fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61 #include "props.h"
  62
  63 struct btrfs_iget_args {
  64         u64 ino;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_path *path, int extent_inserted,
 130                                 struct btrfs_root *root, struct inode *inode,
 131                                 u64 start, size_t size, size_t compressed_size,
 132                                 int compress_type,
 133                                 struct page **compressed_pages)
 134 {
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         unsigned long offset;
 144
 145         if (compressed_size && compressed_pages)
 146                 cur_size = compressed_size;
 147
 148         inode_add_bytes(inode, size);
 149
 150         if (!extent_inserted) {
 151                 struct btrfs_key key;
 152                 size_t datasize;
 153
 154                 key.objectid = btrfs_ino(inode);
 155                 key.offset = start;
 156                 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 157
 158                 datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159                 path->leave_spinning = 1;
 160                 ret = btrfs_insert_empty_item(trans, root, path, &key,
 161                                               datasize);
 162                 if (ret) {
 163                         err = ret;
 164                         goto fail;
 165                 }
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_release_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         return err;
 223 }
 224
 225
 226 /*
 227  * conditionally insert an inline extent into the file.  This
 228  * does the checks required to make sure the data is small enough
 229  * to fit as an inline extent.
 230  */
 231 static noinline int cow_file_range_inline(struct btrfs_root *root,
 232                                           struct inode *inode, u64 start,
 233                                           u64 end, size_t compressed_size,
 234                                           int compress_type,
 235                                           struct page **compressed_pages)
 236 {
 237         struct btrfs_trans_handle *trans;
 238         u64 isize = i_size_read(inode);
 239         u64 actual_end = min(end + 1, isize);
 240         u64 inline_len = actual_end - start;
 241         u64 aligned_end = ALIGN(end, root->sectorsize);
 242         u64 data_len = inline_len;
 243         int ret;
 244         struct btrfs_path *path;
 245         int extent_inserted = 0;
 246         u32 extent_item_size;
 247
 248         if (compressed_size)
 249                 data_len = compressed_size;
 250
 251         if (start > 0 ||
 252             actual_end >= PAGE_CACHE_SIZE ||
 253             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 254             (!compressed_size &&
 255             (actual_end & (root->sectorsize - 1)) == 0) ||
 256             end + 1 < isize ||
 257             data_len > root->fs_info->max_inline) {
 258                 return 1;
 259         }
 260
 261         path = btrfs_alloc_path();
 262         if (!path)
 263                 return -ENOMEM;
 264
 265         trans = btrfs_join_transaction(root);
 266         if (IS_ERR(trans)) {
 267                 btrfs_free_path(path);
 268                 return PTR_ERR(trans);
 269         }
 270         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 271
 272         if (compressed_size && compressed_pages)
 273                 extent_item_size = btrfs_file_extent_calc_inline_size(
 274                    compressed_size);
 275         else
 276                 extent_item_size = btrfs_file_extent_calc_inline_size(
 277                     inline_len);
 278
 279         ret = __btrfs_drop_extents(trans, root, inode, path,
 280                                    start, aligned_end, NULL,
 281                                    1, 1, extent_item_size, &extent_inserted);
 282         if (ret) {
 283                 btrfs_abort_transaction(trans, root, ret);
 284                 goto out;
 285         }
 286
 287         if (isize > actual_end)
 288                 inline_len = min_t(u64, isize, actual_end);
 289         ret = insert_inline_extent(trans, path, extent_inserted,
 290                                    root, inode, start,
 291                                    inline_len, compressed_size,
 292                                    compress_type, compressed_pages);
 293         if (ret && ret != -ENOSPC) {
 294                 btrfs_abort_transaction(trans, root, ret);
 295                 goto out;
 296         } else if (ret == -ENOSPC) {
 297                 ret = 1;
 298                 goto out;
 299         }
 300
 301         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 302         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 303         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 304 out:
 305         btrfs_free_path(path);
 306         btrfs_end_transaction(trans, root);
 307         return ret;
 308 }
 309
 310 struct async_extent {
 311         u64 start;
 312         u64 ram_size;
 313         u64 compressed_size;
 314         struct page **pages;
 315         unsigned long nr_pages;
 316         int compress_type;
 317         struct list_head list;
 318 };
 319
 320 struct async_cow {
 321         struct inode *inode;
 322         struct btrfs_root *root;
 323         struct page *locked_page;
 324         u64 start;
 325         u64 end;
 326         struct list_head extents;
 327         struct btrfs_work work;
 328 };
 329
 330 static noinline int add_async_extent(struct async_cow *cow,
 331                                      u64 start, u64 ram_size,
 332                                      u64 compressed_size,
 333                                      struct page **pages,
 334                                      unsigned long nr_pages,
 335                                      int compress_type)
 336 {
 337         struct async_extent *async_extent;
 338
 339         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 340         BUG_ON(!async_extent); /* -ENOMEM */
 341         async_extent->start = start;
 342         async_extent->ram_size = ram_size;
 343         async_extent->compressed_size = compressed_size;
 344         async_extent->pages = pages;
 345         async_extent->nr_pages = nr_pages;
 346         async_extent->compress_type = compress_type;
 347         list_add_tail(&async_extent->list, &cow->extents);
 348         return 0;
 349 }
 350
 351 /*
 352  * we create compressed extents in two phases.  The first
 353  * phase compresses a range of pages that have already been
 354  * locked (both pages and state bits are locked).
 355  *
 356  * This is done inside an ordered work queue, and the compression
 357  * is spread across many cpus.  The actual IO submission is step
 358  * two, and the ordered work queue takes care of making sure that
 359  * happens in the same order things were put onto the queue by
 360  * writepages and friends.
 361  *
 362  * If this code finds it can't get good compression, it puts an
 363  * entry onto the work queue to write the uncompressed bytes.  This
 364  * makes sure that both compressed inodes and uncompressed inodes
 365  * are written in the same order that the flusher thread sent them
 366  * down.
 367  */
 368 static noinline int compress_file_range(struct inode *inode,
 369                                         struct page *locked_page,
 370                                         u64 start, u64 end,
 371                                         struct async_cow *async_cow,
 372                                         int *num_added)
 373 {
 374         struct btrfs_root *root = BTRFS_I(inode)->root;
 375         u64 num_bytes;
 376         u64 blocksize = root->sectorsize;
 377         u64 actual_end;
 378         u64 isize = i_size_read(inode);
 379         int ret = 0;
 380         struct page **pages = NULL;
 381         unsigned long nr_pages;
 382         unsigned long nr_pages_ret = 0;
 383         unsigned long total_compressed = 0;
 384         unsigned long total_in = 0;
 385         unsigned long max_compressed = 128 * 1024;
 386         unsigned long max_uncompressed = 128 * 1024;
 387         int i;
 388         int will_compress;
 389         int compress_type = root->fs_info->compress_type;
 390         int redirty = 0;
 391
 392         /* if this is a small write inside eof, kick off a defrag */
 393         if ((end - start + 1) < 16 * 1024 &&
 394             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 395                 btrfs_add_inode_defrag(NULL, inode);
 396
 397         actual_end = min_t(u64, isize, end + 1);
 398 again:
 399         will_compress = 0;
 400         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 401         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 402
 403         /*
 404          * we don't want to send crud past the end of i_size through
 405          * compression, that's just a waste of CPU time.  So, if the
 406          * end of the file is before the start of our current
 407          * requested range of bytes, we bail out to the uncompressed
 408          * cleanup code that can deal with all of this.
 409          *
 410          * It isn't really the fastest way to fix things, but this is a
 411          * very uncommon corner.
 412          */
 413         if (actual_end <= start)
 414                 goto cleanup_and_bail_uncompressed;
 415
 416         total_compressed = actual_end - start;
 417
 418         /* we want to make sure that amount of ram required to uncompress
 419          * an extent is reasonable, so we limit the total size in ram
 420          * of a compressed extent to 128k.  This is a crucial number
 421          * because it also controls how easily we can spread reads across
 422          * cpus for decompression.
 423          *
 424          * We also want to make sure the amount of IO required to do
 425          * a random read is reasonably small, so we limit the size of
 426          * a compressed extent to 128k.
 427          */
 428         total_compressed = min(total_compressed, max_uncompressed);
 429         num_bytes = ALIGN(end - start + 1, blocksize);
 430         num_bytes = max(blocksize,  num_bytes);
 431         total_in = 0;
 432         ret = 0;
 433
 434         /*
 435          * we do compression for mount -o compress and when the
 436          * inode has not been flagged as nocompress.  This flag can
 437          * change at any time if we discover bad compression ratios.
 438          */
 439         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 440             (btrfs_test_opt(root, COMPRESS) ||
 441              (BTRFS_I(inode)->force_compress) ||
 442              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 443                 WARN_ON(pages);
 444                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 445                 if (!pages) {
 446                         /* just bail out to the uncompressed code */
 447                         goto cont;
 448                 }
 449
 450                 if (BTRFS_I(inode)->force_compress)
 451                         compress_type = BTRFS_I(inode)->force_compress;
 452
 453                 /*
 454                  * we need to call clear_page_dirty_for_io on each
 455                  * page in the range.  Otherwise applications with the file
 456                  * mmap'd can wander in and change the page contents while
 457                  * we are compressing them.
 458                  *
 459                  * If the compression fails for any reason, we set the pages
 460                  * dirty again later on.
 461                  */
 462                 extent_range_clear_dirty_for_io(inode, start, end);
 463                 redirty = 1;
 464                 ret = btrfs_compress_pages(compress_type,
 465                                            inode->i_mapping, start,
 466                                            total_compressed, pages,
 467                                            nr_pages, &nr_pages_ret,
 468                                            &total_in,
 469                                            &total_compressed,
 470                                            max_compressed);
 471
 472                 if (!ret) {
 473                         unsigned long offset = total_compressed &
 474                                 (PAGE_CACHE_SIZE - 1);
 475                         struct page *page = pages[nr_pages_ret - 1];
 476                         char *kaddr;
 477
 478                         /* zero the tail end of the last page, we might be
 479                          * sending it down to disk
 480                          */
 481                         if (offset) {
 482                                 kaddr = kmap_atomic(page);
 483                                 memset(kaddr + offset, 0,
 484                                        PAGE_CACHE_SIZE - offset);
 485                                 kunmap_atomic(kaddr);
 486                         }
 487                         will_compress = 1;
 488                 }
 489         }
 490 cont:
 491         if (start == 0) {
 492                 /* lets try to make an inline extent */
 493                 if (ret || total_in < (actual_end - start)) {
 494                         /* we didn't compress the entire range, try
 495                          * to make an uncompressed inline extent.
 496                          */
 497                         ret = cow_file_range_inline(root, inode, start, end,
 498                                                     0, 0, NULL);
 499                 } else {
 500                         /* try making a compressed inline extent */
 501                         ret = cow_file_range_inline(root, inode, start, end,
 502                                                     total_compressed,
 503                                                     compress_type, pages);
 504                 }
 505                 if (ret <= 0) {
 506                         unsigned long clear_flags = EXTENT_DELALLOC |
 507                                 EXTENT_DEFRAG;
 508                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 509
 510                         /*
 511                          * inline extent creation worked or returned error,
 512                          * we don't need to create any more async work items.
 513                          * Unlock and free up our temp pages.
 514                          */
 515                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 516                                                      clear_flags, PAGE_UNLOCK |
 517                                                      PAGE_CLEAR_DIRTY |
 518                                                      PAGE_SET_WRITEBACK |
 519                                                      PAGE_END_WRITEBACK);
 520                         goto free_pages_out;
 521                 }
 522         }
 523
 524         if (will_compress) {
 525                 /*
 526                  * we aren't doing an inline extent round the compressed size
 527                  * up to a block size boundary so the allocator does sane
 528                  * things
 529                  */
 530                 total_compressed = ALIGN(total_compressed, blocksize);
 531
 532                 /*
 533                  * one last check to make sure the compression is really a
 534                  * win, compare the page count read with the blocks on disk
 535                  */
 536                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 537                 if (total_compressed >= total_in) {
 538                         will_compress = 0;
 539                 } else {
 540                         num_bytes = total_in;
 541                 }
 542         }
 543         if (!will_compress && pages) {
 544                 /*
 545                  * the compression code ran but failed to make things smaller,
 546                  * free any pages it allocated and our page pointer array
 547                  */
 548                 for (i = 0; i < nr_pages_ret; i++) {
 549                         WARN_ON(pages[i]->mapping);
 550                         page_cache_release(pages[i]);
 551                 }
 552                 kfree(pages);
 553                 pages = NULL;
 554                 total_compressed = 0;
 555                 nr_pages_ret = 0;
 556
 557                 /* flag the file so we don't compress in the future */
 558                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 559                     !(BTRFS_I(inode)->force_compress)) {
 560                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 561                 }
 562         }
 563         if (will_compress) {
 564                 *num_added += 1;
 565
 566                 /* the async work queues will take care of doing actual
 567                  * allocation on disk for these compressed pages,
 568                  * and will submit them to the elevator.
 569                  */
 570                 add_async_extent(async_cow, start, num_bytes,
 571                                  total_compressed, pages, nr_pages_ret,
 572                                  compress_type);
 573
 574                 if (start + num_bytes < end) {
 575                         start += num_bytes;
 576                         pages = NULL;
 577                         cond_resched();
 578                         goto again;
 579                 }
 580         } else {
 581 cleanup_and_bail_uncompressed:
 582                 /*
 583                  * No compression, but we still need to write the pages in
 584                  * the file we've been given so far.  redirty the locked
 585                  * page if it corresponds to our extent and set things up
 586                  * for the async work queue to run cow_file_range to do
 587                  * the normal delalloc dance
 588                  */
 589                 if (page_offset(locked_page) >= start &&
 590                     page_offset(locked_page) <= end) {
 591                         __set_page_dirty_nobuffers(locked_page);
 592                         /* unlocked later on in the async handlers */
 593                 }
 594                 if (redirty)
 595                         extent_range_redirty_for_io(inode, start, end);
 596                 add_async_extent(async_cow, start, end - start + 1,
 597                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 598                 *num_added += 1;
 599         }
 600
 601 out:
 602         return ret;
 603
 604 free_pages_out:
 605         for (i = 0; i < nr_pages_ret; i++) {
 606                 WARN_ON(pages[i]->mapping);
 607                 page_cache_release(pages[i]);
 608         }
 609         kfree(pages);
 610
 611         goto out;
 612 }
 613
 614 /*
 615  * phase two of compressed writeback.  This is the ordered portion
 616  * of the code, which only gets called in the order the work was
 617  * queued.  We walk all the async extents created by compress_file_range
 618  * and send them down to the disk.
 619  */
 620 static noinline int submit_compressed_extents(struct inode *inode,
 621                                               struct async_cow *async_cow)
 622 {
 623         struct async_extent *async_extent;
 624         u64 alloc_hint = 0;
 625         struct btrfs_key ins;
 626         struct extent_map *em;
 627         struct btrfs_root *root = BTRFS_I(inode)->root;
 628         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 629         struct extent_io_tree *io_tree;
 630         int ret = 0;
 631
 632         if (list_empty(&async_cow->extents))
 633                 return 0;
 634
 635 again:
 636         while (!list_empty(&async_cow->extents)) {
 637                 async_extent = list_entry(async_cow->extents.next,
 638                                           struct async_extent, list);
 639                 list_del(&async_extent->list);
 640
 641                 io_tree = &BTRFS_I(inode)->io_tree;
 642
 643 retry:
 644                 /* did the compression code fall back to uncompressed IO? */
 645                 if (!async_extent->pages) {
 646                         int page_started = 0;
 647                         unsigned long nr_written = 0;
 648
 649                         lock_extent(io_tree, async_extent->start,
 650                                          async_extent->start +
 651                                          async_extent->ram_size - 1);
 652
 653                         /* allocate blocks */
 654                         ret = cow_file_range(inode, async_cow->locked_page,
 655                                              async_extent->start,
 656                                              async_extent->start +
 657                                              async_extent->ram_size - 1,
 658                                              &page_started, &nr_written, 0);
 659
 660                         /* JDM XXX */
 661
 662                         /*
 663                          * if page_started, cow_file_range inserted an
 664                          * inline extent and took care of all the unlocking
 665                          * and IO for us.  Otherwise, we need to submit
 666                          * all those pages down to the drive.
 667                          */
 668                         if (!page_started && !ret)
 669                                 extent_write_locked_range(io_tree,
 670                                                   inode, async_extent->start,
 671                                                   async_extent->start +
 672                                                   async_extent->ram_size - 1,
 673                                                   btrfs_get_extent,
 674                                                   WB_SYNC_ALL);
 675                         else if (ret)
 676                                 unlock_page(async_cow->locked_page);
 677                         kfree(async_extent);
 678                         cond_resched();
 679                         continue;
 680                 }
 681
 682                 lock_extent(io_tree, async_extent->start,
 683                             async_extent->start + async_extent->ram_size - 1);
 684
 685                 ret = btrfs_reserve_extent(root,
 686                                            async_extent->compressed_size,
 687                                            async_extent->compressed_size,
 688                                            0, alloc_hint, &ins, 1);
 689                 if (ret) {
 690                         int i;
 691
 692                         for (i = 0; i < async_extent->nr_pages; i++) {
 693                                 WARN_ON(async_extent->pages[i]->mapping);
 694                                 page_cache_release(async_extent->pages[i]);
 695                         }
 696                         kfree(async_extent->pages);
 697                         async_extent->nr_pages = 0;
 698                         async_extent->pages = NULL;
 699
 700                         if (ret == -ENOSPC) {
 701                                 unlock_extent(io_tree, async_extent->start,
 702                                               async_extent->start +
 703                                               async_extent->ram_size - 1);
 704                                 goto retry;
 705                         }
 706                         goto out_free;
 707                 }
 708
 709                 /*
 710                  * here we're doing allocation and writeback of the
 711                  * compressed pages
 712                  */
 713                 btrfs_drop_extent_cache(inode, async_extent->start,
 714                                         async_extent->start +
 715                                         async_extent->ram_size - 1, 0);
 716
 717                 em = alloc_extent_map();
 718                 if (!em) {
 719                         ret = -ENOMEM;
 720                         goto out_free_reserve;
 721                 }
 722                 em->start = async_extent->start;
 723                 em->len = async_extent->ram_size;
 724                 em->orig_start = em->start;
 725                 em->mod_start = em->start;
 726                 em->mod_len = em->len;
 727
 728                 em->block_start = ins.objectid;
 729                 em->block_len = ins.offset;
 730                 em->orig_block_len = ins.offset;
 731                 em->ram_bytes = async_extent->ram_size;
 732                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 733                 em->compress_type = async_extent->compress_type;
 734                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 735                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 736                 em->generation = -1;
 737
 738                 while (1) {
 739                         write_lock(&em_tree->lock);
 740                         ret = add_extent_mapping(em_tree, em, 1);
 741                         write_unlock(&em_tree->lock);
 742                         if (ret != -EEXIST) {
 743                                 free_extent_map(em);
 744                                 break;
 745                         }
 746                         btrfs_drop_extent_cache(inode, async_extent->start,
 747                                                 async_extent->start +
 748                                                 async_extent->ram_size - 1, 0);
 749                 }
 750
 751                 if (ret)
 752                         goto out_free_reserve;
 753
 754                 ret = btrfs_add_ordered_extent_compress(inode,
 755                                                 async_extent->start,
 756                                                 ins.objectid,
 757                                                 async_extent->ram_size,
 758                                                 ins.offset,
 759                                                 BTRFS_ORDERED_COMPRESSED,
 760                                                 async_extent->compress_type);
 761                 if (ret)
 762                         goto out_free_reserve;
 763
 764                 /*
 765                  * clear dirty, set writeback and unlock the pages.
 766                  */
 767                 extent_clear_unlock_delalloc(inode, async_extent->start,
 768                                 async_extent->start +
 769                                 async_extent->ram_size - 1,
 770                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 771                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 772                                 PAGE_SET_WRITEBACK);
 773                 ret = btrfs_submit_compressed_write(inode,
 774                                     async_extent->start,
 775                                     async_extent->ram_size,
 776                                     ins.objectid,
 777                                     ins.offset, async_extent->pages,
 778                                     async_extent->nr_pages);
 779                 alloc_hint = ins.objectid + ins.offset;
 780                 kfree(async_extent);
 781                 if (ret)
 782                         goto out;
 783                 cond_resched();
 784         }
 785         ret = 0;
 786 out:
 787         return ret;
 788 out_free_reserve:
 789         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 790 out_free:
 791         extent_clear_unlock_delalloc(inode, async_extent->start,
 792                                      async_extent->start +
 793                                      async_extent->ram_size - 1,
 794                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 795                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 796                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 797                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 798         kfree(async_extent);
 799         goto again;
 800 }
 801
 802 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 803                                       u64 num_bytes)
 804 {
 805         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 806         struct extent_map *em;
 807         u64 alloc_hint = 0;
 808
 809         read_lock(&em_tree->lock);
 810         em = search_extent_mapping(em_tree, start, num_bytes);
 811         if (em) {
 812                 /*
 813                  * if block start isn't an actual block number then find the
 814                  * first block in this inode and use that as a hint.  If that
 815                  * block is also bogus then just don't worry about it.
 816                  */
 817                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 818                         free_extent_map(em);
 819                         em = search_extent_mapping(em_tree, 0, 0);
 820                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 821                                 alloc_hint = em->block_start;
 822                         if (em)
 823                                 free_extent_map(em);
 824                 } else {
 825                         alloc_hint = em->block_start;
 826                         free_extent_map(em);
 827                 }
 828         }
 829         read_unlock(&em_tree->lock);
 830
 831         return alloc_hint;
 832 }
 833
 834 /*
 835  * when extent_io.c finds a delayed allocation range in the file,
 836  * the call backs end up in this code.  The basic idea is to
 837  * allocate extents on disk for the range, and create ordered data structs
 838  * in ram to track those extents.
 839  *
 840  * locked_page is the page that writepage had locked already.  We use
 841  * it to make sure we don't do extra locks or unlocks.
 842  *
 843  * *page_started is set to one if we unlock locked_page and do everything
 844  * required to start IO on it.  It may be clean and already done with
 845  * IO when we return.
 846  */
 847 static noinline int cow_file_range(struct inode *inode,
 848                                    struct page *locked_page,
 849                                    u64 start, u64 end, int *page_started,
 850                                    unsigned long *nr_written,
 851                                    int unlock)
 852 {
 853         struct btrfs_root *root = BTRFS_I(inode)->root;
 854         u64 alloc_hint = 0;
 855         u64 num_bytes;
 856         unsigned long ram_size;
 857         u64 disk_num_bytes;
 858         u64 cur_alloc_size;
 859         u64 blocksize = root->sectorsize;
 860         struct btrfs_key ins;
 861         struct extent_map *em;
 862         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 863         int ret = 0;
 864
 865         if (btrfs_is_free_space_inode(inode)) {
 866                 WARN_ON_ONCE(1);
 867                 return -EINVAL;
 868         }
 869
 870         num_bytes = ALIGN(end - start + 1, blocksize);
 871         num_bytes = max(blocksize,  num_bytes);
 872         disk_num_bytes = num_bytes;
 873
 874         /* if this is a small write inside eof, kick off defrag */
 875         if (num_bytes < 64 * 1024 &&
 876             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 877                 btrfs_add_inode_defrag(NULL, inode);
 878
 879         if (start == 0) {
 880                 /* lets try to make an inline extent */
 881                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 882                                             NULL);
 883                 if (ret == 0) {
 884                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 885                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 886                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 887                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 888                                      PAGE_END_WRITEBACK);
 889
 890                         *nr_written = *nr_written +
 891                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 892                         *page_started = 1;
 893                         goto out;
 894                 } else if (ret < 0) {
 895                         goto out_unlock;
 896                 }
 897         }
 898
 899         BUG_ON(disk_num_bytes >
 900                btrfs_super_total_bytes(root->fs_info->super_copy));
 901
 902         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 903         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 904
 905         while (disk_num_bytes > 0) {
 906                 unsigned long op;
 907
 908                 cur_alloc_size = disk_num_bytes;
 909                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 910                                            root->sectorsize, 0, alloc_hint,
 911                                            &ins, 1);
 912                 if (ret < 0)
 913                         goto out_unlock;
 914
 915                 em = alloc_extent_map();
 916                 if (!em) {
 917                         ret = -ENOMEM;
 918                         goto out_reserve;
 919                 }
 920                 em->start = start;
 921                 em->orig_start = em->start;
 922                 ram_size = ins.offset;
 923                 em->len = ins.offset;
 924                 em->mod_start = em->start;
 925                 em->mod_len = em->len;
 926
 927                 em->block_start = ins.objectid;
 928                 em->block_len = ins.offset;
 929                 em->orig_block_len = ins.offset;
 930                 em->ram_bytes = ram_size;
 931                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 932                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 933                 em->generation = -1;
 934
 935                 while (1) {
 936                         write_lock(&em_tree->lock);
 937                         ret = add_extent_mapping(em_tree, em, 1);
 938                         write_unlock(&em_tree->lock);
 939                         if (ret != -EEXIST) {
 940                                 free_extent_map(em);
 941                                 break;
 942                         }
 943                         btrfs_drop_extent_cache(inode, start,
 944                                                 start + ram_size - 1, 0);
 945                 }
 946                 if (ret)
 947                         goto out_reserve;
 948
 949                 cur_alloc_size = ins.offset;
 950                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 951                                                ram_size, cur_alloc_size, 0);
 952                 if (ret)
 953                         goto out_reserve;
 954
 955                 if (root->root_key.objectid ==
 956                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 957                         ret = btrfs_reloc_clone_csums(inode, start,
 958                                                       cur_alloc_size);
 959                         if (ret)
 960                                 goto out_reserve;
 961                 }
 962
 963                 if (disk_num_bytes < cur_alloc_size)
 964                         break;
 965
 966                 /* we're not doing compressed IO, don't unlock the first
 967                  * page (which the caller expects to stay locked), don't
 968                  * clear any dirty bits and don't set any writeback bits
 969                  *
 970                  * Do set the Private2 bit so we know this page was properly
 971                  * setup for writepage
 972                  */
 973                 op = unlock ? PAGE_UNLOCK : 0;
 974                 op |= PAGE_SET_PRIVATE2;
 975
 976                 extent_clear_unlock_delalloc(inode, start,
 977                                              start + ram_size - 1, locked_page,
 978                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 979                                              op);
 980                 disk_num_bytes -= cur_alloc_size;
 981                 num_bytes -= cur_alloc_size;
 982                 alloc_hint = ins.objectid + ins.offset;
 983                 start += cur_alloc_size;
 984         }
 985 out:
 986         return ret;
 987
 988 out_reserve:
 989         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 990 out_unlock:
 991         extent_clear_unlock_delalloc(inode, start, end, locked_page,
 992                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
 993                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
 994                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 995                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 996         goto out;
 997 }
 998
 999 /*
1000  * work queue call back to started compression on a file and pages
1001  */
1002 static noinline void async_cow_start(struct btrfs_work *work)
1003 {
1004         struct async_cow *async_cow;
1005         int num_added = 0;
1006         async_cow = container_of(work, struct async_cow, work);
1007
1008         compress_file_range(async_cow->inode, async_cow->locked_page,
1009                             async_cow->start, async_cow->end, async_cow,
1010                             &num_added);
1011         if (num_added == 0) {
1012                 btrfs_add_delayed_iput(async_cow->inode);
1013                 async_cow->inode = NULL;
1014         }
1015 }
1016
1017 /*
1018  * work queue call back to submit previously compressed pages
1019  */
1020 static noinline void async_cow_submit(struct btrfs_work *work)
1021 {
1022         struct async_cow *async_cow;
1023         struct btrfs_root *root;
1024         unsigned long nr_pages;
1025
1026         async_cow = container_of(work, struct async_cow, work);
1027
1028         root = async_cow->root;
1029         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1030                 PAGE_CACHE_SHIFT;
1031
1032         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1033             5 * 1024 * 1024 &&
1034             waitqueue_active(&root->fs_info->async_submit_wait))
1035                 wake_up(&root->fs_info->async_submit_wait);
1036
1037         if (async_cow->inode)
1038                 submit_compressed_extents(async_cow->inode, async_cow);
1039 }
1040
1041 static noinline void async_cow_free(struct btrfs_work *work)
1042 {
1043         struct async_cow *async_cow;
1044         async_cow = container_of(work, struct async_cow, work);
1045         if (async_cow->inode)
1046                 btrfs_add_delayed_iput(async_cow->inode);
1047         kfree(async_cow);
1048 }
1049
1050 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1051                                 u64 start, u64 end, int *page_started,
1052                                 unsigned long *nr_written)
1053 {
1054         struct async_cow *async_cow;
1055         struct btrfs_root *root = BTRFS_I(inode)->root;
1056         unsigned long nr_pages;
1057         u64 cur_end;
1058         int limit = 10 * 1024 * 1024;
1059
1060         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1061                          1, 0, NULL, GFP_NOFS);
1062         while (start < end) {
1063                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1064                 BUG_ON(!async_cow); /* -ENOMEM */
1065                 async_cow->inode = igrab(inode);
1066                 async_cow->root = root;
1067                 async_cow->locked_page = locked_page;
1068                 async_cow->start = start;
1069
1070                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1071                         cur_end = end;
1072                 else
1073                         cur_end = min(end, start + 512 * 1024 - 1);
1074
1075                 async_cow->end = cur_end;
1076                 INIT_LIST_HEAD(&async_cow->extents);
1077
1078                 async_cow->work.func = async_cow_start;
1079                 async_cow->work.ordered_func = async_cow_submit;
1080                 async_cow->work.ordered_free = async_cow_free;
1081                 async_cow->work.flags = 0;
1082
1083                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1084                         PAGE_CACHE_SHIFT;
1085                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1086
1087                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1088                                    &async_cow->work);
1089
1090                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1091                         wait_event(root->fs_info->async_submit_wait,
1092                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1093                             limit));
1094                 }
1095
1096                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1097                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1098                         wait_event(root->fs_info->async_submit_wait,
1099                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1100                            0));
1101                 }
1102
1103                 *nr_written += nr_pages;
1104                 start = cur_end + 1;
1105         }
1106         *page_started = 1;
1107         return 0;
1108 }
1109
1110 static noinline int csum_exist_in_range(struct btrfs_root *root,
1111                                         u64 bytenr, u64 num_bytes)
1112 {
1113         int ret;
1114         struct btrfs_ordered_sum *sums;
1115         LIST_HEAD(list);
1116
1117         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1118                                        bytenr + num_bytes - 1, &list, 0);
1119         if (ret == 0 && list_empty(&list))
1120                 return 0;
1121
1122         while (!list_empty(&list)) {
1123                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1124                 list_del(&sums->list);
1125                 kfree(sums);
1126         }
1127         return 1;
1128 }
1129
1130 /*
1131  * when nowcow writeback call back.  This checks for snapshots or COW copies
1132  * of the extents that exist in the file, and COWs the file as required.
1133  *
1134  * If no cow copies or snapshots exist, we write directly to the existing
1135  * blocks on disk
1136  */
1137 static noinline int run_delalloc_nocow(struct inode *inode,
1138                                        struct page *locked_page,
1139                               u64 start, u64 end, int *page_started, int force,
1140                               unsigned long *nr_written)
1141 {
1142         struct btrfs_root *root = BTRFS_I(inode)->root;
1143         struct btrfs_trans_handle *trans;
1144         struct extent_buffer *leaf;
1145         struct btrfs_path *path;
1146         struct btrfs_file_extent_item *fi;
1147         struct btrfs_key found_key;
1148         u64 cow_start;
1149         u64 cur_offset;
1150         u64 extent_end;
1151         u64 extent_offset;
1152         u64 disk_bytenr;
1153         u64 num_bytes;
1154         u64 disk_num_bytes;
1155         u64 ram_bytes;
1156         int extent_type;
1157         int ret, err;
1158         int type;
1159         int nocow;
1160         int check_prev = 1;
1161         bool nolock;
1162         u64 ino = btrfs_ino(inode);
1163
1164         path = btrfs_alloc_path();
1165         if (!path) {
1166                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1167                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1168                                              EXTENT_DO_ACCOUNTING |
1169                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1170                                              PAGE_CLEAR_DIRTY |
1171                                              PAGE_SET_WRITEBACK |
1172                                              PAGE_END_WRITEBACK);
1173                 return -ENOMEM;
1174         }
1175
1176         nolock = btrfs_is_free_space_inode(inode);
1177
1178         if (nolock)
1179                 trans = btrfs_join_transaction_nolock(root);
1180         else
1181                 trans = btrfs_join_transaction(root);
1182
1183         if (IS_ERR(trans)) {
1184                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1185                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1186                                              EXTENT_DO_ACCOUNTING |
1187                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1188                                              PAGE_CLEAR_DIRTY |
1189                                              PAGE_SET_WRITEBACK |
1190                                              PAGE_END_WRITEBACK);
1191                 btrfs_free_path(path);
1192                 return PTR_ERR(trans);
1193         }
1194
1195         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1196
1197         cow_start = (u64)-1;
1198         cur_offset = start;
1199         while (1) {
1200                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1201                                                cur_offset, 0);
1202                 if (ret < 0)
1203                         goto error;
1204                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1205                         leaf = path->nodes[0];
1206                         btrfs_item_key_to_cpu(leaf, &found_key,
1207                                               path->slots[0] - 1);
1208                         if (found_key.objectid == ino &&
1209                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1210                                 path->slots[0]--;
1211                 }
1212                 check_prev = 0;
1213 next_slot:
1214                 leaf = path->nodes[0];
1215                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1216                         ret = btrfs_next_leaf(root, path);
1217                         if (ret < 0)
1218                                 goto error;
1219                         if (ret > 0)
1220                                 break;
1221                         leaf = path->nodes[0];
1222                 }
1223
1224                 nocow = 0;
1225                 disk_bytenr = 0;
1226                 num_bytes = 0;
1227                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1228
1229                 if (found_key.objectid > ino ||
1230                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1231                     found_key.offset > end)
1232                         break;
1233
1234                 if (found_key.offset > cur_offset) {
1235                         extent_end = found_key.offset;
1236                         extent_type = 0;
1237                         goto out_check;
1238                 }
1239
1240                 fi = btrfs_item_ptr(leaf, path->slots[0],
1241                                     struct btrfs_file_extent_item);
1242                 extent_type = btrfs_file_extent_type(leaf, fi);
1243
1244                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1245                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1246                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1247                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1248                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1249                         extent_end = found_key.offset +
1250                                 btrfs_file_extent_num_bytes(leaf, fi);
1251                         disk_num_bytes =
1252                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1253                         if (extent_end <= start) {
1254                                 path->slots[0]++;
1255                                 goto next_slot;
1256                         }
1257                         if (disk_bytenr == 0)
1258                                 goto out_check;
1259                         if (btrfs_file_extent_compression(leaf, fi) ||
1260                             btrfs_file_extent_encryption(leaf, fi) ||
1261                             btrfs_file_extent_other_encoding(leaf, fi))
1262                                 goto out_check;
1263                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1264                                 goto out_check;
1265                         if (btrfs_extent_readonly(root, disk_bytenr))
1266                                 goto out_check;
1267                         if (btrfs_cross_ref_exist(trans, root, ino,
1268                                                   found_key.offset -
1269                                                   extent_offset, disk_bytenr))
1270                                 goto out_check;
1271                         disk_bytenr += extent_offset;
1272                         disk_bytenr += cur_offset - found_key.offset;
1273                         num_bytes = min(end + 1, extent_end) - cur_offset;
1274                         /*
1275                          * force cow if csum exists in the range.
1276                          * this ensure that csum for a given extent are
1277                          * either valid or do not exist.
1278                          */
1279                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1280                                 goto out_check;
1281                         nocow = 1;
1282                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1283                         extent_end = found_key.offset +
1284                                 btrfs_file_extent_inline_len(leaf, fi);
1285                         extent_end = ALIGN(extent_end, root->sectorsize);
1286                 } else {
1287                         BUG_ON(1);
1288                 }
1289 out_check:
1290                 if (extent_end <= start) {
1291                         path->slots[0]++;
1292                         goto next_slot;
1293                 }
1294                 if (!nocow) {
1295                         if (cow_start == (u64)-1)
1296                                 cow_start = cur_offset;
1297                         cur_offset = extent_end;
1298                         if (cur_offset > end)
1299                                 break;
1300                         path->slots[0]++;
1301                         goto next_slot;
1302                 }
1303
1304                 btrfs_release_path(path);
1305                 if (cow_start != (u64)-1) {
1306                         ret = cow_file_range(inode, locked_page,
1307                                              cow_start, found_key.offset - 1,
1308                                              page_started, nr_written, 1);
1309                         if (ret)
1310                                 goto error;
1311                         cow_start = (u64)-1;
1312                 }
1313
1314                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1315                         struct extent_map *em;
1316                         struct extent_map_tree *em_tree;
1317                         em_tree = &BTRFS_I(inode)->extent_tree;
1318                         em = alloc_extent_map();
1319                         BUG_ON(!em); /* -ENOMEM */
1320                         em->start = cur_offset;
1321                         em->orig_start = found_key.offset - extent_offset;
1322                         em->len = num_bytes;
1323                         em->block_len = num_bytes;
1324                         em->block_start = disk_bytenr;
1325                         em->orig_block_len = disk_num_bytes;
1326                         em->ram_bytes = ram_bytes;
1327                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1328                         em->mod_start = em->start;
1329                         em->mod_len = em->len;
1330                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1331                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1332                         em->generation = -1;
1333                         while (1) {
1334                                 write_lock(&em_tree->lock);
1335                                 ret = add_extent_mapping(em_tree, em, 1);
1336                                 write_unlock(&em_tree->lock);
1337                                 if (ret != -EEXIST) {
1338                                         free_extent_map(em);
1339                                         break;
1340                                 }
1341                                 btrfs_drop_extent_cache(inode, em->start,
1342                                                 em->start + em->len - 1, 0);
1343                         }
1344                         type = BTRFS_ORDERED_PREALLOC;
1345                 } else {
1346                         type = BTRFS_ORDERED_NOCOW;
1347                 }
1348
1349                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1350                                                num_bytes, num_bytes, type);
1351                 BUG_ON(ret); /* -ENOMEM */
1352
1353                 if (root->root_key.objectid ==
1354                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1355                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1356                                                       num_bytes);
1357                         if (ret)
1358                                 goto error;
1359                 }
1360
1361                 extent_clear_unlock_delalloc(inode, cur_offset,
1362                                              cur_offset + num_bytes - 1,
1363                                              locked_page, EXTENT_LOCKED |
1364                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1365                                              PAGE_SET_PRIVATE2);
1366                 cur_offset = extent_end;
1367                 if (cur_offset > end)
1368                         break;
1369         }
1370         btrfs_release_path(path);
1371
1372         if (cur_offset <= end && cow_start == (u64)-1) {
1373                 cow_start = cur_offset;
1374                 cur_offset = end;
1375         }
1376
1377         if (cow_start != (u64)-1) {
1378                 ret = cow_file_range(inode, locked_page, cow_start, end,
1379                                      page_started, nr_written, 1);
1380                 if (ret)
1381                         goto error;
1382         }
1383
1384 error:
1385         err = btrfs_end_transaction(trans, root);
1386         if (!ret)
1387                 ret = err;
1388
1389         if (ret && cur_offset < end)
1390                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1391                                              locked_page, EXTENT_LOCKED |
1392                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1393                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1394                                              PAGE_CLEAR_DIRTY |
1395                                              PAGE_SET_WRITEBACK |
1396                                              PAGE_END_WRITEBACK);
1397         btrfs_free_path(path);
1398         return ret;
1399 }
1400
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405                               u64 start, u64 end, int *page_started,
1406                               unsigned long *nr_written)
1407 {
1408         int ret;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413                                          page_started, 1, nr_written);
1414         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416                                          page_started, 0, nr_written);
1417         } else if (!btrfs_test_opt(root, COMPRESS) &&
1418                    !(BTRFS_I(inode)->force_compress) &&
1419                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420                 ret = cow_file_range(inode, locked_page, start, end,
1421                                       page_started, nr_written, 1);
1422         } else {
1423                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424                         &BTRFS_I(inode)->runtime_flags);
1425                 ret = cow_file_range_async(inode, locked_page, start, end,
1426                                            page_started, nr_written);
1427         }
1428         return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432                                     struct extent_state *orig, u64 split)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(orig->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents++;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450                                     struct extent_state *new,
1451                                     struct extent_state *other)
1452 {
1453         /* not delalloc, ignore it */
1454         if (!(other->state & EXTENT_DELALLOC))
1455                 return;
1456
1457         spin_lock(&BTRFS_I(inode)->lock);
1458         BTRFS_I(inode)->outstanding_extents--;
1459         spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1463                                       struct inode *inode)
1464 {
1465         spin_lock(&root->delalloc_lock);
1466         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1467                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1468                               &root->delalloc_inodes);
1469                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1470                         &BTRFS_I(inode)->runtime_flags);
1471                 root->nr_delalloc_inodes++;
1472                 if (root->nr_delalloc_inodes == 1) {
1473                         spin_lock(&root->fs_info->delalloc_root_lock);
1474                         BUG_ON(!list_empty(&root->delalloc_root));
1475                         list_add_tail(&root->delalloc_root,
1476                                       &root->fs_info->delalloc_roots);
1477                         spin_unlock(&root->fs_info->delalloc_root_lock);
1478                 }
1479         }
1480         spin_unlock(&root->delalloc_lock);
1481 }
1482
1483 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1484                                      struct inode *inode)
1485 {
1486         spin_lock(&root->delalloc_lock);
1487         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1488                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1489                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1490                           &BTRFS_I(inode)->runtime_flags);
1491                 root->nr_delalloc_inodes--;
1492                 if (!root->nr_delalloc_inodes) {
1493                         spin_lock(&root->fs_info->delalloc_root_lock);
1494                         BUG_ON(list_empty(&root->delalloc_root));
1495                         list_del_init(&root->delalloc_root);
1496                         spin_unlock(&root->fs_info->delalloc_root_lock);
1497                 }
1498         }
1499         spin_unlock(&root->delalloc_lock);
1500 }
1501
1502 /*
1503  * extent_io.c set_bit_hook, used to track delayed allocation
1504  * bytes in this file, and to maintain the list of inodes that
1505  * have pending delalloc work to be done.
1506  */
1507 static void btrfs_set_bit_hook(struct inode *inode,
1508                                struct extent_state *state, unsigned long *bits)
1509 {
1510
1511         /*
1512          * set_bit and clear bit hooks normally require _irqsave/restore
1513          * but in this case, we are only testing for the DELALLOC
1514          * bit, which is only set or cleared with irqs on
1515          */
1516         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1517                 struct btrfs_root *root = BTRFS_I(inode)->root;
1518                 u64 len = state->end + 1 - state->start;
1519                 bool do_list = !btrfs_is_free_space_inode(inode);
1520
1521                 if (*bits & EXTENT_FIRST_DELALLOC) {
1522                         *bits &= ~EXTENT_FIRST_DELALLOC;
1523                 } else {
1524                         spin_lock(&BTRFS_I(inode)->lock);
1525                         BTRFS_I(inode)->outstanding_extents++;
1526                         spin_unlock(&BTRFS_I(inode)->lock);
1527                 }
1528
1529                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1530                                      root->fs_info->delalloc_batch);
1531                 spin_lock(&BTRFS_I(inode)->lock);
1532                 BTRFS_I(inode)->delalloc_bytes += len;
1533                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1534                                          &BTRFS_I(inode)->runtime_flags))
1535                         btrfs_add_delalloc_inodes(root, inode);
1536                 spin_unlock(&BTRFS_I(inode)->lock);
1537         }
1538 }
1539
1540 /*
1541  * extent_io.c clear_bit_hook, see set_bit_hook for why
1542  */
1543 static void btrfs_clear_bit_hook(struct inode *inode,
1544                                  struct extent_state *state,
1545                                  unsigned long *bits)
1546 {
1547         /*
1548          * set_bit and clear bit hooks normally require _irqsave/restore
1549          * but in this case, we are only testing for the DELALLOC
1550          * bit, which is only set or cleared with irqs on
1551          */
1552         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1553                 struct btrfs_root *root = BTRFS_I(inode)->root;
1554                 u64 len = state->end + 1 - state->start;
1555                 bool do_list = !btrfs_is_free_space_inode(inode);
1556
1557                 if (*bits & EXTENT_FIRST_DELALLOC) {
1558                         *bits &= ~EXTENT_FIRST_DELALLOC;
1559                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1560                         spin_lock(&BTRFS_I(inode)->lock);
1561                         BTRFS_I(inode)->outstanding_extents--;
1562                         spin_unlock(&BTRFS_I(inode)->lock);
1563                 }
1564
1565                 /*
1566                  * We don't reserve metadata space for space cache inodes so we
1567                  * don't need to call dellalloc_release_metadata if there is an
1568                  * error.
1569                  */
1570                 if (*bits & EXTENT_DO_ACCOUNTING &&
1571                     root != root->fs_info->tree_root)
1572                         btrfs_delalloc_release_metadata(inode, len);
1573
1574                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1575                     && do_list && !(state->state & EXTENT_NORESERVE))
1576                         btrfs_free_reserved_data_space(inode, len);
1577
1578                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1579                                      root->fs_info->delalloc_batch);
1580                 spin_lock(&BTRFS_I(inode)->lock);
1581                 BTRFS_I(inode)->delalloc_bytes -= len;
1582                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1583                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1584                              &BTRFS_I(inode)->runtime_flags))
1585                         btrfs_del_delalloc_inode(root, inode);
1586                 spin_unlock(&BTRFS_I(inode)->lock);
1587         }
1588 }
1589
1590 /*
1591  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1592  * we don't create bios that span stripes or chunks
1593  */
1594 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1595                          size_t size, struct bio *bio,
1596                          unsigned long bio_flags)
1597 {
1598         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1599         u64 logical = (u64)bio->bi_sector << 9;
1600         u64 length = 0;
1601         u64 map_length;
1602         int ret;
1603
1604         if (bio_flags & EXTENT_BIO_COMPRESSED)
1605                 return 0;
1606
1607         length = bio->bi_size;
1608         map_length = length;
1609         ret = btrfs_map_block(root->fs_info, rw, logical,
1610                               &map_length, NULL, 0);
1611         /* Will always return 0 with map_multi == NULL */
1612         BUG_ON(ret < 0);
1613         if (map_length < length + size)
1614                 return 1;
1615         return 0;
1616 }
1617
1618 /*
1619  * in order to insert checksums into the metadata in large chunks,
1620  * we wait until bio submission time.   All the pages in the bio are
1621  * checksummed and sums are attached onto the ordered extent record.
1622  *
1623  * At IO completion time the cums attached on the ordered extent record
1624  * are inserted into the btree
1625  */
1626 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1627                                     struct bio *bio, int mirror_num,
1628                                     unsigned long bio_flags,
1629                                     u64 bio_offset)
1630 {
1631         struct btrfs_root *root = BTRFS_I(inode)->root;
1632         int ret = 0;
1633
1634         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1635         BUG_ON(ret); /* -ENOMEM */
1636         return 0;
1637 }
1638
1639 /*
1640  * in order to insert checksums into the metadata in large chunks,
1641  * we wait until bio submission time.   All the pages in the bio are
1642  * checksummed and sums are attached onto the ordered extent record.
1643  *
1644  * At IO completion time the cums attached on the ordered extent record
1645  * are inserted into the btree
1646  */
1647 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1648                           int mirror_num, unsigned long bio_flags,
1649                           u64 bio_offset)
1650 {
1651         struct btrfs_root *root = BTRFS_I(inode)->root;
1652         int ret;
1653
1654         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1655         if (ret)
1656                 bio_endio(bio, ret);
1657         return ret;
1658 }
1659
1660 /*
1661  * extent_io.c submission hook. This does the right thing for csum calculation
1662  * on write, or reading the csums from the tree before a read
1663  */
1664 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1665                           int mirror_num, unsigned long bio_flags,
1666                           u64 bio_offset)
1667 {
1668         struct btrfs_root *root = BTRFS_I(inode)->root;
1669         int ret = 0;
1670         int skip_sum;
1671         int metadata = 0;
1672         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1673
1674         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1675
1676         if (btrfs_is_free_space_inode(inode))
1677                 metadata = 2;
1678
1679         if (!(rw & REQ_WRITE)) {
1680                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1681                 if (ret)
1682                         goto out;
1683
1684                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1685                         ret = btrfs_submit_compressed_read(inode, bio,
1686                                                            mirror_num,
1687                                                            bio_flags);
1688                         goto out;
1689                 } else if (!skip_sum) {
1690                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1691                         if (ret)
1692                                 goto out;
1693                 }
1694                 goto mapit;
1695         } else if (async && !skip_sum) {
1696                 /* csum items have already been cloned */
1697                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1698                         goto mapit;
1699                 /* we're doing a write, do the async checksumming */
1700                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1701                                    inode, rw, bio, mirror_num,
1702                                    bio_flags, bio_offset,
1703                                    __btrfs_submit_bio_start,
1704                                    __btrfs_submit_bio_done);
1705                 goto out;
1706         } else if (!skip_sum) {
1707                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1708                 if (ret)
1709                         goto out;
1710         }
1711
1712 mapit:
1713         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1714
1715 out:
1716         if (ret < 0)
1717                 bio_endio(bio, ret);
1718         return ret;
1719 }
1720
1721 /*
1722  * given a list of ordered sums record them in the inode.  This happens
1723  * at IO completion time based on sums calculated at bio submission time.
1724  */
1725 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1726                              struct inode *inode, u64 file_offset,
1727                              struct list_head *list)
1728 {
1729         struct btrfs_ordered_sum *sum;
1730
1731         list_for_each_entry(sum, list, list) {
1732                 trans->adding_csums = 1;
1733                 btrfs_csum_file_blocks(trans,
1734                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1735                 trans->adding_csums = 0;
1736         }
1737         return 0;
1738 }
1739
1740 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1741                               struct extent_state **cached_state)
1742 {
1743         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1744         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1745                                    cached_state, GFP_NOFS);
1746 }
1747
1748 /* see btrfs_writepage_start_hook for details on why this is required */
1749 struct btrfs_writepage_fixup {
1750         struct page *page;
1751         struct btrfs_work work;
1752 };
1753
1754 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1755 {
1756         struct btrfs_writepage_fixup *fixup;
1757         struct btrfs_ordered_extent *ordered;
1758         struct extent_state *cached_state = NULL;
1759         struct page *page;
1760         struct inode *inode;
1761         u64 page_start;
1762         u64 page_end;
1763         int ret;
1764
1765         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1766         page = fixup->page;
1767 again:
1768         lock_page(page);
1769         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1770                 ClearPageChecked(page);
1771                 goto out_page;
1772         }
1773
1774         inode = page->mapping->host;
1775         page_start = page_offset(page);
1776         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1777
1778         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1779                          &cached_state);
1780
1781         /* already ordered? We're done */
1782         if (PagePrivate2(page))
1783                 goto out;
1784
1785         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1786         if (ordered) {
1787                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1788                                      page_end, &cached_state, GFP_NOFS);
1789                 unlock_page(page);
1790                 btrfs_start_ordered_extent(inode, ordered, 1);
1791                 btrfs_put_ordered_extent(ordered);
1792                 goto again;
1793         }
1794
1795         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1796         if (ret) {
1797                 mapping_set_error(page->mapping, ret);
1798                 end_extent_writepage(page, ret, page_start, page_end);
1799                 ClearPageChecked(page);
1800                 goto out;
1801          }
1802
1803         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1804         ClearPageChecked(page);
1805         set_page_dirty(page);
1806 out:
1807         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1808                              &cached_state, GFP_NOFS);
1809 out_page:
1810         unlock_page(page);
1811         page_cache_release(page);
1812         kfree(fixup);
1813 }
1814
1815 /*
1816  * There are a few paths in the higher layers of the kernel that directly
1817  * set the page dirty bit without asking the filesystem if it is a
1818  * good idea.  This causes problems because we want to make sure COW
1819  * properly happens and the data=ordered rules are followed.
1820  *
1821  * In our case any range that doesn't have the ORDERED bit set
1822  * hasn't been properly setup for IO.  We kick off an async process
1823  * to fix it up.  The async helper will wait for ordered extents, set
1824  * the delalloc bit and make it safe to write the page.
1825  */
1826 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1827 {
1828         struct inode *inode = page->mapping->host;
1829         struct btrfs_writepage_fixup *fixup;
1830         struct btrfs_root *root = BTRFS_I(inode)->root;
1831
1832         /* this page is properly in the ordered list */
1833         if (TestClearPagePrivate2(page))
1834                 return 0;
1835
1836         if (PageChecked(page))
1837                 return -EAGAIN;
1838
1839         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1840         if (!fixup)
1841                 return -EAGAIN;
1842
1843         SetPageChecked(page);
1844         page_cache_get(page);
1845         fixup->work.func = btrfs_writepage_fixup_worker;
1846         fixup->page = page;
1847         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1848         return -EBUSY;
1849 }
1850
1851 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1852                                        struct inode *inode, u64 file_pos,
1853                                        u64 disk_bytenr, u64 disk_num_bytes,
1854                                        u64 num_bytes, u64 ram_bytes,
1855                                        u8 compression, u8 encryption,
1856                                        u16 other_encoding, int extent_type)
1857 {
1858         struct btrfs_root *root = BTRFS_I(inode)->root;
1859         struct btrfs_file_extent_item *fi;
1860         struct btrfs_path *path;
1861         struct extent_buffer *leaf;
1862         struct btrfs_key ins;
1863         int extent_inserted = 0;
1864         int ret;
1865
1866         path = btrfs_alloc_path();
1867         if (!path)
1868                 return -ENOMEM;
1869
1870         /*
1871          * we may be replacing one extent in the tree with another.
1872          * The new extent is pinned in the extent map, and we don't want
1873          * to drop it from the cache until it is completely in the btree.
1874          *
1875          * So, tell btrfs_drop_extents to leave this extent in the cache.
1876          * the caller is expected to unpin it and allow it to be merged
1877          * with the others.
1878          */
1879         ret = __btrfs_drop_extents(trans, root, inode, path, file_pos,
1880                                    file_pos + num_bytes, NULL, 0,
1881                                    1, sizeof(*fi), &extent_inserted);
1882         if (ret)
1883                 goto out;
1884
1885         if (!extent_inserted) {
1886                 ins.objectid = btrfs_ino(inode);
1887                 ins.offset = file_pos;
1888                 ins.type = BTRFS_EXTENT_DATA_KEY;
1889
1890                 path->leave_spinning = 1;
1891                 ret = btrfs_insert_empty_item(trans, root, path, &ins,
1892                                               sizeof(*fi));
1893                 if (ret)
1894                         goto out;
1895         }
1896         leaf = path->nodes[0];
1897         fi = btrfs_item_ptr(leaf, path->slots[0],
1898                             struct btrfs_file_extent_item);
1899         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1900         btrfs_set_file_extent_type(leaf, fi, extent_type);
1901         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1902         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1903         btrfs_set_file_extent_offset(leaf, fi, 0);
1904         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1905         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1906         btrfs_set_file_extent_compression(leaf, fi, compression);
1907         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1908         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1909
1910         btrfs_mark_buffer_dirty(leaf);
1911         btrfs_release_path(path);
1912
1913         inode_add_bytes(inode, num_bytes);
1914
1915         ins.objectid = disk_bytenr;
1916         ins.offset = disk_num_bytes;
1917         ins.type = BTRFS_EXTENT_ITEM_KEY;
1918         ret = btrfs_alloc_reserved_file_extent(trans, root,
1919                                         root->root_key.objectid,
1920                                         btrfs_ino(inode), file_pos, &ins);
1921 out:
1922         btrfs_free_path(path);
1923
1924         return ret;
1925 }
1926
1927 /* snapshot-aware defrag */
1928 struct sa_defrag_extent_backref {
1929         struct rb_node node;
1930         struct old_sa_defrag_extent *old;
1931         u64 root_id;
1932         u64 inum;
1933         u64 file_pos;
1934         u64 extent_offset;
1935         u64 num_bytes;
1936         u64 generation;
1937 };
1938
1939 struct old_sa_defrag_extent {
1940         struct list_head list;
1941         struct new_sa_defrag_extent *new;
1942
1943         u64 extent_offset;
1944         u64 bytenr;
1945         u64 offset;
1946         u64 len;
1947         int count;
1948 };
1949
1950 struct new_sa_defrag_extent {
1951         struct rb_root root;
1952         struct list_head head;
1953         struct btrfs_path *path;
1954         struct inode *inode;
1955         u64 file_pos;
1956         u64 len;
1957         u64 bytenr;
1958         u64 disk_len;
1959         u8 compress_type;
1960 };
1961
1962 static int backref_comp(struct sa_defrag_extent_backref *b1,
1963                         struct sa_defrag_extent_backref *b2)
1964 {
1965         if (b1->root_id < b2->root_id)
1966                 return -1;
1967         else if (b1->root_id > b2->root_id)
1968                 return 1;
1969
1970         if (b1->inum < b2->inum)
1971                 return -1;
1972         else if (b1->inum > b2->inum)
1973                 return 1;
1974
1975         if (b1->file_pos < b2->file_pos)
1976                 return -1;
1977         else if (b1->file_pos > b2->file_pos)
1978                 return 1;
1979
1980         /*
1981          * [------------------------------] ===> (a range of space)
1982          *     |<--->|   |<---->| =============> (fs/file tree A)
1983          * |<---------------------------->| ===> (fs/file tree B)
1984          *
1985          * A range of space can refer to two file extents in one tree while
1986          * refer to only one file extent in another tree.
1987          *
1988          * So we may process a disk offset more than one time(two extents in A)
1989          * and locate at the same extent(one extent in B), then insert two same
1990          * backrefs(both refer to the extent in B).
1991          */
1992         return 0;
1993 }
1994
1995 static void backref_insert(struct rb_root *root,
1996                            struct sa_defrag_extent_backref *backref)
1997 {
1998         struct rb_node **p = &root->rb_node;
1999         struct rb_node *parent = NULL;
2000         struct sa_defrag_extent_backref *entry;
2001         int ret;
2002
2003         while (*p) {
2004                 parent = *p;
2005                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2006
2007                 ret = backref_comp(backref, entry);
2008                 if (ret < 0)
2009                         p = &(*p)->rb_left;
2010                 else
2011                         p = &(*p)->rb_right;
2012         }
2013
2014         rb_link_node(&backref->node, parent, p);
2015         rb_insert_color(&backref->node, root);
2016 }
2017
2018 /*
2019  * Note the backref might has changed, and in this case we just return 0.
2020  */
2021 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2022                                        void *ctx)
2023 {
2024         struct btrfs_file_extent_item *extent;
2025         struct btrfs_fs_info *fs_info;
2026         struct old_sa_defrag_extent *old = ctx;
2027         struct new_sa_defrag_extent *new = old->new;
2028         struct btrfs_path *path = new->path;
2029         struct btrfs_key key;
2030         struct btrfs_root *root;
2031         struct sa_defrag_extent_backref *backref;
2032         struct extent_buffer *leaf;
2033         struct inode *inode = new->inode;
2034         int slot;
2035         int ret;
2036         u64 extent_offset;
2037         u64 num_bytes;
2038
2039         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2040             inum == btrfs_ino(inode))
2041                 return 0;
2042
2043         key.objectid = root_id;
2044         key.type = BTRFS_ROOT_ITEM_KEY;
2045         key.offset = (u64)-1;
2046
2047         fs_info = BTRFS_I(inode)->root->fs_info;
2048         root = btrfs_read_fs_root_no_name(fs_info, &key);
2049         if (IS_ERR(root)) {
2050                 if (PTR_ERR(root) == -ENOENT)
2051                         return 0;
2052                 WARN_ON(1);
2053                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2054                          inum, offset, root_id);
2055                 return PTR_ERR(root);
2056         }
2057
2058         key.objectid = inum;
2059         key.type = BTRFS_EXTENT_DATA_KEY;
2060         if (offset > (u64)-1 << 32)
2061                 key.offset = 0;
2062         else
2063                 key.offset = offset;
2064
2065         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2066         if (WARN_ON(ret < 0))
2067                 return ret;
2068         ret = 0;
2069
2070         while (1) {
2071                 cond_resched();
2072
2073                 leaf = path->nodes[0];
2074                 slot = path->slots[0];
2075
2076                 if (slot >= btrfs_header_nritems(leaf)) {
2077                         ret = btrfs_next_leaf(root, path);
2078                         if (ret < 0) {
2079                                 goto out;
2080                         } else if (ret > 0) {
2081                                 ret = 0;
2082                                 goto out;
2083                         }
2084                         continue;
2085                 }
2086
2087                 path->slots[0]++;
2088
2089                 btrfs_item_key_to_cpu(leaf, &key, slot);
2090
2091                 if (key.objectid > inum)
2092                         goto out;
2093
2094                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2095                         continue;
2096
2097                 extent = btrfs_item_ptr(leaf, slot,
2098                                         struct btrfs_file_extent_item);
2099
2100                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2101                         continue;
2102
2103                 /*
2104                  * 'offset' refers to the exact key.offset,
2105                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2106                  * (key.offset - extent_offset).
2107                  */
2108                 if (key.offset != offset)
2109                         continue;
2110
2111                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2112                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2113
2114                 if (extent_offset >= old->extent_offset + old->offset +
2115                     old->len || extent_offset + num_bytes <=
2116                     old->extent_offset + old->offset)
2117                         continue;
2118                 break;
2119         }
2120
2121         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2122         if (!backref) {
2123                 ret = -ENOENT;
2124                 goto out;
2125         }
2126
2127         backref->root_id = root_id;
2128         backref->inum = inum;
2129         backref->file_pos = offset;
2130         backref->num_bytes = num_bytes;
2131         backref->extent_offset = extent_offset;
2132         backref->generation = btrfs_file_extent_generation(leaf, extent);
2133         backref->old = old;
2134         backref_insert(&new->root, backref);
2135         old->count++;
2136 out:
2137         btrfs_release_path(path);
2138         WARN_ON(ret);
2139         return ret;
2140 }
2141
2142 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2143                                    struct new_sa_defrag_extent *new)
2144 {
2145         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2146         struct old_sa_defrag_extent *old, *tmp;
2147         int ret;
2148
2149         new->path = path;
2150
2151         list_for_each_entry_safe(old, tmp, &new->head, list) {
2152                 ret = iterate_inodes_from_logical(old->bytenr +
2153                                                   old->extent_offset, fs_info,
2154                                                   path, record_one_backref,
2155                                                   old);
2156                 if (ret < 0 && ret != -ENOENT)
2157                         return false;
2158
2159                 /* no backref to be processed for this extent */
2160                 if (!old->count) {
2161                         list_del(&old->list);
2162                         kfree(old);
2163                 }
2164         }
2165
2166         if (list_empty(&new->head))
2167                 return false;
2168
2169         return true;
2170 }
2171
2172 static int relink_is_mergable(struct extent_buffer *leaf,
2173                               struct btrfs_file_extent_item *fi,
2174                               struct new_sa_defrag_extent *new)
2175 {
2176         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2177                 return 0;
2178
2179         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2180                 return 0;
2181
2182         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2183                 return 0;
2184
2185         if (btrfs_file_extent_encryption(leaf, fi) ||
2186             btrfs_file_extent_other_encoding(leaf, fi))
2187                 return 0;
2188
2189         return 1;
2190 }
2191
2192 /*
2193  * Note the backref might has changed, and in this case we just return 0.
2194  */
2195 static noinline int relink_extent_backref(struct btrfs_path *path,
2196                                  struct sa_defrag_extent_backref *prev,
2197                                  struct sa_defrag_extent_backref *backref)
2198 {
2199         struct btrfs_file_extent_item *extent;
2200         struct btrfs_file_extent_item *item;
2201         struct btrfs_ordered_extent *ordered;
2202         struct btrfs_trans_handle *trans;
2203         struct btrfs_fs_info *fs_info;
2204         struct btrfs_root *root;
2205         struct btrfs_key key;
2206         struct extent_buffer *leaf;
2207         struct old_sa_defrag_extent *old = backref->old;
2208         struct new_sa_defrag_extent *new = old->new;
2209         struct inode *src_inode = new->inode;
2210         struct inode *inode;
2211         struct extent_state *cached = NULL;
2212         int ret = 0;
2213         u64 start;
2214         u64 len;
2215         u64 lock_start;
2216         u64 lock_end;
2217         bool merge = false;
2218         int index;
2219
2220         if (prev && prev->root_id == backref->root_id &&
2221             prev->inum == backref->inum &&
2222             prev->file_pos + prev->num_bytes == backref->file_pos)
2223                 merge = true;
2224
2225         /* step 1: get root */
2226         key.objectid = backref->root_id;
2227         key.type = BTRFS_ROOT_ITEM_KEY;
2228         key.offset = (u64)-1;
2229
2230         fs_info = BTRFS_I(src_inode)->root->fs_info;
2231         index = srcu_read_lock(&fs_info->subvol_srcu);
2232
2233         root = btrfs_read_fs_root_no_name(fs_info, &key);
2234         if (IS_ERR(root)) {
2235                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2236                 if (PTR_ERR(root) == -ENOENT)
2237                         return 0;
2238                 return PTR_ERR(root);
2239         }
2240
2241         /* step 2: get inode */
2242         key.objectid = backref->inum;
2243         key.type = BTRFS_INODE_ITEM_KEY;
2244         key.offset = 0;
2245
2246         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2247         if (IS_ERR(inode)) {
2248                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2249                 return 0;
2250         }
2251
2252         srcu_read_unlock(&fs_info->subvol_srcu, index);
2253
2254         /* step 3: relink backref */
2255         lock_start = backref->file_pos;
2256         lock_end = backref->file_pos + backref->num_bytes - 1;
2257         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2258                          0, &cached);
2259
2260         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2261         if (ordered) {
2262                 btrfs_put_ordered_extent(ordered);
2263                 goto out_unlock;
2264         }
2265
2266         trans = btrfs_join_transaction(root);
2267         if (IS_ERR(trans)) {
2268                 ret = PTR_ERR(trans);
2269                 goto out_unlock;
2270         }
2271
2272         key.objectid = backref->inum;
2273         key.type = BTRFS_EXTENT_DATA_KEY;
2274         key.offset = backref->file_pos;
2275
2276         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2277         if (ret < 0) {
2278                 goto out_free_path;
2279         } else if (ret > 0) {
2280                 ret = 0;
2281                 goto out_free_path;
2282         }
2283
2284         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2285                                 struct btrfs_file_extent_item);
2286
2287         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2288             backref->generation)
2289                 goto out_free_path;
2290
2291         btrfs_release_path(path);
2292
2293         start = backref->file_pos;
2294         if (backref->extent_offset < old->extent_offset + old->offset)
2295                 start += old->extent_offset + old->offset -
2296                          backref->extent_offset;
2297
2298         len = min(backref->extent_offset + backref->num_bytes,
2299                   old->extent_offset + old->offset + old->len);
2300         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2301
2302         ret = btrfs_drop_extents(trans, root, inode, start,
2303                                  start + len, 1);
2304         if (ret)
2305                 goto out_free_path;
2306 again:
2307         key.objectid = btrfs_ino(inode);
2308         key.type = BTRFS_EXTENT_DATA_KEY;
2309         key.offset = start;
2310
2311         path->leave_spinning = 1;
2312         if (merge) {
2313                 struct btrfs_file_extent_item *fi;
2314                 u64 extent_len;
2315                 struct btrfs_key found_key;
2316
2317                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2318                 if (ret < 0)
2319                         goto out_free_path;
2320
2321                 path->slots[0]--;
2322                 leaf = path->nodes[0];
2323                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2324
2325                 fi = btrfs_item_ptr(leaf, path->slots[0],
2326                                     struct btrfs_file_extent_item);
2327                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2328
2329                 if (extent_len + found_key.offset == start &&
2330                     relink_is_mergable(leaf, fi, new)) {
2331                         btrfs_set_file_extent_num_bytes(leaf, fi,
2332                                                         extent_len + len);
2333                         btrfs_mark_buffer_dirty(leaf);
2334                         inode_add_bytes(inode, len);
2335
2336                         ret = 1;
2337                         goto out_free_path;
2338                 } else {
2339                         merge = false;
2340                         btrfs_release_path(path);
2341                         goto again;
2342                 }
2343         }
2344
2345         ret = btrfs_insert_empty_item(trans, root, path, &key,
2346                                         sizeof(*extent));
2347         if (ret) {
2348                 btrfs_abort_transaction(trans, root, ret);
2349                 goto out_free_path;
2350         }
2351
2352         leaf = path->nodes[0];
2353         item = btrfs_item_ptr(leaf, path->slots[0],
2354                                 struct btrfs_file_extent_item);
2355         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2356         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2357         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2358         btrfs_set_file_extent_num_bytes(leaf, item, len);
2359         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2360         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2361         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2362         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2363         btrfs_set_file_extent_encryption(leaf, item, 0);
2364         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2365
2366         btrfs_mark_buffer_dirty(leaf);
2367         inode_add_bytes(inode, len);
2368         btrfs_release_path(path);
2369
2370         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2371                         new->disk_len, 0,
2372                         backref->root_id, backref->inum,
2373                         new->file_pos, 0);      /* start - extent_offset */
2374         if (ret) {
2375                 btrfs_abort_transaction(trans, root, ret);
2376                 goto out_free_path;
2377         }
2378
2379         ret = 1;
2380 out_free_path:
2381         btrfs_release_path(path);
2382         path->leave_spinning = 0;
2383         btrfs_end_transaction(trans, root);
2384 out_unlock:
2385         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2386                              &cached, GFP_NOFS);
2387         iput(inode);
2388         return ret;
2389 }
2390
2391 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2392 {
2393         struct old_sa_defrag_extent *old, *tmp;
2394
2395         if (!new)
2396                 return;
2397
2398         list_for_each_entry_safe(old, tmp, &new->head, list) {
2399                 list_del(&old->list);
2400                 kfree(old);
2401         }
2402         kfree(new);
2403 }
2404
2405 static void relink_file_extents(struct new_sa_defrag_extent *new)
2406 {
2407         struct btrfs_path *path;
2408         struct sa_defrag_extent_backref *backref;
2409         struct sa_defrag_extent_backref *prev = NULL;
2410         struct inode *inode;
2411         struct btrfs_root *root;
2412         struct rb_node *node;
2413         int ret;
2414
2415         inode = new->inode;
2416         root = BTRFS_I(inode)->root;
2417
2418         path = btrfs_alloc_path();
2419         if (!path)
2420                 return;
2421
2422         if (!record_extent_backrefs(path, new)) {
2423                 btrfs_free_path(path);
2424                 goto out;
2425         }
2426         btrfs_release_path(path);
2427
2428         while (1) {
2429                 node = rb_first(&new->root);
2430                 if (!node)
2431                         break;
2432                 rb_erase(node, &new->root);
2433
2434                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2435
2436                 ret = relink_extent_backref(path, prev, backref);
2437                 WARN_ON(ret < 0);
2438
2439                 kfree(prev);
2440
2441                 if (ret == 1)
2442                         prev = backref;
2443                 else
2444                         prev = NULL;
2445                 cond_resched();
2446         }
2447         kfree(prev);
2448
2449         btrfs_free_path(path);
2450 out:
2451         free_sa_defrag_extent(new);
2452
2453         atomic_dec(&root->fs_info->defrag_running);
2454         wake_up(&root->fs_info->transaction_wait);
2455 }
2456
2457 static struct new_sa_defrag_extent *
2458 record_old_file_extents(struct inode *inode,
2459                         struct btrfs_ordered_extent *ordered)
2460 {
2461         struct btrfs_root *root = BTRFS_I(inode)->root;
2462         struct btrfs_path *path;
2463         struct btrfs_key key;
2464         struct old_sa_defrag_extent *old;
2465         struct new_sa_defrag_extent *new;
2466         int ret;
2467
2468         new = kmalloc(sizeof(*new), GFP_NOFS);
2469         if (!new)
2470                 return NULL;
2471
2472         new->inode = inode;
2473         new->file_pos = ordered->file_offset;
2474         new->len = ordered->len;
2475         new->bytenr = ordered->start;
2476         new->disk_len = ordered->disk_len;
2477         new->compress_type = ordered->compress_type;
2478         new->root = RB_ROOT;
2479         INIT_LIST_HEAD(&new->head);
2480
2481         path = btrfs_alloc_path();
2482         if (!path)
2483                 goto out_kfree;
2484
2485         key.objectid = btrfs_ino(inode);
2486         key.type = BTRFS_EXTENT_DATA_KEY;
2487         key.offset = new->file_pos;
2488
2489         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2490         if (ret < 0)
2491                 goto out_free_path;
2492         if (ret > 0 && path->slots[0] > 0)
2493                 path->slots[0]--;
2494
2495         /* find out all the old extents for the file range */
2496         while (1) {
2497                 struct btrfs_file_extent_item *extent;
2498                 struct extent_buffer *l;
2499                 int slot;
2500                 u64 num_bytes;
2501                 u64 offset;
2502                 u64 end;
2503                 u64 disk_bytenr;
2504                 u64 extent_offset;
2505
2506                 l = path->nodes[0];
2507                 slot = path->slots[0];
2508
2509                 if (slot >= btrfs_header_nritems(l)) {
2510                         ret = btrfs_next_leaf(root, path);
2511                         if (ret < 0)
2512                                 goto out_free_path;
2513                         else if (ret > 0)
2514                                 break;
2515                         continue;
2516                 }
2517
2518                 btrfs_item_key_to_cpu(l, &key, slot);
2519
2520                 if (key.objectid != btrfs_ino(inode))
2521                         break;
2522                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2523                         break;
2524                 if (key.offset >= new->file_pos + new->len)
2525                         break;
2526
2527                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2528
2529                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2530                 if (key.offset + num_bytes < new->file_pos)
2531                         goto next;
2532
2533                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2534                 if (!disk_bytenr)
2535                         goto next;
2536
2537                 extent_offset = btrfs_file_extent_offset(l, extent);
2538
2539                 old = kmalloc(sizeof(*old), GFP_NOFS);
2540                 if (!old)
2541                         goto out_free_path;
2542
2543                 offset = max(new->file_pos, key.offset);
2544                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2545
2546                 old->bytenr = disk_bytenr;
2547                 old->extent_offset = extent_offset;
2548                 old->offset = offset - key.offset;
2549                 old->len = end - offset;
2550                 old->new = new;
2551                 old->count = 0;
2552                 list_add_tail(&old->list, &new->head);
2553 next:
2554                 path->slots[0]++;
2555                 cond_resched();
2556         }
2557
2558         btrfs_free_path(path);
2559         atomic_inc(&root->fs_info->defrag_running);
2560
2561         return new;
2562
2563 out_free_path:
2564         btrfs_free_path(path);
2565 out_kfree:
2566         free_sa_defrag_extent(new);
2567         return NULL;
2568 }
2569
2570 /* as ordered data IO finishes, this gets called so we can finish
2571  * an ordered extent if the range of bytes in the file it covers are
2572  * fully written.
2573  */
2574 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2575 {
2576         struct inode *inode = ordered_extent->inode;
2577         struct btrfs_root *root = BTRFS_I(inode)->root;
2578         struct btrfs_trans_handle *trans = NULL;
2579         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2580         struct extent_state *cached_state = NULL;
2581         struct new_sa_defrag_extent *new = NULL;
2582         int compress_type = 0;
2583         int ret = 0;
2584         u64 logical_len = ordered_extent->len;
2585         bool nolock;
2586         bool truncated = false;
2587
2588         nolock = btrfs_is_free_space_inode(inode);
2589
2590         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2591                 ret = -EIO;
2592                 goto out;
2593         }
2594
2595         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2596                 truncated = true;
2597                 logical_len = ordered_extent->truncated_len;
2598                 /* Truncated the entire extent, don't bother adding */
2599                 if (!logical_len)
2600                         goto out;
2601         }
2602
2603         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2604                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2605                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2606                 if (nolock)
2607                         trans = btrfs_join_transaction_nolock(root);
2608                 else
2609                         trans = btrfs_join_transaction(root);
2610                 if (IS_ERR(trans)) {
2611                         ret = PTR_ERR(trans);
2612                         trans = NULL;
2613                         goto out;
2614                 }
2615                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2616                 ret = btrfs_update_inode_fallback(trans, root, inode);
2617                 if (ret) /* -ENOMEM or corruption */
2618                         btrfs_abort_transaction(trans, root, ret);
2619                 goto out;
2620         }
2621
2622         lock_extent_bits(io_tree, ordered_extent->file_offset,
2623                          ordered_extent->file_offset + ordered_extent->len - 1,
2624                          0, &cached_state);
2625
2626         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2627                         ordered_extent->file_offset + ordered_extent->len - 1,
2628                         EXTENT_DEFRAG, 1, cached_state);
2629         if (ret) {
2630                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2631                 if (last_snapshot >= BTRFS_I(inode)->generation)
2632                         /* the inode is shared */
2633                         new = record_old_file_extents(inode, ordered_extent);
2634
2635                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2636                         ordered_extent->file_offset + ordered_extent->len - 1,
2637                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2638         }
2639
2640         if (nolock)
2641                 trans = btrfs_join_transaction_nolock(root);
2642         else
2643                 trans = btrfs_join_transaction(root);
2644         if (IS_ERR(trans)) {
2645                 ret = PTR_ERR(trans);
2646                 trans = NULL;
2647                 goto out_unlock;
2648         }
2649         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2650
2651         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2652                 compress_type = ordered_extent->compress_type;
2653         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2654                 BUG_ON(compress_type);
2655                 ret = btrfs_mark_extent_written(trans, inode,
2656                                                 ordered_extent->file_offset,
2657                                                 ordered_extent->file_offset +
2658                                                 logical_len);
2659         } else {
2660                 BUG_ON(root == root->fs_info->tree_root);
2661                 ret = insert_reserved_file_extent(trans, inode,
2662                                                 ordered_extent->file_offset,
2663                                                 ordered_extent->start,
2664                                                 ordered_extent->disk_len,
2665                                                 logical_len, logical_len,
2666                                                 compress_type, 0, 0,
2667                                                 BTRFS_FILE_EXTENT_REG);
2668         }
2669         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2670                            ordered_extent->file_offset, ordered_extent->len,
2671                            trans->transid);
2672         if (ret < 0) {
2673                 btrfs_abort_transaction(trans, root, ret);
2674                 goto out_unlock;
2675         }
2676
2677         add_pending_csums(trans, inode, ordered_extent->file_offset,
2678                           &ordered_extent->list);
2679
2680         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2681         ret = btrfs_update_inode_fallback(trans, root, inode);
2682         if (ret) { /* -ENOMEM or corruption */
2683                 btrfs_abort_transaction(trans, root, ret);
2684                 goto out_unlock;
2685         }
2686         ret = 0;
2687 out_unlock:
2688         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2689                              ordered_extent->file_offset +
2690                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2691 out:
2692         if (root != root->fs_info->tree_root)
2693                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2694         if (trans)
2695                 btrfs_end_transaction(trans, root);
2696
2697         if (ret || truncated) {
2698                 u64 start, end;
2699
2700                 if (truncated)
2701                         start = ordered_extent->file_offset + logical_len;
2702                 else
2703                         start = ordered_extent->file_offset;
2704                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2705                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2706
2707                 /* Drop the cache for the part of the extent we didn't write. */
2708                 btrfs_drop_extent_cache(inode, start, end, 0);
2709
2710                 /*
2711                  * If the ordered extent had an IOERR or something else went
2712                  * wrong we need to return the space for this ordered extent
2713                  * back to the allocator.  We only free the extent in the
2714                  * truncated case if we didn't write out the extent at all.
2715                  */
2716                 if ((ret || !logical_len) &&
2717                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2718                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2719                         btrfs_free_reserved_extent(root, ordered_extent->start,
2720                                                    ordered_extent->disk_len);
2721         }
2722
2723
2724         /*
2725          * This needs to be done to make sure anybody waiting knows we are done
2726          * updating everything for this ordered extent.
2727          */
2728         btrfs_remove_ordered_extent(inode, ordered_extent);
2729
2730         /* for snapshot-aware defrag */
2731         if (new) {
2732                 if (ret) {
2733                         free_sa_defrag_extent(new);
2734                         atomic_dec(&root->fs_info->defrag_running);
2735                 } else {
2736                         relink_file_extents(new);
2737                 }
2738         }
2739
2740         /* once for us */
2741         btrfs_put_ordered_extent(ordered_extent);
2742         /* once for the tree */
2743         btrfs_put_ordered_extent(ordered_extent);
2744
2745         return ret;
2746 }
2747
2748 static void finish_ordered_fn(struct btrfs_work *work)
2749 {
2750         struct btrfs_ordered_extent *ordered_extent;
2751         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2752         btrfs_finish_ordered_io(ordered_extent);
2753 }
2754
2755 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2756                                 struct extent_state *state, int uptodate)
2757 {
2758         struct inode *inode = page->mapping->host;
2759         struct btrfs_root *root = BTRFS_I(inode)->root;
2760         struct btrfs_ordered_extent *ordered_extent = NULL;
2761         struct btrfs_workers *workers;
2762
2763         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2764
2765         ClearPagePrivate2(page);
2766         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2767                                             end - start + 1, uptodate))
2768                 return 0;
2769
2770         ordered_extent->work.func = finish_ordered_fn;
2771         ordered_extent->work.flags = 0;
2772
2773         if (btrfs_is_free_space_inode(inode))
2774                 workers = &root->fs_info->endio_freespace_worker;
2775         else
2776                 workers = &root->fs_info->endio_write_workers;
2777         btrfs_queue_worker(workers, &ordered_extent->work);
2778
2779         return 0;
2780 }
2781
2782 /*
2783  * when reads are done, we need to check csums to verify the data is correct
2784  * if there's a match, we allow the bio to finish.  If not, the code in
2785  * extent_io.c will try to find good copies for us.
2786  */
2787 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2788                                       u64 phy_offset, struct page *page,
2789                                       u64 start, u64 end, int mirror)
2790 {
2791         size_t offset = start - page_offset(page);
2792         struct inode *inode = page->mapping->host;
2793         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2794         char *kaddr;
2795         struct btrfs_root *root = BTRFS_I(inode)->root;
2796         u32 csum_expected;
2797         u32 csum = ~(u32)0;
2798         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2799                                       DEFAULT_RATELIMIT_BURST);
2800
2801         if (PageChecked(page)) {
2802                 ClearPageChecked(page);
2803                 goto good;
2804         }
2805
2806         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2807                 goto good;
2808
2809         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2810             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2811                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2812                                   GFP_NOFS);
2813                 return 0;
2814         }
2815
2816         phy_offset >>= inode->i_sb->s_blocksize_bits;
2817         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2818
2819         kaddr = kmap_atomic(page);
2820         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2821         btrfs_csum_final(csum, (char *)&csum);
2822         if (csum != csum_expected)
2823                 goto zeroit;
2824
2825         kunmap_atomic(kaddr);
2826 good:
2827         return 0;
2828
2829 zeroit:
2830         if (__ratelimit(&_rs))
2831                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2832                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2833         memset(kaddr + offset, 1, end - start + 1);
2834         flush_dcache_page(page);
2835         kunmap_atomic(kaddr);
2836         if (csum_expected == 0)
2837                 return 0;
2838         return -EIO;
2839 }
2840
2841 struct delayed_iput {
2842         struct list_head list;
2843         struct inode *inode;
2844 };
2845
2846 /* JDM: If this is fs-wide, why can't we add a pointer to
2847  * btrfs_inode instead and avoid the allocation? */
2848 void btrfs_add_delayed_iput(struct inode *inode)
2849 {
2850         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2851         struct delayed_iput *delayed;
2852
2853         if (atomic_add_unless(&inode->i_count, -1, 1))
2854                 return;
2855
2856         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2857         delayed->inode = inode;
2858
2859         spin_lock(&fs_info->delayed_iput_lock);
2860         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2861         spin_unlock(&fs_info->delayed_iput_lock);
2862 }
2863
2864 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2865 {
2866         LIST_HEAD(list);
2867         struct btrfs_fs_info *fs_info = root->fs_info;
2868         struct delayed_iput *delayed;
2869         int empty;
2870
2871         spin_lock(&fs_info->delayed_iput_lock);
2872         empty = list_empty(&fs_info->delayed_iputs);
2873         spin_unlock(&fs_info->delayed_iput_lock);
2874         if (empty)
2875                 return;
2876
2877         spin_lock(&fs_info->delayed_iput_lock);
2878         list_splice_init(&fs_info->delayed_iputs, &list);
2879         spin_unlock(&fs_info->delayed_iput_lock);
2880
2881         while (!list_empty(&list)) {
2882                 delayed = list_entry(list.next, struct delayed_iput, list);
2883                 list_del(&delayed->list);
2884                 iput(delayed->inode);
2885                 kfree(delayed);
2886         }
2887 }
2888
2889 /*
2890  * This is called in transaction commit time. If there are no orphan
2891  * files in the subvolume, it removes orphan item and frees block_rsv
2892  * structure.
2893  */
2894 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2895                               struct btrfs_root *root)
2896 {
2897         struct btrfs_block_rsv *block_rsv;
2898         int ret;
2899
2900         if (atomic_read(&root->orphan_inodes) ||
2901             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2902                 return;
2903
2904         spin_lock(&root->orphan_lock);
2905         if (atomic_read(&root->orphan_inodes)) {
2906                 spin_unlock(&root->orphan_lock);
2907                 return;
2908         }
2909
2910         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2911                 spin_unlock(&root->orphan_lock);
2912                 return;
2913         }
2914
2915         block_rsv = root->orphan_block_rsv;
2916         root->orphan_block_rsv = NULL;
2917         spin_unlock(&root->orphan_lock);
2918
2919         if (root->orphan_item_inserted &&
2920             btrfs_root_refs(&root->root_item) > 0) {
2921                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2922                                             root->root_key.objectid);
2923                 if (ret)
2924                         btrfs_abort_transaction(trans, root, ret);
2925                 else
2926                         root->orphan_item_inserted = 0;
2927         }
2928
2929         if (block_rsv) {
2930                 WARN_ON(block_rsv->size > 0);
2931                 btrfs_free_block_rsv(root, block_rsv);
2932         }
2933 }
2934
2935 /*
2936  * This creates an orphan entry for the given inode in case something goes
2937  * wrong in the middle of an unlink/truncate.
2938  *
2939  * NOTE: caller of this function should reserve 5 units of metadata for
2940  *       this function.
2941  */
2942 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2943 {
2944         struct btrfs_root *root = BTRFS_I(inode)->root;
2945         struct btrfs_block_rsv *block_rsv = NULL;
2946         int reserve = 0;
2947         int insert = 0;
2948         int ret;
2949
2950         if (!root->orphan_block_rsv) {
2951                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2952                 if (!block_rsv)
2953                         return -ENOMEM;
2954         }
2955
2956         spin_lock(&root->orphan_lock);
2957         if (!root->orphan_block_rsv) {
2958                 root->orphan_block_rsv = block_rsv;
2959         } else if (block_rsv) {
2960                 btrfs_free_block_rsv(root, block_rsv);
2961                 block_rsv = NULL;
2962         }
2963
2964         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2965                               &BTRFS_I(inode)->runtime_flags)) {
2966 #if 0
2967                 /*
2968                  * For proper ENOSPC handling, we should do orphan
2969                  * cleanup when mounting. But this introduces backward
2970                  * compatibility issue.
2971                  */
2972                 if (!xchg(&root->orphan_item_inserted, 1))
2973                         insert = 2;
2974                 else
2975                         insert = 1;
2976 #endif
2977                 insert = 1;
2978                 atomic_inc(&root->orphan_inodes);
2979         }
2980
2981         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2982                               &BTRFS_I(inode)->runtime_flags))
2983                 reserve = 1;
2984         spin_unlock(&root->orphan_lock);
2985
2986         /* grab metadata reservation from transaction handle */
2987         if (reserve) {
2988                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2989                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2990         }
2991
2992         /* insert an orphan item to track this unlinked/truncated file */
2993         if (insert >= 1) {
2994                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2995                 if (ret) {
2996                         atomic_dec(&root->orphan_inodes);
2997                         if (reserve) {
2998                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2999                                           &BTRFS_I(inode)->runtime_flags);
3000                                 btrfs_orphan_release_metadata(inode);
3001                         }
3002                         if (ret != -EEXIST) {
3003                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3004                                           &BTRFS_I(inode)->runtime_flags);
3005                                 btrfs_abort_transaction(trans, root, ret);
3006                                 return ret;
3007                         }
3008                 }
3009                 ret = 0;
3010         }
3011
3012         /* insert an orphan item to track subvolume contains orphan files */
3013         if (insert >= 2) {
3014                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3015                                                root->root_key.objectid);
3016                 if (ret && ret != -EEXIST) {
3017                         btrfs_abort_transaction(trans, root, ret);
3018                         return ret;
3019                 }
3020         }
3021         return 0;
3022 }
3023
3024 /*
3025  * We have done the truncate/delete so we can go ahead and remove the orphan
3026  * item for this particular inode.
3027  */
3028 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3029                             struct inode *inode)
3030 {
3031         struct btrfs_root *root = BTRFS_I(inode)->root;
3032         int delete_item = 0;
3033         int release_rsv = 0;
3034         int ret = 0;
3035
3036         spin_lock(&root->orphan_lock);
3037         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3038                                &BTRFS_I(inode)->runtime_flags))
3039                 delete_item = 1;
3040
3041         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3042                                &BTRFS_I(inode)->runtime_flags))
3043                 release_rsv = 1;
3044         spin_unlock(&root->orphan_lock);
3045
3046         if (delete_item) {
3047                 atomic_dec(&root->orphan_inodes);
3048                 if (trans)
3049                         ret = btrfs_del_orphan_item(trans, root,
3050                                                     btrfs_ino(inode));
3051         }
3052
3053         if (release_rsv)
3054                 btrfs_orphan_release_metadata(inode);
3055
3056         return ret;
3057 }
3058
3059 /*
3060  * this cleans up any orphans that may be left on the list from the last use
3061  * of this root.
3062  */
3063 int btrfs_orphan_cleanup(struct btrfs_root *root)
3064 {
3065         struct btrfs_path *path;
3066         struct extent_buffer *leaf;
3067         struct btrfs_key key, found_key;
3068         struct btrfs_trans_handle *trans;
3069         struct inode *inode;
3070         u64 last_objectid = 0;
3071         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3072
3073         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3074                 return 0;
3075
3076         path = btrfs_alloc_path();
3077         if (!path) {
3078                 ret = -ENOMEM;
3079                 goto out;
3080         }
3081         path->reada = -1;
3082
3083         key.objectid = BTRFS_ORPHAN_OBJECTID;
3084         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3085         key.offset = (u64)-1;
3086
3087         while (1) {
3088                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3089                 if (ret < 0)
3090                         goto out;
3091
3092                 /*
3093                  * if ret == 0 means we found what we were searching for, which
3094                  * is weird, but possible, so only screw with path if we didn't
3095                  * find the key and see if we have stuff that matches
3096                  */
3097                 if (ret > 0) {
3098                         ret = 0;
3099                         if (path->slots[0] == 0)
3100                                 break;
3101                         path->slots[0]--;
3102                 }
3103
3104                 /* pull out the item */
3105                 leaf = path->nodes[0];
3106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3107
3108                 /* make sure the item matches what we want */
3109                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3110                         break;
3111                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3112                         break;
3113
3114                 /* release the path since we're done with it */
3115                 btrfs_release_path(path);
3116
3117                 /*
3118                  * this is where we are basically btrfs_lookup, without the
3119                  * crossing root thing.  we store the inode number in the
3120                  * offset of the orphan item.
3121                  */
3122
3123                 if (found_key.offset == last_objectid) {
3124                         btrfs_err(root->fs_info,
3125                                 "Error removing orphan entry, stopping orphan cleanup");
3126                         ret = -EINVAL;
3127                         goto out;
3128                 }
3129
3130                 last_objectid = found_key.offset;
3131
3132                 found_key.objectid = found_key.offset;
3133                 found_key.type = BTRFS_INODE_ITEM_KEY;
3134                 found_key.offset = 0;
3135                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3136                 ret = PTR_ERR_OR_ZERO(inode);
3137                 if (ret && ret != -ESTALE)
3138                         goto out;
3139
3140                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3141                         struct btrfs_root *dead_root;
3142                         struct btrfs_fs_info *fs_info = root->fs_info;
3143                         int is_dead_root = 0;
3144
3145                         /*
3146                          * this is an orphan in the tree root. Currently these
3147                          * could come from 2 sources:
3148                          *  a) a snapshot deletion in progress
3149                          *  b) a free space cache inode
3150                          * We need to distinguish those two, as the snapshot
3151                          * orphan must not get deleted.
3152                          * find_dead_roots already ran before us, so if this
3153                          * is a snapshot deletion, we should find the root
3154                          * in the dead_roots list
3155                          */
3156                         spin_lock(&fs_info->trans_lock);
3157                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3158                                             root_list) {
3159                                 if (dead_root->root_key.objectid ==
3160                                     found_key.objectid) {
3161                                         is_dead_root = 1;
3162                                         break;
3163                                 }
3164                         }
3165                         spin_unlock(&fs_info->trans_lock);
3166                         if (is_dead_root) {
3167                                 /* prevent this orphan from being found again */
3168                                 key.offset = found_key.objectid - 1;
3169                                 continue;
3170                         }
3171                 }
3172                 /*
3173                  * Inode is already gone but the orphan item is still there,
3174                  * kill the orphan item.
3175                  */
3176                 if (ret == -ESTALE) {
3177                         trans = btrfs_start_transaction(root, 1);
3178                         if (IS_ERR(trans)) {
3179                                 ret = PTR_ERR(trans);
3180                                 goto out;
3181                         }
3182                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3183                                 found_key.objectid);
3184                         ret = btrfs_del_orphan_item(trans, root,
3185                                                     found_key.objectid);
3186                         btrfs_end_transaction(trans, root);
3187                         if (ret)
3188                                 goto out;
3189                         continue;
3190                 }
3191
3192                 /*
3193                  * add this inode to the orphan list so btrfs_orphan_del does
3194                  * the proper thing when we hit it
3195                  */
3196                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3197                         &BTRFS_I(inode)->runtime_flags);
3198                 atomic_inc(&root->orphan_inodes);
3199
3200                 /* if we have links, this was a truncate, lets do that */
3201                 if (inode->i_nlink) {
3202                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3203                                 iput(inode);
3204                                 continue;
3205                         }
3206                         nr_truncate++;
3207
3208                         /* 1 for the orphan item deletion. */
3209                         trans = btrfs_start_transaction(root, 1);
3210                         if (IS_ERR(trans)) {
3211                                 iput(inode);
3212                                 ret = PTR_ERR(trans);
3213                                 goto out;
3214                         }
3215                         ret = btrfs_orphan_add(trans, inode);
3216                         btrfs_end_transaction(trans, root);
3217                         if (ret) {
3218                                 iput(inode);
3219                                 goto out;
3220                         }
3221
3222                         ret = btrfs_truncate(inode);
3223                         if (ret)
3224                                 btrfs_orphan_del(NULL, inode);
3225                 } else {
3226                         nr_unlink++;
3227                 }
3228
3229                 /* this will do delete_inode and everything for us */
3230                 iput(inode);
3231                 if (ret)
3232                         goto out;
3233         }
3234         /* release the path since we're done with it */
3235         btrfs_release_path(path);
3236
3237         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3238
3239         if (root->orphan_block_rsv)
3240                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3241                                         (u64)-1);
3242
3243         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3244                 trans = btrfs_join_transaction(root);
3245                 if (!IS_ERR(trans))
3246                         btrfs_end_transaction(trans, root);
3247         }
3248
3249         if (nr_unlink)
3250                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3251         if (nr_truncate)
3252                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3253
3254 out:
3255         if (ret)
3256                 btrfs_crit(root->fs_info,
3257                         "could not do orphan cleanup %d", ret);
3258         btrfs_free_path(path);
3259         return ret;
3260 }
3261
3262 /*
3263  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3264  * don't find any xattrs, we know there can't be any acls.
3265  *
3266  * slot is the slot the inode is in, objectid is the objectid of the inode
3267  */
3268 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3269                                           int slot, u64 objectid,
3270                                           int *first_xattr_slot)
3271 {
3272         u32 nritems = btrfs_header_nritems(leaf);
3273         struct btrfs_key found_key;
3274         static u64 xattr_access = 0;
3275         static u64 xattr_default = 0;
3276         int scanned = 0;
3277
3278         if (!xattr_access) {
3279                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3280                                         strlen(POSIX_ACL_XATTR_ACCESS));
3281                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3282                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3283         }
3284
3285         slot++;
3286         *first_xattr_slot = -1;
3287         while (slot < nritems) {
3288                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3289
3290                 /* we found a different objectid, there must not be acls */
3291                 if (found_key.objectid != objectid)
3292                         return 0;
3293
3294                 /* we found an xattr, assume we've got an acl */
3295                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3296                         if (*first_xattr_slot == -1)
3297                                 *first_xattr_slot = slot;
3298                         if (found_key.offset == xattr_access ||
3299                             found_key.offset == xattr_default)
3300                                 return 1;
3301                 }
3302
3303                 /*
3304                  * we found a key greater than an xattr key, there can't
3305                  * be any acls later on
3306                  */
3307                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3308                         return 0;
3309
3310                 slot++;
3311                 scanned++;
3312
3313                 /*
3314                  * it goes inode, inode backrefs, xattrs, extents,
3315                  * so if there are a ton of hard links to an inode there can
3316                  * be a lot of backrefs.  Don't waste time searching too hard,
3317                  * this is just an optimization
3318                  */
3319                 if (scanned >= 8)
3320                         break;
3321         }
3322         /* we hit the end of the leaf before we found an xattr or
3323          * something larger than an xattr.  We have to assume the inode
3324          * has acls
3325          */
3326         if (*first_xattr_slot == -1)
3327                 *first_xattr_slot = slot;
3328         return 1;
3329 }
3330
3331 /*
3332  * read an inode from the btree into the in-memory inode
3333  */
3334 static void btrfs_read_locked_inode(struct inode *inode)
3335 {
3336         struct btrfs_path *path;
3337         struct extent_buffer *leaf;
3338         struct btrfs_inode_item *inode_item;
3339         struct btrfs_timespec *tspec;
3340         struct btrfs_root *root = BTRFS_I(inode)->root;
3341         struct btrfs_key location;
3342         unsigned long ptr;
3343         int maybe_acls;
3344         u32 rdev;
3345         int ret;
3346         bool filled = false;
3347         int first_xattr_slot;
3348
3349         ret = btrfs_fill_inode(inode, &rdev);
3350         if (!ret)
3351                 filled = true;
3352
3353         path = btrfs_alloc_path();
3354         if (!path)
3355                 goto make_bad;
3356
3357         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3358
3359         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3360         if (ret)
3361                 goto make_bad;
3362
3363         leaf = path->nodes[0];
3364
3365         if (filled)
3366                 goto cache_index;
3367
3368         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3369                                     struct btrfs_inode_item);
3370         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3371         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3372         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3373         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3374         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3375
3376         tspec = btrfs_inode_atime(inode_item);
3377         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3378         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3379
3380         tspec = btrfs_inode_mtime(inode_item);
3381         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3382         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3383
3384         tspec = btrfs_inode_ctime(inode_item);
3385         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3386         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3387
3388         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3389         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3390         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3391
3392         /*
3393          * If we were modified in the current generation and evicted from memory
3394          * and then re-read we need to do a full sync since we don't have any
3395          * idea about which extents were modified before we were evicted from
3396          * cache.
3397          */
3398         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3399                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3400                         &BTRFS_I(inode)->runtime_flags);
3401
3402         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3403         inode->i_generation = BTRFS_I(inode)->generation;
3404         inode->i_rdev = 0;
3405         rdev = btrfs_inode_rdev(leaf, inode_item);
3406
3407         BTRFS_I(inode)->index_cnt = (u64)-1;
3408         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3409
3410 cache_index:
3411         path->slots[0]++;
3412         if (inode->i_nlink != 1 ||
3413             path->slots[0] >= btrfs_header_nritems(leaf))
3414                 goto cache_acl;
3415
3416         btrfs_item_key_to_cpu(leaf, &location, path->slots[0]);
3417         if (location.objectid != btrfs_ino(inode))
3418                 goto cache_acl;
3419
3420         ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3421         if (location.type == BTRFS_INODE_REF_KEY) {
3422                 struct btrfs_inode_ref *ref;
3423
3424                 ref = (struct btrfs_inode_ref *)ptr;
3425                 BTRFS_I(inode)->dir_index = btrfs_inode_ref_index(leaf, ref);
3426         } else if (location.type == BTRFS_INODE_EXTREF_KEY) {
3427                 struct btrfs_inode_extref *extref;
3428
3429                 extref = (struct btrfs_inode_extref *)ptr;
3430                 BTRFS_I(inode)->dir_index = btrfs_inode_extref_index(leaf,
3431                                                                      extref);
3432         }
3433 cache_acl:
3434         /*
3435          * try to precache a NULL acl entry for files that don't have
3436          * any xattrs or acls
3437          */
3438         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3439                                            btrfs_ino(inode), &first_xattr_slot);
3440         if (first_xattr_slot != -1) {
3441                 path->slots[0] = first_xattr_slot;
3442                 ret = btrfs_load_inode_props(inode, path);
3443                 if (ret)
3444                         btrfs_err(root->fs_info,
3445                                   "error loading props for ino %llu (root %llu): %d\n",
3446                                   btrfs_ino(inode),
3447                                   root->root_key.objectid, ret);
3448         }
3449         btrfs_free_path(path);
3450
3451         if (!maybe_acls)
3452                 cache_no_acl(inode);
3453
3454         switch (inode->i_mode & S_IFMT) {
3455         case S_IFREG:
3456                 inode->i_mapping->a_ops = &btrfs_aops;
3457                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3458                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3459                 inode->i_fop = &btrfs_file_operations;
3460                 inode->i_op = &btrfs_file_inode_operations;
3461                 break;
3462         case S_IFDIR:
3463                 inode->i_fop = &btrfs_dir_file_operations;
3464                 if (root == root->fs_info->tree_root)
3465                         inode->i_op = &btrfs_dir_ro_inode_operations;
3466                 else
3467                         inode->i_op = &btrfs_dir_inode_operations;
3468                 break;
3469         case S_IFLNK:
3470                 inode->i_op = &btrfs_symlink_inode_operations;
3471                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3472                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3473                 break;
3474         default:
3475                 inode->i_op = &btrfs_special_inode_operations;
3476                 init_special_inode(inode, inode->i_mode, rdev);
3477                 break;
3478         }
3479
3480         btrfs_update_iflags(inode);
3481         return;
3482
3483 make_bad:
3484         btrfs_free_path(path);
3485         make_bad_inode(inode);
3486 }
3487
3488 /*
3489  * given a leaf and an inode, copy the inode fields into the leaf
3490  */
3491 static void fill_inode_item(struct btrfs_trans_handle *trans,
3492                             struct extent_buffer *leaf,
3493                             struct btrfs_inode_item *item,
3494                             struct inode *inode)
3495 {
3496         struct btrfs_map_token token;
3497
3498         btrfs_init_map_token(&token);
3499
3500         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3501         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3502         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3503                                    &token);
3504         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3505         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3506
3507         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3508                                      inode->i_atime.tv_sec, &token);
3509         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3510                                       inode->i_atime.tv_nsec, &token);
3511
3512         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3513                                      inode->i_mtime.tv_sec, &token);
3514         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3515                                       inode->i_mtime.tv_nsec, &token);
3516
3517         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3518                                      inode->i_ctime.tv_sec, &token);
3519         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3520                                       inode->i_ctime.tv_nsec, &token);
3521
3522         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3523                                      &token);
3524         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3525                                          &token);
3526         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3527         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3528         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3529         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3530         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3531 }
3532
3533 /*
3534  * copy everything in the in-memory inode into the btree.
3535  */
3536 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3537                                 struct btrfs_root *root, struct inode *inode)
3538 {
3539         struct btrfs_inode_item *inode_item;
3540         struct btrfs_path *path;
3541         struct extent_buffer *leaf;
3542         int ret;
3543
3544         path = btrfs_alloc_path();
3545         if (!path)
3546                 return -ENOMEM;
3547
3548         path->leave_spinning = 1;
3549         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3550                                  1);
3551         if (ret) {
3552                 if (ret > 0)
3553                         ret = -ENOENT;
3554                 goto failed;
3555         }
3556
3557         leaf = path->nodes[0];
3558         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3559                                     struct btrfs_inode_item);
3560
3561         fill_inode_item(trans, leaf, inode_item, inode);
3562         btrfs_mark_buffer_dirty(leaf);
3563         btrfs_set_inode_last_trans(trans, inode);
3564         ret = 0;
3565 failed:
3566         btrfs_free_path(path);
3567         return ret;
3568 }
3569
3570 /*
3571  * copy everything in the in-memory inode into the btree.
3572  */
3573 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3574                                 struct btrfs_root *root, struct inode *inode)
3575 {
3576         int ret;
3577
3578         /*
3579          * If the inode is a free space inode, we can deadlock during commit
3580          * if we put it into the delayed code.
3581          *
3582          * The data relocation inode should also be directly updated
3583          * without delay
3584          */
3585         if (!btrfs_is_free_space_inode(inode)
3586             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3587                 btrfs_update_root_times(trans, root);
3588
3589                 ret = btrfs_delayed_update_inode(trans, root, inode);
3590                 if (!ret)
3591                         btrfs_set_inode_last_trans(trans, inode);
3592                 return ret;
3593         }
3594
3595         return btrfs_update_inode_item(trans, root, inode);
3596 }
3597
3598 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3599                                          struct btrfs_root *root,
3600                                          struct inode *inode)
3601 {
3602         int ret;
3603
3604         ret = btrfs_update_inode(trans, root, inode);
3605         if (ret == -ENOSPC)
3606                 return btrfs_update_inode_item(trans, root, inode);
3607         return ret;
3608 }
3609
3610 /*
3611  * unlink helper that gets used here in inode.c and in the tree logging
3612  * recovery code.  It remove a link in a directory with a given name, and
3613  * also drops the back refs in the inode to the directory
3614  */
3615 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3616                                 struct btrfs_root *root,
3617                                 struct inode *dir, struct inode *inode,
3618                                 const char *name, int name_len)
3619 {
3620         struct btrfs_path *path;
3621         int ret = 0;
3622         struct extent_buffer *leaf;
3623         struct btrfs_dir_item *di;
3624         struct btrfs_key key;
3625         u64 index;
3626         u64 ino = btrfs_ino(inode);
3627         u64 dir_ino = btrfs_ino(dir);
3628
3629         path = btrfs_alloc_path();
3630         if (!path) {
3631                 ret = -ENOMEM;
3632                 goto out;
3633         }
3634
3635         path->leave_spinning = 1;
3636         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3637                                     name, name_len, -1);
3638         if (IS_ERR(di)) {
3639                 ret = PTR_ERR(di);
3640                 goto err;
3641         }
3642         if (!di) {
3643                 ret = -ENOENT;
3644                 goto err;
3645         }
3646         leaf = path->nodes[0];
3647         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3648         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3649         if (ret)
3650                 goto err;
3651         btrfs_release_path(path);
3652
3653         /*
3654          * If we don't have dir index, we have to get it by looking up
3655          * the inode ref, since we get the inode ref, remove it directly,
3656          * it is unnecessary to do delayed deletion.
3657          *
3658          * But if we have dir index, needn't search inode ref to get it.
3659          * Since the inode ref is close to the inode item, it is better
3660          * that we delay to delete it, and just do this deletion when
3661          * we update the inode item.
3662          */
3663         if (BTRFS_I(inode)->dir_index) {
3664                 ret = btrfs_delayed_delete_inode_ref(inode);
3665                 if (!ret) {
3666                         index = BTRFS_I(inode)->dir_index;
3667                         goto skip_backref;
3668                 }
3669         }
3670
3671         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3672                                   dir_ino, &index);
3673         if (ret) {
3674                 btrfs_info(root->fs_info,
3675                         "failed to delete reference to %.*s, inode %llu parent %llu",
3676                         name_len, name, ino, dir_ino);
3677                 btrfs_abort_transaction(trans, root, ret);
3678                 goto err;
3679         }
3680 skip_backref:
3681         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3682         if (ret) {
3683                 btrfs_abort_transaction(trans, root, ret);
3684                 goto err;
3685         }
3686
3687         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3688                                          inode, dir_ino);
3689         if (ret != 0 && ret != -ENOENT) {
3690                 btrfs_abort_transaction(trans, root, ret);
3691                 goto err;
3692         }
3693
3694         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3695                                            dir, index);
3696         if (ret == -ENOENT)
3697                 ret = 0;
3698         else if (ret)
3699                 btrfs_abort_transaction(trans, root, ret);
3700 err:
3701         btrfs_free_path(path);
3702         if (ret)
3703                 goto out;
3704
3705         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3706         inode_inc_iversion(inode);
3707         inode_inc_iversion(dir);
3708         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3709         ret = btrfs_update_inode(trans, root, dir);
3710 out:
3711         return ret;
3712 }
3713
3714 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3715                        struct btrfs_root *root,
3716                        struct inode *dir, struct inode *inode,
3717                        const char *name, int name_len)
3718 {
3719         int ret;
3720         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3721         if (!ret) {
3722                 drop_nlink(inode);
3723                 ret = btrfs_update_inode(trans, root, inode);
3724         }
3725         return ret;
3726 }
3727
3728 /*
3729  * helper to start transaction for unlink and rmdir.
3730  *
3731  * unlink and rmdir are special in btrfs, they do not always free space, so
3732  * if we cannot make our reservations the normal way try and see if there is
3733  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3734  * allow the unlink to occur.
3735  */
3736 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3737 {
3738         struct btrfs_trans_handle *trans;
3739         struct btrfs_root *root = BTRFS_I(dir)->root;
3740         int ret;
3741
3742         /*
3743          * 1 for the possible orphan item
3744          * 1 for the dir item
3745          * 1 for the dir index
3746          * 1 for the inode ref
3747          * 1 for the inode
3748          */
3749         trans = btrfs_start_transaction(root, 5);
3750         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3751                 return trans;
3752
3753         if (PTR_ERR(trans) == -ENOSPC) {
3754                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3755
3756                 trans = btrfs_start_transaction(root, 0);
3757                 if (IS_ERR(trans))
3758                         return trans;
3759                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3760                                                &root->fs_info->trans_block_rsv,
3761                                                num_bytes, 5);
3762                 if (ret) {
3763                         btrfs_end_transaction(trans, root);
3764                         return ERR_PTR(ret);
3765                 }
3766                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3767                 trans->bytes_reserved = num_bytes;
3768         }
3769         return trans;
3770 }
3771
3772 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3773 {
3774         struct btrfs_root *root = BTRFS_I(dir)->root;
3775         struct btrfs_trans_handle *trans;
3776         struct inode *inode = dentry->d_inode;
3777         int ret;
3778
3779         trans = __unlink_start_trans(dir);
3780         if (IS_ERR(trans))
3781                 return PTR_ERR(trans);
3782
3783         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3784
3785         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3786                                  dentry->d_name.name, dentry->d_name.len);
3787         if (ret)
3788                 goto out;
3789
3790         if (inode->i_nlink == 0) {
3791                 ret = btrfs_orphan_add(trans, inode);
3792                 if (ret)
3793                         goto out;
3794         }
3795
3796 out:
3797         btrfs_end_transaction(trans, root);
3798         btrfs_btree_balance_dirty(root);
3799         return ret;
3800 }
3801
3802 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3803                         struct btrfs_root *root,
3804                         struct inode *dir, u64 objectid,
3805                         const char *name, int name_len)
3806 {
3807         struct btrfs_path *path;
3808         struct extent_buffer *leaf;
3809         struct btrfs_dir_item *di;
3810         struct btrfs_key key;
3811         u64 index;
3812         int ret;
3813         u64 dir_ino = btrfs_ino(dir);
3814
3815         path = btrfs_alloc_path();
3816         if (!path)
3817                 return -ENOMEM;
3818
3819         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3820                                    name, name_len, -1);
3821         if (IS_ERR_OR_NULL(di)) {
3822                 if (!di)
3823                         ret = -ENOENT;
3824                 else
3825                         ret = PTR_ERR(di);
3826                 goto out;
3827         }
3828
3829         leaf = path->nodes[0];
3830         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3831         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3832         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3833         if (ret) {
3834                 btrfs_abort_transaction(trans, root, ret);
3835                 goto out;
3836         }
3837         btrfs_release_path(path);
3838
3839         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3840                                  objectid, root->root_key.objectid,
3841                                  dir_ino, &index, name, name_len);
3842         if (ret < 0) {
3843                 if (ret != -ENOENT) {
3844                         btrfs_abort_transaction(trans, root, ret);
3845                         goto out;
3846                 }
3847                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3848                                                  name, name_len);
3849                 if (IS_ERR_OR_NULL(di)) {
3850                         if (!di)
3851                                 ret = -ENOENT;
3852                         else
3853                                 ret = PTR_ERR(di);
3854                         btrfs_abort_transaction(trans, root, ret);
3855                         goto out;
3856                 }
3857
3858                 leaf = path->nodes[0];
3859                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3860                 btrfs_release_path(path);
3861                 index = key.offset;
3862         }
3863         btrfs_release_path(path);
3864
3865         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3866         if (ret) {
3867                 btrfs_abort_transaction(trans, root, ret);
3868                 goto out;
3869         }
3870
3871         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3872         inode_inc_iversion(dir);
3873         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3874         ret = btrfs_update_inode_fallback(trans, root, dir);
3875         if (ret)
3876                 btrfs_abort_transaction(trans, root, ret);
3877 out:
3878         btrfs_free_path(path);
3879         return ret;
3880 }
3881
3882 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3883 {
3884         struct inode *inode = dentry->d_inode;
3885         int err = 0;
3886         struct btrfs_root *root = BTRFS_I(dir)->root;
3887         struct btrfs_trans_handle *trans;
3888
3889         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3890                 return -ENOTEMPTY;
3891         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3892                 return -EPERM;
3893
3894         trans = __unlink_start_trans(dir);
3895         if (IS_ERR(trans))
3896                 return PTR_ERR(trans);
3897
3898         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3899                 err = btrfs_unlink_subvol(trans, root, dir,
3900                                           BTRFS_I(inode)->location.objectid,
3901                                           dentry->d_name.name,
3902                                           dentry->d_name.len);
3903                 goto out;
3904         }
3905
3906         err = btrfs_orphan_add(trans, inode);
3907         if (err)
3908                 goto out;
3909
3910         /* now the directory is empty */
3911         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3912                                  dentry->d_name.name, dentry->d_name.len);
3913         if (!err)
3914                 btrfs_i_size_write(inode, 0);
3915 out:
3916         btrfs_end_transaction(trans, root);
3917         btrfs_btree_balance_dirty(root);
3918
3919         return err;
3920 }
3921
3922 /*
3923  * this can truncate away extent items, csum items and directory items.
3924  * It starts at a high offset and removes keys until it can't find
3925  * any higher than new_size
3926  *
3927  * csum items that cross the new i_size are truncated to the new size
3928  * as well.
3929  *
3930  * min_type is the minimum key type to truncate down to.  If set to 0, this
3931  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3932  */
3933 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3934                                struct btrfs_root *root,
3935                                struct inode *inode,
3936                                u64 new_size, u32 min_type)
3937 {
3938         struct btrfs_path *path;
3939         struct extent_buffer *leaf;
3940         struct btrfs_file_extent_item *fi;
3941         struct btrfs_key key;
3942         struct btrfs_key found_key;
3943         u64 extent_start = 0;
3944         u64 extent_num_bytes = 0;
3945         u64 extent_offset = 0;
3946         u64 item_end = 0;
3947         u64 last_size = (u64)-1;
3948         u32 found_type = (u8)-1;
3949         int found_extent;
3950         int del_item;
3951         int pending_del_nr = 0;
3952         int pending_del_slot = 0;
3953         int extent_type = -1;
3954         int ret;
3955         int err = 0;
3956         u64 ino = btrfs_ino(inode);
3957
3958         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3959
3960         path = btrfs_alloc_path();
3961         if (!path)
3962                 return -ENOMEM;
3963         path->reada = -1;
3964
3965         /*
3966          * We want to drop from the next block forward in case this new size is
3967          * not block aligned since we will be keeping the last block of the
3968          * extent just the way it is.
3969          */
3970         if (root->ref_cows || root == root->fs_info->tree_root)
3971                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3972                                         root->sectorsize), (u64)-1, 0);
3973
3974         /*
3975          * This function is also used to drop the items in the log tree before
3976          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3977          * it is used to drop the loged items. So we shouldn't kill the delayed
3978          * items.
3979          */
3980         if (min_type == 0 && root == BTRFS_I(inode)->root)
3981                 btrfs_kill_delayed_inode_items(inode);
3982
3983         key.objectid = ino;
3984         key.offset = (u64)-1;
3985         key.type = (u8)-1;
3986
3987 search_again:
3988         path->leave_spinning = 1;
3989         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3990         if (ret < 0) {
3991                 err = ret;
3992                 goto out;
3993         }
3994
3995         if (ret > 0) {
3996                 /* there are no items in the tree for us to truncate, we're
3997                  * done
3998                  */
3999                 if (path->slots[0] == 0)
4000                         goto out;
4001                 path->slots[0]--;
4002         }
4003
4004         while (1) {
4005                 fi = NULL;
4006                 leaf = path->nodes[0];
4007                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4008                 found_type = btrfs_key_type(&found_key);
4009
4010                 if (found_key.objectid != ino)
4011                         break;
4012
4013                 if (found_type < min_type)
4014                         break;
4015
4016                 item_end = found_key.offset;
4017                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4018                         fi = btrfs_item_ptr(leaf, path->slots[0],
4019                                             struct btrfs_file_extent_item);
4020                         extent_type = btrfs_file_extent_type(leaf, fi);
4021                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4022                                 item_end +=
4023                                     btrfs_file_extent_num_bytes(leaf, fi);
4024                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4025                                 item_end += btrfs_file_extent_inline_len(leaf,
4026                                                                          fi);
4027                         }
4028                         item_end--;
4029                 }
4030                 if (found_type > min_type) {
4031                         del_item = 1;
4032                 } else {
4033                         if (item_end < new_size)
4034                                 break;
4035                         if (found_key.offset >= new_size)
4036                                 del_item = 1;
4037                         else
4038                                 del_item = 0;
4039                 }
4040                 found_extent = 0;
4041                 /* FIXME, shrink the extent if the ref count is only 1 */
4042                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4043                         goto delete;
4044
4045                 if (del_item)
4046                         last_size = found_key.offset;
4047                 else
4048                         last_size = new_size;
4049
4050                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4051                         u64 num_dec;
4052                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4053                         if (!del_item) {
4054                                 u64 orig_num_bytes =
4055                                         btrfs_file_extent_num_bytes(leaf, fi);
4056                                 extent_num_bytes = ALIGN(new_size -
4057                                                 found_key.offset,
4058                                                 root->sectorsize);
4059                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4060                                                          extent_num_bytes);
4061                                 num_dec = (orig_num_bytes -
4062                                            extent_num_bytes);
4063                                 if (root->ref_cows && extent_start != 0)
4064                                         inode_sub_bytes(inode, num_dec);
4065                                 btrfs_mark_buffer_dirty(leaf);
4066                         } else {
4067                                 extent_num_bytes =
4068                                         btrfs_file_extent_disk_num_bytes(leaf,
4069                                                                          fi);
4070                                 extent_offset = found_key.offset -
4071                                         btrfs_file_extent_offset(leaf, fi);
4072
4073                                 /* FIXME blocksize != 4096 */
4074                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4075                                 if (extent_start != 0) {
4076                                         found_extent = 1;
4077                                         if (root->ref_cows)
4078                                                 inode_sub_bytes(inode, num_dec);
4079                                 }
4080                         }
4081                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4082                         /*
4083                          * we can't truncate inline items that have had
4084                          * special encodings
4085                          */
4086                         if (!del_item &&
4087                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4088                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4089                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4090                                 u32 size = new_size - found_key.offset;
4091
4092                                 if (root->ref_cows) {
4093                                         inode_sub_bytes(inode, item_end + 1 -
4094                                                         new_size);
4095                                 }
4096                                 size =
4097                                     btrfs_file_extent_calc_inline_size(size);
4098                                 btrfs_truncate_item(root, path, size, 1);
4099                         } else if (root->ref_cows) {
4100                                 inode_sub_bytes(inode, item_end + 1 -
4101                                                 found_key.offset);
4102                         }
4103                 }
4104 delete:
4105                 if (del_item) {
4106                         if (!pending_del_nr) {
4107                                 /* no pending yet, add ourselves */
4108                                 pending_del_slot = path->slots[0];
4109                                 pending_del_nr = 1;
4110                         } else if (pending_del_nr &&
4111                                    path->slots[0] + 1 == pending_del_slot) {
4112                                 /* hop on the pending chunk */
4113                                 pending_del_nr++;
4114                                 pending_del_slot = path->slots[0];
4115                         } else {
4116                                 BUG();
4117                         }
4118                 } else {
4119                         break;
4120                 }
4121                 if (found_extent && (root->ref_cows ||
4122                                      root == root->fs_info->tree_root)) {
4123                         btrfs_set_path_blocking(path);
4124                         ret = btrfs_free_extent(trans, root, extent_start,
4125                                                 extent_num_bytes, 0,
4126                                                 btrfs_header_owner(leaf),
4127                                                 ino, extent_offset, 0);
4128                         BUG_ON(ret);
4129                 }
4130
4131                 if (found_type == BTRFS_INODE_ITEM_KEY)
4132                         break;
4133
4134                 if (path->slots[0] == 0 ||
4135                     path->slots[0] != pending_del_slot) {
4136                         if (pending_del_nr) {
4137                                 ret = btrfs_del_items(trans, root, path,
4138                                                 pending_del_slot,
4139                                                 pending_del_nr);
4140                                 if (ret) {
4141                                         btrfs_abort_transaction(trans,
4142                                                                 root, ret);
4143                                         goto error;
4144                                 }
4145                                 pending_del_nr = 0;
4146                         }
4147                         btrfs_release_path(path);
4148                         goto search_again;
4149                 } else {
4150                         path->slots[0]--;
4151                 }
4152         }
4153 out:
4154         if (pending_del_nr) {
4155                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4156                                       pending_del_nr);
4157                 if (ret)
4158                         btrfs_abort_transaction(trans, root, ret);
4159         }
4160 error:
4161         if (last_size != (u64)-1)
4162                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4163         btrfs_free_path(path);
4164         return err;
4165 }
4166
4167 /*
4168  * btrfs_truncate_page - read, zero a chunk and write a page
4169  * @inode - inode that we're zeroing
4170  * @from - the offset to start zeroing
4171  * @len - the length to zero, 0 to zero the entire range respective to the
4172  *      offset
4173  * @front - zero up to the offset instead of from the offset on
4174  *
4175  * This will find the page for the "from" offset and cow the page and zero the
4176  * part we want to zero.  This is used with truncate and hole punching.
4177  */
4178 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4179                         int front)
4180 {
4181         struct address_space *mapping = inode->i_mapping;
4182         struct btrfs_root *root = BTRFS_I(inode)->root;
4183         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4184         struct btrfs_ordered_extent *ordered;
4185         struct extent_state *cached_state = NULL;
4186         char *kaddr;
4187         u32 blocksize = root->sectorsize;
4188         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4189         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4190         struct page *page;
4191         gfp_t mask = btrfs_alloc_write_mask(mapping);
4192         int ret = 0;
4193         u64 page_start;
4194         u64 page_end;
4195
4196         if ((offset & (blocksize - 1)) == 0 &&
4197             (!len || ((len & (blocksize - 1)) == 0)))
4198                 goto out;
4199         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4200         if (ret)
4201                 goto out;
4202
4203 again:
4204         page = find_or_create_page(mapping, index, mask);
4205         if (!page) {
4206                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4207                 ret = -ENOMEM;
4208                 goto out;
4209         }
4210
4211         page_start = page_offset(page);
4212         page_end = page_start + PAGE_CACHE_SIZE - 1;
4213
4214         if (!PageUptodate(page)) {
4215                 ret = btrfs_readpage(NULL, page);
4216                 lock_page(page);
4217                 if (page->mapping != mapping) {
4218                         unlock_page(page);
4219                         page_cache_release(page);
4220                         goto again;
4221                 }
4222                 if (!PageUptodate(page)) {
4223                         ret = -EIO;
4224                         goto out_unlock;
4225                 }
4226         }
4227         wait_on_page_writeback(page);
4228
4229         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4230         set_page_extent_mapped(page);
4231
4232         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4233         if (ordered) {
4234                 unlock_extent_cached(io_tree, page_start, page_end,
4235                                      &cached_state, GFP_NOFS);
4236                 unlock_page(page);
4237                 page_cache_release(page);
4238                 btrfs_start_ordered_extent(inode, ordered, 1);
4239                 btrfs_put_ordered_extent(ordered);
4240                 goto again;
4241         }
4242
4243         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4244                           EXTENT_DIRTY | EXTENT_DELALLOC |
4245                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4246                           0, 0, &cached_state, GFP_NOFS);
4247
4248         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4249                                         &cached_state);
4250         if (ret) {
4251                 unlock_extent_cached(io_tree, page_start, page_end,
4252                                      &cached_state, GFP_NOFS);
4253                 goto out_unlock;
4254         }
4255
4256         if (offset != PAGE_CACHE_SIZE) {
4257                 if (!len)
4258                         len = PAGE_CACHE_SIZE - offset;
4259                 kaddr = kmap(page);
4260                 if (front)
4261                         memset(kaddr, 0, offset);
4262                 else
4263                         memset(kaddr + offset, 0, len);
4264                 flush_dcache_page(page);
4265                 kunmap(page);
4266         }
4267         ClearPageChecked(page);
4268         set_page_dirty(page);
4269         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4270                              GFP_NOFS);
4271
4272 out_unlock:
4273         if (ret)
4274                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4275         unlock_page(page);
4276         page_cache_release(page);
4277 out:
4278         return ret;
4279 }
4280
4281 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4282                              u64 offset, u64 len)
4283 {
4284         struct btrfs_trans_handle *trans;
4285         int ret;
4286
4287         /*
4288          * Still need to make sure the inode looks like it's been updated so
4289          * that any holes get logged if we fsync.
4290          */
4291         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4292                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4293                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4294                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4295                 return 0;
4296         }
4297
4298         /*
4299          * 1 - for the one we're dropping
4300          * 1 - for the one we're adding
4301          * 1 - for updating the inode.
4302          */
4303         trans = btrfs_start_transaction(root, 3);
4304         if (IS_ERR(trans))
4305                 return PTR_ERR(trans);
4306
4307         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4308         if (ret) {
4309                 btrfs_abort_transaction(trans, root, ret);
4310                 btrfs_end_transaction(trans, root);
4311                 return ret;
4312         }
4313
4314         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4315                                        0, 0, len, 0, len, 0, 0, 0);
4316         if (ret)
4317                 btrfs_abort_transaction(trans, root, ret);
4318         else
4319                 btrfs_update_inode(trans, root, inode);
4320         btrfs_end_transaction(trans, root);
4321         return ret;
4322 }
4323
4324 /*
4325  * This function puts in dummy file extents for the area we're creating a hole
4326  * for.  So if we are truncating this file to a larger size we need to insert
4327  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4328  * the range between oldsize and size
4329  */
4330 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4331 {
4332         struct btrfs_root *root = BTRFS_I(inode)->root;
4333         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4334         struct extent_map *em = NULL;
4335         struct extent_state *cached_state = NULL;
4336         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4337         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4338         u64 block_end = ALIGN(size, root->sectorsize);
4339         u64 last_byte;
4340         u64 cur_offset;
4341         u64 hole_size;
4342         int err = 0;
4343
4344         /*
4345          * If our size started in the middle of a page we need to zero out the
4346          * rest of the page before we expand the i_size, otherwise we could
4347          * expose stale data.
4348          */
4349         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4350         if (err)
4351                 return err;
4352
4353         if (size <= hole_start)
4354                 return 0;
4355
4356         while (1) {
4357                 struct btrfs_ordered_extent *ordered;
4358
4359                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4360                                  &cached_state);
4361                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4362                                                      block_end - hole_start);
4363                 if (!ordered)
4364                         break;
4365                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4366                                      &cached_state, GFP_NOFS);
4367                 btrfs_start_ordered_extent(inode, ordered, 1);
4368                 btrfs_put_ordered_extent(ordered);
4369         }
4370
4371         cur_offset = hole_start;
4372         while (1) {
4373                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4374                                 block_end - cur_offset, 0);
4375                 if (IS_ERR(em)) {
4376                         err = PTR_ERR(em);
4377                         em = NULL;
4378                         break;
4379                 }
4380                 last_byte = min(extent_map_end(em), block_end);
4381                 last_byte = ALIGN(last_byte , root->sectorsize);
4382                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4383                         struct extent_map *hole_em;
4384                         hole_size = last_byte - cur_offset;
4385
4386                         err = maybe_insert_hole(root, inode, cur_offset,
4387                                                 hole_size);
4388                         if (err)
4389                                 break;
4390                         btrfs_drop_extent_cache(inode, cur_offset,
4391                                                 cur_offset + hole_size - 1, 0);
4392                         hole_em = alloc_extent_map();
4393                         if (!hole_em) {
4394                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4395                                         &BTRFS_I(inode)->runtime_flags);
4396                                 goto next;
4397                         }
4398                         hole_em->start = cur_offset;
4399                         hole_em->len = hole_size;
4400                         hole_em->orig_start = cur_offset;
4401
4402                         hole_em->block_start = EXTENT_MAP_HOLE;
4403                         hole_em->block_len = 0;
4404                         hole_em->orig_block_len = 0;
4405                         hole_em->ram_bytes = hole_size;
4406                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4407                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4408                         hole_em->generation = root->fs_info->generation;
4409
4410                         while (1) {
4411                                 write_lock(&em_tree->lock);
4412                                 err = add_extent_mapping(em_tree, hole_em, 1);
4413                                 write_unlock(&em_tree->lock);
4414                                 if (err != -EEXIST)
4415                                         break;
4416                                 btrfs_drop_extent_cache(inode, cur_offset,
4417                                                         cur_offset +
4418                                                         hole_size - 1, 0);
4419                         }
4420                         free_extent_map(hole_em);
4421                 }
4422 next:
4423                 free_extent_map(em);
4424                 em = NULL;
4425                 cur_offset = last_byte;
4426                 if (cur_offset >= block_end)
4427                         break;
4428         }
4429         free_extent_map(em);
4430         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4431                              GFP_NOFS);
4432         return err;
4433 }
4434
4435 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4436 {
4437         struct btrfs_root *root = BTRFS_I(inode)->root;
4438         struct btrfs_trans_handle *trans;
4439         loff_t oldsize = i_size_read(inode);
4440         loff_t newsize = attr->ia_size;
4441         int mask = attr->ia_valid;
4442         int ret;
4443
4444         /*
4445          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4446          * special case where we need to update the times despite not having
4447          * these flags set.  For all other operations the VFS set these flags
4448          * explicitly if it wants a timestamp update.
4449          */
4450         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4451                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4452
4453         if (newsize > oldsize) {
4454                 truncate_pagecache(inode, newsize);
4455                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4456                 if (ret)
4457                         return ret;
4458
4459                 trans = btrfs_start_transaction(root, 1);
4460                 if (IS_ERR(trans))
4461                         return PTR_ERR(trans);
4462
4463                 i_size_write(inode, newsize);
4464                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4465                 ret = btrfs_update_inode(trans, root, inode);
4466                 btrfs_end_transaction(trans, root);
4467         } else {
4468
4469                 /*
4470                  * We're truncating a file that used to have good data down to
4471                  * zero. Make sure it gets into the ordered flush list so that
4472                  * any new writes get down to disk quickly.
4473                  */
4474                 if (newsize == 0)
4475                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4476                                 &BTRFS_I(inode)->runtime_flags);
4477
4478                 /*
4479                  * 1 for the orphan item we're going to add
4480                  * 1 for the orphan item deletion.
4481                  */
4482                 trans = btrfs_start_transaction(root, 2);
4483                 if (IS_ERR(trans))
4484                         return PTR_ERR(trans);
4485
4486                 /*
4487                  * We need to do this in case we fail at _any_ point during the
4488                  * actual truncate.  Once we do the truncate_setsize we could
4489                  * invalidate pages which forces any outstanding ordered io to
4490                  * be instantly completed which will give us extents that need
4491                  * to be truncated.  If we fail to get an orphan inode down we
4492                  * could have left over extents that were never meant to live,
4493                  * so we need to garuntee from this point on that everything
4494                  * will be consistent.
4495                  */
4496                 ret = btrfs_orphan_add(trans, inode);
4497                 btrfs_end_transaction(trans, root);
4498                 if (ret)
4499                         return ret;
4500
4501                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4502                 truncate_setsize(inode, newsize);
4503
4504                 /* Disable nonlocked read DIO to avoid the end less truncate */
4505                 btrfs_inode_block_unlocked_dio(inode);
4506                 inode_dio_wait(inode);
4507                 btrfs_inode_resume_unlocked_dio(inode);
4508
4509                 ret = btrfs_truncate(inode);
4510                 if (ret && inode->i_nlink) {
4511                         int err;
4512
4513                         /*
4514                          * failed to truncate, disk_i_size is only adjusted down
4515                          * as we remove extents, so it should represent the true
4516                          * size of the inode, so reset the in memory size and
4517                          * delete our orphan entry.
4518                          */
4519                         trans = btrfs_join_transaction(root);
4520                         if (IS_ERR(trans)) {
4521                                 btrfs_orphan_del(NULL, inode);
4522                                 return ret;
4523                         }
4524                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4525                         err = btrfs_orphan_del(trans, inode);
4526                         if (err)
4527                                 btrfs_abort_transaction(trans, root, err);
4528                         btrfs_end_transaction(trans, root);
4529                 }
4530         }
4531
4532         return ret;
4533 }
4534
4535 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4536 {
4537         struct inode *inode = dentry->d_inode;
4538         struct btrfs_root *root = BTRFS_I(inode)->root;
4539         int err;
4540
4541         if (btrfs_root_readonly(root))
4542                 return -EROFS;
4543
4544         err = inode_change_ok(inode, attr);
4545         if (err)
4546                 return err;
4547
4548         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4549                 err = btrfs_setsize(inode, attr);
4550                 if (err)
4551                         return err;
4552         }
4553
4554         if (attr->ia_valid) {
4555                 setattr_copy(inode, attr);
4556                 inode_inc_iversion(inode);
4557                 err = btrfs_dirty_inode(inode);
4558
4559                 if (!err && attr->ia_valid & ATTR_MODE)
4560                         err = btrfs_acl_chmod(inode);
4561         }
4562
4563         return err;
4564 }
4565
4566 /*
4567  * While truncating the inode pages during eviction, we get the VFS calling
4568  * btrfs_invalidatepage() against each page of the inode. This is slow because
4569  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4570  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4571  * extent_state structures over and over, wasting lots of time.
4572  *
4573  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4574  * those expensive operations on a per page basis and do only the ordered io
4575  * finishing, while we release here the extent_map and extent_state structures,
4576  * without the excessive merging and splitting.
4577  */
4578 static void evict_inode_truncate_pages(struct inode *inode)
4579 {
4580         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4581         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4582         struct rb_node *node;
4583
4584         ASSERT(inode->i_state & I_FREEING);
4585         truncate_inode_pages(&inode->i_data, 0);
4586
4587         write_lock(&map_tree->lock);
4588         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4589                 struct extent_map *em;
4590
4591                 node = rb_first(&map_tree->map);
4592                 em = rb_entry(node, struct extent_map, rb_node);
4593                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
4594                 clear_bit(EXTENT_FLAG_LOGGING, &em->flags);
4595                 remove_extent_mapping(map_tree, em);
4596                 free_extent_map(em);
4597         }
4598         write_unlock(&map_tree->lock);
4599
4600         spin_lock(&io_tree->lock);
4601         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4602                 struct extent_state *state;
4603                 struct extent_state *cached_state = NULL;
4604
4605                 node = rb_first(&io_tree->state);
4606                 state = rb_entry(node, struct extent_state, rb_node);
4607                 atomic_inc(&state->refs);
4608                 spin_unlock(&io_tree->lock);
4609
4610                 lock_extent_bits(io_tree, state->start, state->end,
4611                                  0, &cached_state);
4612                 clear_extent_bit(io_tree, state->start, state->end,
4613                                  EXTENT_LOCKED | EXTENT_DIRTY |
4614                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4615                                  EXTENT_DEFRAG, 1, 1,
4616                                  &cached_state, GFP_NOFS);
4617                 free_extent_state(state);
4618
4619                 spin_lock(&io_tree->lock);
4620         }
4621         spin_unlock(&io_tree->lock);
4622 }
4623
4624 void btrfs_evict_inode(struct inode *inode)
4625 {
4626         struct btrfs_trans_handle *trans;
4627         struct btrfs_root *root = BTRFS_I(inode)->root;
4628         struct btrfs_block_rsv *rsv, *global_rsv;
4629         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4630         int ret;
4631
4632         trace_btrfs_inode_evict(inode);
4633
4634         evict_inode_truncate_pages(inode);
4635
4636         if (inode->i_nlink &&
4637             ((btrfs_root_refs(&root->root_item) != 0 &&
4638               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4639              btrfs_is_free_space_inode(inode)))
4640                 goto no_delete;
4641
4642         if (is_bad_inode(inode)) {
4643                 btrfs_orphan_del(NULL, inode);
4644                 goto no_delete;
4645         }
4646         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4647         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4648
4649         if (root->fs_info->log_root_recovering) {
4650                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4651                                  &BTRFS_I(inode)->runtime_flags));
4652                 goto no_delete;
4653         }
4654
4655         if (inode->i_nlink > 0) {
4656                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4657                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4658                 goto no_delete;
4659         }
4660
4661         ret = btrfs_commit_inode_delayed_inode(inode);
4662         if (ret) {
4663                 btrfs_orphan_del(NULL, inode);
4664                 goto no_delete;
4665         }
4666
4667         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4668         if (!rsv) {
4669                 btrfs_orphan_del(NULL, inode);
4670                 goto no_delete;
4671         }
4672         rsv->size = min_size;
4673         rsv->failfast = 1;
4674         global_rsv = &root->fs_info->global_block_rsv;
4675
4676         btrfs_i_size_write(inode, 0);
4677
4678         /*
4679          * This is a bit simpler than btrfs_truncate since we've already
4680          * reserved our space for our orphan item in the unlink, so we just
4681          * need to reserve some slack space in case we add bytes and update
4682          * inode item when doing the truncate.
4683          */
4684         while (1) {
4685                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4686                                              BTRFS_RESERVE_FLUSH_LIMIT);
4687
4688                 /*
4689                  * Try and steal from the global reserve since we will
4690                  * likely not use this space anyway, we want to try as
4691                  * hard as possible to get this to work.
4692                  */
4693                 if (ret)
4694                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4695
4696                 if (ret) {
4697                         btrfs_warn(root->fs_info,
4698                                 "Could not get space for a delete, will truncate on mount %d",
4699                                 ret);
4700                         btrfs_orphan_del(NULL, inode);
4701                         btrfs_free_block_rsv(root, rsv);
4702                         goto no_delete;
4703                 }
4704
4705                 trans = btrfs_join_transaction(root);
4706                 if (IS_ERR(trans)) {
4707                         btrfs_orphan_del(NULL, inode);
4708                         btrfs_free_block_rsv(root, rsv);
4709                         goto no_delete;
4710                 }
4711
4712                 trans->block_rsv = rsv;
4713
4714                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4715                 if (ret != -ENOSPC)
4716                         break;
4717
4718                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4719                 btrfs_end_transaction(trans, root);
4720                 trans = NULL;
4721                 btrfs_btree_balance_dirty(root);
4722         }
4723
4724         btrfs_free_block_rsv(root, rsv);
4725
4726         /*
4727          * Errors here aren't a big deal, it just means we leave orphan items
4728          * in the tree.  They will be cleaned up on the next mount.
4729          */
4730         if (ret == 0) {
4731                 trans->block_rsv = root->orphan_block_rsv;
4732                 btrfs_orphan_del(trans, inode);
4733         } else {
4734                 btrfs_orphan_del(NULL, inode);
4735         }
4736
4737         trans->block_rsv = &root->fs_info->trans_block_rsv;
4738         if (!(root == root->fs_info->tree_root ||
4739               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4740                 btrfs_return_ino(root, btrfs_ino(inode));
4741
4742         btrfs_end_transaction(trans, root);
4743         btrfs_btree_balance_dirty(root);
4744 no_delete:
4745         btrfs_remove_delayed_node(inode);
4746         clear_inode(inode);
4747         return;
4748 }
4749
4750 /*
4751  * this returns the key found in the dir entry in the location pointer.
4752  * If no dir entries were found, location->objectid is 0.
4753  */
4754 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4755                                struct btrfs_key *location)
4756 {
4757         const char *name = dentry->d_name.name;
4758         int namelen = dentry->d_name.len;
4759         struct btrfs_dir_item *di;
4760         struct btrfs_path *path;
4761         struct btrfs_root *root = BTRFS_I(dir)->root;
4762         int ret = 0;
4763
4764         path = btrfs_alloc_path();
4765         if (!path)
4766                 return -ENOMEM;
4767
4768         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4769                                     namelen, 0);
4770         if (IS_ERR(di))
4771                 ret = PTR_ERR(di);
4772
4773         if (IS_ERR_OR_NULL(di))
4774                 goto out_err;
4775
4776         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4777 out:
4778         btrfs_free_path(path);
4779         return ret;
4780 out_err:
4781         location->objectid = 0;
4782         goto out;
4783 }
4784
4785 /*
4786  * when we hit a tree root in a directory, the btrfs part of the inode
4787  * needs to be changed to reflect the root directory of the tree root.  This
4788  * is kind of like crossing a mount point.
4789  */
4790 static int fixup_tree_root_location(struct btrfs_root *root,
4791                                     struct inode *dir,
4792                                     struct dentry *dentry,
4793                                     struct btrfs_key *location,
4794                                     struct btrfs_root **sub_root)
4795 {
4796         struct btrfs_path *path;
4797         struct btrfs_root *new_root;
4798         struct btrfs_root_ref *ref;
4799         struct extent_buffer *leaf;
4800         int ret;
4801         int err = 0;
4802
4803         path = btrfs_alloc_path();
4804         if (!path) {
4805                 err = -ENOMEM;
4806                 goto out;
4807         }
4808
4809         err = -ENOENT;
4810         ret = btrfs_find_item(root->fs_info->tree_root, path,
4811                                 BTRFS_I(dir)->root->root_key.objectid,
4812                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4813         if (ret) {
4814                 if (ret < 0)
4815                         err = ret;
4816                 goto out;
4817         }
4818
4819         leaf = path->nodes[0];
4820         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4821         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4822             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4823                 goto out;
4824
4825         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4826                                    (unsigned long)(ref + 1),
4827                                    dentry->d_name.len);
4828         if (ret)
4829                 goto out;
4830
4831         btrfs_release_path(path);
4832
4833         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4834         if (IS_ERR(new_root)) {
4835                 err = PTR_ERR(new_root);
4836                 goto out;
4837         }
4838
4839         *sub_root = new_root;
4840         location->objectid = btrfs_root_dirid(&new_root->root_item);
4841         location->type = BTRFS_INODE_ITEM_KEY;
4842         location->offset = 0;
4843         err = 0;
4844 out:
4845         btrfs_free_path(path);
4846         return err;
4847 }
4848
4849 static void inode_tree_add(struct inode *inode)
4850 {
4851         struct btrfs_root *root = BTRFS_I(inode)->root;
4852         struct btrfs_inode *entry;
4853         struct rb_node **p;
4854         struct rb_node *parent;
4855         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4856         u64 ino = btrfs_ino(inode);
4857
4858         if (inode_unhashed(inode))
4859                 return;
4860         parent = NULL;
4861         spin_lock(&root->inode_lock);
4862         p = &root->inode_tree.rb_node;
4863         while (*p) {
4864                 parent = *p;
4865                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4866
4867                 if (ino < btrfs_ino(&entry->vfs_inode))
4868                         p = &parent->rb_left;
4869                 else if (ino > btrfs_ino(&entry->vfs_inode))
4870                         p = &parent->rb_right;
4871                 else {
4872                         WARN_ON(!(entry->vfs_inode.i_state &
4873                                   (I_WILL_FREE | I_FREEING)));
4874                         rb_replace_node(parent, new, &root->inode_tree);
4875                         RB_CLEAR_NODE(parent);
4876                         spin_unlock(&root->inode_lock);
4877                         return;
4878                 }
4879         }
4880         rb_link_node(new, parent, p);
4881         rb_insert_color(new, &root->inode_tree);
4882         spin_unlock(&root->inode_lock);
4883 }
4884
4885 static void inode_tree_del(struct inode *inode)
4886 {
4887         struct btrfs_root *root = BTRFS_I(inode)->root;
4888         int empty = 0;
4889
4890         spin_lock(&root->inode_lock);
4891         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4892                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4893                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4894                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4895         }
4896         spin_unlock(&root->inode_lock);
4897
4898         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4899                 synchronize_srcu(&root->fs_info->subvol_srcu);
4900                 spin_lock(&root->inode_lock);
4901                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4902                 spin_unlock(&root->inode_lock);
4903                 if (empty)
4904                         btrfs_add_dead_root(root);
4905         }
4906 }
4907
4908 void btrfs_invalidate_inodes(struct btrfs_root *root)
4909 {
4910         struct rb_node *node;
4911         struct rb_node *prev;
4912         struct btrfs_inode *entry;
4913         struct inode *inode;
4914         u64 objectid = 0;
4915
4916         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4917
4918         spin_lock(&root->inode_lock);
4919 again:
4920         node = root->inode_tree.rb_node;
4921         prev = NULL;
4922         while (node) {
4923                 prev = node;
4924                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4925
4926                 if (objectid < btrfs_ino(&entry->vfs_inode))
4927                         node = node->rb_left;
4928                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4929                         node = node->rb_right;
4930                 else
4931                         break;
4932         }
4933         if (!node) {
4934                 while (prev) {
4935                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4936                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4937                                 node = prev;
4938                                 break;
4939                         }
4940                         prev = rb_next(prev);
4941                 }
4942         }
4943         while (node) {
4944                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4945                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4946                 inode = igrab(&entry->vfs_inode);
4947                 if (inode) {
4948                         spin_unlock(&root->inode_lock);
4949                         if (atomic_read(&inode->i_count) > 1)
4950                                 d_prune_aliases(inode);
4951                         /*
4952                          * btrfs_drop_inode will have it removed from
4953                          * the inode cache when its usage count
4954                          * hits zero.
4955                          */
4956                         iput(inode);
4957                         cond_resched();
4958                         spin_lock(&root->inode_lock);
4959                         goto again;
4960                 }
4961
4962                 if (cond_resched_lock(&root->inode_lock))
4963                         goto again;
4964
4965                 node = rb_next(node);
4966         }
4967         spin_unlock(&root->inode_lock);
4968 }
4969
4970 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4971 {
4972         struct btrfs_iget_args *args = p;
4973         inode->i_ino = args->ino;
4974         BTRFS_I(inode)->root = args->root;
4975         return 0;
4976 }
4977
4978 static int btrfs_find_actor(struct inode *inode, void *opaque)
4979 {
4980         struct btrfs_iget_args *args = opaque;
4981         return args->ino == btrfs_ino(inode) &&
4982                 args->root == BTRFS_I(inode)->root;
4983 }
4984
4985 static struct inode *btrfs_iget_locked(struct super_block *s,
4986                                        u64 objectid,
4987                                        struct btrfs_root *root)
4988 {
4989         struct inode *inode;
4990         struct btrfs_iget_args args;
4991         unsigned long hashval = btrfs_inode_hash(objectid, root);
4992
4993         args.ino = objectid;
4994         args.root = root;
4995
4996         inode = iget5_locked(s, hashval, btrfs_find_actor,
4997                              btrfs_init_locked_inode,
4998                              (void *)&args);
4999         return inode;
5000 }
5001
5002 /* Get an inode object given its location and corresponding root.
5003  * Returns in *is_new if the inode was read from disk
5004  */
5005 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5006                          struct btrfs_root *root, int *new)
5007 {
5008         struct inode *inode;
5009
5010         inode = btrfs_iget_locked(s, location->objectid, root);
5011         if (!inode)
5012                 return ERR_PTR(-ENOMEM);
5013
5014         if (inode->i_state & I_NEW) {
5015                 BTRFS_I(inode)->root = root;
5016                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5017                 btrfs_read_locked_inode(inode);
5018                 if (!is_bad_inode(inode)) {
5019                         inode_tree_add(inode);
5020                         unlock_new_inode(inode);
5021                         if (new)
5022                                 *new = 1;
5023                 } else {
5024                         unlock_new_inode(inode);
5025                         iput(inode);
5026                         inode = ERR_PTR(-ESTALE);
5027                 }
5028         }
5029
5030         return inode;
5031 }
5032
5033 static struct inode *new_simple_dir(struct super_block *s,
5034                                     struct btrfs_key *key,
5035                                     struct btrfs_root *root)
5036 {
5037         struct inode *inode = new_inode(s);
5038
5039         if (!inode)
5040                 return ERR_PTR(-ENOMEM);
5041
5042         BTRFS_I(inode)->root = root;
5043         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5044         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5045
5046         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5047         inode->i_op = &btrfs_dir_ro_inode_operations;
5048         inode->i_fop = &simple_dir_operations;
5049         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5050         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5051
5052         return inode;
5053 }
5054
5055 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5056 {
5057         struct inode *inode;
5058         struct btrfs_root *root = BTRFS_I(dir)->root;
5059         struct btrfs_root *sub_root = root;
5060         struct btrfs_key location;
5061         int index;
5062         int ret = 0;
5063
5064         if (dentry->d_name.len > BTRFS_NAME_LEN)
5065                 return ERR_PTR(-ENAMETOOLONG);
5066
5067         ret = btrfs_inode_by_name(dir, dentry, &location);
5068         if (ret < 0)
5069                 return ERR_PTR(ret);
5070
5071         if (location.objectid == 0)
5072                 return ERR_PTR(-ENOENT);
5073
5074         if (location.type == BTRFS_INODE_ITEM_KEY) {
5075                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5076                 return inode;
5077         }
5078
5079         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5080
5081         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5082         ret = fixup_tree_root_location(root, dir, dentry,
5083                                        &location, &sub_root);
5084         if (ret < 0) {
5085                 if (ret != -ENOENT)
5086                         inode = ERR_PTR(ret);
5087                 else
5088                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5089         } else {
5090                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5091         }
5092         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5093
5094         if (!IS_ERR(inode) && root != sub_root) {
5095                 down_read(&root->fs_info->cleanup_work_sem);
5096                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5097                         ret = btrfs_orphan_cleanup(sub_root);
5098                 up_read(&root->fs_info->cleanup_work_sem);
5099                 if (ret) {
5100                         iput(inode);
5101                         inode = ERR_PTR(ret);
5102                 }
5103         }
5104
5105         return inode;
5106 }
5107
5108 static int btrfs_dentry_delete(const struct dentry *dentry)
5109 {
5110         struct btrfs_root *root;
5111         struct inode *inode = dentry->d_inode;
5112
5113         if (!inode && !IS_ROOT(dentry))
5114                 inode = dentry->d_parent->d_inode;
5115
5116         if (inode) {
5117                 root = BTRFS_I(inode)->root;
5118                 if (btrfs_root_refs(&root->root_item) == 0)
5119                         return 1;
5120
5121                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5122                         return 1;
5123         }
5124         return 0;
5125 }
5126
5127 static void btrfs_dentry_release(struct dentry *dentry)
5128 {
5129         if (dentry->d_fsdata)
5130                 kfree(dentry->d_fsdata);
5131 }
5132
5133 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5134                                    unsigned int flags)
5135 {
5136         struct inode *inode;
5137
5138         inode = btrfs_lookup_dentry(dir, dentry);
5139         if (IS_ERR(inode)) {
5140                 if (PTR_ERR(inode) == -ENOENT)
5141                         inode = NULL;
5142                 else
5143                         return ERR_CAST(inode);
5144         }
5145
5146         return d_splice_alias(inode, dentry);
5147 }
5148
5149 unsigned char btrfs_filetype_table[] = {
5150         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5151 };
5152
5153 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5154 {
5155         struct inode *inode = file_inode(file);
5156         struct btrfs_root *root = BTRFS_I(inode)->root;
5157         struct btrfs_item *item;
5158         struct btrfs_dir_item *di;
5159         struct btrfs_key key;
5160         struct btrfs_key found_key;
5161         struct btrfs_path *path;
5162         struct list_head ins_list;
5163         struct list_head del_list;
5164         int ret;
5165         struct extent_buffer *leaf;
5166         int slot;
5167         unsigned char d_type;
5168         int over = 0;
5169         u32 di_cur;
5170         u32 di_total;
5171         u32 di_len;
5172         int key_type = BTRFS_DIR_INDEX_KEY;
5173         char tmp_name[32];
5174         char *name_ptr;
5175         int name_len;
5176         int is_curr = 0;        /* ctx->pos points to the current index? */
5177
5178         /* FIXME, use a real flag for deciding about the key type */
5179         if (root->fs_info->tree_root == root)
5180                 key_type = BTRFS_DIR_ITEM_KEY;
5181
5182         if (!dir_emit_dots(file, ctx))
5183                 return 0;
5184
5185         path = btrfs_alloc_path();
5186         if (!path)
5187                 return -ENOMEM;
5188
5189         path->reada = 1;
5190
5191         if (key_type == BTRFS_DIR_INDEX_KEY) {
5192                 INIT_LIST_HEAD(&ins_list);
5193                 INIT_LIST_HEAD(&del_list);
5194                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5195         }
5196
5197         btrfs_set_key_type(&key, key_type);
5198         key.offset = ctx->pos;
5199         key.objectid = btrfs_ino(inode);
5200
5201         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5202         if (ret < 0)
5203                 goto err;
5204
5205         while (1) {
5206                 leaf = path->nodes[0];
5207                 slot = path->slots[0];
5208                 if (slot >= btrfs_header_nritems(leaf)) {
5209                         ret = btrfs_next_leaf(root, path);
5210                         if (ret < 0)
5211                                 goto err;
5212                         else if (ret > 0)
5213                                 break;
5214                         continue;
5215                 }
5216
5217                 item = btrfs_item_nr(slot);
5218                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5219
5220                 if (found_key.objectid != key.objectid)
5221                         break;
5222                 if (btrfs_key_type(&found_key) != key_type)
5223                         break;
5224                 if (found_key.offset < ctx->pos)
5225                         goto next;
5226                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5227                     btrfs_should_delete_dir_index(&del_list,
5228                                                   found_key.offset))
5229                         goto next;
5230
5231                 ctx->pos = found_key.offset;
5232                 is_curr = 1;
5233
5234                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5235                 di_cur = 0;
5236                 di_total = btrfs_item_size(leaf, item);
5237
5238                 while (di_cur < di_total) {
5239                         struct btrfs_key location;
5240
5241                         if (verify_dir_item(root, leaf, di))
5242                                 break;
5243
5244                         name_len = btrfs_dir_name_len(leaf, di);
5245                         if (name_len <= sizeof(tmp_name)) {
5246                                 name_ptr = tmp_name;
5247                         } else {
5248                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5249                                 if (!name_ptr) {
5250                                         ret = -ENOMEM;
5251                                         goto err;
5252                                 }
5253                         }
5254                         read_extent_buffer(leaf, name_ptr,
5255                                            (unsigned long)(di + 1), name_len);
5256
5257                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5258                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5259
5260
5261                         /* is this a reference to our own snapshot? If so
5262                          * skip it.
5263                          *
5264                          * In contrast to old kernels, we insert the snapshot's
5265                          * dir item and dir index after it has been created, so
5266                          * we won't find a reference to our own snapshot. We
5267                          * still keep the following code for backward
5268                          * compatibility.
5269                          */
5270                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5271                             location.objectid == root->root_key.objectid) {
5272                                 over = 0;
5273                                 goto skip;
5274                         }
5275                         over = !dir_emit(ctx, name_ptr, name_len,
5276                                        location.objectid, d_type);
5277
5278 skip:
5279                         if (name_ptr != tmp_name)
5280                                 kfree(name_ptr);
5281
5282                         if (over)
5283                                 goto nopos;
5284                         di_len = btrfs_dir_name_len(leaf, di) +
5285                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5286                         di_cur += di_len;
5287                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5288                 }
5289 next:
5290                 path->slots[0]++;
5291         }
5292
5293         if (key_type == BTRFS_DIR_INDEX_KEY) {
5294                 if (is_curr)
5295                         ctx->pos++;
5296                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5297                 if (ret)
5298                         goto nopos;
5299         }
5300
5301         /* Reached end of directory/root. Bump pos past the last item. */
5302         ctx->pos++;
5303
5304         /*
5305          * Stop new entries from being returned after we return the last
5306          * entry.
5307          *
5308          * New directory entries are assigned a strictly increasing
5309          * offset.  This means that new entries created during readdir
5310          * are *guaranteed* to be seen in the future by that readdir.
5311          * This has broken buggy programs which operate on names as
5312          * they're returned by readdir.  Until we re-use freed offsets
5313          * we have this hack to stop new entries from being returned
5314          * under the assumption that they'll never reach this huge
5315          * offset.
5316          *
5317          * This is being careful not to overflow 32bit loff_t unless the
5318          * last entry requires it because doing so has broken 32bit apps
5319          * in the past.
5320          */
5321         if (key_type == BTRFS_DIR_INDEX_KEY) {
5322                 if (ctx->pos >= INT_MAX)
5323                         ctx->pos = LLONG_MAX;
5324                 else
5325                         ctx->pos = INT_MAX;
5326         }
5327 nopos:
5328         ret = 0;
5329 err:
5330         if (key_type == BTRFS_DIR_INDEX_KEY)
5331                 btrfs_put_delayed_items(&ins_list, &del_list);
5332         btrfs_free_path(path);
5333         return ret;
5334 }
5335
5336 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5337 {
5338         struct btrfs_root *root = BTRFS_I(inode)->root;
5339         struct btrfs_trans_handle *trans;
5340         int ret = 0;
5341         bool nolock = false;
5342
5343         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5344                 return 0;
5345
5346         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5347                 nolock = true;
5348
5349         if (wbc->sync_mode == WB_SYNC_ALL) {
5350                 if (nolock)
5351                         trans = btrfs_join_transaction_nolock(root);
5352                 else
5353                         trans = btrfs_join_transaction(root);
5354                 if (IS_ERR(trans))
5355                         return PTR_ERR(trans);
5356                 ret = btrfs_commit_transaction(trans, root);
5357         }
5358         return ret;
5359 }
5360
5361 /*
5362  * This is somewhat expensive, updating the tree every time the
5363  * inode changes.  But, it is most likely to find the inode in cache.
5364  * FIXME, needs more benchmarking...there are no reasons other than performance
5365  * to keep or drop this code.
5366  */
5367 static int btrfs_dirty_inode(struct inode *inode)
5368 {
5369         struct btrfs_root *root = BTRFS_I(inode)->root;
5370         struct btrfs_trans_handle *trans;
5371         int ret;
5372
5373         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5374                 return 0;
5375
5376         trans = btrfs_join_transaction(root);
5377         if (IS_ERR(trans))
5378                 return PTR_ERR(trans);
5379
5380         ret = btrfs_update_inode(trans, root, inode);
5381         if (ret && ret == -ENOSPC) {
5382                 /* whoops, lets try again with the full transaction */
5383                 btrfs_end_transaction(trans, root);
5384                 trans = btrfs_start_transaction(root, 1);
5385                 if (IS_ERR(trans))
5386                         return PTR_ERR(trans);
5387
5388                 ret = btrfs_update_inode(trans, root, inode);
5389         }
5390         btrfs_end_transaction(trans, root);
5391         if (BTRFS_I(inode)->delayed_node)
5392                 btrfs_balance_delayed_items(root);
5393
5394         return ret;
5395 }
5396
5397 /*
5398  * This is a copy of file_update_time.  We need this so we can return error on
5399  * ENOSPC for updating the inode in the case of file write and mmap writes.
5400  */
5401 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5402                              int flags)
5403 {
5404         struct btrfs_root *root = BTRFS_I(inode)->root;
5405
5406         if (btrfs_root_readonly(root))
5407                 return -EROFS;
5408
5409         if (flags & S_VERSION)
5410                 inode_inc_iversion(inode);
5411         if (flags & S_CTIME)
5412                 inode->i_ctime = *now;
5413         if (flags & S_MTIME)
5414                 inode->i_mtime = *now;
5415         if (flags & S_ATIME)
5416                 inode->i_atime = *now;
5417         return btrfs_dirty_inode(inode);
5418 }
5419
5420 /*
5421  * find the highest existing sequence number in a directory
5422  * and then set the in-memory index_cnt variable to reflect
5423  * free sequence numbers
5424  */
5425 static int btrfs_set_inode_index_count(struct inode *inode)
5426 {
5427         struct btrfs_root *root = BTRFS_I(inode)->root;
5428         struct btrfs_key key, found_key;
5429         struct btrfs_path *path;
5430         struct extent_buffer *leaf;
5431         int ret;
5432
5433         key.objectid = btrfs_ino(inode);
5434         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5435         key.offset = (u64)-1;
5436
5437         path = btrfs_alloc_path();
5438         if (!path)
5439                 return -ENOMEM;
5440
5441         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5442         if (ret < 0)
5443                 goto out;
5444         /* FIXME: we should be able to handle this */
5445         if (ret == 0)
5446                 goto out;
5447         ret = 0;
5448
5449         /*
5450          * MAGIC NUMBER EXPLANATION:
5451          * since we search a directory based on f_pos we have to start at 2
5452          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5453          * else has to start at 2
5454          */
5455         if (path->slots[0] == 0) {
5456                 BTRFS_I(inode)->index_cnt = 2;
5457                 goto out;
5458         }
5459
5460         path->slots[0]--;
5461
5462         leaf = path->nodes[0];
5463         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5464
5465         if (found_key.objectid != btrfs_ino(inode) ||
5466             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5467                 BTRFS_I(inode)->index_cnt = 2;
5468                 goto out;
5469         }
5470
5471         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5472 out:
5473         btrfs_free_path(path);
5474         return ret;
5475 }
5476
5477 /*
5478  * helper to find a free sequence number in a given directory.  This current
5479  * code is very simple, later versions will do smarter things in the btree
5480  */
5481 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5482 {
5483         int ret = 0;
5484
5485         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5486                 ret = btrfs_inode_delayed_dir_index_count(dir);
5487                 if (ret) {
5488                         ret = btrfs_set_inode_index_count(dir);
5489                         if (ret)
5490                                 return ret;
5491                 }
5492         }
5493
5494         *index = BTRFS_I(dir)->index_cnt;
5495         BTRFS_I(dir)->index_cnt++;
5496
5497         return ret;
5498 }
5499
5500 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5501                                      struct btrfs_root *root,
5502                                      struct inode *dir,
5503                                      const char *name, int name_len,
5504                                      u64 ref_objectid, u64 objectid,
5505                                      umode_t mode, u64 *index)
5506 {
5507         struct inode *inode;
5508         struct btrfs_inode_item *inode_item;
5509         struct btrfs_key *location;
5510         struct btrfs_path *path;
5511         struct btrfs_inode_ref *ref;
5512         struct btrfs_key key[2];
5513         u32 sizes[2];
5514         unsigned long ptr;
5515         int ret;
5516
5517         path = btrfs_alloc_path();
5518         if (!path)
5519                 return ERR_PTR(-ENOMEM);
5520
5521         inode = new_inode(root->fs_info->sb);
5522         if (!inode) {
5523                 btrfs_free_path(path);
5524                 return ERR_PTR(-ENOMEM);
5525         }
5526
5527         /*
5528          * we have to initialize this early, so we can reclaim the inode
5529          * number if we fail afterwards in this function.
5530          */
5531         inode->i_ino = objectid;
5532
5533         if (dir) {
5534                 trace_btrfs_inode_request(dir);
5535
5536                 ret = btrfs_set_inode_index(dir, index);
5537                 if (ret) {
5538                         btrfs_free_path(path);
5539                         iput(inode);
5540                         return ERR_PTR(ret);
5541                 }
5542         }
5543         /*
5544          * index_cnt is ignored for everything but a dir,
5545          * btrfs_get_inode_index_count has an explanation for the magic
5546          * number
5547          */
5548         BTRFS_I(inode)->index_cnt = 2;
5549         BTRFS_I(inode)->dir_index = *index;
5550         BTRFS_I(inode)->root = root;
5551         BTRFS_I(inode)->generation = trans->transid;
5552         inode->i_generation = BTRFS_I(inode)->generation;
5553
5554         /*
5555          * We could have gotten an inode number from somebody who was fsynced
5556          * and then removed in this same transaction, so let's just set full
5557          * sync since it will be a full sync anyway and this will blow away the
5558          * old info in the log.
5559          */
5560         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5561
5562         key[0].objectid = objectid;
5563         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5564         key[0].offset = 0;
5565
5566         /*
5567          * Start new inodes with an inode_ref. This is slightly more
5568          * efficient for small numbers of hard links since they will
5569          * be packed into one item. Extended refs will kick in if we
5570          * add more hard links than can fit in the ref item.
5571          */
5572         key[1].objectid = objectid;
5573         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5574         key[1].offset = ref_objectid;
5575
5576         sizes[0] = sizeof(struct btrfs_inode_item);
5577         sizes[1] = name_len + sizeof(*ref);
5578
5579         path->leave_spinning = 1;
5580         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5581         if (ret != 0)
5582                 goto fail;
5583
5584         inode_init_owner(inode, dir, mode);
5585         inode_set_bytes(inode, 0);
5586         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5587         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5588                                   struct btrfs_inode_item);
5589         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5590                              sizeof(*inode_item));
5591         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5592
5593         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5594                              struct btrfs_inode_ref);
5595         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5596         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5597         ptr = (unsigned long)(ref + 1);
5598         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5599
5600         btrfs_mark_buffer_dirty(path->nodes[0]);
5601         btrfs_free_path(path);
5602
5603         location = &BTRFS_I(inode)->location;
5604         location->objectid = objectid;
5605         location->offset = 0;
5606         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5607
5608         btrfs_inherit_iflags(inode, dir);
5609
5610         if (S_ISREG(mode)) {
5611                 if (btrfs_test_opt(root, NODATASUM))
5612                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5613                 if (btrfs_test_opt(root, NODATACOW))
5614                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5615                                 BTRFS_INODE_NODATASUM;
5616         }
5617
5618         btrfs_insert_inode_hash(inode);
5619         inode_tree_add(inode);
5620
5621         trace_btrfs_inode_new(inode);
5622         btrfs_set_inode_last_trans(trans, inode);
5623
5624         btrfs_update_root_times(trans, root);
5625
5626         ret = btrfs_inode_inherit_props(trans, inode, dir);
5627         if (ret)
5628                 btrfs_err(root->fs_info,
5629                           "error inheriting props for ino %llu (root %llu): %d",
5630                           btrfs_ino(inode), root->root_key.objectid, ret);
5631
5632         return inode;
5633 fail:
5634         if (dir)
5635                 BTRFS_I(dir)->index_cnt--;
5636         btrfs_free_path(path);
5637         iput(inode);
5638         return ERR_PTR(ret);
5639 }
5640
5641 static inline u8 btrfs_inode_type(struct inode *inode)
5642 {
5643         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5644 }
5645
5646 /*
5647  * utility function to add 'inode' into 'parent_inode' with
5648  * a give name and a given sequence number.
5649  * if 'add_backref' is true, also insert a backref from the
5650  * inode to the parent directory.
5651  */
5652 int btrfs_add_link(struct btrfs_trans_handle *trans,
5653                    struct inode *parent_inode, struct inode *inode,
5654                    const char *name, int name_len, int add_backref, u64 index)
5655 {
5656         int ret = 0;
5657         struct btrfs_key key;
5658         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5659         u64 ino = btrfs_ino(inode);
5660         u64 parent_ino = btrfs_ino(parent_inode);
5661
5662         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5663                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5664         } else {
5665                 key.objectid = ino;
5666                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5667                 key.offset = 0;
5668         }
5669
5670         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5671                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5672                                          key.objectid, root->root_key.objectid,
5673                                          parent_ino, index, name, name_len);
5674         } else if (add_backref) {
5675                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5676                                              parent_ino, index);
5677         }
5678
5679         /* Nothing to clean up yet */
5680         if (ret)
5681                 return ret;
5682
5683         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5684                                     parent_inode, &key,
5685                                     btrfs_inode_type(inode), index);
5686         if (ret == -EEXIST || ret == -EOVERFLOW)
5687                 goto fail_dir_item;
5688         else if (ret) {
5689                 btrfs_abort_transaction(trans, root, ret);
5690                 return ret;
5691         }
5692
5693         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5694                            name_len * 2);
5695         inode_inc_iversion(parent_inode);
5696         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5697         ret = btrfs_update_inode(trans, root, parent_inode);
5698         if (ret)
5699                 btrfs_abort_transaction(trans, root, ret);
5700         return ret;
5701
5702 fail_dir_item:
5703         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5704                 u64 local_index;
5705                 int err;
5706                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5707                                  key.objectid, root->root_key.objectid,
5708                                  parent_ino, &local_index, name, name_len);
5709
5710         } else if (add_backref) {
5711                 u64 local_index;
5712                 int err;
5713
5714                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5715                                           ino, parent_ino, &local_index);
5716         }
5717         return ret;
5718 }
5719
5720 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5721                             struct inode *dir, struct dentry *dentry,
5722                             struct inode *inode, int backref, u64 index)
5723 {
5724         int err = btrfs_add_link(trans, dir, inode,
5725                                  dentry->d_name.name, dentry->d_name.len,
5726                                  backref, index);
5727         if (err > 0)
5728                 err = -EEXIST;
5729         return err;
5730 }
5731
5732 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5733                         umode_t mode, dev_t rdev)
5734 {
5735         struct btrfs_trans_handle *trans;
5736         struct btrfs_root *root = BTRFS_I(dir)->root;
5737         struct inode *inode = NULL;
5738         int err;
5739         int drop_inode = 0;
5740         u64 objectid;
5741         u64 index = 0;
5742
5743         if (!new_valid_dev(rdev))
5744                 return -EINVAL;
5745
5746         /*
5747          * 2 for inode item and ref
5748          * 2 for dir items
5749          * 1 for xattr if selinux is on
5750          */
5751         trans = btrfs_start_transaction(root, 5);
5752         if (IS_ERR(trans))
5753                 return PTR_ERR(trans);
5754
5755         err = btrfs_find_free_ino(root, &objectid);
5756         if (err)
5757                 goto out_unlock;
5758
5759         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5760                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5761                                 mode, &index);
5762         if (IS_ERR(inode)) {
5763                 err = PTR_ERR(inode);
5764                 goto out_unlock;
5765         }
5766
5767         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5768         if (err) {
5769                 drop_inode = 1;
5770                 goto out_unlock;
5771         }
5772
5773         /*
5774         * If the active LSM wants to access the inode during
5775         * d_instantiate it needs these. Smack checks to see
5776         * if the filesystem supports xattrs by looking at the
5777         * ops vector.
5778         */
5779
5780         inode->i_op = &btrfs_special_inode_operations;
5781         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5782         if (err)
5783                 drop_inode = 1;
5784         else {
5785                 init_special_inode(inode, inode->i_mode, rdev);
5786                 btrfs_update_inode(trans, root, inode);
5787                 d_instantiate(dentry, inode);
5788         }
5789 out_unlock:
5790         btrfs_end_transaction(trans, root);
5791         btrfs_btree_balance_dirty(root);
5792         if (drop_inode) {
5793                 inode_dec_link_count(inode);
5794                 iput(inode);
5795         }
5796         return err;
5797 }
5798
5799 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5800                         umode_t mode, bool excl)
5801 {
5802         struct btrfs_trans_handle *trans;
5803         struct btrfs_root *root = BTRFS_I(dir)->root;
5804         struct inode *inode = NULL;
5805         int drop_inode_on_err = 0;
5806         int err;
5807         u64 objectid;
5808         u64 index = 0;
5809
5810         /*
5811          * 2 for inode item and ref
5812          * 2 for dir items
5813          * 1 for xattr if selinux is on
5814          */
5815         trans = btrfs_start_transaction(root, 5);
5816         if (IS_ERR(trans))
5817                 return PTR_ERR(trans);
5818
5819         err = btrfs_find_free_ino(root, &objectid);
5820         if (err)
5821                 goto out_unlock;
5822
5823         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5824                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5825                                 mode, &index);
5826         if (IS_ERR(inode)) {
5827                 err = PTR_ERR(inode);
5828                 goto out_unlock;
5829         }
5830         drop_inode_on_err = 1;
5831
5832         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5833         if (err)
5834                 goto out_unlock;
5835
5836         err = btrfs_update_inode(trans, root, inode);
5837         if (err)
5838                 goto out_unlock;
5839
5840         /*
5841         * If the active LSM wants to access the inode during
5842         * d_instantiate it needs these. Smack checks to see
5843         * if the filesystem supports xattrs by looking at the
5844         * ops vector.
5845         */
5846         inode->i_fop = &btrfs_file_operations;
5847         inode->i_op = &btrfs_file_inode_operations;
5848
5849         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5850         if (err)
5851                 goto out_unlock;
5852
5853         inode->i_mapping->a_ops = &btrfs_aops;
5854         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5855         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5856         d_instantiate(dentry, inode);
5857
5858 out_unlock:
5859         btrfs_end_transaction(trans, root);
5860         if (err && drop_inode_on_err) {
5861                 inode_dec_link_count(inode);
5862                 iput(inode);
5863         }
5864         btrfs_btree_balance_dirty(root);
5865         return err;
5866 }
5867
5868 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5869                       struct dentry *dentry)
5870 {
5871         struct btrfs_trans_handle *trans;
5872         struct btrfs_root *root = BTRFS_I(dir)->root;
5873         struct inode *inode = old_dentry->d_inode;
5874         u64 index;
5875         int err;
5876         int drop_inode = 0;
5877
5878         /* do not allow sys_link's with other subvols of the same device */
5879         if (root->objectid != BTRFS_I(inode)->root->objectid)
5880                 return -EXDEV;
5881
5882         if (inode->i_nlink >= BTRFS_LINK_MAX)
5883                 return -EMLINK;
5884
5885         err = btrfs_set_inode_index(dir, &index);
5886         if (err)
5887                 goto fail;
5888
5889         /*
5890          * 2 items for inode and inode ref
5891          * 2 items for dir items
5892          * 1 item for parent inode
5893          */
5894         trans = btrfs_start_transaction(root, 5);
5895         if (IS_ERR(trans)) {
5896                 err = PTR_ERR(trans);
5897                 goto fail;
5898         }
5899
5900         /* There are several dir indexes for this inode, clear the cache. */
5901         BTRFS_I(inode)->dir_index = 0ULL;
5902         inc_nlink(inode);
5903         inode_inc_iversion(inode);
5904         inode->i_ctime = CURRENT_TIME;
5905         ihold(inode);
5906         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5907
5908         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5909
5910         if (err) {
5911                 drop_inode = 1;
5912         } else {
5913                 struct dentry *parent = dentry->d_parent;
5914                 err = btrfs_update_inode(trans, root, inode);
5915                 if (err)
5916                         goto fail;
5917                 d_instantiate(dentry, inode);
5918                 btrfs_log_new_name(trans, inode, NULL, parent);
5919         }
5920
5921         btrfs_end_transaction(trans, root);
5922 fail:
5923         if (drop_inode) {
5924                 inode_dec_link_count(inode);
5925                 iput(inode);
5926         }
5927         btrfs_btree_balance_dirty(root);
5928         return err;
5929 }
5930
5931 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5932 {
5933         struct inode *inode = NULL;
5934         struct btrfs_trans_handle *trans;
5935         struct btrfs_root *root = BTRFS_I(dir)->root;
5936         int err = 0;
5937         int drop_on_err = 0;
5938         u64 objectid = 0;
5939         u64 index = 0;
5940
5941         /*
5942          * 2 items for inode and ref
5943          * 2 items for dir items
5944          * 1 for xattr if selinux is on
5945          */
5946         trans = btrfs_start_transaction(root, 5);
5947         if (IS_ERR(trans))
5948                 return PTR_ERR(trans);
5949
5950         err = btrfs_find_free_ino(root, &objectid);
5951         if (err)
5952                 goto out_fail;
5953
5954         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5955                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5956                                 S_IFDIR | mode, &index);
5957         if (IS_ERR(inode)) {
5958                 err = PTR_ERR(inode);
5959                 goto out_fail;
5960         }
5961
5962         drop_on_err = 1;
5963
5964         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5965         if (err)
5966                 goto out_fail;
5967
5968         inode->i_op = &btrfs_dir_inode_operations;
5969         inode->i_fop = &btrfs_dir_file_operations;
5970
5971         btrfs_i_size_write(inode, 0);
5972         err = btrfs_update_inode(trans, root, inode);
5973         if (err)
5974                 goto out_fail;
5975
5976         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5977                              dentry->d_name.len, 0, index);
5978         if (err)
5979                 goto out_fail;
5980
5981         d_instantiate(dentry, inode);
5982         drop_on_err = 0;
5983
5984 out_fail:
5985         btrfs_end_transaction(trans, root);
5986         if (drop_on_err)
5987                 iput(inode);
5988         btrfs_btree_balance_dirty(root);
5989         return err;
5990 }
5991
5992 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5993  * and an extent that you want to insert, deal with overlap and insert
5994  * the new extent into the tree.
5995  */
5996 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5997                                 struct extent_map *existing,
5998                                 struct extent_map *em,
5999                                 u64 map_start, u64 map_len)
6000 {
6001         u64 start_diff;
6002
6003         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
6004         start_diff = map_start - em->start;
6005         em->start = map_start;
6006         em->len = map_len;
6007         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
6008             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
6009                 em->block_start += start_diff;
6010                 em->block_len -= start_diff;
6011         }
6012         return add_extent_mapping(em_tree, em, 0);
6013 }
6014
6015 static noinline int uncompress_inline(struct btrfs_path *path,
6016                                       struct inode *inode, struct page *page,
6017                                       size_t pg_offset, u64 extent_offset,
6018                                       struct btrfs_file_extent_item *item)
6019 {
6020         int ret;
6021         struct extent_buffer *leaf = path->nodes[0];
6022         char *tmp;
6023         size_t max_size;
6024         unsigned long inline_size;
6025         unsigned long ptr;
6026         int compress_type;
6027
6028         WARN_ON(pg_offset != 0);
6029         compress_type = btrfs_file_extent_compression(leaf, item);
6030         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6031         inline_size = btrfs_file_extent_inline_item_len(leaf,
6032                                         btrfs_item_nr(path->slots[0]));
6033         tmp = kmalloc(inline_size, GFP_NOFS);
6034         if (!tmp)
6035                 return -ENOMEM;
6036         ptr = btrfs_file_extent_inline_start(item);
6037
6038         read_extent_buffer(leaf, tmp, ptr, inline_size);
6039
6040         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6041         ret = btrfs_decompress(compress_type, tmp, page,
6042                                extent_offset, inline_size, max_size);
6043         if (ret) {
6044                 char *kaddr = kmap_atomic(page);
6045                 unsigned long copy_size = min_t(u64,
6046                                   PAGE_CACHE_SIZE - pg_offset,
6047                                   max_size - extent_offset);
6048                 memset(kaddr + pg_offset, 0, copy_size);
6049                 kunmap_atomic(kaddr);
6050         }
6051         kfree(tmp);
6052         return 0;
6053 }
6054
6055 /*
6056  * a bit scary, this does extent mapping from logical file offset to the disk.
6057  * the ugly parts come from merging extents from the disk with the in-ram
6058  * representation.  This gets more complex because of the data=ordered code,
6059  * where the in-ram extents might be locked pending data=ordered completion.
6060  *
6061  * This also copies inline extents directly into the page.
6062  */
6063
6064 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6065                                     size_t pg_offset, u64 start, u64 len,
6066                                     int create)
6067 {
6068         int ret;
6069         int err = 0;
6070         u64 bytenr;
6071         u64 extent_start = 0;
6072         u64 extent_end = 0;
6073         u64 objectid = btrfs_ino(inode);
6074         u32 found_type;
6075         struct btrfs_path *path = NULL;
6076         struct btrfs_root *root = BTRFS_I(inode)->root;
6077         struct btrfs_file_extent_item *item;
6078         struct extent_buffer *leaf;
6079         struct btrfs_key found_key;
6080         struct extent_map *em = NULL;
6081         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6082         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6083         struct btrfs_trans_handle *trans = NULL;
6084         int compress_type;
6085
6086 again:
6087         read_lock(&em_tree->lock);
6088         em = lookup_extent_mapping(em_tree, start, len);
6089         if (em)
6090                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6091         read_unlock(&em_tree->lock);
6092
6093         if (em) {
6094                 if (em->start > start || em->start + em->len <= start)
6095                         free_extent_map(em);
6096                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6097                         free_extent_map(em);
6098                 else
6099                         goto out;
6100         }
6101         em = alloc_extent_map();
6102         if (!em) {
6103                 err = -ENOMEM;
6104                 goto out;
6105         }
6106         em->bdev = root->fs_info->fs_devices->latest_bdev;
6107         em->start = EXTENT_MAP_HOLE;
6108         em->orig_start = EXTENT_MAP_HOLE;
6109         em->len = (u64)-1;
6110         em->block_len = (u64)-1;
6111
6112         if (!path) {
6113                 path = btrfs_alloc_path();
6114                 if (!path) {
6115                         err = -ENOMEM;
6116                         goto out;
6117                 }
6118                 /*
6119                  * Chances are we'll be called again, so go ahead and do
6120                  * readahead
6121                  */
6122                 path->reada = 1;
6123         }
6124
6125         ret = btrfs_lookup_file_extent(trans, root, path,
6126                                        objectid, start, trans != NULL);
6127         if (ret < 0) {
6128                 err = ret;
6129                 goto out;
6130         }
6131
6132         if (ret != 0) {
6133                 if (path->slots[0] == 0)
6134                         goto not_found;
6135                 path->slots[0]--;
6136         }
6137
6138         leaf = path->nodes[0];
6139         item = btrfs_item_ptr(leaf, path->slots[0],
6140                               struct btrfs_file_extent_item);
6141         /* are we inside the extent that was found? */
6142         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6143         found_type = btrfs_key_type(&found_key);
6144         if (found_key.objectid != objectid ||
6145             found_type != BTRFS_EXTENT_DATA_KEY) {
6146                 /*
6147                  * If we backup past the first extent we want to move forward
6148                  * and see if there is an extent in front of us, otherwise we'll
6149                  * say there is a hole for our whole search range which can
6150                  * cause problems.
6151                  */
6152                 extent_end = start;
6153                 goto next;
6154         }
6155
6156         found_type = btrfs_file_extent_type(leaf, item);
6157         extent_start = found_key.offset;
6158         compress_type = btrfs_file_extent_compression(leaf, item);
6159         if (found_type == BTRFS_FILE_EXTENT_REG ||
6160             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6161                 extent_end = extent_start +
6162                        btrfs_file_extent_num_bytes(leaf, item);
6163         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6164                 size_t size;
6165                 size = btrfs_file_extent_inline_len(leaf, item);
6166                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6167         }
6168 next:
6169         if (start >= extent_end) {
6170                 path->slots[0]++;
6171                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6172                         ret = btrfs_next_leaf(root, path);
6173                         if (ret < 0) {
6174                                 err = ret;
6175                                 goto out;
6176                         }
6177                         if (ret > 0)
6178                                 goto not_found;
6179                         leaf = path->nodes[0];
6180                 }
6181                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6182                 if (found_key.objectid != objectid ||
6183                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6184                         goto not_found;
6185                 if (start + len <= found_key.offset)
6186                         goto not_found;
6187                 em->start = start;
6188                 em->orig_start = start;
6189                 em->len = found_key.offset - start;
6190                 goto not_found_em;
6191         }
6192
6193         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6194         if (found_type == BTRFS_FILE_EXTENT_REG ||
6195             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6196                 em->start = extent_start;
6197                 em->len = extent_end - extent_start;
6198                 em->orig_start = extent_start -
6199                                  btrfs_file_extent_offset(leaf, item);
6200                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6201                                                                       item);
6202                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6203                 if (bytenr == 0) {
6204                         em->block_start = EXTENT_MAP_HOLE;
6205                         goto insert;
6206                 }
6207                 if (compress_type != BTRFS_COMPRESS_NONE) {
6208                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6209                         em->compress_type = compress_type;
6210                         em->block_start = bytenr;
6211                         em->block_len = em->orig_block_len;
6212                 } else {
6213                         bytenr += btrfs_file_extent_offset(leaf, item);
6214                         em->block_start = bytenr;
6215                         em->block_len = em->len;
6216                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6217                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6218                 }
6219                 goto insert;
6220         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6221                 unsigned long ptr;
6222                 char *map;
6223                 size_t size;
6224                 size_t extent_offset;
6225                 size_t copy_size;
6226
6227                 em->block_start = EXTENT_MAP_INLINE;
6228                 if (!page || create) {
6229                         em->start = extent_start;
6230                         em->len = extent_end - extent_start;
6231                         goto out;
6232                 }
6233
6234                 size = btrfs_file_extent_inline_len(leaf, item);
6235                 extent_offset = page_offset(page) + pg_offset - extent_start;
6236                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6237                                 size - extent_offset);
6238                 em->start = extent_start + extent_offset;
6239                 em->len = ALIGN(copy_size, root->sectorsize);
6240                 em->orig_block_len = em->len;
6241                 em->orig_start = em->start;
6242                 if (compress_type) {
6243                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6244                         em->compress_type = compress_type;
6245                 }
6246                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6247                 if (create == 0 && !PageUptodate(page)) {
6248                         if (btrfs_file_extent_compression(leaf, item) !=
6249                             BTRFS_COMPRESS_NONE) {
6250                                 ret = uncompress_inline(path, inode, page,
6251                                                         pg_offset,
6252                                                         extent_offset, item);
6253                                 BUG_ON(ret); /* -ENOMEM */
6254                         } else {
6255                                 map = kmap(page);
6256                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6257                                                    copy_size);
6258                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6259                                         memset(map + pg_offset + copy_size, 0,
6260                                                PAGE_CACHE_SIZE - pg_offset -
6261                                                copy_size);
6262                                 }
6263                                 kunmap(page);
6264                         }
6265                         flush_dcache_page(page);
6266                 } else if (create && PageUptodate(page)) {
6267                         BUG();
6268                         if (!trans) {
6269                                 kunmap(page);
6270                                 free_extent_map(em);
6271                                 em = NULL;
6272
6273                                 btrfs_release_path(path);
6274                                 trans = btrfs_join_transaction(root);
6275
6276                                 if (IS_ERR(trans))
6277                                         return ERR_CAST(trans);
6278                                 goto again;
6279                         }
6280                         map = kmap(page);
6281                         write_extent_buffer(leaf, map + pg_offset, ptr,
6282                                             copy_size);
6283                         kunmap(page);
6284                         btrfs_mark_buffer_dirty(leaf);
6285                 }
6286                 set_extent_uptodate(io_tree, em->start,
6287                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6288                 goto insert;
6289         } else {
6290                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6291         }
6292 not_found:
6293         em->start = start;
6294         em->orig_start = start;
6295         em->len = len;
6296 not_found_em:
6297         em->block_start = EXTENT_MAP_HOLE;
6298         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6299 insert:
6300         btrfs_release_path(path);
6301         if (em->start > start || extent_map_end(em) <= start) {
6302                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6303                         em->start, em->len, start, len);
6304                 err = -EIO;
6305                 goto out;
6306         }
6307
6308         err = 0;
6309         write_lock(&em_tree->lock);
6310         ret = add_extent_mapping(em_tree, em, 0);
6311         /* it is possible that someone inserted the extent into the tree
6312          * while we had the lock dropped.  It is also possible that
6313          * an overlapping map exists in the tree
6314          */
6315         if (ret == -EEXIST) {
6316                 struct extent_map *existing;
6317
6318                 ret = 0;
6319
6320                 existing = lookup_extent_mapping(em_tree, start, len);
6321                 if (existing && (existing->start > start ||
6322                     existing->start + existing->len <= start)) {
6323                         free_extent_map(existing);
6324                         existing = NULL;
6325                 }
6326                 if (!existing) {
6327                         existing = lookup_extent_mapping(em_tree, em->start,
6328                                                          em->len);
6329                         if (existing) {
6330                                 err = merge_extent_mapping(em_tree, existing,
6331                                                            em, start,
6332                                                            root->sectorsize);
6333                                 free_extent_map(existing);
6334                                 if (err) {
6335                                         free_extent_map(em);
6336                                         em = NULL;
6337                                 }
6338                         } else {
6339                                 err = -EIO;
6340                                 free_extent_map(em);
6341                                 em = NULL;
6342                         }
6343                 } else {
6344                         free_extent_map(em);
6345                         em = existing;
6346                         err = 0;
6347                 }
6348         }
6349         write_unlock(&em_tree->lock);
6350 out:
6351
6352         trace_btrfs_get_extent(root, em);
6353
6354         if (path)
6355                 btrfs_free_path(path);
6356         if (trans) {
6357                 ret = btrfs_end_transaction(trans, root);
6358                 if (!err)
6359                         err = ret;
6360         }
6361         if (err) {
6362                 free_extent_map(em);
6363                 return ERR_PTR(err);
6364         }
6365         BUG_ON(!em); /* Error is always set */
6366         return em;
6367 }
6368
6369 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6370                                            size_t pg_offset, u64 start, u64 len,
6371                                            int create)
6372 {
6373         struct extent_map *em;
6374         struct extent_map *hole_em = NULL;
6375         u64 range_start = start;
6376         u64 end;
6377         u64 found;
6378         u64 found_end;
6379         int err = 0;
6380
6381         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6382         if (IS_ERR(em))
6383                 return em;
6384         if (em) {
6385                 /*
6386                  * if our em maps to
6387                  * -  a hole or
6388                  * -  a pre-alloc extent,
6389                  * there might actually be delalloc bytes behind it.
6390                  */
6391                 if (em->block_start != EXTENT_MAP_HOLE &&
6392                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6393                         return em;
6394                 else
6395                         hole_em = em;
6396         }
6397
6398         /* check to see if we've wrapped (len == -1 or similar) */
6399         end = start + len;
6400         if (end < start)
6401                 end = (u64)-1;
6402         else
6403                 end -= 1;
6404
6405         em = NULL;
6406
6407         /* ok, we didn't find anything, lets look for delalloc */
6408         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6409                                  end, len, EXTENT_DELALLOC, 1);
6410         found_end = range_start + found;
6411         if (found_end < range_start)
6412                 found_end = (u64)-1;
6413
6414         /*
6415          * we didn't find anything useful, return
6416          * the original results from get_extent()
6417          */
6418         if (range_start > end || found_end <= start) {
6419                 em = hole_em;
6420                 hole_em = NULL;
6421                 goto out;
6422         }
6423
6424         /* adjust the range_start to make sure it doesn't
6425          * go backwards from the start they passed in
6426          */
6427         range_start = max(start, range_start);
6428         found = found_end - range_start;
6429
6430         if (found > 0) {
6431                 u64 hole_start = start;
6432                 u64 hole_len = len;
6433
6434                 em = alloc_extent_map();
6435                 if (!em) {
6436                         err = -ENOMEM;
6437                         goto out;
6438                 }
6439                 /*
6440                  * when btrfs_get_extent can't find anything it
6441                  * returns one huge hole
6442                  *
6443                  * make sure what it found really fits our range, and
6444                  * adjust to make sure it is based on the start from
6445                  * the caller
6446                  */
6447                 if (hole_em) {
6448                         u64 calc_end = extent_map_end(hole_em);
6449
6450                         if (calc_end <= start || (hole_em->start > end)) {
6451                                 free_extent_map(hole_em);
6452                                 hole_em = NULL;
6453                         } else {
6454                                 hole_start = max(hole_em->start, start);
6455                                 hole_len = calc_end - hole_start;
6456                         }
6457                 }
6458                 em->bdev = NULL;
6459                 if (hole_em && range_start > hole_start) {
6460                         /* our hole starts before our delalloc, so we
6461                          * have to return just the parts of the hole
6462                          * that go until  the delalloc starts
6463                          */
6464                         em->len = min(hole_len,
6465                                       range_start - hole_start);
6466                         em->start = hole_start;
6467                         em->orig_start = hole_start;
6468                         /*
6469                          * don't adjust block start at all,
6470                          * it is fixed at EXTENT_MAP_HOLE
6471                          */
6472                         em->block_start = hole_em->block_start;
6473                         em->block_len = hole_len;
6474                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6475                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6476                 } else {
6477                         em->start = range_start;
6478                         em->len = found;
6479                         em->orig_start = range_start;
6480                         em->block_start = EXTENT_MAP_DELALLOC;
6481                         em->block_len = found;
6482                 }
6483         } else if (hole_em) {
6484                 return hole_em;
6485         }
6486 out:
6487
6488         free_extent_map(hole_em);
6489         if (err) {
6490                 free_extent_map(em);
6491                 return ERR_PTR(err);
6492         }
6493         return em;
6494 }
6495
6496 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6497                                                   u64 start, u64 len)
6498 {
6499         struct btrfs_root *root = BTRFS_I(inode)->root;
6500         struct extent_map *em;
6501         struct btrfs_key ins;
6502         u64 alloc_hint;
6503         int ret;
6504
6505         alloc_hint = get_extent_allocation_hint(inode, start, len);
6506         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6507                                    alloc_hint, &ins, 1);
6508         if (ret)
6509                 return ERR_PTR(ret);
6510
6511         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6512                               ins.offset, ins.offset, ins.offset, 0);
6513         if (IS_ERR(em)) {
6514                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6515                 return em;
6516         }
6517
6518         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6519                                            ins.offset, ins.offset, 0);
6520         if (ret) {
6521                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6522                 free_extent_map(em);
6523                 return ERR_PTR(ret);
6524         }
6525
6526         return em;
6527 }
6528
6529 /*
6530  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6531  * block must be cow'd
6532  */
6533 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6534                               u64 *orig_start, u64 *orig_block_len,
6535                               u64 *ram_bytes)
6536 {
6537         struct btrfs_trans_handle *trans;
6538         struct btrfs_path *path;
6539         int ret;
6540         struct extent_buffer *leaf;
6541         struct btrfs_root *root = BTRFS_I(inode)->root;
6542         struct btrfs_file_extent_item *fi;
6543         struct btrfs_key key;
6544         u64 disk_bytenr;
6545         u64 backref_offset;
6546         u64 extent_end;
6547         u64 num_bytes;
6548         int slot;
6549         int found_type;
6550         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6551
6552         path = btrfs_alloc_path();
6553         if (!path)
6554                 return -ENOMEM;
6555
6556         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6557                                        offset, 0);
6558         if (ret < 0)
6559                 goto out;
6560
6561         slot = path->slots[0];
6562         if (ret == 1) {
6563                 if (slot == 0) {
6564                         /* can't find the item, must cow */
6565                         ret = 0;
6566                         goto out;
6567                 }
6568                 slot--;
6569         }
6570         ret = 0;
6571         leaf = path->nodes[0];
6572         btrfs_item_key_to_cpu(leaf, &key, slot);
6573         if (key.objectid != btrfs_ino(inode) ||
6574             key.type != BTRFS_EXTENT_DATA_KEY) {
6575                 /* not our file or wrong item type, must cow */
6576                 goto out;
6577         }
6578
6579         if (key.offset > offset) {
6580                 /* Wrong offset, must cow */
6581                 goto out;
6582         }
6583
6584         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6585         found_type = btrfs_file_extent_type(leaf, fi);
6586         if (found_type != BTRFS_FILE_EXTENT_REG &&
6587             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6588                 /* not a regular extent, must cow */
6589                 goto out;
6590         }
6591
6592         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6593                 goto out;
6594
6595         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6596         if (extent_end <= offset)
6597                 goto out;
6598
6599         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6600         if (disk_bytenr == 0)
6601                 goto out;
6602
6603         if (btrfs_file_extent_compression(leaf, fi) ||
6604             btrfs_file_extent_encryption(leaf, fi) ||
6605             btrfs_file_extent_other_encoding(leaf, fi))
6606                 goto out;
6607
6608         backref_offset = btrfs_file_extent_offset(leaf, fi);
6609
6610         if (orig_start) {
6611                 *orig_start = key.offset - backref_offset;
6612                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6613                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6614         }
6615
6616         if (btrfs_extent_readonly(root, disk_bytenr))
6617                 goto out;
6618         btrfs_release_path(path);
6619
6620         /*
6621          * look for other files referencing this extent, if we
6622          * find any we must cow
6623          */
6624         trans = btrfs_join_transaction(root);
6625         if (IS_ERR(trans)) {
6626                 ret = 0;
6627                 goto out;
6628         }
6629
6630         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6631                                     key.offset - backref_offset, disk_bytenr);
6632         btrfs_end_transaction(trans, root);
6633         if (ret) {
6634                 ret = 0;
6635                 goto out;
6636         }
6637
6638         /*
6639          * adjust disk_bytenr and num_bytes to cover just the bytes
6640          * in this extent we are about to write.  If there
6641          * are any csums in that range we have to cow in order
6642          * to keep the csums correct
6643          */
6644         disk_bytenr += backref_offset;
6645         disk_bytenr += offset - key.offset;
6646         num_bytes = min(offset + *len, extent_end) - offset;
6647         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6648                                 goto out;
6649         /*
6650          * all of the above have passed, it is safe to overwrite this extent
6651          * without cow
6652          */
6653         *len = num_bytes;
6654         ret = 1;
6655 out:
6656         btrfs_free_path(path);
6657         return ret;
6658 }
6659
6660 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6661                               struct extent_state **cached_state, int writing)
6662 {
6663         struct btrfs_ordered_extent *ordered;
6664         int ret = 0;
6665
6666         while (1) {
6667                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6668                                  0, cached_state);
6669                 /*
6670                  * We're concerned with the entire range that we're going to be
6671                  * doing DIO to, so we need to make sure theres no ordered
6672                  * extents in this range.
6673                  */
6674                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6675                                                      lockend - lockstart + 1);
6676
6677                 /*
6678                  * We need to make sure there are no buffered pages in this
6679                  * range either, we could have raced between the invalidate in
6680                  * generic_file_direct_write and locking the extent.  The
6681                  * invalidate needs to happen so that reads after a write do not
6682                  * get stale data.
6683                  */
6684                 if (!ordered && (!writing ||
6685                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6686                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6687                                     *cached_state)))
6688                         break;
6689
6690                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6691                                      cached_state, GFP_NOFS);
6692
6693                 if (ordered) {
6694                         btrfs_start_ordered_extent(inode, ordered, 1);
6695                         btrfs_put_ordered_extent(ordered);
6696                 } else {
6697                         /* Screw you mmap */
6698                         ret = filemap_write_and_wait_range(inode->i_mapping,
6699                                                            lockstart,
6700                                                            lockend);
6701                         if (ret)
6702                                 break;
6703
6704                         /*
6705                          * If we found a page that couldn't be invalidated just
6706                          * fall back to buffered.
6707                          */
6708                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6709                                         lockstart >> PAGE_CACHE_SHIFT,
6710                                         lockend >> PAGE_CACHE_SHIFT);
6711                         if (ret)
6712                                 break;
6713                 }
6714
6715                 cond_resched();
6716         }
6717
6718         return ret;
6719 }
6720
6721 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6722                                            u64 len, u64 orig_start,
6723                                            u64 block_start, u64 block_len,
6724                                            u64 orig_block_len, u64 ram_bytes,
6725                                            int type)
6726 {
6727         struct extent_map_tree *em_tree;
6728         struct extent_map *em;
6729         struct btrfs_root *root = BTRFS_I(inode)->root;
6730         int ret;
6731
6732         em_tree = &BTRFS_I(inode)->extent_tree;
6733         em = alloc_extent_map();
6734         if (!em)
6735                 return ERR_PTR(-ENOMEM);
6736
6737         em->start = start;
6738         em->orig_start = orig_start;
6739         em->mod_start = start;
6740         em->mod_len = len;
6741         em->len = len;
6742         em->block_len = block_len;
6743         em->block_start = block_start;
6744         em->bdev = root->fs_info->fs_devices->latest_bdev;
6745         em->orig_block_len = orig_block_len;
6746         em->ram_bytes = ram_bytes;
6747         em->generation = -1;
6748         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6749         if (type == BTRFS_ORDERED_PREALLOC)
6750                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6751
6752         do {
6753                 btrfs_drop_extent_cache(inode, em->start,
6754                                 em->start + em->len - 1, 0);
6755                 write_lock(&em_tree->lock);
6756                 ret = add_extent_mapping(em_tree, em, 1);
6757                 write_unlock(&em_tree->lock);
6758         } while (ret == -EEXIST);
6759
6760         if (ret) {
6761                 free_extent_map(em);
6762                 return ERR_PTR(ret);
6763         }
6764
6765         return em;
6766 }
6767
6768
6769 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6770                                    struct buffer_head *bh_result, int create)
6771 {
6772         struct extent_map *em;
6773         struct btrfs_root *root = BTRFS_I(inode)->root;
6774         struct extent_state *cached_state = NULL;
6775         u64 start = iblock << inode->i_blkbits;
6776         u64 lockstart, lockend;
6777         u64 len = bh_result->b_size;
6778         int unlock_bits = EXTENT_LOCKED;
6779         int ret = 0;
6780
6781         if (create)
6782                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6783         else
6784                 len = min_t(u64, len, root->sectorsize);
6785
6786         lockstart = start;
6787         lockend = start + len - 1;
6788
6789         /*
6790          * If this errors out it's because we couldn't invalidate pagecache for
6791          * this range and we need to fallback to buffered.
6792          */
6793         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6794                 return -ENOTBLK;
6795
6796         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6797         if (IS_ERR(em)) {
6798                 ret = PTR_ERR(em);
6799                 goto unlock_err;
6800         }
6801
6802         /*
6803          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6804          * io.  INLINE is special, and we could probably kludge it in here, but
6805          * it's still buffered so for safety lets just fall back to the generic
6806          * buffered path.
6807          *
6808          * For COMPRESSED we _have_ to read the entire extent in so we can
6809          * decompress it, so there will be buffering required no matter what we
6810          * do, so go ahead and fallback to buffered.
6811          *
6812          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6813          * to buffered IO.  Don't blame me, this is the price we pay for using
6814          * the generic code.
6815          */
6816         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6817             em->block_start == EXTENT_MAP_INLINE) {
6818                 free_extent_map(em);
6819                 ret = -ENOTBLK;
6820                 goto unlock_err;
6821         }
6822
6823         /* Just a good old fashioned hole, return */
6824         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6825                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6826                 free_extent_map(em);
6827                 goto unlock_err;
6828         }
6829
6830         /*
6831          * We don't allocate a new extent in the following cases
6832          *
6833          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6834          * existing extent.
6835          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6836          * just use the extent.
6837          *
6838          */
6839         if (!create) {
6840                 len = min(len, em->len - (start - em->start));
6841                 lockstart = start + len;
6842                 goto unlock;
6843         }
6844
6845         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6846             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6847              em->block_start != EXTENT_MAP_HOLE)) {
6848                 int type;
6849                 int ret;
6850                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6851
6852                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6853                         type = BTRFS_ORDERED_PREALLOC;
6854                 else
6855                         type = BTRFS_ORDERED_NOCOW;
6856                 len = min(len, em->len - (start - em->start));
6857                 block_start = em->block_start + (start - em->start);
6858
6859                 if (can_nocow_extent(inode, start, &len, &orig_start,
6860                                      &orig_block_len, &ram_bytes) == 1) {
6861                         if (type == BTRFS_ORDERED_PREALLOC) {
6862                                 free_extent_map(em);
6863                                 em = create_pinned_em(inode, start, len,
6864                                                        orig_start,
6865                                                        block_start, len,
6866                                                        orig_block_len,
6867                                                        ram_bytes, type);
6868                                 if (IS_ERR(em))
6869                                         goto unlock_err;
6870                         }
6871
6872                         ret = btrfs_add_ordered_extent_dio(inode, start,
6873                                            block_start, len, len, type);
6874                         if (ret) {
6875                                 free_extent_map(em);
6876                                 goto unlock_err;
6877                         }
6878                         goto unlock;
6879                 }
6880         }
6881
6882         /*
6883          * this will cow the extent, reset the len in case we changed
6884          * it above
6885          */
6886         len = bh_result->b_size;
6887         free_extent_map(em);
6888         em = btrfs_new_extent_direct(inode, start, len);
6889         if (IS_ERR(em)) {
6890                 ret = PTR_ERR(em);
6891                 goto unlock_err;
6892         }
6893         len = min(len, em->len - (start - em->start));
6894 unlock:
6895         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6896                 inode->i_blkbits;
6897         bh_result->b_size = len;
6898         bh_result->b_bdev = em->bdev;
6899         set_buffer_mapped(bh_result);
6900         if (create) {
6901                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6902                         set_buffer_new(bh_result);
6903
6904                 /*
6905                  * Need to update the i_size under the extent lock so buffered
6906                  * readers will get the updated i_size when we unlock.
6907                  */
6908                 if (start + len > i_size_read(inode))
6909                         i_size_write(inode, start + len);
6910
6911                 spin_lock(&BTRFS_I(inode)->lock);
6912                 BTRFS_I(inode)->outstanding_extents++;
6913                 spin_unlock(&BTRFS_I(inode)->lock);
6914
6915                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6916                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6917                                      &cached_state, GFP_NOFS);
6918                 BUG_ON(ret);
6919         }
6920
6921         /*
6922          * In the case of write we need to clear and unlock the entire range,
6923          * in the case of read we need to unlock only the end area that we
6924          * aren't using if there is any left over space.
6925          */
6926         if (lockstart < lockend) {
6927                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6928                                  lockend, unlock_bits, 1, 0,
6929                                  &cached_state, GFP_NOFS);
6930         } else {
6931                 free_extent_state(cached_state);
6932         }
6933
6934         free_extent_map(em);
6935
6936         return 0;
6937
6938 unlock_err:
6939         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6940                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6941         return ret;
6942 }
6943
6944 static void btrfs_endio_direct_read(struct bio *bio, int err)
6945 {
6946         struct btrfs_dio_private *dip = bio->bi_private;
6947         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6948         struct bio_vec *bvec = bio->bi_io_vec;
6949         struct inode *inode = dip->inode;
6950         struct btrfs_root *root = BTRFS_I(inode)->root;
6951         struct bio *dio_bio;
6952         u32 *csums = (u32 *)dip->csum;
6953         int index = 0;
6954         u64 start;
6955
6956         start = dip->logical_offset;
6957         do {
6958                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6959                         struct page *page = bvec->bv_page;
6960                         char *kaddr;
6961                         u32 csum = ~(u32)0;
6962                         unsigned long flags;
6963
6964                         local_irq_save(flags);
6965                         kaddr = kmap_atomic(page);
6966                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6967                                                csum, bvec->bv_len);
6968                         btrfs_csum_final(csum, (char *)&csum);
6969                         kunmap_atomic(kaddr);
6970                         local_irq_restore(flags);
6971
6972                         flush_dcache_page(bvec->bv_page);
6973                         if (csum != csums[index]) {
6974                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6975                                           btrfs_ino(inode), start, csum,
6976                                           csums[index]);
6977                                 err = -EIO;
6978                         }
6979                 }
6980
6981                 start += bvec->bv_len;
6982                 bvec++;
6983                 index++;
6984         } while (bvec <= bvec_end);
6985
6986         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6987                       dip->logical_offset + dip->bytes - 1);
6988         dio_bio = dip->dio_bio;
6989
6990         kfree(dip);
6991
6992         /* If we had a csum failure make sure to clear the uptodate flag */
6993         if (err)
6994                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6995         dio_end_io(dio_bio, err);
6996         bio_put(bio);
6997 }
6998
6999 static void btrfs_endio_direct_write(struct bio *bio, int err)
7000 {
7001         struct btrfs_dio_private *dip = bio->bi_private;
7002         struct inode *inode = dip->inode;
7003         struct btrfs_root *root = BTRFS_I(inode)->root;
7004         struct btrfs_ordered_extent *ordered = NULL;
7005         u64 ordered_offset = dip->logical_offset;
7006         u64 ordered_bytes = dip->bytes;
7007         struct bio *dio_bio;
7008         int ret;
7009
7010         if (err)
7011                 goto out_done;
7012 again:
7013         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7014                                                    &ordered_offset,
7015                                                    ordered_bytes, !err);
7016         if (!ret)
7017                 goto out_test;
7018
7019         ordered->work.func = finish_ordered_fn;
7020         ordered->work.flags = 0;
7021         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7022                            &ordered->work);
7023 out_test:
7024         /*
7025          * our bio might span multiple ordered extents.  If we haven't
7026          * completed the accounting for the whole dio, go back and try again
7027          */
7028         if (ordered_offset < dip->logical_offset + dip->bytes) {
7029                 ordered_bytes = dip->logical_offset + dip->bytes -
7030                         ordered_offset;
7031                 ordered = NULL;
7032                 goto again;
7033         }
7034 out_done:
7035         dio_bio = dip->dio_bio;
7036
7037         kfree(dip);
7038
7039         /* If we had an error make sure to clear the uptodate flag */
7040         if (err)
7041                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7042         dio_end_io(dio_bio, err);
7043         bio_put(bio);
7044 }
7045
7046 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7047                                     struct bio *bio, int mirror_num,
7048                                     unsigned long bio_flags, u64 offset)
7049 {
7050         int ret;
7051         struct btrfs_root *root = BTRFS_I(inode)->root;
7052         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7053         BUG_ON(ret); /* -ENOMEM */
7054         return 0;
7055 }
7056
7057 static void btrfs_end_dio_bio(struct bio *bio, int err)
7058 {
7059         struct btrfs_dio_private *dip = bio->bi_private;
7060
7061         if (err) {
7062                 btrfs_err(BTRFS_I(dip->inode)->root->fs_info,
7063                           "direct IO failed ino %llu rw %lu sector %#Lx len %u err no %d",
7064                       btrfs_ino(dip->inode), bio->bi_rw,
7065                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7066                 dip->errors = 1;
7067
7068                 /*
7069                  * before atomic variable goto zero, we must make sure
7070                  * dip->errors is perceived to be set.
7071                  */
7072                 smp_mb__before_atomic_dec();
7073         }
7074
7075         /* if there are more bios still pending for this dio, just exit */
7076         if (!atomic_dec_and_test(&dip->pending_bios))
7077                 goto out;
7078
7079         if (dip->errors) {
7080                 bio_io_error(dip->orig_bio);
7081         } else {
7082                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7083                 bio_endio(dip->orig_bio, 0);
7084         }
7085 out:
7086         bio_put(bio);
7087 }
7088
7089 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7090                                        u64 first_sector, gfp_t gfp_flags)
7091 {
7092         int nr_vecs = bio_get_nr_vecs(bdev);
7093         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7094 }
7095
7096 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7097                                          int rw, u64 file_offset, int skip_sum,
7098                                          int async_submit)
7099 {
7100         struct btrfs_dio_private *dip = bio->bi_private;
7101         int write = rw & REQ_WRITE;
7102         struct btrfs_root *root = BTRFS_I(inode)->root;
7103         int ret;
7104
7105         if (async_submit)
7106                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7107
7108         bio_get(bio);
7109
7110         if (!write) {
7111                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7112                 if (ret)
7113                         goto err;
7114         }
7115
7116         if (skip_sum)
7117                 goto map;
7118
7119         if (write && async_submit) {
7120                 ret = btrfs_wq_submit_bio(root->fs_info,
7121                                    inode, rw, bio, 0, 0,
7122                                    file_offset,
7123                                    __btrfs_submit_bio_start_direct_io,
7124                                    __btrfs_submit_bio_done);
7125                 goto err;
7126         } else if (write) {
7127                 /*
7128                  * If we aren't doing async submit, calculate the csum of the
7129                  * bio now.
7130                  */
7131                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7132                 if (ret)
7133                         goto err;
7134         } else if (!skip_sum) {
7135                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7136                                                 file_offset);
7137                 if (ret)
7138                         goto err;
7139         }
7140
7141 map:
7142         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7143 err:
7144         bio_put(bio);
7145         return ret;
7146 }
7147
7148 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7149                                     int skip_sum)
7150 {
7151         struct inode *inode = dip->inode;
7152         struct btrfs_root *root = BTRFS_I(inode)->root;
7153         struct bio *bio;
7154         struct bio *orig_bio = dip->orig_bio;
7155         struct bio_vec *bvec = orig_bio->bi_io_vec;
7156         u64 start_sector = orig_bio->bi_sector;
7157         u64 file_offset = dip->logical_offset;
7158         u64 submit_len = 0;
7159         u64 map_length;
7160         int nr_pages = 0;
7161         int ret = 0;
7162         int async_submit = 0;
7163
7164         map_length = orig_bio->bi_size;
7165         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7166                               &map_length, NULL, 0);
7167         if (ret) {
7168                 bio_put(orig_bio);
7169                 return -EIO;
7170         }
7171
7172         if (map_length >= orig_bio->bi_size) {
7173                 bio = orig_bio;
7174                 goto submit;
7175         }
7176
7177         /* async crcs make it difficult to collect full stripe writes. */
7178         if (btrfs_get_alloc_profile(root, 1) &
7179             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7180                 async_submit = 0;
7181         else
7182                 async_submit = 1;
7183
7184         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7185         if (!bio)
7186                 return -ENOMEM;
7187         bio->bi_private = dip;
7188         bio->bi_end_io = btrfs_end_dio_bio;
7189         atomic_inc(&dip->pending_bios);
7190
7191         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7192                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7193                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7194                                  bvec->bv_offset) < bvec->bv_len)) {
7195                         /*
7196                          * inc the count before we submit the bio so
7197                          * we know the end IO handler won't happen before
7198                          * we inc the count. Otherwise, the dip might get freed
7199                          * before we're done setting it up
7200                          */
7201                         atomic_inc(&dip->pending_bios);
7202                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7203                                                      file_offset, skip_sum,
7204                                                      async_submit);
7205                         if (ret) {
7206                                 bio_put(bio);
7207                                 atomic_dec(&dip->pending_bios);
7208                                 goto out_err;
7209                         }
7210
7211                         start_sector += submit_len >> 9;
7212                         file_offset += submit_len;
7213
7214                         submit_len = 0;
7215                         nr_pages = 0;
7216
7217                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7218                                                   start_sector, GFP_NOFS);
7219                         if (!bio)
7220                                 goto out_err;
7221                         bio->bi_private = dip;
7222                         bio->bi_end_io = btrfs_end_dio_bio;
7223
7224                         map_length = orig_bio->bi_size;
7225                         ret = btrfs_map_block(root->fs_info, rw,
7226                                               start_sector << 9,
7227                                               &map_length, NULL, 0);
7228                         if (ret) {
7229                                 bio_put(bio);
7230                                 goto out_err;
7231                         }
7232                 } else {
7233                         submit_len += bvec->bv_len;
7234                         nr_pages++;
7235                         bvec++;
7236                 }
7237         }
7238
7239 submit:
7240         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7241                                      async_submit);
7242         if (!ret)
7243                 return 0;
7244
7245         bio_put(bio);
7246 out_err:
7247         dip->errors = 1;
7248         /*
7249          * before atomic variable goto zero, we must
7250          * make sure dip->errors is perceived to be set.
7251          */
7252         smp_mb__before_atomic_dec();
7253         if (atomic_dec_and_test(&dip->pending_bios))
7254                 bio_io_error(dip->orig_bio);
7255
7256         /* bio_end_io() will handle error, so we needn't return it */
7257         return 0;
7258 }
7259
7260 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7261                                 struct inode *inode, loff_t file_offset)
7262 {
7263         struct btrfs_root *root = BTRFS_I(inode)->root;
7264         struct btrfs_dio_private *dip;
7265         struct bio *io_bio;
7266         int skip_sum;
7267         int sum_len;
7268         int write = rw & REQ_WRITE;
7269         int ret = 0;
7270         u16 csum_size;
7271
7272         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7273
7274         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7275         if (!io_bio) {
7276                 ret = -ENOMEM;
7277                 goto free_ordered;
7278         }
7279
7280         if (!skip_sum && !write) {
7281                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7282                 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7283                 sum_len *= csum_size;
7284         } else {
7285                 sum_len = 0;
7286         }
7287
7288         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7289         if (!dip) {
7290                 ret = -ENOMEM;
7291                 goto free_io_bio;
7292         }
7293
7294         dip->private = dio_bio->bi_private;
7295         dip->inode = inode;
7296         dip->logical_offset = file_offset;
7297         dip->bytes = dio_bio->bi_size;
7298         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7299         io_bio->bi_private = dip;
7300         dip->errors = 0;
7301         dip->orig_bio = io_bio;
7302         dip->dio_bio = dio_bio;
7303         atomic_set(&dip->pending_bios, 0);
7304
7305         if (write)
7306                 io_bio->bi_end_io = btrfs_endio_direct_write;
7307         else
7308                 io_bio->bi_end_io = btrfs_endio_direct_read;
7309
7310         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7311         if (!ret)
7312                 return;
7313
7314 free_io_bio:
7315         bio_put(io_bio);
7316
7317 free_ordered:
7318         /*
7319          * If this is a write, we need to clean up the reserved space and kill
7320          * the ordered extent.
7321          */
7322         if (write) {
7323                 struct btrfs_ordered_extent *ordered;
7324                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7325                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7326                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7327                         btrfs_free_reserved_extent(root, ordered->start,
7328                                                    ordered->disk_len);
7329                 btrfs_put_ordered_extent(ordered);
7330                 btrfs_put_ordered_extent(ordered);
7331         }
7332         bio_endio(dio_bio, ret);
7333 }
7334
7335 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7336                         const struct iovec *iov, loff_t offset,
7337                         unsigned long nr_segs)
7338 {
7339         int seg;
7340         int i;
7341         size_t size;
7342         unsigned long addr;
7343         unsigned blocksize_mask = root->sectorsize - 1;
7344         ssize_t retval = -EINVAL;
7345         loff_t end = offset;
7346
7347         if (offset & blocksize_mask)
7348                 goto out;
7349
7350         /* Check the memory alignment.  Blocks cannot straddle pages */
7351         for (seg = 0; seg < nr_segs; seg++) {
7352                 addr = (unsigned long)iov[seg].iov_base;
7353                 size = iov[seg].iov_len;
7354                 end += size;
7355                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7356                         goto out;
7357
7358                 /* If this is a write we don't need to check anymore */
7359                 if (rw & WRITE)
7360                         continue;
7361
7362                 /*
7363                  * Check to make sure we don't have duplicate iov_base's in this
7364                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7365                  * when reading back.
7366                  */
7367                 for (i = seg + 1; i < nr_segs; i++) {
7368                         if (iov[seg].iov_base == iov[i].iov_base)
7369                                 goto out;
7370                 }
7371         }
7372         retval = 0;
7373 out:
7374         return retval;
7375 }
7376
7377 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7378                         const struct iovec *iov, loff_t offset,
7379                         unsigned long nr_segs)
7380 {
7381         struct file *file = iocb->ki_filp;
7382         struct inode *inode = file->f_mapping->host;
7383         size_t count = 0;
7384         int flags = 0;
7385         bool wakeup = true;
7386         bool relock = false;
7387         ssize_t ret;
7388
7389         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7390                             offset, nr_segs))
7391                 return 0;
7392
7393         atomic_inc(&inode->i_dio_count);
7394         smp_mb__after_atomic_inc();
7395
7396         /*
7397          * The generic stuff only does filemap_write_and_wait_range, which isn't
7398          * enough if we've written compressed pages to this area, so we need to
7399          * call btrfs_wait_ordered_range to make absolutely sure that any
7400          * outstanding dirty pages are on disk.
7401          */
7402         count = iov_length(iov, nr_segs);
7403         ret = btrfs_wait_ordered_range(inode, offset, count);
7404         if (ret)
7405                 return ret;
7406
7407         if (rw & WRITE) {
7408                 /*
7409                  * If the write DIO is beyond the EOF, we need update
7410                  * the isize, but it is protected by i_mutex. So we can
7411                  * not unlock the i_mutex at this case.
7412                  */
7413                 if (offset + count <= inode->i_size) {
7414                         mutex_unlock(&inode->i_mutex);
7415                         relock = true;
7416                 }
7417                 ret = btrfs_delalloc_reserve_space(inode, count);
7418                 if (ret)
7419                         goto out;
7420         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7421                                      &BTRFS_I(inode)->runtime_flags))) {
7422                 inode_dio_done(inode);
7423                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7424                 wakeup = false;
7425         }
7426
7427         ret = __blockdev_direct_IO(rw, iocb, inode,
7428                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7429                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7430                         btrfs_submit_direct, flags);
7431         if (rw & WRITE) {
7432                 if (ret < 0 && ret != -EIOCBQUEUED)
7433                         btrfs_delalloc_release_space(inode, count);
7434                 else if (ret >= 0 && (size_t)ret < count)
7435                         btrfs_delalloc_release_space(inode,
7436                                                      count - (size_t)ret);
7437                 else
7438                         btrfs_delalloc_release_metadata(inode, 0);
7439         }
7440 out:
7441         if (wakeup)
7442                 inode_dio_done(inode);
7443         if (relock)
7444                 mutex_lock(&inode->i_mutex);
7445
7446         return ret;
7447 }
7448
7449 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7450
7451 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7452                 __u64 start, __u64 len)
7453 {
7454         int     ret;
7455
7456         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7457         if (ret)
7458                 return ret;
7459
7460         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7461 }
7462
7463 int btrfs_readpage(struct file *file, struct page *page)
7464 {
7465         struct extent_io_tree *tree;
7466         tree = &BTRFS_I(page->mapping->host)->io_tree;
7467         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7468 }
7469
7470 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7471 {
7472         struct extent_io_tree *tree;
7473
7474
7475         if (current->flags & PF_MEMALLOC) {
7476                 redirty_page_for_writepage(wbc, page);
7477                 unlock_page(page);
7478                 return 0;
7479         }
7480         tree = &BTRFS_I(page->mapping->host)->io_tree;
7481         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7482 }
7483
7484 static int btrfs_writepages(struct address_space *mapping,
7485                             struct writeback_control *wbc)
7486 {
7487         struct extent_io_tree *tree;
7488
7489         tree = &BTRFS_I(mapping->host)->io_tree;
7490         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7491 }
7492
7493 static int
7494 btrfs_readpages(struct file *file, struct address_space *mapping,
7495                 struct list_head *pages, unsigned nr_pages)
7496 {
7497         struct extent_io_tree *tree;
7498         tree = &BTRFS_I(mapping->host)->io_tree;
7499         return extent_readpages(tree, mapping, pages, nr_pages,
7500                                 btrfs_get_extent);
7501 }
7502 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7503 {
7504         struct extent_io_tree *tree;
7505         struct extent_map_tree *map;
7506         int ret;
7507
7508         tree = &BTRFS_I(page->mapping->host)->io_tree;
7509         map = &BTRFS_I(page->mapping->host)->extent_tree;
7510         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7511         if (ret == 1) {
7512                 ClearPagePrivate(page);
7513                 set_page_private(page, 0);
7514                 page_cache_release(page);
7515         }
7516         return ret;
7517 }
7518
7519 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7520 {
7521         if (PageWriteback(page) || PageDirty(page))
7522                 return 0;
7523         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7524 }
7525
7526 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7527                                  unsigned int length)
7528 {
7529         struct inode *inode = page->mapping->host;
7530         struct extent_io_tree *tree;
7531         struct btrfs_ordered_extent *ordered;
7532         struct extent_state *cached_state = NULL;
7533         u64 page_start = page_offset(page);
7534         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7535         int inode_evicting = inode->i_state & I_FREEING;
7536
7537         /*
7538          * we have the page locked, so new writeback can't start,
7539          * and the dirty bit won't be cleared while we are here.
7540          *
7541          * Wait for IO on this page so that we can safely clear
7542          * the PagePrivate2 bit and do ordered accounting
7543          */
7544         wait_on_page_writeback(page);
7545
7546         tree = &BTRFS_I(inode)->io_tree;
7547         if (offset) {
7548                 btrfs_releasepage(page, GFP_NOFS);
7549                 return;
7550         }
7551
7552         if (!inode_evicting)
7553                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7554         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7555         if (ordered) {
7556                 /*
7557                  * IO on this page will never be started, so we need
7558                  * to account for any ordered extents now
7559                  */
7560                 if (!inode_evicting)
7561                         clear_extent_bit(tree, page_start, page_end,
7562                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7563                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7564                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7565                                          GFP_NOFS);
7566                 /*
7567                  * whoever cleared the private bit is responsible
7568                  * for the finish_ordered_io
7569                  */
7570                 if (TestClearPagePrivate2(page)) {
7571                         struct btrfs_ordered_inode_tree *tree;
7572                         u64 new_len;
7573
7574                         tree = &BTRFS_I(inode)->ordered_tree;
7575
7576                         spin_lock_irq(&tree->lock);
7577                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7578                         new_len = page_start - ordered->file_offset;
7579                         if (new_len < ordered->truncated_len)
7580                                 ordered->truncated_len = new_len;
7581                         spin_unlock_irq(&tree->lock);
7582
7583                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7584                                                            page_start,
7585                                                            PAGE_CACHE_SIZE, 1))
7586                                 btrfs_finish_ordered_io(ordered);
7587                 }
7588                 btrfs_put_ordered_extent(ordered);
7589                 if (!inode_evicting) {
7590                         cached_state = NULL;
7591                         lock_extent_bits(tree, page_start, page_end, 0,
7592                                          &cached_state);
7593                 }
7594         }
7595
7596         if (!inode_evicting) {
7597                 clear_extent_bit(tree, page_start, page_end,
7598                                  EXTENT_LOCKED | EXTENT_DIRTY |
7599                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7600                                  EXTENT_DEFRAG, 1, 1,
7601                                  &cached_state, GFP_NOFS);
7602
7603                 __btrfs_releasepage(page, GFP_NOFS);
7604         }
7605
7606         ClearPageChecked(page);
7607         if (PagePrivate(page)) {
7608                 ClearPagePrivate(page);
7609                 set_page_private(page, 0);
7610                 page_cache_release(page);
7611         }
7612 }
7613
7614 /*
7615  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7616  * called from a page fault handler when a page is first dirtied. Hence we must
7617  * be careful to check for EOF conditions here. We set the page up correctly
7618  * for a written page which means we get ENOSPC checking when writing into
7619  * holes and correct delalloc and unwritten extent mapping on filesystems that
7620  * support these features.
7621  *
7622  * We are not allowed to take the i_mutex here so we have to play games to
7623  * protect against truncate races as the page could now be beyond EOF.  Because
7624  * vmtruncate() writes the inode size before removing pages, once we have the
7625  * page lock we can determine safely if the page is beyond EOF. If it is not
7626  * beyond EOF, then the page is guaranteed safe against truncation until we
7627  * unlock the page.
7628  */
7629 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7630 {
7631         struct page *page = vmf->page;
7632         struct inode *inode = file_inode(vma->vm_file);
7633         struct btrfs_root *root = BTRFS_I(inode)->root;
7634         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7635         struct btrfs_ordered_extent *ordered;
7636         struct extent_state *cached_state = NULL;
7637         char *kaddr;
7638         unsigned long zero_start;
7639         loff_t size;
7640         int ret;
7641         int reserved = 0;
7642         u64 page_start;
7643         u64 page_end;
7644
7645         sb_start_pagefault(inode->i_sb);
7646         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7647         if (!ret) {
7648                 ret = file_update_time(vma->vm_file);
7649                 reserved = 1;
7650         }
7651         if (ret) {
7652                 if (ret == -ENOMEM)
7653                         ret = VM_FAULT_OOM;
7654                 else /* -ENOSPC, -EIO, etc */
7655                         ret = VM_FAULT_SIGBUS;
7656                 if (reserved)
7657                         goto out;
7658                 goto out_noreserve;
7659         }
7660
7661         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7662 again:
7663         lock_page(page);
7664         size = i_size_read(inode);
7665         page_start = page_offset(page);
7666         page_end = page_start + PAGE_CACHE_SIZE - 1;
7667
7668         if ((page->mapping != inode->i_mapping) ||
7669             (page_start >= size)) {
7670                 /* page got truncated out from underneath us */
7671                 goto out_unlock;
7672         }
7673         wait_on_page_writeback(page);
7674
7675         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7676         set_page_extent_mapped(page);
7677
7678         /*
7679          * we can't set the delalloc bits if there are pending ordered
7680          * extents.  Drop our locks and wait for them to finish
7681          */
7682         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7683         if (ordered) {
7684                 unlock_extent_cached(io_tree, page_start, page_end,
7685                                      &cached_state, GFP_NOFS);
7686                 unlock_page(page);
7687                 btrfs_start_ordered_extent(inode, ordered, 1);
7688                 btrfs_put_ordered_extent(ordered);
7689                 goto again;
7690         }
7691
7692         /*
7693          * XXX - page_mkwrite gets called every time the page is dirtied, even
7694          * if it was already dirty, so for space accounting reasons we need to
7695          * clear any delalloc bits for the range we are fixing to save.  There
7696          * is probably a better way to do this, but for now keep consistent with
7697          * prepare_pages in the normal write path.
7698          */
7699         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7700                           EXTENT_DIRTY | EXTENT_DELALLOC |
7701                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7702                           0, 0, &cached_state, GFP_NOFS);
7703
7704         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7705                                         &cached_state);
7706         if (ret) {
7707                 unlock_extent_cached(io_tree, page_start, page_end,
7708                                      &cached_state, GFP_NOFS);
7709                 ret = VM_FAULT_SIGBUS;
7710                 goto out_unlock;
7711         }
7712         ret = 0;
7713
7714         /* page is wholly or partially inside EOF */
7715         if (page_start + PAGE_CACHE_SIZE > size)
7716                 zero_start = size & ~PAGE_CACHE_MASK;
7717         else
7718                 zero_start = PAGE_CACHE_SIZE;
7719
7720         if (zero_start != PAGE_CACHE_SIZE) {
7721                 kaddr = kmap(page);
7722                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7723                 flush_dcache_page(page);
7724                 kunmap(page);
7725         }
7726         ClearPageChecked(page);
7727         set_page_dirty(page);
7728         SetPageUptodate(page);
7729
7730         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7731         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7732         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7733
7734         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7735
7736 out_unlock:
7737         if (!ret) {
7738                 sb_end_pagefault(inode->i_sb);
7739                 return VM_FAULT_LOCKED;
7740         }
7741         unlock_page(page);
7742 out:
7743         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7744 out_noreserve:
7745         sb_end_pagefault(inode->i_sb);
7746         return ret;
7747 }
7748
7749 static int btrfs_truncate(struct inode *inode)
7750 {
7751         struct btrfs_root *root = BTRFS_I(inode)->root;
7752         struct btrfs_block_rsv *rsv;
7753         int ret = 0;
7754         int err = 0;
7755         struct btrfs_trans_handle *trans;
7756         u64 mask = root->sectorsize - 1;
7757         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7758
7759         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7760                                        (u64)-1);
7761         if (ret)
7762                 return ret;
7763
7764         /*
7765          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7766          * 3 things going on here
7767          *
7768          * 1) We need to reserve space for our orphan item and the space to
7769          * delete our orphan item.  Lord knows we don't want to have a dangling
7770          * orphan item because we didn't reserve space to remove it.
7771          *
7772          * 2) We need to reserve space to update our inode.
7773          *
7774          * 3) We need to have something to cache all the space that is going to
7775          * be free'd up by the truncate operation, but also have some slack
7776          * space reserved in case it uses space during the truncate (thank you
7777          * very much snapshotting).
7778          *
7779          * And we need these to all be seperate.  The fact is we can use alot of
7780          * space doing the truncate, and we have no earthly idea how much space
7781          * we will use, so we need the truncate reservation to be seperate so it
7782          * doesn't end up using space reserved for updating the inode or
7783          * removing the orphan item.  We also need to be able to stop the
7784          * transaction and start a new one, which means we need to be able to
7785          * update the inode several times, and we have no idea of knowing how
7786          * many times that will be, so we can't just reserve 1 item for the
7787          * entirety of the opration, so that has to be done seperately as well.
7788          * Then there is the orphan item, which does indeed need to be held on
7789          * to for the whole operation, and we need nobody to touch this reserved
7790          * space except the orphan code.
7791          *
7792          * So that leaves us with
7793          *
7794          * 1) root->orphan_block_rsv - for the orphan deletion.
7795          * 2) rsv - for the truncate reservation, which we will steal from the
7796          * transaction reservation.
7797          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7798          * updating the inode.
7799          */
7800         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7801         if (!rsv)
7802                 return -ENOMEM;
7803         rsv->size = min_size;
7804         rsv->failfast = 1;
7805
7806         /*
7807          * 1 for the truncate slack space
7808          * 1 for updating the inode.
7809          */
7810         trans = btrfs_start_transaction(root, 2);
7811         if (IS_ERR(trans)) {
7812                 err = PTR_ERR(trans);
7813                 goto out;
7814         }
7815
7816         /* Migrate the slack space for the truncate to our reserve */
7817         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7818                                       min_size);
7819         BUG_ON(ret);
7820
7821         /*
7822          * setattr is responsible for setting the ordered_data_close flag,
7823          * but that is only tested during the last file release.  That
7824          * could happen well after the next commit, leaving a great big
7825          * window where new writes may get lost if someone chooses to write
7826          * to this file after truncating to zero
7827          *
7828          * The inode doesn't have any dirty data here, and so if we commit
7829          * this is a noop.  If someone immediately starts writing to the inode
7830          * it is very likely we'll catch some of their writes in this
7831          * transaction, and the commit will find this file on the ordered
7832          * data list with good things to send down.
7833          *
7834          * This is a best effort solution, there is still a window where
7835          * using truncate to replace the contents of the file will
7836          * end up with a zero length file after a crash.
7837          */
7838         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7839                                            &BTRFS_I(inode)->runtime_flags))
7840                 btrfs_add_ordered_operation(trans, root, inode);
7841
7842         /*
7843          * So if we truncate and then write and fsync we normally would just
7844          * write the extents that changed, which is a problem if we need to
7845          * first truncate that entire inode.  So set this flag so we write out
7846          * all of the extents in the inode to the sync log so we're completely
7847          * safe.
7848          */
7849         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7850         trans->block_rsv = rsv;
7851
7852         while (1) {
7853                 ret = btrfs_truncate_inode_items(trans, root, inode,
7854                                                  inode->i_size,
7855                                                  BTRFS_EXTENT_DATA_KEY);
7856                 if (ret != -ENOSPC) {
7857                         err = ret;
7858                         break;
7859                 }
7860
7861                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7862                 ret = btrfs_update_inode(trans, root, inode);
7863                 if (ret) {
7864                         err = ret;
7865                         break;
7866                 }
7867
7868                 btrfs_end_transaction(trans, root);
7869                 btrfs_btree_balance_dirty(root);
7870
7871                 trans = btrfs_start_transaction(root, 2);
7872                 if (IS_ERR(trans)) {
7873                         ret = err = PTR_ERR(trans);
7874                         trans = NULL;
7875                         break;
7876                 }
7877
7878                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7879                                               rsv, min_size);
7880                 BUG_ON(ret);    /* shouldn't happen */
7881                 trans->block_rsv = rsv;
7882         }
7883
7884         if (ret == 0 && inode->i_nlink > 0) {
7885                 trans->block_rsv = root->orphan_block_rsv;
7886                 ret = btrfs_orphan_del(trans, inode);
7887                 if (ret)
7888                         err = ret;
7889         }
7890
7891         if (trans) {
7892                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7893                 ret = btrfs_update_inode(trans, root, inode);
7894                 if (ret && !err)
7895                         err = ret;
7896
7897                 ret = btrfs_end_transaction(trans, root);
7898                 btrfs_btree_balance_dirty(root);
7899         }
7900
7901 out:
7902         btrfs_free_block_rsv(root, rsv);
7903
7904         if (ret && !err)
7905                 err = ret;
7906
7907         return err;
7908 }
7909
7910 /*
7911  * create a new subvolume directory/inode (helper for the ioctl).
7912  */
7913 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7914                              struct btrfs_root *new_root,
7915                              struct btrfs_root *parent_root,
7916                              u64 new_dirid)
7917 {
7918         struct inode *inode;
7919         int err;
7920         u64 index = 0;
7921
7922         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7923                                 new_dirid, new_dirid,
7924                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7925                                 &index);
7926         if (IS_ERR(inode))
7927                 return PTR_ERR(inode);
7928         inode->i_op = &btrfs_dir_inode_operations;
7929         inode->i_fop = &btrfs_dir_file_operations;
7930
7931         set_nlink(inode, 1);
7932         btrfs_i_size_write(inode, 0);
7933
7934         err = btrfs_subvol_inherit_props(trans, new_root, parent_root);
7935         if (err)
7936                 btrfs_err(new_root->fs_info,
7937                           "error inheriting subvolume %llu properties: %d\n",
7938                           new_root->root_key.objectid, err);
7939
7940         err = btrfs_update_inode(trans, new_root, inode);
7941
7942         iput(inode);
7943         return err;
7944 }
7945
7946 struct inode *btrfs_alloc_inode(struct super_block *sb)
7947 {
7948         struct btrfs_inode *ei;
7949         struct inode *inode;
7950
7951         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7952         if (!ei)
7953                 return NULL;
7954
7955         ei->root = NULL;
7956         ei->generation = 0;
7957         ei->last_trans = 0;
7958         ei->last_sub_trans = 0;
7959         ei->logged_trans = 0;
7960         ei->delalloc_bytes = 0;
7961         ei->disk_i_size = 0;
7962         ei->flags = 0;
7963         ei->csum_bytes = 0;
7964         ei->index_cnt = (u64)-1;
7965         ei->dir_index = 0;
7966         ei->last_unlink_trans = 0;
7967         ei->last_log_commit = 0;
7968
7969         spin_lock_init(&ei->lock);
7970         ei->outstanding_extents = 0;
7971         ei->reserved_extents = 0;
7972
7973         ei->runtime_flags = 0;
7974         ei->force_compress = BTRFS_COMPRESS_NONE;
7975
7976         ei->delayed_node = NULL;
7977
7978         inode = &ei->vfs_inode;
7979         extent_map_tree_init(&ei->extent_tree);
7980         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7981         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7982         ei->io_tree.track_uptodate = 1;
7983         ei->io_failure_tree.track_uptodate = 1;
7984         atomic_set(&ei->sync_writers, 0);
7985         mutex_init(&ei->log_mutex);
7986         mutex_init(&ei->delalloc_mutex);
7987         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7988         INIT_LIST_HEAD(&ei->delalloc_inodes);
7989         INIT_LIST_HEAD(&ei->ordered_operations);
7990         RB_CLEAR_NODE(&ei->rb_node);
7991
7992         return inode;
7993 }
7994
7995 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7996 void btrfs_test_destroy_inode(struct inode *inode)
7997 {
7998         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7999         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8000 }
8001 #endif
8002
8003 static void btrfs_i_callback(struct rcu_head *head)
8004 {
8005         struct inode *inode = container_of(head, struct inode, i_rcu);
8006         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
8007 }
8008
8009 void btrfs_destroy_inode(struct inode *inode)
8010 {
8011         struct btrfs_ordered_extent *ordered;
8012         struct btrfs_root *root = BTRFS_I(inode)->root;
8013
8014         WARN_ON(!hlist_empty(&inode->i_dentry));
8015         WARN_ON(inode->i_data.nrpages);
8016         WARN_ON(BTRFS_I(inode)->outstanding_extents);
8017         WARN_ON(BTRFS_I(inode)->reserved_extents);
8018         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
8019         WARN_ON(BTRFS_I(inode)->csum_bytes);
8020
8021         /*
8022          * This can happen where we create an inode, but somebody else also
8023          * created the same inode and we need to destroy the one we already
8024          * created.
8025          */
8026         if (!root)
8027                 goto free;
8028
8029         /*
8030          * Make sure we're properly removed from the ordered operation
8031          * lists.
8032          */
8033         smp_mb();
8034         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
8035                 spin_lock(&root->fs_info->ordered_root_lock);
8036                 list_del_init(&BTRFS_I(inode)->ordered_operations);
8037                 spin_unlock(&root->fs_info->ordered_root_lock);
8038         }
8039
8040         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
8041                      &BTRFS_I(inode)->runtime_flags)) {
8042                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
8043                         btrfs_ino(inode));
8044                 atomic_dec(&root->orphan_inodes);
8045         }
8046
8047         while (1) {
8048                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
8049                 if (!ordered)
8050                         break;
8051                 else {
8052                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
8053                                 ordered->file_offset, ordered->len);
8054                         btrfs_remove_ordered_extent(inode, ordered);
8055                         btrfs_put_ordered_extent(ordered);
8056                         btrfs_put_ordered_extent(ordered);
8057                 }
8058         }
8059         inode_tree_del(inode);
8060         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8061 free:
8062         call_rcu(&inode->i_rcu, btrfs_i_callback);
8063 }
8064
8065 int btrfs_drop_inode(struct inode *inode)
8066 {
8067         struct btrfs_root *root = BTRFS_I(inode)->root;
8068
8069         if (root == NULL)
8070                 return 1;
8071
8072         /* the snap/subvol tree is on deleting */
8073         if (btrfs_root_refs(&root->root_item) == 0)
8074                 return 1;
8075         else
8076                 return generic_drop_inode(inode);
8077 }
8078
8079 static void init_once(void *foo)
8080 {
8081         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8082
8083         inode_init_once(&ei->vfs_inode);
8084 }
8085
8086 void btrfs_destroy_cachep(void)
8087 {
8088         /*
8089          * Make sure all delayed rcu free inodes are flushed before we
8090          * destroy cache.
8091          */
8092         rcu_barrier();
8093         if (btrfs_inode_cachep)
8094                 kmem_cache_destroy(btrfs_inode_cachep);
8095         if (btrfs_trans_handle_cachep)
8096                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8097         if (btrfs_transaction_cachep)
8098                 kmem_cache_destroy(btrfs_transaction_cachep);
8099         if (btrfs_path_cachep)
8100                 kmem_cache_destroy(btrfs_path_cachep);
8101         if (btrfs_free_space_cachep)
8102                 kmem_cache_destroy(btrfs_free_space_cachep);
8103         if (btrfs_delalloc_work_cachep)
8104                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8105 }
8106
8107 int btrfs_init_cachep(void)
8108 {
8109         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8110                         sizeof(struct btrfs_inode), 0,
8111                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8112         if (!btrfs_inode_cachep)
8113                 goto fail;
8114
8115         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8116                         sizeof(struct btrfs_trans_handle), 0,
8117                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8118         if (!btrfs_trans_handle_cachep)
8119                 goto fail;
8120
8121         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8122                         sizeof(struct btrfs_transaction), 0,
8123                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8124         if (!btrfs_transaction_cachep)
8125                 goto fail;
8126
8127         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8128                         sizeof(struct btrfs_path), 0,
8129                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8130         if (!btrfs_path_cachep)
8131                 goto fail;
8132
8133         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8134                         sizeof(struct btrfs_free_space), 0,
8135                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8136         if (!btrfs_free_space_cachep)
8137                 goto fail;
8138
8139         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8140                         sizeof(struct btrfs_delalloc_work), 0,
8141                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8142                         NULL);
8143         if (!btrfs_delalloc_work_cachep)
8144                 goto fail;
8145
8146         return 0;
8147 fail:
8148         btrfs_destroy_cachep();
8149         return -ENOMEM;
8150 }
8151
8152 static int btrfs_getattr(struct vfsmount *mnt,
8153                          struct dentry *dentry, struct kstat *stat)
8154 {
8155         u64 delalloc_bytes;
8156         struct inode *inode = dentry->d_inode;
8157         u32 blocksize = inode->i_sb->s_blocksize;
8158
8159         generic_fillattr(inode, stat);
8160         stat->dev = BTRFS_I(inode)->root->anon_dev;
8161         stat->blksize = PAGE_CACHE_SIZE;
8162
8163         spin_lock(&BTRFS_I(inode)->lock);
8164         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8165         spin_unlock(&BTRFS_I(inode)->lock);
8166         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8167                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8168         return 0;
8169 }
8170
8171 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8172                            struct inode *new_dir, struct dentry *new_dentry)
8173 {
8174         struct btrfs_trans_handle *trans;
8175         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8176         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8177         struct inode *new_inode = new_dentry->d_inode;
8178         struct inode *old_inode = old_dentry->d_inode;
8179         struct timespec ctime = CURRENT_TIME;
8180         u64 index = 0;
8181         u64 root_objectid;
8182         int ret;
8183         u64 old_ino = btrfs_ino(old_inode);
8184
8185         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8186                 return -EPERM;
8187
8188         /* we only allow rename subvolume link between subvolumes */
8189         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8190                 return -EXDEV;
8191
8192         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8193             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8194                 return -ENOTEMPTY;
8195
8196         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8197             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8198                 return -ENOTEMPTY;
8199
8200
8201         /* check for collisions, even if the  name isn't there */
8202         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8203                              new_dentry->d_name.name,
8204                              new_dentry->d_name.len);
8205
8206         if (ret) {
8207                 if (ret == -EEXIST) {
8208                         /* we shouldn't get
8209                          * eexist without a new_inode */
8210                         if (WARN_ON(!new_inode)) {
8211                                 return ret;
8212                         }
8213                 } else {
8214                         /* maybe -EOVERFLOW */
8215                         return ret;
8216                 }
8217         }
8218         ret = 0;
8219
8220         /*
8221          * we're using rename to replace one file with another.
8222          * and the replacement file is large.  Start IO on it now so
8223          * we don't add too much work to the end of the transaction
8224          */
8225         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8226             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8227                 filemap_flush(old_inode->i_mapping);
8228
8229         /* close the racy window with snapshot create/destroy ioctl */
8230         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8231                 down_read(&root->fs_info->subvol_sem);
8232         /*
8233          * We want to reserve the absolute worst case amount of items.  So if
8234          * both inodes are subvols and we need to unlink them then that would
8235          * require 4 item modifications, but if they are both normal inodes it
8236          * would require 5 item modifications, so we'll assume their normal
8237          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8238          * should cover the worst case number of items we'll modify.
8239          */
8240         trans = btrfs_start_transaction(root, 11);
8241         if (IS_ERR(trans)) {
8242                 ret = PTR_ERR(trans);
8243                 goto out_notrans;
8244         }
8245
8246         if (dest != root)
8247                 btrfs_record_root_in_trans(trans, dest);
8248
8249         ret = btrfs_set_inode_index(new_dir, &index);
8250         if (ret)
8251                 goto out_fail;
8252
8253         BTRFS_I(old_inode)->dir_index = 0ULL;
8254         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8255                 /* force full log commit if subvolume involved. */
8256                 root->fs_info->last_trans_log_full_commit = trans->transid;
8257         } else {
8258                 ret = btrfs_insert_inode_ref(trans, dest,
8259                                              new_dentry->d_name.name,
8260                                              new_dentry->d_name.len,
8261                                              old_ino,
8262                                              btrfs_ino(new_dir), index);
8263                 if (ret)
8264                         goto out_fail;
8265                 /*
8266                  * this is an ugly little race, but the rename is required
8267                  * to make sure that if we crash, the inode is either at the
8268                  * old name or the new one.  pinning the log transaction lets
8269                  * us make sure we don't allow a log commit to come in after
8270                  * we unlink the name but before we add the new name back in.
8271                  */
8272                 btrfs_pin_log_trans(root);
8273         }
8274         /*
8275          * make sure the inode gets flushed if it is replacing
8276          * something.
8277          */
8278         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8279                 btrfs_add_ordered_operation(trans, root, old_inode);
8280
8281         inode_inc_iversion(old_dir);
8282         inode_inc_iversion(new_dir);
8283         inode_inc_iversion(old_inode);
8284         old_dir->i_ctime = old_dir->i_mtime = ctime;
8285         new_dir->i_ctime = new_dir->i_mtime = ctime;
8286         old_inode->i_ctime = ctime;
8287
8288         if (old_dentry->d_parent != new_dentry->d_parent)
8289                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8290
8291         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8292                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8293                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8294                                         old_dentry->d_name.name,
8295                                         old_dentry->d_name.len);
8296         } else {
8297                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8298                                         old_dentry->d_inode,
8299                                         old_dentry->d_name.name,
8300                                         old_dentry->d_name.len);
8301                 if (!ret)
8302                         ret = btrfs_update_inode(trans, root, old_inode);
8303         }
8304         if (ret) {
8305                 btrfs_abort_transaction(trans, root, ret);
8306                 goto out_fail;
8307         }
8308
8309         if (new_inode) {
8310                 inode_inc_iversion(new_inode);
8311                 new_inode->i_ctime = CURRENT_TIME;
8312                 if (unlikely(btrfs_ino(new_inode) ==
8313                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8314                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8315                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8316                                                 root_objectid,
8317                                                 new_dentry->d_name.name,
8318                                                 new_dentry->d_name.len);
8319                         BUG_ON(new_inode->i_nlink == 0);
8320                 } else {
8321                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8322                                                  new_dentry->d_inode,
8323                                                  new_dentry->d_name.name,
8324                                                  new_dentry->d_name.len);
8325                 }
8326                 if (!ret && new_inode->i_nlink == 0)
8327                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8328                 if (ret) {
8329                         btrfs_abort_transaction(trans, root, ret);
8330                         goto out_fail;
8331                 }
8332         }
8333
8334         ret = btrfs_add_link(trans, new_dir, old_inode,
8335                              new_dentry->d_name.name,
8336                              new_dentry->d_name.len, 0, index);
8337         if (ret) {
8338                 btrfs_abort_transaction(trans, root, ret);
8339                 goto out_fail;
8340         }
8341
8342         if (old_inode->i_nlink == 1)
8343                 BTRFS_I(old_inode)->dir_index = index;
8344
8345         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8346                 struct dentry *parent = new_dentry->d_parent;
8347                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8348                 btrfs_end_log_trans(root);
8349         }
8350 out_fail:
8351         btrfs_end_transaction(trans, root);
8352 out_notrans:
8353         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8354                 up_read(&root->fs_info->subvol_sem);
8355
8356         return ret;
8357 }
8358
8359 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8360 {
8361         struct btrfs_delalloc_work *delalloc_work;
8362         struct inode *inode;
8363
8364         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8365                                      work);
8366         inode = delalloc_work->inode;
8367         if (delalloc_work->wait) {
8368                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8369         } else {
8370                 filemap_flush(inode->i_mapping);
8371                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8372                              &BTRFS_I(inode)->runtime_flags))
8373                         filemap_flush(inode->i_mapping);
8374         }
8375
8376         if (delalloc_work->delay_iput)
8377                 btrfs_add_delayed_iput(inode);
8378         else
8379                 iput(inode);
8380         complete(&delalloc_work->completion);
8381 }
8382
8383 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8384                                                     int wait, int delay_iput)
8385 {
8386         struct btrfs_delalloc_work *work;
8387
8388         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8389         if (!work)
8390                 return NULL;
8391
8392         init_completion(&work->completion);
8393         INIT_LIST_HEAD(&work->list);
8394         work->inode = inode;
8395         work->wait = wait;
8396         work->delay_iput = delay_iput;
8397         work->work.func = btrfs_run_delalloc_work;
8398
8399         return work;
8400 }
8401
8402 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8403 {
8404         wait_for_completion(&work->completion);
8405         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8406 }
8407
8408 /*
8409  * some fairly slow code that needs optimization. This walks the list
8410  * of all the inodes with pending delalloc and forces them to disk.
8411  */
8412 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8413 {
8414         struct btrfs_inode *binode;
8415         struct inode *inode;
8416         struct btrfs_delalloc_work *work, *next;
8417         struct list_head works;
8418         struct list_head splice;
8419         int ret = 0;
8420
8421         INIT_LIST_HEAD(&works);
8422         INIT_LIST_HEAD(&splice);
8423
8424         spin_lock(&root->delalloc_lock);
8425         list_splice_init(&root->delalloc_inodes, &splice);
8426         while (!list_empty(&splice)) {
8427                 binode = list_entry(splice.next, struct btrfs_inode,
8428                                     delalloc_inodes);
8429
8430                 list_move_tail(&binode->delalloc_inodes,
8431                                &root->delalloc_inodes);
8432                 inode = igrab(&binode->vfs_inode);
8433                 if (!inode) {
8434                         cond_resched_lock(&root->delalloc_lock);
8435                         continue;
8436                 }
8437                 spin_unlock(&root->delalloc_lock);
8438
8439                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8440                 if (unlikely(!work)) {
8441                         if (delay_iput)
8442                                 btrfs_add_delayed_iput(inode);
8443                         else
8444                                 iput(inode);
8445                         ret = -ENOMEM;
8446                         goto out;
8447                 }
8448                 list_add_tail(&work->list, &works);
8449                 btrfs_queue_worker(&root->fs_info->flush_workers,
8450                                    &work->work);
8451
8452                 cond_resched();
8453                 spin_lock(&root->delalloc_lock);
8454         }
8455         spin_unlock(&root->delalloc_lock);
8456
8457         list_for_each_entry_safe(work, next, &works, list) {
8458                 list_del_init(&work->list);
8459                 btrfs_wait_and_free_delalloc_work(work);
8460         }
8461         return 0;
8462 out:
8463         list_for_each_entry_safe(work, next, &works, list) {
8464                 list_del_init(&work->list);
8465                 btrfs_wait_and_free_delalloc_work(work);
8466         }
8467
8468         if (!list_empty_careful(&splice)) {
8469                 spin_lock(&root->delalloc_lock);
8470                 list_splice_tail(&splice, &root->delalloc_inodes);
8471                 spin_unlock(&root->delalloc_lock);
8472         }
8473         return ret;
8474 }
8475
8476 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8477 {
8478         int ret;
8479
8480         if (test_bit(BTRFS_FS_STATE_ERROR, &root->fs_info->fs_state))
8481                 return -EROFS;
8482
8483         ret = __start_delalloc_inodes(root, delay_iput);
8484         /*
8485          * the filemap_flush will queue IO into the worker threads, but
8486          * we have to make sure the IO is actually started and that
8487          * ordered extents get created before we return
8488          */
8489         atomic_inc(&root->fs_info->async_submit_draining);
8490         while (atomic_read(&root->fs_info->nr_async_submits) ||
8491               atomic_read(&root->fs_info->async_delalloc_pages)) {
8492                 wait_event(root->fs_info->async_submit_wait,
8493                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8494                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8495         }
8496         atomic_dec(&root->fs_info->async_submit_draining);
8497         return ret;
8498 }
8499
8500 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8501 {
8502         struct btrfs_root *root;
8503         struct list_head splice;
8504         int ret;
8505
8506         if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state))
8507                 return -EROFS;
8508
8509         INIT_LIST_HEAD(&splice);
8510
8511         spin_lock(&fs_info->delalloc_root_lock);
8512         list_splice_init(&fs_info->delalloc_roots, &splice);
8513         while (!list_empty(&splice)) {
8514                 root = list_first_entry(&splice, struct btrfs_root,
8515                                         delalloc_root);
8516                 root = btrfs_grab_fs_root(root);
8517                 BUG_ON(!root);
8518                 list_move_tail(&root->delalloc_root,
8519                                &fs_info->delalloc_roots);
8520                 spin_unlock(&fs_info->delalloc_root_lock);
8521
8522                 ret = __start_delalloc_inodes(root, delay_iput);
8523                 btrfs_put_fs_root(root);
8524                 if (ret)
8525                         goto out;
8526
8527                 spin_lock(&fs_info->delalloc_root_lock);
8528         }
8529         spin_unlock(&fs_info->delalloc_root_lock);
8530
8531         atomic_inc(&fs_info->async_submit_draining);
8532         while (atomic_read(&fs_info->nr_async_submits) ||
8533               atomic_read(&fs_info->async_delalloc_pages)) {
8534                 wait_event(fs_info->async_submit_wait,
8535                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8536                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8537         }
8538         atomic_dec(&fs_info->async_submit_draining);
8539         return 0;
8540 out:
8541         if (!list_empty_careful(&splice)) {
8542                 spin_lock(&fs_info->delalloc_root_lock);
8543                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8544                 spin_unlock(&fs_info->delalloc_root_lock);
8545         }
8546         return ret;
8547 }
8548
8549 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8550                          const char *symname)
8551 {
8552         struct btrfs_trans_handle *trans;
8553         struct btrfs_root *root = BTRFS_I(dir)->root;
8554         struct btrfs_path *path;
8555         struct btrfs_key key;
8556         struct inode *inode = NULL;
8557         int err;
8558         int drop_inode = 0;
8559         u64 objectid;
8560         u64 index = 0;
8561         int name_len;
8562         int datasize;
8563         unsigned long ptr;
8564         struct btrfs_file_extent_item *ei;
8565         struct extent_buffer *leaf;
8566
8567         name_len = strlen(symname);
8568         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8569                 return -ENAMETOOLONG;
8570
8571         /*
8572          * 2 items for inode item and ref
8573          * 2 items for dir items
8574          * 1 item for xattr if selinux is on
8575          */
8576         trans = btrfs_start_transaction(root, 5);
8577         if (IS_ERR(trans))
8578                 return PTR_ERR(trans);
8579
8580         err = btrfs_find_free_ino(root, &objectid);
8581         if (err)
8582                 goto out_unlock;
8583
8584         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8585                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8586                                 S_IFLNK|S_IRWXUGO, &index);
8587         if (IS_ERR(inode)) {
8588                 err = PTR_ERR(inode);
8589                 goto out_unlock;
8590         }
8591
8592         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8593         if (err) {
8594                 drop_inode = 1;
8595                 goto out_unlock;
8596         }
8597
8598         /*
8599         * If the active LSM wants to access the inode during
8600         * d_instantiate it needs these. Smack checks to see
8601         * if the filesystem supports xattrs by looking at the
8602         * ops vector.
8603         */
8604         inode->i_fop = &btrfs_file_operations;
8605         inode->i_op = &btrfs_file_inode_operations;
8606
8607         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8608         if (err)
8609                 drop_inode = 1;
8610         else {
8611                 inode->i_mapping->a_ops = &btrfs_aops;
8612                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8613                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8614         }
8615         if (drop_inode)
8616                 goto out_unlock;
8617
8618         path = btrfs_alloc_path();
8619         if (!path) {
8620                 err = -ENOMEM;
8621                 drop_inode = 1;
8622                 goto out_unlock;
8623         }
8624         key.objectid = btrfs_ino(inode);
8625         key.offset = 0;
8626         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8627         datasize = btrfs_file_extent_calc_inline_size(name_len);
8628         err = btrfs_insert_empty_item(trans, root, path, &key,
8629                                       datasize);
8630         if (err) {
8631                 drop_inode = 1;
8632                 btrfs_free_path(path);
8633                 goto out_unlock;
8634         }
8635         leaf = path->nodes[0];
8636         ei = btrfs_item_ptr(leaf, path->slots[0],
8637                             struct btrfs_file_extent_item);
8638         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8639         btrfs_set_file_extent_type(leaf, ei,
8640                                    BTRFS_FILE_EXTENT_INLINE);
8641         btrfs_set_file_extent_encryption(leaf, ei, 0);
8642         btrfs_set_file_extent_compression(leaf, ei, 0);
8643         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8644         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8645
8646         ptr = btrfs_file_extent_inline_start(ei);
8647         write_extent_buffer(leaf, symname, ptr, name_len);
8648         btrfs_mark_buffer_dirty(leaf);
8649         btrfs_free_path(path);
8650
8651         inode->i_op = &btrfs_symlink_inode_operations;
8652         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8653         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8654         inode_set_bytes(inode, name_len);
8655         btrfs_i_size_write(inode, name_len);
8656         err = btrfs_update_inode(trans, root, inode);
8657         if (err)
8658                 drop_inode = 1;
8659
8660 out_unlock:
8661         if (!err)
8662                 d_instantiate(dentry, inode);
8663         btrfs_end_transaction(trans, root);
8664         if (drop_inode) {
8665                 inode_dec_link_count(inode);
8666                 iput(inode);
8667         }
8668         btrfs_btree_balance_dirty(root);
8669         return err;
8670 }
8671
8672 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8673                                        u64 start, u64 num_bytes, u64 min_size,
8674                                        loff_t actual_len, u64 *alloc_hint,
8675                                        struct btrfs_trans_handle *trans)
8676 {
8677         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8678         struct extent_map *em;
8679         struct btrfs_root *root = BTRFS_I(inode)->root;
8680         struct btrfs_key ins;
8681         u64 cur_offset = start;
8682         u64 i_size;
8683         u64 cur_bytes;
8684         int ret = 0;
8685         bool own_trans = true;
8686
8687         if (trans)
8688                 own_trans = false;
8689         while (num_bytes > 0) {
8690                 if (own_trans) {
8691                         trans = btrfs_start_transaction(root, 3);
8692                         if (IS_ERR(trans)) {
8693                                 ret = PTR_ERR(trans);
8694                                 break;
8695                         }
8696                 }
8697
8698                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8699                 cur_bytes = max(cur_bytes, min_size);
8700                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8701                                            *alloc_hint, &ins, 1);
8702                 if (ret) {
8703                         if (own_trans)
8704                                 btrfs_end_transaction(trans, root);
8705                         break;
8706                 }
8707
8708                 ret = insert_reserved_file_extent(trans, inode,
8709                                                   cur_offset, ins.objectid,
8710                                                   ins.offset, ins.offset,
8711                                                   ins.offset, 0, 0, 0,
8712                                                   BTRFS_FILE_EXTENT_PREALLOC);
8713                 if (ret) {
8714                         btrfs_free_reserved_extent(root, ins.objectid,
8715                                                    ins.offset);
8716                         btrfs_abort_transaction(trans, root, ret);
8717                         if (own_trans)
8718                                 btrfs_end_transaction(trans, root);
8719                         break;
8720                 }
8721                 btrfs_drop_extent_cache(inode, cur_offset,
8722                                         cur_offset + ins.offset -1, 0);
8723
8724                 em = alloc_extent_map();
8725                 if (!em) {
8726                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8727                                 &BTRFS_I(inode)->runtime_flags);
8728                         goto next;
8729                 }
8730
8731                 em->start = cur_offset;
8732                 em->orig_start = cur_offset;
8733                 em->len = ins.offset;
8734                 em->block_start = ins.objectid;
8735                 em->block_len = ins.offset;
8736                 em->orig_block_len = ins.offset;
8737                 em->ram_bytes = ins.offset;
8738                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8739                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8740                 em->generation = trans->transid;
8741
8742                 while (1) {
8743                         write_lock(&em_tree->lock);
8744                         ret = add_extent_mapping(em_tree, em, 1);
8745                         write_unlock(&em_tree->lock);
8746                         if (ret != -EEXIST)
8747                                 break;
8748                         btrfs_drop_extent_cache(inode, cur_offset,
8749                                                 cur_offset + ins.offset - 1,
8750                                                 0);
8751                 }
8752                 free_extent_map(em);
8753 next:
8754                 num_bytes -= ins.offset;
8755                 cur_offset += ins.offset;
8756                 *alloc_hint = ins.objectid + ins.offset;
8757
8758                 inode_inc_iversion(inode);
8759                 inode->i_ctime = CURRENT_TIME;
8760                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8761                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8762                     (actual_len > inode->i_size) &&
8763                     (cur_offset > inode->i_size)) {
8764                         if (cur_offset > actual_len)
8765                                 i_size = actual_len;
8766                         else
8767                                 i_size = cur_offset;
8768                         i_size_write(inode, i_size);
8769                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8770                 }
8771
8772                 ret = btrfs_update_inode(trans, root, inode);
8773
8774                 if (ret) {
8775                         btrfs_abort_transaction(trans, root, ret);
8776                         if (own_trans)
8777                                 btrfs_end_transaction(trans, root);
8778                         break;
8779                 }
8780
8781                 if (own_trans)
8782                         btrfs_end_transaction(trans, root);
8783         }
8784         return ret;
8785 }
8786
8787 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8788                               u64 start, u64 num_bytes, u64 min_size,
8789                               loff_t actual_len, u64 *alloc_hint)
8790 {
8791         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8792                                            min_size, actual_len, alloc_hint,
8793                                            NULL);
8794 }
8795
8796 int btrfs_prealloc_file_range_trans(struct inode *inode,
8797                                     struct btrfs_trans_handle *trans, int mode,
8798                                     u64 start, u64 num_bytes, u64 min_size,
8799                                     loff_t actual_len, u64 *alloc_hint)
8800 {
8801         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8802                                            min_size, actual_len, alloc_hint, trans);
8803 }
8804
8805 static int btrfs_set_page_dirty(struct page *page)
8806 {
8807         return __set_page_dirty_nobuffers(page);
8808 }
8809
8810 static int btrfs_permission(struct inode *inode, int mask)
8811 {
8812         struct btrfs_root *root = BTRFS_I(inode)->root;
8813         umode_t mode = inode->i_mode;
8814
8815         if (mask & MAY_WRITE &&
8816             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8817                 if (btrfs_root_readonly(root))
8818                         return -EROFS;
8819                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8820                         return -EACCES;
8821         }
8822         return generic_permission(inode, mask);
8823 }
8824
8825 static const struct inode_operations btrfs_dir_inode_operations = {
8826         .getattr        = btrfs_getattr,
8827         .lookup         = btrfs_lookup,
8828         .create         = btrfs_create,
8829         .unlink         = btrfs_unlink,
8830         .link           = btrfs_link,
8831         .mkdir          = btrfs_mkdir,
8832         .rmdir          = btrfs_rmdir,
8833         .rename         = btrfs_rename,
8834         .symlink        = btrfs_symlink,
8835         .setattr        = btrfs_setattr,
8836         .mknod          = btrfs_mknod,
8837         .setxattr       = btrfs_setxattr,
8838         .getxattr       = btrfs_getxattr,
8839         .listxattr      = btrfs_listxattr,
8840         .removexattr    = btrfs_removexattr,
8841         .permission     = btrfs_permission,
8842         .get_acl        = btrfs_get_acl,
8843         .update_time    = btrfs_update_time,
8844 };
8845 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8846         .lookup         = btrfs_lookup,
8847         .permission     = btrfs_permission,
8848         .get_acl        = btrfs_get_acl,
8849         .update_time    = btrfs_update_time,
8850 };
8851
8852 static const struct file_operations btrfs_dir_file_operations = {
8853         .llseek         = generic_file_llseek,
8854         .read           = generic_read_dir,
8855         .iterate        = btrfs_real_readdir,
8856         .unlocked_ioctl = btrfs_ioctl,
8857 #ifdef CONFIG_COMPAT
8858         .compat_ioctl   = btrfs_ioctl,
8859 #endif
8860         .release        = btrfs_release_file,
8861         .fsync          = btrfs_sync_file,
8862 };
8863
8864 static struct extent_io_ops btrfs_extent_io_ops = {
8865         .fill_delalloc = run_delalloc_range,
8866         .submit_bio_hook = btrfs_submit_bio_hook,
8867         .merge_bio_hook = btrfs_merge_bio_hook,
8868         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8869         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8870         .writepage_start_hook = btrfs_writepage_start_hook,
8871         .set_bit_hook = btrfs_set_bit_hook,
8872         .clear_bit_hook = btrfs_clear_bit_hook,
8873         .merge_extent_hook = btrfs_merge_extent_hook,
8874         .split_extent_hook = btrfs_split_extent_hook,
8875 };
8876
8877 /*
8878  * btrfs doesn't support the bmap operation because swapfiles
8879  * use bmap to make a mapping of extents in the file.  They assume
8880  * these extents won't change over the life of the file and they
8881  * use the bmap result to do IO directly to the drive.
8882  *
8883  * the btrfs bmap call would return logical addresses that aren't
8884  * suitable for IO and they also will change frequently as COW
8885  * operations happen.  So, swapfile + btrfs == corruption.
8886  *
8887  * For now we're avoiding this by dropping bmap.
8888  */
8889 static const struct address_space_operations btrfs_aops = {
8890         .readpage       = btrfs_readpage,
8891         .writepage      = btrfs_writepage,
8892         .writepages     = btrfs_writepages,
8893         .readpages      = btrfs_readpages,
8894         .direct_IO      = btrfs_direct_IO,
8895         .invalidatepage = btrfs_invalidatepage,
8896         .releasepage    = btrfs_releasepage,
8897         .set_page_dirty = btrfs_set_page_dirty,
8898         .error_remove_page = generic_error_remove_page,
8899 };
8900
8901 static const struct address_space_operations btrfs_symlink_aops = {
8902         .readpage       = btrfs_readpage,
8903         .writepage      = btrfs_writepage,
8904         .invalidatepage = btrfs_invalidatepage,
8905         .releasepage    = btrfs_releasepage,
8906 };
8907
8908 static const struct inode_operations btrfs_file_inode_operations = {
8909         .getattr        = btrfs_getattr,
8910         .setattr        = btrfs_setattr,
8911         .setxattr       = btrfs_setxattr,
8912         .getxattr       = btrfs_getxattr,
8913         .listxattr      = btrfs_listxattr,
8914         .removexattr    = btrfs_removexattr,
8915         .permission     = btrfs_permission,
8916         .fiemap         = btrfs_fiemap,
8917         .get_acl        = btrfs_get_acl,
8918         .update_time    = btrfs_update_time,
8919 };
8920 static const struct inode_operations btrfs_special_inode_operations = {
8921         .getattr        = btrfs_getattr,
8922         .setattr        = btrfs_setattr,
8923         .permission     = btrfs_permission,
8924         .setxattr       = btrfs_setxattr,
8925         .getxattr       = btrfs_getxattr,
8926         .listxattr      = btrfs_listxattr,
8927         .removexattr    = btrfs_removexattr,
8928         .get_acl        = btrfs_get_acl,
8929         .update_time    = btrfs_update_time,
8930 };
8931 static const struct inode_operations btrfs_symlink_inode_operations = {
8932         .readlink       = generic_readlink,
8933         .follow_link    = page_follow_link_light,
8934         .put_link       = page_put_link,
8935         .getattr        = btrfs_getattr,
8936         .setattr        = btrfs_setattr,
8937         .permission     = btrfs_permission,
8938         .setxattr       = btrfs_setxattr,
8939         .getxattr       = btrfs_getxattr,
8940         .listxattr      = btrfs_listxattr,
8941         .removexattr    = btrfs_removexattr,
8942         .get_acl        = btrfs_get_acl,
8943         .update_time    = btrfs_update_time,
8944 };
8945
8946 const struct dentry_operations btrfs_dentry_operations = {
8947         .d_delete       = btrfs_dentry_delete,
8948         .d_release      = btrfs_dentry_release,
8949 };