG:
2678                 inode->i_mapping->a_ops = &btrfs_aops;
2679                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2680                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2681                 inode->i_fop = &btrfs_file_operations;
2682                 inode->i_op = &btrfs_file_inode_operations;
2683                 break;
2684         case S_IFDIR:
2685                 inode->i_fop = &btrfs_dir_file_operations;
2686                 if (root == root->fs_info->tree_root)
2687                         inode->i_op = &btrfs_dir_ro_inode_operations;
2688                 else
2689                         inode->i_op = &btrfs_dir_inode_operations;
2690                 break;
2691         case S_IFLNK:
2692                 inode->i_op = &btrfs_symlink_inode_operations;
2693                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2694                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2695                 break;
2696         default:
2697                 inode->i_op = &btrfs_special_inode_operations;
2698                 init_special_inode(inode, inode->i_mode, rdev);
2699                 break;
2700         }
2701
2702         btrfs_update_iflags(inode);
2703         return;
2704
2705 make_bad:
2706         btrfs_free_path(path);
2707         make_bad_inode(inode);
2708 }
2709
2710 /*
2711  * given a leaf and an inode, copy the inode fields into the leaf
2712  */
2713 static void fill_inode_item(struct btrfs_trans_handle *trans,
2714                             struct extent_buffer *leaf,
2715                             struct btrfs_inode_item *item,
2716                             struct inode *inode)
2717 {
2718         struct btrfs_map_token token;
2719
2720         btrfs_init_map_token(&token);
2721
2722         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2723         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2724         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
2725                                    &token);
2726         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2727         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2728
2729         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2730                                      inode->i_atime.tv_sec, &token);
2731         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2732                                       inode->i_atime.tv_nsec, &token);
2733
2734         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2735                                      inode->i_mtime.tv_sec, &token);
2736         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2737                                       inode->i_mtime.tv_nsec, &token);
2738
2739         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2740                                      inode->i_ctime.tv_sec, &token);
2741         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2742                                       inode->i_ctime.tv_nsec, &token);
2743
2744         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2745                                      &token);
2746         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
2747                                          &token);
2748         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2749         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2750         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
2751         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
2752         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
2753 }
2754
2755 /*
2756  * copy everything in the in-memory inode into the btree.
2757  */
2758 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2759                                 struct btrfs_root *root, struct inode *inode)
2760 {
2761         struct btrfs_inode_item *inode_item;
2762         struct btrfs_path *path;
2763         struct extent_buffer *leaf;
2764         int ret;
2765
2766         path = btrfs_alloc_path();
2767         if (!path)
2768                 return -ENOMEM;
2769
2770         path->leave_spinning = 1;
2771         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2772                                  1);
2773         if (ret) {
2774                 if (ret > 0)
2775                         ret = -ENOENT;
2776                 goto failed;
2777         }
2778
2779         btrfs_unlock_up_safe(path, 1);
2780         leaf = path->nodes[0];
2781         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2782                                     struct btrfs_inode_item);
2783
2784         fill_inode_item(trans, leaf, inode_item, inode);
2785         btrfs_mark_buffer_dirty(leaf);
2786         btrfs_set_inode_last_trans(trans, inode);
2787         ret = 0;
2788 failed:
2789         btrfs_free_path(path);
2790         return ret;
2791 }
2792
2793 /*
2794  * copy everything in the in-memory inode into the btree.
2795  */
2796 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2797                                 struct btrfs_root *root, struct inode *inode)
2798 {
2799         int ret;
2800
2801         /*
2802          * If the inode is a free space inode, we can deadlock during commit
2803          * if we put it into the delayed code.
2804          *
2805          * The data relocation inode should also be directly updated
2806          * without delay
2807          */
2808         if (!btrfs_is_free_space_inode(inode)
2809             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2810                 btrfs_update_root_times(trans, root);
2811
2812                 ret = btrfs_delayed_update_inode(trans, root, inode);
2813                 if (!ret)
2814                         btrfs_set_inode_last_trans(trans, inode);
2815                 return ret;
2816         }
2817
2818         return btrfs_update_inode_item(trans, root, inode);
2819 }
2820
2821 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2822                                          struct btrfs_root *root,
2823                                          struct inode *inode)
2824 {
2825         int ret;
2826
2827         ret = btrfs_update_inode(trans, root, inode);
2828         if (ret == -ENOSPC)
2829                 return btrfs_update_inode_item(trans, root, inode);
2830         return ret;
2831 }
2832
2833 /*
2834  * unlink helper that gets used here in inode.c and in the tree logging
2835  * recovery code.  It remove a link in a directory with a given name, and
2836  * also drops the back refs in the inode to the directory
2837  */
2838 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2839                                 struct btrfs_root *root,
2840                                 struct inode *dir, struct inode *inode,
2841                                 const char *name, int name_len)
2842 {
2843         struct btrfs_path *path;
2844         int ret = 0;
2845         struct extent_buffer *leaf;
2846         struct btrfs_dir_item *di;
2847         struct btrfs_key key;
2848         u64 index;
2849         u64 ino = btrfs_ino(inode);
2850         u64 dir_ino = btrfs_ino(dir);
2851
2852         path = btrfs_alloc_path();
2853         if (!path) {
2854                 ret = -ENOMEM;
2855                 goto out;
2856         }
2857
2858         path->leave_spinning = 1;
2859         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2860                                     name, name_len, -1);
2861         if (IS_ERR(di)) {
2862                 ret = PTR_ERR(di);
2863                 goto err;
2864         }
2865         if (!di) {
2866                 ret = -ENOENT;
2867                 goto err;
2868         }
2869         leaf = path->nodes[0];
2870         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2871         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2872         if (ret)
2873                 goto err;
2874         btrfs_release_path(path);
2875
2876         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2877                                   dir_ino, &index);
2878         if (ret) {
2879                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2880                        "inode %llu parent %llu\n", name_len, name,
2881                        (unsigned long long)ino, (unsigned long long)dir_ino);
2882                 btrfs_abort_transaction(trans, root, ret);
2883                 goto err;
2884         }
2885
2886         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2887         if (ret) {
2888                 btrfs_abort_transaction(trans, root, ret);
2889                 goto err;
2890         }
2891
2892         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2893                                          inode, dir_ino);
2894         if (ret != 0 && ret != -ENOENT) {
2895                 btrfs_abort_transaction(trans, root, ret);
2896                 goto err;
2897         }
2898
2899         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2900                                            dir, index);
2901         if (ret == -ENOENT)
2902                 ret = 0;
2903 err:
2904         btrfs_free_path(path);
2905         if (ret)
2906                 goto out;
2907
2908         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2909         inode_inc_iversion(inode);
2910         inode_inc_iversion(dir);
2911         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2912         ret = btrfs_update_inode(trans, root, dir);
2913 out:
2914         return ret;
2915 }
2916
2917 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2918                        struct btrfs_root *root,
2919                        struct inode *dir, struct inode *inode,
2920                        const char *name, int name_len)
2921 {
2922         int ret;
2923         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2924         if (!ret) {
2925                 btrfs_drop_nlink(inode);
2926                 ret = btrfs_update_inode(trans, root, inode);
2927         }
2928         return ret;
2929 }
2930
2931
2932 /* helper to check if there is any shared block in the path */
2933 static int check_path_shared(struct btrfs_root *root,
2934                              struct btrfs_path *path)
2935 {
2936         struct extent_buffer *eb;
2937         int level;
2938         u64 refs = 1;
2939
2940         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2941                 int ret;
2942
2943                 if (!path->nodes[level])
2944                         break;
2945                 eb = path->nodes[level];
2946                 if (!btrfs_block_can_be_shared(root, eb))
2947                         continue;
2948                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2949                                                &refs, NULL);
2950                 if (refs > 1)
2951                         return 1;
2952         }
2953         return 0;
2954 }
2955
2956 /*
2957  * helper to start transaction for unlink and rmdir.
2958  *
2959  * unlink and rmdir are special in btrfs, they do not always free space.
2960  * so in enospc case, we should make sure they will free space before
2961  * allowing them to use the global metadata reservation.
2962  */
2963 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2964                                                        struct dentry *dentry)
2965 {
2966         struct btrfs_trans_handle *trans;
2967         struct btrfs_root *root = BTRFS_I(dir)->root;
2968         struct btrfs_path *path;
2969         struct btrfs_dir_item *di;
2970         struct inode *inode = dentry->d_inode;
2971         u64 index;
2972         int check_link = 1;
2973         int err = -ENOSPC;
2974         int ret;
2975         u64 ino = btrfs_ino(inode);
2976         u64 dir_ino = btrfs_ino(dir);
2977
2978         /*
2979          * 1 for the possible orphan item
2980          * 1 for the dir item
2981          * 1 for the dir index
2982          * 1 for the inode ref
2983          * 1 for the inode ref in the tree log
2984          * 2 for the dir entries in the log
2985          * 1 for the inode
2986          */
2987         trans = btrfs_start_transaction(root, 8);
2988         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2989                 return trans;
2990
2991         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2992                 return ERR_PTR(-ENOSPC);
2993
2994         /* check if there is someone else holds reference */
2995         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2996                 return ERR_PTR(-ENOSPC);
2997
2998         if (atomic_read(&inode->i_count) > 2)
2999                 return ERR_PTR(-ENOSPC);
3000
3001         if (xchg(&root->fs_info->enospc_unlink, 1))
3002                 return ERR_PTR(-ENOSPC);
3003
3004         path = btrfs_alloc_path();
3005         if (!path) {
3006                 root->fs_info->enospc_unlink = 0;
3007                 return ERR_PTR(-ENOMEM);
3008         }
3009
3010         /* 1 for the orphan item */
3011         trans = btrfs_start_transaction(root, 1);
3012         if (IS_ERR(trans)) {
3013                 btrfs_free_path(path);
3014                 root->fs_info->enospc_unlink = 0;
3015                 return trans;
3016         }
3017
3018         path->skip_locking = 1;
3019         path->search_commit_root = 1;
3020
3021         ret = btrfs_lookup_inode(trans, root, path,
3022                                 &BTRFS_I(dir)->location, 0);
3023         if (ret < 0) {
3024                 err = ret;
3025                 goto out;
3026         }
3027         if (ret == 0) {
3028                 if (check_path_shared(root, path))
3029                         goto out;
3030         } else {
3031                 check_link = 0;
3032         }
3033         btrfs_release_path(path);
3034
3035         ret = btrfs_lookup_inode(trans, root, path,
3036                                 &BTRFS_I(inode)->location, 0);
3037         if (ret < 0) {
3038                 err = ret;
3039                 goto out;
3040         }
3041         if (ret == 0) {
3042                 if (check_path_shared(root, path))
3043                         goto out;
3044         } else {
3045                 check_link = 0;
3046         }
3047         btrfs_release_path(path);
3048
3049         if (ret == 0 && S_ISREG(inode->i_mode)) {
3050                 ret = btrfs_lookup_file_extent(trans, root, path,
3051                                                ino, (u64)-1, 0);
3052                 if (ret < 0) {
3053                         err = ret;
3054                         goto out;
3055                 }
3056                 BUG_ON(ret == 0); /* Corruption */
3057                 if (check_path_shared(root, path))
3058                         goto out;
3059                 btrfs_release_path(path);
3060         }
3061
3062         if (!check_link) {
3063                 err = 0;
3064                 goto out;
3065         }
3066
3067         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3068                                 dentry->d_name.name, dentry->d_name.len, 0);
3069         if (IS_ERR(di)) {
3070                 err = PTR_ERR(di);
3071                 goto out;
3072         }
3073         if (di) {
3074                 if (check_path_shared(root, path))
3075                         goto out;
3076         } else {
3077                 err = 0;
3078                 goto out;
3079         }
3080         btrfs_release_path(path);
3081
3082         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3083                                         dentry->d_name.len, ino, dir_ino, 0,
3084                                         &index);
3085         if (ret) {
3086                 err = ret;
3087                 goto out;
3088         }
3089
3090         if (check_path_shared(root, path))
3091                 goto out;
3092
3093         btrfs_release_path(path);
3094
3095         /*
3096          * This is a commit root search, if we can lookup inode item and other
3097          * relative items in the commit root, it means the transaction of
3098          * dir/file creation has been committed, and the dir index item that we
3099          * delay to insert has also been inserted into the commit root. So
3100          * we needn't worry about the delayed insertion of the dir index item
3101          * here.
3102          */
3103         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3104                                 dentry->d_name.name, dentry->d_name.len, 0);
3105         if (IS_ERR(di)) {
3106                 err = PTR_ERR(di);
3107                 goto out;
3108         }
3109         BUG_ON(ret == -ENOENT);
3110         if (check_path_shared(root, path))
3111                 goto out;
3112
3113         err = 0;
3114 out:
3115         btrfs_free_path(path);
3116         /* Migrate the orphan reservation over */
3117         if (!err)
3118                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3119                                 &root->fs_info->global_block_rsv,
3120                                 trans->bytes_reserved);
3121
3122         if (err) {
3123                 btrfs_end_transaction(trans, root);
3124                 root->fs_info->enospc_unlink = 0;
3125                 return ERR_PTR(err);
3126         }
3127
3128         trans->block_rsv = &root->fs_info->global_block_rsv;
3129         return trans;
3130 }
3131
3132 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3133                                struct btrfs_root *root)
3134 {
3135         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3136                 btrfs_block_rsv_release(root, trans->block_rsv,
3137                                         trans->bytes_reserved);
3138                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3139                 BUG_ON(!root->fs_info->enospc_unlink);
3140                 root->fs_info->enospc_unlink = 0;
3141         }
3142         btrfs_end_transaction(trans, root);
3143 }
3144
3145 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3146 {
3147         struct btrfs_root *root = BTRFS_I(dir)->root;
3148         struct btrfs_trans_handle *trans;
3149         struct inode *inode = dentry->d_inode;
3150         int ret;
3151
3152         trans = __unlink_start_trans(dir, dentry);
3153         if (IS_ERR(trans))
3154                 return PTR_ERR(trans);
3155
3156         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3157
3158         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3159                                  dentry->d_name.name, dentry->d_name.len);
3160         if (ret)
3161                 goto out;
3162
3163         if (inode->i_nlink == 0) {
3164                 ret = btrfs_orphan_add(trans, inode);
3165                 if (ret)
3166                         goto out;
3167         }
3168
3169 out:
3170         __unlink_end_trans(trans, root);
3171         btrfs_btree_balance_dirty(root);
3172         return ret;
3173 }
3174
3175 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3176                         struct btrfs_root *root,
3177                         struct inode *dir, u64 objectid,
3178                         const char *name, int name_len)
3179 {
3180         struct btrfs_path *path;
3181         struct extent_buffer *leaf;
3182         struct btrfs_dir_item *di;
3183         struct btrfs_key key;
3184         u64 index;
3185         int ret;
3186         u64 dir_ino = btrfs_ino(dir);
3187
3188         path = btrfs_alloc_path();
3189         if (!path)
3190                 return -ENOMEM;
3191
3192         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3193                                    name, name_len, -1);
3194         if (IS_ERR_OR_NULL(di)) {
3195                 if (!di)
3196                         ret = -ENOENT;
3197                 else
3198                         ret = PTR_ERR(di);
3199                 goto out;
3200         }
3201
3202         leaf = path->nodes[0];
3203         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3204         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3205         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3206         if (ret) {
3207                 btrfs_abort_transaction(trans, root, ret);
3208                 goto out;
3209         }
3210         btrfs_release_path(path);
3211
3212         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3213                                  objectid, root->root_key.objectid,
3214                                  dir_ino, &index, name, name_len);
3215         if (ret < 0) {
3216                 if (ret != -ENOENT) {
3217                         btrfs_abort_transaction(trans, root, ret);
3218                         goto out;
3219                 }
3220                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3221                                                  name, name_len);
3222                 if (IS_ERR_OR_NULL(di)) {
3223                         if (!di)
3224                                 ret = -ENOENT;
3225                         else
3226                                 ret = PTR_ERR(di);
3227                         btrfs_abort_transaction(trans, root, ret);
3228                         goto out;
3229                 }
3230
3231                 leaf = path->nodes[0];
3232                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3233                 btrfs_release_path(path);
3234                 index = key.offset;
3235         }
3236         btrfs_release_path(path);
3237
3238         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3239         if (ret) {
3240                 btrfs_abort_transaction(trans, root, ret);
3241                 goto out;
3242         }
3243
3244         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3245         inode_inc_iversion(dir);
3246         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3247         ret = btrfs_update_inode_fallback(trans, root, dir);
3248         if (ret)
3249                 btrfs_abort_transaction(trans, root, ret);
3250 out:
3251         btrfs_free_path(path);
3252         return ret;
3253 }
3254
3255 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3256 {
3257         struct inode *inode = dentry->d_inode;
3258         int err = 0;
3259         struct btrfs_root *root = BTRFS_I(dir)->root;
3260         struct btrfs_trans_handle *trans;
3261
3262         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3263                 return -ENOTEMPTY;
3264         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3265                 return -EPERM;
3266
3267         trans = __unlink_start_trans(dir, dentry);
3268         if (IS_ERR(trans))
3269                 return PTR_ERR(trans);
3270
3271         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3272                 err = btrfs_unlink_subvol(trans, root, dir,
3273                                           BTRFS_I(inode)->location.objectid,
3274                                           dentry->d_name.name,
3275                                           dentry->d_name.len);
3276                 goto out;
3277         }
3278
3279         err = btrfs_orphan_add(trans, inode);
3280         if (err)
3281                 goto out;
3282
3283         /* now the directory is empty */
3284         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3285                                  dentry->d_name.name, dentry->d_name.len);
3286         if (!err)
3287                 btrfs_i_size_write(inode, 0);
3288 out:
3289         __unlink_end_trans(trans, root);
3290         btrfs_btree_balance_dirty(root);
3291
3292         return err;
3293 }
3294
3295 /*
3296  * this can truncate away extent items, csum items and directory items.
3297  * It starts at a high offset and removes keys until it can't find
3298  * any higher than new_size
3299  *
3300  * csum items that cross the new i_size are truncated to the new size
3301  * as well.
3302  *
3303  * min_type is the minimum key type to truncate down to.  If set to 0, this
3304  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3305  */
3306 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3307                                struct btrfs_root *root,
3308                                struct inode *inode,
3309                                u64 new_size, u32 min_type)
3310 {
3311         struct btrfs_path *path;
3312         struct extent_buffer *leaf;
3313         struct btrfs_file_extent_item *fi;
3314         struct btrfs_key key;
3315         struct btrfs_key found_key;
3316         u64 extent_start = 0;
3317         u64 extent_num_bytes = 0;
3318         u64 extent_offset = 0;
3319         u64 item_end = 0;
3320         u64 mask = root->sectorsize - 1;
3321         u32 found_type = (u8)-1;
3322         int found_extent;
3323         int del_item;
3324         int pending_del_nr = 0;
3325         int pending_del_slot = 0;
3326         int extent_type = -1;
3327         int ret;
3328         int err = 0;
3329         u64 ino = btrfs_ino(inode);
3330
3331         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3332
3333         path = btrfs_alloc_path();
3334         if (!path)
3335                 return -ENOMEM;
3336         path->reada = -1;
3337
3338         /*
3339          * We want to drop from the next block forward in case this new size is
3340          * not block aligned since we will be keeping the last block of the
3341          * extent just the way it is.
3342          */
3343         if (root->ref_cows || root == root->fs_info->tree_root)
3344                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3345
3346         /*
3347          * This function is also used to drop the items in the log tree before
3348          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3349          * it is used to drop the loged items. So we shouldn't kill the delayed
3350          * items.
3351          */
3352         if (min_type == 0 && root == BTRFS_I(inode)->root)
3353                 btrfs_kill_delayed_inode_items(inode);
3354
3355         key.objectid = ino;
3356         key.offset = (u64)-1;
3357         key.type = (u8)-1;
3358
3359 search_again:
3360         path->leave_spinning = 1;
3361         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3362         if (ret < 0) {
3363                 err = ret;
3364                 goto out;
3365         }
3366
3367         if (ret > 0) {
3368                 /* there are no items in the tree for us to truncate, we're
3369                  * done
3370                  */
3371                 if (path->slots[0] == 0)
3372                         goto out;
3373                 path->slots[0]--;
3374         }
3375
3376         while (1) {
3377                 fi = NULL;
3378                 leaf = path->nodes[0];
3379                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3380                 found_type = btrfs_key_type(&found_key);
3381
3382                 if (found_key.objectid != ino)
3383                         break;
3384
3385                 if (found_type < min_type)
3386                         break;
3387
3388                 item_end = found_key.offset;
3389                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3390                         fi = btrfs_item_ptr(leaf, path->slots[0],
3391                                             struct btrfs_file_extent_item);
3392                         extent_type = btrfs_file_extent_type(leaf, fi);
3393                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3394                                 item_end +=
3395                                     btrfs_file_extent_num_bytes(leaf, fi);
3396                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3397                                 item_end += btrfs_file_extent_inline_len(leaf,
3398                                                                          fi);
3399                         }
3400                         item_end--;
3401                 }
3402                 if (found_type > min_type) {
3403                         del_item = 1;
3404                 } else {
3405                         if (item_end < new_size)
3406                                 break;
3407                         if (found_key.offset >= new_size)
3408                                 del_item = 1;
3409                         else
3410                                 del_item = 0;
3411                 }
3412                 found_extent = 0;
3413                 /* FIXME, shrink the extent if the ref count is only 1 */
3414                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3415                         goto delete;
3416
3417                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3418                         u64 num_dec;
3419                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3420                         if (!del_item) {
3421                                 u64 orig_num_bytes =
3422                                         btrfs_file_extent_num_bytes(leaf, fi);
3423                                 extent_num_bytes = new_size -
3424                                         found_key.offset + root->sectorsize - 1;
3425                                 extent_num_bytes = extent_num_bytes &
3426                                         ~((u64)root->sectorsize - 1);
3427                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3428                                                          extent_num_bytes);
3429                                 num_dec = (orig_num_bytes -
3430                                            extent_num_bytes);
3431                                 if (root->ref_cows && extent_start != 0)
3432                                         inode_sub_bytes(inode, num_dec);
3433                                 btrfs_mark_buffer_dirty(leaf);
3434                         } else {
3435                                 extent_num_bytes =
3436                                         btrfs_file_extent_disk_num_bytes(leaf,
3437                                                                          fi);
3438                                 extent_offset = found_key.offset -
3439                                         btrfs_file_extent_offset(leaf, fi);
3440
3441                                 /* FIXME blocksize != 4096 */
3442                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3443                                 if (extent_start != 0) {
3444                                         found_extent = 1;
3445                                         if (root->ref_cows)
3446                                                 inode_sub_bytes(inode, num_dec);
3447                                 }
3448                         }
3449                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3450                         /*
3451                          * we can't truncate inline items that have had
3452                          * special encodings
3453                          */
3454                         if (!del_item &&
3455                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3456                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3457                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3458                                 u32 size = new_size - found_key.offset;
3459
3460                                 if (root->ref_cows) {
3461                                         inode_sub_bytes(inode, item_end + 1 -
3462                                                         new_size);
3463                                 }
3464                                 size =
3465                                     btrfs_file_extent_calc_inline_size(size);
3466                                 btrfs_truncate_item(trans, root, path,
3467                                                     size, 1);
3468                         } else if (root->ref_cows) {
3469                                 inode_sub_bytes(inode, item_end + 1 -
3470                                                 found_key.offset);
3471                         }
3472                 }
3473 delete:
3474                 if (del_item) {
3475                         if (!pending_del_nr) {
3476                                 /* no pending yet, add ourselves */
3477                                 pending_del_slot = path->slots[0];
3478                                 pending_del_nr = 1;
3479                         } else if (pending_del_nr &&
3480                                    path->slots[0] + 1 == pending_del_slot) {
3481                                 /* hop on the pending chunk */
3482                                 pending_del_nr++;
3483                                 pending_del_slot = path->slots[0];
3484                         } else {
3485                                 BUG();
3486                         }
3487                 } else {
3488                         break;
3489                 }
3490                 if (found_extent && (root->ref_cows ||
3491                                      root == root->fs_info->tree_root)) {
3492                         btrfs_set_path_blocking(path);
3493                         ret = btrfs_free_extent(trans, root, extent_start,
3494                                                 extent_num_bytes, 0,
3495                                                 btrfs_header_owner(leaf),
3496                                                 ino, extent_offset, 0);
3497                         BUG_ON(ret);
3498                 }
3499
3500                 if (found_type == BTRFS_INODE_ITEM_KEY)
3501                         break;
3502
3503                 if (path->slots[0] == 0 ||
3504                     path->slots[0] != pending_del_slot) {
3505                         if (pending_del_nr) {
3506                                 ret = btrfs_del_items(trans, root, path,
3507                                                 pending_del_slot,
3508                                                 pending_del_nr);
3509                                 if (ret) {
3510                                         btrfs_abort_transaction(trans,
3511                                                                 root, ret);
3512                                         goto error;
3513                                 }
3514                                 pending_del_nr = 0;
3515                         }
3516                         btrfs_release_path(path);
3517                         goto search_again;
3518                 } else {
3519                         path->slots[0]--;
3520                 }
3521         }
3522 out:
3523         if (pending_del_nr) {
3524                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3525                                       pending_del_nr);
3526                 if (ret)
3527                         btrfs_abort_transaction(trans, root, ret);
3528         }
3529 error:
3530         btrfs_free_path(path);
3531         return err;
3532 }
3533
3534 /*
3535  * btrfs_truncate_page - read, zero a chunk and write a page
3536  * @inode - inode that we're zeroing
3537  * @from - the offset to start zeroing
3538  * @len - the length to zero, 0 to zero the entire range respective to the
3539  *      offset
3540  * @front - zero up to the offset instead of from the offset on
3541  *
3542  * This will find the page for the "from" offset and cow the page and zero the
3543  * part we want to zero.  This is used with truncate and hole punching.
3544  */
3545 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3546                         int front)
3547 {
3548         struct address_space *mapping = inode->i_mapping;
3549         struct btrfs_root *root = BTRFS_I(inode)->root;
3550         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3551         struct btrfs_ordered_extent *ordered;
3552         struct extent_state *cached_state = NULL;
3553         char *kaddr;
3554         u32 blocksize = root->sectorsize;
3555         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3556         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3557         struct page *page;
3558         gfp_t mask = btrfs_alloc_write_mask(mapping);
3559         int ret = 0;
3560         u64 page_start;
3561         u64 page_end;
3562
3563         if ((offset & (blocksize - 1)) == 0 &&
3564             (!len || ((len & (blocksize - 1)) == 0)))
3565                 goto out;
3566         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3567         if (ret)
3568                 goto out;
3569
3570 again:
3571         page = find_or_create_page(mapping, index, mask);
3572         if (!page) {
3573                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3574                 ret = -ENOMEM;
3575                 goto out;
3576         }
3577
3578         page_start = page_offset(page);
3579         page_end = page_start + PAGE_CACHE_SIZE - 1;
3580
3581         if (!PageUptodate(page)) {
3582                 ret = btrfs_readpage(NULL, page);
3583                 lock_page(page);
3584                 if (page->mapping != mapping) {
3585                         unlock_page(page);
3586                         page_cache_release(page);
3587                         goto again;
3588                 }
3589                 if (!PageUptodate(page)) {
3590                         ret = -EIO;
3591                         goto out_unlock;
3592                 }
3593         }
3594         wait_on_page_writeback(page);
3595
3596         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3597         set_page_extent_mapped(page);
3598
3599         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3600         if (ordered) {
3601                 unlock_extent_cached(io_tree, page_start, page_end,
3602                                      &cached_state, GFP_NOFS);
3603                 unlock_page(page);
3604                 page_cache_release(page);
3605                 btrfs_start_ordered_extent(inode, ordered, 1);
3606                 btrfs_put_ordered_extent(ordered);
3607                 goto again;
3608         }
3609
3610         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3611                           EXTENT_DIRTY | EXTENT_DELALLOC |
3612                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3613                           0, 0, &cached_state, GFP_NOFS);
3614
3615         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3616                                         &cached_state);
3617         if (ret) {
3618                 unlock_extent_cached(io_tree, page_start, page_end,
3619                                      &cached_state, GFP_NOFS);
3620                 goto out_unlock;
3621         }
3622
3623         if (offset != PAGE_CACHE_SIZE) {
3624                 if (!len)
3625                         len = PAGE_CACHE_SIZE - offset;
3626                 kaddr = kmap(page);
3627                 if (front)
3628                         memset(kaddr, 0, offset);
3629                 else
3630                         memset(kaddr + offset, 0, len);
3631                 flush_dcache_page(page);
3632                 kunmap(page);
3633         }
3634         ClearPageChecked(page);
3635         set_page_dirty(page);
3636         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3637                              GFP_NOFS);
3638
3639 out_unlock:
3640         if (ret)
3641                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3642         unlock_page(page);
3643         page_cache_release(page);
3644 out:
3645         return ret;
3646 }
3647
3648 /*
3649  * This function puts in dummy file extents for the area we're creating a hole
3650  * for.  So if we are truncating this file to a larger size we need to insert
3651  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3652  * the range between oldsize and size
3653  */
3654 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3655 {
3656         struct btrfs_trans_handle *trans;
3657         struct btrfs_root *root = BTRFS_I(inode)->root;
3658         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3659         struct extent_map *em = NULL;
3660         struct extent_state *cached_state = NULL;
3661         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3662         u64 mask = root->sectorsize - 1;
3663         u64 hole_start = (oldsize + mask) & ~mask;
3664         u64 block_end = (size + mask) & ~mask;
3665         u64 last_byte;
3666         u64 cur_offset;
3667         u64 hole_size;
3668         int err = 0;
3669
3670         if (size <= hole_start)
3671                 return 0;
3672
3673         while (1) {
3674                 struct btrfs_ordered_extent *ordered;
3675                 btrfs_wait_ordered_range(inode, hole_start,
3676                                          block_end - hole_start);
3677                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3678                                  &cached_state);
3679                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3680                 if (!ordered)
3681                         break;
3682                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3683                                      &cached_state, GFP_NOFS);
3684                 btrfs_put_ordered_extent(ordered);
3685         }
3686
3687         cur_offset = hole_start;
3688         while (1) {
3689                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3690                                 block_end - cur_offset, 0);
3691                 if (IS_ERR(em)) {
3692                         err = PTR_ERR(em);
3693                         em = NULL;
3694                         break;
3695                 }
3696                 last_byte = min(extent_map_end(em), block_end);
3697                 last_byte = (last_byte + mask) & ~mask;
3698                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3699                         struct extent_map *hole_em;
3700                         hole_size = last_byte - cur_offset;
3701
3702                         trans = btrfs_start_transaction(root, 3);
3703                         if (IS_ERR(trans)) {
3704                                 err = PTR_ERR(trans);
3705                                 break;
3706                         }
3707
3708                         err = btrfs_drop_extents(trans, root, inode,
3709                                                  cur_offset,
3710                                                  cur_offset + hole_size, 1);
3711                         if (err) {
3712                                 btrfs_abort_transaction(trans, root, err);
3713                                 btrfs_end_transaction(trans, root);
3714                                 break;
3715                         }
3716
3717                         err = btrfs_insert_file_extent(trans, root,
3718                                         btrfs_ino(inode), cur_offset, 0,
3719                                         0, hole_size, 0, hole_size,
3720                                         0, 0, 0);
3721                         if (err) {
3722                                 btrfs_abort_transaction(trans, root, err);
3723                                 btrfs_end_transaction(trans, root);
3724                                 break;
3725                         }
3726
3727                         btrfs_drop_extent_cache(inode, cur_offset,
3728                                                 cur_offset + hole_size - 1, 0);
3729                         hole_em = alloc_extent_map();
3730                         if (!hole_em) {
3731                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3732                                         &BTRFS_I(inode)->runtime_flags);
3733                                 goto next;
3734                         }
3735                         hole_em->start = cur_offset;
3736                         hole_em->len = hole_size;
3737                         hole_em->orig_start = cur_offset;
3738
3739                         hole_em->block_start = EXTENT_MAP_HOLE;
3740                         hole_em->block_len = 0;
3741                         hole_em->orig_block_len = 0;
3742                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3743                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3744                         hole_em->generation = trans->transid;
3745
3746                         while (1) {
3747                                 write_lock(&em_tree->lock);
3748                                 err = add_extent_mapping(em_tree, hole_em);
3749                                 if (!err)
3750                                         list_move(&hole_em->list,
3751                                                   &em_tree->modified_extents);
3752                                 write_unlock(&em_tree->lock);
3753                                 if (err != -EEXIST)
3754                                         break;
3755                                 btrfs_drop_extent_cache(inode, cur_offset,
3756                                                         cur_offset +
3757                                                         hole_size - 1, 0);
3758                         }
3759                         free_extent_map(hole_em);
3760 next:
3761                         btrfs_update_inode(trans, root, inode);
3762                         btrfs_end_transaction(trans, root);
3763                 }
3764                 free_extent_map(em);
3765                 em = NULL;
3766                 cur_offset = last_byte;
3767                 if (cur_offset >= block_end)
3768                         break;
3769         }
3770
3771         free_extent_map(em);
3772         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3773                              GFP_NOFS);
3774         return err;
3775 }
3776
3777 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
3778 {
3779         struct btrfs_root *root = BTRFS_I(inode)->root;
3780         struct btrfs_trans_handle *trans;
3781         loff_t oldsize = i_size_read(inode);
3782         loff_t newsize = attr->ia_size;
3783         int mask = attr->ia_valid;
3784         int ret;
3785
3786         if (newsize == oldsize)
3787                 return 0;
3788
3789         /*
3790          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3791          * special case where we need to update the times despite not having
3792          * these flags set.  For all other operations the VFS set these flags
3793          * explicitly if it wants a timestamp update.
3794          */
3795         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3796                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3797
3798         if (newsize > oldsize) {
3799                 truncate_pagecache(inode, oldsize, newsize);
3800                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3801                 if (ret)
3802                         return ret;
3803
3804                 trans = btrfs_start_transaction(root, 1);
3805                 if (IS_ERR(trans))
3806                         return PTR_ERR(trans);
3807
3808                 i_size_write(inode, newsize);
3809                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3810                 ret = btrfs_update_inode(trans, root, inode);
3811                 btrfs_end_transaction(trans, root);
3812         } else {
3813
3814                 /*
3815                  * We're truncating a file that used to have good data down to
3816                  * zero. Make sure it gets into the ordered flush list so that
3817                  * any new writes get down to disk quickly.
3818                  */
3819                 if (newsize == 0)
3820                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3821                                 &BTRFS_I(inode)->runtime_flags);
3822
3823                 /*
3824                  * 1 for the orphan item we're going to add
3825                  * 1 for the orphan item deletion.
3826                  */
3827                 trans = btrfs_start_transaction(root, 2);
3828                 if (IS_ERR(trans))
3829                         return PTR_ERR(trans);
3830
3831                 /*
3832                  * We need to do this in case we fail at _any_ point during the
3833                  * actual truncate.  Once we do the truncate_setsize we could
3834                  * invalidate pages which forces any outstanding ordered io to
3835                  * be instantly completed which will give us extents that need
3836                  * to be truncated.  If we fail to get an orphan inode down we
3837                  * could have left over extents that were never meant to live,
3838                  * so we need to garuntee from this point on that everything
3839                  * will be consistent.
3840                  */
3841                 ret = btrfs_orphan_add(trans, inode);
3842                 btrfs_end_transaction(trans, root);
3843                 if (ret)
3844                         return ret;
3845
3846                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3847                 truncate_setsize(inode, newsize);
3848                 ret = btrfs_truncate(inode);
3849                 if (ret && inode->i_nlink)
3850                         btrfs_orphan_del(NULL, inode);
3851         }
3852
3853         return ret;
3854 }
3855
3856 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3857 {
3858         struct inode *inode = dentry->d_inode;
3859         struct btrfs_root *root = BTRFS_I(inode)->root;
3860         int err;
3861
3862         if (btrfs_root_readonly(root))
3863                 return -EROFS;
3864
3865         err = inode_change_ok(inode, attr);
3866         if (err)
3867                 return err;
3868
3869         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3870                 err = btrfs_setsize(inode, attr);
3871                 if (err)
3872                         return err;
3873         }
3874
3875         if (attr->ia_valid) {
3876                 setattr_copy(inode, attr);
3877                 inode_inc_iversion(inode);
3878                 err = btrfs_dirty_inode(inode);
3879
3880                 if (!err && attr->ia_valid & ATTR_MODE)
3881                         err = btrfs_acl_chmod(inode);
3882         }
3883
3884         return err;
3885 }
3886
3887 void btrfs_evict_inode(struct inode *inode)
3888 {
3889         struct btrfs_trans_handle *trans;
3890         struct btrfs_root *root = BTRFS_I(inode)->root;
3891         struct btrfs_block_rsv *rsv, *global_rsv;
3892         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3893         int ret;
3894
3895         trace_btrfs_inode_evict(inode);
3896
3897         truncate_inode_pages(&inode->i_data, 0);
3898         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3899                                btrfs_is_free_space_inode(inode)))
3900                 goto no_delete;
3901
3902         if (is_bad_inode(inode)) {
3903                 btrfs_orphan_del(NULL, inode);
3904                 goto no_delete;
3905         }
3906         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3907         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3908
3909         if (root->fs_info->log_root_recovering) {
3910                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3911                                  &BTRFS_I(inode)->runtime_flags));
3912                 goto no_delete;
3913         }
3914
3915         if (inode->i_nlink > 0) {
3916                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3917                 goto no_delete;
3918         }
3919
3920         ret = btrfs_commit_inode_delayed_inode(inode);
3921         if (ret) {
3922                 btrfs_orphan_del(NULL, inode);
3923                 goto no_delete;
3924         }
3925
3926         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3927         if (!rsv) {
3928                 btrfs_orphan_del(NULL, inode);
3929                 goto no_delete;
3930         }
3931         rsv->size = min_size;
3932         rsv->failfast = 1;
3933         global_rsv = &root->fs_info->global_block_rsv;
3934
3935         btrfs_i_size_write(inode, 0);
3936
3937         /*
3938          * This is a bit simpler than btrfs_truncate since we've already
3939          * reserved our space for our orphan item in the unlink, so we just
3940          * need to reserve some slack space in case we add bytes and update
3941          * inode item when doing the truncate.
3942          */
3943         while (1) {
3944                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3945                                              BTRFS_RESERVE_FLUSH_LIMIT);
3946
3947                 /*
3948                  * Try and steal from the global reserve since we will
3949                  * likely not use this space anyway, we want to try as
3950                  * hard as possible to get this to work.
3951                  */
3952                 if (ret)
3953                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3954
3955                 if (ret) {
3956                         printk(KERN_WARNING "Could not get space for a "
3957                                "delete, will truncate on mount %d\n", ret);
3958                         btrfs_orphan_del(NULL, inode);
3959                         btrfs_free_block_rsv(root, rsv);
3960                         goto no_delete;
3961                 }
3962
3963                 trans = btrfs_join_transaction(root);
3964                 if (IS_ERR(trans)) {
3965                         btrfs_orphan_del(NULL, inode);
3966                         btrfs_free_block_rsv(root, rsv);
3967                         goto no_delete;
3968                 }
3969
3970                 trans->block_rsv = rsv;
3971
3972                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3973                 if (ret != -ENOSPC)
3974                         break;
3975
3976                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3977                 btrfs_end_transaction(trans, root);
3978                 trans = NULL;
3979                 btrfs_btree_balance_dirty(root);
3980         }
3981
3982         btrfs_free_block_rsv(root, rsv);
3983
3984         if (ret == 0) {
3985                 trans->block_rsv = root->orphan_block_rsv;
3986                 ret = btrfs_orphan_del(trans, inode);
3987                 BUG_ON(ret);
3988         }
3989
3990         trans->block_rsv = &root->fs_info->trans_block_rsv;
3991         if (!(root == root->fs_info->tree_root ||
3992               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3993                 btrfs_return_ino(root, btrfs_ino(inode));
3994
3995         btrfs_end_transaction(trans, root);
3996         btrfs_btree_balance_dirty(root);
3997 no_delete:
3998         clear_inode(inode);
3999         return;
4000 }
4001
4002 /*
4003  * this returns the key found in the dir entry in the location pointer.
4004  * If no dir entries were found, location->objectid is 0.
4005  */
4006 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4007                                struct btrfs_key *location)
4008 {
4009         const char *name = dentry->d_name.name;
4010         int namelen = dentry->d_name.len;
4011         struct btrfs_dir_item *di;
4012         struct btrfs_path *path;
4013         struct btrfs_root *root = BTRFS_I(dir)->root;
4014         int ret = 0;
4015
4016         path = btrfs_alloc_path();
4017         if (!path)
4018                 return -ENOMEM;
4019
4020         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4021                                     namelen, 0);
4022         if (IS_ERR(di))
4023                 ret = PTR_ERR(di);
4024
4025         if (IS_ERR_OR_NULL(di))
4026                 goto out_err;
4027
4028         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4029 out:
4030         btrfs_free_path(path);
4031         return ret;
4032 out_err:
4033         location->objectid = 0;
4034         goto out;
4035 }
4036
4037 /*
4038  * when we hit a tree root in a directory, the btrfs part of the inode
4039  * needs to be changed to reflect the root directory of the tree root.  This
4040  * is kind of like crossing a mount point.
4041  */
4042 static int fixup_tree_root_location(struct btrfs_root *root,
4043                                     struct inode *dir,
4044                                     struct dentry *dentry,
4045                                     struct btrfs_key *location,
4046                                     struct btrfs_root **sub_root)
4047 {
4048         struct btrfs_path *path;
4049         struct btrfs_root *new_root;
4050         struct btrfs_root_ref *ref;
4051         struct extent_buffer *leaf;
4052         int ret;
4053         int err = 0;
4054
4055         path = btrfs_alloc_path();
4056         if (!path) {
4057                 err = -ENOMEM;
4058                 goto out;
4059         }
4060
4061         err = -ENOENT;
4062         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4063                                   BTRFS_I(dir)->root->root_key.objectid,
4064                                   location->objectid);
4065         if (ret) {
4066                 if (ret < 0)
4067                         err = ret;
4068                 goto out;
4069         }
4070
4071         leaf = path->nodes[0];
4072         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4073         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4074             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4075                 goto out;
4076
4077         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4078                                    (unsigned long)(ref + 1),
4079                                    dentry->d_name.len);
4080         if (ret)
4081                 goto out;
4082
4083         btrfs_release_path(path);
4084
4085         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4086         if (IS_ERR(new_root)) {
4087                 err = PTR_ERR(new_root);
4088                 goto out;
4089         }
4090
4091         if (btrfs_root_refs(&new_root->root_item) == 0) {
4092                 err = -ENOENT;
4093                 goto out;
4094         }
4095
4096         *sub_root = new_root;
4097         location->objectid = btrfs_root_dirid(&new_root->root_item);
4098         location->type = BTRFS_INODE_ITEM_KEY;
4099         location->offset = 0;
4100         err = 0;
4101 out:
4102         btrfs_free_path(path);
4103         return err;
4104 }
4105
4106 static void inode_tree_add(struct inode *inode)
4107 {
4108         struct btrfs_root *root = BTRFS_I(inode)->root;
4109         struct btrfs_inode *entry;
4110         struct rb_node **p;
4111         struct rb_node *parent;
4112         u64 ino = btrfs_ino(inode);
4113 again:
4114         p = &root->inode_tree.rb_node;
4115         parent = NULL;
4116
4117         if (inode_unhashed(inode))
4118                 return;
4119
4120         spin_lock(&root->inode_lock);
4121         while (*p) {
4122                 parent = *p;
4123                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4124
4125                 if (ino < btrfs_ino(&entry->vfs_inode))
4126                         p = &parent->rb_left;
4127                 else if (ino > btrfs_ino(&entry->vfs_inode))
4128                         p = &parent->rb_right;
4129                 else {
4130                         WARN_ON(!(entry->vfs_inode.i_state &
4131                                   (I_WILL_FREE | I_FREEING)));
4132                         rb_erase(parent, &root->inode_tree);
4133                         RB_CLEAR_NODE(parent);
4134                         spin_unlock(&root->inode_lock);
4135                         goto again;
4136                 }
4137         }
4138         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4139         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4140         spin_unlock(&root->inode_lock);
4141 }
4142
4143 static void inode_tree_del(struct inode *inode)
4144 {
4145         struct btrfs_root *root = BTRFS_I(inode)->root;
4146         int empty = 0;
4147
4148         spin_lock(&root->inode_lock);
4149         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4150                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4151                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4152                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4153         }
4154         spin_unlock(&root->inode_lock);
4155
4156         /*
4157          * Free space cache has inodes in the tree root, but the tree root has a
4158          * root_refs of 0, so this could end up dropping the tree root as a
4159          * snapshot, so we need the extra !root->fs_info->tree_root check to
4160          * make sure we don't drop it.
4161          */
4162         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4163             root != root->fs_info->tree_root) {
4164                 synchronize_srcu(&root->fs_info->subvol_srcu);
4165                 spin_lock(&root->inode_lock);
4166                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4167                 spin_unlock(&root->inode_lock);
4168                 if (empty)
4169                         btrfs_add_dead_root(root);
4170         }
4171 }
4172
4173 void btrfs_invalidate_inodes(struct btrfs_root *root)
4174 {
4175         struct rb_node *node;
4176         struct rb_node *prev;
4177         struct btrfs_inode *entry;
4178         struct inode *inode;
4179         u64 objectid = 0;
4180
4181         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4182
4183         spin_lock(&root->inode_lock);
4184 again:
4185         node = root->inode_tree.rb_node;
4186         prev = NULL;
4187         while (node) {
4188                 prev = node;
4189                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4190
4191                 if (objectid < btrfs_ino(&entry->vfs_inode))
4192                         node = node->rb_left;
4193                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4194                         node = node->rb_right;
4195                 else
4196                         break;
4197         }
4198         if (!node) {
4199                 while (prev) {
4200                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4201                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4202                                 node = prev;
4203                                 break;
4204                         }
4205                         prev = rb_next(prev);
4206                 }
4207         }
4208         while (node) {
4209                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4210                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4211                 inode = igrab(&entry->vfs_inode);
4212                 if (inode) {
4213                         spin_unlock(&root->inode_lock);
4214                         if (atomic_read(&inode->i_count) > 1)
4215                                 d_prune_aliases(inode);
4216                         /*
4217                          * btrfs_drop_inode will have it removed from
4218                          * the inode cache when its usage count
4219                          * hits zero.
4220                          */
4221                         iput(inode);
4222                         cond_resched();
4223                         spin_lock(&root->inode_lock);
4224                         goto again;
4225                 }
4226
4227                 if (cond_resched_lock(&root->inode_lock))
4228                         goto again;
4229
4230                 node = rb_next(node);
4231         }
4232         spin_unlock(&root->inode_lock);
4233 }
4234
4235 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4236 {
4237         struct btrfs_iget_args *args = p;
4238         inode->i_ino = args->ino;
4239         BTRFS_I(inode)->root = args->root;
4240         return 0;
4241 }
4242
4243 static int btrfs_find_actor(struct inode *inode, void *opaque)
4244 {
4245         struct btrfs_iget_args *args = opaque;
4246         return args->ino == btrfs_ino(inode) &&
4247                 args->root == BTRFS_I(inode)->root;
4248 }
4249
4250 static struct inode *btrfs_iget_locked(struct super_block *s,
4251                                        u64 objectid,
4252                                        struct btrfs_root *root)
4253 {
4254         struct inode *inode;
4255         struct btrfs_iget_args args;
4256         args.ino = objectid;
4257         args.root = root;
4258
4259         inode = iget5_locked(s, objectid, btrfs_find_actor,
4260                              btrfs_init_locked_inode,
4261                              (void *)&args);
4262         return inode;
4263 }
4264
4265 /* Get an inode object given its location and corresponding root.
4266  * Returns in *is_new if the inode was read from disk
4267  */
4268 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4269                          struct btrfs_root *root, int *new)
4270 {
4271         struct inode *inode;
4272
4273         inode = btrfs_iget_locked(s, location->objectid, root);
4274         if (!inode)
4275                 return ERR_PTR(-ENOMEM);
4276
4277         if (inode->i_state & I_NEW) {
4278                 BTRFS_I(inode)->root = root;
4279                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4280                 btrfs_read_locked_inode(inode);
4281                 if (!is_bad_inode(inode)) {
4282                         inode_tree_add(inode);
4283                         unlock_new_inode(inode);
4284                         if (new)
4285                                 *new = 1;
4286                 } else {
4287                         unlock_new_inode(inode);
4288                         iput(inode);
4289                         inode = ERR_PTR(-ESTALE);
4290                 }
4291         }
4292
4293         return inode;
4294 }
4295
4296 static struct inode *new_simple_dir(struct super_block *s,
4297                                     struct btrfs_key *key,
4298                                     struct btrfs_root *root)
4299 {
4300         struct inode *inode = new_inode(s);
4301
4302         if (!inode)
4303                 return ERR_PTR(-ENOMEM);
4304
4305         BTRFS_I(inode)->root = root;
4306         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4307         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4308
4309         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4310         inode->i_op = &btrfs_dir_ro_inode_operations;
4311         inode->i_fop = &simple_dir_operations;
4312         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4313         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4314
4315         return inode;
4316 }
4317
4318 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4319 {
4320         struct inode *inode;
4321         struct btrfs_root *root = BTRFS_I(dir)->root;
4322         struct btrfs_root *sub_root = root;
4323         struct btrfs_key location;
4324         int index;
4325         int ret = 0;
4326
4327         if (dentry->d_name.len > BTRFS_NAME_LEN)
4328                 return ERR_PTR(-ENAMETOOLONG);
4329
4330         if (unlikely(d_need_lookup(dentry))) {
4331                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4332                 kfree(dentry->d_fsdata);
4333                 dentry->d_fsdata = NULL;
4334                 /* This thing is hashed, drop it for now */
4335                 d_drop(dentry);
4336         } else {
4337                 ret = btrfs_inode_by_name(dir, dentry, &location);
4338         }
4339
4340         if (ret < 0)
4341                 return ERR_PTR(ret);
4342
4343         if (location.objectid == 0)
4344                 return NULL;
4345
4346         if (location.type == BTRFS_INODE_ITEM_KEY) {
4347                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4348                 return inode;
4349         }
4350
4351         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4352
4353         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4354         ret = fixup_tree_root_location(root, dir, dentry,
4355                                        &location, &sub_root);
4356         if (ret < 0) {
4357                 if (ret != -ENOENT)
4358                         inode = ERR_PTR(ret);
4359                 else
4360                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4361         } else {
4362                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4363         }
4364         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4365
4366         if (!IS_ERR(inode) && root != sub_root) {
4367                 down_read(&root->fs_info->cleanup_work_sem);
4368                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4369                         ret = btrfs_orphan_cleanup(sub_root);
4370                 up_read(&root->fs_info->cleanup_work_sem);
4371                 if (ret)
4372                         inode = ERR_PTR(ret);
4373         }
4374
4375         return inode;
4376 }
4377
4378 static int btrfs_dentry_delete(const struct dentry *dentry)
4379 {
4380         struct btrfs_root *root;
4381         struct inode *inode = dentry->d_inode;
4382
4383         if (!inode && !IS_ROOT(dentry))
4384                 inode = dentry->d_parent->d_inode;
4385
4386         if (inode) {
4387                 root = BTRFS_I(inode)->root;
4388                 if (btrfs_root_refs(&root->root_item) == 0)
4389                         return 1;
4390
4391                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4392                         return 1;
4393         }
4394         return 0;
4395 }
4396
4397 static void btrfs_dentry_release(struct dentry *dentry)
4398 {
4399         if (dentry->d_fsdata)
4400                 kfree(dentry->d_fsdata);
4401 }
4402
4403 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4404                                    unsigned int flags)
4405 {
4406         struct dentry *ret;
4407
4408         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4409         if (unlikely(d_need_lookup(dentry))) {
4410                 spin_lock(&dentry->d_lock);
4411                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4412                 spin_unlock(&dentry->d_lock);
4413         }
4414         return ret;
4415 }
4416
4417 unsigned char btrfs_filetype_table[] = {
4418         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4419 };
4420
4421 static int btrfs_real_readdir(struct file *filp, void *dirent,
4422                               filldir_t filldir)
4423 {
4424         struct inode *inode = filp->f_dentry->d_inode;
4425         struct btrfs_root *root = BTRFS_I(inode)->root;
4426         struct btrfs_item *item;
4427         struct btrfs_dir_item *di;
4428         struct btrfs_key key;
4429         struct btrfs_key found_key;
4430         struct btrfs_path *path;
4431         struct list_head ins_list;
4432         struct list_head del_list;
4433         int ret;
4434         struct extent_buffer *leaf;
4435         int slot;
4436         unsigned char d_type;
4437         int over = 0;
4438         u32 di_cur;
4439         u32 di_total;
4440         u32 di_len;
4441         int key_type = BTRFS_DIR_INDEX_KEY;
4442         char tmp_name[32];
4443         char *name_ptr;
4444         int name_len;
4445         int is_curr = 0;        /* filp->f_pos points to the current index? */
4446
4447         /* FIXME, use a real flag for deciding about the key type */
4448         if (root->fs_info->tree_root == root)
4449                 key_type = BTRFS_DIR_ITEM_KEY;
4450
4451         /* special case for "." */
4452         if (filp->f_pos == 0) {
4453                 over = filldir(dirent, ".", 1,
4454                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4455                 if (over)
4456                         return 0;
4457                 filp->f_pos = 1;
4458         }
4459         /* special case for .., just use the back ref */
4460         if (filp->f_pos == 1) {
4461                 u64 pino = parent_ino(filp->f_path.dentry);
4462                 over = filldir(dirent, "..", 2,
4463                                filp->f_pos, pino, DT_DIR);
4464                 if (over)
4465                         return 0;
4466                 filp->f_pos = 2;
4467         }
4468         path = btrfs_alloc_path();
4469         if (!path)
4470                 return -ENOMEM;
4471
4472         path->reada = 1;
4473
4474         if (key_type == BTRFS_DIR_INDEX_KEY) {
4475                 INIT_LIST_HEAD(&ins_list);
4476                 INIT_LIST_HEAD(&del_list);
4477                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4478         }
4479
4480         btrfs_set_key_type(&key, key_type);
4481         key.offset = filp->f_pos;
4482         key.objectid = btrfs_ino(inode);
4483
4484         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4485         if (ret < 0)
4486                 goto err;
4487
4488         while (1) {
4489                 leaf = path->nodes[0];
4490                 slot = path->slots[0];
4491                 if (slot >= btrfs_header_nritems(leaf)) {
4492                         ret = btrfs_next_leaf(root, path);
4493                         if (ret < 0)
4494                                 goto err;
4495                         else if (ret > 0)
4496                                 break;
4497                         continue;
4498                 }
4499
4500                 item = btrfs_item_nr(leaf, slot);
4501                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4502
4503                 if (found_key.objectid != key.objectid)
4504                         break;
4505                 if (btrfs_key_type(&found_key) != key_type)
4506                         break;
4507                 if (found_key.offset < filp->f_pos)
4508                         goto next;
4509                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4510                     btrfs_should_delete_dir_index(&del_list,
4511                                                   found_key.offset))
4512                         goto next;
4513
4514                 filp->f_pos = found_key.offset;
4515                 is_curr = 1;
4516
4517                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4518                 di_cur = 0;
4519                 di_total = btrfs_item_size(leaf, item);
4520
4521                 while (di_cur < di_total) {
4522                         struct btrfs_key location;
4523
4524                         if (verify_dir_item(root, leaf, di))
4525                                 break;
4526
4527                         name_len = btrfs_dir_name_len(leaf, di);
4528                         if (name_len <= sizeof(tmp_name)) {
4529                                 name_ptr = tmp_name;
4530                         } else {
4531                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4532                                 if (!name_ptr) {
4533                                         ret = -ENOMEM;
4534                                         goto err;
4535                                 }
4536                         }
4537                         read_extent_buffer(leaf, name_ptr,
4538                                            (unsigned long)(di + 1), name_len);
4539
4540                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4541                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4542
4543
4544                         /* is this a reference to our own snapshot? If so
4545                          * skip it.
4546                          *
4547                          * In contrast to old kernels, we insert the snapshot's
4548                          * dir item and dir index after it has been created, so
4549                          * we won't find a reference to our own snapshot. We
4550                          * still keep the following code for backward
4551                          * compatibility.
4552                          */
4553                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4554                             location.objectid == root->root_key.objectid) {
4555                                 over = 0;
4556                                 goto skip;
4557                         }
4558                         over = filldir(dirent, name_ptr, name_len,
4559                                        found_key.offset, location.objectid,
4560                                        d_type);
4561
4562 skip:
4563                         if (name_ptr != tmp_name)
4564                                 kfree(name_ptr);
4565
4566                         if (over)
4567                                 goto nopos;
4568                         di_len = btrfs_dir_name_len(leaf, di) +
4569                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4570                         di_cur += di_len;
4571                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4572                 }
4573 next:
4574                 path->slots[0]++;
4575         }
4576
4577         if (key_type == BTRFS_DIR_INDEX_KEY) {
4578                 if (is_curr)
4579                         filp->f_pos++;
4580                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4581                                                       &ins_list);
4582                 if (ret)
4583                         goto nopos;
4584         }
4585
4586         /* Reached end of directory/root. Bump pos past the last item. */
4587         if (key_type == BTRFS_DIR_INDEX_KEY)
4588                 /*
4589                  * 32-bit glibc will use getdents64, but then strtol -
4590                  * so the last number we can serve is this.
4591                  */
4592                 filp->f_pos = 0x7fffffff;
4593         else
4594                 filp->f_pos++;
4595 nopos:
4596         ret = 0;
4597 err:
4598         if (key_type == BTRFS_DIR_INDEX_KEY)
4599                 btrfs_put_delayed_items(&ins_list, &del_list);
4600         btrfs_free_path(path);
4601         return ret;
4602 }
4603
4604 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4605 {
4606         struct btrfs_root *root = BTRFS_I(inode)->root;
4607         struct btrfs_trans_handle *trans;
4608         int ret = 0;
4609         bool nolock = false;
4610
4611         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4612                 return 0;
4613
4614         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4615                 nolock = true;
4616
4617         if (wbc->sync_mode == WB_SYNC_ALL) {
4618                 if (nolock)
4619                         trans = btrfs_join_transaction_nolock(root);
4620                 else
4621                         trans = btrfs_join_transaction(root);
4622                 if (IS_ERR(trans))
4623                         return PTR_ERR(trans);
4624                 ret = btrfs_commit_transaction(trans, root);
4625         }
4626         return ret;
4627 }
4628
4629 /*
4630  * This is somewhat expensive, updating the tree every time the
4631  * inode changes.  But, it is most likely to find the inode in cache.
4632  * FIXME, needs more benchmarking...there are no reasons other than performance
4633  * to keep or drop this code.
4634  */
4635 int btrfs_dirty_inode(struct inode *inode)
4636 {
4637         struct btrfs_root *root = BTRFS_I(inode)->root;
4638         struct btrfs_trans_handle *trans;
4639         int ret;
4640
4641         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4642                 return 0;
4643
4644         trans = btrfs_join_transaction(root);
4645         if (IS_ERR(trans))
4646                 return PTR_ERR(trans);
4647
4648         ret = btrfs_update_inode(trans, root, inode);
4649         if (ret && ret == -ENOSPC) {
4650                 /* whoops, lets try again with the full transaction */
4651                 btrfs_end_transaction(trans, root);
4652                 trans = btrfs_start_transaction(root, 1);
4653                 if (IS_ERR(trans))
4654                         return PTR_ERR(trans);
4655
4656                 ret = btrfs_update_inode(trans, root, inode);
4657         }
4658         btrfs_end_transaction(trans, root);
4659         if (BTRFS_I(inode)->delayed_node)
4660                 btrfs_balance_delayed_items(root);
4661
4662         return ret;
4663 }
4664
4665 /*
4666  * This is a copy of file_update_time.  We need this so we can return error on
4667  * ENOSPC for updating the inode in the case of file write and mmap writes.
4668  */
4669 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4670                              int flags)
4671 {
4672         struct btrfs_root *root = BTRFS_I(inode)->root;
4673
4674         if (btrfs_root_readonly(root))
4675                 return -EROFS;
4676
4677         if (flags & S_VERSION)
4678                 inode_inc_iversion(inode);
4679         if (flags & S_CTIME)
4680                 inode->i_ctime = *now;
4681         if (flags & S_MTIME)
4682                 inode->i_mtime = *now;
4683         if (flags & S_ATIME)
4684                 inode->i_atime = *now;
4685         return btrfs_dirty_inode(inode);
4686 }
4687
4688 /*
4689  * find the highest existing sequence number in a directory
4690  * and then set the in-memory index_cnt variable to reflect
4691  * free sequence numbers
4692  */
4693 static int btrfs_set_inode_index_count(struct inode *inode)
4694 {
4695         struct btrfs_root *root = BTRFS_I(inode)->root;
4696         struct btrfs_key key, found_key;
4697         struct btrfs_path *path;
4698         struct extent_buffer *leaf;
4699         int ret;
4700
4701         key.objectid = btrfs_ino(inode);
4702         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4703         key.offset = (u64)-1;
4704
4705         path = btrfs_alloc_path();
4706         if (!path)
4707                 return -ENOMEM;
4708
4709         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4710         if (ret < 0)
4711                 goto out;
4712         /* FIXME: we should be able to handle this */
4713         if (ret == 0)
4714                 goto out;
4715         ret = 0;
4716
4717         /*
4718          * MAGIC NUMBER EXPLANATION:
4719          * since we search a directory based on f_pos we have to start at 2
4720          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4721          * else has to start at 2
4722          */
4723         if (path->slots[0] == 0) {
4724                 BTRFS_I(inode)->index_cnt = 2;
4725                 goto out;
4726         }
4727
4728         path->slots[0]--;
4729
4730         leaf = path->nodes[0];
4731         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4732
4733         if (found_key.objectid != btrfs_ino(inode) ||
4734             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4735                 BTRFS_I(inode)->index_cnt = 2;
4736                 goto out;
4737         }
4738
4739         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4740 out:
4741         btrfs_free_path(path);
4742         return ret;
4743 }
4744
4745 /*
4746  * helper to find a free sequence number in a given directory.  This current
4747  * code is very simple, later versions will do smarter things in the btree
4748  */
4749 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4750 {
4751         int ret = 0;
4752
4753         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4754                 ret = btrfs_inode_delayed_dir_index_count(dir);
4755                 if (ret) {
4756                         ret = btrfs_set_inode_index_count(dir);
4757                         if (ret)
4758                                 return ret;
4759                 }
4760         }
4761
4762         *index = BTRFS_I(dir)->index_cnt;
4763         BTRFS_I(dir)->index_cnt++;
4764
4765         return ret;
4766 }
4767
4768 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4769                                      struct btrfs_root *root,
4770                                      struct inode *dir,
4771                                      const char *name, int name_len,
4772                                      u64 ref_objectid, u64 objectid,
4773                                      umode_t mode, u64 *index)
4774 {
4775         struct inode *inode;
4776         struct btrfs_inode_item *inode_item;
4777         struct btrfs_key *location;
4778         struct btrfs_path *path;
4779         struct btrfs_inode_ref *ref;
4780         struct btrfs_key key[2];
4781         u32 sizes[2];
4782         unsigned long ptr;
4783         int ret;
4784         int owner;
4785
4786         path = btrfs_alloc_path();
4787         if (!path)
4788                 return ERR_PTR(-ENOMEM);
4789
4790         inode = new_inode(root->fs_info->sb);
4791         if (!inode) {
4792                 btrfs_free_path(path);
4793                 return ERR_PTR(-ENOMEM);
4794         }
4795
4796         /*
4797          * we have to initialize this early, so we can reclaim the inode
4798          * number if we fail afterwards in this function.
4799          */
4800         inode->i_ino = objectid;
4801
4802         if (dir) {
4803                 trace_btrfs_inode_request(dir);
4804
4805                 ret = btrfs_set_inode_index(dir, index);
4806                 if (ret) {
4807                         btrfs_free_path(path);
4808                         iput(inode);
4809                         return ERR_PTR(ret);
4810                 }
4811         }
4812         /*
4813          * index_cnt is ignored for everything but a dir,
4814          * btrfs_get_inode_index_count has an explanation for the magic
4815          * number
4816          */
4817         BTRFS_I(inode)->index_cnt = 2;
4818         BTRFS_I(inode)->root = root;
4819         BTRFS_I(inode)->generation = trans->transid;
4820         inode->i_generation = BTRFS_I(inode)->generation;
4821
4822         /*
4823          * We could have gotten an inode number from somebody who was fsynced
4824          * and then removed in this same transaction, so let's just set full
4825          * sync since it will be a full sync anyway and this will blow away the
4826          * old info in the log.
4827          */
4828         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4829
4830         if (S_ISDIR(mode))
4831                 owner = 0;
4832         else
4833                 owner = 1;
4834
4835         key[0].objectid = objectid;
4836         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4837         key[0].offset = 0;
4838
4839         /*
4840          * Start new inodes with an inode_ref. This is slightly more
4841          * efficient for small numbers of hard links since they will
4842          * be packed into one item. Extended refs will kick in if we
4843          * add more hard links than can fit in the ref item.
4844          */
4845         key[1].objectid = objectid;
4846         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4847         key[1].offset = ref_objectid;
4848
4849         sizes[0] = sizeof(struct btrfs_inode_item);
4850         sizes[1] = name_len + sizeof(*ref);
4851
4852         path->leave_spinning = 1;
4853         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4854         if (ret != 0)
4855                 goto fail;
4856
4857         inode_init_owner(inode, dir, mode);
4858         inode_set_bytes(inode, 0);
4859         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4860         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4861                                   struct btrfs_inode_item);
4862         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4863                              sizeof(*inode_item));
4864         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4865
4866         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4867                              struct btrfs_inode_ref);
4868         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4869         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4870         ptr = (unsigned long)(ref + 1);
4871         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4872
4873         btrfs_mark_buffer_dirty(path->nodes[0]);
4874         btrfs_free_path(path);
4875
4876         location = &BTRFS_I(inode)->location;
4877         location->objectid = objectid;
4878         location->offset = 0;
4879         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4880
4881         btrfs_inherit_iflags(inode, dir);
4882
4883         if (S_ISREG(mode)) {
4884                 if (btrfs_test_opt(root, NODATASUM))
4885                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4886                 if (btrfs_test_opt(root, NODATACOW))
4887                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4888         }
4889
4890         insert_inode_hash(inode);
4891         inode_tree_add(inode);
4892
4893         trace_btrfs_inode_new(inode);
4894         btrfs_set_inode_last_trans(trans, inode);
4895
4896         btrfs_update_root_times(trans, root);
4897
4898         return inode;
4899 fail:
4900         if (dir)
4901                 BTRFS_I(dir)->index_cnt--;
4902         btrfs_free_path(path);
4903         iput(inode);
4904         return ERR_PTR(ret);
4905 }
4906
4907 static inline u8 btrfs_inode_type(struct inode *inode)
4908 {
4909         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4910 }
4911
4912 /*
4913  * utility function to add 'inode' into 'parent_inode' with
4914  * a give name and a given sequence number.
4915  * if 'add_backref' is true, also insert a backref from the
4916  * inode to the parent directory.
4917  */
4918 int btrfs_add_link(struct btrfs_trans_handle *trans,
4919                    struct inode *parent_inode, struct inode *inode,
4920                    const char *name, int name_len, int add_backref, u64 index)
4921 {
4922         int ret = 0;
4923         struct btrfs_key key;
4924         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4925         u64 ino = btrfs_ino(inode);
4926         u64 parent_ino = btrfs_ino(parent_inode);
4927
4928         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4929                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4930         } else {
4931                 key.objectid = ino;
4932                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4933                 key.offset = 0;
4934         }
4935
4936         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4937                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4938                                          key.objectid, root->root_key.objectid,
4939                                          parent_ino, index, name, name_len);
4940         } else if (add_backref) {
4941                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4942                                              parent_ino, index);
4943         }
4944
4945         /* Nothing to clean up yet */
4946         if (ret)
4947                 return ret;
4948
4949         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4950                                     parent_inode, &key,
4951                                     btrfs_inode_type(inode), index);
4952         if (ret == -EEXIST || ret == -EOVERFLOW)
4953                 goto fail_dir_item;
4954         else if (ret) {
4955                 btrfs_abort_transaction(trans, root, ret);
4956                 return ret;
4957         }
4958
4959         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4960                            name_len * 2);
4961         inode_inc_iversion(parent_inode);
4962         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4963         ret = btrfs_update_inode(trans, root, parent_inode);
4964         if (ret)
4965                 btrfs_abort_transaction(trans, root, ret);
4966         return ret;
4967
4968 fail_dir_item:
4969         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4970                 u64 local_index;
4971                 int err;
4972                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4973                                  key.objectid, root->root_key.objectid,
4974                                  parent_ino, &local_index, name, name_len);
4975
4976         } else if (add_backref) {
4977                 u64 local_index;
4978                 int err;
4979
4980                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4981                                           ino, parent_ino, &local_index);
4982         }
4983         return ret;
4984 }
4985
4986 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4987                             struct inode *dir, struct dentry *dentry,
4988                             struct inode *inode, int backref, u64 index)
4989 {
4990         int err = btrfs_add_link(trans, dir, inode,
4991                                  dentry->d_name.name, dentry->d_name.len,
4992                                  backref, index);
4993         if (err > 0)
4994                 err = -EEXIST;
4995         return err;
4996 }
4997
4998 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4999                         umode_t mode, dev_t rdev)
5000 {
5001         struct btrfs_trans_handle *trans;
5002         struct btrfs_root *root = BTRFS_I(dir)->root;
5003         struct inode *inode = NULL;
5004         int err;
5005         int drop_inode = 0;
5006         u64 objectid;
5007         u64 index = 0;
5008
5009         if (!new_valid_dev(rdev))
5010                 return -EINVAL;
5011
5012         /*
5013          * 2 for inode item and ref
5014          * 2 for dir items
5015          * 1 for xattr if selinux is on
5016          */
5017         trans = btrfs_start_transaction(root, 5);
5018         if (IS_ERR(trans))
5019                 return PTR_ERR(trans);
5020
5021         err = btrfs_find_free_ino(root, &objectid);
5022         if (err)
5023                 goto out_unlock;
5024
5025         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5026                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5027                                 mode, &index);
5028         if (IS_ERR(inode)) {
5029                 err = PTR_ERR(inode);
5030                 goto out_unlock;
5031         }
5032
5033         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5034         if (err) {
5035                 drop_inode = 1;
5036                 goto out_unlock;
5037         }
5038
5039         /*
5040         * If the active LSM wants to access the inode during
5041         * d_instantiate it needs these. Smack checks to see
5042         * if the filesystem supports xattrs by looking at the
5043         * ops vector.
5044         */
5045
5046         inode->i_op = &btrfs_special_inode_operations;
5047         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5048         if (err)
5049                 drop_inode = 1;
5050         else {
5051                 init_special_inode(inode, inode->i_mode, rdev);
5052                 btrfs_update_inode(trans, root, inode);
5053                 d_instantiate(dentry, inode);
5054         }
5055 out_unlock:
5056         btrfs_end_transaction(trans, root);
5057         btrfs_btree_balance_dirty(root);
5058         if (drop_inode) {
5059                 inode_dec_link_count(inode);
5060                 iput(inode);
5061         }
5062         return err;
5063 }
5064
5065 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5066                         umode_t mode, bool excl)
5067 {
5068         struct btrfs_trans_handle *trans;
5069         struct btrfs_root *root = BTRFS_I(dir)->root;
5070         struct inode *inode = NULL;
5071         int drop_inode_on_err = 0;
5072         int err;
5073         u64 objectid;
5074         u64 index = 0;
5075
5076         /*
5077          * 2 for inode item and ref
5078          * 2 for dir items
5079          * 1 for xattr if selinux is on
5080          */
5081         trans = btrfs_start_transaction(root, 5);
5082         if (IS_ERR(trans))
5083                 return PTR_ERR(trans);
5084
5085         err = btrfs_find_free_ino(root, &objectid);
5086         if (err)
5087                 goto out_unlock;
5088
5089         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5090                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5091                                 mode, &index);
5092         if (IS_ERR(inode)) {
5093                 err = PTR_ERR(inode);
5094                 goto out_unlock;
5095         }
5096         drop_inode_on_err = 1;
5097
5098         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5099         if (err)
5100                 goto out_unlock;
5101
5102         err = btrfs_update_inode(trans, root, inode);
5103         if (err)
5104                 goto out_unlock;
5105
5106         /*
5107         * If the active LSM wants to access the inode during
5108         * d_instantiate it needs these. Smack checks to see
5109         * if the filesystem supports xattrs by looking at the
5110         * ops vector.
5111         */
5112         inode->i_fop = &btrfs_file_operations;
5113         inode->i_op = &btrfs_file_inode_operations;
5114
5115         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5116         if (err)
5117                 goto out_unlock;
5118
5119         inode->i_mapping->a_ops = &btrfs_aops;
5120         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5121         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5122         d_instantiate(dentry, inode);
5123
5124 out_unlock:
5125         btrfs_end_transaction(trans, root);
5126         if (err && drop_inode_on_err) {
5127                 inode_dec_link_count(inode);
5128                 iput(inode);
5129         }
5130         btrfs_btree_balance_dirty(root);
5131         return err;
5132 }
5133
5134 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5135                       struct dentry *dentry)
5136 {
5137         struct btrfs_trans_handle *trans;
5138         struct btrfs_root *root = BTRFS_I(dir)->root;
5139         struct inode *inode = old_dentry->d_inode;
5140         u64 index;
5141         int err;
5142         int drop_inode = 0;
5143
5144         /* do not allow sys_link's with other subvols of the same device */
5145         if (root->objectid != BTRFS_I(inode)->root->objectid)
5146                 return -EXDEV;
5147
5148         if (inode->i_nlink >= BTRFS_LINK_MAX)
5149                 return -EMLINK;
5150
5151         err = btrfs_set_inode_index(dir, &index);
5152         if (err)
5153                 goto fail;
5154
5155         /*
5156          * 2 items for inode and inode ref
5157          * 2 items for dir items
5158          * 1 item for parent inode
5159          */
5160         trans = btrfs_start_transaction(root, 5);
5161         if (IS_ERR(trans)) {
5162                 err = PTR_ERR(trans);
5163                 goto fail;
5164         }
5165
5166         btrfs_inc_nlink(inode);
5167         inode_inc_iversion(inode);
5168         inode->i_ctime = CURRENT_TIME;
5169         ihold(inode);
5170         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5171
5172         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5173
5174         if (err) {
5175                 drop_inode = 1;
5176         } else {
5177                 struct dentry *parent = dentry->d_parent;
5178                 err = btrfs_update_inode(trans, root, inode);
5179                 if (err)
5180                         goto fail;
5181                 d_instantiate(dentry, inode);
5182                 btrfs_log_new_name(trans, inode, NULL, parent);
5183         }
5184
5185         btrfs_end_transaction(trans, root);
5186 fail:
5187         if (drop_inode) {
5188                 inode_dec_link_count(inode);
5189                 iput(inode);
5190         }
5191         btrfs_btree_balance_dirty(root);
5192         return err;
5193 }
5194
5195 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5196 {
5197         struct inode *inode = NULL;
5198         struct btrfs_trans_handle *trans;
5199         struct btrfs_root *root = BTRFS_I(dir)->root;
5200         int err = 0;
5201         int drop_on_err = 0;
5202         u64 objectid = 0;
5203         u64 index = 0;
5204
5205         /*
5206          * 2 items for inode and ref
5207          * 2 items for dir items
5208          * 1 for xattr if selinux is on
5209          */
5210         trans = btrfs_start_transaction(root, 5);
5211         if (IS_ERR(trans))
5212                 return PTR_ERR(trans);
5213
5214         err = btrfs_find_free_ino(root, &objectid);
5215         if (err)
5216                 goto out_fail;
5217
5218         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5219                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5220                                 S_IFDIR | mode, &index);
5221         if (IS_ERR(inode)) {
5222                 err = PTR_ERR(inode);
5223                 goto out_fail;
5224         }
5225
5226         drop_on_err = 1;
5227
5228         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5229         if (err)
5230                 goto out_fail;
5231
5232         inode->i_op = &btrfs_dir_inode_operations;
5233         inode->i_fop = &btrfs_dir_file_operations;
5234
5235         btrfs_i_size_write(inode, 0);
5236         err = btrfs_update_inode(trans, root, inode);
5237         if (err)
5238                 goto out_fail;
5239
5240         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5241                              dentry->d_name.len, 0, index);
5242         if (err)
5243                 goto out_fail;
5244
5245         d_instantiate(dentry, inode);
5246         drop_on_err = 0;
5247
5248 out_fail:
5249         btrfs_end_transaction(trans, root);
5250         if (drop_on_err)
5251                 iput(inode);
5252         btrfs_btree_balance_dirty(root);
5253         return err;
5254 }
5255
5256 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5257  * and an extent that you want to insert, deal with overlap and insert
5258  * the new extent into the tree.
5259  */
5260 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5261                                 struct extent_map *existing,
5262                                 struct extent_map *em,
5263                                 u64 map_start, u64 map_len)
5264 {
5265         u64 start_diff;
5266
5267         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5268         start_diff = map_start - em->start;
5269         em->start = map_start;
5270         em->len = map_len;
5271         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5272             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5273                 em->block_start += start_diff;
5274                 em->block_len -= start_diff;
5275         }
5276         return add_extent_mapping(em_tree, em);
5277 }
5278
5279 static noinline int uncompress_inline(struct btrfs_path *path,
5280                                       struct inode *inode, struct page *page,
5281                                       size_t pg_offset, u64 extent_offset,
5282                                       struct btrfs_file_extent_item *item)
5283 {
5284         int ret;
5285         struct extent_buffer *leaf = path->nodes[0];
5286         char *tmp;
5287         size_t max_size;
5288         unsigned long inline_size;
5289         unsigned long ptr;
5290         int compress_type;
5291
5292         WARN_ON(pg_offset != 0);
5293         compress_type = btrfs_file_extent_compression(leaf, item);
5294         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5295         inline_size = btrfs_file_extent_inline_item_len(leaf,
5296                                         btrfs_item_nr(leaf, path->slots[0]));
5297         tmp = kmalloc(inline_size, GFP_NOFS);
5298         if (!tmp)
5299                 return -ENOMEM;
5300         ptr = btrfs_file_extent_inline_start(item);
5301
5302         read_extent_buffer(leaf, tmp, ptr, inline_size);
5303
5304         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5305         ret = btrfs_decompress(compress_type, tmp, page,
5306                                extent_offset, inline_size, max_size);
5307         if (ret) {
5308                 char *kaddr = kmap_atomic(page);
5309                 unsigned long copy_size = min_t(u64,
5310                                   PAGE_CACHE_SIZE - pg_offset,
5311                                   max_size - extent_offset);
5312                 memset(kaddr + pg_offset, 0, copy_size);
5313                 kunmap_atomic(kaddr);
5314         }
5315         kfree(tmp);
5316         return 0;
5317 }
5318
5319 /*
5320  * a bit scary, this does extent mapping from logical file offset to the disk.
5321  * the ugly parts come from merging extents from the disk with the in-ram
5322  * representation.  This gets more complex because of the data=ordered code,
5323  * where the in-ram extents might be locked pending data=ordered completion.
5324  *
5325  * This also copies inline extents directly into the page.
5326  */
5327
5328 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5329                                     size_t pg_offset, u64 start, u64 len,
5330                                     int create)
5331 {
5332         int ret;
5333         int err = 0;
5334         u64 bytenr;
5335         u64 extent_start = 0;
5336         u64 extent_end = 0;
5337         u64 objectid = btrfs_ino(inode);
5338         u32 found_type;
5339         struct btrfs_path *path = NULL;
5340         struct btrfs_root *root = BTRFS_I(inode)->root;
5341         struct btrfs_file_extent_item *item;
5342         struct extent_buffer *leaf;
5343         struct btrfs_key found_key;
5344         struct extent_map *em = NULL;
5345         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5346         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5347         struct btrfs_trans_handle *trans = NULL;
5348         int compress_type;
5349
5350 again:
5351         read_lock(&em_tree->lock);
5352         em = lookup_extent_mapping(em_tree, start, len);
5353         if (em)
5354                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5355         read_unlock(&em_tree->lock);
5356
5357         if (em) {
5358                 if (em->start > start || em->start + em->len <= start)
5359                         free_extent_map(em);
5360                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5361                         free_extent_map(em);
5362                 else
5363                         goto out;
5364         }
5365         em = alloc_extent_map();
5366         if (!em) {
5367                 err = -ENOMEM;
5368                 goto out;
5369         }
5370         em->bdev = root->fs_info->fs_devices->latest_bdev;
5371         em->start = EXTENT_MAP_HOLE;
5372         em->orig_start = EXTENT_MAP_HOLE;
5373         em->len = (u64)-1;
5374         em->block_len = (u64)-1;
5375
5376         if (!path) {
5377                 path = btrfs_alloc_path();
5378                 if (!path) {
5379                         err = -ENOMEM;
5380                         goto out;
5381                 }
5382                 /*
5383                  * Chances are we'll be called again, so go ahead and do
5384                  * readahead
5385                  */
5386                 path->reada = 1;
5387         }
5388
5389         ret = btrfs_lookup_file_extent(trans, root, path,
5390                                        objectid, start, trans != NULL);
5391         if (ret < 0) {
5392                 err = ret;
5393                 goto out;
5394         }
5395
5396         if (ret != 0) {
5397                 if (path->slots[0] == 0)
5398                         goto not_found;
5399                 path->slots[0]--;
5400         }
5401
5402         leaf = path->nodes[0];
5403         item = btrfs_item_ptr(leaf, path->slots[0],
5404                               struct btrfs_file_extent_item);
5405         /* are we inside the extent that was found? */
5406         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5407         found_type = btrfs_key_type(&found_key);
5408         if (found_key.objectid != objectid ||
5409             found_type != BTRFS_EXTENT_DATA_KEY) {
5410                 goto not_found;
5411         }
5412
5413         found_type = btrfs_file_extent_type(leaf, item);
5414         extent_start = found_key.offset;
5415         compress_type = btrfs_file_extent_compression(leaf, item);
5416         if (found_type == BTRFS_FILE_EXTENT_REG ||
5417             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5418                 extent_end = extent_start +
5419                        btrfs_file_extent_num_bytes(leaf, item);
5420         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5421                 size_t size;
5422                 size = btrfs_file_extent_inline_len(leaf, item);
5423                 extent_end = (extent_start + size + root->sectorsize - 1) &
5424                         ~((u64)root->sectorsize - 1);
5425         }
5426
5427         if (start >= extent_end) {
5428                 path->slots[0]++;
5429                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5430                         ret = btrfs_next_leaf(root, path);
5431                         if (ret < 0) {
5432                                 err = ret;
5433                                 goto out;
5434                         }
5435                         if (ret > 0)
5436                                 goto not_found;
5437                         leaf = path->nodes[0];
5438                 }
5439                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5440                 if (found_key.objectid != objectid ||
5441                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5442                         goto not_found;
5443                 if (start + len <= found_key.offset)
5444                         goto not_found;
5445                 em->start = start;
5446                 em->orig_start = start;
5447                 em->len = found_key.offset - start;
5448                 goto not_found_em;
5449         }
5450
5451         if (found_type == BTRFS_FILE_EXTENT_REG ||
5452             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5453                 em->start = extent_start;
5454                 em->len = extent_end - extent_start;
5455                 em->orig_start = extent_start -
5456                                  btrfs_file_extent_offset(leaf, item);
5457                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5458                                                                       item);
5459                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5460                 if (bytenr == 0) {
5461                         em->block_start = EXTENT_MAP_HOLE;
5462                         goto insert;
5463                 }
5464                 if (compress_type != BTRFS_COMPRESS_NONE) {
5465                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5466                         em->compress_type = compress_type;
5467                         em->block_start = bytenr;
5468                         em->block_len = em->orig_block_len;
5469                 } else {
5470                         bytenr += btrfs_file_extent_offset(leaf, item);
5471                         em->block_start = bytenr;
5472                         em->block_len = em->len;
5473                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5474                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5475                 }
5476                 goto insert;
5477         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5478                 unsigned long ptr;
5479                 char *map;
5480                 size_t size;
5481                 size_t extent_offset;
5482                 size_t copy_size;
5483
5484                 em->block_start = EXTENT_MAP_INLINE;
5485                 if (!page || create) {
5486                         em->start = extent_start;
5487                         em->len = extent_end - extent_start;
5488                         goto out;
5489                 }
5490
5491                 size = btrfs_file_extent_inline_len(leaf, item);
5492                 extent_offset = page_offset(page) + pg_offset - extent_start;
5493                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5494                                 size - extent_offset);
5495                 em->start = extent_start + extent_offset;
5496                 em->len = (copy_size + root->sectorsize - 1) &
5497                         ~((u64)root->sectorsize - 1);
5498                 em->orig_block_len = em->len;
5499                 em->orig_start = em->start;
5500                 if (compress_type) {
5501                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5502                         em->compress_type = compress_type;
5503                 }
5504                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5505                 if (create == 0 && !PageUptodate(page)) {
5506                         if (btrfs_file_extent_compression(leaf, item) !=
5507                             BTRFS_COMPRESS_NONE) {
5508                                 ret = uncompress_inline(path, inode, page,
5509                                                         pg_offset,
5510                                                         extent_offset, item);
5511                                 BUG_ON(ret); /* -ENOMEM */
5512                         } else {
5513                                 map = kmap(page);
5514                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5515                                                    copy_size);
5516                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5517                                         memset(map + pg_offset + copy_size, 0,
5518                                                PAGE_CACHE_SIZE - pg_offset -
5519                                                copy_size);
5520                                 }
5521                                 kunmap(page);
5522                         }
5523                         flush_dcache_page(page);
5524                 } else if (create && PageUptodate(page)) {
5525                         BUG();
5526                         if (!trans) {
5527                                 kunmap(page);
5528                                 free_extent_map(em);
5529                                 em = NULL;
5530
5531                                 btrfs_release_path(path);
5532                                 trans = btrfs_join_transaction(root);
5533
5534                                 if (IS_ERR(trans))
5535                                         return ERR_CAST(trans);
5536                                 goto again;
5537                         }
5538                         map = kmap(page);
5539                         write_extent_buffer(leaf, map + pg_offset, ptr,
5540                                             copy_size);
5541                         kunmap(page);
5542                         btrfs_mark_buffer_dirty(leaf);
5543                 }
5544                 set_extent_uptodate(io_tree, em->start,
5545                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5546                 goto insert;
5547         } else {
5548                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5549         }
5550 not_found:
5551         em->start = start;
5552         em->orig_start = start;
5553         em->len = len;
5554 not_found_em:
5555         em->block_start = EXTENT_MAP_HOLE;
5556         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5557 insert:
5558         btrfs_release_path(path);
5559         if (em->start > start || extent_map_end(em) <= start) {
5560                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5561                        "[%llu %llu]\n", (unsigned long long)em->start,
5562                        (unsigned long long)em->len,
5563                        (unsigned long long)start,
5564                        (unsigned long long)len);
5565                 err = -EIO;
5566                 goto out;
5567         }
5568
5569         err = 0;
5570         write_lock(&em_tree->lock);
5571         ret = add_extent_mapping(em_tree, em);
5572         /* it is possible that someone inserted the extent into the tree
5573          * while we had the lock dropped.  It is also possible that
5574          * an overlapping map exists in the tree
5575          */
5576         if (ret == -EEXIST) {
5577                 struct extent_map *existing;
5578
5579                 ret = 0;
5580
5581                 existing = lookup_extent_mapping(em_tree, start, len);
5582                 if (existing && (existing->start > start ||
5583                     existing->start + existing->len <= start)) {
5584                         free_extent_map(existing);
5585                         existing = NULL;
5586                 }
5587                 if (!existing) {
5588                         existing = lookup_extent_mapping(em_tree, em->start,
5589                                                          em->len);
5590                         if (existing) {
5591                                 err = merge_extent_mapping(em_tree, existing,
5592                                                            em, start,
5593                                                            root->sectorsize);
5594                                 free_extent_map(existing);
5595                                 if (err) {
5596                                         free_extent_map(em);
5597                                         em = NULL;
5598                                 }
5599                         } else {
5600                                 err = -EIO;
5601                                 free_extent_map(em);
5602                                 em = NULL;
5603                         }
5604                 } else {
5605                         free_extent_map(em);
5606                         em = existing;
5607                         err = 0;
5608                 }
5609         }
5610         write_unlock(&em_tree->lock);
5611 out:
5612
5613         if (em)
5614                 trace_btrfs_get_extent(root, em);
5615
5616         if (path)
5617                 btrfs_free_path(path);
5618         if (trans) {
5619                 ret = btrfs_end_transaction(trans, root);
5620                 if (!err)
5621                         err = ret;
5622         }
5623         if (err) {
5624                 free_extent_map(em);
5625                 return ERR_PTR(err);
5626         }
5627         BUG_ON(!em); /* Error is always set */
5628         return em;
5629 }
5630
5631 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5632                                            size_t pg_offset, u64 start, u64 len,
5633                                            int create)
5634 {
5635         struct extent_map *em;
5636         struct extent_map *hole_em = NULL;
5637         u64 range_start = start;
5638         u64 end;
5639         u64 found;
5640         u64 found_end;
5641         int err = 0;
5642
5643         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5644         if (IS_ERR(em))
5645                 return em;
5646         if (em) {
5647                 /*
5648                  * if our em maps to
5649                  * -  a hole or
5650                  * -  a pre-alloc extent,
5651                  * there might actually be delalloc bytes behind it.
5652                  */
5653                 if (em->block_start != EXTENT_MAP_HOLE &&
5654                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5655                         return em;
5656                 else
5657                         hole_em = em;
5658         }
5659
5660         /* check to see if we've wrapped (len == -1 or similar) */
5661         end = start + len;
5662         if (end < start)
5663                 end = (u64)-1;
5664         else
5665                 end -= 1;
5666
5667         em = NULL;
5668
5669         /* ok, we didn't find anything, lets look for delalloc */
5670         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5671                                  end, len, EXTENT_DELALLOC, 1);
5672         found_end = range_start + found;
5673         if (found_end < range_start)
5674                 found_end = (u64)-1;
5675
5676         /*
5677          * we didn't find anything useful, return
5678          * the original results from get_extent()
5679          */
5680         if (range_start > end || found_end <= start) {
5681                 em = hole_em;
5682                 hole_em = NULL;
5683                 goto out;
5684         }
5685
5686         /* adjust the range_start to make sure it doesn't
5687          * go backwards from the start they passed in
5688          */
5689         range_start = max(start,range_start);
5690         found = found_end - range_start;
5691
5692         if (found > 0) {
5693                 u64 hole_start = start;
5694                 u64 hole_len = len;
5695
5696                 em = alloc_extent_map();
5697                 if (!em) {
5698                         err = -ENOMEM;
5699                         goto out;
5700                 }
5701                 /*
5702                  * when btrfs_get_extent can't find anything it
5703                  * returns one huge hole
5704                  *
5705                  * make sure what it found really fits our range, and
5706                  * adjust to make sure it is based on the start from
5707                  * the caller
5708                  */
5709                 if (hole_em) {
5710                         u64 calc_end = extent_map_end(hole_em);
5711
5712                         if (calc_end <= start || (hole_em->start > end)) {
5713                                 free_extent_map(hole_em);
5714                                 hole_em = NULL;
5715                         } else {
5716                                 hole_start = max(hole_em->start, start);
5717                                 hole_len = calc_end - hole_start;
5718                         }
5719                 }
5720                 em->bdev = NULL;
5721                 if (hole_em && range_start > hole_start) {
5722                         /* our hole starts before our delalloc, so we
5723                          * have to return just the parts of the hole
5724                          * that go until  the delalloc starts
5725                          */
5726                         em->len = min(hole_len,
5727                                       range_start - hole_start);
5728                         em->start = hole_start;
5729                         em->orig_start = hole_start;
5730                         /*
5731                          * don't adjust block start at all,
5732                          * it is fixed at EXTENT_MAP_HOLE
5733                          */
5734                         em->block_start = hole_em->block_start;
5735                         em->block_len = hole_len;
5736                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5737                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5738                 } else {
5739                         em->start = range_start;
5740                         em->len = found;
5741                         em->orig_start = range_start;
5742                         em->block_start = EXTENT_MAP_DELALLOC;
5743                         em->block_len = found;
5744                 }
5745         } else if (hole_em) {
5746                 return hole_em;
5747         }
5748 out:
5749
5750         free_extent_map(hole_em);
5751         if (err) {
5752                 free_extent_map(em);
5753                 return ERR_PTR(err);
5754         }
5755         return em;
5756 }
5757
5758 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5759                                                   u64 start, u64 len)
5760 {
5761         struct btrfs_root *root = BTRFS_I(inode)->root;
5762         struct btrfs_trans_handle *trans;
5763         struct extent_map *em;
5764         struct btrfs_key ins;
5765         u64 alloc_hint;
5766         int ret;
5767
5768         trans = btrfs_join_transaction(root);
5769         if (IS_ERR(trans))
5770                 return ERR_CAST(trans);
5771
5772         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5773
5774         alloc_hint = get_extent_allocation_hint(inode, start, len);
5775         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5776                                    alloc_hint, &ins, 1);
5777         if (ret) {
5778                 em = ERR_PTR(ret);
5779                 goto out;
5780         }
5781
5782         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5783                               ins.offset, ins.offset, 0);
5784         if (IS_ERR(em))
5785                 goto out;
5786
5787         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5788                                            ins.offset, ins.offset, 0);
5789         if (ret) {
5790                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5791                 em = ERR_PTR(ret);
5792         }
5793 out:
5794         btrfs_end_transaction(trans, root);
5795         return em;
5796 }
5797
5798 /*
5799  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5800  * block must be cow'd
5801  */
5802 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5803                                       struct inode *inode, u64 offset, u64 len)
5804 {
5805         struct btrfs_path *path;
5806         int ret;
5807         struct extent_buffer *leaf;
5808         struct btrfs_root *root = BTRFS_I(inode)->root;
5809         struct btrfs_file_extent_item *fi;
5810         struct btrfs_key key;
5811         u64 disk_bytenr;
5812         u64 backref_offset;
5813         u64 extent_end;
5814         u64 num_bytes;
5815         int slot;
5816         int found_type;
5817
5818         path = btrfs_alloc_path();
5819         if (!path)
5820                 return -ENOMEM;
5821
5822         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5823                                        offset, 0);
5824         if (ret < 0)
5825                 goto out;
5826
5827         slot = path->slots[0];
5828         if (ret == 1) {
5829                 if (slot == 0) {
5830                         /* can't find the item, must cow */
5831                         ret = 0;
5832                         goto out;
5833                 }
5834                 slot--;
5835         }
5836         ret = 0;
5837         leaf = path->nodes[0];
5838         btrfs_item_key_to_cpu(leaf, &key, slot);
5839         if (key.objectid != btrfs_ino(inode) ||
5840             key.type != BTRFS_EXTENT_DATA_KEY) {
5841                 /* not our file or wrong item type, must cow */
5842                 goto out;
5843         }
5844
5845         if (key.offset > offset) {
5846                 /* Wrong offset, must cow */
5847                 goto out;
5848         }
5849
5850         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5851         found_type = btrfs_file_extent_type(leaf, fi);
5852         if (found_type != BTRFS_FILE_EXTENT_REG &&
5853             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5854                 /* not a regular extent, must cow */
5855                 goto out;
5856         }
5857         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5858         backref_offset = btrfs_file_extent_offset(leaf, fi);
5859
5860         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5861         if (extent_end < offset + len) {
5862                 /* extent doesn't include our full range, must cow */
5863                 goto out;
5864         }
5865
5866         if (btrfs_extent_readonly(root, disk_bytenr))
5867                 goto out;
5868
5869         /*
5870          * look for other files referencing this extent, if we
5871          * find any we must cow
5872          */
5873         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5874                                   key.offset - backref_offset, disk_bytenr))
5875                 goto out;
5876
5877         /*
5878          * adjust disk_bytenr and num_bytes to cover just the bytes
5879          * in this extent we are about to write.  If there
5880          * are any csums in that range we have to cow in order
5881          * to keep the csums correct
5882          */
5883         disk_bytenr += backref_offset;
5884         disk_bytenr += offset - key.offset;
5885         num_bytes = min(offset + len, extent_end) - offset;
5886         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5887                                 goto out;
5888         /*
5889          * all of the above have passed, it is safe to overwrite this extent
5890          * without cow
5891          */
5892         ret = 1;
5893 out:
5894         btrfs_free_path(path);
5895         return ret;
5896 }
5897
5898 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5899                               struct extent_state **cached_state, int writing)
5900 {
5901         struct btrfs_ordered_extent *ordered;
5902         int ret = 0;
5903
5904         while (1) {
5905                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5906                                  0, cached_state);
5907                 /*
5908                  * We're concerned with the entire range that we're going to be
5909                  * doing DIO to, so we need to make sure theres no ordered
5910                  * extents in this range.
5911                  */
5912                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5913                                                      lockend - lockstart + 1);
5914
5915                 /*
5916                  * We need to make sure there are no buffered pages in this
5917                  * range either, we could have raced between the invalidate in
5918                  * generic_file_direct_write and locking the extent.  The
5919                  * invalidate needs to happen so that reads after a write do not
5920                  * get stale data.
5921                  */
5922                 if (!ordered && (!writing ||
5923                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5924                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5925                                     *cached_state)))
5926                         break;
5927
5928                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5929                                      cached_state, GFP_NOFS);
5930
5931                 if (ordered) {
5932                         btrfs_start_ordered_extent(inode, ordered, 1);
5933                         btrfs_put_ordered_extent(ordered);
5934                 } else {
5935                         /* Screw you mmap */
5936                         ret = filemap_write_and_wait_range(inode->i_mapping,
5937                                                            lockstart,
5938                                                            lockend);
5939                         if (ret)
5940                                 break;
5941
5942                         /*
5943                          * If we found a page that couldn't be invalidated just
5944                          * fall back to buffered.
5945                          */
5946                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5947                                         lockstart >> PAGE_CACHE_SHIFT,
5948                                         lockend >> PAGE_CACHE_SHIFT);
5949                         if (ret)
5950                                 break;
5951                 }
5952
5953                 cond_resched();
5954         }
5955
5956         return ret;
5957 }
5958
5959 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5960                                            u64 len, u64 orig_start,
5961                                            u64 block_start, u64 block_len,
5962                                            u64 orig_block_len, int type)
5963 {
5964         struct extent_map_tree *em_tree;
5965         struct extent_map *em;
5966         struct btrfs_root *root = BTRFS_I(inode)->root;
5967         int ret;
5968
5969         em_tree = &BTRFS_I(inode)->extent_tree;
5970         em = alloc_extent_map();
5971         if (!em)
5972                 return ERR_PTR(-ENOMEM);
5973
5974         em->start = start;
5975         em->orig_start = orig_start;
5976         em->mod_start = start;
5977         em->mod_len = len;
5978         em->len = len;
5979         em->block_len = block_len;
5980         em->block_start = block_start;
5981         em->bdev = root->fs_info->fs_devices->latest_bdev;
5982         em->orig_block_len = orig_block_len;
5983         em->generation = -1;
5984         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5985         if (type == BTRFS_ORDERED_PREALLOC)
5986                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5987
5988         do {
5989                 btrfs_drop_extent_cache(inode, em->start,
5990                                 em->start + em->len - 1, 0);
5991                 write_lock(&em_tree->lock);
5992                 ret = add_extent_mapping(em_tree, em);
5993                 if (!ret)
5994                         list_move(&em->list,
5995                                   &em_tree->modified_extents);
5996                 write_unlock(&em_tree->lock);
5997         } while (ret == -EEXIST);
5998
5999         if (ret) {
6000                 free_extent_map(em);
6001                 return ERR_PTR(ret);
6002         }
6003
6004         return em;
6005 }
6006
6007
6008 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6009                                    struct buffer_head *bh_result, int create)
6010 {
6011         struct extent_map *em;
6012         struct btrfs_root *root = BTRFS_I(inode)->root;
6013         struct extent_state *cached_state = NULL;
6014         u64 start = iblock << inode->i_blkbits;
6015         u64 lockstart, lockend;
6016         u64 len = bh_result->b_size;
6017         struct btrfs_trans_handle *trans;
6018         int unlock_bits = EXTENT_LOCKED;
6019         int ret;
6020
6021         if (create) {
6022                 ret = btrfs_delalloc_reserve_space(inode, len);
6023                 if (ret)
6024                         return ret;
6025                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6026         } else {
6027                 len = min_t(u64, len, root->sectorsize);
6028         }
6029
6030         lockstart = start;
6031         lockend = start + len - 1;
6032
6033         /*
6034          * If this errors out it's because we couldn't invalidate pagecache for
6035          * this range and we need to fallback to buffered.
6036          */
6037         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6038                 return -ENOTBLK;
6039
6040         if (create) {
6041                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6042                                      lockend, EXTENT_DELALLOC, NULL,
6043                                      &cached_state, GFP_NOFS);
6044                 if (ret)
6045                         goto unlock_err;
6046         }
6047
6048         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6049         if (IS_ERR(em)) {
6050                 ret = PTR_ERR(em);
6051                 goto unlock_err;
6052         }
6053
6054         /*
6055          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6056          * io.  INLINE is special, and we could probably kludge it in here, but
6057          * it's still buffered so for safety lets just fall back to the generic
6058          * buffered path.
6059          *
6060          * For COMPRESSED we _have_ to read the entire extent in so we can
6061          * decompress it, so there will be buffering required no matter what we
6062          * do, so go ahead and fallback to buffered.
6063          *
6064          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6065          * to buffered IO.  Don't blame me, this is the price we pay for using
6066          * the generic code.
6067          */
6068         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6069             em->block_start == EXTENT_MAP_INLINE) {
6070                 free_extent_map(em);
6071                 ret = -ENOTBLK;
6072                 goto unlock_err;
6073         }
6074
6075         /* Just a good old fashioned hole, return */
6076         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6077                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6078                 free_extent_map(em);
6079                 ret = 0;
6080                 goto unlock_err;
6081         }
6082
6083         /*
6084          * We don't allocate a new extent in the following cases
6085          *
6086          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6087          * existing extent.
6088          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6089          * just use the extent.
6090          *
6091          */
6092         if (!create) {
6093                 len = min(len, em->len - (start - em->start));
6094                 lockstart = start + len;
6095                 goto unlock;
6096         }
6097
6098         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6099             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6100              em->block_start != EXTENT_MAP_HOLE)) {
6101                 int type;
6102                 int ret;
6103                 u64 block_start;
6104
6105                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6106                         type = BTRFS_ORDERED_PREALLOC;
6107                 else
6108                         type = BTRFS_ORDERED_NOCOW;
6109                 len = min(len, em->len - (start - em->start));
6110                 block_start = em->block_start + (start - em->start);
6111
6112                 /*
6113                  * we're not going to log anything, but we do need
6114                  * to make sure the current transaction stays open
6115                  * while we look for nocow cross refs
6116                  */
6117                 trans = btrfs_join_transaction(root);
6118                 if (IS_ERR(trans))
6119                         goto must_cow;
6120
6121                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6122                         u64 orig_start = em->orig_start;
6123                         u64 orig_block_len = em->orig_block_len;
6124
6125                         if (type == BTRFS_ORDERED_PREALLOC) {
6126                                 free_extent_map(em);
6127                                 em = create_pinned_em(inode, start, len,
6128                                                        orig_start,
6129                                                        block_start, len,
6130                                                        orig_block_len, type);
6131                                 if (IS_ERR(em)) {
6132                                         btrfs_end_transaction(trans, root);
6133                                         goto unlock_err;
6134                                 }
6135                         }
6136
6137                         ret = btrfs_add_ordered_extent_dio(inode, start,
6138                                            block_start, len, len, type);
6139                         btrfs_end_transaction(trans, root);
6140                         if (ret) {
6141                                 free_extent_map(em);
6142                                 goto unlock_err;
6143                         }
6144                         goto unlock;
6145                 }
6146                 btrfs_end_transaction(trans, root);
6147         }
6148 must_cow:
6149         /*
6150          * this will cow the extent, reset the len in case we changed
6151          * it above
6152          */
6153         len = bh_result->b_size;
6154         free_extent_map(em);
6155         em = btrfs_new_extent_direct(inode, start, len);
6156         if (IS_ERR(em)) {
6157                 ret = PTR_ERR(em);
6158                 goto unlock_err;
6159         }
6160         len = min(len, em->len - (start - em->start));
6161 unlock:
6162         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6163                 inode->i_blkbits;
6164         bh_result->b_size = len;
6165         bh_result->b_bdev = em->bdev;
6166         set_buffer_mapped(bh_result);
6167         if (create) {
6168                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6169                         set_buffer_new(bh_result);
6170
6171                 /*
6172                  * Need to update the i_size under the extent lock so buffered
6173                  * readers will get the updated i_size when we unlock.
6174                  */
6175                 if (start + len > i_size_read(inode))
6176                         i_size_write(inode, start + len);
6177         }
6178
6179         /*
6180          * In the case of write we need to clear and unlock the entire range,
6181          * in the case of read we need to unlock only the end area that we
6182          * aren't using if there is any left over space.
6183          */
6184         if (lockstart < lockend) {
6185                 if (create && len < lockend - lockstart) {
6186                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6187                                          lockstart + len - 1,
6188                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6189                                          &cached_state, GFP_NOFS);
6190                         /*
6191                          * Beside unlock, we also need to cleanup reserved space
6192                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6193                          */
6194                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6195                                          lockstart + len, lockend,
6196                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6197                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6198                 } else {
6199                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6200                                          lockend, unlock_bits, 1, 0,
6201                                          &cached_state, GFP_NOFS);
6202                 }
6203         } else {
6204                 free_extent_state(cached_state);
6205         }
6206
6207         free_extent_map(em);
6208
6209         return 0;
6210
6211 unlock_err:
6212         if (create)
6213                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6214
6215         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6216                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6217         return ret;
6218 }
6219
6220 struct btrfs_dio_private {
6221         struct inode *inode;
6222         u64 logical_offset;
6223         u64 disk_bytenr;
6224         u64 bytes;
6225         void *private;
6226
6227         /* number of bios pending for this dio */
6228         atomic_t pending_bios;
6229
6230         /* IO errors */
6231         int errors;
6232
6233         struct bio *orig_bio;
6234 };
6235
6236 static void btrfs_endio_direct_read(struct bio *bio, int err)
6237 {
6238         struct btrfs_dio_private *dip = bio->bi_private;
6239         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6240         struct bio_vec *bvec = bio->bi_io_vec;
6241         struct inode *inode = dip->inode;
6242         struct btrfs_root *root = BTRFS_I(inode)->root;
6243         u64 start;
6244
6245         start = dip->logical_offset;
6246         do {
6247                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6248                         struct page *page = bvec->bv_page;
6249                         char *kaddr;
6250                         u32 csum = ~(u32)0;
6251                         u64 private = ~(u32)0;
6252                         unsigned long flags;
6253
6254                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6255                                               start, &private))
6256                                 goto failed;
6257                         local_irq_save(flags);
6258                         kaddr = kmap_atomic(page);
6259                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6260                                                csum, bvec->bv_len);
6261                         btrfs_csum_final(csum, (char *)&csum);
6262                         kunmap_atomic(kaddr);
6263                         local_irq_restore(flags);
6264
6265                         flush_dcache_page(bvec->bv_page);
6266                         if (csum != private) {
6267 failed:
6268                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6269                                       " %llu csum %u private %u\n",
6270                                       (unsigned long long)btrfs_ino(inode),
6271                                       (unsigned long long)start,
6272                                       csum, (unsigned)private);
6273                                 err = -EIO;
6274                         }
6275                 }
6276
6277                 start += bvec->bv_len;
6278                 bvec++;
6279         } while (bvec <= bvec_end);
6280
6281         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6282                       dip->logical_offset + dip->bytes - 1);
6283         bio->bi_private = dip->private;
6284
6285         kfree(dip);
6286
6287         /* If we had a csum failure make sure to clear the uptodate flag */
6288         if (err)
6289                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6290         dio_end_io(bio, err);
6291 }
6292
6293 static void btrfs_endio_direct_write(struct bio *bio, int err)
6294 {
6295         struct btrfs_dio_private *dip = bio->bi_private;
6296         struct inode *inode = dip->inode;
6297         struct btrfs_root *root = BTRFS_I(inode)->root;
6298         struct btrfs_ordered_extent *ordered = NULL;
6299         u64 ordered_offset = dip->logical_offset;
6300         u64 ordered_bytes = dip->bytes;
6301         int ret;
6302
6303         if (err)
6304                 goto out_done;
6305 again:
6306         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6307                                                    &ordered_offset,
6308                                                    ordered_bytes, !err);
6309         if (!ret)
6310                 goto out_test;
6311
6312         ordered->work.func = finish_ordered_fn;
6313         ordered->work.flags = 0;
6314         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6315                            &ordered->work);
6316 out_test:
6317         /*
6318          * our bio might span multiple ordered extents.  If we haven't
6319          * completed the accounting for the whole dio, go back and try again
6320          */
6321         if (ordered_offset < dip->logical_offset + dip->bytes) {
6322                 ordered_bytes = dip->logical_offset + dip->bytes -
6323                         ordered_offset;
6324                 ordered = NULL;
6325                 goto again;
6326         }
6327 out_done:
6328         bio->bi_private = dip->private;
6329
6330         kfree(dip);
6331
6332         /* If we had an error make sure to clear the uptodate flag */
6333         if (err)
6334                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6335         dio_end_io(bio, err);
6336 }
6337
6338 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6339                                     struct bio *bio, int mirror_num,
6340                                     unsigned long bio_flags, u64 offset)
6341 {
6342         int ret;
6343         struct btrfs_root *root = BTRFS_I(inode)->root;
6344         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6345         BUG_ON(ret); /* -ENOMEM */
6346         return 0;
6347 }
6348
6349 static void btrfs_end_dio_bio(struct bio *bio, int err)
6350 {
6351         struct btrfs_dio_private *dip = bio->bi_private;
6352
6353         if (err) {
6354                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6355                       "sector %#Lx len %u err no %d\n",
6356                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6357                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6358                 dip->errors = 1;
6359
6360                 /*
6361                  * before atomic variable goto zero, we must make sure
6362                  * dip->errors is perceived to be set.
6363                  */
6364                 smp_mb__before_atomic_dec();
6365         }
6366
6367         /* if there are more bios still pending for this dio, just exit */
6368         if (!atomic_dec_and_test(&dip->pending_bios))
6369                 goto out;
6370
6371         if (dip->errors)
6372                 bio_io_error(dip->orig_bio);
6373         else {
6374                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6375                 bio_endio(dip->orig_bio, 0);
6376         }
6377 out:
6378         bio_put(bio);
6379 }
6380
6381 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6382                                        u64 first_sector, gfp_t gfp_flags)
6383 {
6384         int nr_vecs = bio_get_nr_vecs(bdev);
6385         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6386 }
6387
6388 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6389                                          int rw, u64 file_offset, int skip_sum,
6390                                          int async_submit)
6391 {
6392         int write = rw & REQ_WRITE;
6393         struct btrfs_root *root = BTRFS_I(inode)->root;
6394         int ret;
6395
6396         if (async_submit)
6397                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6398
6399         bio_get(bio);
6400
6401         if (!write) {
6402                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6403                 if (ret)
6404                         goto err;
6405         }
6406
6407         if (skip_sum)
6408                 goto map;
6409
6410         if (write && async_submit) {
6411                 ret = btrfs_wq_submit_bio(root->fs_info,
6412                                    inode, rw, bio, 0, 0,
6413                                    file_offset,
6414                                    __btrfs_submit_bio_start_direct_io,
6415                                    __btrfs_submit_bio_done);
6416                 goto err;
6417         } else if (write) {
6418                 /*
6419                  * If we aren't doing async submit, calculate the csum of the
6420                  * bio now.
6421                  */
6422                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6423                 if (ret)
6424                         goto err;
6425         } else if (!skip_sum) {
6426                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6427                 if (ret)
6428                         goto err;
6429         }
6430
6431 map:
6432         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6433 err:
6434         bio_put(bio);
6435         return ret;
6436 }
6437
6438 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6439                                     int skip_sum)
6440 {
6441         struct inode *inode = dip->inode;
6442         struct btrfs_root *root = BTRFS_I(inode)->root;
6443         struct bio *bio;
6444         struct bio *orig_bio = dip->orig_bio;
6445         struct bio_vec *bvec = orig_bio->bi_io_vec;
6446         u64 start_sector = orig_bio->bi_sector;
6447         u64 file_offset = dip->logical_offset;
6448         u64 submit_len = 0;
6449         u64 map_length;
6450         int nr_pages = 0;
6451         int ret = 0;
6452         int async_submit = 0;
6453
6454         map_length = orig_bio->bi_size;
6455         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6456                               &map_length, NULL, 0);
6457         if (ret) {
6458                 bio_put(orig_bio);
6459                 return -EIO;
6460         }
6461
6462         if (map_length >= orig_bio->bi_size) {
6463                 bio = orig_bio;
6464                 goto submit;
6465         }
6466
6467         async_submit = 1;
6468         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6469         if (!bio)
6470                 return -ENOMEM;
6471         bio->bi_private = dip;
6472         bio->bi_end_io = btrfs_end_dio_bio;
6473         atomic_inc(&dip->pending_bios);
6474
6475         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6476                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6477                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6478                                  bvec->bv_offset) < bvec->bv_len)) {
6479                         /*
6480                          * inc the count before we submit the bio so
6481                          * we know the end IO handler won't happen before
6482                          * we inc the count. Otherwise, the dip might get freed
6483                          * before we're done setting it up
6484                          */
6485                         atomic_inc(&dip->pending_bios);
6486                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6487                                                      file_offset, skip_sum,
6488                                                      async_submit);
6489                         if (ret) {
6490                                 bio_put(bio);
6491                                 atomic_dec(&dip->pending_bios);
6492                                 goto out_err;
6493                         }
6494
6495                         start_sector += submit_len >> 9;
6496                         file_offset += submit_len;
6497
6498                         submit_len = 0;
6499                         nr_pages = 0;
6500
6501                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6502                                                   start_sector, GFP_NOFS);
6503                         if (!bio)
6504                                 goto out_err;
6505                         bio->bi_private = dip;
6506                         bio->bi_end_io = btrfs_end_dio_bio;
6507
6508                         map_length = orig_bio->bi_size;
6509                         ret = btrfs_map_block(root->fs_info, READ,
6510                                               start_sector << 9,
6511                                               &map_length, NULL, 0);
6512                         if (ret) {
6513                                 bio_put(bio);
6514                                 goto out_err;
6515                         }
6516                 } else {
6517                         submit_len += bvec->bv_len;
6518                         nr_pages ++;
6519                         bvec++;
6520                 }
6521         }
6522
6523 submit:
6524         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6525                                      async_submit);
6526         if (!ret)
6527                 return 0;
6528
6529         bio_put(bio);
6530 out_err:
6531         dip->errors = 1;
6532         /*
6533          * before atomic variable goto zero, we must
6534          * make sure dip->errors is perceived to be set.
6535          */
6536         smp_mb__before_atomic_dec();
6537         if (atomic_dec_and_test(&dip->pending_bios))
6538                 bio_io_error(dip->orig_bio);
6539
6540         /* bio_end_io() will handle error, so we needn't return it */
6541         return 0;
6542 }
6543
6544 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6545                                 loff_t file_offset)
6546 {
6547         struct btrfs_root *root = BTRFS_I(inode)->root;
6548         struct btrfs_dio_private *dip;
6549         struct bio_vec *bvec = bio->bi_io_vec;
6550         int skip_sum;
6551         int write = rw & REQ_WRITE;
6552         int ret = 0;
6553
6554         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6555
6556         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6557         if (!dip) {
6558                 ret = -ENOMEM;
6559                 goto free_ordered;
6560         }
6561
6562         dip->private = bio->bi_private;
6563         dip->inode = inode;
6564         dip->logical_offset = file_offset;
6565
6566         dip->bytes = 0;
6567         do {
6568                 dip->bytes += bvec->bv_len;
6569                 bvec++;
6570         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6571
6572         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6573         bio->bi_private = dip;
6574         dip->errors = 0;
6575         dip->orig_bio = bio;
6576         atomic_set(&dip->pending_bios, 0);
6577
6578         if (write)
6579                 bio->bi_end_io = btrfs_endio_direct_write;
6580         else
6581                 bio->bi_end_io = btrfs_endio_direct_read;
6582
6583         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6584         if (!ret)
6585                 return;
6586 free_ordered:
6587         /*
6588          * If this is a write, we need to clean up the reserved space and kill
6589          * the ordered extent.
6590          */
6591         if (write) {
6592                 struct btrfs_ordered_extent *ordered;
6593                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6594                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6595                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6596                         btrfs_free_reserved_extent(root, ordered->start,
6597                                                    ordered->disk_len);
6598                 btrfs_put_ordered_extent(ordered);
6599                 btrfs_put_ordered_extent(ordered);
6600         }
6601         bio_endio(bio, ret);
6602 }
6603
6604 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6605                         const struct iovec *iov, loff_t offset,
6606                         unsigned long nr_segs)
6607 {
6608         int seg;
6609         int i;
6610         size_t size;
6611         unsigned long addr;
6612         unsigned blocksize_mask = root->sectorsize - 1;
6613         ssize_t retval = -EINVAL;
6614         loff_t end = offset;
6615
6616         if (offset & blocksize_mask)
6617                 goto out;
6618
6619         /* Check the memory alignment.  Blocks cannot straddle pages */
6620         for (seg = 0; seg < nr_segs; seg++) {
6621                 addr = (unsigned long)iov[seg].iov_base;
6622                 size = iov[seg].iov_len;
6623                 end += size;
6624                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6625                         goto out;
6626
6627                 /* If this is a write we don't need to check anymore */
6628                 if (rw & WRITE)
6629                         continue;
6630
6631                 /*
6632                  * Check to make sure we don't have duplicate iov_base's in this
6633                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6634                  * when reading back.
6635                  */
6636                 for (i = seg + 1; i < nr_segs; i++) {
6637                         if (iov[seg].iov_base == iov[i].iov_base)
6638                                 goto out;
6639                 }
6640         }
6641         retval = 0;
6642 out:
6643         return retval;
6644 }
6645
6646 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6647                         const struct iovec *iov, loff_t offset,
6648                         unsigned long nr_segs)
6649 {
6650         struct file *file = iocb->ki_filp;
6651         struct inode *inode = file->f_mapping->host;
6652
6653         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6654                             offset, nr_segs))
6655                 return 0;
6656
6657         return __blockdev_direct_IO(rw, iocb, inode,
6658                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6659                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6660                    btrfs_submit_direct, 0);
6661 }
6662
6663 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6664
6665 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6666                 __u64 start, __u64 len)
6667 {
6668         int     ret;
6669
6670         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6671         if (ret)
6672                 return ret;
6673
6674         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6675 }
6676
6677 int btrfs_readpage(struct file *file, struct page *page)
6678 {
6679         struct extent_io_tree *tree;
6680         tree = &BTRFS_I(page->mapping->host)->io_tree;
6681         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6682 }
6683
6684 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6685 {
6686         struct extent_io_tree *tree;
6687
6688
6689         if (current->flags & PF_MEMALLOC) {
6690                 redirty_page_for_writepage(wbc, page);
6691                 unlock_page(page);
6692                 return 0;
6693         }
6694         tree = &BTRFS_I(page->mapping->host)->io_tree;
6695         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6696 }
6697
6698 int btrfs_writepages(struct address_space *mapping,
6699                      struct writeback_control *wbc)
6700 {
6701         struct extent_io_tree *tree;
6702
6703         tree = &BTRFS_I(mapping->host)->io_tree;
6704         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6705 }
6706
6707 static int
6708 btrfs_readpages(struct file *file, struct address_space *mapping,
6709                 struct list_head *pages, unsigned nr_pages)
6710 {
6711         struct extent_io_tree *tree;
6712         tree = &BTRFS_I(mapping->host)->io_tree;
6713         return extent_readpages(tree, mapping, pages, nr_pages,
6714                                 btrfs_get_extent);
6715 }
6716 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6717 {
6718         struct extent_io_tree *tree;
6719         struct extent_map_tree *map;
6720         int ret;
6721
6722         tree = &BTRFS_I(page->mapping->host)->io_tree;
6723         map = &BTRFS_I(page->mapping->host)->extent_tree;
6724         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6725         if (ret == 1) {
6726                 ClearPagePrivate(page);
6727                 set_page_private(page, 0);
6728                 page_cache_release(page);
6729         }
6730         return ret;
6731 }
6732
6733 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6734 {
6735         if (PageWriteback(page) || PageDirty(page))
6736                 return 0;
6737         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6738 }
6739
6740 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6741 {
6742         struct inode *inode = page->mapping->host;
6743         struct extent_io_tree *tree;
6744         struct btrfs_ordered_extent *ordered;
6745         struct extent_state *cached_state = NULL;
6746         u64 page_start = page_offset(page);
6747         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6748
6749         /*
6750          * we have the page locked, so new writeback can't start,
6751          * and the dirty bit won't be cleared while we are here.
6752          *
6753          * Wait for IO on this page so that we can safely clear
6754          * the PagePrivate2 bit and do ordered accounting
6755          */
6756         wait_on_page_writeback(page);
6757
6758         tree = &BTRFS_I(inode)->io_tree;
6759         if (offset) {
6760                 btrfs_releasepage(page, GFP_NOFS);
6761                 return;
6762         }
6763         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6764         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
6765         if (ordered) {
6766                 /*
6767                  * IO on this page will never be started, so we need
6768                  * to account for any ordered extents now
6769                  */
6770                 clear_extent_bit(tree, page_start, page_end,
6771                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6772                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6773                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6774                 /*
6775                  * whoever cleared the private bit is responsible
6776                  * for the finish_ordered_io
6777                  */
6778                 if (TestClearPagePrivate2(page) &&
6779                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6780                                                    PAGE_CACHE_SIZE, 1)) {
6781                         btrfs_finish_ordered_io(ordered);
6782                 }
6783                 btrfs_put_ordered_extent(ordered);
6784                 cached_state = NULL;
6785                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6786         }
6787         clear_extent_bit(tree, page_start, page_end,
6788                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6789                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6790                  &cached_state, GFP_NOFS);
6791         __btrfs_releasepage(page, GFP_NOFS);
6792
6793         ClearPageChecked(page);
6794         if (PagePrivate(page)) {
6795                 ClearPagePrivate(page);
6796                 set_page_private(page, 0);
6797                 page_cache_release(page);
6798         }
6799 }
6800
6801 /*
6802  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6803  * called from a page fault handler when a page is first dirtied. Hence we must
6804  * be careful to check for EOF conditions here. We set the page up correctly
6805  * for a written page which means we get ENOSPC checking when writing into
6806  * holes and correct delalloc and unwritten extent mapping on filesystems that
6807  * support these features.
6808  *
6809  * We are not allowed to take the i_mutex here so we have to play games to
6810  * protect against truncate races as the page could now be beyond EOF.  Because
6811  * vmtruncate() writes the inode size before removing pages, once we have the
6812  * page lock we can determine safely if the page is beyond EOF. If it is not
6813  * beyond EOF, then the page is guaranteed safe against truncation until we
6814  * unlock the page.
6815  */
6816 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6817 {
6818         struct page *page = vmf->page;
6819         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6820         struct btrfs_root *root = BTRFS_I(inode)->root;
6821         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6822         struct btrfs_ordered_extent *ordered;
6823         struct extent_state *cached_state = NULL;
6824         char *kaddr;
6825         unsigned long zero_start;
6826         loff_t size;
6827         int ret;
6828         int reserved = 0;
6829         u64 page_start;
6830         u64 page_end;
6831
6832         sb_start_pagefault(inode->i_sb);
6833         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6834         if (!ret) {
6835                 ret = file_update_time(vma->vm_file);
6836                 reserved = 1;
6837         }
6838         if (ret) {
6839                 if (ret == -ENOMEM)
6840                         ret = VM_FAULT_OOM;
6841                 else /* -ENOSPC, -EIO, etc */
6842                         ret = VM_FAULT_SIGBUS;
6843                 if (reserved)
6844                         goto out;
6845                 goto out_noreserve;
6846         }
6847
6848         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6849 again:
6850         lock_page(page);
6851         size = i_size_read(inode);
6852         page_start = page_offset(page);
6853         page_end = page_start + PAGE_CACHE_SIZE - 1;
6854
6855         if ((page->mapping != inode->i_mapping) ||
6856             (page_start >= size)) {
6857                 /* page got truncated out from underneath us */
6858                 goto out_unlock;
6859         }
6860         wait_on_page_writeback(page);
6861
6862         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6863         set_page_extent_mapped(page);
6864
6865         /*
6866          * we can't set the delalloc bits if there are pending ordered
6867          * extents.  Drop our locks and wait for them to finish
6868          */
6869         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6870         if (ordered) {
6871                 unlock_extent_cached(io_tree, page_start, page_end,
6872                                      &cached_state, GFP_NOFS);
6873                 unlock_page(page);
6874                 btrfs_start_ordered_extent(inode, ordered, 1);
6875                 btrfs_put_ordered_extent(ordered);
6876                 goto again;
6877         }
6878
6879         /*
6880          * XXX - page_mkwrite gets called every time the page is dirtied, even
6881          * if it was already dirty, so for space accounting reasons we need to
6882          * clear any delalloc bits for the range we are fixing to save.  There
6883          * is probably a better way to do this, but for now keep consistent with
6884          * prepare_pages in the normal write path.
6885          */
6886         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6887                           EXTENT_DIRTY | EXTENT_DELALLOC |
6888                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6889                           0, 0, &cached_state, GFP_NOFS);
6890
6891         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6892                                         &cached_state);
6893         if (ret) {
6894                 unlock_extent_cached(io_tree, page_start, page_end,
6895                                      &cached_state, GFP_NOFS);
6896                 ret = VM_FAULT_SIGBUS;
6897                 goto out_unlock;
6898         }
6899         ret = 0;
6900
6901         /* page is wholly or partially inside EOF */
6902         if (page_start + PAGE_CACHE_SIZE > size)
6903                 zero_start = size & ~PAGE_CACHE_MASK;
6904         else
6905                 zero_start = PAGE_CACHE_SIZE;
6906
6907         if (zero_start != PAGE_CACHE_SIZE) {
6908                 kaddr = kmap(page);
6909                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6910                 flush_dcache_page(page);
6911                 kunmap(page);
6912         }
6913         ClearPageChecked(page);
6914         set_page_dirty(page);
6915         SetPageUptodate(page);
6916
6917         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6918         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6919         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6920
6921         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6922
6923 out_unlock:
6924         if (!ret) {
6925                 sb_end_pagefault(inode->i_sb);
6926                 return VM_FAULT_LOCKED;
6927         }
6928         unlock_page(page);
6929 out:
6930         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6931 out_noreserve:
6932         sb_end_pagefault(inode->i_sb);
6933         return ret;
6934 }
6935
6936 static int btrfs_truncate(struct inode *inode)
6937 {
6938         struct btrfs_root *root = BTRFS_I(inode)->root;
6939         struct btrfs_block_rsv *rsv;
6940         int ret;
6941         int err = 0;
6942         struct btrfs_trans_handle *trans;
6943         u64 mask = root->sectorsize - 1;
6944         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6945
6946         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6947         if (ret)
6948                 return ret;
6949
6950         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6951         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6952
6953         /*
6954          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6955          * 3 things going on here
6956          *
6957          * 1) We need to reserve space for our orphan item and the space to
6958          * delete our orphan item.  Lord knows we don't want to have a dangling
6959          * orphan item because we didn't reserve space to remove it.
6960          *
6961          * 2) We need to reserve space to update our inode.
6962          *
6963          * 3) We need to have something to cache all the space that is going to
6964          * be free'd up by the truncate operation, but also have some slack
6965          * space reserved in case it uses space during the truncate (thank you
6966          * very much snapshotting).
6967          *
6968          * And we need these to all be seperate.  The fact is we can use alot of
6969          * space doing the truncate, and we have no earthly idea how much space
6970          * we will use, so we need the truncate reservation to be seperate so it
6971          * doesn't end up using space reserved for updating the inode or
6972          * removing the orphan item.  We also need to be able to stop the
6973          * transaction and start a new one, which means we need to be able to
6974          * update the inode several times, and we have no idea of knowing how
6975          * many times that will be, so we can't just reserve 1 item for the
6976          * entirety of the opration, so that has to be done seperately as well.
6977          * Then there is the orphan item, which does indeed need to be held on
6978          * to for the whole operation, and we need nobody to touch this reserved
6979          * space except the orphan code.
6980          *
6981          * So that leaves us with
6982          *
6983          * 1) root->orphan_block_rsv - for the orphan deletion.
6984          * 2) rsv - for the truncate reservation, which we will steal from the
6985          * transaction reservation.
6986          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6987          * updating the inode.
6988          */
6989         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6990         if (!rsv)
6991                 return -ENOMEM;
6992         rsv->size = min_size;
6993         rsv->failfast = 1;
6994
6995         /*
6996          * 1 for the truncate slack space
6997          * 1 for updating the inode.
6998          */
6999         trans = btrfs_start_transaction(root, 2);
7000         if (IS_ERR(trans)) {
7001                 err = PTR_ERR(trans);
7002                 goto out;
7003         }
7004
7005         /* Migrate the slack space for the truncate to our reserve */
7006         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7007                                       min_size);
7008         BUG_ON(ret);
7009
7010         /*
7011          * setattr is responsible for setting the ordered_data_close flag,
7012          * but that is only tested during the last file release.  That
7013          * could happen well after the next commit, leaving a great big
7014          * window where new writes may get lost if someone chooses to write
7015          * to this file after truncating to zero
7016          *
7017          * The inode doesn't have any dirty data here, and so if we commit
7018          * this is a noop.  If someone immediately starts writing to the inode
7019          * it is very likely we'll catch some of their writes in this
7020          * transaction, and the commit will find this file on the ordered
7021          * data list with good things to send down.
7022          *
7023          * This is a best effort solution, there is still a window where
7024          * using truncate to replace the contents of the file will
7025          * end up with a zero length file after a crash.
7026          */
7027         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7028                                            &BTRFS_I(inode)->runtime_flags))
7029                 btrfs_add_ordered_operation(trans, root, inode);
7030
7031         /*
7032          * So if we truncate and then write and fsync we normally would just
7033          * write the extents that changed, which is a problem if we need to
7034          * first truncate that entire inode.  So set this flag so we write out
7035          * all of the extents in the inode to the sync log so we're completely
7036          * safe.
7037          */
7038         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7039         trans->block_rsv = rsv;
7040
7041         while (1) {
7042                 ret = btrfs_truncate_inode_items(trans, root, inode,
7043                                                  inode->i_size,
7044                                                  BTRFS_EXTENT_DATA_KEY);
7045                 if (ret != -ENOSPC) {
7046                         err = ret;
7047                         break;
7048                 }
7049
7050                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7051                 ret = btrfs_update_inode(trans, root, inode);
7052                 if (ret) {
7053                         err = ret;
7054                         break;
7055                 }
7056
7057                 btrfs_end_transaction(trans, root);
7058                 btrfs_btree_balance_dirty(root);
7059
7060                 trans = btrfs_start_transaction(root, 2);
7061                 if (IS_ERR(trans)) {
7062                         ret = err = PTR_ERR(trans);
7063                         trans = NULL;
7064                         break;
7065                 }
7066
7067                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7068                                               rsv, min_size);
7069                 BUG_ON(ret);    /* shouldn't happen */
7070                 trans->block_rsv = rsv;
7071         }
7072
7073         if (ret == 0 && inode->i_nlink > 0) {
7074                 trans->block_rsv = root->orphan_block_rsv;
7075                 ret = btrfs_orphan_del(trans, inode);
7076                 if (ret)
7077                         err = ret;
7078         }
7079
7080         if (trans) {
7081                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7082                 ret = btrfs_update_inode(trans, root, inode);
7083                 if (ret && !err)
7084                         err = ret;
7085
7086                 ret = btrfs_end_transaction(trans, root);
7087                 btrfs_btree_balance_dirty(root);
7088         }
7089
7090 out:
7091         btrfs_free_block_rsv(root, rsv);
7092
7093         if (ret && !err)
7094                 err = ret;
7095
7096         return err;
7097 }
7098
7099 /*
7100  * create a new subvolume directory/inode (helper for the ioctl).
7101  */
7102 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7103                              struct btrfs_root *new_root, u64 new_dirid)
7104 {
7105         struct inode *inode;
7106         int err;
7107         u64 index = 0;
7108
7109         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7110                                 new_dirid, new_dirid,
7111                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7112                                 &index);
7113         if (IS_ERR(inode))
7114                 return PTR_ERR(inode);
7115         inode->i_op = &btrfs_dir_inode_operations;
7116         inode->i_fop = &btrfs_dir_file_operations;
7117
7118         set_nlink(inode, 1);
7119         btrfs_i_size_write(inode, 0);
7120
7121         err = btrfs_update_inode(trans, new_root, inode);
7122
7123         iput(inode);
7124         return err;
7125 }
7126
7127 struct inode *btrfs_alloc_inode(struct super_block *sb)
7128 {
7129         struct btrfs_inode *ei;
7130         struct inode *inode;
7131
7132         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7133         if (!ei)
7134                 return NULL;
7135
7136         ei->root = NULL;
7137         ei->generation = 0;
7138         ei->last_trans = 0;
7139         ei->last_sub_trans = 0;
7140         ei->logged_trans = 0;
7141         ei->delalloc_bytes = 0;
7142         ei->disk_i_size = 0;
7143         ei->flags = 0;
7144         ei->csum_bytes = 0;
7145         ei->index_cnt = (u64)-1;
7146         ei->last_unlink_trans = 0;
7147         ei->last_log_commit = 0;
7148
7149         spin_lock_init(&ei->lock);
7150         ei->outstanding_extents = 0;
7151         ei->reserved_extents = 0;
7152
7153         ei->runtime_flags = 0;
7154         ei->force_compress = BTRFS_COMPRESS_NONE;
7155
7156         ei->delayed_node = NULL;
7157
7158         inode = &ei->vfs_inode;
7159         extent_map_tree_init(&ei->extent_tree);
7160         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7161         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7162         ei->io_tree.track_uptodate = 1;
7163         ei->io_failure_tree.track_uptodate = 1;
7164         atomic_set(&ei->sync_writers, 0);
7165         mutex_init(&ei->log_mutex);
7166         mutex_init(&ei->delalloc_mutex);
7167         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7168         INIT_LIST_HEAD(&ei->delalloc_inodes);
7169         INIT_LIST_HEAD(&ei->ordered_operations);
7170         RB_CLEAR_NODE(&ei->rb_node);
7171
7172         return inode;
7173 }
7174
7175 static void btrfs_i_callback(struct rcu_head *head)
7176 {
7177         struct inode *inode = container_of(head, struct inode, i_rcu);
7178         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7179 }
7180
7181 void btrfs_destroy_inode(struct inode *inode)
7182 {
7183         struct btrfs_ordered_extent *ordered;
7184         struct btrfs_root *root = BTRFS_I(inode)->root;
7185
7186         WARN_ON(!hlist_empty(&inode->i_dentry));
7187         WARN_ON(inode->i_data.nrpages);
7188         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7189         WARN_ON(BTRFS_I(inode)->reserved_extents);
7190         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7191         WARN_ON(BTRFS_I(inode)->csum_bytes);
7192
7193         /*
7194          * This can happen where we create an inode, but somebody else also
7195          * created the same inode and we need to destroy the one we already
7196          * created.
7197          */
7198         if (!root)
7199                 goto free;
7200
7201         /*
7202          * Make sure we're properly removed from the ordered operation
7203          * lists.
7204          */
7205         smp_mb();
7206         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7207                 spin_lock(&root->fs_info->ordered_extent_lock);
7208                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7209                 spin_unlock(&root->fs_info->ordered_extent_lock);
7210         }
7211
7212         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7213                      &BTRFS_I(inode)->runtime_flags)) {
7214                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7215                        (unsigned long long)btrfs_ino(inode));
7216                 atomic_dec(&root->orphan_inodes);
7217         }
7218
7219         while (1) {
7220                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7221                 if (!ordered)
7222                         break;
7223                 else {
7224                         printk(KERN_ERR "btrfs found ordered "
7225                                "extent %llu %llu on inode cleanup\n",
7226                                (unsigned long long)ordered->file_offset,
7227                                (unsigned long long)ordered->len);
7228                         btrfs_remove_ordered_extent(inode, ordered);
7229                         btrfs_put_ordered_extent(ordered);
7230                         btrfs_put_ordered_extent(ordered);
7231                 }
7232         }
7233         inode_tree_del(inode);
7234         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7235 free:
7236         btrfs_remove_delayed_node(inode);
7237         call_rcu(&inode->i_rcu, btrfs_i_callback);
7238 }
7239
7240 int btrfs_drop_inode(struct inode *inode)
7241 {
7242         struct btrfs_root *root = BTRFS_I(inode)->root;
7243
7244         if (btrfs_root_refs(&root->root_item) == 0 &&
7245             !btrfs_is_free_space_inode(inode))
7246                 return 1;
7247         else
7248                 return generic_drop_inode(inode);
7249 }
7250
7251 static void init_once(void *foo)
7252 {
7253         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7254
7255         inode_init_once(&ei->vfs_inode);
7256 }
7257
7258 void btrfs_destroy_cachep(void)
7259 {
7260         /*
7261          * Make sure all delayed rcu free inodes are flushed before we
7262          * destroy cache.
7263          */
7264         rcu_barrier();
7265         if (btrfs_inode_cachep)
7266                 kmem_cache_destroy(btrfs_inode_cachep);
7267         if (btrfs_trans_handle_cachep)
7268                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7269         if (btrfs_transaction_cachep)
7270                 kmem_cache_destroy(btrfs_transaction_cachep);
7271         if (btrfs_path_cachep)
7272                 kmem_cache_destroy(btrfs_path_cachep);
7273         if (btrfs_free_space_cachep)
7274                 kmem_cache_destroy(btrfs_free_space_cachep);
7275         if (btrfs_delalloc_work_cachep)
7276                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7277 }
7278
7279 int btrfs_init_cachep(void)
7280 {
7281         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7282                         sizeof(struct btrfs_inode), 0,
7283                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7284         if (!btrfs_inode_cachep)
7285                 goto fail;
7286
7287         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7288                         sizeof(struct btrfs_trans_handle), 0,
7289                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7290         if (!btrfs_trans_handle_cachep)
7291                 goto fail;
7292
7293         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7294                         sizeof(struct btrfs_transaction), 0,
7295                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7296         if (!btrfs_transaction_cachep)
7297                 goto fail;
7298
7299         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7300                         sizeof(struct btrfs_path), 0,
7301                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7302         if (!btrfs_path_cachep)
7303                 goto fail;
7304
7305         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7306                         sizeof(struct btrfs_free_space), 0,
7307                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7308         if (!btrfs_free_space_cachep)
7309                 goto fail;
7310
7311         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7312                         sizeof(struct btrfs_delalloc_work), 0,
7313                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7314                         NULL);
7315         if (!btrfs_delalloc_work_cachep)
7316                 goto fail;
7317
7318         return 0;
7319 fail:
7320         btrfs_destroy_cachep();
7321         return -ENOMEM;
7322 }
7323
7324 static int btrfs_getattr(struct vfsmount *mnt,
7325                          struct dentry *dentry, struct kstat *stat)
7326 {
7327         struct inode *inode = dentry->d_inode;
7328         u32 blocksize = inode->i_sb->s_blocksize;
7329
7330         generic_fillattr(inode, stat);
7331         stat->dev = BTRFS_I(inode)->root->anon_dev;
7332         stat->blksize = PAGE_CACHE_SIZE;
7333         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7334                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7335         return 0;
7336 }
7337
7338 /*
7339  * If a file is moved, it will inherit the cow and compression flags of the new
7340  * directory.
7341  */
7342 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7343 {
7344         struct btrfs_inode *b_dir = BTRFS_I(dir);
7345         struct btrfs_inode *b_inode = BTRFS_I(inode);
7346
7347         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7348                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7349         else
7350                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7351
7352         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7353                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7354                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7355         } else {
7356                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7357                                     BTRFS_INODE_NOCOMPRESS);
7358         }
7359 }
7360
7361 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7362                            struct inode *new_dir, struct dentry *new_dentry)
7363 {
7364         struct btrfs_trans_handle *trans;
7365         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7366         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7367         struct inode *new_inode = new_dentry->d_inode;
7368         struct inode *old_inode = old_dentry->d_inode;
7369         struct timespec ctime = CURRENT_TIME;
7370         u64 index = 0;
7371         u64 root_objectid;
7372         int ret;
7373         u64 old_ino = btrfs_ino(old_inode);
7374
7375         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7376                 return -EPERM;
7377
7378         /* we only allow rename subvolume link between subvolumes */
7379         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7380                 return -EXDEV;
7381
7382         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7383             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7384                 return -ENOTEMPTY;
7385
7386         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7387             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7388                 return -ENOTEMPTY;
7389
7390
7391         /* check for collisions, even if the  name isn't there */
7392         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7393                              new_dentry->d_name.name,
7394                              new_dentry->d_name.len);
7395
7396         if (ret) {
7397                 if (ret == -EEXIST) {
7398                         /* we shouldn't get
7399                          * eexist without a new_inode */
7400                         if (!new_inode) {
7401                                 WARN_ON(1);
7402                                 return ret;
7403                         }
7404                 } else {
7405                         /* maybe -EOVERFLOW */
7406                         return ret;
7407                 }
7408         }
7409         ret = 0;
7410
7411         /*
7412          * we're using rename to replace one file with another.
7413          * and the replacement file is large.  Start IO on it now so
7414          * we don't add too much work to the end of the transaction
7415          */
7416         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7417             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7418                 filemap_flush(old_inode->i_mapping);
7419
7420         /* close the racy window with snapshot create/destroy ioctl */
7421         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7422                 down_read(&root->fs_info->subvol_sem);
7423         /*
7424          * We want to reserve the absolute worst case amount of items.  So if
7425          * both inodes are subvols and we need to unlink them then that would
7426          * require 4 item modifications, but if they are both normal inodes it
7427          * would require 5 item modifications, so we'll assume their normal
7428          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7429          * should cover the worst case number of items we'll modify.
7430          */
7431         trans = btrfs_start_transaction(root, 20);
7432         if (IS_ERR(trans)) {
7433                 ret = PTR_ERR(trans);
7434                 goto out_notrans;
7435         }
7436
7437         if (dest != root)
7438                 btrfs_record_root_in_trans(trans, dest);
7439
7440         ret = btrfs_set_inode_index(new_dir, &index);
7441         if (ret)
7442                 goto out_fail;
7443
7444         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7445                 /* force full log commit if subvolume involved. */
7446                 root->fs_info->last_trans_log_full_commit = trans->transid;
7447         } else {
7448                 ret = btrfs_insert_inode_ref(trans, dest,
7449                                              new_dentry->d_name.name,
7450                                              new_dentry->d_name.len,
7451                                              old_ino,
7452                                              btrfs_ino(new_dir), index);
7453                 if (ret)
7454                         goto out_fail;
7455                 /*
7456                  * this is an ugly little race, but the rename is required
7457                  * to make sure that if we crash, the inode is either at the
7458                  * old name or the new one.  pinning the log transaction lets
7459                  * us make sure we don't allow a log commit to come in after
7460                  * we unlink the name but before we add the new name back in.
7461                  */
7462                 btrfs_pin_log_trans(root);
7463         }
7464         /*
7465          * make sure the inode gets flushed if it is replacing
7466          * something.
7467          */
7468         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7469                 btrfs_add_ordered_operation(trans, root, old_inode);
7470
7471         inode_inc_iversion(old_dir);
7472         inode_inc_iversion(new_dir);
7473         inode_inc_iversion(old_inode);
7474         old_dir->i_ctime = old_dir->i_mtime = ctime;
7475         new_dir->i_ctime = new_dir->i_mtime = ctime;
7476         old_inode->i_ctime = ctime;
7477
7478         if (old_dentry->d_parent != new_dentry->d_parent)
7479                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7480
7481         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7482                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7483                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7484                                         old_dentry->d_name.name,
7485                                         old_dentry->d_name.len);
7486         } else {
7487                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7488                                         old_dentry->d_inode,
7489                                         old_dentry->d_name.name,
7490                                         old_dentry->d_name.len);
7491                 if (!ret)
7492                         ret = btrfs_update_inode(trans, root, old_inode);
7493         }
7494         if (ret) {
7495                 btrfs_abort_transaction(trans, root, ret);
7496                 goto out_fail;
7497         }
7498
7499         if (new_inode) {
7500                 inode_inc_iversion(new_inode);
7501                 new_inode->i_ctime = CURRENT_TIME;
7502                 if (unlikely(btrfs_ino(new_inode) ==
7503                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7504                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7505                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7506                                                 root_objectid,
7507                                                 new_dentry->d_name.name,
7508                                                 new_dentry->d_name.len);
7509                         BUG_ON(new_inode->i_nlink == 0);
7510                 } else {
7511                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7512                                                  new_dentry->d_inode,
7513                                                  new_dentry->d_name.name,
7514                                                  new_dentry->d_name.len);
7515                 }
7516                 if (!ret && new_inode->i_nlink == 0) {
7517                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7518                         BUG_ON(ret);
7519                 }
7520                 if (ret) {
7521                         btrfs_abort_transaction(trans, root, ret);
7522                         goto out_fail;
7523                 }
7524         }
7525
7526         fixup_inode_flags(new_dir, old_inode);
7527
7528         ret = btrfs_add_link(trans, new_dir, old_inode,
7529                              new_dentry->d_name.name,
7530                              new_dentry->d_name.len, 0, index);
7531         if (ret) {
7532                 btrfs_abort_transaction(trans, root, ret);
7533                 goto out_fail;
7534         }
7535
7536         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7537                 struct dentry *parent = new_dentry->d_parent;
7538                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7539                 btrfs_end_log_trans(root);
7540         }
7541 out_fail:
7542         btrfs_end_transaction(trans, root);
7543 out_notrans:
7544         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7545                 up_read(&root->fs_info->subvol_sem);
7546
7547         return ret;
7548 }
7549
7550 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7551 {
7552         struct btrfs_delalloc_work *delalloc_work;
7553
7554         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7555                                      work);
7556         if (delalloc_work->wait)
7557                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7558         else
7559                 filemap_flush(delalloc_work->inode->i_mapping);
7560
7561         if (delalloc_work->delay_iput)
7562                 btrfs_add_delayed_iput(delalloc_work->inode);
7563         else
7564                 iput(delalloc_work->inode);
7565         complete(&delalloc_work->completion);
7566 }
7567
7568 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7569                                                     int wait, int delay_iput)
7570 {
7571         struct btrfs_delalloc_work *work;
7572
7573         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7574         if (!work)
7575                 return NULL;
7576
7577         init_completion(&work->completion);
7578         INIT_LIST_HEAD(&work->list);
7579         work->inode = inode;
7580         work->wait = wait;
7581         work->delay_iput = delay_iput;
7582         work->work.func = btrfs_run_delalloc_work;
7583
7584         return work;
7585 }
7586
7587 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7588 {
7589         wait_for_completion(&work->completion);
7590         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7591 }
7592
7593 /*
7594  * some fairly slow code that needs optimization. This walks the list
7595  * of all the inodes with pending delalloc and forces them to disk.
7596  */
7597 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7598 {
7599         struct btrfs_inode *binode;
7600         struct inode *inode;
7601         struct btrfs_delalloc_work *work, *next;
7602         struct list_head works;
7603         struct list_head splice;
7604         int ret = 0;
7605
7606         if (root->fs_info->sb->s_flags & MS_RDONLY)
7607                 return -EROFS;
7608
7609         INIT_LIST_HEAD(&works);
7610         INIT_LIST_HEAD(&splice);
7611
7612         spin_lock(&root->fs_info->delalloc_lock);
7613         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7614         while (!list_empty(&splice)) {
7615                 binode = list_entry(splice.next, struct btrfs_inode,
7616                                     delalloc_inodes);
7617
7618                 list_del_init(&binode->delalloc_inodes);
7619
7620                 inode = igrab(&binode->vfs_inode);
7621                 if (!inode)
7622                         continue;
7623
7624                 list_add_tail(&binode->delalloc_inodes,
7625                               &root->fs_info->delalloc_inodes);
7626                 spin_unlock(&root->fs_info->delalloc_lock);
7627
7628                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7629                 if (unlikely(!work)) {
7630                         ret = -ENOMEM;
7631                         goto out;
7632                 }
7633                 list_add_tail(&work->list, &works);
7634                 btrfs_queue_worker(&root->fs_info->flush_workers,
7635                                    &work->work);
7636
7637                 cond_resched();
7638                 spin_lock(&root->fs_info->delalloc_lock);
7639         }
7640         spin_unlock(&root->fs_info->delalloc_lock);
7641
7642         list_for_each_entry_safe(work, next, &works, list) {
7643                 list_del_init(&work->list);
7644                 btrfs_wait_and_free_delalloc_work(work);
7645         }
7646
7647         /* the filemap_flush will queue IO into the worker threads, but
7648          * we have to make sure the IO is actually started and that
7649          * ordered extents get created before we return
7650          */
7651         atomic_inc(&root->fs_info->async_submit_draining);
7652         while (atomic_read(&root->fs_info->nr_async_submits) ||
7653               atomic_read(&root->fs_info->async_delalloc_pages)) {
7654                 wait_event(root->fs_info->async_submit_wait,
7655                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7656                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7657         }
7658         atomic_dec(&root->fs_info->async_submit_draining);
7659         return 0;
7660 out:
7661         list_for_each_entry_safe(work, next, &works, list) {
7662                 list_del_init(&work->list);
7663                 btrfs_wait_and_free_delalloc_work(work);
7664         }
7665
7666         if (!list_empty_careful(&splice)) {
7667                 spin_lock(&root->fs_info->delalloc_lock);
7668                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7669                 spin_unlock(&root->fs_info->delalloc_lock);
7670         }
7671         return ret;
7672 }
7673
7674 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7675                          const char *symname)
7676 {
7677         struct btrfs_trans_handle *trans;
7678         struct btrfs_root *root = BTRFS_I(dir)->root;
7679         struct btrfs_path *path;
7680         struct btrfs_key key;
7681         struct inode *inode = NULL;
7682         int err;
7683         int drop_inode = 0;
7684         u64 objectid;
7685         u64 index = 0 ;
7686         int name_len;
7687         int datasize;
7688         unsigned long ptr;
7689         struct btrfs_file_extent_item *ei;
7690         struct extent_buffer *leaf;
7691
7692         name_len = strlen(symname) + 1;
7693         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7694                 return -ENAMETOOLONG;
7695
7696         /*
7697          * 2 items for inode item and ref
7698          * 2 items for dir items
7699          * 1 item for xattr if selinux is on
7700          */
7701         trans = btrfs_start_transaction(root, 5);
7702         if (IS_ERR(trans))
7703                 return PTR_ERR(trans);
7704
7705         err = btrfs_find_free_ino(root, &objectid);
7706         if (err)
7707                 goto out_unlock;
7708
7709         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7710                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7711                                 S_IFLNK|S_IRWXUGO, &index);
7712         if (IS_ERR(inode)) {
7713                 err = PTR_ERR(inode);
7714                 goto out_unlock;
7715         }
7716
7717         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7718         if (err) {
7719                 drop_inode = 1;
7720                 goto out_unlock;
7721         }
7722
7723         /*
7724         * If the active LSM wants to access the inode during
7725         * d_instantiate it needs these. Smack checks to see
7726         * if the filesystem supports xattrs by looking at the
7727         * ops vector.
7728         */
7729         inode->i_fop = &btrfs_file_operations;
7730         inode->i_op = &btrfs_file_inode_operations;
7731
7732         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7733         if (err)
7734                 drop_inode = 1;
7735         else {
7736                 inode->i_mapping->a_ops = &btrfs_aops;
7737                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7738                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7739         }
7740         if (drop_inode)
7741                 goto out_unlock;
7742
7743         path = btrfs_alloc_path();
7744         if (!path) {
7745                 err = -ENOMEM;
7746                 drop_inode = 1;
7747                 goto out_unlock;
7748         }
7749         key.objectid = btrfs_ino(inode);
7750         key.offset = 0;
7751         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7752         datasize = btrfs_file_extent_calc_inline_size(name_len);
7753         err = btrfs_insert_empty_item(trans, root, path, &key,
7754                                       datasize);
7755         if (err) {
7756                 drop_inode = 1;
7757                 btrfs_free_path(path);
7758                 goto out_unlock;
7759         }
7760         leaf = path->nodes[0];
7761         ei = btrfs_item_ptr(leaf, path->slots[0],
7762                             struct btrfs_file_extent_item);
7763         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7764         btrfs_set_file_extent_type(leaf, ei,
7765                                    BTRFS_FILE_EXTENT_INLINE);
7766         btrfs_set_file_extent_encryption(leaf, ei, 0);
7767         btrfs_set_file_extent_compression(leaf, ei, 0);
7768         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7769         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7770
7771         ptr = btrfs_file_extent_inline_start(ei);
7772         write_extent_buffer(leaf, symname, ptr, name_len);
7773         btrfs_mark_buffer_dirty(leaf);
7774         btrfs_free_path(path);
7775
7776         inode->i_op = &btrfs_symlink_inode_operations;
7777         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7778         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7779         inode_set_bytes(inode, name_len);
7780         btrfs_i_size_write(inode, name_len - 1);
7781         err = btrfs_update_inode(trans, root, inode);
7782         if (err)
7783                 drop_inode = 1;
7784
7785 out_unlock:
7786         if (!err)
7787                 d_instantiate(dentry, inode);
7788         btrfs_end_transaction(trans, root);
7789         if (drop_inode) {
7790                 inode_dec_link_count(inode);
7791                 iput(inode);
7792         }
7793         btrfs_btree_balance_dirty(root);
7794         return err;
7795 }
7796
7797 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7798                                        u64 start, u64 num_bytes, u64 min_size,
7799                                        loff_t actual_len, u64 *alloc_hint,
7800                                        struct btrfs_trans_handle *trans)
7801 {
7802         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7803         struct extent_map *em;
7804         struct btrfs_root *root = BTRFS_I(inode)->root;
7805         struct btrfs_key ins;
7806         u64 cur_offset = start;
7807         u64 i_size;
7808         int ret = 0;
7809         bool own_trans = true;
7810
7811         if (trans)
7812                 own_trans = false;
7813         while (num_bytes > 0) {
7814                 if (own_trans) {
7815                         trans = btrfs_start_transaction(root, 3);
7816                         if (IS_ERR(trans)) {
7817                                 ret = PTR_ERR(trans);
7818                                 break;
7819                         }
7820                 }
7821
7822                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7823                                            0, *alloc_hint, &ins, 1);
7824                 if (ret) {
7825                         if (own_trans)
7826                                 btrfs_end_transaction(trans, root);
7827                         break;
7828                 }
7829
7830                 ret = insert_reserved_file_extent(trans, inode,
7831                                                   cur_offset, ins.objectid,
7832                                                   ins.offset, ins.offset,
7833                                                   ins.offset, 0, 0, 0,
7834                                                   BTRFS_FILE_EXTENT_PREALLOC);
7835                 if (ret) {
7836                         btrfs_abort_transaction(trans, root, ret);
7837                         if (own_trans)
7838                                 btrfs_end_transaction(trans, root);
7839                         break;
7840                 }
7841                 btrfs_drop_extent_cache(inode, cur_offset,
7842                                         cur_offset + ins.offset -1, 0);
7843
7844                 em = alloc_extent_map();
7845                 if (!em) {
7846                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7847                                 &BTRFS_I(inode)->runtime_flags);
7848                         goto next;
7849                 }
7850
7851                 em->start = cur_offset;
7852                 em->orig_start = cur_offset;
7853                 em->len = ins.offset;
7854                 em->block_start = ins.objectid;
7855                 em->block_len = ins.offset;
7856                 em->orig_block_len = ins.offset;
7857                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7858                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7859                 em->generation = trans->transid;
7860
7861                 while (1) {
7862                         write_lock(&em_tree->lock);
7863                         ret = add_extent_mapping(em_tree, em);
7864                         if (!ret)
7865                                 list_move(&em->list,
7866                                           &em_tree->modified_extents);
7867                         write_unlock(&em_tree->lock);
7868                         if (ret != -EEXIST)
7869                                 break;
7870                         btrfs_drop_extent_cache(inode, cur_offset,
7871                                                 cur_offset + ins.offset - 1,
7872                                                 0);
7873                 }
7874                 free_extent_map(em);
7875 next:
7876                 num_bytes -= ins.offset;
7877                 cur_offset += ins.offset;
7878                 *alloc_hint = ins.objectid + ins.offset;
7879
7880                 inode_inc_iversion(inode);
7881                 inode->i_ctime = CURRENT_TIME;
7882                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7883                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7884                     (actual_len > inode->i_size) &&
7885                     (cur_offset > inode->i_size)) {
7886                         if (cur_offset > actual_len)
7887                                 i_size = actual_len;
7888                         else
7889                                 i_size = cur_offset;
7890                         i_size_write(inode, i_size);
7891                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7892                 }
7893
7894                 ret = btrfs_update_inode(trans, root, inode);
7895
7896                 if (ret) {
7897                         btrfs_abort_transaction(trans, root, ret);
7898                         if (own_trans)
7899                                 btrfs_end_transaction(trans, root);
7900                         break;
7901                 }
7902
7903                 if (own_trans)
7904                         btrfs_end_transaction(trans, root);
7905         }
7906         return ret;
7907 }
7908
7909 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7910                               u64 start, u64 num_bytes, u64 min_size,
7911                               loff_t actual_len, u64 *alloc_hint)
7912 {
7913         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7914                                            min_size, actual_len, alloc_hint,
7915                                            NULL);
7916 }
7917
7918 int btrfs_prealloc_file_range_trans(struct inode *inode,
7919                                     struct btrfs_trans_handle *trans, int mode,
7920                                     u64 start, u64 num_bytes, u64 min_size,
7921                                     loff_t actual_len, u64 *alloc_hint)
7922 {
7923         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7924                                            min_size, actual_len, alloc_hint, trans);
7925 }
7926
7927 static int btrfs_set_page_dirty(struct page *page)
7928 {
7929         return __set_page_dirty_nobuffers(page);
7930 }
7931
7932 static int btrfs_permission(struct inode *inode, int mask)
7933 {
7934         struct btrfs_root *root = BTRFS_I(inode)->root;
7935         umode_t mode = inode->i_mode;
7936
7937         if (mask & MAY_WRITE &&
7938             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7939                 if (btrfs_root_readonly(root))
7940                         return -EROFS;
7941                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7942                         return -EACCES;
7943         }
7944         return generic_permission(inode, mask);
7945 }
7946
7947 static const struct inode_operations btrfs_dir_inode_operations = {
7948         .getattr        = btrfs_getattr,
7949         .lookup         = btrfs_lookup,
7950         .create         = btrfs_create,
7951         .unlink         = btrfs_unlink,
7952         .link           = btrfs_link,
7953         .mkdir          = btrfs_mkdir,
7954         .rmdir          = btrfs_rmdir,
7955         .rename         = btrfs_rename,
7956         .symlink        = btrfs_symlink,
7957         .setattr        = btrfs_setattr,
7958         .mknod          = btrfs_mknod,
7959         .setxattr       = btrfs_setxattr,
7960         .getxattr       = btrfs_getxattr,
7961         .listxattr      = btrfs_listxattr,
7962         .removexattr    = btrfs_removexattr,
7963         .permission     = btrfs_permission,
7964         .get_acl        = btrfs_get_acl,
7965 };
7966 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7967         .lookup         = btrfs_lookup,
7968         .permission     = btrfs_permission,
7969         .get_acl        = btrfs_get_acl,
7970 };
7971
7972 static const struct file_operations btrfs_dir_file_operations = {
7973         .llseek         = generic_file_llseek,
7974         .read           = generic_read_dir,
7975         .readdir        = btrfs_real_readdir,
7976         .unlocked_ioctl = btrfs_ioctl,
7977 #ifdef CONFIG_COMPAT
7978         .compat_ioctl   = btrfs_ioctl,
7979 #endif
7980         .release        = btrfs_release_file,
7981         .fsync          = btrfs_sync_file,
7982 };
7983
7984 static struct extent_io_ops btrfs_extent_io_ops = {
7985         .fill_delalloc = run_delalloc_range,
7986         .submit_bio_hook = btrfs_submit_bio_hook,
7987         .merge_bio_hook = btrfs_merge_bio_hook,
7988         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7989         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7990         .writepage_start_hook = btrfs_writepage_start_hook,
7991         .set_bit_hook = btrfs_set_bit_hook,
7992         .clear_bit_hook = btrfs_clear_bit_hook,
7993         .merge_extent_hook = btrfs_merge_extent_hook,
7994         .split_extent_hook = btrfs_split_extent_hook,
7995 };
7996
7997 /*
7998  * btrfs doesn't support the bmap operation because swapfiles
7999  * use bmap to make a mapping of extents in the file.  They assume
8000  * these extents won't change over the life of the file and they
8001  * use the bmap result to do IO directly to the drive.
8002  *
8003  * the btrfs bmap call would return logical addresses that aren't
8004  * suitable for IO and they also will change frequently as COW
8005  * operations happen.  So, swapfile + btrfs == corruption.
8006  *
8007  * For now we're avoiding this by dropping bmap.
8008  */
8009 static const struct address_space_operations btrfs_aops = {
8010         .readpage       = btrfs_readpage,
8011         .writepage      = btrfs_writepage,
8012         .writepages     = btrfs_writepages,
8013         .readpages      = btrfs_readpages,
8014         .direct_IO      = btrfs_direct_IO,
8015         .invalidatepage = btrfs_invalidatepage,
8016         .releasepage    = btrfs_releasepage,
8017         .set_page_dirty = btrfs_set_page_dirty,
8018         .error_remove_page = generic_error_remove_page,
8019 };
8020
8021 static const struct address_space_operations btrfs_symlink_aops = {
8022         .readpage       = btrfs_readpage,
8023         .writepage      = btrfs_writepage,
8024         .invalidatepage = btrfs_invalidatepage,
8025         .releasepage    = btrfs_releasepage,
8026 };
8027
8028 static const struct inode_operations btrfs_file_inode_operations = {
8029         .getattr        = btrfs_getattr,
8030         .setattr        = btrfs_setattr,
8031         .setxattr       = btrfs_setxattr,
8032         .getxattr       = btrfs_getxattr,
8033         .listxattr      = btrfs_listxattr,
8034         .removexattr    = btrfs_removexattr,
8035         .permission     = btrfs_permission,
8036         .fiemap         = btrfs_fiemap,
8037         .get_acl        = btrfs_get_acl,
8038         .update_time    = btrfs_update_time,
8039 };
8040 static const struct inode_operations btrfs_special_inode_operations = {
8041         .getattr        = btrfs_getattr,
8042         .setattr        = btrfs_setattr,
8043         .permission     = btrfs_permission,
8044         .setxattr       = btrfs_setxattr,
8045         .getxattr       = btrfs_getxattr,
8046         .listxattr      = btrfs_listxattr,
8047         .removexattr    = btrfs_removexattr,
8048         .get_acl        = btrfs_get_acl,
8049         .update_time    = btrfs_update_time,
8050 };
8051 static const struct inode_operations btrfs_symlink_inode_operations = {
8052         .readlink       = generic_readlink,
8053         .follow_link    = page_follow_link_light,
8054         .put_link       = page_put_link,
8055         .getattr        = btrfs_getattr,
8056         .setattr        = btrfs_setattr,
8057         .permission     = btrfs_permission,
8058         .setxattr       = btrfs_setxattr,
8059         .getxattr       = btrfs_getxattr,
8060         .listxattr      = btrfs_listxattr,
8061         .removexattr    = btrfs_removexattr,
8062         .get_acl        = btrfs_get_acl,
8063         .update_time    = btrfs_update_time,
8064 };
8065
8066 const struct dentry_operations btrfs_dentry_operations = {
8067         .d_delete       = btrfs_dentry_delete,
8068         .d_release      = btrfs_dentry_release,
8069 };