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 "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 /*
 754  * when extent_io.c finds a delayed allocation range in the file,
 755  * the call backs end up in this code.  The basic idea is to
 756  * allocate extents on disk for the range, and create ordered data structs
 757  * in ram to track those extents.
 758  *
 759  * locked_page is the page that writepage had locked already.  We use
 760  * it to make sure we don't do extra locks or unlocks.
 761  *
 762  * *page_started is set to one if we unlock locked_page and do everything
 763  * required to start IO on it.  It may be clean and already done with
 764  * IO when we return.
 765  */
 766 static noinline int cow_file_range(struct inode *inode,
 767                                    struct page *locked_page,
 768                                    u64 start, u64 end, int *page_started,
 769                                    unsigned long *nr_written,
 770                                    int unlock)
 771 {
 772         struct btrfs_root *root = BTRFS_I(inode)->root;
 773         struct btrfs_trans_handle *trans;
 774         u64 alloc_hint = 0;
 775         u64 num_bytes;
 776         unsigned long ram_size;
 777         u64 disk_num_bytes;
 778         u64 cur_alloc_size;
 779         u64 blocksize = root->sectorsize;
 780         struct btrfs_key ins;
 781         struct extent_map *em;
 782         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 783         int ret = 0;
 784
 785         BUG_ON(btrfs_is_free_space_inode(root, inode));
 786         trans = btrfs_join_transaction(root);
 787         BUG_ON(IS_ERR(trans));
 788         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 789
 790         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 791         num_bytes = max(blocksize,  num_bytes);
 792         disk_num_bytes = num_bytes;
 793         ret = 0;
 794
 795         /* if this is a small write inside eof, kick off defrag */
 796         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 797                 btrfs_add_inode_defrag(trans, inode);
 798
 799         if (start == 0) {
 800                 /* lets try to make an inline extent */
 801                 ret = cow_file_range_inline(trans, root, inode,
 802                                             start, end, 0, 0, NULL);
 803                 if (ret == 0) {
 804                         extent_clear_unlock_delalloc(inode,
 805                                      &BTRFS_I(inode)->io_tree,
 806                                      start, end, NULL,
 807                                      EXTENT_CLEAR_UNLOCK_PAGE |
 808                                      EXTENT_CLEAR_UNLOCK |
 809                                      EXTENT_CLEAR_DELALLOC |
 810                                      EXTENT_CLEAR_DIRTY |
 811                                      EXTENT_SET_WRITEBACK |
 812                                      EXTENT_END_WRITEBACK);
 813
 814                         *nr_written = *nr_written +
 815                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 816                         *page_started = 1;
 817                         ret = 0;
 818                         goto out;
 819                 }
 820         }
 821
 822         BUG_ON(disk_num_bytes >
 823                btrfs_super_total_bytes(&root->fs_info->super_copy));
 824
 825         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 826         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 827
 828         while (disk_num_bytes > 0) {
 829                 unsigned long op;
 830
 831                 cur_alloc_size = disk_num_bytes;
 832                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 833                                            root->sectorsize, 0, alloc_hint,
 834                                            (u64)-1, &ins, 1);
 835                 BUG_ON(ret);
 836
 837                 em = alloc_extent_map();
 838                 BUG_ON(!em);
 839                 em->start = start;
 840                 em->orig_start = em->start;
 841                 ram_size = ins.offset;
 842                 em->len = ins.offset;
 843
 844                 em->block_start = ins.objectid;
 845                 em->block_len = ins.offset;
 846                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 847                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 848
 849                 while (1) {
 850                         write_lock(&em_tree->lock);
 851                         ret = add_extent_mapping(em_tree, em);
 852                         write_unlock(&em_tree->lock);
 853                         if (ret != -EEXIST) {
 854                                 free_extent_map(em);
 855                                 break;
 856                         }
 857                         btrfs_drop_extent_cache(inode, start,
 858                                                 start + ram_size - 1, 0);
 859                 }
 860
 861                 cur_alloc_size = ins.offset;
 862                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 863                                                ram_size, cur_alloc_size, 0);
 864                 BUG_ON(ret);
 865
 866                 if (root->root_key.objectid ==
 867                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 868                         ret = btrfs_reloc_clone_csums(inode, start,
 869                                                       cur_alloc_size);
 870                         BUG_ON(ret);
 871                 }
 872
 873                 if (disk_num_bytes < cur_alloc_size)
 874                         break;
 875
 876                 /* we're not doing compressed IO, don't unlock the first
 877                  * page (which the caller expects to stay locked), don't
 878                  * clear any dirty bits and don't set any writeback bits
 879                  *
 880                  * Do set the Private2 bit so we know this page was properly
 881                  * setup for writepage
 882                  */
 883                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 884                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 885                         EXTENT_SET_PRIVATE2;
 886
 887                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 888                                              start, start + ram_size - 1,
 889                                              locked_page, op);
 890                 disk_num_bytes -= cur_alloc_size;
 891                 num_bytes -= cur_alloc_size;
 892                 alloc_hint = ins.objectid + ins.offset;
 893                 start += cur_alloc_size;
 894         }
 895 out:
 896         ret = 0;
 897         btrfs_end_transaction(trans, root);
 898
 899         return ret;
 900 }
 901
 902 /*
 903  * work queue call back to started compression on a file and pages
 904  */
 905 static noinline void async_cow_start(struct btrfs_work *work)
 906 {
 907         struct async_cow *async_cow;
 908         int num_added = 0;
 909         async_cow = container_of(work, struct async_cow, work);
 910
 911         compress_file_range(async_cow->inode, async_cow->locked_page,
 912                             async_cow->start, async_cow->end, async_cow,
 913                             &num_added);
 914         if (num_added == 0)
 915                 async_cow->inode = NULL;
 916 }
 917
 918 /*
 919  * work queue call back to submit previously compressed pages
 920  */
 921 static noinline void async_cow_submit(struct btrfs_work *work)
 922 {
 923         struct async_cow *async_cow;
 924         struct btrfs_root *root;
 925         unsigned long nr_pages;
 926
 927         async_cow = container_of(work, struct async_cow, work);
 928
 929         root = async_cow->root;
 930         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 931                 PAGE_CACHE_SHIFT;
 932
 933         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 934
 935         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 936             5 * 1042 * 1024 &&
 937             waitqueue_active(&root->fs_info->async_submit_wait))
 938                 wake_up(&root->fs_info->async_submit_wait);
 939
 940         if (async_cow->inode)
 941                 submit_compressed_extents(async_cow->inode, async_cow);
 942 }
 943
 944 static noinline void async_cow_free(struct btrfs_work *work)
 945 {
 946         struct async_cow *async_cow;
 947         async_cow = container_of(work, struct async_cow, work);
 948         kfree(async_cow);
 949 }
 950
 951 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 952                                 u64 start, u64 end, int *page_started,
 953                                 unsigned long *nr_written)
 954 {
 955         struct async_cow *async_cow;
 956         struct btrfs_root *root = BTRFS_I(inode)->root;
 957         unsigned long nr_pages;
 958         u64 cur_end;
 959         int limit = 10 * 1024 * 1042;
 960
 961         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 962                          1, 0, NULL, GFP_NOFS);
 963         while (start < end) {
 964                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 965                 BUG_ON(!async_cow);
 966                 async_cow->inode = inode;
 967                 async_cow->root = root;
 968                 async_cow->locked_page = locked_page;
 969                 async_cow->start = start;
 970
 971                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 972                         cur_end = end;
 973                 else
 974                         cur_end = min(end, start + 512 * 1024 - 1);
 975
 976                 async_cow->end = cur_end;
 977                 INIT_LIST_HEAD(&async_cow->extents);
 978
 979                 async_cow->work.func = async_cow_start;
 980                 async_cow->work.ordered_func = async_cow_submit;
 981                 async_cow->work.ordered_free = async_cow_free;
 982                 async_cow->work.flags = 0;
 983
 984                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 985                         PAGE_CACHE_SHIFT;
 986                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 987
 988                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 989                                    &async_cow->work);
 990
 991                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 992                         wait_event(root->fs_info->async_submit_wait,
 993                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 994                             limit));
 995                 }
 996
 997                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 998                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1001                            0));
1002                 }
1003
1004                 *nr_written += nr_pages;
1005                 start = cur_end + 1;
1006         }
1007         *page_started = 1;
1008         return 0;
1009 }
1010
1011 static noinline int csum_exist_in_range(struct btrfs_root *root,
1012                                         u64 bytenr, u64 num_bytes)
1013 {
1014         int ret;
1015         struct btrfs_ordered_sum *sums;
1016         LIST_HEAD(list);
1017
1018         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1019                                        bytenr + num_bytes - 1, &list, 0);
1020         if (ret == 0 && list_empty(&list))
1021                 return 0;
1022
1023         while (!list_empty(&list)) {
1024                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025                 list_del(&sums->list);
1026                 kfree(sums);
1027         }
1028         return 1;
1029 }
1030
1031 /*
1032  * when nowcow writeback call back.  This checks for snapshots or COW copies
1033  * of the extents that exist in the file, and COWs the file as required.
1034  *
1035  * If no cow copies or snapshots exist, we write directly to the existing
1036  * blocks on disk
1037  */
1038 static noinline int run_delalloc_nocow(struct inode *inode,
1039                                        struct page *locked_page,
1040                               u64 start, u64 end, int *page_started, int force,
1041                               unsigned long *nr_written)
1042 {
1043         struct btrfs_root *root = BTRFS_I(inode)->root;
1044         struct btrfs_trans_handle *trans;
1045         struct extent_buffer *leaf;
1046         struct btrfs_path *path;
1047         struct btrfs_file_extent_item *fi;
1048         struct btrfs_key found_key;
1049         u64 cow_start;
1050         u64 cur_offset;
1051         u64 extent_end;
1052         u64 extent_offset;
1053         u64 disk_bytenr;
1054         u64 num_bytes;
1055         int extent_type;
1056         int ret;
1057         int type;
1058         int nocow;
1059         int check_prev = 1;
1060         bool nolock;
1061         u64 ino = btrfs_ino(inode);
1062
1063         path = btrfs_alloc_path();
1064         if (!path)
1065                 return -ENOMEM;
1066
1067         nolock = btrfs_is_free_space_inode(root, inode);
1068
1069         if (nolock)
1070                 trans = btrfs_join_transaction_nolock(root);
1071         else
1072                 trans = btrfs_join_transaction(root);
1073
1074         BUG_ON(IS_ERR(trans));
1075         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1076
1077         cow_start = (u64)-1;
1078         cur_offset = start;
1079         while (1) {
1080                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1081                                                cur_offset, 0);
1082                 BUG_ON(ret < 0);
1083                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084                         leaf = path->nodes[0];
1085                         btrfs_item_key_to_cpu(leaf, &found_key,
1086                                               path->slots[0] - 1);
1087                         if (found_key.objectid == ino &&
1088                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1089                                 path->slots[0]--;
1090                 }
1091                 check_prev = 0;
1092 next_slot:
1093                 leaf = path->nodes[0];
1094                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095                         ret = btrfs_next_leaf(root, path);
1096                         if (ret < 0)
1097                                 BUG_ON(1);
1098                         if (ret > 0)
1099                                 break;
1100                         leaf = path->nodes[0];
1101                 }
1102
1103                 nocow = 0;
1104                 disk_bytenr = 0;
1105                 num_bytes = 0;
1106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
1108                 if (found_key.objectid > ino ||
1109                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110                     found_key.offset > end)
1111                         break;
1112
1113                 if (found_key.offset > cur_offset) {
1114                         extent_end = found_key.offset;
1115                         extent_type = 0;
1116                         goto out_check;
1117                 }
1118
1119                 fi = btrfs_item_ptr(leaf, path->slots[0],
1120                                     struct btrfs_file_extent_item);
1121                 extent_type = btrfs_file_extent_type(leaf, fi);
1122
1123                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1127                         extent_end = found_key.offset +
1128                                 btrfs_file_extent_num_bytes(leaf, fi);
1129                         if (extent_end <= start) {
1130                                 path->slots[0]++;
1131                                 goto next_slot;
1132                         }
1133                         if (disk_bytenr == 0)
1134                                 goto out_check;
1135                         if (btrfs_file_extent_compression(leaf, fi) ||
1136                             btrfs_file_extent_encryption(leaf, fi) ||
1137                             btrfs_file_extent_other_encoding(leaf, fi))
1138                                 goto out_check;
1139                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140                                 goto out_check;
1141                         if (btrfs_extent_readonly(root, disk_bytenr))
1142                                 goto out_check;
1143                         if (btrfs_cross_ref_exist(trans, root, ino,
1144                                                   found_key.offset -
1145                                                   extent_offset, disk_bytenr))
1146                                 goto out_check;
1147                         disk_bytenr += extent_offset;
1148                         disk_bytenr += cur_offset - found_key.offset;
1149                         num_bytes = min(end + 1, extent_end) - cur_offset;
1150                         /*
1151                          * force cow if csum exists in the range.
1152                          * this ensure that csum for a given extent are
1153                          * either valid or do not exist.
1154                          */
1155                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156                                 goto out_check;
1157                         nocow = 1;
1158                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159                         extent_end = found_key.offset +
1160                                 btrfs_file_extent_inline_len(leaf, fi);
1161                         extent_end = ALIGN(extent_end, root->sectorsize);
1162                 } else {
1163                         BUG_ON(1);
1164                 }
1165 out_check:
1166                 if (extent_end <= start) {
1167                         path->slots[0]++;
1168                         goto next_slot;
1169                 }
1170                 if (!nocow) {
1171                         if (cow_start == (u64)-1)
1172                                 cow_start = cur_offset;
1173                         cur_offset = extent_end;
1174                         if (cur_offset > end)
1175                                 break;
1176                         path->slots[0]++;
1177                         goto next_slot;
1178                 }
1179
1180                 btrfs_release_path(path);
1181                 if (cow_start != (u64)-1) {
1182                         ret = cow_file_range(inode, locked_page, cow_start,
1183                                         found_key.offset - 1, page_started,
1184                                         nr_written, 1);
1185                         BUG_ON(ret);
1186                         cow_start = (u64)-1;
1187                 }
1188
1189                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190                         struct extent_map *em;
1191                         struct extent_map_tree *em_tree;
1192                         em_tree = &BTRFS_I(inode)->extent_tree;
1193                         em = alloc_extent_map();
1194                         BUG_ON(!em);
1195                         em->start = cur_offset;
1196                         em->orig_start = em->start;
1197                         em->len = num_bytes;
1198                         em->block_len = num_bytes;
1199                         em->block_start = disk_bytenr;
1200                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1201                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202                         while (1) {
1203                                 write_lock(&em_tree->lock);
1204                                 ret = add_extent_mapping(em_tree, em);
1205                                 write_unlock(&em_tree->lock);
1206                                 if (ret != -EEXIST) {
1207                                         free_extent_map(em);
1208                                         break;
1209                                 }
1210                                 btrfs_drop_extent_cache(inode, em->start,
1211                                                 em->start + em->len - 1, 0);
1212                         }
1213                         type = BTRFS_ORDERED_PREALLOC;
1214                 } else {
1215                         type = BTRFS_ORDERED_NOCOW;
1216                 }
1217
1218                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219                                                num_bytes, num_bytes, type);
1220                 BUG_ON(ret);
1221
1222                 if (root->root_key.objectid ==
1223                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225                                                       num_bytes);
1226                         BUG_ON(ret);
1227                 }
1228
1229                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230                                 cur_offset, cur_offset + num_bytes - 1,
1231                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233                                 EXTENT_SET_PRIVATE2);
1234                 cur_offset = extent_end;
1235                 if (cur_offset > end)
1236                         break;
1237         }
1238         btrfs_release_path(path);
1239
1240         if (cur_offset <= end && cow_start == (u64)-1)
1241                 cow_start = cur_offset;
1242         if (cow_start != (u64)-1) {
1243                 ret = cow_file_range(inode, locked_page, cow_start, end,
1244                                      page_started, nr_written, 1);
1245                 BUG_ON(ret);
1246         }
1247
1248         if (nolock) {
1249                 ret = btrfs_end_transaction_nolock(trans, root);
1250                 BUG_ON(ret);
1251         } else {
1252                 ret = btrfs_end_transaction(trans, root);
1253                 BUG_ON(ret);
1254         }
1255         btrfs_free_path(path);
1256         return 0;
1257 }
1258
1259 /*
1260  * extent_io.c call back to do delayed allocation processing
1261  */
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263                               u64 start, u64 end, int *page_started,
1264                               unsigned long *nr_written)
1265 {
1266         int ret;
1267         struct btrfs_root *root = BTRFS_I(inode)->root;
1268
1269         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271                                          page_started, 1, nr_written);
1272         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274                                          page_started, 0, nr_written);
1275         else if (!btrfs_test_opt(root, COMPRESS) &&
1276                  !(BTRFS_I(inode)->force_compress) &&
1277                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278                 ret = cow_file_range(inode, locked_page, start, end,
1279                                       page_started, nr_written, 1);
1280         else
1281                 ret = cow_file_range_async(inode, locked_page, start, end,
1282                                            page_started, nr_written);
1283         return ret;
1284 }
1285
1286 static void btrfs_split_extent_hook(struct inode *inode,
1287                                     struct extent_state *orig, u64 split)
1288 {
1289         /* not delalloc, ignore it */
1290         if (!(orig->state & EXTENT_DELALLOC))
1291                 return;
1292
1293         spin_lock(&BTRFS_I(inode)->lock);
1294         BTRFS_I(inode)->outstanding_extents++;
1295         spin_unlock(&BTRFS_I(inode)->lock);
1296 }
1297
1298 /*
1299  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300  * extents so we can keep track of new extents that are just merged onto old
1301  * extents, such as when we are doing sequential writes, so we can properly
1302  * account for the metadata space we'll need.
1303  */
1304 static void btrfs_merge_extent_hook(struct inode *inode,
1305                                     struct extent_state *new,
1306                                     struct extent_state *other)
1307 {
1308         /* not delalloc, ignore it */
1309         if (!(other->state & EXTENT_DELALLOC))
1310                 return;
1311
1312         spin_lock(&BTRFS_I(inode)->lock);
1313         BTRFS_I(inode)->outstanding_extents--;
1314         spin_unlock(&BTRFS_I(inode)->lock);
1315 }
1316
1317 /*
1318  * extent_io.c set_bit_hook, used to track delayed allocation
1319  * bytes in this file, and to maintain the list of inodes that
1320  * have pending delalloc work to be done.
1321  */
1322 static void btrfs_set_bit_hook(struct inode *inode,
1323                                struct extent_state *state, int *bits)
1324 {
1325
1326         /*
1327          * set_bit and clear bit hooks normally require _irqsave/restore
1328          * but in this case, we are only testing for the DELALLOC
1329          * bit, which is only set or cleared with irqs on
1330          */
1331         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1332                 struct btrfs_root *root = BTRFS_I(inode)->root;
1333                 u64 len = state->end + 1 - state->start;
1334                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1335
1336                 if (*bits & EXTENT_FIRST_DELALLOC) {
1337                         *bits &= ~EXTENT_FIRST_DELALLOC;
1338                 } else {
1339                         spin_lock(&BTRFS_I(inode)->lock);
1340                         BTRFS_I(inode)->outstanding_extents++;
1341                         spin_unlock(&BTRFS_I(inode)->lock);
1342                 }
1343
1344                 spin_lock(&root->fs_info->delalloc_lock);
1345                 BTRFS_I(inode)->delalloc_bytes += len;
1346                 root->fs_info->delalloc_bytes += len;
1347                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1348                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349                                       &root->fs_info->delalloc_inodes);
1350                 }
1351                 spin_unlock(&root->fs_info->delalloc_lock);
1352         }
1353 }
1354
1355 /*
1356  * extent_io.c clear_bit_hook, see set_bit_hook for why
1357  */
1358 static void btrfs_clear_bit_hook(struct inode *inode,
1359                                  struct extent_state *state, int *bits)
1360 {
1361         /*
1362          * set_bit and clear bit hooks normally require _irqsave/restore
1363          * but in this case, we are only testing for the DELALLOC
1364          * bit, which is only set or cleared with irqs on
1365          */
1366         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1367                 struct btrfs_root *root = BTRFS_I(inode)->root;
1368                 u64 len = state->end + 1 - state->start;
1369                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1370
1371                 if (*bits & EXTENT_FIRST_DELALLOC) {
1372                         *bits &= ~EXTENT_FIRST_DELALLOC;
1373                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374                         spin_lock(&BTRFS_I(inode)->lock);
1375                         BTRFS_I(inode)->outstanding_extents--;
1376                         spin_unlock(&BTRFS_I(inode)->lock);
1377                 }
1378
1379                 if (*bits & EXTENT_DO_ACCOUNTING)
1380                         btrfs_delalloc_release_metadata(inode, len);
1381
1382                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1383                     && do_list)
1384                         btrfs_free_reserved_data_space(inode, len);
1385
1386                 spin_lock(&root->fs_info->delalloc_lock);
1387                 root->fs_info->delalloc_bytes -= len;
1388                 BTRFS_I(inode)->delalloc_bytes -= len;
1389
1390                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1391                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1393                 }
1394                 spin_unlock(&root->fs_info->delalloc_lock);
1395         }
1396 }
1397
1398 /*
1399  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400  * we don't create bios that span stripes or chunks
1401  */
1402 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1403                          size_t size, struct bio *bio,
1404                          unsigned long bio_flags)
1405 {
1406         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407         struct btrfs_mapping_tree *map_tree;
1408         u64 logical = (u64)bio->bi_sector << 9;
1409         u64 length = 0;
1410         u64 map_length;
1411         int ret;
1412
1413         if (bio_flags & EXTENT_BIO_COMPRESSED)
1414                 return 0;
1415
1416         length = bio->bi_size;
1417         map_tree = &root->fs_info->mapping_tree;
1418         map_length = length;
1419         ret = btrfs_map_block(map_tree, READ, logical,
1420                               &map_length, NULL, 0);
1421
1422         if (map_length < length + size)
1423                 return 1;
1424         return ret;
1425 }
1426
1427 /*
1428  * in order to insert checksums into the metadata in large chunks,
1429  * we wait until bio submission time.   All the pages in the bio are
1430  * checksummed and sums are attached onto the ordered extent record.
1431  *
1432  * At IO completion time the cums attached on the ordered extent record
1433  * are inserted into the btree
1434  */
1435 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436                                     struct bio *bio, int mirror_num,
1437                                     unsigned long bio_flags,
1438                                     u64 bio_offset)
1439 {
1440         struct btrfs_root *root = BTRFS_I(inode)->root;
1441         int ret = 0;
1442
1443         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1444         BUG_ON(ret);
1445         return 0;
1446 }
1447
1448 /*
1449  * in order to insert checksums into the metadata in large chunks,
1450  * we wait until bio submission time.   All the pages in the bio are
1451  * checksummed and sums are attached onto the ordered extent record.
1452  *
1453  * At IO completion time the cums attached on the ordered extent record
1454  * are inserted into the btree
1455  */
1456 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1457                           int mirror_num, unsigned long bio_flags,
1458                           u64 bio_offset)
1459 {
1460         struct btrfs_root *root = BTRFS_I(inode)->root;
1461         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1462 }
1463
1464 /*
1465  * extent_io.c submission hook. This does the right thing for csum calculation
1466  * on write, or reading the csums from the tree before a read
1467  */
1468 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1469                           int mirror_num, unsigned long bio_flags,
1470                           u64 bio_offset)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         int ret = 0;
1474         int skip_sum;
1475
1476         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1477
1478         if (btrfs_is_free_space_inode(root, inode))
1479                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1480         else
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1482         BUG_ON(ret);
1483
1484         if (!(rw & REQ_WRITE)) {
1485                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1486                         return btrfs_submit_compressed_read(inode, bio,
1487                                                     mirror_num, bio_flags);
1488                 } else if (!skip_sum) {
1489                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1490                         if (ret)
1491                                 return ret;
1492                 }
1493                 goto mapit;
1494         } else if (!skip_sum) {
1495                 /* csum items have already been cloned */
1496                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1497                         goto mapit;
1498                 /* we're doing a write, do the async checksumming */
1499                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1500                                    inode, rw, bio, mirror_num,
1501                                    bio_flags, bio_offset,
1502                                    __btrfs_submit_bio_start,
1503                                    __btrfs_submit_bio_done);
1504         }
1505
1506 mapit:
1507         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1508 }
1509
1510 /*
1511  * given a list of ordered sums record them in the inode.  This happens
1512  * at IO completion time based on sums calculated at bio submission time.
1513  */
1514 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1515                              struct inode *inode, u64 file_offset,
1516                              struct list_head *list)
1517 {
1518         struct btrfs_ordered_sum *sum;
1519
1520         list_for_each_entry(sum, list, list) {
1521                 btrfs_csum_file_blocks(trans,
1522                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1523         }
1524         return 0;
1525 }
1526
1527 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528                               struct extent_state **cached_state)
1529 {
1530         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1531                 WARN_ON(1);
1532         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1533                                    cached_state, GFP_NOFS);
1534 }
1535
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup {
1538         struct page *page;
1539         struct btrfs_work work;
1540 };
1541
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1543 {
1544         struct btrfs_writepage_fixup *fixup;
1545         struct btrfs_ordered_extent *ordered;
1546         struct extent_state *cached_state = NULL;
1547         struct page *page;
1548         struct inode *inode;
1549         u64 page_start;
1550         u64 page_end;
1551
1552         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1553         page = fixup->page;
1554 again:
1555         lock_page(page);
1556         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557                 ClearPageChecked(page);
1558                 goto out_page;
1559         }
1560
1561         inode = page->mapping->host;
1562         page_start = page_offset(page);
1563         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1564
1565         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566                          &cached_state, GFP_NOFS);
1567
1568         /* already ordered? We're done */
1569         if (PagePrivate2(page))
1570                 goto out;
1571
1572         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1573         if (ordered) {
1574                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575                                      page_end, &cached_state, GFP_NOFS);
1576                 unlock_page(page);
1577                 btrfs_start_ordered_extent(inode, ordered, 1);
1578                 goto again;
1579         }
1580
1581         BUG();
1582         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1583         ClearPageChecked(page);
1584 out:
1585         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586                              &cached_state, GFP_NOFS);
1587 out_page:
1588         unlock_page(page);
1589         page_cache_release(page);
1590         kfree(fixup);
1591 }
1592
1593 /*
1594  * There are a few paths in the higher layers of the kernel that directly
1595  * set the page dirty bit without asking the filesystem if it is a
1596  * good idea.  This causes problems because we want to make sure COW
1597  * properly happens and the data=ordered rules are followed.
1598  *
1599  * In our case any range that doesn't have the ORDERED bit set
1600  * hasn't been properly setup for IO.  We kick off an async process
1601  * to fix it up.  The async helper will wait for ordered extents, set
1602  * the delalloc bit and make it safe to write the page.
1603  */
1604 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1605 {
1606         struct inode *inode = page->mapping->host;
1607         struct btrfs_writepage_fixup *fixup;
1608         struct btrfs_root *root = BTRFS_I(inode)->root;
1609
1610         /* this page is properly in the ordered list */
1611         if (TestClearPagePrivate2(page))
1612                 return 0;
1613
1614         if (PageChecked(page))
1615                 return -EAGAIN;
1616
1617         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1618         if (!fixup)
1619                 return -EAGAIN;
1620
1621         SetPageChecked(page);
1622         page_cache_get(page);
1623         fixup->work.func = btrfs_writepage_fixup_worker;
1624         fixup->page = page;
1625         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1626         return -EAGAIN;
1627 }
1628
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630                                        struct inode *inode, u64 file_pos,
1631                                        u64 disk_bytenr, u64 disk_num_bytes,
1632                                        u64 num_bytes, u64 ram_bytes,
1633                                        u8 compression, u8 encryption,
1634                                        u16 other_encoding, int extent_type)
1635 {
1636         struct btrfs_root *root = BTRFS_I(inode)->root;
1637         struct btrfs_file_extent_item *fi;
1638         struct btrfs_path *path;
1639         struct extent_buffer *leaf;
1640         struct btrfs_key ins;
1641         u64 hint;
1642         int ret;
1643
1644         path = btrfs_alloc_path();
1645         if (!path)
1646                 return -ENOMEM;
1647
1648         path->leave_spinning = 1;
1649
1650         /*
1651          * we may be replacing one extent in the tree with another.
1652          * The new extent is pinned in the extent map, and we don't want
1653          * to drop it from the cache until it is completely in the btree.
1654          *
1655          * So, tell btrfs_drop_extents to leave this extent in the cache.
1656          * the caller is expected to unpin it and allow it to be merged
1657          * with the others.
1658          */
1659         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1660                                  &hint, 0);
1661         BUG_ON(ret);
1662
1663         ins.objectid = btrfs_ino(inode);
1664         ins.offset = file_pos;
1665         ins.type = BTRFS_EXTENT_DATA_KEY;
1666         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1667         BUG_ON(ret);
1668         leaf = path->nodes[0];
1669         fi = btrfs_item_ptr(leaf, path->slots[0],
1670                             struct btrfs_file_extent_item);
1671         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672         btrfs_set_file_extent_type(leaf, fi, extent_type);
1673         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675         btrfs_set_file_extent_offset(leaf, fi, 0);
1676         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678         btrfs_set_file_extent_compression(leaf, fi, compression);
1679         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1681
1682         btrfs_unlock_up_safe(path, 1);
1683         btrfs_set_lock_blocking(leaf);
1684
1685         btrfs_mark_buffer_dirty(leaf);
1686
1687         inode_add_bytes(inode, num_bytes);
1688
1689         ins.objectid = disk_bytenr;
1690         ins.offset = disk_num_bytes;
1691         ins.type = BTRFS_EXTENT_ITEM_KEY;
1692         ret = btrfs_alloc_reserved_file_extent(trans, root,
1693                                         root->root_key.objectid,
1694                                         btrfs_ino(inode), file_pos, &ins);
1695         BUG_ON(ret);
1696         btrfs_free_path(path);
1697
1698         return 0;
1699 }
1700
1701 /*
1702  * helper function for btrfs_finish_ordered_io, this
1703  * just reads in some of the csum leaves to prime them into ram
1704  * before we start the transaction.  It limits the amount of btree
1705  * reads required while inside the transaction.
1706  */
1707 /* as ordered data IO finishes, this gets called so we can finish
1708  * an ordered extent if the range of bytes in the file it covers are
1709  * fully written.
1710  */
1711 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1712 {
1713         struct btrfs_root *root = BTRFS_I(inode)->root;
1714         struct btrfs_trans_handle *trans = NULL;
1715         struct btrfs_ordered_extent *ordered_extent = NULL;
1716         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1717         struct extent_state *cached_state = NULL;
1718         int compress_type = 0;
1719         int ret;
1720         bool nolock;
1721
1722         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1723                                              end - start + 1);
1724         if (!ret)
1725                 return 0;
1726         BUG_ON(!ordered_extent);
1727
1728         nolock = btrfs_is_free_space_inode(root, inode);
1729
1730         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731                 BUG_ON(!list_empty(&ordered_extent->list));
1732                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1733                 if (!ret) {
1734                         if (nolock)
1735                                 trans = btrfs_join_transaction_nolock(root);
1736                         else
1737                                 trans = btrfs_join_transaction(root);
1738                         BUG_ON(IS_ERR(trans));
1739                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740                         ret = btrfs_update_inode(trans, root, inode);
1741                         BUG_ON(ret);
1742                 }
1743                 goto out;
1744         }
1745
1746         lock_extent_bits(io_tree, ordered_extent->file_offset,
1747                          ordered_extent->file_offset + ordered_extent->len - 1,
1748                          0, &cached_state, GFP_NOFS);
1749
1750         if (nolock)
1751                 trans = btrfs_join_transaction_nolock(root);
1752         else
1753                 trans = btrfs_join_transaction(root);
1754         BUG_ON(IS_ERR(trans));
1755         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756
1757         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1758                 compress_type = ordered_extent->compress_type;
1759         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1760                 BUG_ON(compress_type);
1761                 ret = btrfs_mark_extent_written(trans, inode,
1762                                                 ordered_extent->file_offset,
1763                                                 ordered_extent->file_offset +
1764                                                 ordered_extent->len);
1765                 BUG_ON(ret);
1766         } else {
1767                 BUG_ON(root == root->fs_info->tree_root);
1768                 ret = insert_reserved_file_extent(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->start,
1771                                                 ordered_extent->disk_len,
1772                                                 ordered_extent->len,
1773                                                 ordered_extent->len,
1774                                                 compress_type, 0, 0,
1775                                                 BTRFS_FILE_EXTENT_REG);
1776                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777                                    ordered_extent->file_offset,
1778                                    ordered_extent->len);
1779                 BUG_ON(ret);
1780         }
1781         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782                              ordered_extent->file_offset +
1783                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1784
1785         add_pending_csums(trans, inode, ordered_extent->file_offset,
1786                           &ordered_extent->list);
1787
1788         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1789         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790                 ret = btrfs_update_inode(trans, root, inode);
1791                 BUG_ON(ret);
1792         }
1793         ret = 0;
1794 out:
1795         if (root != root->fs_info->tree_root)
1796                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1797         if (trans) {
1798                 if (nolock)
1799                         btrfs_end_transaction_nolock(trans, root);
1800                 else
1801                         btrfs_end_transaction(trans, root);
1802         }
1803
1804         /* once for us */
1805         btrfs_put_ordered_extent(ordered_extent);
1806         /* once for the tree */
1807         btrfs_put_ordered_extent(ordered_extent);
1808
1809         return 0;
1810 }
1811
1812 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1813                                 struct extent_state *state, int uptodate)
1814 {
1815         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1816
1817         ClearPagePrivate2(page);
1818         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1819 }
1820
1821 /*
1822  * When IO fails, either with EIO or csum verification fails, we
1823  * try other mirrors that might have a good copy of the data.  This
1824  * io_failure_record is used to record state as we go through all the
1825  * mirrors.  If another mirror has good data, the page is set up to date
1826  * and things continue.  If a good mirror can't be found, the original
1827  * bio end_io callback is called to indicate things have failed.
1828  */
1829 struct io_failure_record {
1830         struct page *page;
1831         u64 start;
1832         u64 len;
1833         u64 logical;
1834         unsigned long bio_flags;
1835         int last_mirror;
1836 };
1837
1838 static int btrfs_io_failed_hook(struct bio *failed_bio,
1839                          struct page *page, u64 start, u64 end,
1840                          struct extent_state *state)
1841 {
1842         struct io_failure_record *failrec = NULL;
1843         u64 private;
1844         struct extent_map *em;
1845         struct inode *inode = page->mapping->host;
1846         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1848         struct bio *bio;
1849         int num_copies;
1850         int ret;
1851         int rw;
1852         u64 logical;
1853
1854         ret = get_state_private(failure_tree, start, &private);
1855         if (ret) {
1856                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1857                 if (!failrec)
1858                         return -ENOMEM;
1859                 failrec->start = start;
1860                 failrec->len = end - start + 1;
1861                 failrec->last_mirror = 0;
1862                 failrec->bio_flags = 0;
1863
1864                 read_lock(&em_tree->lock);
1865                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866                 if (em->start > start || em->start + em->len < start) {
1867                         free_extent_map(em);
1868                         em = NULL;
1869                 }
1870                 read_unlock(&em_tree->lock);
1871
1872                 if (IS_ERR_OR_NULL(em)) {
1873                         kfree(failrec);
1874                         return -EIO;
1875                 }
1876                 logical = start - em->start;
1877                 logical = em->block_start + logical;
1878                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879                         logical = em->block_start;
1880                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881                         extent_set_compress_type(&failrec->bio_flags,
1882                                                  em->compress_type);
1883                 }
1884                 failrec->logical = logical;
1885                 free_extent_map(em);
1886                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887                                 EXTENT_DIRTY, GFP_NOFS);
1888                 set_state_private(failure_tree, start,
1889                                  (u64)(unsigned long)failrec);
1890         } else {
1891                 failrec = (struct io_failure_record *)(unsigned long)private;
1892         }
1893         num_copies = btrfs_num_copies(
1894                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895                               failrec->logical, failrec->len);
1896         failrec->last_mirror++;
1897         if (!state) {
1898                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900                                                     failrec->start,
1901                                                     EXTENT_LOCKED);
1902                 if (state && state->start != failrec->start)
1903                         state = NULL;
1904                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1905         }
1906         if (!state || failrec->last_mirror > num_copies) {
1907                 set_state_private(failure_tree, failrec->start, 0);
1908                 clear_extent_bits(failure_tree, failrec->start,
1909                                   failrec->start + failrec->len - 1,
1910                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911                 kfree(failrec);
1912                 return -EIO;
1913         }
1914         bio = bio_alloc(GFP_NOFS, 1);
1915         bio->bi_private = state;
1916         bio->bi_end_io = failed_bio->bi_end_io;
1917         bio->bi_sector = failrec->logical >> 9;
1918         bio->bi_bdev = failed_bio->bi_bdev;
1919         bio->bi_size = 0;
1920
1921         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922         if (failed_bio->bi_rw & REQ_WRITE)
1923                 rw = WRITE;
1924         else
1925                 rw = READ;
1926
1927         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928                                                       failrec->last_mirror,
1929                                                       failrec->bio_flags, 0);
1930         return ret;
1931 }
1932
1933 /*
1934  * each time an IO finishes, we do a fast check in the IO failure tree
1935  * to see if we need to process or clean up an io_failure_record
1936  */
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1938 {
1939         u64 private;
1940         u64 private_failure;
1941         struct io_failure_record *failure;
1942         int ret;
1943
1944         private = 0;
1945         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948                                         start, &private_failure);
1949                 if (ret == 0) {
1950                         failure = (struct io_failure_record *)(unsigned long)
1951                                    private_failure;
1952                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953                                           failure->start, 0);
1954                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955                                           failure->start,
1956                                           failure->start + failure->len - 1,
1957                                           EXTENT_DIRTY | EXTENT_LOCKED,
1958                                           GFP_NOFS);
1959                         kfree(failure);
1960                 }
1961         }
1962         return 0;
1963 }
1964
1965 /*
1966  * when reads are done, we need to check csums to verify the data is correct
1967  * if there's a match, we allow the bio to finish.  If not, we go through
1968  * the io_failure_record routines to find good copies
1969  */
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971                                struct extent_state *state)
1972 {
1973         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974         struct inode *inode = page->mapping->host;
1975         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1976         char *kaddr;
1977         u64 private = ~(u32)0;
1978         int ret;
1979         struct btrfs_root *root = BTRFS_I(inode)->root;
1980         u32 csum = ~(u32)0;
1981
1982         if (PageChecked(page)) {
1983                 ClearPageChecked(page);
1984                 goto good;
1985         }
1986
1987         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1988                 goto good;
1989
1990         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993                                   GFP_NOFS);
1994                 return 0;
1995         }
1996
1997         if (state && state->start == start) {
1998                 private = state->private;
1999                 ret = 0;
2000         } else {
2001                 ret = get_state_private(io_tree, start, &private);
2002         }
2003         kaddr = kmap_atomic(page, KM_USER0);
2004         if (ret)
2005                 goto zeroit;
2006
2007         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2008         btrfs_csum_final(csum, (char *)&csum);
2009         if (csum != private)
2010                 goto zeroit;
2011
2012         kunmap_atomic(kaddr, KM_USER0);
2013 good:
2014         /* if the io failure tree for this inode is non-empty,
2015          * check to see if we've recovered from a failed IO
2016          */
2017         btrfs_clean_io_failures(inode, start);
2018         return 0;
2019
2020 zeroit:
2021         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2022                        "private %llu\n",
2023                        (unsigned long long)btrfs_ino(page->mapping->host),
2024                        (unsigned long long)start, csum,
2025                        (unsigned long long)private);
2026         memset(kaddr + offset, 1, end - start + 1);
2027         flush_dcache_page(page);
2028         kunmap_atomic(kaddr, KM_USER0);
2029         if (private == 0)
2030                 return 0;
2031         return -EIO;
2032 }
2033
2034 struct delayed_iput {
2035         struct list_head list;
2036         struct inode *inode;
2037 };
2038
2039 void btrfs_add_delayed_iput(struct inode *inode)
2040 {
2041         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042         struct delayed_iput *delayed;
2043
2044         if (atomic_add_unless(&inode->i_count, -1, 1))
2045                 return;
2046
2047         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048         delayed->inode = inode;
2049
2050         spin_lock(&fs_info->delayed_iput_lock);
2051         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052         spin_unlock(&fs_info->delayed_iput_lock);
2053 }
2054
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056 {
2057         LIST_HEAD(list);
2058         struct btrfs_fs_info *fs_info = root->fs_info;
2059         struct delayed_iput *delayed;
2060         int empty;
2061
2062         spin_lock(&fs_info->delayed_iput_lock);
2063         empty = list_empty(&fs_info->delayed_iputs);
2064         spin_unlock(&fs_info->delayed_iput_lock);
2065         if (empty)
2066                 return;
2067
2068         down_read(&root->fs_info->cleanup_work_sem);
2069         spin_lock(&fs_info->delayed_iput_lock);
2070         list_splice_init(&fs_info->delayed_iputs, &list);
2071         spin_unlock(&fs_info->delayed_iput_lock);
2072
2073         while (!list_empty(&list)) {
2074                 delayed = list_entry(list.next, struct delayed_iput, list);
2075                 list_del(&delayed->list);
2076                 iput(delayed->inode);
2077                 kfree(delayed);
2078         }
2079         up_read(&root->fs_info->cleanup_work_sem);
2080 }
2081
2082 enum btrfs_orphan_cleanup_state {
2083         ORPHAN_CLEANUP_STARTED  = 1,
2084         ORPHAN_CLEANUP_DONE     = 2,
2085 };
2086
2087 /*
2088  * This is called in transaction commmit time. If there are no orphan
2089  * files in the subvolume, it removes orphan item and frees block_rsv
2090  * structure.
2091  */
2092 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2093                               struct btrfs_root *root)
2094 {
2095         int ret;
2096
2097         if (!list_empty(&root->orphan_list) ||
2098             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2099                 return;
2100
2101         if (root->orphan_item_inserted &&
2102             btrfs_root_refs(&root->root_item) > 0) {
2103                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2104                                             root->root_key.objectid);
2105                 BUG_ON(ret);
2106                 root->orphan_item_inserted = 0;
2107         }
2108
2109         if (root->orphan_block_rsv) {
2110                 WARN_ON(root->orphan_block_rsv->size > 0);
2111                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2112                 root->orphan_block_rsv = NULL;
2113         }
2114 }
2115
2116 /*
2117  * This creates an orphan entry for the given inode in case something goes
2118  * wrong in the middle of an unlink/truncate.
2119  *
2120  * NOTE: caller of this function should reserve 5 units of metadata for
2121  *       this function.
2122  */
2123 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2124 {
2125         struct btrfs_root *root = BTRFS_I(inode)->root;
2126         struct btrfs_block_rsv *block_rsv = NULL;
2127         int reserve = 0;
2128         int insert = 0;
2129         int ret;
2130
2131         if (!root->orphan_block_rsv) {
2132                 block_rsv = btrfs_alloc_block_rsv(root);
2133                 if (!block_rsv)
2134                         return -ENOMEM;
2135         }
2136
2137         spin_lock(&root->orphan_lock);
2138         if (!root->orphan_block_rsv) {
2139                 root->orphan_block_rsv = block_rsv;
2140         } else if (block_rsv) {
2141                 btrfs_free_block_rsv(root, block_rsv);
2142                 block_rsv = NULL;
2143         }
2144
2145         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2146                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2147 #if 0
2148                 /*
2149                  * For proper ENOSPC handling, we should do orphan
2150                  * cleanup when mounting. But this introduces backward
2151                  * compatibility issue.
2152                  */
2153                 if (!xchg(&root->orphan_item_inserted, 1))
2154                         insert = 2;
2155                 else
2156                         insert = 1;
2157 #endif
2158                 insert = 1;
2159         }
2160
2161         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2162                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2163                 reserve = 1;
2164         }
2165         spin_unlock(&root->orphan_lock);
2166
2167         /* grab metadata reservation from transaction handle */
2168         if (reserve) {
2169                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2170                 BUG_ON(ret);
2171         }
2172
2173         /* insert an orphan item to track this unlinked/truncated file */
2174         if (insert >= 1) {
2175                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2176                 BUG_ON(ret);
2177         }
2178
2179         /* insert an orphan item to track subvolume contains orphan files */
2180         if (insert >= 2) {
2181                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2182                                                root->root_key.objectid);
2183                 BUG_ON(ret);
2184         }
2185         return 0;
2186 }
2187
2188 /*
2189  * We have done the truncate/delete so we can go ahead and remove the orphan
2190  * item for this particular inode.
2191  */
2192 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         int delete_item = 0;
2196         int release_rsv = 0;
2197         int ret = 0;
2198
2199         spin_lock(&root->orphan_lock);
2200         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2201                 list_del_init(&BTRFS_I(inode)->i_orphan);
2202                 delete_item = 1;
2203         }
2204
2205         if (BTRFS_I(inode)->orphan_meta_reserved) {
2206                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2207                 release_rsv = 1;
2208         }
2209         spin_unlock(&root->orphan_lock);
2210
2211         if (trans && delete_item) {
2212                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2213                 BUG_ON(ret);
2214         }
2215
2216         if (release_rsv)
2217                 btrfs_orphan_release_metadata(inode);
2218
2219         return 0;
2220 }
2221
2222 /*
2223  * this cleans up any orphans that may be left on the list from the last use
2224  * of this root.
2225  */
2226 int btrfs_orphan_cleanup(struct btrfs_root *root)
2227 {
2228         struct btrfs_path *path;
2229         struct extent_buffer *leaf;
2230         struct btrfs_key key, found_key;
2231         struct btrfs_trans_handle *trans;
2232         struct inode *inode;
2233         u64 last_objectid = 0;
2234         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2235
2236         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2237                 return 0;
2238
2239         path = btrfs_alloc_path();
2240         if (!path) {
2241                 ret = -ENOMEM;
2242                 goto out;
2243         }
2244         path->reada = -1;
2245
2246         key.objectid = BTRFS_ORPHAN_OBJECTID;
2247         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2248         key.offset = (u64)-1;
2249
2250         while (1) {
2251                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2252                 if (ret < 0)
2253                         goto out;
2254
2255                 /*
2256                  * if ret == 0 means we found what we were searching for, which
2257                  * is weird, but possible, so only screw with path if we didn't
2258                  * find the key and see if we have stuff that matches
2259                  */
2260                 if (ret > 0) {
2261                         ret = 0;
2262                         if (path->slots[0] == 0)
2263                                 break;
2264                         path->slots[0]--;
2265                 }
2266
2267                 /* pull out the item */
2268                 leaf = path->nodes[0];
2269                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2270
2271                 /* make sure the item matches what we want */
2272                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2273                         break;
2274                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2275                         break;
2276
2277                 /* release the path since we're done with it */
2278                 btrfs_release_path(path);
2279
2280                 /*
2281                  * this is where we are basically btrfs_lookup, without the
2282                  * crossing root thing.  we store the inode number in the
2283                  * offset of the orphan item.
2284                  */
2285
2286                 if (found_key.offset == last_objectid) {
2287                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2288                                "stopping orphan cleanup\n");
2289                         ret = -EINVAL;
2290                         goto out;
2291                 }
2292
2293                 last_objectid = found_key.offset;
2294
2295                 found_key.objectid = found_key.offset;
2296                 found_key.type = BTRFS_INODE_ITEM_KEY;
2297                 found_key.offset = 0;
2298                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2299                 ret = PTR_RET(inode);
2300                 if (ret && ret != -ESTALE)
2301                         goto out;
2302
2303                 /*
2304                  * Inode is already gone but the orphan item is still there,
2305                  * kill the orphan item.
2306                  */
2307                 if (ret == -ESTALE) {
2308                         trans = btrfs_start_transaction(root, 1);
2309                         if (IS_ERR(trans)) {
2310                                 ret = PTR_ERR(trans);
2311                                 goto out;
2312                         }
2313                         ret = btrfs_del_orphan_item(trans, root,
2314                                                     found_key.objectid);
2315                         BUG_ON(ret);
2316                         btrfs_end_transaction(trans, root);
2317                         continue;
2318                 }
2319
2320                 /*
2321                  * add this inode to the orphan list so btrfs_orphan_del does
2322                  * the proper thing when we hit it
2323                  */
2324                 spin_lock(&root->orphan_lock);
2325                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2326                 spin_unlock(&root->orphan_lock);
2327
2328                 /* if we have links, this was a truncate, lets do that */
2329                 if (inode->i_nlink) {
2330                         if (!S_ISREG(inode->i_mode)) {
2331                                 WARN_ON(1);
2332                                 iput(inode);
2333                                 continue;
2334                         }
2335                         nr_truncate++;
2336                         ret = btrfs_truncate(inode);
2337                 } else {
2338                         nr_unlink++;
2339                 }
2340
2341                 /* this will do delete_inode and everything for us */
2342                 iput(inode);
2343                 if (ret)
2344                         goto out;
2345         }
2346         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2347
2348         if (root->orphan_block_rsv)
2349                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2350                                         (u64)-1);
2351
2352         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2353                 trans = btrfs_join_transaction(root);
2354                 if (!IS_ERR(trans))
2355                         btrfs_end_transaction(trans, root);
2356         }
2357
2358         if (nr_unlink)
2359                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2360         if (nr_truncate)
2361                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2362
2363 out:
2364         if (ret)
2365                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2366         btrfs_free_path(path);
2367         return ret;
2368 }
2369
2370 /*
2371  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2372  * don't find any xattrs, we know there can't be any acls.
2373  *
2374  * slot is the slot the inode is in, objectid is the objectid of the inode
2375  */
2376 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2377                                           int slot, u64 objectid)
2378 {
2379         u32 nritems = btrfs_header_nritems(leaf);
2380         struct btrfs_key found_key;
2381         int scanned = 0;
2382
2383         slot++;
2384         while (slot < nritems) {
2385                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2386
2387                 /* we found a different objectid, there must not be acls */
2388                 if (found_key.objectid != objectid)
2389                         return 0;
2390
2391                 /* we found an xattr, assume we've got an acl */
2392                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2393                         return 1;
2394
2395                 /*
2396                  * we found a key greater than an xattr key, there can't
2397                  * be any acls later on
2398                  */
2399                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2400                         return 0;
2401
2402                 slot++;
2403                 scanned++;
2404
2405                 /*
2406                  * it goes inode, inode backrefs, xattrs, extents,
2407                  * so if there are a ton of hard links to an inode there can
2408                  * be a lot of backrefs.  Don't waste time searching too hard,
2409                  * this is just an optimization
2410                  */
2411                 if (scanned >= 8)
2412                         break;
2413         }
2414         /* we hit the end of the leaf before we found an xattr or
2415          * something larger than an xattr.  We have to assume the inode
2416          * has acls
2417          */
2418         return 1;
2419 }
2420
2421 /*
2422  * read an inode from the btree into the in-memory inode
2423  */
2424 static void btrfs_read_locked_inode(struct inode *inode)
2425 {
2426         struct btrfs_path *path;
2427         struct extent_buffer *leaf;
2428         struct btrfs_inode_item *inode_item;
2429         struct btrfs_timespec *tspec;
2430         struct btrfs_root *root = BTRFS_I(inode)->root;
2431         struct btrfs_key location;
2432         int maybe_acls;
2433         u32 rdev;
2434         int ret;
2435         bool filled = false;
2436
2437         ret = btrfs_fill_inode(inode, &rdev);
2438         if (!ret)
2439                 filled = true;
2440
2441         path = btrfs_alloc_path();
2442         if (!path)
2443                 goto make_bad;
2444
2445         path->leave_spinning = 1;
2446         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2447
2448         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2449         if (ret)
2450                 goto make_bad;
2451
2452         leaf = path->nodes[0];
2453
2454         if (filled)
2455                 goto cache_acl;
2456
2457         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458                                     struct btrfs_inode_item);
2459         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2460         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2461         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2462         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2463         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2464
2465         tspec = btrfs_inode_atime(inode_item);
2466         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2467         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2468
2469         tspec = btrfs_inode_mtime(inode_item);
2470         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2472
2473         tspec = btrfs_inode_ctime(inode_item);
2474         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2476
2477         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2478         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2479         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2480         inode->i_generation = BTRFS_I(inode)->generation;
2481         inode->i_rdev = 0;
2482         rdev = btrfs_inode_rdev(leaf, inode_item);
2483
2484         BTRFS_I(inode)->index_cnt = (u64)-1;
2485         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2486 cache_acl:
2487         /*
2488          * try to precache a NULL acl entry for files that don't have
2489          * any xattrs or acls
2490          */
2491         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2492                                            btrfs_ino(inode));
2493         if (!maybe_acls)
2494                 cache_no_acl(inode);
2495
2496         btrfs_free_path(path);
2497
2498         switch (inode->i_mode & S_IFMT) {
2499         case S_IFREG:
2500                 inode->i_mapping->a_ops = &btrfs_aops;
2501                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2502                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2503                 inode->i_fop = &btrfs_file_operations;
2504                 inode->i_op = &btrfs_file_inode_operations;
2505                 break;
2506         case S_IFDIR:
2507                 inode->i_fop = &btrfs_dir_file_operations;
2508                 if (root == root->fs_info->tree_root)
2509                         inode->i_op = &btrfs_dir_ro_inode_operations;
2510                 else
2511                         inode->i_op = &btrfs_dir_inode_operations;
2512                 break;
2513         case S_IFLNK:
2514                 inode->i_op = &btrfs_symlink_inode_operations;
2515                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2516                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2517                 break;
2518         default:
2519                 inode->i_op = &btrfs_special_inode_operations;
2520                 init_special_inode(inode, inode->i_mode, rdev);
2521                 break;
2522         }
2523
2524         btrfs_update_iflags(inode);
2525         return;
2526
2527 make_bad:
2528         btrfs_free_path(path);
2529         make_bad_inode(inode);
2530 }
2531
2532 /*
2533  * given a leaf and an inode, copy the inode fields into the leaf
2534  */
2535 static void fill_inode_item(struct btrfs_trans_handle *trans,
2536                             struct extent_buffer *leaf,
2537                             struct btrfs_inode_item *item,
2538                             struct inode *inode)
2539 {
2540         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2541         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2542         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2543         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2544         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2545
2546         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2547                                inode->i_atime.tv_sec);
2548         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2549                                 inode->i_atime.tv_nsec);
2550
2551         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2552                                inode->i_mtime.tv_sec);
2553         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2554                                 inode->i_mtime.tv_nsec);
2555
2556         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2557                                inode->i_ctime.tv_sec);
2558         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2559                                 inode->i_ctime.tv_nsec);
2560
2561         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2562         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2563         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2564         btrfs_set_inode_transid(leaf, item, trans->transid);
2565         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2566         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2567         btrfs_set_inode_block_group(leaf, item, 0);
2568 }
2569
2570 /*
2571  * copy everything in the in-memory inode into the btree.
2572  */
2573 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2574                                 struct btrfs_root *root, struct inode *inode)
2575 {
2576         struct btrfs_inode_item *inode_item;
2577         struct btrfs_path *path;
2578         struct extent_buffer *leaf;
2579         int ret;
2580
2581         /*
2582          * If the inode is a free space inode, we can deadlock during commit
2583          * if we put it into the delayed code.
2584          *
2585          * The data relocation inode should also be directly updated
2586          * without delay
2587          */
2588         if (!btrfs_is_free_space_inode(root, inode)
2589             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2590                 ret = btrfs_delayed_update_inode(trans, root, inode);
2591                 if (!ret)
2592                         btrfs_set_inode_last_trans(trans, inode);
2593                 return ret;
2594         }
2595
2596         path = btrfs_alloc_path();
2597         if (!path)
2598                 return -ENOMEM;
2599
2600         path->leave_spinning = 1;
2601         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2602                                  1);
2603         if (ret) {
2604                 if (ret > 0)
2605                         ret = -ENOENT;
2606                 goto failed;
2607         }
2608
2609         btrfs_unlock_up_safe(path, 1);
2610         leaf = path->nodes[0];
2611         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2612                                     struct btrfs_inode_item);
2613
2614         fill_inode_item(trans, leaf, inode_item, inode);
2615         btrfs_mark_buffer_dirty(leaf);
2616         btrfs_set_inode_last_trans(trans, inode);
2617         ret = 0;
2618 failed:
2619         btrfs_free_path(path);
2620         return ret;
2621 }
2622
2623 /*
2624  * unlink helper that gets used here in inode.c and in the tree logging
2625  * recovery code.  It remove a link in a directory with a given name, and
2626  * also drops the back refs in the inode to the directory
2627  */
2628 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2629                                 struct btrfs_root *root,
2630                                 struct inode *dir, struct inode *inode,
2631                                 const char *name, int name_len)
2632 {
2633         struct btrfs_path *path;
2634         int ret = 0;
2635         struct extent_buffer *leaf;
2636         struct btrfs_dir_item *di;
2637         struct btrfs_key key;
2638         u64 index;
2639         u64 ino = btrfs_ino(inode);
2640         u64 dir_ino = btrfs_ino(dir);
2641
2642         path = btrfs_alloc_path();
2643         if (!path) {
2644                 ret = -ENOMEM;
2645                 goto out;
2646         }
2647
2648         path->leave_spinning = 1;
2649         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2650                                     name, name_len, -1);
2651         if (IS_ERR(di)) {
2652                 ret = PTR_ERR(di);
2653                 goto err;
2654         }
2655         if (!di) {
2656                 ret = -ENOENT;
2657                 goto err;
2658         }
2659         leaf = path->nodes[0];
2660         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2661         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2662         if (ret)
2663                 goto err;
2664         btrfs_release_path(path);
2665
2666         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2667                                   dir_ino, &index);
2668         if (ret) {
2669                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2670                        "inode %llu parent %llu\n", name_len, name,
2671                        (unsigned long long)ino, (unsigned long long)dir_ino);
2672                 goto err;
2673         }
2674
2675         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2676         if (ret)
2677                 goto err;
2678
2679         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2680                                          inode, dir_ino);
2681         BUG_ON(ret != 0 && ret != -ENOENT);
2682
2683         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2684                                            dir, index);
2685         if (ret == -ENOENT)
2686                 ret = 0;
2687 err:
2688         btrfs_free_path(path);
2689         if (ret)
2690                 goto out;
2691
2692         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2693         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2694         btrfs_update_inode(trans, root, dir);
2695 out:
2696         return ret;
2697 }
2698
2699 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2700                        struct btrfs_root *root,
2701                        struct inode *dir, struct inode *inode,
2702                        const char *name, int name_len)
2703 {
2704         int ret;
2705         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2706         if (!ret) {
2707                 btrfs_drop_nlink(inode);
2708                 ret = btrfs_update_inode(trans, root, inode);
2709         }
2710         return ret;
2711 }
2712
2713
2714 /* helper to check if there is any shared block in the path */
2715 static int check_path_shared(struct btrfs_root *root,
2716                              struct btrfs_path *path)
2717 {
2718         struct extent_buffer *eb;
2719         int level;
2720         u64 refs = 1;
2721
2722         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2723                 int ret;
2724
2725                 if (!path->nodes[level])
2726                         break;
2727                 eb = path->nodes[level];
2728                 if (!btrfs_block_can_be_shared(root, eb))
2729                         continue;
2730                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2731                                                &refs, NULL);
2732                 if (refs > 1)
2733                         return 1;
2734         }
2735         return 0;
2736 }
2737
2738 /*
2739  * helper to start transaction for unlink and rmdir.
2740  *
2741  * unlink and rmdir are special in btrfs, they do not always free space.
2742  * so in enospc case, we should make sure they will free space before
2743  * allowing them to use the global metadata reservation.
2744  */
2745 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2746                                                        struct dentry *dentry)
2747 {
2748         struct btrfs_trans_handle *trans;
2749         struct btrfs_root *root = BTRFS_I(dir)->root;
2750         struct btrfs_path *path;
2751         struct btrfs_inode_ref *ref;
2752         struct btrfs_dir_item *di;
2753         struct inode *inode = dentry->d_inode;
2754         u64 index;
2755         int check_link = 1;
2756         int err = -ENOSPC;
2757         int ret;
2758         u64 ino = btrfs_ino(inode);
2759         u64 dir_ino = btrfs_ino(dir);
2760
2761         /*
2762          * 1 for the possible orphan item
2763          * 1 for the dir item
2764          * 1 for the dir index
2765          * 1 for the inode ref
2766          * 1 for the inode ref in the tree log
2767          * 2 for the dir entries in the log
2768          * 1 for the inode
2769          */
2770         trans = btrfs_start_transaction(root, 8);
2771         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2772                 return trans;
2773
2774         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2775                 return ERR_PTR(-ENOSPC);
2776
2777         /* check if there is someone else holds reference */
2778         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2779                 return ERR_PTR(-ENOSPC);
2780
2781         if (atomic_read(&inode->i_count) > 2)
2782                 return ERR_PTR(-ENOSPC);
2783
2784         if (xchg(&root->fs_info->enospc_unlink, 1))
2785                 return ERR_PTR(-ENOSPC);
2786
2787         path = btrfs_alloc_path();
2788         if (!path) {
2789                 root->fs_info->enospc_unlink = 0;
2790                 return ERR_PTR(-ENOMEM);
2791         }
2792
2793         /* 1 for the orphan item */
2794         trans = btrfs_start_transaction(root, 1);
2795         if (IS_ERR(trans)) {
2796                 btrfs_free_path(path);
2797                 root->fs_info->enospc_unlink = 0;
2798                 return trans;
2799         }
2800
2801         path->skip_locking = 1;
2802         path->search_commit_root = 1;
2803
2804         ret = btrfs_lookup_inode(trans, root, path,
2805                                 &BTRFS_I(dir)->location, 0);
2806         if (ret < 0) {
2807                 err = ret;
2808                 goto out;
2809         }
2810         if (ret == 0) {
2811                 if (check_path_shared(root, path))
2812                         goto out;
2813         } else {
2814                 check_link = 0;
2815         }
2816         btrfs_release_path(path);
2817
2818         ret = btrfs_lookup_inode(trans, root, path,
2819                                 &BTRFS_I(inode)->location, 0);
2820         if (ret < 0) {
2821                 err = ret;
2822                 goto out;
2823         }
2824         if (ret == 0) {
2825                 if (check_path_shared(root, path))
2826                         goto out;
2827         } else {
2828                 check_link = 0;
2829         }
2830         btrfs_release_path(path);
2831
2832         if (ret == 0 && S_ISREG(inode->i_mode)) {
2833                 ret = btrfs_lookup_file_extent(trans, root, path,
2834                                                ino, (u64)-1, 0);
2835                 if (ret < 0) {
2836                         err = ret;
2837                         goto out;
2838                 }
2839                 BUG_ON(ret == 0);
2840                 if (check_path_shared(root, path))
2841                         goto out;
2842                 btrfs_release_path(path);
2843         }
2844
2845         if (!check_link) {
2846                 err = 0;
2847                 goto out;
2848         }
2849
2850         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2851                                 dentry->d_name.name, dentry->d_name.len, 0);
2852         if (IS_ERR(di)) {
2853                 err = PTR_ERR(di);
2854                 goto out;
2855         }
2856         if (di) {
2857                 if (check_path_shared(root, path))
2858                         goto out;
2859         } else {
2860                 err = 0;
2861                 goto out;
2862         }
2863         btrfs_release_path(path);
2864
2865         ref = btrfs_lookup_inode_ref(trans, root, path,
2866                                 dentry->d_name.name, dentry->d_name.len,
2867                                 ino, dir_ino, 0);
2868         if (IS_ERR(ref)) {
2869                 err = PTR_ERR(ref);
2870                 goto out;
2871         }
2872         BUG_ON(!ref);
2873         if (check_path_shared(root, path))
2874                 goto out;
2875         index = btrfs_inode_ref_index(path->nodes[0], ref);
2876         btrfs_release_path(path);
2877
2878         /*
2879          * This is a commit root search, if we can lookup inode item and other
2880          * relative items in the commit root, it means the transaction of
2881          * dir/file creation has been committed, and the dir index item that we
2882          * delay to insert has also been inserted into the commit root. So
2883          * we needn't worry about the delayed insertion of the dir index item
2884          * here.
2885          */
2886         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2887                                 dentry->d_name.name, dentry->d_name.len, 0);
2888         if (IS_ERR(di)) {
2889                 err = PTR_ERR(di);
2890                 goto out;
2891         }
2892         BUG_ON(ret == -ENOENT);
2893         if (check_path_shared(root, path))
2894                 goto out;
2895
2896         err = 0;
2897 out:
2898         btrfs_free_path(path);
2899         /* Migrate the orphan reservation over */
2900         if (!err)
2901                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2902                                 &root->fs_info->global_block_rsv,
2903                                 btrfs_calc_trans_metadata_size(root, 1));
2904
2905         if (err) {
2906                 btrfs_end_transaction(trans, root);
2907                 root->fs_info->enospc_unlink = 0;
2908                 return ERR_PTR(err);
2909         }
2910
2911         trans->block_rsv = &root->fs_info->global_block_rsv;
2912         return trans;
2913 }
2914
2915 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2916                                struct btrfs_root *root)
2917 {
2918         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2919                 BUG_ON(!root->fs_info->enospc_unlink);
2920                 root->fs_info->enospc_unlink = 0;
2921         }
2922         btrfs_end_transaction_throttle(trans, root);
2923 }
2924
2925 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2926 {
2927         struct btrfs_root *root = BTRFS_I(dir)->root;
2928         struct btrfs_trans_handle *trans;
2929         struct inode *inode = dentry->d_inode;
2930         int ret;
2931         unsigned long nr = 0;
2932
2933         trans = __unlink_start_trans(dir, dentry);
2934         if (IS_ERR(trans))
2935                 return PTR_ERR(trans);
2936
2937         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2938
2939         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2940                                  dentry->d_name.name, dentry->d_name.len);
2941         if (ret)
2942                 goto out;
2943
2944         if (inode->i_nlink == 0) {
2945                 ret = btrfs_orphan_add(trans, inode);
2946                 if (ret)
2947                         goto out;
2948         }
2949
2950 out:
2951         nr = trans->blocks_used;
2952         __unlink_end_trans(trans, root);
2953         btrfs_btree_balance_dirty(root, nr);
2954         return ret;
2955 }
2956
2957 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2958                         struct btrfs_root *root,
2959                         struct inode *dir, u64 objectid,
2960                         const char *name, int name_len)
2961 {
2962         struct btrfs_path *path;
2963         struct extent_buffer *leaf;
2964         struct btrfs_dir_item *di;
2965         struct btrfs_key key;
2966         u64 index;
2967         int ret;
2968         u64 dir_ino = btrfs_ino(dir);
2969
2970         path = btrfs_alloc_path();
2971         if (!path)
2972                 return -ENOMEM;
2973
2974         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2975                                    name, name_len, -1);
2976         BUG_ON(IS_ERR_OR_NULL(di));
2977
2978         leaf = path->nodes[0];
2979         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2980         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2981         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2982         BUG_ON(ret);
2983         btrfs_release_path(path);
2984
2985         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2986                                  objectid, root->root_key.objectid,
2987                                  dir_ino, &index, name, name_len);
2988         if (ret < 0) {
2989                 BUG_ON(ret != -ENOENT);
2990                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2991                                                  name, name_len);
2992                 BUG_ON(IS_ERR_OR_NULL(di));
2993
2994                 leaf = path->nodes[0];
2995                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2996                 btrfs_release_path(path);
2997                 index = key.offset;
2998         }
2999         btrfs_release_path(path);
3000
3001         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3002         BUG_ON(ret);
3003
3004         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3005         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3006         ret = btrfs_update_inode(trans, root, dir);
3007         BUG_ON(ret);
3008
3009         btrfs_free_path(path);
3010         return 0;
3011 }
3012
3013 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3014 {
3015         struct inode *inode = dentry->d_inode;
3016         int err = 0;
3017         struct btrfs_root *root = BTRFS_I(dir)->root;
3018         struct btrfs_trans_handle *trans;
3019         unsigned long nr = 0;
3020
3021         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3022             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3023                 return -ENOTEMPTY;
3024
3025         trans = __unlink_start_trans(dir, dentry);
3026         if (IS_ERR(trans))
3027                 return PTR_ERR(trans);
3028
3029         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3030                 err = btrfs_unlink_subvol(trans, root, dir,
3031                                           BTRFS_I(inode)->location.objectid,
3032                                           dentry->d_name.name,
3033                                           dentry->d_name.len);
3034                 goto out;
3035         }
3036
3037         err = btrfs_orphan_add(trans, inode);
3038         if (err)
3039                 goto out;
3040
3041         /* now the directory is empty */
3042         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3043                                  dentry->d_name.name, dentry->d_name.len);
3044         if (!err)
3045                 btrfs_i_size_write(inode, 0);
3046 out:
3047         nr = trans->blocks_used;
3048         __unlink_end_trans(trans, root);
3049         btrfs_btree_balance_dirty(root, nr);
3050
3051         return err;
3052 }
3053
3054 /*
3055  * this can truncate away extent items, csum items and directory items.
3056  * It starts at a high offset and removes keys until it can't find
3057  * any higher than new_size
3058  *
3059  * csum items that cross the new i_size are truncated to the new size
3060  * as well.
3061  *
3062  * min_type is the minimum key type to truncate down to.  If set to 0, this
3063  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3064  */
3065 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3066                                struct btrfs_root *root,
3067                                struct inode *inode,
3068                                u64 new_size, u32 min_type)
3069 {
3070         struct btrfs_path *path;
3071         struct extent_buffer *leaf;
3072         struct btrfs_file_extent_item *fi;
3073         struct btrfs_key key;
3074         struct btrfs_key found_key;
3075         u64 extent_start = 0;
3076         u64 extent_num_bytes = 0;
3077         u64 extent_offset = 0;
3078         u64 item_end = 0;
3079         u64 mask = root->sectorsize - 1;
3080         u32 found_type = (u8)-1;
3081         int found_extent;
3082         int del_item;
3083         int pending_del_nr = 0;
3084         int pending_del_slot = 0;
3085         int extent_type = -1;
3086         int encoding;
3087         int ret;
3088         int err = 0;
3089         u64 ino = btrfs_ino(inode);
3090
3091         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3092
3093         path = btrfs_alloc_path();
3094         if (!path)
3095                 return -ENOMEM;
3096         path->reada = -1;
3097
3098         if (root->ref_cows || root == root->fs_info->tree_root)
3099                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3100
3101         /*
3102          * This function is also used to drop the items in the log tree before
3103          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3104          * it is used to drop the loged items. So we shouldn't kill the delayed
3105          * items.
3106          */
3107         if (min_type == 0 && root == BTRFS_I(inode)->root)
3108                 btrfs_kill_delayed_inode_items(inode);
3109
3110         key.objectid = ino;
3111         key.offset = (u64)-1;
3112         key.type = (u8)-1;
3113
3114 search_again:
3115         path->leave_spinning = 1;
3116         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3117         if (ret < 0) {
3118                 err = ret;
3119                 goto out;
3120         }
3121
3122         if (ret > 0) {
3123                 /* there are no items in the tree for us to truncate, we're
3124                  * done
3125                  */
3126                 if (path->slots[0] == 0)
3127                         goto out;
3128                 path->slots[0]--;
3129         }
3130
3131         while (1) {
3132                 fi = NULL;
3133                 leaf = path->nodes[0];
3134                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3135                 found_type = btrfs_key_type(&found_key);
3136                 encoding = 0;
3137
3138                 if (found_key.objectid != ino)
3139                         break;
3140
3141                 if (found_type < min_type)
3142                         break;
3143
3144                 item_end = found_key.offset;
3145                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3146                         fi = btrfs_item_ptr(leaf, path->slots[0],
3147                                             struct btrfs_file_extent_item);
3148                         extent_type = btrfs_file_extent_type(leaf, fi);
3149                         encoding = btrfs_file_extent_compression(leaf, fi);
3150                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3151                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3152
3153                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3154                                 item_end +=
3155                                     btrfs_file_extent_num_bytes(leaf, fi);
3156                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3157                                 item_end += btrfs_file_extent_inline_len(leaf,
3158                                                                          fi);
3159                         }
3160                         item_end--;
3161                 }
3162                 if (found_type > min_type) {
3163                         del_item = 1;
3164                 } else {
3165                         if (item_end < new_size)
3166                                 break;
3167                         if (found_key.offset >= new_size)
3168                                 del_item = 1;
3169                         else
3170                                 del_item = 0;
3171                 }
3172                 found_extent = 0;
3173                 /* FIXME, shrink the extent if the ref count is only 1 */
3174                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3175                         goto delete;
3176
3177                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3178                         u64 num_dec;
3179                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3180                         if (!del_item && !encoding) {
3181                                 u64 orig_num_bytes =
3182                                         btrfs_file_extent_num_bytes(leaf, fi);
3183                                 extent_num_bytes = new_size -
3184                                         found_key.offset + root->sectorsize - 1;
3185                                 extent_num_bytes = extent_num_bytes &
3186                                         ~((u64)root->sectorsize - 1);
3187                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3188                                                          extent_num_bytes);
3189                                 num_dec = (orig_num_bytes -
3190                                            extent_num_bytes);
3191                                 if (root->ref_cows && extent_start != 0)
3192                                         inode_sub_bytes(inode, num_dec);
3193                                 btrfs_mark_buffer_dirty(leaf);
3194                         } else {
3195                                 extent_num_bytes =
3196                                         btrfs_file_extent_disk_num_bytes(leaf,
3197                                                                          fi);
3198                                 extent_offset = found_key.offset -
3199                                         btrfs_file_extent_offset(leaf, fi);
3200
3201                                 /* FIXME blocksize != 4096 */
3202                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3203                                 if (extent_start != 0) {
3204                                         found_extent = 1;
3205                                         if (root->ref_cows)
3206                                                 inode_sub_bytes(inode, num_dec);
3207                                 }
3208                         }
3209                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3210                         /*
3211                          * we can't truncate inline items that have had
3212                          * special encodings
3213                          */
3214                         if (!del_item &&
3215                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3216                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3217                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3218                                 u32 size = new_size - found_key.offset;
3219
3220                                 if (root->ref_cows) {
3221                                         inode_sub_bytes(inode, item_end + 1 -
3222                                                         new_size);
3223                                 }
3224                                 size =
3225                                     btrfs_file_extent_calc_inline_size(size);
3226                                 ret = btrfs_truncate_item(trans, root, path,
3227                                                           size, 1);
3228                         } else if (root->ref_cows) {
3229                                 inode_sub_bytes(inode, item_end + 1 -
3230                                                 found_key.offset);
3231                         }
3232                 }
3233 delete:
3234                 if (del_item) {
3235                         if (!pending_del_nr) {
3236                                 /* no pending yet, add ourselves */
3237                                 pending_del_slot = path->slots[0];
3238                                 pending_del_nr = 1;
3239                         } else if (pending_del_nr &&
3240                                    path->slots[0] + 1 == pending_del_slot) {
3241                                 /* hop on the pending chunk */
3242                                 pending_del_nr++;
3243                                 pending_del_slot = path->slots[0];
3244                         } else {
3245                                 BUG();
3246                         }
3247                 } else {
3248                         break;
3249                 }
3250                 if (found_extent && (root->ref_cows ||
3251                                      root == root->fs_info->tree_root)) {
3252                         btrfs_set_path_blocking(path);
3253                         ret = btrfs_free_extent(trans, root, extent_start,
3254                                                 extent_num_bytes, 0,
3255                                                 btrfs_header_owner(leaf),
3256                                                 ino, extent_offset);
3257                         BUG_ON(ret);
3258                 }
3259
3260                 if (found_type == BTRFS_INODE_ITEM_KEY)
3261                         break;
3262
3263                 if (path->slots[0] == 0 ||
3264                     path->slots[0] != pending_del_slot) {
3265                         if (root->ref_cows &&
3266                             BTRFS_I(inode)->location.objectid !=
3267                                                 BTRFS_FREE_INO_OBJECTID) {
3268                                 err = -EAGAIN;
3269                                 goto out;
3270                         }
3271                         if (pending_del_nr) {
3272                                 ret = btrfs_del_items(trans, root, path,
3273                                                 pending_del_slot,
3274                                                 pending_del_nr);
3275                                 BUG_ON(ret);
3276                                 pending_del_nr = 0;
3277                         }
3278                         btrfs_release_path(path);
3279                         goto search_again;
3280                 } else {
3281                         path->slots[0]--;
3282                 }
3283         }
3284 out:
3285         if (pending_del_nr) {
3286                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3287                                       pending_del_nr);
3288                 BUG_ON(ret);
3289         }
3290         btrfs_free_path(path);
3291         return err;
3292 }
3293
3294 /*
3295  * taken from block_truncate_page, but does cow as it zeros out
3296  * any bytes left in the last page in the file.
3297  */
3298 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3299 {
3300         struct inode *inode = mapping->host;
3301         struct btrfs_root *root = BTRFS_I(inode)->root;
3302         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3303         struct btrfs_ordered_extent *ordered;
3304         struct extent_state *cached_state = NULL;
3305         char *kaddr;
3306         u32 blocksize = root->sectorsize;
3307         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3308         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3309         struct page *page;
3310         gfp_t mask = btrfs_alloc_write_mask(mapping);
3311         int ret = 0;
3312         u64 page_start;
3313         u64 page_end;
3314
3315         if ((offset & (blocksize - 1)) == 0)
3316                 goto out;
3317         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3318         if (ret)
3319                 goto out;
3320
3321         ret = -ENOMEM;
3322 again:
3323         page = find_or_create_page(mapping, index, mask);
3324         if (!page) {
3325                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3326                 goto out;
3327         }
3328
3329         page_start = page_offset(page);
3330         page_end = page_start + PAGE_CACHE_SIZE - 1;
3331
3332         if (!PageUptodate(page)) {
3333                 ret = btrfs_readpage(NULL, page);
3334                 lock_page(page);
3335                 if (page->mapping != mapping) {
3336                         unlock_page(page);
3337                         page_cache_release(page);
3338                         goto again;
3339                 }
3340                 if (!PageUptodate(page)) {
3341                         ret = -EIO;
3342                         goto out_unlock;
3343                 }
3344         }
3345         wait_on_page_writeback(page);
3346
3347         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3348                          GFP_NOFS);
3349         set_page_extent_mapped(page);
3350
3351         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3352         if (ordered) {
3353                 unlock_extent_cached(io_tree, page_start, page_end,
3354                                      &cached_state, GFP_NOFS);
3355                 unlock_page(page);
3356                 page_cache_release(page);
3357                 btrfs_start_ordered_extent(inode, ordered, 1);
3358                 btrfs_put_ordered_extent(ordered);
3359                 goto again;
3360         }
3361
3362         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3363                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3364                           0, 0, &cached_state, GFP_NOFS);
3365
3366         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3367                                         &cached_state);
3368         if (ret) {
3369                 unlock_extent_cached(io_tree, page_start, page_end,
3370                                      &cached_state, GFP_NOFS);
3371                 goto out_unlock;
3372         }
3373
3374         ret = 0;
3375         if (offset != PAGE_CACHE_SIZE) {
3376                 kaddr = kmap(page);
3377                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3378                 flush_dcache_page(page);
3379                 kunmap(page);
3380         }
3381         ClearPageChecked(page);
3382         set_page_dirty(page);
3383         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3384                              GFP_NOFS);
3385
3386 out_unlock:
3387         if (ret)
3388                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3389         unlock_page(page);
3390         page_cache_release(page);
3391 out:
3392         return ret;
3393 }
3394
3395 /*
3396  * This function puts in dummy file extents for the area we're creating a hole
3397  * for.  So if we are truncating this file to a larger size we need to insert
3398  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3399  * the range between oldsize and size
3400  */
3401 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3402 {
3403         struct btrfs_trans_handle *trans;
3404         struct btrfs_root *root = BTRFS_I(inode)->root;
3405         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3406         struct extent_map *em = NULL;
3407         struct extent_state *cached_state = NULL;
3408         u64 mask = root->sectorsize - 1;
3409         u64 hole_start = (oldsize + mask) & ~mask;
3410         u64 block_end = (size + mask) & ~mask;
3411         u64 last_byte;
3412         u64 cur_offset;
3413         u64 hole_size;
3414         int err = 0;
3415
3416         if (size <= hole_start)
3417                 return 0;
3418
3419         while (1) {
3420                 struct btrfs_ordered_extent *ordered;
3421                 btrfs_wait_ordered_range(inode, hole_start,
3422                                          block_end - hole_start);
3423                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3424                                  &cached_state, GFP_NOFS);
3425                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3426                 if (!ordered)
3427                         break;
3428                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3429                                      &cached_state, GFP_NOFS);
3430                 btrfs_put_ordered_extent(ordered);
3431         }
3432
3433         cur_offset = hole_start;
3434         while (1) {
3435                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3436                                 block_end - cur_offset, 0);
3437                 BUG_ON(IS_ERR_OR_NULL(em));
3438                 last_byte = min(extent_map_end(em), block_end);
3439                 last_byte = (last_byte + mask) & ~mask;
3440                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3441                         u64 hint_byte = 0;
3442                         hole_size = last_byte - cur_offset;
3443
3444                         trans = btrfs_start_transaction(root, 2);
3445                         if (IS_ERR(trans)) {
3446                                 err = PTR_ERR(trans);
3447                                 break;
3448                         }
3449
3450                         err = btrfs_drop_extents(trans, inode, cur_offset,
3451                                                  cur_offset + hole_size,
3452                                                  &hint_byte, 1);
3453                         if (err) {
3454                                 btrfs_end_transaction(trans, root);
3455                                 break;
3456                         }
3457
3458                         err = btrfs_insert_file_extent(trans, root,
3459                                         btrfs_ino(inode), cur_offset, 0,
3460                                         0, hole_size, 0, hole_size,
3461                                         0, 0, 0);
3462                         if (err) {
3463                                 btrfs_end_transaction(trans, root);
3464                                 break;
3465                         }
3466
3467                         btrfs_drop_extent_cache(inode, hole_start,
3468                                         last_byte - 1, 0);
3469
3470                         btrfs_end_transaction(trans, root);
3471                 }
3472                 free_extent_map(em);
3473                 em = NULL;
3474                 cur_offset = last_byte;
3475                 if (cur_offset >= block_end)
3476                         break;
3477         }
3478
3479         free_extent_map(em);
3480         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3481                              GFP_NOFS);
3482         return err;
3483 }
3484
3485 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3486 {
3487         loff_t oldsize = i_size_read(inode);
3488         int ret;
3489
3490         if (newsize == oldsize)
3491                 return 0;
3492
3493         if (newsize > oldsize) {
3494                 i_size_write(inode, newsize);
3495                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3496                 truncate_pagecache(inode, oldsize, newsize);
3497                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3498                 if (ret) {
3499                         btrfs_setsize(inode, oldsize);
3500                         return ret;
3501                 }
3502
3503                 mark_inode_dirty(inode);
3504         } else {
3505
3506                 /*
3507                  * We're truncating a file that used to have good data down to
3508                  * zero. Make sure it gets into the ordered flush list so that
3509                  * any new writes get down to disk quickly.
3510                  */
3511                 if (newsize == 0)
3512                         BTRFS_I(inode)->ordered_data_close = 1;
3513
3514                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3515                 truncate_setsize(inode, newsize);
3516                 ret = btrfs_truncate(inode);
3517         }
3518
3519         return ret;
3520 }
3521
3522 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3523 {
3524         struct inode *inode = dentry->d_inode;
3525         struct btrfs_root *root = BTRFS_I(inode)->root;
3526         int err;
3527
3528         if (btrfs_root_readonly(root))
3529                 return -EROFS;
3530
3531         err = inode_change_ok(inode, attr);
3532         if (err)
3533                 return err;
3534
3535         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3536                 err = btrfs_setsize(inode, attr->ia_size);
3537                 if (err)
3538                         return err;
3539         }
3540
3541         if (attr->ia_valid) {
3542                 setattr_copy(inode, attr);
3543                 mark_inode_dirty(inode);
3544
3545                 if (attr->ia_valid & ATTR_MODE)
3546                         err = btrfs_acl_chmod(inode);
3547         }
3548
3549         return err;
3550 }
3551
3552 void btrfs_evict_inode(struct inode *inode)
3553 {
3554         struct btrfs_trans_handle *trans;
3555         struct btrfs_root *root = BTRFS_I(inode)->root;
3556         struct btrfs_block_rsv *rsv, *global_rsv;
3557         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3558         unsigned long nr;
3559         int ret;
3560
3561         trace_btrfs_inode_evict(inode);
3562
3563         truncate_inode_pages(&inode->i_data, 0);
3564         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3565                                btrfs_is_free_space_inode(root, inode)))
3566                 goto no_delete;
3567
3568         if (is_bad_inode(inode)) {
3569                 btrfs_orphan_del(NULL, inode);
3570                 goto no_delete;
3571         }
3572         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3573         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3574
3575         if (root->fs_info->log_root_recovering) {
3576                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3577                 goto no_delete;
3578         }
3579
3580         if (inode->i_nlink > 0) {
3581                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3582                 goto no_delete;
3583         }
3584
3585         rsv = btrfs_alloc_block_rsv(root);
3586         if (!rsv) {
3587                 btrfs_orphan_del(NULL, inode);
3588                 goto no_delete;
3589         }
3590         rsv->size = min_size;
3591         global_rsv = &root->fs_info->global_block_rsv;
3592
3593         btrfs_i_size_write(inode, 0);
3594
3595         /*
3596          * This is a bit simpler than btrfs_truncate since
3597          *
3598          * 1) We've already reserved our space for our orphan item in the
3599          *    unlink.
3600          * 2) We're going to delete the inode item, so we don't need to update
3601          *    it at all.
3602          *
3603          * So we just need to reserve some slack space in case we add bytes when
3604          * doing the truncate.
3605          */
3606         while (1) {
3607                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
3608
3609                 /*
3610                  * Try and steal from the global reserve since we will
3611                  * likely not use this space anyway, we want to try as
3612                  * hard as possible to get this to work.
3613                  */
3614                 if (ret)
3615                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3616
3617                 if (ret) {
3618                         printk(KERN_WARNING "Could not get space for a "
3619                                "delete, will truncate on mount %d\n", ret);
3620                         btrfs_orphan_del(NULL, inode);
3621                         btrfs_free_block_rsv(root, rsv);
3622                         goto no_delete;
3623                 }
3624
3625                 trans = btrfs_start_transaction(root, 0);
3626                 if (IS_ERR(trans)) {
3627                         btrfs_orphan_del(NULL, inode);
3628                         btrfs_free_block_rsv(root, rsv);
3629                         goto no_delete;
3630                 }
3631
3632                 trans->block_rsv = rsv;
3633
3634                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3635                 if (ret != -EAGAIN)
3636                         break;
3637
3638                 nr = trans->blocks_used;
3639                 btrfs_end_transaction(trans, root);
3640                 trans = NULL;
3641                 btrfs_btree_balance_dirty(root, nr);
3642         }
3643
3644         btrfs_free_block_rsv(root, rsv);
3645
3646         if (ret == 0) {
3647                 trans->block_rsv = root->orphan_block_rsv;
3648                 ret = btrfs_orphan_del(trans, inode);
3649                 BUG_ON(ret);
3650         }
3651
3652         trans->block_rsv = &root->fs_info->trans_block_rsv;
3653         if (!(root == root->fs_info->tree_root ||
3654               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3655                 btrfs_return_ino(root, btrfs_ino(inode));
3656
3657         nr = trans->blocks_used;
3658         btrfs_end_transaction(trans, root);
3659         btrfs_btree_balance_dirty(root, nr);
3660 no_delete:
3661         end_writeback(inode);
3662         return;
3663 }
3664
3665 /*
3666  * this returns the key found in the dir entry in the location pointer.
3667  * If no dir entries were found, location->objectid is 0.
3668  */
3669 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3670                                struct btrfs_key *location)
3671 {
3672         const char *name = dentry->d_name.name;
3673         int namelen = dentry->d_name.len;
3674         struct btrfs_dir_item *di;
3675         struct btrfs_path *path;
3676         struct btrfs_root *root = BTRFS_I(dir)->root;
3677         int ret = 0;
3678
3679         path = btrfs_alloc_path();
3680         if (!path)
3681                 return -ENOMEM;
3682
3683         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3684                                     namelen, 0);
3685         if (IS_ERR(di))
3686                 ret = PTR_ERR(di);
3687
3688         if (IS_ERR_OR_NULL(di))
3689                 goto out_err;
3690
3691         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3692 out:
3693         btrfs_free_path(path);
3694         return ret;
3695 out_err:
3696         location->objectid = 0;
3697         goto out;
3698 }
3699
3700 /*
3701  * when we hit a tree root in a directory, the btrfs part of the inode
3702  * needs to be changed to reflect the root directory of the tree root.  This
3703  * is kind of like crossing a mount point.
3704  */
3705 static int fixup_tree_root_location(struct btrfs_root *root,
3706                                     struct inode *dir,
3707                                     struct dentry *dentry,
3708                                     struct btrfs_key *location,
3709                                     struct btrfs_root **sub_root)
3710 {
3711         struct btrfs_path *path;
3712         struct btrfs_root *new_root;
3713         struct btrfs_root_ref *ref;
3714         struct extent_buffer *leaf;
3715         int ret;
3716         int err = 0;
3717
3718         path = btrfs_alloc_path();
3719         if (!path) {
3720                 err = -ENOMEM;
3721                 goto out;
3722         }
3723
3724         err = -ENOENT;
3725         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3726                                   BTRFS_I(dir)->root->root_key.objectid,
3727                                   location->objectid);
3728         if (ret) {
3729                 if (ret < 0)
3730                         err = ret;
3731                 goto out;
3732         }
3733
3734         leaf = path->nodes[0];
3735         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3736         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3737             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3738                 goto out;
3739
3740         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3741                                    (unsigned long)(ref + 1),
3742                                    dentry->d_name.len);
3743         if (ret)
3744                 goto out;
3745
3746         btrfs_release_path(path);
3747
3748         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3749         if (IS_ERR(new_root)) {
3750                 err = PTR_ERR(new_root);
3751                 goto out;
3752         }
3753
3754         if (btrfs_root_refs(&new_root->root_item) == 0) {
3755                 err = -ENOENT;
3756                 goto out;
3757         }
3758
3759         *sub_root = new_root;
3760         location->objectid = btrfs_root_dirid(&new_root->root_item);
3761         location->type = BTRFS_INODE_ITEM_KEY;
3762         location->offset = 0;
3763         err = 0;
3764 out:
3765         btrfs_free_path(path);
3766         return err;
3767 }
3768
3769 static void inode_tree_add(struct inode *inode)
3770 {
3771         struct btrfs_root *root = BTRFS_I(inode)->root;
3772         struct btrfs_inode *entry;
3773         struct rb_node **p;
3774         struct rb_node *parent;
3775         u64 ino = btrfs_ino(inode);
3776 again:
3777         p = &root->inode_tree.rb_node;
3778         parent = NULL;
3779
3780         if (inode_unhashed(inode))
3781                 return;
3782
3783         spin_lock(&root->inode_lock);
3784         while (*p) {
3785                 parent = *p;
3786                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3787
3788                 if (ino < btrfs_ino(&entry->vfs_inode))
3789                         p = &parent->rb_left;
3790                 else if (ino > btrfs_ino(&entry->vfs_inode))
3791                         p = &parent->rb_right;
3792                 else {
3793                         WARN_ON(!(entry->vfs_inode.i_state &
3794                                   (I_WILL_FREE | I_FREEING)));
3795                         rb_erase(parent, &root->inode_tree);
3796                         RB_CLEAR_NODE(parent);
3797                         spin_unlock(&root->inode_lock);
3798                         goto again;
3799                 }
3800         }
3801         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3802         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3803         spin_unlock(&root->inode_lock);
3804 }
3805
3806 static void inode_tree_del(struct inode *inode)
3807 {
3808         struct btrfs_root *root = BTRFS_I(inode)->root;
3809         int empty = 0;
3810
3811         spin_lock(&root->inode_lock);
3812         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3813                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3814                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3815                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3816         }
3817         spin_unlock(&root->inode_lock);
3818
3819         /*
3820          * Free space cache has inodes in the tree root, but the tree root has a
3821          * root_refs of 0, so this could end up dropping the tree root as a
3822          * snapshot, so we need the extra !root->fs_info->tree_root check to
3823          * make sure we don't drop it.
3824          */
3825         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3826             root != root->fs_info->tree_root) {
3827                 synchronize_srcu(&root->fs_info->subvol_srcu);
3828                 spin_lock(&root->inode_lock);
3829                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3830                 spin_unlock(&root->inode_lock);
3831                 if (empty)
3832                         btrfs_add_dead_root(root);
3833         }
3834 }
3835
3836 int btrfs_invalidate_inodes(struct btrfs_root *root)
3837 {
3838         struct rb_node *node;
3839         struct rb_node *prev;
3840         struct btrfs_inode *entry;
3841         struct inode *inode;
3842         u64 objectid = 0;
3843
3844         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3845
3846         spin_lock(&root->inode_lock);
3847 again:
3848         node = root->inode_tree.rb_node;
3849         prev = NULL;
3850         while (node) {
3851                 prev = node;
3852                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3853
3854                 if (objectid < btrfs_ino(&entry->vfs_inode))
3855                         node = node->rb_left;
3856                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3857                         node = node->rb_right;
3858                 else
3859                         break;
3860         }
3861         if (!node) {
3862                 while (prev) {
3863                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3864                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3865                                 node = prev;
3866                                 break;
3867                         }
3868                         prev = rb_next(prev);
3869                 }
3870         }
3871         while (node) {
3872                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3873                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3874                 inode = igrab(&entry->vfs_inode);
3875                 if (inode) {
3876                         spin_unlock(&root->inode_lock);
3877                         if (atomic_read(&inode->i_count) > 1)
3878                                 d_prune_aliases(inode);
3879                         /*
3880                          * btrfs_drop_inode will have it removed from
3881                          * the inode cache when its usage count
3882                          * hits zero.
3883                          */
3884                         iput(inode);
3885                         cond_resched();
3886                         spin_lock(&root->inode_lock);
3887                         goto again;
3888                 }
3889
3890                 if (cond_resched_lock(&root->inode_lock))
3891                         goto again;
3892
3893                 node = rb_next(node);
3894         }
3895         spin_unlock(&root->inode_lock);
3896         return 0;
3897 }
3898
3899 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3900 {
3901         struct btrfs_iget_args *args = p;
3902         inode->i_ino = args->ino;
3903         BTRFS_I(inode)->root = args->root;
3904         btrfs_set_inode_space_info(args->root, inode);
3905         return 0;
3906 }
3907
3908 static int btrfs_find_actor(struct inode *inode, void *opaque)
3909 {
3910         struct btrfs_iget_args *args = opaque;
3911         return args->ino == btrfs_ino(inode) &&
3912                 args->root == BTRFS_I(inode)->root;
3913 }
3914
3915 static struct inode *btrfs_iget_locked(struct super_block *s,
3916                                        u64 objectid,
3917                                        struct btrfs_root *root)
3918 {
3919         struct inode *inode;
3920         struct btrfs_iget_args args;
3921         args.ino = objectid;
3922         args.root = root;
3923
3924         inode = iget5_locked(s, objectid, btrfs_find_actor,
3925                              btrfs_init_locked_inode,
3926                              (void *)&args);
3927         return inode;
3928 }
3929
3930 /* Get an inode object given its location and corresponding root.
3931  * Returns in *is_new if the inode was read from disk
3932  */
3933 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3934                          struct btrfs_root *root, int *new)
3935 {
3936         struct inode *inode;
3937
3938         inode = btrfs_iget_locked(s, location->objectid, root);
3939         if (!inode)
3940                 return ERR_PTR(-ENOMEM);
3941
3942         if (inode->i_state & I_NEW) {
3943                 BTRFS_I(inode)->root = root;
3944                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3945                 btrfs_read_locked_inode(inode);
3946                 if (!is_bad_inode(inode)) {
3947                         inode_tree_add(inode);
3948                         unlock_new_inode(inode);
3949                         if (new)
3950                                 *new = 1;
3951                 } else {
3952                         unlock_new_inode(inode);
3953                         iput(inode);
3954                         inode = ERR_PTR(-ESTALE);
3955                 }
3956         }
3957
3958         return inode;
3959 }
3960
3961 static struct inode *new_simple_dir(struct super_block *s,
3962                                     struct btrfs_key *key,
3963                                     struct btrfs_root *root)
3964 {
3965         struct inode *inode = new_inode(s);
3966
3967         if (!inode)
3968                 return ERR_PTR(-ENOMEM);
3969
3970         BTRFS_I(inode)->root = root;
3971         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3972         BTRFS_I(inode)->dummy_inode = 1;
3973
3974         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3975         inode->i_op = &simple_dir_inode_operations;
3976         inode->i_fop = &simple_dir_operations;
3977         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3978         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3979
3980         return inode;
3981 }
3982
3983 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3984 {
3985         struct inode *inode;
3986         struct btrfs_root *root = BTRFS_I(dir)->root;
3987         struct btrfs_root *sub_root = root;
3988         struct btrfs_key location;
3989         int index;
3990         int ret = 0;
3991
3992         if (dentry->d_name.len > BTRFS_NAME_LEN)
3993                 return ERR_PTR(-ENAMETOOLONG);
3994
3995         if (unlikely(d_need_lookup(dentry))) {
3996                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3997                 kfree(dentry->d_fsdata);
3998                 dentry->d_fsdata = NULL;
3999                 /* This thing is hashed, drop it for now */
4000                 d_drop(dentry);
4001         } else {
4002                 ret = btrfs_inode_by_name(dir, dentry, &location);
4003         }
4004
4005         if (ret < 0)
4006                 return ERR_PTR(ret);
4007
4008         if (location.objectid == 0)
4009                 return NULL;
4010
4011         if (location.type == BTRFS_INODE_ITEM_KEY) {
4012                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4013                 return inode;
4014         }
4015
4016         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4017
4018         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4019         ret = fixup_tree_root_location(root, dir, dentry,
4020                                        &location, &sub_root);
4021         if (ret < 0) {
4022                 if (ret != -ENOENT)
4023                         inode = ERR_PTR(ret);
4024                 else
4025                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4026         } else {
4027                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4028         }
4029         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4030
4031         if (!IS_ERR(inode) && root != sub_root) {
4032                 down_read(&root->fs_info->cleanup_work_sem);
4033                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4034                         ret = btrfs_orphan_cleanup(sub_root);
4035                 up_read(&root->fs_info->cleanup_work_sem);
4036                 if (ret)
4037                         inode = ERR_PTR(ret);
4038         }
4039
4040         return inode;
4041 }
4042
4043 static int btrfs_dentry_delete(const struct dentry *dentry)
4044 {
4045         struct btrfs_root *root;
4046
4047         if (!dentry->d_inode && !IS_ROOT(dentry))
4048                 dentry = dentry->d_parent;
4049
4050         if (dentry->d_inode) {
4051                 root = BTRFS_I(dentry->d_inode)->root;
4052                 if (btrfs_root_refs(&root->root_item) == 0)
4053                         return 1;
4054         }
4055         return 0;
4056 }
4057
4058 static void btrfs_dentry_release(struct dentry *dentry)
4059 {
4060         if (dentry->d_fsdata)
4061                 kfree(dentry->d_fsdata);
4062 }
4063
4064 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4065                                    struct nameidata *nd)
4066 {
4067         struct dentry *ret;
4068
4069         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4070         if (unlikely(d_need_lookup(dentry))) {
4071                 spin_lock(&dentry->d_lock);
4072                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4073                 spin_unlock(&dentry->d_lock);
4074         }
4075         return ret;
4076 }
4077
4078 unsigned char btrfs_filetype_table[] = {
4079         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4080 };
4081
4082 static int btrfs_real_readdir(struct file *filp, void *dirent,
4083                               filldir_t filldir)
4084 {
4085         struct inode *inode = filp->f_dentry->d_inode;
4086         struct btrfs_root *root = BTRFS_I(inode)->root;
4087         struct btrfs_item *item;
4088         struct btrfs_dir_item *di;
4089         struct btrfs_key key;
4090         struct btrfs_key found_key;
4091         struct btrfs_path *path;
4092         struct list_head ins_list;
4093         struct list_head del_list;
4094         struct qstr q;
4095         int ret;
4096         struct extent_buffer *leaf;
4097         int slot;
4098         unsigned char d_type;
4099         int over = 0;
4100         u32 di_cur;
4101         u32 di_total;
4102         u32 di_len;
4103         int key_type = BTRFS_DIR_INDEX_KEY;
4104         char tmp_name[32];
4105         char *name_ptr;
4106         int name_len;
4107         int is_curr = 0;        /* filp->f_pos points to the current index? */
4108
4109         /* FIXME, use a real flag for deciding about the key type */
4110         if (root->fs_info->tree_root == root)
4111                 key_type = BTRFS_DIR_ITEM_KEY;
4112
4113         /* special case for "." */
4114         if (filp->f_pos == 0) {
4115                 over = filldir(dirent, ".", 1,
4116                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4117                 if (over)
4118                         return 0;
4119                 filp->f_pos = 1;
4120         }
4121         /* special case for .., just use the back ref */
4122         if (filp->f_pos == 1) {
4123                 u64 pino = parent_ino(filp->f_path.dentry);
4124                 over = filldir(dirent, "..", 2,
4125                                filp->f_pos, pino, DT_DIR);
4126                 if (over)
4127                         return 0;
4128                 filp->f_pos = 2;
4129         }
4130         path = btrfs_alloc_path();
4131         if (!path)
4132                 return -ENOMEM;
4133
4134         path->reada = 1;
4135
4136         if (key_type == BTRFS_DIR_INDEX_KEY) {
4137                 INIT_LIST_HEAD(&ins_list);
4138                 INIT_LIST_HEAD(&del_list);
4139                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4140         }
4141
4142         btrfs_set_key_type(&key, key_type);
4143         key.offset = filp->f_pos;
4144         key.objectid = btrfs_ino(inode);
4145
4146         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4147         if (ret < 0)
4148                 goto err;
4149
4150         while (1) {
4151                 leaf = path->nodes[0];
4152                 slot = path->slots[0];
4153                 if (slot >= btrfs_header_nritems(leaf)) {
4154                         ret = btrfs_next_leaf(root, path);
4155                         if (ret < 0)
4156                                 goto err;
4157                         else if (ret > 0)
4158                                 break;
4159                         continue;
4160                 }
4161
4162                 item = btrfs_item_nr(leaf, slot);
4163                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4164
4165                 if (found_key.objectid != key.objectid)
4166                         break;
4167                 if (btrfs_key_type(&found_key) != key_type)
4168                         break;
4169                 if (found_key.offset < filp->f_pos)
4170                         goto next;
4171                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4172                     btrfs_should_delete_dir_index(&del_list,
4173                                                   found_key.offset))
4174                         goto next;
4175
4176                 filp->f_pos = found_key.offset;
4177                 is_curr = 1;
4178
4179                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4180                 di_cur = 0;
4181                 di_total = btrfs_item_size(leaf, item);
4182
4183                 while (di_cur < di_total) {
4184                         struct btrfs_key location;
4185                         struct dentry *tmp;
4186
4187                         if (verify_dir_item(root, leaf, di))
4188                                 break;
4189
4190                         name_len = btrfs_dir_name_len(leaf, di);
4191                         if (name_len <= sizeof(tmp_name)) {
4192                                 name_ptr = tmp_name;
4193                         } else {
4194                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4195                                 if (!name_ptr) {
4196                                         ret = -ENOMEM;
4197                                         goto err;
4198                                 }
4199                         }
4200                         read_extent_buffer(leaf, name_ptr,
4201                                            (unsigned long)(di + 1), name_len);
4202
4203                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4204                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4205
4206                         q.name = name_ptr;
4207                         q.len = name_len;
4208                         q.hash = full_name_hash(q.name, q.len);
4209                         tmp = d_lookup(filp->f_dentry, &q);
4210                         if (!tmp) {
4211                                 struct btrfs_key *newkey;
4212
4213                                 newkey = kzalloc(sizeof(struct btrfs_key),
4214                                                  GFP_NOFS);
4215                                 if (!newkey)
4216                                         goto no_dentry;
4217                                 tmp = d_alloc(filp->f_dentry, &q);
4218                                 if (!tmp) {
4219                                         kfree(newkey);
4220                                         dput(tmp);
4221                                         goto no_dentry;
4222                                 }
4223                                 memcpy(newkey, &location,
4224                                        sizeof(struct btrfs_key));
4225                                 tmp->d_fsdata = newkey;
4226                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4227                                 d_rehash(tmp);
4228                                 dput(tmp);
4229                         } else {
4230                                 dput(tmp);
4231                         }
4232 no_dentry:
4233                         /* is this a reference to our own snapshot? If so
4234                          * skip it
4235                          */
4236                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4237                             location.objectid == root->root_key.objectid) {
4238                                 over = 0;
4239                                 goto skip;
4240                         }
4241                         over = filldir(dirent, name_ptr, name_len,
4242                                        found_key.offset, location.objectid,
4243                                        d_type);
4244
4245 skip:
4246                         if (name_ptr != tmp_name)
4247                                 kfree(name_ptr);
4248
4249                         if (over)
4250                                 goto nopos;
4251                         di_len = btrfs_dir_name_len(leaf, di) +
4252                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4253                         di_cur += di_len;
4254                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4255                 }
4256 next:
4257                 path->slots[0]++;
4258         }
4259
4260         if (key_type == BTRFS_DIR_INDEX_KEY) {
4261                 if (is_curr)
4262                         filp->f_pos++;
4263                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4264                                                       &ins_list);
4265                 if (ret)
4266                         goto nopos;
4267         }
4268
4269         /* Reached end of directory/root. Bump pos past the last item. */
4270         if (key_type == BTRFS_DIR_INDEX_KEY)
4271                 /*
4272                  * 32-bit glibc will use getdents64, but then strtol -
4273                  * so the last number we can serve is this.
4274                  */
4275                 filp->f_pos = 0x7fffffff;
4276         else
4277                 filp->f_pos++;
4278 nopos:
4279         ret = 0;
4280 err:
4281         if (key_type == BTRFS_DIR_INDEX_KEY)
4282                 btrfs_put_delayed_items(&ins_list, &del_list);
4283         btrfs_free_path(path);
4284         return ret;
4285 }
4286
4287 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4288 {
4289         struct btrfs_root *root = BTRFS_I(inode)->root;
4290         struct btrfs_trans_handle *trans;
4291         int ret = 0;
4292         bool nolock = false;
4293
4294         if (BTRFS_I(inode)->dummy_inode)
4295                 return 0;
4296
4297         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4298                 nolock = true;
4299
4300         if (wbc->sync_mode == WB_SYNC_ALL) {
4301                 if (nolock)
4302                         trans = btrfs_join_transaction_nolock(root);
4303                 else
4304                         trans = btrfs_join_transaction(root);
4305                 if (IS_ERR(trans))
4306                         return PTR_ERR(trans);
4307                 if (nolock)
4308                         ret = btrfs_end_transaction_nolock(trans, root);
4309                 else
4310                         ret = btrfs_commit_transaction(trans, root);
4311         }
4312         return ret;
4313 }
4314
4315 /*
4316  * This is somewhat expensive, updating the tree every time the
4317  * inode changes.  But, it is most likely to find the inode in cache.
4318  * FIXME, needs more benchmarking...there are no reasons other than performance
4319  * to keep or drop this code.
4320  */
4321 void btrfs_dirty_inode(struct inode *inode, int flags)
4322 {
4323         struct btrfs_root *root = BTRFS_I(inode)->root;
4324         struct btrfs_trans_handle *trans;
4325         int ret;
4326
4327         if (BTRFS_I(inode)->dummy_inode)
4328                 return;
4329
4330         trans = btrfs_join_transaction(root);
4331         BUG_ON(IS_ERR(trans));
4332
4333         ret = btrfs_update_inode(trans, root, inode);
4334         if (ret && ret == -ENOSPC) {
4335                 /* whoops, lets try again with the full transaction */
4336                 btrfs_end_transaction(trans, root);
4337                 trans = btrfs_start_transaction(root, 1);
4338                 if (IS_ERR(trans)) {
4339                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4340                                        "dirty  inode %llu error %ld\n",
4341                                        (unsigned long long)btrfs_ino(inode),
4342                                        PTR_ERR(trans));
4343                         return;
4344                 }
4345
4346                 ret = btrfs_update_inode(trans, root, inode);
4347                 if (ret) {
4348                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4349                                        "dirty  inode %llu error %d\n",
4350                                        (unsigned long long)btrfs_ino(inode),
4351                                        ret);
4352                 }
4353         }
4354         btrfs_end_transaction(trans, root);
4355         if (BTRFS_I(inode)->delayed_node)
4356                 btrfs_balance_delayed_items(root);
4357 }
4358
4359 /*
4360  * find the highest existing sequence number in a directory
4361  * and then set the in-memory index_cnt variable to reflect
4362  * free sequence numbers
4363  */
4364 static int btrfs_set_inode_index_count(struct inode *inode)
4365 {
4366         struct btrfs_root *root = BTRFS_I(inode)->root;
4367         struct btrfs_key key, found_key;
4368         struct btrfs_path *path;
4369         struct extent_buffer *leaf;
4370         int ret;
4371
4372         key.objectid = btrfs_ino(inode);
4373         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4374         key.offset = (u64)-1;
4375
4376         path = btrfs_alloc_path();
4377         if (!path)
4378                 return -ENOMEM;
4379
4380         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4381         if (ret < 0)
4382                 goto out;
4383         /* FIXME: we should be able to handle this */
4384         if (ret == 0)
4385                 goto out;
4386         ret = 0;
4387
4388         /*
4389          * MAGIC NUMBER EXPLANATION:
4390          * since we search a directory based on f_pos we have to start at 2
4391          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4392          * else has to start at 2
4393          */
4394         if (path->slots[0] == 0) {
4395                 BTRFS_I(inode)->index_cnt = 2;
4396                 goto out;
4397         }
4398
4399         path->slots[0]--;
4400
4401         leaf = path->nodes[0];
4402         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4403
4404         if (found_key.objectid != btrfs_ino(inode) ||
4405             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4406                 BTRFS_I(inode)->index_cnt = 2;
4407                 goto out;
4408         }
4409
4410         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4411 out:
4412         btrfs_free_path(path);
4413         return ret;
4414 }
4415
4416 /*
4417  * helper to find a free sequence number in a given directory.  This current
4418  * code is very simple, later versions will do smarter things in the btree
4419  */
4420 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4421 {
4422         int ret = 0;
4423
4424         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4425                 ret = btrfs_inode_delayed_dir_index_count(dir);
4426                 if (ret) {
4427                         ret = btrfs_set_inode_index_count(dir);
4428                         if (ret)
4429                                 return ret;
4430                 }
4431         }
4432
4433         *index = BTRFS_I(dir)->index_cnt;
4434         BTRFS_I(dir)->index_cnt++;
4435
4436         return ret;
4437 }
4438
4439 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4440                                      struct btrfs_root *root,
4441                                      struct inode *dir,
4442                                      const char *name, int name_len,
4443                                      u64 ref_objectid, u64 objectid, int mode,
4444                                      u64 *index)
4445 {
4446         struct inode *inode;
4447         struct btrfs_inode_item *inode_item;
4448         struct btrfs_key *location;
4449         struct btrfs_path *path;
4450         struct btrfs_inode_ref *ref;
4451         struct btrfs_key key[2];
4452         u32 sizes[2];
4453         unsigned long ptr;
4454         int ret;
4455         int owner;
4456
4457         path = btrfs_alloc_path();
4458         if (!path)
4459                 return ERR_PTR(-ENOMEM);
4460
4461         inode = new_inode(root->fs_info->sb);
4462         if (!inode) {
4463                 btrfs_free_path(path);
4464                 return ERR_PTR(-ENOMEM);
4465         }
4466
4467         /*
4468          * we have to initialize this early, so we can reclaim the inode
4469          * number if we fail afterwards in this function.
4470          */
4471         inode->i_ino = objectid;
4472
4473         if (dir) {
4474                 trace_btrfs_inode_request(dir);
4475
4476                 ret = btrfs_set_inode_index(dir, index);
4477                 if (ret) {
4478                         btrfs_free_path(path);
4479                         iput(inode);
4480                         return ERR_PTR(ret);
4481                 }
4482         }
4483         /*
4484          * index_cnt is ignored for everything but a dir,
4485          * btrfs_get_inode_index_count has an explanation for the magic
4486          * number
4487          */
4488         BTRFS_I(inode)->index_cnt = 2;
4489         BTRFS_I(inode)->root = root;
4490         BTRFS_I(inode)->generation = trans->transid;
4491         inode->i_generation = BTRFS_I(inode)->generation;
4492         btrfs_set_inode_space_info(root, inode);
4493
4494         if (S_ISDIR(mode))
4495                 owner = 0;
4496         else
4497                 owner = 1;
4498
4499         key[0].objectid = objectid;
4500         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4501         key[0].offset = 0;
4502
4503         key[1].objectid = objectid;
4504         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4505         key[1].offset = ref_objectid;
4506
4507         sizes[0] = sizeof(struct btrfs_inode_item);
4508         sizes[1] = name_len + sizeof(*ref);
4509
4510         path->leave_spinning = 1;
4511         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4512         if (ret != 0)
4513                 goto fail;
4514
4515         inode_init_owner(inode, dir, mode);
4516         inode_set_bytes(inode, 0);
4517         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4518         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4519                                   struct btrfs_inode_item);
4520         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4521
4522         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4523                              struct btrfs_inode_ref);
4524         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4525         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4526         ptr = (unsigned long)(ref + 1);
4527         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4528
4529         btrfs_mark_buffer_dirty(path->nodes[0]);
4530         btrfs_free_path(path);
4531
4532         location = &BTRFS_I(inode)->location;
4533         location->objectid = objectid;
4534         location->offset = 0;
4535         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4536
4537         btrfs_inherit_iflags(inode, dir);
4538
4539         if (S_ISREG(mode)) {
4540                 if (btrfs_test_opt(root, NODATASUM))
4541                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4542                 if (btrfs_test_opt(root, NODATACOW) ||
4543                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4544                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4545         }
4546
4547         insert_inode_hash(inode);
4548         inode_tree_add(inode);
4549
4550         trace_btrfs_inode_new(inode);
4551         btrfs_set_inode_last_trans(trans, inode);
4552
4553         return inode;
4554 fail:
4555         if (dir)
4556                 BTRFS_I(dir)->index_cnt--;
4557         btrfs_free_path(path);
4558         iput(inode);
4559         return ERR_PTR(ret);
4560 }
4561
4562 static inline u8 btrfs_inode_type(struct inode *inode)
4563 {
4564         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4565 }
4566
4567 /*
4568  * utility function to add 'inode' into 'parent_inode' with
4569  * a give name and a given sequence number.
4570  * if 'add_backref' is true, also insert a backref from the
4571  * inode to the parent directory.
4572  */
4573 int btrfs_add_link(struct btrfs_trans_handle *trans,
4574                    struct inode *parent_inode, struct inode *inode,
4575                    const char *name, int name_len, int add_backref, u64 index)
4576 {
4577         int ret = 0;
4578         struct btrfs_key key;
4579         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4580         u64 ino = btrfs_ino(inode);
4581         u64 parent_ino = btrfs_ino(parent_inode);
4582
4583         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4584                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4585         } else {
4586                 key.objectid = ino;
4587                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4588                 key.offset = 0;
4589         }
4590
4591         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4592                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4593                                          key.objectid, root->root_key.objectid,
4594                                          parent_ino, index, name, name_len);
4595         } else if (add_backref) {
4596                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4597                                              parent_ino, index);
4598         }
4599
4600         if (ret == 0) {
4601                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4602                                             parent_inode, &key,
4603                                             btrfs_inode_type(inode), index);
4604                 BUG_ON(ret);
4605
4606                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4607                                    name_len * 2);
4608                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4609                 ret = btrfs_update_inode(trans, root, parent_inode);
4610         }
4611         return ret;
4612 }
4613
4614 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4615                             struct inode *dir, struct dentry *dentry,
4616                             struct inode *inode, int backref, u64 index)
4617 {
4618         int err = btrfs_add_link(trans, dir, inode,
4619                                  dentry->d_name.name, dentry->d_name.len,
4620                                  backref, index);
4621         if (!err) {
4622                 d_instantiate(dentry, inode);
4623                 return 0;
4624         }
4625         if (err > 0)
4626                 err = -EEXIST;
4627         return err;
4628 }
4629
4630 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4631                         int mode, dev_t rdev)
4632 {
4633         struct btrfs_trans_handle *trans;
4634         struct btrfs_root *root = BTRFS_I(dir)->root;
4635         struct inode *inode = NULL;
4636         int err;
4637         int drop_inode = 0;
4638         u64 objectid;
4639         unsigned long nr = 0;
4640         u64 index = 0;
4641
4642         if (!new_valid_dev(rdev))
4643                 return -EINVAL;
4644
4645         /*
4646          * 2 for inode item and ref
4647          * 2 for dir items
4648          * 1 for xattr if selinux is on
4649          */
4650         trans = btrfs_start_transaction(root, 5);
4651         if (IS_ERR(trans))
4652                 return PTR_ERR(trans);
4653
4654         err = btrfs_find_free_ino(root, &objectid);
4655         if (err)
4656                 goto out_unlock;
4657
4658         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4659                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4660                                 mode, &index);
4661         if (IS_ERR(inode)) {
4662                 err = PTR_ERR(inode);
4663                 goto out_unlock;
4664         }
4665
4666         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4667         if (err) {
4668                 drop_inode = 1;
4669                 goto out_unlock;
4670         }
4671
4672         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4673         if (err)
4674                 drop_inode = 1;
4675         else {
4676                 inode->i_op = &btrfs_special_inode_operations;
4677                 init_special_inode(inode, inode->i_mode, rdev);
4678                 btrfs_update_inode(trans, root, inode);
4679         }
4680 out_unlock:
4681         nr = trans->blocks_used;
4682         btrfs_end_transaction_throttle(trans, root);
4683         btrfs_btree_balance_dirty(root, nr);
4684         if (drop_inode) {
4685                 inode_dec_link_count(inode);
4686                 iput(inode);
4687         }
4688         return err;
4689 }
4690
4691 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4692                         int mode, struct nameidata *nd)
4693 {
4694         struct btrfs_trans_handle *trans;
4695         struct btrfs_root *root = BTRFS_I(dir)->root;
4696         struct inode *inode = NULL;
4697         int drop_inode = 0;
4698         int err;
4699         unsigned long nr = 0;
4700         u64 objectid;
4701         u64 index = 0;
4702
4703         /*
4704          * 2 for inode item and ref
4705          * 2 for dir items
4706          * 1 for xattr if selinux is on
4707          */
4708         trans = btrfs_start_transaction(root, 5);
4709         if (IS_ERR(trans))
4710                 return PTR_ERR(trans);
4711
4712         err = btrfs_find_free_ino(root, &objectid);
4713         if (err)
4714                 goto out_unlock;
4715
4716         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4717                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4718                                 mode, &index);
4719         if (IS_ERR(inode)) {
4720                 err = PTR_ERR(inode);
4721                 goto out_unlock;
4722         }
4723
4724         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4725         if (err) {
4726                 drop_inode = 1;
4727                 goto out_unlock;
4728         }
4729
4730         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4731         if (err)
4732                 drop_inode = 1;
4733         else {
4734                 inode->i_mapping->a_ops = &btrfs_aops;
4735                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4736                 inode->i_fop = &btrfs_file_operations;
4737                 inode->i_op = &btrfs_file_inode_operations;
4738                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4739         }
4740 out_unlock:
4741         nr = trans->blocks_used;
4742         btrfs_end_transaction_throttle(trans, root);
4743         if (drop_inode) {
4744                 inode_dec_link_count(inode);
4745                 iput(inode);
4746         }
4747         btrfs_btree_balance_dirty(root, nr);
4748         return err;
4749 }
4750
4751 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4752                       struct dentry *dentry)
4753 {
4754         struct btrfs_trans_handle *trans;
4755         struct btrfs_root *root = BTRFS_I(dir)->root;
4756         struct inode *inode = old_dentry->d_inode;
4757         u64 index;
4758         unsigned long nr = 0;
4759         int err;
4760         int drop_inode = 0;
4761
4762         /* do not allow sys_link's with other subvols of the same device */
4763         if (root->objectid != BTRFS_I(inode)->root->objectid)
4764                 return -EXDEV;
4765
4766         if (inode->i_nlink == ~0U)
4767                 return -EMLINK;
4768
4769         err = btrfs_set_inode_index(dir, &index);
4770         if (err)
4771                 goto fail;
4772
4773         /*
4774          * 2 items for inode and inode ref
4775          * 2 items for dir items
4776          * 1 item for parent inode
4777          */
4778         trans = btrfs_start_transaction(root, 5);
4779         if (IS_ERR(trans)) {
4780                 err = PTR_ERR(trans);
4781                 goto fail;
4782         }
4783
4784         btrfs_inc_nlink(inode);
4785         inode->i_ctime = CURRENT_TIME;
4786         ihold(inode);
4787
4788         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4789
4790         if (err) {
4791                 drop_inode = 1;
4792         } else {
4793                 struct dentry *parent = dentry->d_parent;
4794                 err = btrfs_update_inode(trans, root, inode);
4795                 BUG_ON(err);
4796                 btrfs_log_new_name(trans, inode, NULL, parent);
4797         }
4798
4799         nr = trans->blocks_used;
4800         btrfs_end_transaction_throttle(trans, root);
4801 fail:
4802         if (drop_inode) {
4803                 inode_dec_link_count(inode);
4804                 iput(inode);
4805         }
4806         btrfs_btree_balance_dirty(root, nr);
4807         return err;
4808 }
4809
4810 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4811 {
4812         struct inode *inode = NULL;
4813         struct btrfs_trans_handle *trans;
4814         struct btrfs_root *root = BTRFS_I(dir)->root;
4815         int err = 0;
4816         int drop_on_err = 0;
4817         u64 objectid = 0;
4818         u64 index = 0;
4819         unsigned long nr = 1;
4820
4821         /*
4822          * 2 items for inode and ref
4823          * 2 items for dir items
4824          * 1 for xattr if selinux is on
4825          */
4826         trans = btrfs_start_transaction(root, 5);
4827         if (IS_ERR(trans))
4828                 return PTR_ERR(trans);
4829
4830         err = btrfs_find_free_ino(root, &objectid);
4831         if (err)
4832                 goto out_fail;
4833
4834         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4835                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4836                                 S_IFDIR | mode, &index);
4837         if (IS_ERR(inode)) {
4838                 err = PTR_ERR(inode);
4839                 goto out_fail;
4840         }
4841
4842         drop_on_err = 1;
4843
4844         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4845         if (err)
4846                 goto out_fail;
4847
4848         inode->i_op = &btrfs_dir_inode_operations;
4849         inode->i_fop = &btrfs_dir_file_operations;
4850
4851         btrfs_i_size_write(inode, 0);
4852         err = btrfs_update_inode(trans, root, inode);
4853         if (err)
4854                 goto out_fail;
4855
4856         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4857                              dentry->d_name.len, 0, index);
4858         if (err)
4859                 goto out_fail;
4860
4861         d_instantiate(dentry, inode);
4862         drop_on_err = 0;
4863
4864 out_fail:
4865         nr = trans->blocks_used;
4866         btrfs_end_transaction_throttle(trans, root);
4867         if (drop_on_err)
4868                 iput(inode);
4869         btrfs_btree_balance_dirty(root, nr);
4870         return err;
4871 }
4872
4873 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4874  * and an extent that you want to insert, deal with overlap and insert
4875  * the new extent into the tree.
4876  */
4877 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4878                                 struct extent_map *existing,
4879                                 struct extent_map *em,
4880                                 u64 map_start, u64 map_len)
4881 {
4882         u64 start_diff;
4883
4884         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4885         start_diff = map_start - em->start;
4886         em->start = map_start;
4887         em->len = map_len;
4888         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4889             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4890                 em->block_start += start_diff;
4891                 em->block_len -= start_diff;
4892         }
4893         return add_extent_mapping(em_tree, em);
4894 }
4895
4896 static noinline int uncompress_inline(struct btrfs_path *path,
4897                                       struct inode *inode, struct page *page,
4898                                       size_t pg_offset, u64 extent_offset,
4899                                       struct btrfs_file_extent_item *item)
4900 {
4901         int ret;
4902         struct extent_buffer *leaf = path->nodes[0];
4903         char *tmp;
4904         size_t max_size;
4905         unsigned long inline_size;
4906         unsigned long ptr;
4907         int compress_type;
4908
4909         WARN_ON(pg_offset != 0);
4910         compress_type = btrfs_file_extent_compression(leaf, item);
4911         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4912         inline_size = btrfs_file_extent_inline_item_len(leaf,
4913                                         btrfs_item_nr(leaf, path->slots[0]));
4914         tmp = kmalloc(inline_size, GFP_NOFS);
4915         if (!tmp)
4916                 return -ENOMEM;
4917         ptr = btrfs_file_extent_inline_start(item);
4918
4919         read_extent_buffer(leaf, tmp, ptr, inline_size);
4920
4921         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4922         ret = btrfs_decompress(compress_type, tmp, page,
4923                                extent_offset, inline_size, max_size);
4924         if (ret) {
4925                 char *kaddr = kmap_atomic(page, KM_USER0);
4926                 unsigned long copy_size = min_t(u64,
4927                                   PAGE_CACHE_SIZE - pg_offset,
4928                                   max_size - extent_offset);
4929                 memset(kaddr + pg_offset, 0, copy_size);
4930                 kunmap_atomic(kaddr, KM_USER0);
4931         }
4932         kfree(tmp);
4933         return 0;
4934 }
4935
4936 /*
4937  * a bit scary, this does extent mapping from logical file offset to the disk.
4938  * the ugly parts come from merging extents from the disk with the in-ram
4939  * representation.  This gets more complex because of the data=ordered code,
4940  * where the in-ram extents might be locked pending data=ordered completion.
4941  *
4942  * This also copies inline extents directly into the page.
4943  */
4944
4945 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4946                                     size_t pg_offset, u64 start, u64 len,
4947                                     int create)
4948 {
4949         int ret;
4950         int err = 0;
4951         u64 bytenr;
4952         u64 extent_start = 0;
4953         u64 extent_end = 0;
4954         u64 objectid = btrfs_ino(inode);
4955         u32 found_type;
4956         struct btrfs_path *path = NULL;
4957         struct btrfs_root *root = BTRFS_I(inode)->root;
4958         struct btrfs_file_extent_item *item;
4959         struct extent_buffer *leaf;
4960         struct btrfs_key found_key;
4961         struct extent_map *em = NULL;
4962         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4963         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4964         struct btrfs_trans_handle *trans = NULL;
4965         int compress_type;
4966
4967 again:
4968         read_lock(&em_tree->lock);
4969         em = lookup_extent_mapping(em_tree, start, len);
4970         if (em)
4971                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4972         read_unlock(&em_tree->lock);
4973
4974         if (em) {
4975                 if (em->start > start || em->start + em->len <= start)
4976                         free_extent_map(em);
4977                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4978                         free_extent_map(em);
4979                 else
4980                         goto out;
4981         }
4982         em = alloc_extent_map();
4983         if (!em) {
4984                 err = -ENOMEM;
4985                 goto out;
4986         }
4987         em->bdev = root->fs_info->fs_devices->latest_bdev;
4988         em->start = EXTENT_MAP_HOLE;
4989         em->orig_start = EXTENT_MAP_HOLE;
4990         em->len = (u64)-1;
4991         em->block_len = (u64)-1;
4992
4993         if (!path) {
4994                 path = btrfs_alloc_path();
4995                 if (!path) {
4996                         err = -ENOMEM;
4997                         goto out;
4998                 }
4999                 /*
5000                  * Chances are we'll be called again, so go ahead and do
5001                  * readahead
5002                  */
5003                 path->reada = 1;
5004         }
5005
5006         ret = btrfs_lookup_file_extent(trans, root, path,
5007                                        objectid, start, trans != NULL);
5008         if (ret < 0) {
5009                 err = ret;
5010                 goto out;
5011         }
5012
5013         if (ret != 0) {
5014                 if (path->slots[0] == 0)
5015                         goto not_found;
5016                 path->slots[0]--;
5017         }
5018
5019         leaf = path->nodes[0];
5020         item = btrfs_item_ptr(leaf, path->slots[0],
5021                               struct btrfs_file_extent_item);
5022         /* are we inside the extent that was found? */
5023         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5024         found_type = btrfs_key_type(&found_key);
5025         if (found_key.objectid != objectid ||
5026             found_type != BTRFS_EXTENT_DATA_KEY) {
5027                 goto not_found;
5028         }
5029
5030         found_type = btrfs_file_extent_type(leaf, item);
5031         extent_start = found_key.offset;
5032         compress_type = btrfs_file_extent_compression(leaf, item);
5033         if (found_type == BTRFS_FILE_EXTENT_REG ||
5034             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5035                 extent_end = extent_start +
5036                        btrfs_file_extent_num_bytes(leaf, item);
5037         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5038                 size_t size;
5039                 size = btrfs_file_extent_inline_len(leaf, item);
5040                 extent_end = (extent_start + size + root->sectorsize - 1) &
5041                         ~((u64)root->sectorsize - 1);
5042         }
5043
5044         if (start >= extent_end) {
5045                 path->slots[0]++;
5046                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5047                         ret = btrfs_next_leaf(root, path);
5048                         if (ret < 0) {
5049                                 err = ret;
5050                                 goto out;
5051                         }
5052                         if (ret > 0)
5053                                 goto not_found;
5054                         leaf = path->nodes[0];
5055                 }
5056                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5057                 if (found_key.objectid != objectid ||
5058                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5059                         goto not_found;
5060                 if (start + len <= found_key.offset)
5061                         goto not_found;
5062                 em->start = start;
5063                 em->len = found_key.offset - start;
5064                 goto not_found_em;
5065         }
5066
5067         if (found_type == BTRFS_FILE_EXTENT_REG ||
5068             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5069                 em->start = extent_start;
5070                 em->len = extent_end - extent_start;
5071                 em->orig_start = extent_start -
5072                                  btrfs_file_extent_offset(leaf, item);
5073                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5074                 if (bytenr == 0) {
5075                         em->block_start = EXTENT_MAP_HOLE;
5076                         goto insert;
5077                 }
5078                 if (compress_type != BTRFS_COMPRESS_NONE) {
5079                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5080                         em->compress_type = compress_type;
5081                         em->block_start = bytenr;
5082                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5083                                                                          item);
5084                 } else {
5085                         bytenr += btrfs_file_extent_offset(leaf, item);
5086                         em->block_start = bytenr;
5087                         em->block_len = em->len;
5088                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5089                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5090                 }
5091                 goto insert;
5092         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5093                 unsigned long ptr;
5094                 char *map;
5095                 size_t size;
5096                 size_t extent_offset;
5097                 size_t copy_size;
5098
5099                 em->block_start = EXTENT_MAP_INLINE;
5100                 if (!page || create) {
5101                         em->start = extent_start;
5102                         em->len = extent_end - extent_start;
5103                         goto out;
5104                 }
5105
5106                 size = btrfs_file_extent_inline_len(leaf, item);
5107                 extent_offset = page_offset(page) + pg_offset - extent_start;
5108                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5109                                 size - extent_offset);
5110                 em->start = extent_start + extent_offset;
5111                 em->len = (copy_size + root->sectorsize - 1) &
5112                         ~((u64)root->sectorsize - 1);
5113                 em->orig_start = EXTENT_MAP_INLINE;
5114                 if (compress_type) {
5115                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5116                         em->compress_type = compress_type;
5117                 }
5118                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5119                 if (create == 0 && !PageUptodate(page)) {
5120                         if (btrfs_file_extent_compression(leaf, item) !=
5121                             BTRFS_COMPRESS_NONE) {
5122                                 ret = uncompress_inline(path, inode, page,
5123                                                         pg_offset,
5124                                                         extent_offset, item);
5125                                 BUG_ON(ret);
5126                         } else {
5127                                 map = kmap(page);
5128                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5129                                                    copy_size);
5130                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5131                                         memset(map + pg_offset + copy_size, 0,
5132                                                PAGE_CACHE_SIZE - pg_offset -
5133                                                copy_size);
5134                                 }
5135                                 kunmap(page);
5136                         }
5137                         flush_dcache_page(page);
5138                 } else if (create && PageUptodate(page)) {
5139                         WARN_ON(1);
5140                         if (!trans) {
5141                                 kunmap(page);
5142                                 free_extent_map(em);
5143                                 em = NULL;
5144
5145                                 btrfs_release_path(path);
5146                                 trans = btrfs_join_transaction(root);
5147
5148                                 if (IS_ERR(trans))
5149                                         return ERR_CAST(trans);
5150                                 goto again;
5151                         }
5152                         map = kmap(page);
5153                         write_extent_buffer(leaf, map + pg_offset, ptr,
5154                                             copy_size);
5155                         kunmap(page);
5156                         btrfs_mark_buffer_dirty(leaf);
5157                 }
5158                 set_extent_uptodate(io_tree, em->start,
5159                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5160                 goto insert;
5161         } else {
5162                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5163                 WARN_ON(1);
5164         }
5165 not_found:
5166         em->start = start;
5167         em->len = len;
5168 not_found_em:
5169         em->block_start = EXTENT_MAP_HOLE;
5170         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5171 insert:
5172         btrfs_release_path(path);
5173         if (em->start > start || extent_map_end(em) <= start) {
5174                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5175                        "[%llu %llu]\n", (unsigned long long)em->start,
5176                        (unsigned long long)em->len,
5177                        (unsigned long long)start,
5178                        (unsigned long long)len);
5179                 err = -EIO;
5180                 goto out;
5181         }
5182
5183         err = 0;
5184         write_lock(&em_tree->lock);
5185         ret = add_extent_mapping(em_tree, em);
5186         /* it is possible that someone inserted the extent into the tree
5187          * while we had the lock dropped.  It is also possible that
5188          * an overlapping map exists in the tree
5189          */
5190         if (ret == -EEXIST) {
5191                 struct extent_map *existing;
5192
5193                 ret = 0;
5194
5195                 existing = lookup_extent_mapping(em_tree, start, len);
5196                 if (existing && (existing->start > start ||
5197                     existing->start + existing->len <= start)) {
5198                         free_extent_map(existing);
5199                         existing = NULL;
5200                 }
5201                 if (!existing) {
5202                         existing = lookup_extent_mapping(em_tree, em->start,
5203                                                          em->len);
5204                         if (existing) {
5205                                 err = merge_extent_mapping(em_tree, existing,
5206                                                            em, start,
5207                                                            root->sectorsize);
5208                                 free_extent_map(existing);
5209                                 if (err) {
5210                                         free_extent_map(em);
5211                                         em = NULL;
5212                                 }
5213                         } else {
5214                                 err = -EIO;
5215                                 free_extent_map(em);
5216                                 em = NULL;
5217                         }
5218                 } else {
5219                         free_extent_map(em);
5220                         em = existing;
5221                         err = 0;
5222                 }
5223         }
5224         write_unlock(&em_tree->lock);
5225 out:
5226
5227         trace_btrfs_get_extent(root, em);
5228
5229         if (path)
5230                 btrfs_free_path(path);
5231         if (trans) {
5232                 ret = btrfs_end_transaction(trans, root);
5233                 if (!err)
5234                         err = ret;
5235         }
5236         if (err) {
5237                 free_extent_map(em);
5238                 return ERR_PTR(err);
5239         }
5240         return em;
5241 }
5242
5243 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5244                                            size_t pg_offset, u64 start, u64 len,
5245                                            int create)
5246 {
5247         struct extent_map *em;
5248         struct extent_map *hole_em = NULL;
5249         u64 range_start = start;
5250         u64 end;
5251         u64 found;
5252         u64 found_end;
5253         int err = 0;
5254
5255         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5256         if (IS_ERR(em))
5257                 return em;
5258         if (em) {
5259                 /*
5260                  * if our em maps to a hole, there might
5261                  * actually be delalloc bytes behind it
5262                  */
5263                 if (em->block_start != EXTENT_MAP_HOLE)
5264                         return em;
5265                 else
5266                         hole_em = em;
5267         }
5268
5269         /* check to see if we've wrapped (len == -1 or similar) */
5270         end = start + len;
5271         if (end < start)
5272                 end = (u64)-1;
5273         else
5274                 end -= 1;
5275
5276         em = NULL;
5277
5278         /* ok, we didn't find anything, lets look for delalloc */
5279         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5280                                  end, len, EXTENT_DELALLOC, 1);
5281         found_end = range_start + found;
5282         if (found_end < range_start)
5283                 found_end = (u64)-1;
5284
5285         /*
5286          * we didn't find anything useful, return
5287          * the original results from get_extent()
5288          */
5289         if (range_start > end || found_end <= start) {
5290                 em = hole_em;
5291                 hole_em = NULL;
5292                 goto out;
5293         }
5294
5295         /* adjust the range_start to make sure it doesn't
5296          * go backwards from the start they passed in
5297          */
5298         range_start = max(start,range_start);
5299         found = found_end - range_start;
5300
5301         if (found > 0) {
5302                 u64 hole_start = start;
5303                 u64 hole_len = len;
5304
5305                 em = alloc_extent_map();
5306                 if (!em) {
5307                         err = -ENOMEM;
5308                         goto out;
5309                 }
5310                 /*
5311                  * when btrfs_get_extent can't find anything it
5312                  * returns one huge hole
5313                  *
5314                  * make sure what it found really fits our range, and
5315                  * adjust to make sure it is based on the start from
5316                  * the caller
5317                  */
5318                 if (hole_em) {
5319                         u64 calc_end = extent_map_end(hole_em);
5320
5321                         if (calc_end <= start || (hole_em->start > end)) {
5322                                 free_extent_map(hole_em);
5323                                 hole_em = NULL;
5324                         } else {
5325                                 hole_start = max(hole_em->start, start);
5326                                 hole_len = calc_end - hole_start;
5327                         }
5328                 }
5329                 em->bdev = NULL;
5330                 if (hole_em && range_start > hole_start) {
5331                         /* our hole starts before our delalloc, so we
5332                          * have to return just the parts of the hole
5333                          * that go until  the delalloc starts
5334                          */
5335                         em->len = min(hole_len,
5336                                       range_start - hole_start);
5337                         em->start = hole_start;
5338                         em->orig_start = hole_start;
5339                         /*
5340                          * don't adjust block start at all,
5341                          * it is fixed at EXTENT_MAP_HOLE
5342                          */
5343                         em->block_start = hole_em->block_start;
5344                         em->block_len = hole_len;
5345                 } else {
5346                         em->start = range_start;
5347                         em->len = found;
5348                         em->orig_start = range_start;
5349                         em->block_start = EXTENT_MAP_DELALLOC;
5350                         em->block_len = found;
5351                 }
5352         } else if (hole_em) {
5353                 return hole_em;
5354         }
5355 out:
5356
5357         free_extent_map(hole_em);
5358         if (err) {
5359                 free_extent_map(em);
5360                 return ERR_PTR(err);
5361         }
5362         return em;
5363 }
5364
5365 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5366                                                   struct extent_map *em,
5367                                                   u64 start, u64 len)
5368 {
5369         struct btrfs_root *root = BTRFS_I(inode)->root;
5370         struct btrfs_trans_handle *trans;
5371         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5372         struct btrfs_key ins;
5373         u64 alloc_hint;
5374         int ret;
5375         bool insert = false;
5376
5377         /*
5378          * Ok if the extent map we looked up is a hole and is for the exact
5379          * range we want, there is no reason to allocate a new one, however if
5380          * it is not right then we need to free this one and drop the cache for
5381          * our range.
5382          */
5383         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5384             em->len != len) {
5385                 free_extent_map(em);
5386                 em = NULL;
5387                 insert = true;
5388                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5389         }
5390
5391         trans = btrfs_join_transaction(root);
5392         if (IS_ERR(trans))
5393                 return ERR_CAST(trans);
5394
5395         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5396                 btrfs_add_inode_defrag(trans, inode);
5397
5398         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5399
5400         alloc_hint = get_extent_allocation_hint(inode, start, len);
5401         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5402                                    alloc_hint, (u64)-1, &ins, 1);
5403         if (ret) {
5404                 em = ERR_PTR(ret);
5405                 goto out;
5406         }
5407
5408         if (!em) {
5409                 em = alloc_extent_map();
5410                 if (!em) {
5411                         em = ERR_PTR(-ENOMEM);
5412                         goto out;
5413                 }
5414         }
5415
5416         em->start = start;
5417         em->orig_start = em->start;
5418         em->len = ins.offset;
5419
5420         em->block_start = ins.objectid;
5421         em->block_len = ins.offset;
5422         em->bdev = root->fs_info->fs_devices->latest_bdev;
5423
5424         /*
5425          * We need to do this because if we're using the original em we searched
5426          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5427          */
5428         em->flags = 0;
5429         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5430
5431         while (insert) {
5432                 write_lock(&em_tree->lock);
5433                 ret = add_extent_mapping(em_tree, em);
5434                 write_unlock(&em_tree->lock);
5435                 if (ret != -EEXIST)
5436                         break;
5437                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5438         }
5439
5440         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5441                                            ins.offset, ins.offset, 0);
5442         if (ret) {
5443                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5444                 em = ERR_PTR(ret);
5445         }
5446 out:
5447         btrfs_end_transaction(trans, root);
5448         return em;
5449 }
5450
5451 /*
5452  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5453  * block must be cow'd
5454  */
5455 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5456                                       struct inode *inode, u64 offset, u64 len)
5457 {
5458         struct btrfs_path *path;
5459         int ret;
5460         struct extent_buffer *leaf;
5461         struct btrfs_root *root = BTRFS_I(inode)->root;
5462         struct btrfs_file_extent_item *fi;
5463         struct btrfs_key key;
5464         u64 disk_bytenr;
5465         u64 backref_offset;
5466         u64 extent_end;
5467         u64 num_bytes;
5468         int slot;
5469         int found_type;
5470
5471         path = btrfs_alloc_path();
5472         if (!path)
5473                 return -ENOMEM;
5474
5475         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5476                                        offset, 0);
5477         if (ret < 0)
5478                 goto out;
5479
5480         slot = path->slots[0];
5481         if (ret == 1) {
5482                 if (slot == 0) {
5483                         /* can't find the item, must cow */
5484                         ret = 0;
5485                         goto out;
5486                 }
5487                 slot--;
5488         }
5489         ret = 0;
5490         leaf = path->nodes[0];
5491         btrfs_item_key_to_cpu(leaf, &key, slot);
5492         if (key.objectid != btrfs_ino(inode) ||
5493             key.type != BTRFS_EXTENT_DATA_KEY) {
5494                 /* not our file or wrong item type, must cow */
5495                 goto out;
5496         }
5497
5498         if (key.offset > offset) {
5499                 /* Wrong offset, must cow */
5500                 goto out;
5501         }
5502
5503         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5504         found_type = btrfs_file_extent_type(leaf, fi);
5505         if (found_type != BTRFS_FILE_EXTENT_REG &&
5506             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5507                 /* not a regular extent, must cow */
5508                 goto out;
5509         }
5510         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5511         backref_offset = btrfs_file_extent_offset(leaf, fi);
5512
5513         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5514         if (extent_end < offset + len) {
5515                 /* extent doesn't include our full range, must cow */
5516                 goto out;
5517         }
5518
5519         if (btrfs_extent_readonly(root, disk_bytenr))
5520                 goto out;
5521
5522         /*
5523          * look for other files referencing this extent, if we
5524          * find any we must cow
5525          */
5526         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5527                                   key.offset - backref_offset, disk_bytenr))
5528                 goto out;
5529
5530         /*
5531          * adjust disk_bytenr and num_bytes to cover just the bytes
5532          * in this extent we are about to write.  If there
5533          * are any csums in that range we have to cow in order
5534          * to keep the csums correct
5535          */
5536         disk_bytenr += backref_offset;
5537         disk_bytenr += offset - key.offset;
5538         num_bytes = min(offset + len, extent_end) - offset;
5539         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5540                                 goto out;
5541         /*
5542          * all of the above have passed, it is safe to overwrite this extent
5543          * without cow
5544          */
5545         ret = 1;
5546 out:
5547         btrfs_free_path(path);
5548         return ret;
5549 }
5550
5551 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5552                                    struct buffer_head *bh_result, int create)
5553 {
5554         struct extent_map *em;
5555         struct btrfs_root *root = BTRFS_I(inode)->root;
5556         u64 start = iblock << inode->i_blkbits;
5557         u64 len = bh_result->b_size;
5558         struct btrfs_trans_handle *trans;
5559
5560         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5561         if (IS_ERR(em))
5562                 return PTR_ERR(em);
5563
5564         /*
5565          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5566          * io.  INLINE is special, and we could probably kludge it in here, but
5567          * it's still buffered so for safety lets just fall back to the generic
5568          * buffered path.
5569          *
5570          * For COMPRESSED we _have_ to read the entire extent in so we can
5571          * decompress it, so there will be buffering required no matter what we
5572          * do, so go ahead and fallback to buffered.
5573          *
5574          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5575          * to buffered IO.  Don't blame me, this is the price we pay for using
5576          * the generic code.
5577          */
5578         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5579             em->block_start == EXTENT_MAP_INLINE) {
5580                 free_extent_map(em);
5581                 return -ENOTBLK;
5582         }
5583
5584         /* Just a good old fashioned hole, return */
5585         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5586                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5587                 free_extent_map(em);
5588                 /* DIO will do one hole at a time, so just unlock a sector */
5589                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5590                               start + root->sectorsize - 1, GFP_NOFS);
5591                 return 0;
5592         }
5593
5594         /*
5595          * We don't allocate a new extent in the following cases
5596          *
5597          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5598          * existing extent.
5599          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5600          * just use the extent.
5601          *
5602          */
5603         if (!create) {
5604                 len = em->len - (start - em->start);
5605                 goto map;
5606         }
5607
5608         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5609             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5610              em->block_start != EXTENT_MAP_HOLE)) {
5611                 int type;
5612                 int ret;
5613                 u64 block_start;
5614
5615                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5616                         type = BTRFS_ORDERED_PREALLOC;
5617                 else
5618                         type = BTRFS_ORDERED_NOCOW;
5619                 len = min(len, em->len - (start - em->start));
5620                 block_start = em->block_start + (start - em->start);
5621
5622                 /*
5623                  * we're not going to log anything, but we do need
5624                  * to make sure the current transaction stays open
5625                  * while we look for nocow cross refs
5626                  */
5627                 trans = btrfs_join_transaction(root);
5628                 if (IS_ERR(trans))
5629                         goto must_cow;
5630
5631                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5632                         ret = btrfs_add_ordered_extent_dio(inode, start,
5633                                            block_start, len, len, type);
5634                         btrfs_end_transaction(trans, root);
5635                         if (ret) {
5636                                 free_extent_map(em);
5637                                 return ret;
5638                         }
5639                         goto unlock;
5640                 }
5641                 btrfs_end_transaction(trans, root);
5642         }
5643 must_cow:
5644         /*
5645          * this will cow the extent, reset the len in case we changed
5646          * it above
5647          */
5648         len = bh_result->b_size;
5649         em = btrfs_new_extent_direct(inode, em, start, len);
5650         if (IS_ERR(em))
5651                 return PTR_ERR(em);
5652         len = min(len, em->len - (start - em->start));
5653 unlock:
5654         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5655                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5656                           0, NULL, GFP_NOFS);
5657 map:
5658         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5659                 inode->i_blkbits;
5660         bh_result->b_size = len;
5661         bh_result->b_bdev = em->bdev;
5662         set_buffer_mapped(bh_result);
5663         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5664                 set_buffer_new(bh_result);
5665
5666         free_extent_map(em);
5667
5668         return 0;
5669 }
5670
5671 struct btrfs_dio_private {
5672         struct inode *inode;
5673         u64 logical_offset;
5674         u64 disk_bytenr;
5675         u64 bytes;
5676         u32 *csums;
5677         void *private;
5678
5679         /* number of bios pending for this dio */
5680         atomic_t pending_bios;
5681
5682         /* IO errors */
5683         int errors;
5684
5685         struct bio *orig_bio;
5686 };
5687
5688 static void btrfs_endio_direct_read(struct bio *bio, int err)
5689 {
5690         struct btrfs_dio_private *dip = bio->bi_private;
5691         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5692         struct bio_vec *bvec = bio->bi_io_vec;
5693         struct inode *inode = dip->inode;
5694         struct btrfs_root *root = BTRFS_I(inode)->root;
5695         u64 start;
5696         u32 *private = dip->csums;
5697
5698         start = dip->logical_offset;
5699         do {
5700                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5701                         struct page *page = bvec->bv_page;
5702                         char *kaddr;
5703                         u32 csum = ~(u32)0;
5704                         unsigned long flags;
5705
5706                         local_irq_save(flags);
5707                         kaddr = kmap_atomic(page, KM_IRQ0);
5708                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5709                                                csum, bvec->bv_len);
5710                         btrfs_csum_final(csum, (char *)&csum);
5711                         kunmap_atomic(kaddr, KM_IRQ0);
5712                         local_irq_restore(flags);
5713
5714                         flush_dcache_page(bvec->bv_page);
5715                         if (csum != *private) {
5716                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5717                                       " %llu csum %u private %u\n",
5718                                       (unsigned long long)btrfs_ino(inode),
5719                                       (unsigned long long)start,
5720                                       csum, *private);
5721                                 err = -EIO;
5722                         }
5723                 }
5724
5725                 start += bvec->bv_len;
5726                 private++;
5727                 bvec++;
5728         } while (bvec <= bvec_end);
5729
5730         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5731                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5732         bio->bi_private = dip->private;
5733
5734         kfree(dip->csums);
5735         kfree(dip);
5736
5737         /* If we had a csum failure make sure to clear the uptodate flag */
5738         if (err)
5739                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5740         dio_end_io(bio, err);
5741 }
5742
5743 static void btrfs_endio_direct_write(struct bio *bio, int err)
5744 {
5745         struct btrfs_dio_private *dip = bio->bi_private;
5746         struct inode *inode = dip->inode;
5747         struct btrfs_root *root = BTRFS_I(inode)->root;
5748         struct btrfs_trans_handle *trans;
5749         struct btrfs_ordered_extent *ordered = NULL;
5750         struct extent_state *cached_state = NULL;
5751         u64 ordered_offset = dip->logical_offset;
5752         u64 ordered_bytes = dip->bytes;
5753         int ret;
5754
5755         if (err)
5756                 goto out_done;
5757 again:
5758         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5759                                                    &ordered_offset,
5760                                                    ordered_bytes);
5761         if (!ret)
5762                 goto out_test;
5763
5764         BUG_ON(!ordered);
5765
5766         trans = btrfs_join_transaction(root);
5767         if (IS_ERR(trans)) {
5768                 err = -ENOMEM;
5769                 goto out;
5770         }
5771         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5772
5773         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5774                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5775                 if (!ret)
5776                         ret = btrfs_update_inode(trans, root, inode);
5777                 err = ret;
5778                 goto out;
5779         }
5780
5781         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5782                          ordered->file_offset + ordered->len - 1, 0,
5783                          &cached_state, GFP_NOFS);
5784
5785         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5786                 ret = btrfs_mark_extent_written(trans, inode,
5787                                                 ordered->file_offset,
5788                                                 ordered->file_offset +
5789                                                 ordered->len);
5790                 if (ret) {
5791                         err = ret;
5792                         goto out_unlock;
5793                 }
5794         } else {
5795                 ret = insert_reserved_file_extent(trans, inode,
5796                                                   ordered->file_offset,
5797                                                   ordered->start,
5798                                                   ordered->disk_len,
5799                                                   ordered->len,
5800                                                   ordered->len,
5801                                                   0, 0, 0,
5802                                                   BTRFS_FILE_EXTENT_REG);
5803                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5804                                    ordered->file_offset, ordered->len);
5805                 if (ret) {
5806                         err = ret;
5807                         WARN_ON(1);
5808                         goto out_unlock;
5809                 }
5810         }
5811
5812         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5813         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5814         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5815                 btrfs_update_inode(trans, root, inode);
5816         ret = 0;
5817 out_unlock:
5818         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5819                              ordered->file_offset + ordered->len - 1,
5820                              &cached_state, GFP_NOFS);
5821 out:
5822         btrfs_delalloc_release_metadata(inode, ordered->len);
5823         btrfs_end_transaction(trans, root);
5824         ordered_offset = ordered->file_offset + ordered->len;
5825         btrfs_put_ordered_extent(ordered);
5826         btrfs_put_ordered_extent(ordered);
5827
5828 out_test:
5829         /*
5830          * our bio might span multiple ordered extents.  If we haven't
5831          * completed the accounting for the whole dio, go back and try again
5832          */
5833         if (ordered_offset < dip->logical_offset + dip->bytes) {
5834                 ordered_bytes = dip->logical_offset + dip->bytes -
5835                         ordered_offset;
5836                 goto again;
5837         }
5838 out_done:
5839         bio->bi_private = dip->private;
5840
5841         kfree(dip->csums);
5842         kfree(dip);
5843
5844         /* If we had an error make sure to clear the uptodate flag */
5845         if (err)
5846                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5847         dio_end_io(bio, err);
5848 }
5849
5850 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5851                                     struct bio *bio, int mirror_num,
5852                                     unsigned long bio_flags, u64 offset)
5853 {
5854         int ret;
5855         struct btrfs_root *root = BTRFS_I(inode)->root;
5856         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5857         BUG_ON(ret);
5858         return 0;
5859 }
5860
5861 static void btrfs_end_dio_bio(struct bio *bio, int err)
5862 {
5863         struct btrfs_dio_private *dip = bio->bi_private;
5864
5865         if (err) {
5866                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5867                       "sector %#Lx len %u err no %d\n",
5868                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5869                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5870                 dip->errors = 1;
5871
5872                 /*
5873                  * before atomic variable goto zero, we must make sure
5874                  * dip->errors is perceived to be set.
5875                  */
5876                 smp_mb__before_atomic_dec();
5877         }
5878
5879         /* if there are more bios still pending for this dio, just exit */
5880         if (!atomic_dec_and_test(&dip->pending_bios))
5881                 goto out;
5882
5883         if (dip->errors)
5884                 bio_io_error(dip->orig_bio);
5885         else {
5886                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5887                 bio_endio(dip->orig_bio, 0);
5888         }
5889 out:
5890         bio_put(bio);
5891 }
5892
5893 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5894                                        u64 first_sector, gfp_t gfp_flags)
5895 {
5896         int nr_vecs = bio_get_nr_vecs(bdev);
5897         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5898 }
5899
5900 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5901                                          int rw, u64 file_offset, int skip_sum,
5902                                          u32 *csums, int async_submit)
5903 {
5904         int write = rw & REQ_WRITE;
5905         struct btrfs_root *root = BTRFS_I(inode)->root;
5906         int ret;
5907
5908         bio_get(bio);
5909         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5910         if (ret)
5911                 goto err;
5912
5913         if (skip_sum)
5914                 goto map;
5915
5916         if (write && async_submit) {
5917                 ret = btrfs_wq_submit_bio(root->fs_info,
5918                                    inode, rw, bio, 0, 0,
5919                                    file_offset,
5920                                    __btrfs_submit_bio_start_direct_io,
5921                                    __btrfs_submit_bio_done);
5922                 goto err;
5923         } else if (write) {
5924                 /*
5925                  * If we aren't doing async submit, calculate the csum of the
5926                  * bio now.
5927                  */
5928                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5929                 if (ret)
5930                         goto err;
5931         } else if (!skip_sum) {
5932                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5933                                           file_offset, csums);
5934                 if (ret)
5935                         goto err;
5936         }
5937
5938 map:
5939         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5940 err:
5941         bio_put(bio);
5942         return ret;
5943 }
5944
5945 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5946                                     int skip_sum)
5947 {
5948         struct inode *inode = dip->inode;
5949         struct btrfs_root *root = BTRFS_I(inode)->root;
5950         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5951         struct bio *bio;
5952         struct bio *orig_bio = dip->orig_bio;
5953         struct bio_vec *bvec = orig_bio->bi_io_vec;
5954         u64 start_sector = orig_bio->bi_sector;
5955         u64 file_offset = dip->logical_offset;
5956         u64 submit_len = 0;
5957         u64 map_length;
5958         int nr_pages = 0;
5959         u32 *csums = dip->csums;
5960         int ret = 0;
5961         int async_submit = 0;
5962         int write = rw & REQ_WRITE;
5963
5964         map_length = orig_bio->bi_size;
5965         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5966                               &map_length, NULL, 0);
5967         if (ret) {
5968                 bio_put(orig_bio);
5969                 return -EIO;
5970         }
5971
5972         if (map_length >= orig_bio->bi_size) {
5973                 bio = orig_bio;
5974                 goto submit;
5975         }
5976
5977         async_submit = 1;
5978         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5979         if (!bio)
5980                 return -ENOMEM;
5981         bio->bi_private = dip;
5982         bio->bi_end_io = btrfs_end_dio_bio;
5983         atomic_inc(&dip->pending_bios);
5984
5985         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5986                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5987                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5988                                  bvec->bv_offset) < bvec->bv_len)) {
5989                         /*
5990                          * inc the count before we submit the bio so
5991                          * we know the end IO handler won't happen before
5992                          * we inc the count. Otherwise, the dip might get freed
5993                          * before we're done setting it up
5994                          */
5995                         atomic_inc(&dip->pending_bios);
5996                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5997                                                      file_offset, skip_sum,
5998                                                      csums, async_submit);
5999                         if (ret) {
6000                                 bio_put(bio);
6001                                 atomic_dec(&dip->pending_bios);
6002                                 goto out_err;
6003                         }
6004
6005                         /* Write's use the ordered csums */
6006                         if (!write && !skip_sum)
6007                                 csums = csums + nr_pages;
6008                         start_sector += submit_len >> 9;
6009                         file_offset += submit_len;
6010
6011                         submit_len = 0;
6012                         nr_pages = 0;
6013
6014                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6015                                                   start_sector, GFP_NOFS);
6016                         if (!bio)
6017                                 goto out_err;
6018                         bio->bi_private = dip;
6019                         bio->bi_end_io = btrfs_end_dio_bio;
6020
6021                         map_length = orig_bio->bi_size;
6022                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6023                                               &map_length, NULL, 0);
6024                         if (ret) {
6025                                 bio_put(bio);
6026                                 goto out_err;
6027                         }
6028                 } else {
6029                         submit_len += bvec->bv_len;
6030                         nr_pages ++;
6031                         bvec++;
6032                 }
6033         }
6034
6035 submit:
6036         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6037                                      csums, async_submit);
6038         if (!ret)
6039                 return 0;
6040
6041         bio_put(bio);
6042 out_err:
6043         dip->errors = 1;
6044         /*
6045          * before atomic variable goto zero, we must
6046          * make sure dip->errors is perceived to be set.
6047          */
6048         smp_mb__before_atomic_dec();
6049         if (atomic_dec_and_test(&dip->pending_bios))
6050                 bio_io_error(dip->orig_bio);
6051
6052         /* bio_end_io() will handle error, so we needn't return it */
6053         return 0;
6054 }
6055
6056 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6057                                 loff_t file_offset)
6058 {
6059         struct btrfs_root *root = BTRFS_I(inode)->root;
6060         struct btrfs_dio_private *dip;
6061         struct bio_vec *bvec = bio->bi_io_vec;
6062         int skip_sum;
6063         int write = rw & REQ_WRITE;
6064         int ret = 0;
6065
6066         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6067
6068         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6069         if (!dip) {
6070                 ret = -ENOMEM;
6071                 goto free_ordered;
6072         }
6073         dip->csums = NULL;
6074
6075         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6076         if (!write && !skip_sum) {
6077                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6078                 if (!dip->csums) {
6079                         kfree(dip);
6080                         ret = -ENOMEM;
6081                         goto free_ordered;
6082                 }
6083         }
6084
6085         dip->private = bio->bi_private;
6086         dip->inode = inode;
6087         dip->logical_offset = file_offset;
6088
6089         dip->bytes = 0;
6090         do {
6091                 dip->bytes += bvec->bv_len;
6092                 bvec++;
6093         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6094
6095         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6096         bio->bi_private = dip;
6097         dip->errors = 0;
6098         dip->orig_bio = bio;
6099         atomic_set(&dip->pending_bios, 0);
6100
6101         if (write)
6102                 bio->bi_end_io = btrfs_endio_direct_write;
6103         else
6104                 bio->bi_end_io = btrfs_endio_direct_read;
6105
6106         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6107         if (!ret)
6108                 return;
6109 free_ordered:
6110         /*
6111          * If this is a write, we need to clean up the reserved space and kill
6112          * the ordered extent.
6113          */
6114         if (write) {
6115                 struct btrfs_ordered_extent *ordered;
6116                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6117                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6118                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6119                         btrfs_free_reserved_extent(root, ordered->start,
6120                                                    ordered->disk_len);
6121                 btrfs_put_ordered_extent(ordered);
6122                 btrfs_put_ordered_extent(ordered);
6123         }
6124         bio_endio(bio, ret);
6125 }
6126
6127 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6128                         const struct iovec *iov, loff_t offset,
6129                         unsigned long nr_segs)
6130 {
6131         int seg;
6132         int i;
6133         size_t size;
6134         unsigned long addr;
6135         unsigned blocksize_mask = root->sectorsize - 1;
6136         ssize_t retval = -EINVAL;
6137         loff_t end = offset;
6138
6139         if (offset & blocksize_mask)
6140                 goto out;
6141
6142         /* Check the memory alignment.  Blocks cannot straddle pages */
6143         for (seg = 0; seg < nr_segs; seg++) {
6144                 addr = (unsigned long)iov[seg].iov_base;
6145                 size = iov[seg].iov_len;
6146                 end += size;
6147                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6148                         goto out;
6149
6150                 /* If this is a write we don't need to check anymore */
6151                 if (rw & WRITE)
6152                         continue;
6153
6154                 /*
6155                  * Check to make sure we don't have duplicate iov_base's in this
6156                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6157                  * when reading back.
6158                  */
6159                 for (i = seg + 1; i < nr_segs; i++) {
6160                         if (iov[seg].iov_base == iov[i].iov_base)
6161                                 goto out;
6162                 }
6163         }
6164         retval = 0;
6165 out:
6166         return retval;
6167 }
6168 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6169                         const struct iovec *iov, loff_t offset,
6170                         unsigned long nr_segs)
6171 {
6172         struct file *file = iocb->ki_filp;
6173         struct inode *inode = file->f_mapping->host;
6174         struct btrfs_ordered_extent *ordered;
6175         struct extent_state *cached_state = NULL;
6176         u64 lockstart, lockend;
6177         ssize_t ret;
6178         int writing = rw & WRITE;
6179         int write_bits = 0;
6180         size_t count = iov_length(iov, nr_segs);
6181
6182         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6183                             offset, nr_segs)) {
6184                 return 0;
6185         }
6186
6187         lockstart = offset;
6188         lockend = offset + count - 1;
6189
6190         if (writing) {
6191                 ret = btrfs_delalloc_reserve_space(inode, count);
6192                 if (ret)
6193                         goto out;
6194         }
6195
6196         while (1) {
6197                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6198                                  0, &cached_state, GFP_NOFS);
6199                 /*
6200                  * We're concerned with the entire range that we're going to be
6201                  * doing DIO to, so we need to make sure theres no ordered
6202                  * extents in this range.
6203                  */
6204                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6205                                                      lockend - lockstart + 1);
6206                 if (!ordered)
6207                         break;
6208                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6209                                      &cached_state, GFP_NOFS);
6210                 btrfs_start_ordered_extent(inode, ordered, 1);
6211                 btrfs_put_ordered_extent(ordered);
6212                 cond_resched();
6213         }
6214
6215         /*
6216          * we don't use btrfs_set_extent_delalloc because we don't want
6217          * the dirty or uptodate bits
6218          */
6219         if (writing) {
6220                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6221                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6222                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6223                                      GFP_NOFS);
6224                 if (ret) {
6225                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6226                                          lockend, EXTENT_LOCKED | write_bits,
6227                                          1, 0, &cached_state, GFP_NOFS);
6228                         goto out;
6229                 }
6230         }
6231
6232         free_extent_state(cached_state);
6233         cached_state = NULL;
6234
6235         ret = __blockdev_direct_IO(rw, iocb, inode,
6236                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6237                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6238                    btrfs_submit_direct, 0);
6239
6240         if (ret < 0 && ret != -EIOCBQUEUED) {
6241                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6242                               offset + iov_length(iov, nr_segs) - 1,
6243                               EXTENT_LOCKED | write_bits, 1, 0,
6244                               &cached_state, GFP_NOFS);
6245         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6246                 /*
6247                  * We're falling back to buffered, unlock the section we didn't
6248                  * do IO on.
6249                  */
6250                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6251                               offset + iov_length(iov, nr_segs) - 1,
6252                               EXTENT_LOCKED | write_bits, 1, 0,
6253                               &cached_state, GFP_NOFS);
6254         }
6255 out:
6256         free_extent_state(cached_state);
6257         return ret;
6258 }
6259
6260 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6261                 __u64 start, __u64 len)
6262 {
6263         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6264 }
6265
6266 int btrfs_readpage(struct file *file, struct page *page)
6267 {
6268         struct extent_io_tree *tree;
6269         tree = &BTRFS_I(page->mapping->host)->io_tree;
6270         return extent_read_full_page(tree, page, btrfs_get_extent);
6271 }
6272
6273 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6274 {
6275         struct extent_io_tree *tree;
6276
6277
6278         if (current->flags & PF_MEMALLOC) {
6279                 redirty_page_for_writepage(wbc, page);
6280                 unlock_page(page);
6281                 return 0;
6282         }
6283         tree = &BTRFS_I(page->mapping->host)->io_tree;
6284         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6285 }
6286
6287 int btrfs_writepages(struct address_space *mapping,
6288                      struct writeback_control *wbc)
6289 {
6290         struct extent_io_tree *tree;
6291
6292         tree = &BTRFS_I(mapping->host)->io_tree;
6293         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6294 }
6295
6296 static int
6297 btrfs_readpages(struct file *file, struct address_space *mapping,
6298                 struct list_head *pages, unsigned nr_pages)
6299 {
6300         struct extent_io_tree *tree;
6301         tree = &BTRFS_I(mapping->host)->io_tree;
6302         return extent_readpages(tree, mapping, pages, nr_pages,
6303                                 btrfs_get_extent);
6304 }
6305 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6306 {
6307         struct extent_io_tree *tree;
6308         struct extent_map_tree *map;
6309         int ret;
6310
6311         tree = &BTRFS_I(page->mapping->host)->io_tree;
6312         map = &BTRFS_I(page->mapping->host)->extent_tree;
6313         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6314         if (ret == 1) {
6315                 ClearPagePrivate(page);
6316                 set_page_private(page, 0);
6317                 page_cache_release(page);
6318         }
6319         return ret;
6320 }
6321
6322 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6323 {
6324         if (PageWriteback(page) || PageDirty(page))
6325                 return 0;
6326         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6327 }
6328
6329 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6330 {
6331         struct extent_io_tree *tree;
6332         struct btrfs_ordered_extent *ordered;
6333         struct extent_state *cached_state = NULL;
6334         u64 page_start = page_offset(page);
6335         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6336
6337
6338         /*
6339          * we have the page locked, so new writeback can't start,
6340          * and the dirty bit won't be cleared while we are here.
6341          *
6342          * Wait for IO on this page so that we can safely clear
6343          * the PagePrivate2 bit and do ordered accounting
6344          */
6345         wait_on_page_writeback(page);
6346
6347         tree = &BTRFS_I(page->mapping->host)->io_tree;
6348         if (offset) {
6349                 btrfs_releasepage(page, GFP_NOFS);
6350                 return;
6351         }
6352         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6353                          GFP_NOFS);
6354         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6355                                            page_offset(page));
6356         if (ordered) {
6357                 /*
6358                  * IO on this page will never be started, so we need
6359                  * to account for any ordered extents now
6360                  */
6361                 clear_extent_bit(tree, page_start, page_end,
6362                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6363                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6364                                  &cached_state, GFP_NOFS);
6365                 /*
6366                  * whoever cleared the private bit is responsible
6367                  * for the finish_ordered_io
6368                  */
6369                 if (TestClearPagePrivate2(page)) {
6370                         btrfs_finish_ordered_io(page->mapping->host,
6371                                                 page_start, page_end);
6372                 }
6373                 btrfs_put_ordered_extent(ordered);
6374                 cached_state = NULL;
6375                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6376                                  GFP_NOFS);
6377         }
6378         clear_extent_bit(tree, page_start, page_end,
6379                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6380                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6381         __btrfs_releasepage(page, GFP_NOFS);
6382
6383         ClearPageChecked(page);
6384         if (PagePrivate(page)) {
6385                 ClearPagePrivate(page);
6386                 set_page_private(page, 0);
6387                 page_cache_release(page);
6388         }
6389 }
6390
6391 /*
6392  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6393  * called from a page fault handler when a page is first dirtied. Hence we must
6394  * be careful to check for EOF conditions here. We set the page up correctly
6395  * for a written page which means we get ENOSPC checking when writing into
6396  * holes and correct delalloc and unwritten extent mapping on filesystems that
6397  * support these features.
6398  *
6399  * We are not allowed to take the i_mutex here so we have to play games to
6400  * protect against truncate races as the page could now be beyond EOF.  Because
6401  * vmtruncate() writes the inode size before removing pages, once we have the
6402  * page lock we can determine safely if the page is beyond EOF. If it is not
6403  * beyond EOF, then the page is guaranteed safe against truncation until we
6404  * unlock the page.
6405  */
6406 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6407 {
6408         struct page *page = vmf->page;
6409         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6410         struct btrfs_root *root = BTRFS_I(inode)->root;
6411         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6412         struct btrfs_ordered_extent *ordered;
6413         struct extent_state *cached_state = NULL;
6414         char *kaddr;
6415         unsigned long zero_start;
6416         loff_t size;
6417         int ret;
6418         u64 page_start;
6419         u64 page_end;
6420
6421         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6422         if (ret) {
6423                 if (ret == -ENOMEM)
6424                         ret = VM_FAULT_OOM;
6425                 else /* -ENOSPC, -EIO, etc */
6426                         ret = VM_FAULT_SIGBUS;
6427                 goto out;
6428         }
6429
6430         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6431 again:
6432         lock_page(page);
6433         size = i_size_read(inode);
6434         page_start = page_offset(page);
6435         page_end = page_start + PAGE_CACHE_SIZE - 1;
6436
6437         if ((page->mapping != inode->i_mapping) ||
6438             (page_start >= size)) {
6439                 /* page got truncated out from underneath us */
6440                 goto out_unlock;
6441         }
6442         wait_on_page_writeback(page);
6443
6444         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6445                          GFP_NOFS);
6446         set_page_extent_mapped(page);
6447
6448         /*
6449          * we can't set the delalloc bits if there are pending ordered
6450          * extents.  Drop our locks and wait for them to finish
6451          */
6452         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6453         if (ordered) {
6454                 unlock_extent_cached(io_tree, page_start, page_end,
6455                                      &cached_state, GFP_NOFS);
6456                 unlock_page(page);
6457                 btrfs_start_ordered_extent(inode, ordered, 1);
6458                 btrfs_put_ordered_extent(ordered);
6459                 goto again;
6460         }
6461
6462         /*
6463          * XXX - page_mkwrite gets called every time the page is dirtied, even
6464          * if it was already dirty, so for space accounting reasons we need to
6465          * clear any delalloc bits for the range we are fixing to save.  There
6466          * is probably a better way to do this, but for now keep consistent with
6467          * prepare_pages in the normal write path.
6468          */
6469         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6470                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6471                           0, 0, &cached_state, GFP_NOFS);
6472
6473         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6474                                         &cached_state);
6475         if (ret) {
6476                 unlock_extent_cached(io_tree, page_start, page_end,
6477                                      &cached_state, GFP_NOFS);
6478                 ret = VM_FAULT_SIGBUS;
6479                 goto out_unlock;
6480         }
6481         ret = 0;
6482
6483         /* page is wholly or partially inside EOF */
6484         if (page_start + PAGE_CACHE_SIZE > size)
6485                 zero_start = size & ~PAGE_CACHE_MASK;
6486         else
6487                 zero_start = PAGE_CACHE_SIZE;
6488
6489         if (zero_start != PAGE_CACHE_SIZE) {
6490                 kaddr = kmap(page);
6491                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6492                 flush_dcache_page(page);
6493                 kunmap(page);
6494         }
6495         ClearPageChecked(page);
6496         set_page_dirty(page);
6497         SetPageUptodate(page);
6498
6499         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6500         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6501
6502         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6503
6504 out_unlock:
6505         if (!ret)
6506                 return VM_FAULT_LOCKED;
6507         unlock_page(page);
6508         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6509 out:
6510         return ret;
6511 }
6512
6513 static int btrfs_truncate(struct inode *inode)
6514 {
6515         struct btrfs_root *root = BTRFS_I(inode)->root;
6516         struct btrfs_block_rsv *rsv;
6517         int ret;
6518         int err = 0;
6519         struct btrfs_trans_handle *trans;
6520         unsigned long nr;
6521         u64 mask = root->sectorsize - 1;
6522         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6523
6524         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6525         if (ret)
6526                 return ret;
6527
6528         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6529         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6530
6531         /*
6532          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6533          * 3 things going on here
6534          *
6535          * 1) We need to reserve space for our orphan item and the space to
6536          * delete our orphan item.  Lord knows we don't want to have a dangling
6537          * orphan item because we didn't reserve space to remove it.
6538          *
6539          * 2) We need to reserve space to update our inode.
6540          *
6541          * 3) We need to have something to cache all the space that is going to
6542          * be free'd up by the truncate operation, but also have some slack
6543          * space reserved in case it uses space during the truncate (thank you
6544          * very much snapshotting).
6545          *
6546          * And we need these to all be seperate.  The fact is we can use alot of
6547          * space doing the truncate, and we have no earthly idea how much space
6548          * we will use, so we need the truncate reservation to be seperate so it
6549          * doesn't end up using space reserved for updating the inode or
6550          * removing the orphan item.  We also need to be able to stop the
6551          * transaction and start a new one, which means we need to be able to
6552          * update the inode several times, and we have no idea of knowing how
6553          * many times that will be, so we can't just reserve 1 item for the
6554          * entirety of the opration, so that has to be done seperately as well.
6555          * Then there is the orphan item, which does indeed need to be held on
6556          * to for the whole operation, and we need nobody to touch this reserved
6557          * space except the orphan code.
6558          *
6559          * So that leaves us with
6560          *
6561          * 1) root->orphan_block_rsv - for the orphan deletion.
6562          * 2) rsv - for the truncate reservation, which we will steal from the
6563          * transaction reservation.
6564          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6565          * updating the inode.
6566          */
6567         rsv = btrfs_alloc_block_rsv(root);
6568         if (!rsv)
6569                 return -ENOMEM;
6570         rsv->size = min_size;
6571
6572         /*
6573          * 1 for the truncate slack space
6574          * 1 for the orphan item we're going to add
6575          * 1 for the orphan item deletion
6576          * 1 for updating the inode.
6577          */
6578         trans = btrfs_start_transaction(root, 4);
6579         if (IS_ERR(trans)) {
6580                 err = PTR_ERR(trans);
6581                 goto out;
6582         }
6583
6584         /* Migrate the slack space for the truncate to our reserve */
6585         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6586                                       min_size);
6587         BUG_ON(ret);
6588
6589         ret = btrfs_orphan_add(trans, inode);
6590         if (ret) {
6591                 btrfs_end_transaction(trans, root);
6592                 goto out;
6593         }
6594
6595         /*
6596          * setattr is responsible for setting the ordered_data_close flag,
6597          * but that is only tested during the last file release.  That
6598          * could happen well after the next commit, leaving a great big
6599          * window where new writes may get lost if someone chooses to write
6600          * to this file after truncating to zero
6601          *
6602          * The inode doesn't have any dirty data here, and so if we commit
6603          * this is a noop.  If someone immediately starts writing to the inode
6604          * it is very likely we'll catch some of their writes in this
6605          * transaction, and the commit will find this file on the ordered
6606          * data list with good things to send down.
6607          *
6608          * This is a best effort solution, there is still a window where
6609          * using truncate to replace the contents of the file will
6610          * end up with a zero length file after a crash.
6611          */
6612         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6613                 btrfs_add_ordered_operation(trans, root, inode);
6614
6615         while (1) {
6616                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
6617                 if (ret) {
6618                         /*
6619                          * This can only happen with the original transaction we
6620                          * started above, every other time we shouldn't have a
6621                          * transaction started yet.
6622                          */
6623                         if (ret == -EAGAIN)
6624                                 goto end_trans;
6625                         err = ret;
6626                         break;
6627                 }
6628
6629                 if (!trans) {
6630                         /* Just need the 1 for updating the inode */
6631                         trans = btrfs_start_transaction(root, 1);
6632                         if (IS_ERR(trans)) {
6633                                 err = PTR_ERR(trans);
6634                                 goto out;
6635                         }
6636                 }
6637
6638                 trans->block_rsv = rsv;
6639
6640                 ret = btrfs_truncate_inode_items(trans, root, inode,
6641                                                  inode->i_size,
6642                                                  BTRFS_EXTENT_DATA_KEY);
6643                 if (ret != -EAGAIN) {
6644                         err = ret;
6645                         break;
6646                 }
6647
6648                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6649                 ret = btrfs_update_inode(trans, root, inode);
6650                 if (ret) {
6651                         err = ret;
6652                         break;
6653                 }
6654 end_trans:
6655                 nr = trans->blocks_used;
6656                 btrfs_end_transaction(trans, root);
6657                 trans = NULL;
6658                 btrfs_btree_balance_dirty(root, nr);
6659         }
6660
6661         if (ret == 0 && inode->i_nlink > 0) {
6662                 trans->block_rsv = root->orphan_block_rsv;
6663                 ret = btrfs_orphan_del(trans, inode);
6664                 if (ret)
6665                         err = ret;
6666         } else if (ret && inode->i_nlink > 0) {
6667                 /*
6668                  * Failed to do the truncate, remove us from the in memory
6669                  * orphan list.
6670                  */
6671                 ret = btrfs_orphan_del(NULL, inode);
6672         }
6673
6674         trans->block_rsv = &root->fs_info->trans_block_rsv;
6675         ret = btrfs_update_inode(trans, root, inode);
6676         if (ret && !err)
6677                 err = ret;
6678
6679         nr = trans->blocks_used;
6680         ret = btrfs_end_transaction_throttle(trans, root);
6681         btrfs_btree_balance_dirty(root, nr);
6682
6683 out:
6684         btrfs_free_block_rsv(root, rsv);
6685
6686         if (ret && !err)
6687                 err = ret;
6688
6689         return err;
6690 }
6691
6692 /*
6693  * create a new subvolume directory/inode (helper for the ioctl).
6694  */
6695 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6696                              struct btrfs_root *new_root, u64 new_dirid)
6697 {
6698         struct inode *inode;
6699         int err;
6700         u64 index = 0;
6701
6702         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6703                                 new_dirid, S_IFDIR | 0700, &index);
6704         if (IS_ERR(inode))
6705                 return PTR_ERR(inode);
6706         inode->i_op = &btrfs_dir_inode_operations;
6707         inode->i_fop = &btrfs_dir_file_operations;
6708
6709         inode->i_nlink = 1;
6710         btrfs_i_size_write(inode, 0);
6711
6712         err = btrfs_update_inode(trans, new_root, inode);
6713         BUG_ON(err);
6714
6715         iput(inode);
6716         return 0;
6717 }
6718
6719 struct inode *btrfs_alloc_inode(struct super_block *sb)
6720 {
6721         struct btrfs_inode *ei;
6722         struct inode *inode;
6723
6724         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6725         if (!ei)
6726                 return NULL;
6727
6728         ei->root = NULL;
6729         ei->space_info = NULL;
6730         ei->generation = 0;
6731         ei->sequence = 0;
6732         ei->last_trans = 0;
6733         ei->last_sub_trans = 0;
6734         ei->logged_trans = 0;
6735         ei->delalloc_bytes = 0;
6736         ei->disk_i_size = 0;
6737         ei->flags = 0;
6738         ei->csum_bytes = 0;
6739         ei->index_cnt = (u64)-1;
6740         ei->last_unlink_trans = 0;
6741
6742         spin_lock_init(&ei->lock);
6743         ei->outstanding_extents = 0;
6744         ei->reserved_extents = 0;
6745
6746         ei->ordered_data_close = 0;
6747         ei->orphan_meta_reserved = 0;
6748         ei->dummy_inode = 0;
6749         ei->in_defrag = 0;
6750         ei->force_compress = BTRFS_COMPRESS_NONE;
6751
6752         ei->delayed_node = NULL;
6753
6754         inode = &ei->vfs_inode;
6755         extent_map_tree_init(&ei->extent_tree);
6756         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6757         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6758         mutex_init(&ei->log_mutex);
6759         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6760         INIT_LIST_HEAD(&ei->i_orphan);
6761         INIT_LIST_HEAD(&ei->delalloc_inodes);
6762         INIT_LIST_HEAD(&ei->ordered_operations);
6763         RB_CLEAR_NODE(&ei->rb_node);
6764
6765         return inode;
6766 }
6767
6768 static void btrfs_i_callback(struct rcu_head *head)
6769 {
6770         struct inode *inode = container_of(head, struct inode, i_rcu);
6771         INIT_LIST_HEAD(&inode->i_dentry);
6772         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6773 }
6774
6775 void btrfs_destroy_inode(struct inode *inode)
6776 {
6777         struct btrfs_ordered_extent *ordered;
6778         struct btrfs_root *root = BTRFS_I(inode)->root;
6779
6780         WARN_ON(!list_empty(&inode->i_dentry));
6781         WARN_ON(inode->i_data.nrpages);
6782         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6783         WARN_ON(BTRFS_I(inode)->reserved_extents);
6784         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6785         WARN_ON(BTRFS_I(inode)->csum_bytes);
6786
6787         /*
6788          * This can happen where we create an inode, but somebody else also
6789          * created the same inode and we need to destroy the one we already
6790          * created.
6791          */
6792         if (!root)
6793                 goto free;
6794
6795         /*
6796          * Make sure we're properly removed from the ordered operation
6797          * lists.
6798          */
6799         smp_mb();
6800         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6801                 spin_lock(&root->fs_info->ordered_extent_lock);
6802                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6803                 spin_unlock(&root->fs_info->ordered_extent_lock);
6804         }
6805
6806         spin_lock(&root->orphan_lock);
6807         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6808                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6809                        (unsigned long long)btrfs_ino(inode));
6810                 list_del_init(&BTRFS_I(inode)->i_orphan);
6811         }
6812         spin_unlock(&root->orphan_lock);
6813
6814         while (1) {
6815                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6816                 if (!ordered)
6817                         break;
6818                 else {
6819                         printk(KERN_ERR "btrfs found ordered "
6820                                "extent %llu %llu on inode cleanup\n",
6821                                (unsigned long long)ordered->file_offset,
6822                                (unsigned long long)ordered->len);
6823                         btrfs_remove_ordered_extent(inode, ordered);
6824                         btrfs_put_ordered_extent(ordered);
6825                         btrfs_put_ordered_extent(ordered);
6826                 }
6827         }
6828         inode_tree_del(inode);
6829         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6830 free:
6831         btrfs_remove_delayed_node(inode);
6832         call_rcu(&inode->i_rcu, btrfs_i_callback);
6833 }
6834
6835 int btrfs_drop_inode(struct inode *inode)
6836 {
6837         struct btrfs_root *root = BTRFS_I(inode)->root;
6838
6839         if (btrfs_root_refs(&root->root_item) == 0 &&
6840             !btrfs_is_free_space_inode(root, inode))
6841                 return 1;
6842         else
6843                 return generic_drop_inode(inode);
6844 }
6845
6846 static void init_once(void *foo)
6847 {
6848         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6849
6850         inode_init_once(&ei->vfs_inode);
6851 }
6852
6853 void btrfs_destroy_cachep(void)
6854 {
6855         if (btrfs_inode_cachep)
6856                 kmem_cache_destroy(btrfs_inode_cachep);
6857         if (btrfs_trans_handle_cachep)
6858                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6859         if (btrfs_transaction_cachep)
6860                 kmem_cache_destroy(btrfs_transaction_cachep);
6861         if (btrfs_path_cachep)
6862                 kmem_cache_destroy(btrfs_path_cachep);
6863         if (btrfs_free_space_cachep)
6864                 kmem_cache_destroy(btrfs_free_space_cachep);
6865 }
6866
6867 int btrfs_init_cachep(void)
6868 {
6869         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6870                         sizeof(struct btrfs_inode), 0,
6871                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6872         if (!btrfs_inode_cachep)
6873                 goto fail;
6874
6875         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6876                         sizeof(struct btrfs_trans_handle), 0,
6877                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6878         if (!btrfs_trans_handle_cachep)
6879                 goto fail;
6880
6881         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6882                         sizeof(struct btrfs_transaction), 0,
6883                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6884         if (!btrfs_transaction_cachep)
6885                 goto fail;
6886
6887         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6888                         sizeof(struct btrfs_path), 0,
6889                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6890         if (!btrfs_path_cachep)
6891                 goto fail;
6892
6893         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6894                         sizeof(struct btrfs_free_space), 0,
6895                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6896         if (!btrfs_free_space_cachep)
6897                 goto fail;
6898
6899         return 0;
6900 fail:
6901         btrfs_destroy_cachep();
6902         return -ENOMEM;
6903 }
6904
6905 static int btrfs_getattr(struct vfsmount *mnt,
6906                          struct dentry *dentry, struct kstat *stat)
6907 {
6908         struct inode *inode = dentry->d_inode;
6909         generic_fillattr(inode, stat);
6910         stat->dev = BTRFS_I(inode)->root->anon_dev;
6911         stat->blksize = PAGE_CACHE_SIZE;
6912         stat->blocks = (inode_get_bytes(inode) +
6913                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6914         return 0;
6915 }
6916
6917 /*
6918  * If a file is moved, it will inherit the cow and compression flags of the new
6919  * directory.
6920  */
6921 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6922 {
6923         struct btrfs_inode *b_dir = BTRFS_I(dir);
6924         struct btrfs_inode *b_inode = BTRFS_I(inode);
6925
6926         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6927                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6928         else
6929                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6930
6931         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6932                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6933         else
6934                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6935 }
6936
6937 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6938                            struct inode *new_dir, struct dentry *new_dentry)
6939 {
6940         struct btrfs_trans_handle *trans;
6941         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6942         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6943         struct inode *new_inode = new_dentry->d_inode;
6944         struct inode *old_inode = old_dentry->d_inode;
6945         struct timespec ctime = CURRENT_TIME;
6946         u64 index = 0;
6947         u64 root_objectid;
6948         int ret;
6949         u64 old_ino = btrfs_ino(old_inode);
6950
6951         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6952                 return -EPERM;
6953
6954         /* we only allow rename subvolume link between subvolumes */
6955         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6956                 return -EXDEV;
6957
6958         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6959             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6960                 return -ENOTEMPTY;
6961
6962         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6963             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6964                 return -ENOTEMPTY;
6965         /*
6966          * we're using rename to replace one file with another.
6967          * and the replacement file is large.  Start IO on it now so
6968          * we don't add too much work to the end of the transaction
6969          */
6970         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6971             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6972                 filemap_flush(old_inode->i_mapping);
6973
6974         /* close the racy window with snapshot create/destroy ioctl */
6975         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6976                 down_read(&root->fs_info->subvol_sem);
6977         /*
6978          * We want to reserve the absolute worst case amount of items.  So if
6979          * both inodes are subvols and we need to unlink them then that would
6980          * require 4 item modifications, but if they are both normal inodes it
6981          * would require 5 item modifications, so we'll assume their normal
6982          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6983          * should cover the worst case number of items we'll modify.
6984          */
6985         trans = btrfs_start_transaction(root, 20);
6986         if (IS_ERR(trans)) {
6987                 ret = PTR_ERR(trans);
6988                 goto out_notrans;
6989         }
6990
6991         if (dest != root)
6992                 btrfs_record_root_in_trans(trans, dest);
6993
6994         ret = btrfs_set_inode_index(new_dir, &index);
6995         if (ret)
6996                 goto out_fail;
6997
6998         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6999                 /* force full log commit if subvolume involved. */
7000                 root->fs_info->last_trans_log_full_commit = trans->transid;
7001         } else {
7002                 ret = btrfs_insert_inode_ref(trans, dest,
7003                                              new_dentry->d_name.name,
7004                                              new_dentry->d_name.len,
7005                                              old_ino,
7006                                              btrfs_ino(new_dir), index);
7007                 if (ret)
7008                         goto out_fail;
7009                 /*
7010                  * this is an ugly little race, but the rename is required
7011                  * to make sure that if we crash, the inode is either at the
7012                  * old name or the new one.  pinning the log transaction lets
7013                  * us make sure we don't allow a log commit to come in after
7014                  * we unlink the name but before we add the new name back in.
7015                  */
7016                 btrfs_pin_log_trans(root);
7017         }
7018         /*
7019          * make sure the inode gets flushed if it is replacing
7020          * something.
7021          */
7022         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7023                 btrfs_add_ordered_operation(trans, root, old_inode);
7024
7025         old_dir->i_ctime = old_dir->i_mtime = ctime;
7026         new_dir->i_ctime = new_dir->i_mtime = ctime;
7027         old_inode->i_ctime = ctime;
7028
7029         if (old_dentry->d_parent != new_dentry->d_parent)
7030                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7031
7032         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7033                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7034                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7035                                         old_dentry->d_name.name,
7036                                         old_dentry->d_name.len);
7037         } else {
7038                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7039                                         old_dentry->d_inode,
7040                                         old_dentry->d_name.name,
7041                                         old_dentry->d_name.len);
7042                 if (!ret)
7043                         ret = btrfs_update_inode(trans, root, old_inode);
7044         }
7045         BUG_ON(ret);
7046
7047         if (new_inode) {
7048                 new_inode->i_ctime = CURRENT_TIME;
7049                 if (unlikely(btrfs_ino(new_inode) ==
7050                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7051                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7052                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7053                                                 root_objectid,
7054                                                 new_dentry->d_name.name,
7055                                                 new_dentry->d_name.len);
7056                         BUG_ON(new_inode->i_nlink == 0);
7057                 } else {
7058                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7059                                                  new_dentry->d_inode,
7060                                                  new_dentry->d_name.name,
7061                                                  new_dentry->d_name.len);
7062                 }
7063                 BUG_ON(ret);
7064                 if (new_inode->i_nlink == 0) {
7065                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7066                         BUG_ON(ret);
7067                 }
7068         }
7069
7070         fixup_inode_flags(new_dir, old_inode);
7071
7072         ret = btrfs_add_link(trans, new_dir, old_inode,
7073                              new_dentry->d_name.name,
7074                              new_dentry->d_name.len, 0, index);
7075         BUG_ON(ret);
7076
7077         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7078                 struct dentry *parent = new_dentry->d_parent;
7079                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7080                 btrfs_end_log_trans(root);
7081         }
7082 out_fail:
7083         btrfs_end_transaction_throttle(trans, root);
7084 out_notrans:
7085         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7086                 up_read(&root->fs_info->subvol_sem);
7087
7088         return ret;
7089 }
7090
7091 /*
7092  * some fairly slow code that needs optimization. This walks the list
7093  * of all the inodes with pending delalloc and forces them to disk.
7094  */
7095 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7096 {
7097         struct list_head *head = &root->fs_info->delalloc_inodes;
7098         struct btrfs_inode *binode;
7099         struct inode *inode;
7100
7101         if (root->fs_info->sb->s_flags & MS_RDONLY)
7102                 return -EROFS;
7103
7104         spin_lock(&root->fs_info->delalloc_lock);
7105         while (!list_empty(head)) {
7106                 binode = list_entry(head->next, struct btrfs_inode,
7107                                     delalloc_inodes);
7108                 inode = igrab(&binode->vfs_inode);
7109                 if (!inode)
7110                         list_del_init(&binode->delalloc_inodes);
7111                 spin_unlock(&root->fs_info->delalloc_lock);
7112                 if (inode) {
7113                         filemap_flush(inode->i_mapping);
7114                         if (delay_iput)
7115                                 btrfs_add_delayed_iput(inode);
7116                         else
7117                                 iput(inode);
7118                 }
7119                 cond_resched();
7120                 spin_lock(&root->fs_info->delalloc_lock);
7121         }
7122         spin_unlock(&root->fs_info->delalloc_lock);
7123
7124         /* the filemap_flush will queue IO into the worker threads, but
7125          * we have to make sure the IO is actually started and that
7126          * ordered extents get created before we return
7127          */
7128         atomic_inc(&root->fs_info->async_submit_draining);
7129         while (atomic_read(&root->fs_info->nr_async_submits) ||
7130               atomic_read(&root->fs_info->async_delalloc_pages)) {
7131                 wait_event(root->fs_info->async_submit_wait,
7132                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7133                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7134         }
7135         atomic_dec(&root->fs_info->async_submit_draining);
7136         return 0;
7137 }
7138
7139 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7140                          const char *symname)
7141 {
7142         struct btrfs_trans_handle *trans;
7143         struct btrfs_root *root = BTRFS_I(dir)->root;
7144         struct btrfs_path *path;
7145         struct btrfs_key key;
7146         struct inode *inode = NULL;
7147         int err;
7148         int drop_inode = 0;
7149         u64 objectid;
7150         u64 index = 0 ;
7151         int name_len;
7152         int datasize;
7153         unsigned long ptr;
7154         struct btrfs_file_extent_item *ei;
7155         struct extent_buffer *leaf;
7156         unsigned long nr = 0;
7157
7158         name_len = strlen(symname) + 1;
7159         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7160                 return -ENAMETOOLONG;
7161
7162         /*
7163          * 2 items for inode item and ref
7164          * 2 items for dir items
7165          * 1 item for xattr if selinux is on
7166          */
7167         trans = btrfs_start_transaction(root, 5);
7168         if (IS_ERR(trans))
7169                 return PTR_ERR(trans);
7170
7171         err = btrfs_find_free_ino(root, &objectid);
7172         if (err)
7173                 goto out_unlock;
7174
7175         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7176                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7177                                 S_IFLNK|S_IRWXUGO, &index);
7178         if (IS_ERR(inode)) {
7179                 err = PTR_ERR(inode);
7180                 goto out_unlock;
7181         }
7182
7183         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7184         if (err) {
7185                 drop_inode = 1;
7186                 goto out_unlock;
7187         }
7188
7189         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7190         if (err)
7191                 drop_inode = 1;
7192         else {
7193                 inode->i_mapping->a_ops = &btrfs_aops;
7194                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7195                 inode->i_fop = &btrfs_file_operations;
7196                 inode->i_op = &btrfs_file_inode_operations;
7197                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7198         }
7199         if (drop_inode)
7200                 goto out_unlock;
7201
7202         path = btrfs_alloc_path();
7203         if (!path) {
7204                 err = -ENOMEM;
7205                 drop_inode = 1;
7206                 goto out_unlock;
7207         }
7208         key.objectid = btrfs_ino(inode);
7209         key.offset = 0;
7210         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7211         datasize = btrfs_file_extent_calc_inline_size(name_len);
7212         err = btrfs_insert_empty_item(trans, root, path, &key,
7213                                       datasize);
7214         if (err) {
7215                 drop_inode = 1;
7216                 btrfs_free_path(path);
7217                 goto out_unlock;
7218         }
7219         leaf = path->nodes[0];
7220         ei = btrfs_item_ptr(leaf, path->slots[0],
7221                             struct btrfs_file_extent_item);
7222         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7223         btrfs_set_file_extent_type(leaf, ei,
7224                                    BTRFS_FILE_EXTENT_INLINE);
7225         btrfs_set_file_extent_encryption(leaf, ei, 0);
7226         btrfs_set_file_extent_compression(leaf, ei, 0);
7227         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7228         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7229
7230         ptr = btrfs_file_extent_inline_start(ei);
7231         write_extent_buffer(leaf, symname, ptr, name_len);
7232         btrfs_mark_buffer_dirty(leaf);
7233         btrfs_free_path(path);
7234
7235         inode->i_op = &btrfs_symlink_inode_operations;
7236         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7237         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7238         inode_set_bytes(inode, name_len);
7239         btrfs_i_size_write(inode, name_len - 1);
7240         err = btrfs_update_inode(trans, root, inode);
7241         if (err)
7242                 drop_inode = 1;
7243
7244 out_unlock:
7245         nr = trans->blocks_used;
7246         btrfs_end_transaction_throttle(trans, root);
7247         if (drop_inode) {
7248                 inode_dec_link_count(inode);
7249                 iput(inode);
7250         }
7251         btrfs_btree_balance_dirty(root, nr);
7252         return err;
7253 }
7254
7255 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7256                                        u64 start, u64 num_bytes, u64 min_size,
7257                                        loff_t actual_len, u64 *alloc_hint,
7258                                        struct btrfs_trans_handle *trans)
7259 {
7260         struct btrfs_root *root = BTRFS_I(inode)->root;
7261         struct btrfs_key ins;
7262         u64 cur_offset = start;
7263         u64 i_size;
7264         int ret = 0;
7265         bool own_trans = true;
7266
7267         if (trans)
7268                 own_trans = false;
7269         while (num_bytes > 0) {
7270                 if (own_trans) {
7271                         trans = btrfs_start_transaction(root, 3);
7272                         if (IS_ERR(trans)) {
7273                                 ret = PTR_ERR(trans);
7274                                 break;
7275                         }
7276                 }
7277
7278                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7279                                            0, *alloc_hint, (u64)-1, &ins, 1);
7280                 if (ret) {
7281                         if (own_trans)
7282                                 btrfs_end_transaction(trans, root);
7283                         break;
7284                 }
7285
7286                 ret = insert_reserved_file_extent(trans, inode,
7287                                                   cur_offset, ins.objectid,
7288                                                   ins.offset, ins.offset,
7289                                                   ins.offset, 0, 0, 0,
7290                                                   BTRFS_FILE_EXTENT_PREALLOC);
7291                 BUG_ON(ret);
7292                 btrfs_drop_extent_cache(inode, cur_offset,
7293                                         cur_offset + ins.offset -1, 0);
7294
7295                 num_bytes -= ins.offset;
7296                 cur_offset += ins.offset;
7297                 *alloc_hint = ins.objectid + ins.offset;
7298
7299                 inode->i_ctime = CURRENT_TIME;
7300                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7301                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7302                     (actual_len > inode->i_size) &&
7303                     (cur_offset > inode->i_size)) {
7304                         if (cur_offset > actual_len)
7305                                 i_size = actual_len;
7306                         else
7307                                 i_size = cur_offset;
7308                         i_size_write(inode, i_size);
7309                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7310                 }
7311
7312                 ret = btrfs_update_inode(trans, root, inode);
7313                 BUG_ON(ret);
7314
7315                 if (own_trans)
7316                         btrfs_end_transaction(trans, root);
7317         }
7318         return ret;
7319 }
7320
7321 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7322                               u64 start, u64 num_bytes, u64 min_size,
7323                               loff_t actual_len, u64 *alloc_hint)
7324 {
7325         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7326                                            min_size, actual_len, alloc_hint,
7327                                            NULL);
7328 }
7329
7330 int btrfs_prealloc_file_range_trans(struct inode *inode,
7331                                     struct btrfs_trans_handle *trans, int mode,
7332                                     u64 start, u64 num_bytes, u64 min_size,
7333                                     loff_t actual_len, u64 *alloc_hint)
7334 {
7335         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7336                                            min_size, actual_len, alloc_hint, trans);
7337 }
7338
7339 static int btrfs_set_page_dirty(struct page *page)
7340 {
7341         return __set_page_dirty_nobuffers(page);
7342 }
7343
7344 static int btrfs_permission(struct inode *inode, int mask)
7345 {
7346         struct btrfs_root *root = BTRFS_I(inode)->root;
7347         umode_t mode = inode->i_mode;
7348
7349         if (mask & MAY_WRITE &&
7350             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7351                 if (btrfs_root_readonly(root))
7352                         return -EROFS;
7353                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7354                         return -EACCES;
7355         }
7356         return generic_permission(inode, mask);
7357 }
7358
7359 static const struct inode_operations btrfs_dir_inode_operations = {
7360         .getattr        = btrfs_getattr,
7361         .lookup         = btrfs_lookup,
7362         .create         = btrfs_create,
7363         .unlink         = btrfs_unlink,
7364         .link           = btrfs_link,
7365         .mkdir          = btrfs_mkdir,
7366         .rmdir          = btrfs_rmdir,
7367         .rename         = btrfs_rename,
7368         .symlink        = btrfs_symlink,
7369         .setattr        = btrfs_setattr,
7370         .mknod          = btrfs_mknod,
7371         .setxattr       = btrfs_setxattr,
7372         .getxattr       = btrfs_getxattr,
7373         .listxattr      = btrfs_listxattr,
7374         .removexattr    = btrfs_removexattr,
7375         .permission     = btrfs_permission,
7376         .get_acl        = btrfs_get_acl,
7377 };
7378 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7379         .lookup         = btrfs_lookup,
7380         .permission     = btrfs_permission,
7381         .get_acl        = btrfs_get_acl,
7382 };
7383
7384 static const struct file_operations btrfs_dir_file_operations = {
7385         .llseek         = generic_file_llseek,
7386         .read           = generic_read_dir,
7387         .readdir        = btrfs_real_readdir,
7388         .unlocked_ioctl = btrfs_ioctl,
7389 #ifdef CONFIG_COMPAT
7390         .compat_ioctl   = btrfs_ioctl,
7391 #endif
7392         .release        = btrfs_release_file,
7393         .fsync          = btrfs_sync_file,
7394 };
7395
7396 static struct extent_io_ops btrfs_extent_io_ops = {
7397         .fill_delalloc = run_delalloc_range,
7398         .submit_bio_hook = btrfs_submit_bio_hook,
7399         .merge_bio_hook = btrfs_merge_bio_hook,
7400         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7401         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7402         .writepage_start_hook = btrfs_writepage_start_hook,
7403         .readpage_io_failed_hook = btrfs_io_failed_hook,
7404         .set_bit_hook = btrfs_set_bit_hook,
7405         .clear_bit_hook = btrfs_clear_bit_hook,
7406         .merge_extent_hook = btrfs_merge_extent_hook,
7407         .split_extent_hook = btrfs_split_extent_hook,
7408 };
7409
7410 /*
7411  * btrfs doesn't support the bmap operation because swapfiles
7412  * use bmap to make a mapping of extents in the file.  They assume
7413  * these extents won't change over the life of the file and they
7414  * use the bmap result to do IO directly to the drive.
7415  *
7416  * the btrfs bmap call would return logical addresses that aren't
7417  * suitable for IO and they also will change frequently as COW
7418  * operations happen.  So, swapfile + btrfs == corruption.
7419  *
7420  * For now we're avoiding this by dropping bmap.
7421  */
7422 static const struct address_space_operations btrfs_aops = {
7423         .readpage       = btrfs_readpage,
7424         .writepage      = btrfs_writepage,
7425         .writepages     = btrfs_writepages,
7426         .readpages      = btrfs_readpages,
7427         .direct_IO      = btrfs_direct_IO,
7428         .invalidatepage = btrfs_invalidatepage,
7429         .releasepage    = btrfs_releasepage,
7430         .set_page_dirty = btrfs_set_page_dirty,
7431         .error_remove_page = generic_error_remove_page,
7432 };
7433
7434 static const struct address_space_operations btrfs_symlink_aops = {
7435         .readpage       = btrfs_readpage,
7436         .writepage      = btrfs_writepage,
7437         .invalidatepage = btrfs_invalidatepage,
7438         .releasepage    = btrfs_releasepage,
7439 };
7440
7441 static const struct inode_operations btrfs_file_inode_operations = {
7442         .getattr        = btrfs_getattr,
7443         .setattr        = btrfs_setattr,
7444         .setxattr       = btrfs_setxattr,
7445         .getxattr       = btrfs_getxattr,
7446         .listxattr      = btrfs_listxattr,
7447         .removexattr    = btrfs_removexattr,
7448         .permission     = btrfs_permission,
7449         .fiemap         = btrfs_fiemap,
7450         .get_acl        = btrfs_get_acl,
7451 };
7452 static const struct inode_operations btrfs_special_inode_operations = {
7453         .getattr        = btrfs_getattr,
7454         .setattr        = btrfs_setattr,
7455         .permission     = btrfs_permission,
7456         .setxattr       = btrfs_setxattr,
7457         .getxattr       = btrfs_getxattr,
7458         .listxattr      = btrfs_listxattr,
7459         .removexattr    = btrfs_removexattr,
7460         .get_acl        = btrfs_get_acl,
7461 };
7462 static const struct inode_operations btrfs_symlink_inode_operations = {
7463         .readlink       = generic_readlink,
7464         .follow_link    = page_follow_link_light,
7465         .put_link       = page_put_link,
7466         .getattr        = btrfs_getattr,
7467         .permission     = btrfs_permission,
7468         .setxattr       = btrfs_setxattr,
7469         .getxattr       = btrfs_getxattr,
7470         .listxattr      = btrfs_listxattr,
7471         .removexattr    = btrfs_removexattr,
7472         .get_acl        = btrfs_get_acl,
7473 };
7474
7475 const struct dentry_operations btrfs_dentry_operations = {
7476         .d_delete       = btrfs_dentry_delete,
7477         .d_release      = btrfs_dentry_release,
7478 };