fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 static inline bool is_free_space_inode(struct btrfs_root *root,
 754                                        struct inode *inode)
 755 {
 756         if (root == root->fs_info->tree_root ||
 757             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 758                 return true;
 759         return false;
 760 }
 761
 762 /*
 763  * when extent_io.c finds a delayed allocation range in the file,
 764  * the call backs end up in this code.  The basic idea is to
 765  * allocate extents on disk for the range, and create ordered data structs
 766  * in ram to track those extents.
 767  *
 768  * locked_page is the page that writepage had locked already.  We use
 769  * it to make sure we don't do extra locks or unlocks.
 770  *
 771  * *page_started is set to one if we unlock locked_page and do everything
 772  * required to start IO on it.  It may be clean and already done with
 773  * IO when we return.
 774  */
 775 static noinline int cow_file_range(struct inode *inode,
 776                                    struct page *locked_page,
 777                                    u64 start, u64 end, int *page_started,
 778                                    unsigned long *nr_written,
 779                                    int unlock)
 780 {
 781         struct btrfs_root *root = BTRFS_I(inode)->root;
 782         struct btrfs_trans_handle *trans;
 783         u64 alloc_hint = 0;
 784         u64 num_bytes;
 785         unsigned long ram_size;
 786         u64 disk_num_bytes;
 787         u64 cur_alloc_size;
 788         u64 blocksize = root->sectorsize;
 789         struct btrfs_key ins;
 790         struct extent_map *em;
 791         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 792         int ret = 0;
 793
 794         BUG_ON(is_free_space_inode(root, inode));
 795         trans = btrfs_join_transaction(root);
 796         BUG_ON(IS_ERR(trans));
 797         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 798
 799         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 800         num_bytes = max(blocksize,  num_bytes);
 801         disk_num_bytes = num_bytes;
 802         ret = 0;
 803
 804         /* if this is a small write inside eof, kick off defrag */
 805         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 806                 btrfs_add_inode_defrag(trans, inode);
 807
 808         if (start == 0) {
 809                 /* lets try to make an inline extent */
 810                 ret = cow_file_range_inline(trans, root, inode,
 811                                             start, end, 0, 0, NULL);
 812                 if (ret == 0) {
 813                         extent_clear_unlock_delalloc(inode,
 814                                      &BTRFS_I(inode)->io_tree,
 815                                      start, end, NULL,
 816                                      EXTENT_CLEAR_UNLOCK_PAGE |
 817                                      EXTENT_CLEAR_UNLOCK |
 818                                      EXTENT_CLEAR_DELALLOC |
 819                                      EXTENT_CLEAR_DIRTY |
 820                                      EXTENT_SET_WRITEBACK |
 821                                      EXTENT_END_WRITEBACK);
 822
 823                         *nr_written = *nr_written +
 824                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 825                         *page_started = 1;
 826                         ret = 0;
 827                         goto out;
 828                 }
 829         }
 830
 831         BUG_ON(disk_num_bytes >
 832                btrfs_super_total_bytes(&root->fs_info->super_copy));
 833
 834         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 835         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 836
 837         while (disk_num_bytes > 0) {
 838                 unsigned long op;
 839
 840                 cur_alloc_size = disk_num_bytes;
 841                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 842                                            root->sectorsize, 0, alloc_hint,
 843                                            (u64)-1, &ins, 1);
 844                 BUG_ON(ret);
 845
 846                 em = alloc_extent_map();
 847                 BUG_ON(!em);
 848                 em->start = start;
 849                 em->orig_start = em->start;
 850                 ram_size = ins.offset;
 851                 em->len = ins.offset;
 852
 853                 em->block_start = ins.objectid;
 854                 em->block_len = ins.offset;
 855                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 856                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 857
 858                 while (1) {
 859                         write_lock(&em_tree->lock);
 860                         ret = add_extent_mapping(em_tree, em);
 861                         write_unlock(&em_tree->lock);
 862                         if (ret != -EEXIST) {
 863                                 free_extent_map(em);
 864                                 break;
 865                         }
 866                         btrfs_drop_extent_cache(inode, start,
 867                                                 start + ram_size - 1, 0);
 868                 }
 869
 870                 cur_alloc_size = ins.offset;
 871                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 872                                                ram_size, cur_alloc_size, 0);
 873                 BUG_ON(ret);
 874
 875                 if (root->root_key.objectid ==
 876                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 877                         ret = btrfs_reloc_clone_csums(inode, start,
 878                                                       cur_alloc_size);
 879                         BUG_ON(ret);
 880                 }
 881
 882                 if (disk_num_bytes < cur_alloc_size)
 883                         break;
 884
 885                 /* we're not doing compressed IO, don't unlock the first
 886                  * page (which the caller expects to stay locked), don't
 887                  * clear any dirty bits and don't set any writeback bits
 888                  *
 889                  * Do set the Private2 bit so we know this page was properly
 890                  * setup for writepage
 891                  */
 892                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 893                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 894                         EXTENT_SET_PRIVATE2;
 895
 896                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 897                                              start, start + ram_size - 1,
 898                                              locked_page, op);
 899                 disk_num_bytes -= cur_alloc_size;
 900                 num_bytes -= cur_alloc_size;
 901                 alloc_hint = ins.objectid + ins.offset;
 902                 start += cur_alloc_size;
 903         }
 904 out:
 905         ret = 0;
 906         btrfs_end_transaction(trans, root);
 907
 908         return ret;
 909 }
 910
 911 /*
 912  * work queue call back to started compression on a file and pages
 913  */
 914 static noinline void async_cow_start(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         int num_added = 0;
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         compress_file_range(async_cow->inode, async_cow->locked_page,
 921                             async_cow->start, async_cow->end, async_cow,
 922                             &num_added);
 923         if (num_added == 0)
 924                 async_cow->inode = NULL;
 925 }
 926
 927 /*
 928  * work queue call back to submit previously compressed pages
 929  */
 930 static noinline void async_cow_submit(struct btrfs_work *work)
 931 {
 932         struct async_cow *async_cow;
 933         struct btrfs_root *root;
 934         unsigned long nr_pages;
 935
 936         async_cow = container_of(work, struct async_cow, work);
 937
 938         root = async_cow->root;
 939         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 940                 PAGE_CACHE_SHIFT;
 941
 942         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 943
 944         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 945             5 * 1042 * 1024 &&
 946             waitqueue_active(&root->fs_info->async_submit_wait))
 947                 wake_up(&root->fs_info->async_submit_wait);
 948
 949         if (async_cow->inode)
 950                 submit_compressed_extents(async_cow->inode, async_cow);
 951 }
 952
 953 static noinline void async_cow_free(struct btrfs_work *work)
 954 {
 955         struct async_cow *async_cow;
 956         async_cow = container_of(work, struct async_cow, work);
 957         kfree(async_cow);
 958 }
 959
 960 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 961                                 u64 start, u64 end, int *page_started,
 962                                 unsigned long *nr_written)
 963 {
 964         struct async_cow *async_cow;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         unsigned long nr_pages;
 967         u64 cur_end;
 968         int limit = 10 * 1024 * 1042;
 969
 970         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 971                          1, 0, NULL, GFP_NOFS);
 972         while (start < end) {
 973                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 974                 BUG_ON(!async_cow);
 975                 async_cow->inode = inode;
 976                 async_cow->root = root;
 977                 async_cow->locked_page = locked_page;
 978                 async_cow->start = start;
 979
 980                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 981                         cur_end = end;
 982                 else
 983                         cur_end = min(end, start + 512 * 1024 - 1);
 984
 985                 async_cow->end = cur_end;
 986                 INIT_LIST_HEAD(&async_cow->extents);
 987
 988                 async_cow->work.func = async_cow_start;
 989                 async_cow->work.ordered_func = async_cow_submit;
 990                 async_cow->work.ordered_free = async_cow_free;
 991                 async_cow->work.flags = 0;
 992
 993                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 994                         PAGE_CACHE_SHIFT;
 995                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 996
 997                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 998                                    &async_cow->work);
 999
1000                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1001                         wait_event(root->fs_info->async_submit_wait,
1002                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1003                             limit));
1004                 }
1005
1006                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1007                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1008                         wait_event(root->fs_info->async_submit_wait,
1009                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010                            0));
1011                 }
1012
1013                 *nr_written += nr_pages;
1014                 start = cur_end + 1;
1015         }
1016         *page_started = 1;
1017         return 0;
1018 }
1019
1020 static noinline int csum_exist_in_range(struct btrfs_root *root,
1021                                         u64 bytenr, u64 num_bytes)
1022 {
1023         int ret;
1024         struct btrfs_ordered_sum *sums;
1025         LIST_HEAD(list);
1026
1027         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1028                                        bytenr + num_bytes - 1, &list, 0);
1029         if (ret == 0 && list_empty(&list))
1030                 return 0;
1031
1032         while (!list_empty(&list)) {
1033                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1034                 list_del(&sums->list);
1035                 kfree(sums);
1036         }
1037         return 1;
1038 }
1039
1040 /*
1041  * when nowcow writeback call back.  This checks for snapshots or COW copies
1042  * of the extents that exist in the file, and COWs the file as required.
1043  *
1044  * If no cow copies or snapshots exist, we write directly to the existing
1045  * blocks on disk
1046  */
1047 static noinline int run_delalloc_nocow(struct inode *inode,
1048                                        struct page *locked_page,
1049                               u64 start, u64 end, int *page_started, int force,
1050                               unsigned long *nr_written)
1051 {
1052         struct btrfs_root *root = BTRFS_I(inode)->root;
1053         struct btrfs_trans_handle *trans;
1054         struct extent_buffer *leaf;
1055         struct btrfs_path *path;
1056         struct btrfs_file_extent_item *fi;
1057         struct btrfs_key found_key;
1058         u64 cow_start;
1059         u64 cur_offset;
1060         u64 extent_end;
1061         u64 extent_offset;
1062         u64 disk_bytenr;
1063         u64 num_bytes;
1064         int extent_type;
1065         int ret;
1066         int type;
1067         int nocow;
1068         int check_prev = 1;
1069         bool nolock;
1070         u64 ino = btrfs_ino(inode);
1071
1072         path = btrfs_alloc_path();
1073         BUG_ON(!path);
1074
1075         nolock = is_free_space_inode(root, inode);
1076
1077         if (nolock)
1078                 trans = btrfs_join_transaction_nolock(root);
1079         else
1080                 trans = btrfs_join_transaction(root);
1081
1082         BUG_ON(IS_ERR(trans));
1083         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1084
1085         cow_start = (u64)-1;
1086         cur_offset = start;
1087         while (1) {
1088                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089                                                cur_offset, 0);
1090                 BUG_ON(ret < 0);
1091                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1092                         leaf = path->nodes[0];
1093                         btrfs_item_key_to_cpu(leaf, &found_key,
1094                                               path->slots[0] - 1);
1095                         if (found_key.objectid == ino &&
1096                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1097                                 path->slots[0]--;
1098                 }
1099                 check_prev = 0;
1100 next_slot:
1101                 leaf = path->nodes[0];
1102                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1103                         ret = btrfs_next_leaf(root, path);
1104                         if (ret < 0)
1105                                 BUG_ON(1);
1106                         if (ret > 0)
1107                                 break;
1108                         leaf = path->nodes[0];
1109                 }
1110
1111                 nocow = 0;
1112                 disk_bytenr = 0;
1113                 num_bytes = 0;
1114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1115
1116                 if (found_key.objectid > ino ||
1117                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1118                     found_key.offset > end)
1119                         break;
1120
1121                 if (found_key.offset > cur_offset) {
1122                         extent_end = found_key.offset;
1123                         extent_type = 0;
1124                         goto out_check;
1125                 }
1126
1127                 fi = btrfs_item_ptr(leaf, path->slots[0],
1128                                     struct btrfs_file_extent_item);
1129                 extent_type = btrfs_file_extent_type(leaf, fi);
1130
1131                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1132                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1133                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1134                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1135                         extent_end = found_key.offset +
1136                                 btrfs_file_extent_num_bytes(leaf, fi);
1137                         if (extent_end <= start) {
1138                                 path->slots[0]++;
1139                                 goto next_slot;
1140                         }
1141                         if (disk_bytenr == 0)
1142                                 goto out_check;
1143                         if (btrfs_file_extent_compression(leaf, fi) ||
1144                             btrfs_file_extent_encryption(leaf, fi) ||
1145                             btrfs_file_extent_other_encoding(leaf, fi))
1146                                 goto out_check;
1147                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1148                                 goto out_check;
1149                         if (btrfs_extent_readonly(root, disk_bytenr))
1150                                 goto out_check;
1151                         if (btrfs_cross_ref_exist(trans, root, ino,
1152                                                   found_key.offset -
1153                                                   extent_offset, disk_bytenr))
1154                                 goto out_check;
1155                         disk_bytenr += extent_offset;
1156                         disk_bytenr += cur_offset - found_key.offset;
1157                         num_bytes = min(end + 1, extent_end) - cur_offset;
1158                         /*
1159                          * force cow if csum exists in the range.
1160                          * this ensure that csum for a given extent are
1161                          * either valid or do not exist.
1162                          */
1163                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164                                 goto out_check;
1165                         nocow = 1;
1166                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1167                         extent_end = found_key.offset +
1168                                 btrfs_file_extent_inline_len(leaf, fi);
1169                         extent_end = ALIGN(extent_end, root->sectorsize);
1170                 } else {
1171                         BUG_ON(1);
1172                 }
1173 out_check:
1174                 if (extent_end <= start) {
1175                         path->slots[0]++;
1176                         goto next_slot;
1177                 }
1178                 if (!nocow) {
1179                         if (cow_start == (u64)-1)
1180                                 cow_start = cur_offset;
1181                         cur_offset = extent_end;
1182                         if (cur_offset > end)
1183                                 break;
1184                         path->slots[0]++;
1185                         goto next_slot;
1186                 }
1187
1188                 btrfs_release_path(path);
1189                 if (cow_start != (u64)-1) {
1190                         ret = cow_file_range(inode, locked_page, cow_start,
1191                                         found_key.offset - 1, page_started,
1192                                         nr_written, 1);
1193                         BUG_ON(ret);
1194                         cow_start = (u64)-1;
1195                 }
1196
1197                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1198                         struct extent_map *em;
1199                         struct extent_map_tree *em_tree;
1200                         em_tree = &BTRFS_I(inode)->extent_tree;
1201                         em = alloc_extent_map();
1202                         BUG_ON(!em);
1203                         em->start = cur_offset;
1204                         em->orig_start = em->start;
1205                         em->len = num_bytes;
1206                         em->block_len = num_bytes;
1207                         em->block_start = disk_bytenr;
1208                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1209                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1210                         while (1) {
1211                                 write_lock(&em_tree->lock);
1212                                 ret = add_extent_mapping(em_tree, em);
1213                                 write_unlock(&em_tree->lock);
1214                                 if (ret != -EEXIST) {
1215                                         free_extent_map(em);
1216                                         break;
1217                                 }
1218                                 btrfs_drop_extent_cache(inode, em->start,
1219                                                 em->start + em->len - 1, 0);
1220                         }
1221                         type = BTRFS_ORDERED_PREALLOC;
1222                 } else {
1223                         type = BTRFS_ORDERED_NOCOW;
1224                 }
1225
1226                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1227                                                num_bytes, num_bytes, type);
1228                 BUG_ON(ret);
1229
1230                 if (root->root_key.objectid ==
1231                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1232                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233                                                       num_bytes);
1234                         BUG_ON(ret);
1235                 }
1236
1237                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1238                                 cur_offset, cur_offset + num_bytes - 1,
1239                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1240                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1241                                 EXTENT_SET_PRIVATE2);
1242                 cur_offset = extent_end;
1243                 if (cur_offset > end)
1244                         break;
1245         }
1246         btrfs_release_path(path);
1247
1248         if (cur_offset <= end && cow_start == (u64)-1)
1249                 cow_start = cur_offset;
1250         if (cow_start != (u64)-1) {
1251                 ret = cow_file_range(inode, locked_page, cow_start, end,
1252                                      page_started, nr_written, 1);
1253                 BUG_ON(ret);
1254         }
1255
1256         if (nolock) {
1257                 ret = btrfs_end_transaction_nolock(trans, root);
1258                 BUG_ON(ret);
1259         } else {
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262         }
1263         btrfs_free_path(path);
1264         return 0;
1265 }
1266
1267 /*
1268  * extent_io.c call back to do delayed allocation processing
1269  */
1270 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1271                               u64 start, u64 end, int *page_started,
1272                               unsigned long *nr_written)
1273 {
1274         int ret;
1275         struct btrfs_root *root = BTRFS_I(inode)->root;
1276
1277         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1278                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1279                                          page_started, 1, nr_written);
1280         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1281                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1282                                          page_started, 0, nr_written);
1283         else if (!btrfs_test_opt(root, COMPRESS) &&
1284                  !(BTRFS_I(inode)->force_compress) &&
1285                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1286                 ret = cow_file_range(inode, locked_page, start, end,
1287                                       page_started, nr_written, 1);
1288         else
1289                 ret = cow_file_range_async(inode, locked_page, start, end,
1290                                            page_started, nr_written);
1291         return ret;
1292 }
1293
1294 static int btrfs_split_extent_hook(struct inode *inode,
1295                                    struct extent_state *orig, u64 split)
1296 {
1297         /* not delalloc, ignore it */
1298         if (!(orig->state & EXTENT_DELALLOC))
1299                 return 0;
1300
1301         spin_lock(&BTRFS_I(inode)->lock);
1302         BTRFS_I(inode)->outstanding_extents++;
1303         spin_unlock(&BTRFS_I(inode)->lock);
1304         return 0;
1305 }
1306
1307 /*
1308  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1309  * extents so we can keep track of new extents that are just merged onto old
1310  * extents, such as when we are doing sequential writes, so we can properly
1311  * account for the metadata space we'll need.
1312  */
1313 static int btrfs_merge_extent_hook(struct inode *inode,
1314                                    struct extent_state *new,
1315                                    struct extent_state *other)
1316 {
1317         /* not delalloc, ignore it */
1318         if (!(other->state & EXTENT_DELALLOC))
1319                 return 0;
1320
1321         spin_lock(&BTRFS_I(inode)->lock);
1322         BTRFS_I(inode)->outstanding_extents--;
1323         spin_unlock(&BTRFS_I(inode)->lock);
1324         return 0;
1325 }
1326
1327 /*
1328  * extent_io.c set_bit_hook, used to track delayed allocation
1329  * bytes in this file, and to maintain the list of inodes that
1330  * have pending delalloc work to be done.
1331  */
1332 static int btrfs_set_bit_hook(struct inode *inode,
1333                               struct extent_state *state, int *bits)
1334 {
1335
1336         /*
1337          * set_bit and clear bit hooks normally require _irqsave/restore
1338          * but in this case, we are only testing for the DELALLOC
1339          * bit, which is only set or cleared with irqs on
1340          */
1341         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1342                 struct btrfs_root *root = BTRFS_I(inode)->root;
1343                 u64 len = state->end + 1 - state->start;
1344                 bool do_list = !is_free_space_inode(root, inode);
1345
1346                 if (*bits & EXTENT_FIRST_DELALLOC) {
1347                         *bits &= ~EXTENT_FIRST_DELALLOC;
1348                 } else {
1349                         spin_lock(&BTRFS_I(inode)->lock);
1350                         BTRFS_I(inode)->outstanding_extents++;
1351                         spin_unlock(&BTRFS_I(inode)->lock);
1352                 }
1353
1354                 spin_lock(&root->fs_info->delalloc_lock);
1355                 BTRFS_I(inode)->delalloc_bytes += len;
1356                 root->fs_info->delalloc_bytes += len;
1357                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1358                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1359                                       &root->fs_info->delalloc_inodes);
1360                 }
1361                 spin_unlock(&root->fs_info->delalloc_lock);
1362         }
1363         return 0;
1364 }
1365
1366 /*
1367  * extent_io.c clear_bit_hook, see set_bit_hook for why
1368  */
1369 static int btrfs_clear_bit_hook(struct inode *inode,
1370                                 struct extent_state *state, int *bits)
1371 {
1372         /*
1373          * set_bit and clear bit hooks normally require _irqsave/restore
1374          * but in this case, we are only testing for the DELALLOC
1375          * bit, which is only set or cleared with irqs on
1376          */
1377         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1378                 struct btrfs_root *root = BTRFS_I(inode)->root;
1379                 u64 len = state->end + 1 - state->start;
1380                 bool do_list = !is_free_space_inode(root, inode);
1381
1382                 if (*bits & EXTENT_FIRST_DELALLOC) {
1383                         *bits &= ~EXTENT_FIRST_DELALLOC;
1384                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1385                         spin_lock(&BTRFS_I(inode)->lock);
1386                         BTRFS_I(inode)->outstanding_extents--;
1387                         spin_unlock(&BTRFS_I(inode)->lock);
1388                 }
1389
1390                 if (*bits & EXTENT_DO_ACCOUNTING)
1391                         btrfs_delalloc_release_metadata(inode, len);
1392
1393                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1394                     && do_list)
1395                         btrfs_free_reserved_data_space(inode, len);
1396
1397                 spin_lock(&root->fs_info->delalloc_lock);
1398                 root->fs_info->delalloc_bytes -= len;
1399                 BTRFS_I(inode)->delalloc_bytes -= len;
1400
1401                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1402                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1403                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1404                 }
1405                 spin_unlock(&root->fs_info->delalloc_lock);
1406         }
1407         return 0;
1408 }
1409
1410 /*
1411  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1412  * we don't create bios that span stripes or chunks
1413  */
1414 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1415                          size_t size, struct bio *bio,
1416                          unsigned long bio_flags)
1417 {
1418         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1419         struct btrfs_mapping_tree *map_tree;
1420         u64 logical = (u64)bio->bi_sector << 9;
1421         u64 length = 0;
1422         u64 map_length;
1423         int ret;
1424
1425         if (bio_flags & EXTENT_BIO_COMPRESSED)
1426                 return 0;
1427
1428         length = bio->bi_size;
1429         map_tree = &root->fs_info->mapping_tree;
1430         map_length = length;
1431         ret = btrfs_map_block(map_tree, READ, logical,
1432                               &map_length, NULL, 0);
1433
1434         if (map_length < length + size)
1435                 return 1;
1436         return ret;
1437 }
1438
1439 /*
1440  * in order to insert checksums into the metadata in large chunks,
1441  * we wait until bio submission time.   All the pages in the bio are
1442  * checksummed and sums are attached onto the ordered extent record.
1443  *
1444  * At IO completion time the cums attached on the ordered extent record
1445  * are inserted into the btree
1446  */
1447 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1448                                     struct bio *bio, int mirror_num,
1449                                     unsigned long bio_flags,
1450                                     u64 bio_offset)
1451 {
1452         struct btrfs_root *root = BTRFS_I(inode)->root;
1453         int ret = 0;
1454
1455         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1456         BUG_ON(ret);
1457         return 0;
1458 }
1459
1460 /*
1461  * in order to insert checksums into the metadata in large chunks,
1462  * we wait until bio submission time.   All the pages in the bio are
1463  * checksummed and sums are attached onto the ordered extent record.
1464  *
1465  * At IO completion time the cums attached on the ordered extent record
1466  * are inserted into the btree
1467  */
1468 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1469                           int mirror_num, unsigned long bio_flags,
1470                           u64 bio_offset)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1474 }
1475
1476 /*
1477  * extent_io.c submission hook. This does the right thing for csum calculation
1478  * on write, or reading the csums from the tree before a read
1479  */
1480 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1481                           int mirror_num, unsigned long bio_flags,
1482                           u64 bio_offset)
1483 {
1484         struct btrfs_root *root = BTRFS_I(inode)->root;
1485         int ret = 0;
1486         int skip_sum;
1487
1488         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1489
1490         if (is_free_space_inode(root, inode))
1491                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1492         else
1493                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1494         BUG_ON(ret);
1495
1496         if (!(rw & REQ_WRITE)) {
1497                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1498                         return btrfs_submit_compressed_read(inode, bio,
1499                                                     mirror_num, bio_flags);
1500                 } else if (!skip_sum) {
1501                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1502                         if (ret)
1503                                 return ret;
1504                 }
1505                 goto mapit;
1506         } else if (!skip_sum) {
1507                 /* csum items have already been cloned */
1508                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1509                         goto mapit;
1510                 /* we're doing a write, do the async checksumming */
1511                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1512                                    inode, rw, bio, mirror_num,
1513                                    bio_flags, bio_offset,
1514                                    __btrfs_submit_bio_start,
1515                                    __btrfs_submit_bio_done);
1516         }
1517
1518 mapit:
1519         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1520 }
1521
1522 /*
1523  * given a list of ordered sums record them in the inode.  This happens
1524  * at IO completion time based on sums calculated at bio submission time.
1525  */
1526 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1527                              struct inode *inode, u64 file_offset,
1528                              struct list_head *list)
1529 {
1530         struct btrfs_ordered_sum *sum;
1531
1532         list_for_each_entry(sum, list, list) {
1533                 btrfs_csum_file_blocks(trans,
1534                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1535         }
1536         return 0;
1537 }
1538
1539 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1540                               struct extent_state **cached_state)
1541 {
1542         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1543                 WARN_ON(1);
1544         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1545                                    cached_state, GFP_NOFS);
1546 }
1547
1548 /* see btrfs_writepage_start_hook for details on why this is required */
1549 struct btrfs_writepage_fixup {
1550         struct page *page;
1551         struct btrfs_work work;
1552 };
1553
1554 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1555 {
1556         struct btrfs_writepage_fixup *fixup;
1557         struct btrfs_ordered_extent *ordered;
1558         struct extent_state *cached_state = NULL;
1559         struct page *page;
1560         struct inode *inode;
1561         u64 page_start;
1562         u64 page_end;
1563
1564         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1565         page = fixup->page;
1566 again:
1567         lock_page(page);
1568         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1569                 ClearPageChecked(page);
1570                 goto out_page;
1571         }
1572
1573         inode = page->mapping->host;
1574         page_start = page_offset(page);
1575         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1576
1577         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1578                          &cached_state, GFP_NOFS);
1579
1580         /* already ordered? We're done */
1581         if (PagePrivate2(page))
1582                 goto out;
1583
1584         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1585         if (ordered) {
1586                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1587                                      page_end, &cached_state, GFP_NOFS);
1588                 unlock_page(page);
1589                 btrfs_start_ordered_extent(inode, ordered, 1);
1590                 goto again;
1591         }
1592
1593         BUG();
1594         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1595         ClearPageChecked(page);
1596 out:
1597         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1598                              &cached_state, GFP_NOFS);
1599 out_page:
1600         unlock_page(page);
1601         page_cache_release(page);
1602         kfree(fixup);
1603 }
1604
1605 /*
1606  * There are a few paths in the higher layers of the kernel that directly
1607  * set the page dirty bit without asking the filesystem if it is a
1608  * good idea.  This causes problems because we want to make sure COW
1609  * properly happens and the data=ordered rules are followed.
1610  *
1611  * In our case any range that doesn't have the ORDERED bit set
1612  * hasn't been properly setup for IO.  We kick off an async process
1613  * to fix it up.  The async helper will wait for ordered extents, set
1614  * the delalloc bit and make it safe to write the page.
1615  */
1616 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1617 {
1618         struct inode *inode = page->mapping->host;
1619         struct btrfs_writepage_fixup *fixup;
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621
1622         /* this page is properly in the ordered list */
1623         if (TestClearPagePrivate2(page))
1624                 return 0;
1625
1626         if (PageChecked(page))
1627                 return -EAGAIN;
1628
1629         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1630         if (!fixup)
1631                 return -EAGAIN;
1632
1633         SetPageChecked(page);
1634         page_cache_get(page);
1635         fixup->work.func = btrfs_writepage_fixup_worker;
1636         fixup->page = page;
1637         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1638         return -EAGAIN;
1639 }
1640
1641 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1642                                        struct inode *inode, u64 file_pos,
1643                                        u64 disk_bytenr, u64 disk_num_bytes,
1644                                        u64 num_bytes, u64 ram_bytes,
1645                                        u8 compression, u8 encryption,
1646                                        u16 other_encoding, int extent_type)
1647 {
1648         struct btrfs_root *root = BTRFS_I(inode)->root;
1649         struct btrfs_file_extent_item *fi;
1650         struct btrfs_path *path;
1651         struct extent_buffer *leaf;
1652         struct btrfs_key ins;
1653         u64 hint;
1654         int ret;
1655
1656         path = btrfs_alloc_path();
1657         BUG_ON(!path);
1658
1659         path->leave_spinning = 1;
1660
1661         /*
1662          * we may be replacing one extent in the tree with another.
1663          * The new extent is pinned in the extent map, and we don't want
1664          * to drop it from the cache until it is completely in the btree.
1665          *
1666          * So, tell btrfs_drop_extents to leave this extent in the cache.
1667          * the caller is expected to unpin it and allow it to be merged
1668          * with the others.
1669          */
1670         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1671                                  &hint, 0);
1672         BUG_ON(ret);
1673
1674         ins.objectid = btrfs_ino(inode);
1675         ins.offset = file_pos;
1676         ins.type = BTRFS_EXTENT_DATA_KEY;
1677         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1678         BUG_ON(ret);
1679         leaf = path->nodes[0];
1680         fi = btrfs_item_ptr(leaf, path->slots[0],
1681                             struct btrfs_file_extent_item);
1682         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1683         btrfs_set_file_extent_type(leaf, fi, extent_type);
1684         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1685         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1686         btrfs_set_file_extent_offset(leaf, fi, 0);
1687         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1688         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1689         btrfs_set_file_extent_compression(leaf, fi, compression);
1690         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1691         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1692
1693         btrfs_unlock_up_safe(path, 1);
1694         btrfs_set_lock_blocking(leaf);
1695
1696         btrfs_mark_buffer_dirty(leaf);
1697
1698         inode_add_bytes(inode, num_bytes);
1699
1700         ins.objectid = disk_bytenr;
1701         ins.offset = disk_num_bytes;
1702         ins.type = BTRFS_EXTENT_ITEM_KEY;
1703         ret = btrfs_alloc_reserved_file_extent(trans, root,
1704                                         root->root_key.objectid,
1705                                         btrfs_ino(inode), file_pos, &ins);
1706         BUG_ON(ret);
1707         btrfs_free_path(path);
1708
1709         return 0;
1710 }
1711
1712 /*
1713  * helper function for btrfs_finish_ordered_io, this
1714  * just reads in some of the csum leaves to prime them into ram
1715  * before we start the transaction.  It limits the amount of btree
1716  * reads required while inside the transaction.
1717  */
1718 /* as ordered data IO finishes, this gets called so we can finish
1719  * an ordered extent if the range of bytes in the file it covers are
1720  * fully written.
1721  */
1722 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1723 {
1724         struct btrfs_root *root = BTRFS_I(inode)->root;
1725         struct btrfs_trans_handle *trans = NULL;
1726         struct btrfs_ordered_extent *ordered_extent = NULL;
1727         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1728         struct extent_state *cached_state = NULL;
1729         int compress_type = 0;
1730         int ret;
1731         bool nolock;
1732
1733         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1734                                              end - start + 1);
1735         if (!ret)
1736                 return 0;
1737         BUG_ON(!ordered_extent);
1738
1739         nolock = is_free_space_inode(root, inode);
1740
1741         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1742                 BUG_ON(!list_empty(&ordered_extent->list));
1743                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1744                 if (!ret) {
1745                         if (nolock)
1746                                 trans = btrfs_join_transaction_nolock(root);
1747                         else
1748                                 trans = btrfs_join_transaction(root);
1749                         BUG_ON(IS_ERR(trans));
1750                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1751                         ret = btrfs_update_inode(trans, root, inode);
1752                         BUG_ON(ret);
1753                 }
1754                 goto out;
1755         }
1756
1757         lock_extent_bits(io_tree, ordered_extent->file_offset,
1758                          ordered_extent->file_offset + ordered_extent->len - 1,
1759                          0, &cached_state, GFP_NOFS);
1760
1761         if (nolock)
1762                 trans = btrfs_join_transaction_nolock(root);
1763         else
1764                 trans = btrfs_join_transaction(root);
1765         BUG_ON(IS_ERR(trans));
1766         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1767
1768         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1769                 compress_type = ordered_extent->compress_type;
1770         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1771                 BUG_ON(compress_type);
1772                 ret = btrfs_mark_extent_written(trans, inode,
1773                                                 ordered_extent->file_offset,
1774                                                 ordered_extent->file_offset +
1775                                                 ordered_extent->len);
1776                 BUG_ON(ret);
1777         } else {
1778                 BUG_ON(root == root->fs_info->tree_root);
1779                 ret = insert_reserved_file_extent(trans, inode,
1780                                                 ordered_extent->file_offset,
1781                                                 ordered_extent->start,
1782                                                 ordered_extent->disk_len,
1783                                                 ordered_extent->len,
1784                                                 ordered_extent->len,
1785                                                 compress_type, 0, 0,
1786                                                 BTRFS_FILE_EXTENT_REG);
1787                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1788                                    ordered_extent->file_offset,
1789                                    ordered_extent->len);
1790                 BUG_ON(ret);
1791         }
1792         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1793                              ordered_extent->file_offset +
1794                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1795
1796         add_pending_csums(trans, inode, ordered_extent->file_offset,
1797                           &ordered_extent->list);
1798
1799         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1800         if (!ret) {
1801                 ret = btrfs_update_inode(trans, root, inode);
1802                 BUG_ON(ret);
1803         }
1804         ret = 0;
1805 out:
1806         if (nolock) {
1807                 if (trans)
1808                         btrfs_end_transaction_nolock(trans, root);
1809         } else {
1810                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1811                 if (trans)
1812                         btrfs_end_transaction(trans, root);
1813         }
1814
1815         /* once for us */
1816         btrfs_put_ordered_extent(ordered_extent);
1817         /* once for the tree */
1818         btrfs_put_ordered_extent(ordered_extent);
1819
1820         return 0;
1821 }
1822
1823 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1824                                 struct extent_state *state, int uptodate)
1825 {
1826         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1827
1828         ClearPagePrivate2(page);
1829         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1830 }
1831
1832 /*
1833  * When IO fails, either with EIO or csum verification fails, we
1834  * try other mirrors that might have a good copy of the data.  This
1835  * io_failure_record is used to record state as we go through all the
1836  * mirrors.  If another mirror has good data, the page is set up to date
1837  * and things continue.  If a good mirror can't be found, the original
1838  * bio end_io callback is called to indicate things have failed.
1839  */
1840 struct io_failure_record {
1841         struct page *page;
1842         u64 start;
1843         u64 len;
1844         u64 logical;
1845         unsigned long bio_flags;
1846         int last_mirror;
1847 };
1848
1849 static int btrfs_io_failed_hook(struct bio *failed_bio,
1850                          struct page *page, u64 start, u64 end,
1851                          struct extent_state *state)
1852 {
1853         struct io_failure_record *failrec = NULL;
1854         u64 private;
1855         struct extent_map *em;
1856         struct inode *inode = page->mapping->host;
1857         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1858         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1859         struct bio *bio;
1860         int num_copies;
1861         int ret;
1862         int rw;
1863         u64 logical;
1864
1865         ret = get_state_private(failure_tree, start, &private);
1866         if (ret) {
1867                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1868                 if (!failrec)
1869                         return -ENOMEM;
1870                 failrec->start = start;
1871                 failrec->len = end - start + 1;
1872                 failrec->last_mirror = 0;
1873                 failrec->bio_flags = 0;
1874
1875                 read_lock(&em_tree->lock);
1876                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1877                 if (em->start > start || em->start + em->len < start) {
1878                         free_extent_map(em);
1879                         em = NULL;
1880                 }
1881                 read_unlock(&em_tree->lock);
1882
1883                 if (IS_ERR_OR_NULL(em)) {
1884                         kfree(failrec);
1885                         return -EIO;
1886                 }
1887                 logical = start - em->start;
1888                 logical = em->block_start + logical;
1889                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1890                         logical = em->block_start;
1891                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1892                         extent_set_compress_type(&failrec->bio_flags,
1893                                                  em->compress_type);
1894                 }
1895                 failrec->logical = logical;
1896                 free_extent_map(em);
1897                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1898                                 EXTENT_DIRTY, GFP_NOFS);
1899                 set_state_private(failure_tree, start,
1900                                  (u64)(unsigned long)failrec);
1901         } else {
1902                 failrec = (struct io_failure_record *)(unsigned long)private;
1903         }
1904         num_copies = btrfs_num_copies(
1905                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1906                               failrec->logical, failrec->len);
1907         failrec->last_mirror++;
1908         if (!state) {
1909                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1910                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1911                                                     failrec->start,
1912                                                     EXTENT_LOCKED);
1913                 if (state && state->start != failrec->start)
1914                         state = NULL;
1915                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1916         }
1917         if (!state || failrec->last_mirror > num_copies) {
1918                 set_state_private(failure_tree, failrec->start, 0);
1919                 clear_extent_bits(failure_tree, failrec->start,
1920                                   failrec->start + failrec->len - 1,
1921                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1922                 kfree(failrec);
1923                 return -EIO;
1924         }
1925         bio = bio_alloc(GFP_NOFS, 1);
1926         bio->bi_private = state;
1927         bio->bi_end_io = failed_bio->bi_end_io;
1928         bio->bi_sector = failrec->logical >> 9;
1929         bio->bi_bdev = failed_bio->bi_bdev;
1930         bio->bi_size = 0;
1931
1932         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1933         if (failed_bio->bi_rw & REQ_WRITE)
1934                 rw = WRITE;
1935         else
1936                 rw = READ;
1937
1938         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1939                                                       failrec->last_mirror,
1940                                                       failrec->bio_flags, 0);
1941         return ret;
1942 }
1943
1944 /*
1945  * each time an IO finishes, we do a fast check in the IO failure tree
1946  * to see if we need to process or clean up an io_failure_record
1947  */
1948 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1949 {
1950         u64 private;
1951         u64 private_failure;
1952         struct io_failure_record *failure;
1953         int ret;
1954
1955         private = 0;
1956         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1957                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1958                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1959                                         start, &private_failure);
1960                 if (ret == 0) {
1961                         failure = (struct io_failure_record *)(unsigned long)
1962                                    private_failure;
1963                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1964                                           failure->start, 0);
1965                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1966                                           failure->start,
1967                                           failure->start + failure->len - 1,
1968                                           EXTENT_DIRTY | EXTENT_LOCKED,
1969                                           GFP_NOFS);
1970                         kfree(failure);
1971                 }
1972         }
1973         return 0;
1974 }
1975
1976 /*
1977  * when reads are done, we need to check csums to verify the data is correct
1978  * if there's a match, we allow the bio to finish.  If not, we go through
1979  * the io_failure_record routines to find good copies
1980  */
1981 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1982                                struct extent_state *state)
1983 {
1984         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1985         struct inode *inode = page->mapping->host;
1986         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1987         char *kaddr;
1988         u64 private = ~(u32)0;
1989         int ret;
1990         struct btrfs_root *root = BTRFS_I(inode)->root;
1991         u32 csum = ~(u32)0;
1992
1993         if (PageChecked(page)) {
1994                 ClearPageChecked(page);
1995                 goto good;
1996         }
1997
1998         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1999                 goto good;
2000
2001         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2002             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2003                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2004                                   GFP_NOFS);
2005                 return 0;
2006         }
2007
2008         if (state && state->start == start) {
2009                 private = state->private;
2010                 ret = 0;
2011         } else {
2012                 ret = get_state_private(io_tree, start, &private);
2013         }
2014         kaddr = kmap_atomic(page, KM_USER0);
2015         if (ret)
2016                 goto zeroit;
2017
2018         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2019         btrfs_csum_final(csum, (char *)&csum);
2020         if (csum != private)
2021                 goto zeroit;
2022
2023         kunmap_atomic(kaddr, KM_USER0);
2024 good:
2025         /* if the io failure tree for this inode is non-empty,
2026          * check to see if we've recovered from a failed IO
2027          */
2028         btrfs_clean_io_failures(inode, start);
2029         return 0;
2030
2031 zeroit:
2032         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2033                        "private %llu\n",
2034                        (unsigned long long)btrfs_ino(page->mapping->host),
2035                        (unsigned long long)start, csum,
2036                        (unsigned long long)private);
2037         memset(kaddr + offset, 1, end - start + 1);
2038         flush_dcache_page(page);
2039         kunmap_atomic(kaddr, KM_USER0);
2040         if (private == 0)
2041                 return 0;
2042         return -EIO;
2043 }
2044
2045 struct delayed_iput {
2046         struct list_head list;
2047         struct inode *inode;
2048 };
2049
2050 void btrfs_add_delayed_iput(struct inode *inode)
2051 {
2052         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2053         struct delayed_iput *delayed;
2054
2055         if (atomic_add_unless(&inode->i_count, -1, 1))
2056                 return;
2057
2058         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2059         delayed->inode = inode;
2060
2061         spin_lock(&fs_info->delayed_iput_lock);
2062         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2063         spin_unlock(&fs_info->delayed_iput_lock);
2064 }
2065
2066 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2067 {
2068         LIST_HEAD(list);
2069         struct btrfs_fs_info *fs_info = root->fs_info;
2070         struct delayed_iput *delayed;
2071         int empty;
2072
2073         spin_lock(&fs_info->delayed_iput_lock);
2074         empty = list_empty(&fs_info->delayed_iputs);
2075         spin_unlock(&fs_info->delayed_iput_lock);
2076         if (empty)
2077                 return;
2078
2079         down_read(&root->fs_info->cleanup_work_sem);
2080         spin_lock(&fs_info->delayed_iput_lock);
2081         list_splice_init(&fs_info->delayed_iputs, &list);
2082         spin_unlock(&fs_info->delayed_iput_lock);
2083
2084         while (!list_empty(&list)) {
2085                 delayed = list_entry(list.next, struct delayed_iput, list);
2086                 list_del(&delayed->list);
2087                 iput(delayed->inode);
2088                 kfree(delayed);
2089         }
2090         up_read(&root->fs_info->cleanup_work_sem);
2091 }
2092
2093 /*
2094  * calculate extra metadata reservation when snapshotting a subvolume
2095  * contains orphan files.
2096  */
2097 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2098                                 struct btrfs_pending_snapshot *pending,
2099                                 u64 *bytes_to_reserve)
2100 {
2101         struct btrfs_root *root;
2102         struct btrfs_block_rsv *block_rsv;
2103         u64 num_bytes;
2104         int index;
2105
2106         root = pending->root;
2107         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2108                 return;
2109
2110         block_rsv = root->orphan_block_rsv;
2111
2112         /* orphan block reservation for the snapshot */
2113         num_bytes = block_rsv->size;
2114
2115         /*
2116          * after the snapshot is created, COWing tree blocks may use more
2117          * space than it frees. So we should make sure there is enough
2118          * reserved space.
2119          */
2120         index = trans->transid & 0x1;
2121         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2122                 num_bytes += block_rsv->size -
2123                              (block_rsv->reserved + block_rsv->freed[index]);
2124         }
2125
2126         *bytes_to_reserve += num_bytes;
2127 }
2128
2129 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2130                                 struct btrfs_pending_snapshot *pending)
2131 {
2132         struct btrfs_root *root = pending->root;
2133         struct btrfs_root *snap = pending->snap;
2134         struct btrfs_block_rsv *block_rsv;
2135         u64 num_bytes;
2136         int index;
2137         int ret;
2138
2139         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2140                 return;
2141
2142         /* refill source subvolume's orphan block reservation */
2143         block_rsv = root->orphan_block_rsv;
2144         index = trans->transid & 0x1;
2145         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2146                 num_bytes = block_rsv->size -
2147                             (block_rsv->reserved + block_rsv->freed[index]);
2148                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2149                                               root->orphan_block_rsv,
2150                                               num_bytes);
2151                 BUG_ON(ret);
2152         }
2153
2154         /* setup orphan block reservation for the snapshot */
2155         block_rsv = btrfs_alloc_block_rsv(snap);
2156         BUG_ON(!block_rsv);
2157
2158         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2159         snap->orphan_block_rsv = block_rsv;
2160
2161         num_bytes = root->orphan_block_rsv->size;
2162         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2163                                       block_rsv, num_bytes);
2164         BUG_ON(ret);
2165
2166 #if 0
2167         /* insert orphan item for the snapshot */
2168         WARN_ON(!root->orphan_item_inserted);
2169         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2170                                        snap->root_key.objectid);
2171         BUG_ON(ret);
2172         snap->orphan_item_inserted = 1;
2173 #endif
2174 }
2175
2176 enum btrfs_orphan_cleanup_state {
2177         ORPHAN_CLEANUP_STARTED  = 1,
2178         ORPHAN_CLEANUP_DONE     = 2,
2179 };
2180
2181 /*
2182  * This is called in transaction commmit time. If there are no orphan
2183  * files in the subvolume, it removes orphan item and frees block_rsv
2184  * structure.
2185  */
2186 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2187                               struct btrfs_root *root)
2188 {
2189         int ret;
2190
2191         if (!list_empty(&root->orphan_list) ||
2192             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2193                 return;
2194
2195         if (root->orphan_item_inserted &&
2196             btrfs_root_refs(&root->root_item) > 0) {
2197                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2198                                             root->root_key.objectid);
2199                 BUG_ON(ret);
2200                 root->orphan_item_inserted = 0;
2201         }
2202
2203         if (root->orphan_block_rsv) {
2204                 WARN_ON(root->orphan_block_rsv->size > 0);
2205                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2206                 root->orphan_block_rsv = NULL;
2207         }
2208 }
2209
2210 /*
2211  * This creates an orphan entry for the given inode in case something goes
2212  * wrong in the middle of an unlink/truncate.
2213  *
2214  * NOTE: caller of this function should reserve 5 units of metadata for
2215  *       this function.
2216  */
2217 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2218 {
2219         struct btrfs_root *root = BTRFS_I(inode)->root;
2220         struct btrfs_block_rsv *block_rsv = NULL;
2221         int reserve = 0;
2222         int insert = 0;
2223         int ret;
2224
2225         if (!root->orphan_block_rsv) {
2226                 block_rsv = btrfs_alloc_block_rsv(root);
2227                 BUG_ON(!block_rsv);
2228         }
2229
2230         spin_lock(&root->orphan_lock);
2231         if (!root->orphan_block_rsv) {
2232                 root->orphan_block_rsv = block_rsv;
2233         } else if (block_rsv) {
2234                 btrfs_free_block_rsv(root, block_rsv);
2235                 block_rsv = NULL;
2236         }
2237
2238         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2239                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2240 #if 0
2241                 /*
2242                  * For proper ENOSPC handling, we should do orphan
2243                  * cleanup when mounting. But this introduces backward
2244                  * compatibility issue.
2245                  */
2246                 if (!xchg(&root->orphan_item_inserted, 1))
2247                         insert = 2;
2248                 else
2249                         insert = 1;
2250 #endif
2251                 insert = 1;
2252         }
2253
2254         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2255                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2256                 reserve = 1;
2257         }
2258         spin_unlock(&root->orphan_lock);
2259
2260         if (block_rsv)
2261                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2262
2263         /* grab metadata reservation from transaction handle */
2264         if (reserve) {
2265                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2266                 BUG_ON(ret);
2267         }
2268
2269         /* insert an orphan item to track this unlinked/truncated file */
2270         if (insert >= 1) {
2271                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2272                 BUG_ON(ret);
2273         }
2274
2275         /* insert an orphan item to track subvolume contains orphan files */
2276         if (insert >= 2) {
2277                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2278                                                root->root_key.objectid);
2279                 BUG_ON(ret);
2280         }
2281         return 0;
2282 }
2283
2284 /*
2285  * We have done the truncate/delete so we can go ahead and remove the orphan
2286  * item for this particular inode.
2287  */
2288 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2289 {
2290         struct btrfs_root *root = BTRFS_I(inode)->root;
2291         int delete_item = 0;
2292         int release_rsv = 0;
2293         int ret = 0;
2294
2295         spin_lock(&root->orphan_lock);
2296         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2297                 list_del_init(&BTRFS_I(inode)->i_orphan);
2298                 delete_item = 1;
2299         }
2300
2301         if (BTRFS_I(inode)->orphan_meta_reserved) {
2302                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2303                 release_rsv = 1;
2304         }
2305         spin_unlock(&root->orphan_lock);
2306
2307         if (trans && delete_item) {
2308                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2309                 BUG_ON(ret);
2310         }
2311
2312         if (release_rsv)
2313                 btrfs_orphan_release_metadata(inode);
2314
2315         return 0;
2316 }
2317
2318 /*
2319  * this cleans up any orphans that may be left on the list from the last use
2320  * of this root.
2321  */
2322 int btrfs_orphan_cleanup(struct btrfs_root *root)
2323 {
2324         struct btrfs_path *path;
2325         struct extent_buffer *leaf;
2326         struct btrfs_key key, found_key;
2327         struct btrfs_trans_handle *trans;
2328         struct inode *inode;
2329         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2330
2331         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2332                 return 0;
2333
2334         path = btrfs_alloc_path();
2335         if (!path) {
2336                 ret = -ENOMEM;
2337                 goto out;
2338         }
2339         path->reada = -1;
2340
2341         key.objectid = BTRFS_ORPHAN_OBJECTID;
2342         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2343         key.offset = (u64)-1;
2344
2345         while (1) {
2346                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2347                 if (ret < 0)
2348                         goto out;
2349
2350                 /*
2351                  * if ret == 0 means we found what we were searching for, which
2352                  * is weird, but possible, so only screw with path if we didn't
2353                  * find the key and see if we have stuff that matches
2354                  */
2355                 if (ret > 0) {
2356                         ret = 0;
2357                         if (path->slots[0] == 0)
2358                                 break;
2359                         path->slots[0]--;
2360                 }
2361
2362                 /* pull out the item */
2363                 leaf = path->nodes[0];
2364                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2365
2366                 /* make sure the item matches what we want */
2367                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2368                         break;
2369                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2370                         break;
2371
2372                 /* release the path since we're done with it */
2373                 btrfs_release_path(path);
2374
2375                 /*
2376                  * this is where we are basically btrfs_lookup, without the
2377                  * crossing root thing.  we store the inode number in the
2378                  * offset of the orphan item.
2379                  */
2380                 found_key.objectid = found_key.offset;
2381                 found_key.type = BTRFS_INODE_ITEM_KEY;
2382                 found_key.offset = 0;
2383                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2384                 if (IS_ERR(inode)) {
2385                         ret = PTR_ERR(inode);
2386                         goto out;
2387                 }
2388
2389                 /*
2390                  * add this inode to the orphan list so btrfs_orphan_del does
2391                  * the proper thing when we hit it
2392                  */
2393                 spin_lock(&root->orphan_lock);
2394                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2395                 spin_unlock(&root->orphan_lock);
2396
2397                 /*
2398                  * if this is a bad inode, means we actually succeeded in
2399                  * removing the inode, but not the orphan record, which means
2400                  * we need to manually delete the orphan since iput will just
2401                  * do a destroy_inode
2402                  */
2403                 if (is_bad_inode(inode)) {
2404                         trans = btrfs_start_transaction(root, 0);
2405                         if (IS_ERR(trans)) {
2406                                 ret = PTR_ERR(trans);
2407                                 goto out;
2408                         }
2409                         btrfs_orphan_del(trans, inode);
2410                         btrfs_end_transaction(trans, root);
2411                         iput(inode);
2412                         continue;
2413                 }
2414
2415                 /* if we have links, this was a truncate, lets do that */
2416                 if (inode->i_nlink) {
2417                         if (!S_ISREG(inode->i_mode)) {
2418                                 WARN_ON(1);
2419                                 iput(inode);
2420                                 continue;
2421                         }
2422                         nr_truncate++;
2423                         ret = btrfs_truncate(inode);
2424                 } else {
2425                         nr_unlink++;
2426                 }
2427
2428                 /* this will do delete_inode and everything for us */
2429                 iput(inode);
2430                 if (ret)
2431                         goto out;
2432         }
2433         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2434
2435         if (root->orphan_block_rsv)
2436                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2437                                         (u64)-1);
2438
2439         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2440                 trans = btrfs_join_transaction(root);
2441                 if (!IS_ERR(trans))
2442                         btrfs_end_transaction(trans, root);
2443         }
2444
2445         if (nr_unlink)
2446                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2447         if (nr_truncate)
2448                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2449
2450 out:
2451         if (ret)
2452                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2453         btrfs_free_path(path);
2454         return ret;
2455 }
2456
2457 /*
2458  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2459  * don't find any xattrs, we know there can't be any acls.
2460  *
2461  * slot is the slot the inode is in, objectid is the objectid of the inode
2462  */
2463 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2464                                           int slot, u64 objectid)
2465 {
2466         u32 nritems = btrfs_header_nritems(leaf);
2467         struct btrfs_key found_key;
2468         int scanned = 0;
2469
2470         slot++;
2471         while (slot < nritems) {
2472                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2473
2474                 /* we found a different objectid, there must not be acls */
2475                 if (found_key.objectid != objectid)
2476                         return 0;
2477
2478                 /* we found an xattr, assume we've got an acl */
2479                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2480                         return 1;
2481
2482                 /*
2483                  * we found a key greater than an xattr key, there can't
2484                  * be any acls later on
2485                  */
2486                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2487                         return 0;
2488
2489                 slot++;
2490                 scanned++;
2491
2492                 /*
2493                  * it goes inode, inode backrefs, xattrs, extents,
2494                  * so if there are a ton of hard links to an inode there can
2495                  * be a lot of backrefs.  Don't waste time searching too hard,
2496                  * this is just an optimization
2497                  */
2498                 if (scanned >= 8)
2499                         break;
2500         }
2501         /* we hit the end of the leaf before we found an xattr or
2502          * something larger than an xattr.  We have to assume the inode
2503          * has acls
2504          */
2505         return 1;
2506 }
2507
2508 /*
2509  * read an inode from the btree into the in-memory inode
2510  */
2511 static void btrfs_read_locked_inode(struct inode *inode)
2512 {
2513         struct btrfs_path *path;
2514         struct extent_buffer *leaf;
2515         struct btrfs_inode_item *inode_item;
2516         struct btrfs_timespec *tspec;
2517         struct btrfs_root *root = BTRFS_I(inode)->root;
2518         struct btrfs_key location;
2519         int maybe_acls;
2520         u32 rdev;
2521         int ret;
2522         bool filled = false;
2523
2524         ret = btrfs_fill_inode(inode, &rdev);
2525         if (!ret)
2526                 filled = true;
2527
2528         path = btrfs_alloc_path();
2529         BUG_ON(!path);
2530         path->leave_spinning = 1;
2531         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2532
2533         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2534         if (ret)
2535                 goto make_bad;
2536
2537         leaf = path->nodes[0];
2538
2539         if (filled)
2540                 goto cache_acl;
2541
2542         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2543                                     struct btrfs_inode_item);
2544         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2545         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2546         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2547         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2548         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2549
2550         tspec = btrfs_inode_atime(inode_item);
2551         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2552         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2553
2554         tspec = btrfs_inode_mtime(inode_item);
2555         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2556         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2557
2558         tspec = btrfs_inode_ctime(inode_item);
2559         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2560         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2561
2562         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2563         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2564         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2565         inode->i_generation = BTRFS_I(inode)->generation;
2566         inode->i_rdev = 0;
2567         rdev = btrfs_inode_rdev(leaf, inode_item);
2568
2569         BTRFS_I(inode)->index_cnt = (u64)-1;
2570         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2571 cache_acl:
2572         /*
2573          * try to precache a NULL acl entry for files that don't have
2574          * any xattrs or acls
2575          */
2576         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2577                                            btrfs_ino(inode));
2578         if (!maybe_acls)
2579                 cache_no_acl(inode);
2580
2581         btrfs_free_path(path);
2582
2583         switch (inode->i_mode & S_IFMT) {
2584         case S_IFREG:
2585                 inode->i_mapping->a_ops = &btrfs_aops;
2586                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2587                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2588                 inode->i_fop = &btrfs_file_operations;
2589                 inode->i_op = &btrfs_file_inode_operations;
2590                 break;
2591         case S_IFDIR:
2592                 inode->i_fop = &btrfs_dir_file_operations;
2593                 if (root == root->fs_info->tree_root)
2594                         inode->i_op = &btrfs_dir_ro_inode_operations;
2595                 else
2596                         inode->i_op = &btrfs_dir_inode_operations;
2597                 break;
2598         case S_IFLNK:
2599                 inode->i_op = &btrfs_symlink_inode_operations;
2600                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2601                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2602                 break;
2603         default:
2604                 inode->i_op = &btrfs_special_inode_operations;
2605                 init_special_inode(inode, inode->i_mode, rdev);
2606                 break;
2607         }
2608
2609         btrfs_update_iflags(inode);
2610         return;
2611
2612 make_bad:
2613         btrfs_free_path(path);
2614         make_bad_inode(inode);
2615 }
2616
2617 /*
2618  * given a leaf and an inode, copy the inode fields into the leaf
2619  */
2620 static void fill_inode_item(struct btrfs_trans_handle *trans,
2621                             struct extent_buffer *leaf,
2622                             struct btrfs_inode_item *item,
2623                             struct inode *inode)
2624 {
2625         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2626         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2627         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2628         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2629         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2630
2631         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2632                                inode->i_atime.tv_sec);
2633         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2634                                 inode->i_atime.tv_nsec);
2635
2636         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2637                                inode->i_mtime.tv_sec);
2638         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2639                                 inode->i_mtime.tv_nsec);
2640
2641         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2642                                inode->i_ctime.tv_sec);
2643         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2644                                 inode->i_ctime.tv_nsec);
2645
2646         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2647         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2648         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2649         btrfs_set_inode_transid(leaf, item, trans->transid);
2650         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2651         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2652         btrfs_set_inode_block_group(leaf, item, 0);
2653 }
2654
2655 /*
2656  * copy everything in the in-memory inode into the btree.
2657  */
2658 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2659                                 struct btrfs_root *root, struct inode *inode)
2660 {
2661         struct btrfs_inode_item *inode_item;
2662         struct btrfs_path *path;
2663         struct extent_buffer *leaf;
2664         int ret;
2665
2666         /*
2667          * If the inode is a free space inode, we can deadlock during commit
2668          * if we put it into the delayed code.
2669          *
2670          * The data relocation inode should also be directly updated
2671          * without delay
2672          */
2673         if (!is_free_space_inode(root, inode)
2674             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2675                 ret = btrfs_delayed_update_inode(trans, root, inode);
2676                 if (!ret)
2677                         btrfs_set_inode_last_trans(trans, inode);
2678                 return ret;
2679         }
2680
2681         path = btrfs_alloc_path();
2682         if (!path)
2683                 return -ENOMEM;
2684
2685         path->leave_spinning = 1;
2686         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2687                                  1);
2688         if (ret) {
2689                 if (ret > 0)
2690                         ret = -ENOENT;
2691                 goto failed;
2692         }
2693
2694         btrfs_unlock_up_safe(path, 1);
2695         leaf = path->nodes[0];
2696         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2697                                     struct btrfs_inode_item);
2698
2699         fill_inode_item(trans, leaf, inode_item, inode);
2700         btrfs_mark_buffer_dirty(leaf);
2701         btrfs_set_inode_last_trans(trans, inode);
2702         ret = 0;
2703 failed:
2704         btrfs_free_path(path);
2705         return ret;
2706 }
2707
2708 /*
2709  * unlink helper that gets used here in inode.c and in the tree logging
2710  * recovery code.  It remove a link in a directory with a given name, and
2711  * also drops the back refs in the inode to the directory
2712  */
2713 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2714                                 struct btrfs_root *root,
2715                                 struct inode *dir, struct inode *inode,
2716                                 const char *name, int name_len)
2717 {
2718         struct btrfs_path *path;
2719         int ret = 0;
2720         struct extent_buffer *leaf;
2721         struct btrfs_dir_item *di;
2722         struct btrfs_key key;
2723         u64 index;
2724         u64 ino = btrfs_ino(inode);
2725         u64 dir_ino = btrfs_ino(dir);
2726
2727         path = btrfs_alloc_path();
2728         if (!path) {
2729                 ret = -ENOMEM;
2730                 goto out;
2731         }
2732
2733         path->leave_spinning = 1;
2734         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2735                                     name, name_len, -1);
2736         if (IS_ERR(di)) {
2737                 ret = PTR_ERR(di);
2738                 goto err;
2739         }
2740         if (!di) {
2741                 ret = -ENOENT;
2742                 goto err;
2743         }
2744         leaf = path->nodes[0];
2745         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2746         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2747         if (ret)
2748                 goto err;
2749         btrfs_release_path(path);
2750
2751         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2752                                   dir_ino, &index);
2753         if (ret) {
2754                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2755                        "inode %llu parent %llu\n", name_len, name,
2756                        (unsigned long long)ino, (unsigned long long)dir_ino);
2757                 goto err;
2758         }
2759
2760         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2761         if (ret)
2762                 goto err;
2763
2764         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2765                                          inode, dir_ino);
2766         BUG_ON(ret != 0 && ret != -ENOENT);
2767
2768         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2769                                            dir, index);
2770         if (ret == -ENOENT)
2771                 ret = 0;
2772 err:
2773         btrfs_free_path(path);
2774         if (ret)
2775                 goto out;
2776
2777         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2778         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2779         btrfs_update_inode(trans, root, dir);
2780 out:
2781         return ret;
2782 }
2783
2784 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2785                        struct btrfs_root *root,
2786                        struct inode *dir, struct inode *inode,
2787                        const char *name, int name_len)
2788 {
2789         int ret;
2790         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2791         if (!ret) {
2792                 btrfs_drop_nlink(inode);
2793                 ret = btrfs_update_inode(trans, root, inode);
2794         }
2795         return ret;
2796 }
2797
2798
2799 /* helper to check if there is any shared block in the path */
2800 static int check_path_shared(struct btrfs_root *root,
2801                              struct btrfs_path *path)
2802 {
2803         struct extent_buffer *eb;
2804         int level;
2805         u64 refs = 1;
2806
2807         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2808                 int ret;
2809
2810                 if (!path->nodes[level])
2811                         break;
2812                 eb = path->nodes[level];
2813                 if (!btrfs_block_can_be_shared(root, eb))
2814                         continue;
2815                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2816                                                &refs, NULL);
2817                 if (refs > 1)
2818                         return 1;
2819         }
2820         return 0;
2821 }
2822
2823 /*
2824  * helper to start transaction for unlink and rmdir.
2825  *
2826  * unlink and rmdir are special in btrfs, they do not always free space.
2827  * so in enospc case, we should make sure they will free space before
2828  * allowing them to use the global metadata reservation.
2829  */
2830 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2831                                                        struct dentry *dentry)
2832 {
2833         struct btrfs_trans_handle *trans;
2834         struct btrfs_root *root = BTRFS_I(dir)->root;
2835         struct btrfs_path *path;
2836         struct btrfs_inode_ref *ref;
2837         struct btrfs_dir_item *di;
2838         struct inode *inode = dentry->d_inode;
2839         u64 index;
2840         int check_link = 1;
2841         int err = -ENOSPC;
2842         int ret;
2843         u64 ino = btrfs_ino(inode);
2844         u64 dir_ino = btrfs_ino(dir);
2845
2846         trans = btrfs_start_transaction(root, 10);
2847         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2848                 return trans;
2849
2850         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2851                 return ERR_PTR(-ENOSPC);
2852
2853         /* check if there is someone else holds reference */
2854         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2855                 return ERR_PTR(-ENOSPC);
2856
2857         if (atomic_read(&inode->i_count) > 2)
2858                 return ERR_PTR(-ENOSPC);
2859
2860         if (xchg(&root->fs_info->enospc_unlink, 1))
2861                 return ERR_PTR(-ENOSPC);
2862
2863         path = btrfs_alloc_path();
2864         if (!path) {
2865                 root->fs_info->enospc_unlink = 0;
2866                 return ERR_PTR(-ENOMEM);
2867         }
2868
2869         trans = btrfs_start_transaction(root, 0);
2870         if (IS_ERR(trans)) {
2871                 btrfs_free_path(path);
2872                 root->fs_info->enospc_unlink = 0;
2873                 return trans;
2874         }
2875
2876         path->skip_locking = 1;
2877         path->search_commit_root = 1;
2878
2879         ret = btrfs_lookup_inode(trans, root, path,
2880                                 &BTRFS_I(dir)->location, 0);
2881         if (ret < 0) {
2882                 err = ret;
2883                 goto out;
2884         }
2885         if (ret == 0) {
2886                 if (check_path_shared(root, path))
2887                         goto out;
2888         } else {
2889                 check_link = 0;
2890         }
2891         btrfs_release_path(path);
2892
2893         ret = btrfs_lookup_inode(trans, root, path,
2894                                 &BTRFS_I(inode)->location, 0);
2895         if (ret < 0) {
2896                 err = ret;
2897                 goto out;
2898         }
2899         if (ret == 0) {
2900                 if (check_path_shared(root, path))
2901                         goto out;
2902         } else {
2903                 check_link = 0;
2904         }
2905         btrfs_release_path(path);
2906
2907         if (ret == 0 && S_ISREG(inode->i_mode)) {
2908                 ret = btrfs_lookup_file_extent(trans, root, path,
2909                                                ino, (u64)-1, 0);
2910                 if (ret < 0) {
2911                         err = ret;
2912                         goto out;
2913                 }
2914                 BUG_ON(ret == 0);
2915                 if (check_path_shared(root, path))
2916                         goto out;
2917                 btrfs_release_path(path);
2918         }
2919
2920         if (!check_link) {
2921                 err = 0;
2922                 goto out;
2923         }
2924
2925         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2926                                 dentry->d_name.name, dentry->d_name.len, 0);
2927         if (IS_ERR(di)) {
2928                 err = PTR_ERR(di);
2929                 goto out;
2930         }
2931         if (di) {
2932                 if (check_path_shared(root, path))
2933                         goto out;
2934         } else {
2935                 err = 0;
2936                 goto out;
2937         }
2938         btrfs_release_path(path);
2939
2940         ref = btrfs_lookup_inode_ref(trans, root, path,
2941                                 dentry->d_name.name, dentry->d_name.len,
2942                                 ino, dir_ino, 0);
2943         if (IS_ERR(ref)) {
2944                 err = PTR_ERR(ref);
2945                 goto out;
2946         }
2947         BUG_ON(!ref);
2948         if (check_path_shared(root, path))
2949                 goto out;
2950         index = btrfs_inode_ref_index(path->nodes[0], ref);
2951         btrfs_release_path(path);
2952
2953         /*
2954          * This is a commit root search, if we can lookup inode item and other
2955          * relative items in the commit root, it means the transaction of
2956          * dir/file creation has been committed, and the dir index item that we
2957          * delay to insert has also been inserted into the commit root. So
2958          * we needn't worry about the delayed insertion of the dir index item
2959          * here.
2960          */
2961         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2962                                 dentry->d_name.name, dentry->d_name.len, 0);
2963         if (IS_ERR(di)) {
2964                 err = PTR_ERR(di);
2965                 goto out;
2966         }
2967         BUG_ON(ret == -ENOENT);
2968         if (check_path_shared(root, path))
2969                 goto out;
2970
2971         err = 0;
2972 out:
2973         btrfs_free_path(path);
2974         if (err) {
2975                 btrfs_end_transaction(trans, root);
2976                 root->fs_info->enospc_unlink = 0;
2977                 return ERR_PTR(err);
2978         }
2979
2980         trans->block_rsv = &root->fs_info->global_block_rsv;
2981         return trans;
2982 }
2983
2984 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2985                                struct btrfs_root *root)
2986 {
2987         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2988                 BUG_ON(!root->fs_info->enospc_unlink);
2989                 root->fs_info->enospc_unlink = 0;
2990         }
2991         btrfs_end_transaction_throttle(trans, root);
2992 }
2993
2994 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2995 {
2996         struct btrfs_root *root = BTRFS_I(dir)->root;
2997         struct btrfs_trans_handle *trans;
2998         struct inode *inode = dentry->d_inode;
2999         int ret;
3000         unsigned long nr = 0;
3001
3002         trans = __unlink_start_trans(dir, dentry);
3003         if (IS_ERR(trans))
3004                 return PTR_ERR(trans);
3005
3006         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3007
3008         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3009                                  dentry->d_name.name, dentry->d_name.len);
3010         BUG_ON(ret);
3011
3012         if (inode->i_nlink == 0) {
3013                 ret = btrfs_orphan_add(trans, inode);
3014                 BUG_ON(ret);
3015         }
3016
3017         nr = trans->blocks_used;
3018         __unlink_end_trans(trans, root);
3019         btrfs_btree_balance_dirty(root, nr);
3020         return ret;
3021 }
3022
3023 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3024                         struct btrfs_root *root,
3025                         struct inode *dir, u64 objectid,
3026                         const char *name, int name_len)
3027 {
3028         struct btrfs_path *path;
3029         struct extent_buffer *leaf;
3030         struct btrfs_dir_item *di;
3031         struct btrfs_key key;
3032         u64 index;
3033         int ret;
3034         u64 dir_ino = btrfs_ino(dir);
3035
3036         path = btrfs_alloc_path();
3037         if (!path)
3038                 return -ENOMEM;
3039
3040         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3041                                    name, name_len, -1);
3042         BUG_ON(IS_ERR_OR_NULL(di));
3043
3044         leaf = path->nodes[0];
3045         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3046         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3047         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3048         BUG_ON(ret);
3049         btrfs_release_path(path);
3050
3051         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3052                                  objectid, root->root_key.objectid,
3053                                  dir_ino, &index, name, name_len);
3054         if (ret < 0) {
3055                 BUG_ON(ret != -ENOENT);
3056                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3057                                                  name, name_len);
3058                 BUG_ON(IS_ERR_OR_NULL(di));
3059
3060                 leaf = path->nodes[0];
3061                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3062                 btrfs_release_path(path);
3063                 index = key.offset;
3064         }
3065         btrfs_release_path(path);
3066
3067         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3068         BUG_ON(ret);
3069
3070         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3071         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3072         ret = btrfs_update_inode(trans, root, dir);
3073         BUG_ON(ret);
3074
3075         btrfs_free_path(path);
3076         return 0;
3077 }
3078
3079 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3080 {
3081         struct inode *inode = dentry->d_inode;
3082         int err = 0;
3083         struct btrfs_root *root = BTRFS_I(dir)->root;
3084         struct btrfs_trans_handle *trans;
3085         unsigned long nr = 0;
3086
3087         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3088             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3089                 return -ENOTEMPTY;
3090
3091         trans = __unlink_start_trans(dir, dentry);
3092         if (IS_ERR(trans))
3093                 return PTR_ERR(trans);
3094
3095         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3096                 err = btrfs_unlink_subvol(trans, root, dir,
3097                                           BTRFS_I(inode)->location.objectid,
3098                                           dentry->d_name.name,
3099                                           dentry->d_name.len);
3100                 goto out;
3101         }
3102
3103         err = btrfs_orphan_add(trans, inode);
3104         if (err)
3105                 goto out;
3106
3107         /* now the directory is empty */
3108         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3109                                  dentry->d_name.name, dentry->d_name.len);
3110         if (!err)
3111                 btrfs_i_size_write(inode, 0);
3112 out:
3113         nr = trans->blocks_used;
3114         __unlink_end_trans(trans, root);
3115         btrfs_btree_balance_dirty(root, nr);
3116
3117         return err;
3118 }
3119
3120 /*
3121  * this can truncate away extent items, csum items and directory items.
3122  * It starts at a high offset and removes keys until it can't find
3123  * any higher than new_size
3124  *
3125  * csum items that cross the new i_size are truncated to the new size
3126  * as well.
3127  *
3128  * min_type is the minimum key type to truncate down to.  If set to 0, this
3129  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3130  */
3131 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3132                                struct btrfs_root *root,
3133                                struct inode *inode,
3134                                u64 new_size, u32 min_type)
3135 {
3136         struct btrfs_path *path;
3137         struct extent_buffer *leaf;
3138         struct btrfs_file_extent_item *fi;
3139         struct btrfs_key key;
3140         struct btrfs_key found_key;
3141         u64 extent_start = 0;
3142         u64 extent_num_bytes = 0;
3143         u64 extent_offset = 0;
3144         u64 item_end = 0;
3145         u64 mask = root->sectorsize - 1;
3146         u32 found_type = (u8)-1;
3147         int found_extent;
3148         int del_item;
3149         int pending_del_nr = 0;
3150         int pending_del_slot = 0;
3151         int extent_type = -1;
3152         int encoding;
3153         int ret;
3154         int err = 0;
3155         u64 ino = btrfs_ino(inode);
3156
3157         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3158
3159         if (root->ref_cows || root == root->fs_info->tree_root)
3160                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3161
3162         /*
3163          * This function is also used to drop the items in the log tree before
3164          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3165          * it is used to drop the loged items. So we shouldn't kill the delayed
3166          * items.
3167          */
3168         if (min_type == 0 && root == BTRFS_I(inode)->root)
3169                 btrfs_kill_delayed_inode_items(inode);
3170
3171         path = btrfs_alloc_path();
3172         BUG_ON(!path);
3173         path->reada = -1;
3174
3175         key.objectid = ino;
3176         key.offset = (u64)-1;
3177         key.type = (u8)-1;
3178
3179 search_again:
3180         path->leave_spinning = 1;
3181         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3182         if (ret < 0) {
3183                 err = ret;
3184                 goto out;
3185         }
3186
3187         if (ret > 0) {
3188                 /* there are no items in the tree for us to truncate, we're
3189                  * done
3190                  */
3191                 if (path->slots[0] == 0)
3192                         goto out;
3193                 path->slots[0]--;
3194         }
3195
3196         while (1) {
3197                 fi = NULL;
3198                 leaf = path->nodes[0];
3199                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3200                 found_type = btrfs_key_type(&found_key);
3201                 encoding = 0;
3202
3203                 if (found_key.objectid != ino)
3204                         break;
3205
3206                 if (found_type < min_type)
3207                         break;
3208
3209                 item_end = found_key.offset;
3210                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3211                         fi = btrfs_item_ptr(leaf, path->slots[0],
3212                                             struct btrfs_file_extent_item);
3213                         extent_type = btrfs_file_extent_type(leaf, fi);
3214                         encoding = btrfs_file_extent_compression(leaf, fi);
3215                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3216                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3217
3218                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3219                                 item_end +=
3220                                     btrfs_file_extent_num_bytes(leaf, fi);
3221                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3222                                 item_end += btrfs_file_extent_inline_len(leaf,
3223                                                                          fi);
3224                         }
3225                         item_end--;
3226                 }
3227                 if (found_type > min_type) {
3228                         del_item = 1;
3229                 } else {
3230                         if (item_end < new_size)
3231                                 break;
3232                         if (found_key.offset >= new_size)
3233                                 del_item = 1;
3234                         else
3235                                 del_item = 0;
3236                 }
3237                 found_extent = 0;
3238                 /* FIXME, shrink the extent if the ref count is only 1 */
3239                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3240                         goto delete;
3241
3242                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3243                         u64 num_dec;
3244                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3245                         if (!del_item && !encoding) {
3246                                 u64 orig_num_bytes =
3247                                         btrfs_file_extent_num_bytes(leaf, fi);
3248                                 extent_num_bytes = new_size -
3249                                         found_key.offset + root->sectorsize - 1;
3250                                 extent_num_bytes = extent_num_bytes &
3251                                         ~((u64)root->sectorsize - 1);
3252                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3253                                                          extent_num_bytes);
3254                                 num_dec = (orig_num_bytes -
3255                                            extent_num_bytes);
3256                                 if (root->ref_cows && extent_start != 0)
3257                                         inode_sub_bytes(inode, num_dec);
3258                                 btrfs_mark_buffer_dirty(leaf);
3259                         } else {
3260                                 extent_num_bytes =
3261                                         btrfs_file_extent_disk_num_bytes(leaf,
3262                                                                          fi);
3263                                 extent_offset = found_key.offset -
3264                                         btrfs_file_extent_offset(leaf, fi);
3265
3266                                 /* FIXME blocksize != 4096 */
3267                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3268                                 if (extent_start != 0) {
3269                                         found_extent = 1;
3270                                         if (root->ref_cows)
3271                                                 inode_sub_bytes(inode, num_dec);
3272                                 }
3273                         }
3274                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3275                         /*
3276                          * we can't truncate inline items that have had
3277                          * special encodings
3278                          */
3279                         if (!del_item &&
3280                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3281                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3282                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3283                                 u32 size = new_size - found_key.offset;
3284
3285                                 if (root->ref_cows) {
3286                                         inode_sub_bytes(inode, item_end + 1 -
3287                                                         new_size);
3288                                 }
3289                                 size =
3290                                     btrfs_file_extent_calc_inline_size(size);
3291                                 ret = btrfs_truncate_item(trans, root, path,
3292                                                           size, 1);
3293                         } else if (root->ref_cows) {
3294                                 inode_sub_bytes(inode, item_end + 1 -
3295                                                 found_key.offset);
3296                         }
3297                 }
3298 delete:
3299                 if (del_item) {
3300                         if (!pending_del_nr) {
3301                                 /* no pending yet, add ourselves */
3302                                 pending_del_slot = path->slots[0];
3303                                 pending_del_nr = 1;
3304                         } else if (pending_del_nr &&
3305                                    path->slots[0] + 1 == pending_del_slot) {
3306                                 /* hop on the pending chunk */
3307                                 pending_del_nr++;
3308                                 pending_del_slot = path->slots[0];
3309                         } else {
3310                                 BUG();
3311                         }
3312                 } else {
3313                         break;
3314                 }
3315                 if (found_extent && (root->ref_cows ||
3316                                      root == root->fs_info->tree_root)) {
3317                         btrfs_set_path_blocking(path);
3318                         ret = btrfs_free_extent(trans, root, extent_start,
3319                                                 extent_num_bytes, 0,
3320                                                 btrfs_header_owner(leaf),
3321                                                 ino, extent_offset);
3322                         BUG_ON(ret);
3323                 }
3324
3325                 if (found_type == BTRFS_INODE_ITEM_KEY)
3326                         break;
3327
3328                 if (path->slots[0] == 0 ||
3329                     path->slots[0] != pending_del_slot) {
3330                         if (root->ref_cows &&
3331                             BTRFS_I(inode)->location.objectid !=
3332                                                 BTRFS_FREE_INO_OBJECTID) {
3333                                 err = -EAGAIN;
3334                                 goto out;
3335                         }
3336                         if (pending_del_nr) {
3337                                 ret = btrfs_del_items(trans, root, path,
3338                                                 pending_del_slot,
3339                                                 pending_del_nr);
3340                                 BUG_ON(ret);
3341                                 pending_del_nr = 0;
3342                         }
3343                         btrfs_release_path(path);
3344                         goto search_again;
3345                 } else {
3346                         path->slots[0]--;
3347                 }
3348         }
3349 out:
3350         if (pending_del_nr) {
3351                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3352                                       pending_del_nr);
3353                 BUG_ON(ret);
3354         }
3355         btrfs_free_path(path);
3356         return err;
3357 }
3358
3359 /*
3360  * taken from block_truncate_page, but does cow as it zeros out
3361  * any bytes left in the last page in the file.
3362  */
3363 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3364 {
3365         struct inode *inode = mapping->host;
3366         struct btrfs_root *root = BTRFS_I(inode)->root;
3367         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3368         struct btrfs_ordered_extent *ordered;
3369         struct extent_state *cached_state = NULL;
3370         char *kaddr;
3371         u32 blocksize = root->sectorsize;
3372         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3373         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3374         struct page *page;
3375         int ret = 0;
3376         u64 page_start;
3377         u64 page_end;
3378
3379         if ((offset & (blocksize - 1)) == 0)
3380                 goto out;
3381         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3382         if (ret)
3383                 goto out;
3384
3385         ret = -ENOMEM;
3386 again:
3387         page = find_or_create_page(mapping, index, GFP_NOFS);
3388         if (!page) {
3389                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3390                 goto out;
3391         }
3392
3393         page_start = page_offset(page);
3394         page_end = page_start + PAGE_CACHE_SIZE - 1;
3395
3396         if (!PageUptodate(page)) {
3397                 ret = btrfs_readpage(NULL, page);
3398                 lock_page(page);
3399                 if (page->mapping != mapping) {
3400                         unlock_page(page);
3401                         page_cache_release(page);
3402                         goto again;
3403                 }
3404                 if (!PageUptodate(page)) {
3405                         ret = -EIO;
3406                         goto out_unlock;
3407                 }
3408         }
3409         wait_on_page_writeback(page);
3410
3411         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3412                          GFP_NOFS);
3413         set_page_extent_mapped(page);
3414
3415         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3416         if (ordered) {
3417                 unlock_extent_cached(io_tree, page_start, page_end,
3418                                      &cached_state, GFP_NOFS);
3419                 unlock_page(page);
3420                 page_cache_release(page);
3421                 btrfs_start_ordered_extent(inode, ordered, 1);
3422                 btrfs_put_ordered_extent(ordered);
3423                 goto again;
3424         }
3425
3426         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3427                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3428                           0, 0, &cached_state, GFP_NOFS);
3429
3430         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3431                                         &cached_state);
3432         if (ret) {
3433                 unlock_extent_cached(io_tree, page_start, page_end,
3434                                      &cached_state, GFP_NOFS);
3435                 goto out_unlock;
3436         }
3437
3438         ret = 0;
3439         if (offset != PAGE_CACHE_SIZE) {
3440                 kaddr = kmap(page);
3441                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3442                 flush_dcache_page(page);
3443                 kunmap(page);
3444         }
3445         ClearPageChecked(page);
3446         set_page_dirty(page);
3447         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3448                              GFP_NOFS);
3449
3450 out_unlock:
3451         if (ret)
3452                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3453         unlock_page(page);
3454         page_cache_release(page);
3455 out:
3456         return ret;
3457 }
3458
3459 /*
3460  * This function puts in dummy file extents for the area we're creating a hole
3461  * for.  So if we are truncating this file to a larger size we need to insert
3462  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3463  * the range between oldsize and size
3464  */
3465 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3466 {
3467         struct btrfs_trans_handle *trans;
3468         struct btrfs_root *root = BTRFS_I(inode)->root;
3469         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3470         struct extent_map *em = NULL;
3471         struct extent_state *cached_state = NULL;
3472         u64 mask = root->sectorsize - 1;
3473         u64 hole_start = (oldsize + mask) & ~mask;
3474         u64 block_end = (size + mask) & ~mask;
3475         u64 last_byte;
3476         u64 cur_offset;
3477         u64 hole_size;
3478         int err = 0;
3479
3480         if (size <= hole_start)
3481                 return 0;
3482
3483         while (1) {
3484                 struct btrfs_ordered_extent *ordered;
3485                 btrfs_wait_ordered_range(inode, hole_start,
3486                                          block_end - hole_start);
3487                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3488                                  &cached_state, GFP_NOFS);
3489                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3490                 if (!ordered)
3491                         break;
3492                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3493                                      &cached_state, GFP_NOFS);
3494                 btrfs_put_ordered_extent(ordered);
3495         }
3496
3497         cur_offset = hole_start;
3498         while (1) {
3499                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3500                                 block_end - cur_offset, 0);
3501                 BUG_ON(IS_ERR_OR_NULL(em));
3502                 last_byte = min(extent_map_end(em), block_end);
3503                 last_byte = (last_byte + mask) & ~mask;
3504                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3505                         u64 hint_byte = 0;
3506                         hole_size = last_byte - cur_offset;
3507
3508                         trans = btrfs_start_transaction(root, 2);
3509                         if (IS_ERR(trans)) {
3510                                 err = PTR_ERR(trans);
3511                                 break;
3512                         }
3513
3514                         err = btrfs_drop_extents(trans, inode, cur_offset,
3515                                                  cur_offset + hole_size,
3516                                                  &hint_byte, 1);
3517                         if (err)
3518                                 break;
3519
3520                         err = btrfs_insert_file_extent(trans, root,
3521                                         btrfs_ino(inode), cur_offset, 0,
3522                                         0, hole_size, 0, hole_size,
3523                                         0, 0, 0);
3524                         if (err)
3525                                 break;
3526
3527                         btrfs_drop_extent_cache(inode, hole_start,
3528                                         last_byte - 1, 0);
3529
3530                         btrfs_end_transaction(trans, root);
3531                 }
3532                 free_extent_map(em);
3533                 em = NULL;
3534                 cur_offset = last_byte;
3535                 if (cur_offset >= block_end)
3536                         break;
3537         }
3538
3539         free_extent_map(em);
3540         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3541                              GFP_NOFS);
3542         return err;
3543 }
3544
3545 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3546 {
3547         loff_t oldsize = i_size_read(inode);
3548         int ret;
3549
3550         if (newsize == oldsize)
3551                 return 0;
3552
3553         if (newsize > oldsize) {
3554                 i_size_write(inode, newsize);
3555                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3556                 truncate_pagecache(inode, oldsize, newsize);
3557                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3558                 if (ret) {
3559                         btrfs_setsize(inode, oldsize);
3560                         return ret;
3561                 }
3562
3563                 mark_inode_dirty(inode);
3564         } else {
3565
3566                 /*
3567                  * We're truncating a file that used to have good data down to
3568                  * zero. Make sure it gets into the ordered flush list so that
3569                  * any new writes get down to disk quickly.
3570                  */
3571                 if (newsize == 0)
3572                         BTRFS_I(inode)->ordered_data_close = 1;
3573
3574                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3575                 truncate_setsize(inode, newsize);
3576                 ret = btrfs_truncate(inode);
3577         }
3578
3579         return ret;
3580 }
3581
3582 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3583 {
3584         struct inode *inode = dentry->d_inode;
3585         struct btrfs_root *root = BTRFS_I(inode)->root;
3586         int err;
3587
3588         if (btrfs_root_readonly(root))
3589                 return -EROFS;
3590
3591         err = inode_change_ok(inode, attr);
3592         if (err)
3593                 return err;
3594
3595         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3596                 err = btrfs_setsize(inode, attr->ia_size);
3597                 if (err)
3598                         return err;
3599         }
3600
3601         if (attr->ia_valid) {
3602                 setattr_copy(inode, attr);
3603                 mark_inode_dirty(inode);
3604
3605                 if (attr->ia_valid & ATTR_MODE)
3606                         err = btrfs_acl_chmod(inode);
3607         }
3608
3609         return err;
3610 }
3611
3612 void btrfs_evict_inode(struct inode *inode)
3613 {
3614         struct btrfs_trans_handle *trans;
3615         struct btrfs_root *root = BTRFS_I(inode)->root;
3616         unsigned long nr;
3617         int ret;
3618
3619         trace_btrfs_inode_evict(inode);
3620
3621         truncate_inode_pages(&inode->i_data, 0);
3622         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3623                                is_free_space_inode(root, inode)))
3624                 goto no_delete;
3625
3626         if (is_bad_inode(inode)) {
3627                 btrfs_orphan_del(NULL, inode);
3628                 goto no_delete;
3629         }
3630         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3631         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3632
3633         if (root->fs_info->log_root_recovering) {
3634                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3635                 goto no_delete;
3636         }
3637
3638         if (inode->i_nlink > 0) {
3639                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3640                 goto no_delete;
3641         }
3642
3643         btrfs_i_size_write(inode, 0);
3644
3645         while (1) {
3646                 trans = btrfs_join_transaction(root);
3647                 BUG_ON(IS_ERR(trans));
3648                 trans->block_rsv = root->orphan_block_rsv;
3649
3650                 ret = btrfs_block_rsv_check(trans, root,
3651                                             root->orphan_block_rsv, 0, 5);
3652                 if (ret) {
3653                         BUG_ON(ret != -EAGAIN);
3654                         ret = btrfs_commit_transaction(trans, root);
3655                         BUG_ON(ret);
3656                         continue;
3657                 }
3658
3659                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3660                 if (ret != -EAGAIN)
3661                         break;
3662
3663                 nr = trans->blocks_used;
3664                 btrfs_end_transaction(trans, root);
3665                 trans = NULL;
3666                 btrfs_btree_balance_dirty(root, nr);
3667
3668         }
3669
3670         if (ret == 0) {
3671                 ret = btrfs_orphan_del(trans, inode);
3672                 BUG_ON(ret);
3673         }
3674
3675         if (!(root == root->fs_info->tree_root ||
3676               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3677                 btrfs_return_ino(root, btrfs_ino(inode));
3678
3679         nr = trans->blocks_used;
3680         btrfs_end_transaction(trans, root);
3681         btrfs_btree_balance_dirty(root, nr);
3682 no_delete:
3683         end_writeback(inode);
3684         return;
3685 }
3686
3687 /*
3688  * this returns the key found in the dir entry in the location pointer.
3689  * If no dir entries were found, location->objectid is 0.
3690  */
3691 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3692                                struct btrfs_key *location)
3693 {
3694         const char *name = dentry->d_name.name;
3695         int namelen = dentry->d_name.len;
3696         struct btrfs_dir_item *di;
3697         struct btrfs_path *path;
3698         struct btrfs_root *root = BTRFS_I(dir)->root;
3699         int ret = 0;
3700
3701         path = btrfs_alloc_path();
3702         BUG_ON(!path);
3703
3704         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3705                                     namelen, 0);
3706         if (IS_ERR(di))
3707                 ret = PTR_ERR(di);
3708
3709         if (IS_ERR_OR_NULL(di))
3710                 goto out_err;
3711
3712         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3713 out:
3714         btrfs_free_path(path);
3715         return ret;
3716 out_err:
3717         location->objectid = 0;
3718         goto out;
3719 }
3720
3721 /*
3722  * when we hit a tree root in a directory, the btrfs part of the inode
3723  * needs to be changed to reflect the root directory of the tree root.  This
3724  * is kind of like crossing a mount point.
3725  */
3726 static int fixup_tree_root_location(struct btrfs_root *root,
3727                                     struct inode *dir,
3728                                     struct dentry *dentry,
3729                                     struct btrfs_key *location,
3730                                     struct btrfs_root **sub_root)
3731 {
3732         struct btrfs_path *path;
3733         struct btrfs_root *new_root;
3734         struct btrfs_root_ref *ref;
3735         struct extent_buffer *leaf;
3736         int ret;
3737         int err = 0;
3738
3739         path = btrfs_alloc_path();
3740         if (!path) {
3741                 err = -ENOMEM;
3742                 goto out;
3743         }
3744
3745         err = -ENOENT;
3746         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3747                                   BTRFS_I(dir)->root->root_key.objectid,
3748                                   location->objectid);
3749         if (ret) {
3750                 if (ret < 0)
3751                         err = ret;
3752                 goto out;
3753         }
3754
3755         leaf = path->nodes[0];
3756         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3757         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3758             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3759                 goto out;
3760
3761         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3762                                    (unsigned long)(ref + 1),
3763                                    dentry->d_name.len);
3764         if (ret)
3765                 goto out;
3766
3767         btrfs_release_path(path);
3768
3769         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3770         if (IS_ERR(new_root)) {
3771                 err = PTR_ERR(new_root);
3772                 goto out;
3773         }
3774
3775         if (btrfs_root_refs(&new_root->root_item) == 0) {
3776                 err = -ENOENT;
3777                 goto out;
3778         }
3779
3780         *sub_root = new_root;
3781         location->objectid = btrfs_root_dirid(&new_root->root_item);
3782         location->type = BTRFS_INODE_ITEM_KEY;
3783         location->offset = 0;
3784         err = 0;
3785 out:
3786         btrfs_free_path(path);
3787         return err;
3788 }
3789
3790 static void inode_tree_add(struct inode *inode)
3791 {
3792         struct btrfs_root *root = BTRFS_I(inode)->root;
3793         struct btrfs_inode *entry;
3794         struct rb_node **p;
3795         struct rb_node *parent;
3796         u64 ino = btrfs_ino(inode);
3797 again:
3798         p = &root->inode_tree.rb_node;
3799         parent = NULL;
3800
3801         if (inode_unhashed(inode))
3802                 return;
3803
3804         spin_lock(&root->inode_lock);
3805         while (*p) {
3806                 parent = *p;
3807                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3808
3809                 if (ino < btrfs_ino(&entry->vfs_inode))
3810                         p = &parent->rb_left;
3811                 else if (ino > btrfs_ino(&entry->vfs_inode))
3812                         p = &parent->rb_right;
3813                 else {
3814                         WARN_ON(!(entry->vfs_inode.i_state &
3815                                   (I_WILL_FREE | I_FREEING)));
3816                         rb_erase(parent, &root->inode_tree);
3817                         RB_CLEAR_NODE(parent);
3818                         spin_unlock(&root->inode_lock);
3819                         goto again;
3820                 }
3821         }
3822         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3823         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3824         spin_unlock(&root->inode_lock);
3825 }
3826
3827 static void inode_tree_del(struct inode *inode)
3828 {
3829         struct btrfs_root *root = BTRFS_I(inode)->root;
3830         int empty = 0;
3831
3832         spin_lock(&root->inode_lock);
3833         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3834                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3835                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3836                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3837         }
3838         spin_unlock(&root->inode_lock);
3839
3840         /*
3841          * Free space cache has inodes in the tree root, but the tree root has a
3842          * root_refs of 0, so this could end up dropping the tree root as a
3843          * snapshot, so we need the extra !root->fs_info->tree_root check to
3844          * make sure we don't drop it.
3845          */
3846         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3847             root != root->fs_info->tree_root) {
3848                 synchronize_srcu(&root->fs_info->subvol_srcu);
3849                 spin_lock(&root->inode_lock);
3850                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3851                 spin_unlock(&root->inode_lock);
3852                 if (empty)
3853                         btrfs_add_dead_root(root);
3854         }
3855 }
3856
3857 int btrfs_invalidate_inodes(struct btrfs_root *root)
3858 {
3859         struct rb_node *node;
3860         struct rb_node *prev;
3861         struct btrfs_inode *entry;
3862         struct inode *inode;
3863         u64 objectid = 0;
3864
3865         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3866
3867         spin_lock(&root->inode_lock);
3868 again:
3869         node = root->inode_tree.rb_node;
3870         prev = NULL;
3871         while (node) {
3872                 prev = node;
3873                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3874
3875                 if (objectid < btrfs_ino(&entry->vfs_inode))
3876                         node = node->rb_left;
3877                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3878                         node = node->rb_right;
3879                 else
3880                         break;
3881         }
3882         if (!node) {
3883                 while (prev) {
3884                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3885                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3886                                 node = prev;
3887                                 break;
3888                         }
3889                         prev = rb_next(prev);
3890                 }
3891         }
3892         while (node) {
3893                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3894                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3895                 inode = igrab(&entry->vfs_inode);
3896                 if (inode) {
3897                         spin_unlock(&root->inode_lock);
3898                         if (atomic_read(&inode->i_count) > 1)
3899                                 d_prune_aliases(inode);
3900                         /*
3901                          * btrfs_drop_inode will have it removed from
3902                          * the inode cache when its usage count
3903                          * hits zero.
3904                          */
3905                         iput(inode);
3906                         cond_resched();
3907                         spin_lock(&root->inode_lock);
3908                         goto again;
3909                 }
3910
3911                 if (cond_resched_lock(&root->inode_lock))
3912                         goto again;
3913
3914                 node = rb_next(node);
3915         }
3916         spin_unlock(&root->inode_lock);
3917         return 0;
3918 }
3919
3920 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3921 {
3922         struct btrfs_iget_args *args = p;
3923         inode->i_ino = args->ino;
3924         BTRFS_I(inode)->root = args->root;
3925         btrfs_set_inode_space_info(args->root, inode);
3926         return 0;
3927 }
3928
3929 static int btrfs_find_actor(struct inode *inode, void *opaque)
3930 {
3931         struct btrfs_iget_args *args = opaque;
3932         return args->ino == btrfs_ino(inode) &&
3933                 args->root == BTRFS_I(inode)->root;
3934 }
3935
3936 static struct inode *btrfs_iget_locked(struct super_block *s,
3937                                        u64 objectid,
3938                                        struct btrfs_root *root)
3939 {
3940         struct inode *inode;
3941         struct btrfs_iget_args args;
3942         args.ino = objectid;
3943         args.root = root;
3944
3945         inode = iget5_locked(s, objectid, btrfs_find_actor,
3946                              btrfs_init_locked_inode,
3947                              (void *)&args);
3948         return inode;
3949 }
3950
3951 /* Get an inode object given its location and corresponding root.
3952  * Returns in *is_new if the inode was read from disk
3953  */
3954 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3955                          struct btrfs_root *root, int *new)
3956 {
3957         struct inode *inode;
3958
3959         inode = btrfs_iget_locked(s, location->objectid, root);
3960         if (!inode)
3961                 return ERR_PTR(-ENOMEM);
3962
3963         if (inode->i_state & I_NEW) {
3964                 BTRFS_I(inode)->root = root;
3965                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3966                 btrfs_read_locked_inode(inode);
3967                 inode_tree_add(inode);
3968                 unlock_new_inode(inode);
3969                 if (new)
3970                         *new = 1;
3971         }
3972
3973         return inode;
3974 }
3975
3976 static struct inode *new_simple_dir(struct super_block *s,
3977                                     struct btrfs_key *key,
3978                                     struct btrfs_root *root)
3979 {
3980         struct inode *inode = new_inode(s);
3981
3982         if (!inode)
3983                 return ERR_PTR(-ENOMEM);
3984
3985         BTRFS_I(inode)->root = root;
3986         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3987         BTRFS_I(inode)->dummy_inode = 1;
3988
3989         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3990         inode->i_op = &simple_dir_inode_operations;
3991         inode->i_fop = &simple_dir_operations;
3992         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3993         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3994
3995         return inode;
3996 }
3997
3998 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3999 {
4000         struct inode *inode;
4001         struct btrfs_root *root = BTRFS_I(dir)->root;
4002         struct btrfs_root *sub_root = root;
4003         struct btrfs_key location;
4004         int index;
4005         int ret;
4006
4007         if (dentry->d_name.len > BTRFS_NAME_LEN)
4008                 return ERR_PTR(-ENAMETOOLONG);
4009
4010         ret = btrfs_inode_by_name(dir, dentry, &location);
4011
4012         if (ret < 0)
4013                 return ERR_PTR(ret);
4014
4015         if (location.objectid == 0)
4016                 return NULL;
4017
4018         if (location.type == BTRFS_INODE_ITEM_KEY) {
4019                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4020                 return inode;
4021         }
4022
4023         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4024
4025         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4026         ret = fixup_tree_root_location(root, dir, dentry,
4027                                        &location, &sub_root);
4028         if (ret < 0) {
4029                 if (ret != -ENOENT)
4030                         inode = ERR_PTR(ret);
4031                 else
4032                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4033         } else {
4034                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4035         }
4036         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4037
4038         if (!IS_ERR(inode) && root != sub_root) {
4039                 down_read(&root->fs_info->cleanup_work_sem);
4040                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4041                         ret = btrfs_orphan_cleanup(sub_root);
4042                 up_read(&root->fs_info->cleanup_work_sem);
4043                 if (ret)
4044                         inode = ERR_PTR(ret);
4045         }
4046
4047         return inode;
4048 }
4049
4050 static int btrfs_dentry_delete(const struct dentry *dentry)
4051 {
4052         struct btrfs_root *root;
4053
4054         if (!dentry->d_inode && !IS_ROOT(dentry))
4055                 dentry = dentry->d_parent;
4056
4057         if (dentry->d_inode) {
4058                 root = BTRFS_I(dentry->d_inode)->root;
4059                 if (btrfs_root_refs(&root->root_item) == 0)
4060                         return 1;
4061         }
4062         return 0;
4063 }
4064
4065 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4066                                    struct nameidata *nd)
4067 {
4068         struct inode *inode;
4069
4070         inode = btrfs_lookup_dentry(dir, dentry);
4071         if (IS_ERR(inode))
4072                 return ERR_CAST(inode);
4073
4074         return d_splice_alias(inode, dentry);
4075 }
4076
4077 unsigned char btrfs_filetype_table[] = {
4078         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4079 };
4080
4081 static int btrfs_real_readdir(struct file *filp, void *dirent,
4082                               filldir_t filldir)
4083 {
4084         struct inode *inode = filp->f_dentry->d_inode;
4085         struct btrfs_root *root = BTRFS_I(inode)->root;
4086         struct btrfs_item *item;
4087         struct btrfs_dir_item *di;
4088         struct btrfs_key key;
4089         struct btrfs_key found_key;
4090         struct btrfs_path *path;
4091         struct list_head ins_list;
4092         struct list_head del_list;
4093         int ret;
4094         struct extent_buffer *leaf;
4095         int slot;
4096         unsigned char d_type;
4097         int over = 0;
4098         u32 di_cur;
4099         u32 di_total;
4100         u32 di_len;
4101         int key_type = BTRFS_DIR_INDEX_KEY;
4102         char tmp_name[32];
4103         char *name_ptr;
4104         int name_len;
4105         int is_curr = 0;        /* filp->f_pos points to the current index? */
4106
4107         /* FIXME, use a real flag for deciding about the key type */
4108         if (root->fs_info->tree_root == root)
4109                 key_type = BTRFS_DIR_ITEM_KEY;
4110
4111         /* special case for "." */
4112         if (filp->f_pos == 0) {
4113                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4114                 if (over)
4115                         return 0;
4116                 filp->f_pos = 1;
4117         }
4118         /* special case for .., just use the back ref */
4119         if (filp->f_pos == 1) {
4120                 u64 pino = parent_ino(filp->f_path.dentry);
4121                 over = filldir(dirent, "..", 2,
4122                                2, pino, DT_DIR);
4123                 if (over)
4124                         return 0;
4125                 filp->f_pos = 2;
4126         }
4127         path = btrfs_alloc_path();
4128         if (!path)
4129                 return -ENOMEM;
4130
4131         path->reada = 1;
4132
4133         if (key_type == BTRFS_DIR_INDEX_KEY) {
4134                 INIT_LIST_HEAD(&ins_list);
4135                 INIT_LIST_HEAD(&del_list);
4136                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4137         }
4138
4139         btrfs_set_key_type(&key, key_type);
4140         key.offset = filp->f_pos;
4141         key.objectid = btrfs_ino(inode);
4142
4143         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4144         if (ret < 0)
4145                 goto err;
4146
4147         while (1) {
4148                 leaf = path->nodes[0];
4149                 slot = path->slots[0];
4150                 if (slot >= btrfs_header_nritems(leaf)) {
4151                         ret = btrfs_next_leaf(root, path);
4152                         if (ret < 0)
4153                                 goto err;
4154                         else if (ret > 0)
4155                                 break;
4156                         continue;
4157                 }
4158
4159                 item = btrfs_item_nr(leaf, slot);
4160                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4161
4162                 if (found_key.objectid != key.objectid)
4163                         break;
4164                 if (btrfs_key_type(&found_key) != key_type)
4165                         break;
4166                 if (found_key.offset < filp->f_pos)
4167                         goto next;
4168                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4169                     btrfs_should_delete_dir_index(&del_list,
4170                                                   found_key.offset))
4171                         goto next;
4172
4173                 filp->f_pos = found_key.offset;
4174                 is_curr = 1;
4175
4176                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4177                 di_cur = 0;
4178                 di_total = btrfs_item_size(leaf, item);
4179
4180                 while (di_cur < di_total) {
4181                         struct btrfs_key location;
4182
4183                         if (verify_dir_item(root, leaf, di))
4184                                 break;
4185
4186                         name_len = btrfs_dir_name_len(leaf, di);
4187                         if (name_len <= sizeof(tmp_name)) {
4188                                 name_ptr = tmp_name;
4189                         } else {
4190                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4191                                 if (!name_ptr) {
4192                                         ret = -ENOMEM;
4193                                         goto err;
4194                                 }
4195                         }
4196                         read_extent_buffer(leaf, name_ptr,
4197                                            (unsigned long)(di + 1), name_len);
4198
4199                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4200                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4201
4202                         /* is this a reference to our own snapshot? If so
4203                          * skip it
4204                          */
4205                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4206                             location.objectid == root->root_key.objectid) {
4207                                 over = 0;
4208                                 goto skip;
4209                         }
4210                         over = filldir(dirent, name_ptr, name_len,
4211                                        found_key.offset, location.objectid,
4212                                        d_type);
4213
4214 skip:
4215                         if (name_ptr != tmp_name)
4216                                 kfree(name_ptr);
4217
4218                         if (over)
4219                                 goto nopos;
4220                         di_len = btrfs_dir_name_len(leaf, di) +
4221                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4222                         di_cur += di_len;
4223                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4224                 }
4225 next:
4226                 path->slots[0]++;
4227         }
4228
4229         if (key_type == BTRFS_DIR_INDEX_KEY) {
4230                 if (is_curr)
4231                         filp->f_pos++;
4232                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4233                                                       &ins_list);
4234                 if (ret)
4235                         goto nopos;
4236         }
4237
4238         /* Reached end of directory/root. Bump pos past the last item. */
4239         if (key_type == BTRFS_DIR_INDEX_KEY)
4240                 /*
4241                  * 32-bit glibc will use getdents64, but then strtol -
4242                  * so the last number we can serve is this.
4243                  */
4244                 filp->f_pos = 0x7fffffff;
4245         else
4246                 filp->f_pos++;
4247 nopos:
4248         ret = 0;
4249 err:
4250         if (key_type == BTRFS_DIR_INDEX_KEY)
4251                 btrfs_put_delayed_items(&ins_list, &del_list);
4252         btrfs_free_path(path);
4253         return ret;
4254 }
4255
4256 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4257 {
4258         struct btrfs_root *root = BTRFS_I(inode)->root;
4259         struct btrfs_trans_handle *trans;
4260         int ret = 0;
4261         bool nolock = false;
4262
4263         if (BTRFS_I(inode)->dummy_inode)
4264                 return 0;
4265
4266         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4267                 nolock = true;
4268
4269         if (wbc->sync_mode == WB_SYNC_ALL) {
4270                 if (nolock)
4271                         trans = btrfs_join_transaction_nolock(root);
4272                 else
4273                         trans = btrfs_join_transaction(root);
4274                 if (IS_ERR(trans))
4275                         return PTR_ERR(trans);
4276                 if (nolock)
4277                         ret = btrfs_end_transaction_nolock(trans, root);
4278                 else
4279                         ret = btrfs_commit_transaction(trans, root);
4280         }
4281         return ret;
4282 }
4283
4284 /*
4285  * This is somewhat expensive, updating the tree every time the
4286  * inode changes.  But, it is most likely to find the inode in cache.
4287  * FIXME, needs more benchmarking...there are no reasons other than performance
4288  * to keep or drop this code.
4289  */
4290 void btrfs_dirty_inode(struct inode *inode)
4291 {
4292         struct btrfs_root *root = BTRFS_I(inode)->root;
4293         struct btrfs_trans_handle *trans;
4294         int ret;
4295
4296         if (BTRFS_I(inode)->dummy_inode)
4297                 return;
4298
4299         trans = btrfs_join_transaction(root);
4300         BUG_ON(IS_ERR(trans));
4301
4302         ret = btrfs_update_inode(trans, root, inode);
4303         if (ret && ret == -ENOSPC) {
4304                 /* whoops, lets try again with the full transaction */
4305                 btrfs_end_transaction(trans, root);
4306                 trans = btrfs_start_transaction(root, 1);
4307                 if (IS_ERR(trans)) {
4308                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4309                                        "dirty  inode %llu error %ld\n",
4310                                        (unsigned long long)btrfs_ino(inode),
4311                                        PTR_ERR(trans));
4312                         return;
4313                 }
4314
4315                 ret = btrfs_update_inode(trans, root, inode);
4316                 if (ret) {
4317                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4318                                        "dirty  inode %llu error %d\n",
4319                                        (unsigned long long)btrfs_ino(inode),
4320                                        ret);
4321                 }
4322         }
4323         btrfs_end_transaction(trans, root);
4324         if (BTRFS_I(inode)->delayed_node)
4325                 btrfs_balance_delayed_items(root);
4326 }
4327
4328 /*
4329  * find the highest existing sequence number in a directory
4330  * and then set the in-memory index_cnt variable to reflect
4331  * free sequence numbers
4332  */
4333 static int btrfs_set_inode_index_count(struct inode *inode)
4334 {
4335         struct btrfs_root *root = BTRFS_I(inode)->root;
4336         struct btrfs_key key, found_key;
4337         struct btrfs_path *path;
4338         struct extent_buffer *leaf;
4339         int ret;
4340
4341         key.objectid = btrfs_ino(inode);
4342         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4343         key.offset = (u64)-1;
4344
4345         path = btrfs_alloc_path();
4346         if (!path)
4347                 return -ENOMEM;
4348
4349         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4350         if (ret < 0)
4351                 goto out;
4352         /* FIXME: we should be able to handle this */
4353         if (ret == 0)
4354                 goto out;
4355         ret = 0;
4356
4357         /*
4358          * MAGIC NUMBER EXPLANATION:
4359          * since we search a directory based on f_pos we have to start at 2
4360          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4361          * else has to start at 2
4362          */
4363         if (path->slots[0] == 0) {
4364                 BTRFS_I(inode)->index_cnt = 2;
4365                 goto out;
4366         }
4367
4368         path->slots[0]--;
4369
4370         leaf = path->nodes[0];
4371         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4372
4373         if (found_key.objectid != btrfs_ino(inode) ||
4374             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4375                 BTRFS_I(inode)->index_cnt = 2;
4376                 goto out;
4377         }
4378
4379         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4380 out:
4381         btrfs_free_path(path);
4382         return ret;
4383 }
4384
4385 /*
4386  * helper to find a free sequence number in a given directory.  This current
4387  * code is very simple, later versions will do smarter things in the btree
4388  */
4389 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4390 {
4391         int ret = 0;
4392
4393         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4394                 ret = btrfs_inode_delayed_dir_index_count(dir);
4395                 if (ret) {
4396                         ret = btrfs_set_inode_index_count(dir);
4397                         if (ret)
4398                                 return ret;
4399                 }
4400         }
4401
4402         *index = BTRFS_I(dir)->index_cnt;
4403         BTRFS_I(dir)->index_cnt++;
4404
4405         return ret;
4406 }
4407
4408 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4409                                      struct btrfs_root *root,
4410                                      struct inode *dir,
4411                                      const char *name, int name_len,
4412                                      u64 ref_objectid, u64 objectid, int mode,
4413                                      u64 *index)
4414 {
4415         struct inode *inode;
4416         struct btrfs_inode_item *inode_item;
4417         struct btrfs_key *location;
4418         struct btrfs_path *path;
4419         struct btrfs_inode_ref *ref;
4420         struct btrfs_key key[2];
4421         u32 sizes[2];
4422         unsigned long ptr;
4423         int ret;
4424         int owner;
4425
4426         path = btrfs_alloc_path();
4427         BUG_ON(!path);
4428
4429         inode = new_inode(root->fs_info->sb);
4430         if (!inode) {
4431                 btrfs_free_path(path);
4432                 return ERR_PTR(-ENOMEM);
4433         }
4434
4435         /*
4436          * we have to initialize this early, so we can reclaim the inode
4437          * number if we fail afterwards in this function.
4438          */
4439         inode->i_ino = objectid;
4440
4441         if (dir) {
4442                 trace_btrfs_inode_request(dir);
4443
4444                 ret = btrfs_set_inode_index(dir, index);
4445                 if (ret) {
4446                         btrfs_free_path(path);
4447                         iput(inode);
4448                         return ERR_PTR(ret);
4449                 }
4450         }
4451         /*
4452          * index_cnt is ignored for everything but a dir,
4453          * btrfs_get_inode_index_count has an explanation for the magic
4454          * number
4455          */
4456         BTRFS_I(inode)->index_cnt = 2;
4457         BTRFS_I(inode)->root = root;
4458         BTRFS_I(inode)->generation = trans->transid;
4459         inode->i_generation = BTRFS_I(inode)->generation;
4460         btrfs_set_inode_space_info(root, inode);
4461
4462         if (mode & S_IFDIR)
4463                 owner = 0;
4464         else
4465                 owner = 1;
4466
4467         key[0].objectid = objectid;
4468         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4469         key[0].offset = 0;
4470
4471         key[1].objectid = objectid;
4472         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4473         key[1].offset = ref_objectid;
4474
4475         sizes[0] = sizeof(struct btrfs_inode_item);
4476         sizes[1] = name_len + sizeof(*ref);
4477
4478         path->leave_spinning = 1;
4479         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4480         if (ret != 0)
4481                 goto fail;
4482
4483         inode_init_owner(inode, dir, mode);
4484         inode_set_bytes(inode, 0);
4485         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4486         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4487                                   struct btrfs_inode_item);
4488         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4489
4490         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4491                              struct btrfs_inode_ref);
4492         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4493         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4494         ptr = (unsigned long)(ref + 1);
4495         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4496
4497         btrfs_mark_buffer_dirty(path->nodes[0]);
4498         btrfs_free_path(path);
4499
4500         location = &BTRFS_I(inode)->location;
4501         location->objectid = objectid;
4502         location->offset = 0;
4503         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4504
4505         btrfs_inherit_iflags(inode, dir);
4506
4507         if ((mode & S_IFREG)) {
4508                 if (btrfs_test_opt(root, NODATASUM))
4509                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4510                 if (btrfs_test_opt(root, NODATACOW) ||
4511                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4512                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4513         }
4514
4515         insert_inode_hash(inode);
4516         inode_tree_add(inode);
4517
4518         trace_btrfs_inode_new(inode);
4519         btrfs_set_inode_last_trans(trans, inode);
4520
4521         return inode;
4522 fail:
4523         if (dir)
4524                 BTRFS_I(dir)->index_cnt--;
4525         btrfs_free_path(path);
4526         iput(inode);
4527         return ERR_PTR(ret);
4528 }
4529
4530 static inline u8 btrfs_inode_type(struct inode *inode)
4531 {
4532         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4533 }
4534
4535 /*
4536  * utility function to add 'inode' into 'parent_inode' with
4537  * a give name and a given sequence number.
4538  * if 'add_backref' is true, also insert a backref from the
4539  * inode to the parent directory.
4540  */
4541 int btrfs_add_link(struct btrfs_trans_handle *trans,
4542                    struct inode *parent_inode, struct inode *inode,
4543                    const char *name, int name_len, int add_backref, u64 index)
4544 {
4545         int ret = 0;
4546         struct btrfs_key key;
4547         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4548         u64 ino = btrfs_ino(inode);
4549         u64 parent_ino = btrfs_ino(parent_inode);
4550
4551         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4552                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4553         } else {
4554                 key.objectid = ino;
4555                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4556                 key.offset = 0;
4557         }
4558
4559         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4560                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4561                                          key.objectid, root->root_key.objectid,
4562                                          parent_ino, index, name, name_len);
4563         } else if (add_backref) {
4564                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4565                                              parent_ino, index);
4566         }
4567
4568         if (ret == 0) {
4569                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4570                                             parent_inode, &key,
4571                                             btrfs_inode_type(inode), index);
4572                 BUG_ON(ret);
4573
4574                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4575                                    name_len * 2);
4576                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4577                 ret = btrfs_update_inode(trans, root, parent_inode);
4578         }
4579         return ret;
4580 }
4581
4582 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4583                             struct inode *dir, struct dentry *dentry,
4584                             struct inode *inode, int backref, u64 index)
4585 {
4586         int err = btrfs_add_link(trans, dir, inode,
4587                                  dentry->d_name.name, dentry->d_name.len,
4588                                  backref, index);
4589         if (!err) {
4590                 d_instantiate(dentry, inode);
4591                 return 0;
4592         }
4593         if (err > 0)
4594                 err = -EEXIST;
4595         return err;
4596 }
4597
4598 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4599                         int mode, dev_t rdev)
4600 {
4601         struct btrfs_trans_handle *trans;
4602         struct btrfs_root *root = BTRFS_I(dir)->root;
4603         struct inode *inode = NULL;
4604         int err;
4605         int drop_inode = 0;
4606         u64 objectid;
4607         unsigned long nr = 0;
4608         u64 index = 0;
4609
4610         if (!new_valid_dev(rdev))
4611                 return -EINVAL;
4612
4613         /*
4614          * 2 for inode item and ref
4615          * 2 for dir items
4616          * 1 for xattr if selinux is on
4617          */
4618         trans = btrfs_start_transaction(root, 5);
4619         if (IS_ERR(trans))
4620                 return PTR_ERR(trans);
4621
4622         err = btrfs_find_free_ino(root, &objectid);
4623         if (err)
4624                 goto out_unlock;
4625
4626         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4627                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4628                                 mode, &index);
4629         if (IS_ERR(inode)) {
4630                 err = PTR_ERR(inode);
4631                 goto out_unlock;
4632         }
4633
4634         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4635         if (err) {
4636                 drop_inode = 1;
4637                 goto out_unlock;
4638         }
4639
4640         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4641         if (err)
4642                 drop_inode = 1;
4643         else {
4644                 inode->i_op = &btrfs_special_inode_operations;
4645                 init_special_inode(inode, inode->i_mode, rdev);
4646                 btrfs_update_inode(trans, root, inode);
4647         }
4648 out_unlock:
4649         nr = trans->blocks_used;
4650         btrfs_end_transaction_throttle(trans, root);
4651         btrfs_btree_balance_dirty(root, nr);
4652         if (drop_inode) {
4653                 inode_dec_link_count(inode);
4654                 iput(inode);
4655         }
4656         return err;
4657 }
4658
4659 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4660                         int mode, struct nameidata *nd)
4661 {
4662         struct btrfs_trans_handle *trans;
4663         struct btrfs_root *root = BTRFS_I(dir)->root;
4664         struct inode *inode = NULL;
4665         int drop_inode = 0;
4666         int err;
4667         unsigned long nr = 0;
4668         u64 objectid;
4669         u64 index = 0;
4670
4671         /*
4672          * 2 for inode item and ref
4673          * 2 for dir items
4674          * 1 for xattr if selinux is on
4675          */
4676         trans = btrfs_start_transaction(root, 5);
4677         if (IS_ERR(trans))
4678                 return PTR_ERR(trans);
4679
4680         err = btrfs_find_free_ino(root, &objectid);
4681         if (err)
4682                 goto out_unlock;
4683
4684         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4685                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4686                                 mode, &index);
4687         if (IS_ERR(inode)) {
4688                 err = PTR_ERR(inode);
4689                 goto out_unlock;
4690         }
4691
4692         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4693         if (err) {
4694                 drop_inode = 1;
4695                 goto out_unlock;
4696         }
4697
4698         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4699         if (err)
4700                 drop_inode = 1;
4701         else {
4702                 inode->i_mapping->a_ops = &btrfs_aops;
4703                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4704                 inode->i_fop = &btrfs_file_operations;
4705                 inode->i_op = &btrfs_file_inode_operations;
4706                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4707         }
4708 out_unlock:
4709         nr = trans->blocks_used;
4710         btrfs_end_transaction_throttle(trans, root);
4711         if (drop_inode) {
4712                 inode_dec_link_count(inode);
4713                 iput(inode);
4714         }
4715         btrfs_btree_balance_dirty(root, nr);
4716         return err;
4717 }
4718
4719 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4720                       struct dentry *dentry)
4721 {
4722         struct btrfs_trans_handle *trans;
4723         struct btrfs_root *root = BTRFS_I(dir)->root;
4724         struct inode *inode = old_dentry->d_inode;
4725         u64 index;
4726         unsigned long nr = 0;
4727         int err;
4728         int drop_inode = 0;
4729
4730         /* do not allow sys_link's with other subvols of the same device */
4731         if (root->objectid != BTRFS_I(inode)->root->objectid)
4732                 return -EXDEV;
4733
4734         if (inode->i_nlink == ~0U)
4735                 return -EMLINK;
4736
4737         err = btrfs_set_inode_index(dir, &index);
4738         if (err)
4739                 goto fail;
4740
4741         /*
4742          * 2 items for inode and inode ref
4743          * 2 items for dir items
4744          * 1 item for parent inode
4745          */
4746         trans = btrfs_start_transaction(root, 5);
4747         if (IS_ERR(trans)) {
4748                 err = PTR_ERR(trans);
4749                 goto fail;
4750         }
4751
4752         btrfs_inc_nlink(inode);
4753         inode->i_ctime = CURRENT_TIME;
4754         ihold(inode);
4755
4756         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4757
4758         if (err) {
4759                 drop_inode = 1;
4760         } else {
4761                 struct dentry *parent = dget_parent(dentry);
4762                 err = btrfs_update_inode(trans, root, inode);
4763                 BUG_ON(err);
4764                 btrfs_log_new_name(trans, inode, NULL, parent);
4765                 dput(parent);
4766         }
4767
4768         nr = trans->blocks_used;
4769         btrfs_end_transaction_throttle(trans, root);
4770 fail:
4771         if (drop_inode) {
4772                 inode_dec_link_count(inode);
4773                 iput(inode);
4774         }
4775         btrfs_btree_balance_dirty(root, nr);
4776         return err;
4777 }
4778
4779 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4780 {
4781         struct inode *inode = NULL;
4782         struct btrfs_trans_handle *trans;
4783         struct btrfs_root *root = BTRFS_I(dir)->root;
4784         int err = 0;
4785         int drop_on_err = 0;
4786         u64 objectid = 0;
4787         u64 index = 0;
4788         unsigned long nr = 1;
4789
4790         /*
4791          * 2 items for inode and ref
4792          * 2 items for dir items
4793          * 1 for xattr if selinux is on
4794          */
4795         trans = btrfs_start_transaction(root, 5);
4796         if (IS_ERR(trans))
4797                 return PTR_ERR(trans);
4798
4799         err = btrfs_find_free_ino(root, &objectid);
4800         if (err)
4801                 goto out_fail;
4802
4803         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4804                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4805                                 S_IFDIR | mode, &index);
4806         if (IS_ERR(inode)) {
4807                 err = PTR_ERR(inode);
4808                 goto out_fail;
4809         }
4810
4811         drop_on_err = 1;
4812
4813         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4814         if (err)
4815                 goto out_fail;
4816
4817         inode->i_op = &btrfs_dir_inode_operations;
4818         inode->i_fop = &btrfs_dir_file_operations;
4819
4820         btrfs_i_size_write(inode, 0);
4821         err = btrfs_update_inode(trans, root, inode);
4822         if (err)
4823                 goto out_fail;
4824
4825         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4826                              dentry->d_name.len, 0, index);
4827         if (err)
4828                 goto out_fail;
4829
4830         d_instantiate(dentry, inode);
4831         drop_on_err = 0;
4832
4833 out_fail:
4834         nr = trans->blocks_used;
4835         btrfs_end_transaction_throttle(trans, root);
4836         if (drop_on_err)
4837                 iput(inode);
4838         btrfs_btree_balance_dirty(root, nr);
4839         return err;
4840 }
4841
4842 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4843  * and an extent that you want to insert, deal with overlap and insert
4844  * the new extent into the tree.
4845  */
4846 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4847                                 struct extent_map *existing,
4848                                 struct extent_map *em,
4849                                 u64 map_start, u64 map_len)
4850 {
4851         u64 start_diff;
4852
4853         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4854         start_diff = map_start - em->start;
4855         em->start = map_start;
4856         em->len = map_len;
4857         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4858             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4859                 em->block_start += start_diff;
4860                 em->block_len -= start_diff;
4861         }
4862         return add_extent_mapping(em_tree, em);
4863 }
4864
4865 static noinline int uncompress_inline(struct btrfs_path *path,
4866                                       struct inode *inode, struct page *page,
4867                                       size_t pg_offset, u64 extent_offset,
4868                                       struct btrfs_file_extent_item *item)
4869 {
4870         int ret;
4871         struct extent_buffer *leaf = path->nodes[0];
4872         char *tmp;
4873         size_t max_size;
4874         unsigned long inline_size;
4875         unsigned long ptr;
4876         int compress_type;
4877
4878         WARN_ON(pg_offset != 0);
4879         compress_type = btrfs_file_extent_compression(leaf, item);
4880         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4881         inline_size = btrfs_file_extent_inline_item_len(leaf,
4882                                         btrfs_item_nr(leaf, path->slots[0]));
4883         tmp = kmalloc(inline_size, GFP_NOFS);
4884         if (!tmp)
4885                 return -ENOMEM;
4886         ptr = btrfs_file_extent_inline_start(item);
4887
4888         read_extent_buffer(leaf, tmp, ptr, inline_size);
4889
4890         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4891         ret = btrfs_decompress(compress_type, tmp, page,
4892                                extent_offset, inline_size, max_size);
4893         if (ret) {
4894                 char *kaddr = kmap_atomic(page, KM_USER0);
4895                 unsigned long copy_size = min_t(u64,
4896                                   PAGE_CACHE_SIZE - pg_offset,
4897                                   max_size - extent_offset);
4898                 memset(kaddr + pg_offset, 0, copy_size);
4899                 kunmap_atomic(kaddr, KM_USER0);
4900         }
4901         kfree(tmp);
4902         return 0;
4903 }
4904
4905 /*
4906  * a bit scary, this does extent mapping from logical file offset to the disk.
4907  * the ugly parts come from merging extents from the disk with the in-ram
4908  * representation.  This gets more complex because of the data=ordered code,
4909  * where the in-ram extents might be locked pending data=ordered completion.
4910  *
4911  * This also copies inline extents directly into the page.
4912  */
4913
4914 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4915                                     size_t pg_offset, u64 start, u64 len,
4916                                     int create)
4917 {
4918         int ret;
4919         int err = 0;
4920         u64 bytenr;
4921         u64 extent_start = 0;
4922         u64 extent_end = 0;
4923         u64 objectid = btrfs_ino(inode);
4924         u32 found_type;
4925         struct btrfs_path *path = NULL;
4926         struct btrfs_root *root = BTRFS_I(inode)->root;
4927         struct btrfs_file_extent_item *item;
4928         struct extent_buffer *leaf;
4929         struct btrfs_key found_key;
4930         struct extent_map *em = NULL;
4931         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4932         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4933         struct btrfs_trans_handle *trans = NULL;
4934         int compress_type;
4935
4936 again:
4937         read_lock(&em_tree->lock);
4938         em = lookup_extent_mapping(em_tree, start, len);
4939         if (em)
4940                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4941         read_unlock(&em_tree->lock);
4942
4943         if (em) {
4944                 if (em->start > start || em->start + em->len <= start)
4945                         free_extent_map(em);
4946                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4947                         free_extent_map(em);
4948                 else
4949                         goto out;
4950         }
4951         em = alloc_extent_map();
4952         if (!em) {
4953                 err = -ENOMEM;
4954                 goto out;
4955         }
4956         em->bdev = root->fs_info->fs_devices->latest_bdev;
4957         em->start = EXTENT_MAP_HOLE;
4958         em->orig_start = EXTENT_MAP_HOLE;
4959         em->len = (u64)-1;
4960         em->block_len = (u64)-1;
4961
4962         if (!path) {
4963                 path = btrfs_alloc_path();
4964                 if (!path) {
4965                         err = -ENOMEM;
4966                         goto out;
4967                 }
4968                 /*
4969                  * Chances are we'll be called again, so go ahead and do
4970                  * readahead
4971                  */
4972                 path->reada = 1;
4973         }
4974
4975         ret = btrfs_lookup_file_extent(trans, root, path,
4976                                        objectid, start, trans != NULL);
4977         if (ret < 0) {
4978                 err = ret;
4979                 goto out;
4980         }
4981
4982         if (ret != 0) {
4983                 if (path->slots[0] == 0)
4984                         goto not_found;
4985                 path->slots[0]--;
4986         }
4987
4988         leaf = path->nodes[0];
4989         item = btrfs_item_ptr(leaf, path->slots[0],
4990                               struct btrfs_file_extent_item);
4991         /* are we inside the extent that was found? */
4992         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4993         found_type = btrfs_key_type(&found_key);
4994         if (found_key.objectid != objectid ||
4995             found_type != BTRFS_EXTENT_DATA_KEY) {
4996                 goto not_found;
4997         }
4998
4999         found_type = btrfs_file_extent_type(leaf, item);
5000         extent_start = found_key.offset;
5001         compress_type = btrfs_file_extent_compression(leaf, item);
5002         if (found_type == BTRFS_FILE_EXTENT_REG ||
5003             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5004                 extent_end = extent_start +
5005                        btrfs_file_extent_num_bytes(leaf, item);
5006         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5007                 size_t size;
5008                 size = btrfs_file_extent_inline_len(leaf, item);
5009                 extent_end = (extent_start + size + root->sectorsize - 1) &
5010                         ~((u64)root->sectorsize - 1);
5011         }
5012
5013         if (start >= extent_end) {
5014                 path->slots[0]++;
5015                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5016                         ret = btrfs_next_leaf(root, path);
5017                         if (ret < 0) {
5018                                 err = ret;
5019                                 goto out;
5020                         }
5021                         if (ret > 0)
5022                                 goto not_found;
5023                         leaf = path->nodes[0];
5024                 }
5025                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5026                 if (found_key.objectid != objectid ||
5027                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5028                         goto not_found;
5029                 if (start + len <= found_key.offset)
5030                         goto not_found;
5031                 em->start = start;
5032                 em->len = found_key.offset - start;
5033                 goto not_found_em;
5034         }
5035
5036         if (found_type == BTRFS_FILE_EXTENT_REG ||
5037             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5038                 em->start = extent_start;
5039                 em->len = extent_end - extent_start;
5040                 em->orig_start = extent_start -
5041                                  btrfs_file_extent_offset(leaf, item);
5042                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5043                 if (bytenr == 0) {
5044                         em->block_start = EXTENT_MAP_HOLE;
5045                         goto insert;
5046                 }
5047                 if (compress_type != BTRFS_COMPRESS_NONE) {
5048                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5049                         em->compress_type = compress_type;
5050                         em->block_start = bytenr;
5051                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5052                                                                          item);
5053                 } else {
5054                         bytenr += btrfs_file_extent_offset(leaf, item);
5055                         em->block_start = bytenr;
5056                         em->block_len = em->len;
5057                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5058                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5059                 }
5060                 goto insert;
5061         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5062                 unsigned long ptr;
5063                 char *map;
5064                 size_t size;
5065                 size_t extent_offset;
5066                 size_t copy_size;
5067
5068                 em->block_start = EXTENT_MAP_INLINE;
5069                 if (!page || create) {
5070                         em->start = extent_start;
5071                         em->len = extent_end - extent_start;
5072                         goto out;
5073                 }
5074
5075                 size = btrfs_file_extent_inline_len(leaf, item);
5076                 extent_offset = page_offset(page) + pg_offset - extent_start;
5077                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5078                                 size - extent_offset);
5079                 em->start = extent_start + extent_offset;
5080                 em->len = (copy_size + root->sectorsize - 1) &
5081                         ~((u64)root->sectorsize - 1);
5082                 em->orig_start = EXTENT_MAP_INLINE;
5083                 if (compress_type) {
5084                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5085                         em->compress_type = compress_type;
5086                 }
5087                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5088                 if (create == 0 && !PageUptodate(page)) {
5089                         if (btrfs_file_extent_compression(leaf, item) !=
5090                             BTRFS_COMPRESS_NONE) {
5091                                 ret = uncompress_inline(path, inode, page,
5092                                                         pg_offset,
5093                                                         extent_offset, item);
5094                                 BUG_ON(ret);
5095                         } else {
5096                                 map = kmap(page);
5097                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5098                                                    copy_size);
5099                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5100                                         memset(map + pg_offset + copy_size, 0,
5101                                                PAGE_CACHE_SIZE - pg_offset -
5102                                                copy_size);
5103                                 }
5104                                 kunmap(page);
5105                         }
5106                         flush_dcache_page(page);
5107                 } else if (create && PageUptodate(page)) {
5108                         WARN_ON(1);
5109                         if (!trans) {
5110                                 kunmap(page);
5111                                 free_extent_map(em);
5112                                 em = NULL;
5113
5114                                 btrfs_release_path(path);
5115                                 trans = btrfs_join_transaction(root);
5116
5117                                 if (IS_ERR(trans))
5118                                         return ERR_CAST(trans);
5119                                 goto again;
5120                         }
5121                         map = kmap(page);
5122                         write_extent_buffer(leaf, map + pg_offset, ptr,
5123                                             copy_size);
5124                         kunmap(page);
5125                         btrfs_mark_buffer_dirty(leaf);
5126                 }
5127                 set_extent_uptodate(io_tree, em->start,
5128                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5129                 goto insert;
5130         } else {
5131                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5132                 WARN_ON(1);
5133         }
5134 not_found:
5135         em->start = start;
5136         em->len = len;
5137 not_found_em:
5138         em->block_start = EXTENT_MAP_HOLE;
5139         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5140 insert:
5141         btrfs_release_path(path);
5142         if (em->start > start || extent_map_end(em) <= start) {
5143                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5144                        "[%llu %llu]\n", (unsigned long long)em->start,
5145                        (unsigned long long)em->len,
5146                        (unsigned long long)start,
5147                        (unsigned long long)len);
5148                 err = -EIO;
5149                 goto out;
5150         }
5151
5152         err = 0;
5153         write_lock(&em_tree->lock);
5154         ret = add_extent_mapping(em_tree, em);
5155         /* it is possible that someone inserted the extent into the tree
5156          * while we had the lock dropped.  It is also possible that
5157          * an overlapping map exists in the tree
5158          */
5159         if (ret == -EEXIST) {
5160                 struct extent_map *existing;
5161
5162                 ret = 0;
5163
5164                 existing = lookup_extent_mapping(em_tree, start, len);
5165                 if (existing && (existing->start > start ||
5166                     existing->start + existing->len <= start)) {
5167                         free_extent_map(existing);
5168                         existing = NULL;
5169                 }
5170                 if (!existing) {
5171                         existing = lookup_extent_mapping(em_tree, em->start,
5172                                                          em->len);
5173                         if (existing) {
5174                                 err = merge_extent_mapping(em_tree, existing,
5175                                                            em, start,
5176                                                            root->sectorsize);
5177                                 free_extent_map(existing);
5178                                 if (err) {
5179                                         free_extent_map(em);
5180                                         em = NULL;
5181                                 }
5182                         } else {
5183                                 err = -EIO;
5184                                 free_extent_map(em);
5185                                 em = NULL;
5186                         }
5187                 } else {
5188                         free_extent_map(em);
5189                         em = existing;
5190                         err = 0;
5191                 }
5192         }
5193         write_unlock(&em_tree->lock);
5194 out:
5195
5196         trace_btrfs_get_extent(root, em);
5197
5198         if (path)
5199                 btrfs_free_path(path);
5200         if (trans) {
5201                 ret = btrfs_end_transaction(trans, root);
5202                 if (!err)
5203                         err = ret;
5204         }
5205         if (err) {
5206                 free_extent_map(em);
5207                 return ERR_PTR(err);
5208         }
5209         return em;
5210 }
5211
5212 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5213                                            size_t pg_offset, u64 start, u64 len,
5214                                            int create)
5215 {
5216         struct extent_map *em;
5217         struct extent_map *hole_em = NULL;
5218         u64 range_start = start;
5219         u64 end;
5220         u64 found;
5221         u64 found_end;
5222         int err = 0;
5223
5224         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5225         if (IS_ERR(em))
5226                 return em;
5227         if (em) {
5228                 /*
5229                  * if our em maps to a hole, there might
5230                  * actually be delalloc bytes behind it
5231                  */
5232                 if (em->block_start != EXTENT_MAP_HOLE)
5233                         return em;
5234                 else
5235                         hole_em = em;
5236         }
5237
5238         /* check to see if we've wrapped (len == -1 or similar) */
5239         end = start + len;
5240         if (end < start)
5241                 end = (u64)-1;
5242         else
5243                 end -= 1;
5244
5245         em = NULL;
5246
5247         /* ok, we didn't find anything, lets look for delalloc */
5248         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5249                                  end, len, EXTENT_DELALLOC, 1);
5250         found_end = range_start + found;
5251         if (found_end < range_start)
5252                 found_end = (u64)-1;
5253
5254         /*
5255          * we didn't find anything useful, return
5256          * the original results from get_extent()
5257          */
5258         if (range_start > end || found_end <= start) {
5259                 em = hole_em;
5260                 hole_em = NULL;
5261                 goto out;
5262         }
5263
5264         /* adjust the range_start to make sure it doesn't
5265          * go backwards from the start they passed in
5266          */
5267         range_start = max(start,range_start);
5268         found = found_end - range_start;
5269
5270         if (found > 0) {
5271                 u64 hole_start = start;
5272                 u64 hole_len = len;
5273
5274                 em = alloc_extent_map();
5275                 if (!em) {
5276                         err = -ENOMEM;
5277                         goto out;
5278                 }
5279                 /*
5280                  * when btrfs_get_extent can't find anything it
5281                  * returns one huge hole
5282                  *
5283                  * make sure what it found really fits our range, and
5284                  * adjust to make sure it is based on the start from
5285                  * the caller
5286                  */
5287                 if (hole_em) {
5288                         u64 calc_end = extent_map_end(hole_em);
5289
5290                         if (calc_end <= start || (hole_em->start > end)) {
5291                                 free_extent_map(hole_em);
5292                                 hole_em = NULL;
5293                         } else {
5294                                 hole_start = max(hole_em->start, start);
5295                                 hole_len = calc_end - hole_start;
5296                         }
5297                 }
5298                 em->bdev = NULL;
5299                 if (hole_em && range_start > hole_start) {
5300                         /* our hole starts before our delalloc, so we
5301                          * have to return just the parts of the hole
5302                          * that go until  the delalloc starts
5303                          */
5304                         em->len = min(hole_len,
5305                                       range_start - hole_start);
5306                         em->start = hole_start;
5307                         em->orig_start = hole_start;
5308                         /*
5309                          * don't adjust block start at all,
5310                          * it is fixed at EXTENT_MAP_HOLE
5311                          */
5312                         em->block_start = hole_em->block_start;
5313                         em->block_len = hole_len;
5314                 } else {
5315                         em->start = range_start;
5316                         em->len = found;
5317                         em->orig_start = range_start;
5318                         em->block_start = EXTENT_MAP_DELALLOC;
5319                         em->block_len = found;
5320                 }
5321         } else if (hole_em) {
5322                 return hole_em;
5323         }
5324 out:
5325
5326         free_extent_map(hole_em);
5327         if (err) {
5328                 free_extent_map(em);
5329                 return ERR_PTR(err);
5330         }
5331         return em;
5332 }
5333
5334 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5335                                                   struct extent_map *em,
5336                                                   u64 start, u64 len)
5337 {
5338         struct btrfs_root *root = BTRFS_I(inode)->root;
5339         struct btrfs_trans_handle *trans;
5340         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5341         struct btrfs_key ins;
5342         u64 alloc_hint;
5343         int ret;
5344         bool insert = false;
5345
5346         /*
5347          * Ok if the extent map we looked up is a hole and is for the exact
5348          * range we want, there is no reason to allocate a new one, however if
5349          * it is not right then we need to free this one and drop the cache for
5350          * our range.
5351          */
5352         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5353             em->len != len) {
5354                 free_extent_map(em);
5355                 em = NULL;
5356                 insert = true;
5357                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5358         }
5359
5360         trans = btrfs_join_transaction(root);
5361         if (IS_ERR(trans))
5362                 return ERR_CAST(trans);
5363
5364         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5365                 btrfs_add_inode_defrag(trans, inode);
5366
5367         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5368
5369         alloc_hint = get_extent_allocation_hint(inode, start, len);
5370         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5371                                    alloc_hint, (u64)-1, &ins, 1);
5372         if (ret) {
5373                 em = ERR_PTR(ret);
5374                 goto out;
5375         }
5376
5377         if (!em) {
5378                 em = alloc_extent_map();
5379                 if (!em) {
5380                         em = ERR_PTR(-ENOMEM);
5381                         goto out;
5382                 }
5383         }
5384
5385         em->start = start;
5386         em->orig_start = em->start;
5387         em->len = ins.offset;
5388
5389         em->block_start = ins.objectid;
5390         em->block_len = ins.offset;
5391         em->bdev = root->fs_info->fs_devices->latest_bdev;
5392
5393         /*
5394          * We need to do this because if we're using the original em we searched
5395          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5396          */
5397         em->flags = 0;
5398         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5399
5400         while (insert) {
5401                 write_lock(&em_tree->lock);
5402                 ret = add_extent_mapping(em_tree, em);
5403                 write_unlock(&em_tree->lock);
5404                 if (ret != -EEXIST)
5405                         break;
5406                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5407         }
5408
5409         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5410                                            ins.offset, ins.offset, 0);
5411         if (ret) {
5412                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5413                 em = ERR_PTR(ret);
5414         }
5415 out:
5416         btrfs_end_transaction(trans, root);
5417         return em;
5418 }
5419
5420 /*
5421  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5422  * block must be cow'd
5423  */
5424 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5425                                       struct inode *inode, u64 offset, u64 len)
5426 {
5427         struct btrfs_path *path;
5428         int ret;
5429         struct extent_buffer *leaf;
5430         struct btrfs_root *root = BTRFS_I(inode)->root;
5431         struct btrfs_file_extent_item *fi;
5432         struct btrfs_key key;
5433         u64 disk_bytenr;
5434         u64 backref_offset;
5435         u64 extent_end;
5436         u64 num_bytes;
5437         int slot;
5438         int found_type;
5439
5440         path = btrfs_alloc_path();
5441         if (!path)
5442                 return -ENOMEM;
5443
5444         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5445                                        offset, 0);
5446         if (ret < 0)
5447                 goto out;
5448
5449         slot = path->slots[0];
5450         if (ret == 1) {
5451                 if (slot == 0) {
5452                         /* can't find the item, must cow */
5453                         ret = 0;
5454                         goto out;
5455                 }
5456                 slot--;
5457         }
5458         ret = 0;
5459         leaf = path->nodes[0];
5460         btrfs_item_key_to_cpu(leaf, &key, slot);
5461         if (key.objectid != btrfs_ino(inode) ||
5462             key.type != BTRFS_EXTENT_DATA_KEY) {
5463                 /* not our file or wrong item type, must cow */
5464                 goto out;
5465         }
5466
5467         if (key.offset > offset) {
5468                 /* Wrong offset, must cow */
5469                 goto out;
5470         }
5471
5472         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5473         found_type = btrfs_file_extent_type(leaf, fi);
5474         if (found_type != BTRFS_FILE_EXTENT_REG &&
5475             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5476                 /* not a regular extent, must cow */
5477                 goto out;
5478         }
5479         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5480         backref_offset = btrfs_file_extent_offset(leaf, fi);
5481
5482         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5483         if (extent_end < offset + len) {
5484                 /* extent doesn't include our full range, must cow */
5485                 goto out;
5486         }
5487
5488         if (btrfs_extent_readonly(root, disk_bytenr))
5489                 goto out;
5490
5491         /*
5492          * look for other files referencing this extent, if we
5493          * find any we must cow
5494          */
5495         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5496                                   key.offset - backref_offset, disk_bytenr))
5497                 goto out;
5498
5499         /*
5500          * adjust disk_bytenr and num_bytes to cover just the bytes
5501          * in this extent we are about to write.  If there
5502          * are any csums in that range we have to cow in order
5503          * to keep the csums correct
5504          */
5505         disk_bytenr += backref_offset;
5506         disk_bytenr += offset - key.offset;
5507         num_bytes = min(offset + len, extent_end) - offset;
5508         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5509                                 goto out;
5510         /*
5511          * all of the above have passed, it is safe to overwrite this extent
5512          * without cow
5513          */
5514         ret = 1;
5515 out:
5516         btrfs_free_path(path);
5517         return ret;
5518 }
5519
5520 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5521                                    struct buffer_head *bh_result, int create)
5522 {
5523         struct extent_map *em;
5524         struct btrfs_root *root = BTRFS_I(inode)->root;
5525         u64 start = iblock << inode->i_blkbits;
5526         u64 len = bh_result->b_size;
5527         struct btrfs_trans_handle *trans;
5528
5529         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5530         if (IS_ERR(em))
5531                 return PTR_ERR(em);
5532
5533         /*
5534          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5535          * io.  INLINE is special, and we could probably kludge it in here, but
5536          * it's still buffered so for safety lets just fall back to the generic
5537          * buffered path.
5538          *
5539          * For COMPRESSED we _have_ to read the entire extent in so we can
5540          * decompress it, so there will be buffering required no matter what we
5541          * do, so go ahead and fallback to buffered.
5542          *
5543          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5544          * to buffered IO.  Don't blame me, this is the price we pay for using
5545          * the generic code.
5546          */
5547         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5548             em->block_start == EXTENT_MAP_INLINE) {
5549                 free_extent_map(em);
5550                 return -ENOTBLK;
5551         }
5552
5553         /* Just a good old fashioned hole, return */
5554         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5555                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5556                 free_extent_map(em);
5557                 /* DIO will do one hole at a time, so just unlock a sector */
5558                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5559                               start + root->sectorsize - 1, GFP_NOFS);
5560                 return 0;
5561         }
5562
5563         /*
5564          * We don't allocate a new extent in the following cases
5565          *
5566          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5567          * existing extent.
5568          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5569          * just use the extent.
5570          *
5571          */
5572         if (!create) {
5573                 len = em->len - (start - em->start);
5574                 goto map;
5575         }
5576
5577         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5578             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5579              em->block_start != EXTENT_MAP_HOLE)) {
5580                 int type;
5581                 int ret;
5582                 u64 block_start;
5583
5584                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5585                         type = BTRFS_ORDERED_PREALLOC;
5586                 else
5587                         type = BTRFS_ORDERED_NOCOW;
5588                 len = min(len, em->len - (start - em->start));
5589                 block_start = em->block_start + (start - em->start);
5590
5591                 /*
5592                  * we're not going to log anything, but we do need
5593                  * to make sure the current transaction stays open
5594                  * while we look for nocow cross refs
5595                  */
5596                 trans = btrfs_join_transaction(root);
5597                 if (IS_ERR(trans))
5598                         goto must_cow;
5599
5600                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5601                         ret = btrfs_add_ordered_extent_dio(inode, start,
5602                                            block_start, len, len, type);
5603                         btrfs_end_transaction(trans, root);
5604                         if (ret) {
5605                                 free_extent_map(em);
5606                                 return ret;
5607                         }
5608                         goto unlock;
5609                 }
5610                 btrfs_end_transaction(trans, root);
5611         }
5612 must_cow:
5613         /*
5614          * this will cow the extent, reset the len in case we changed
5615          * it above
5616          */
5617         len = bh_result->b_size;
5618         em = btrfs_new_extent_direct(inode, em, start, len);
5619         if (IS_ERR(em))
5620                 return PTR_ERR(em);
5621         len = min(len, em->len - (start - em->start));
5622 unlock:
5623         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5624                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5625                           0, NULL, GFP_NOFS);
5626 map:
5627         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5628                 inode->i_blkbits;
5629         bh_result->b_size = len;
5630         bh_result->b_bdev = em->bdev;
5631         set_buffer_mapped(bh_result);
5632         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5633                 set_buffer_new(bh_result);
5634
5635         free_extent_map(em);
5636
5637         return 0;
5638 }
5639
5640 struct btrfs_dio_private {
5641         struct inode *inode;
5642         u64 logical_offset;
5643         u64 disk_bytenr;
5644         u64 bytes;
5645         u32 *csums;
5646         void *private;
5647
5648         /* number of bios pending for this dio */
5649         atomic_t pending_bios;
5650
5651         /* IO errors */
5652         int errors;
5653
5654         struct bio *orig_bio;
5655 };
5656
5657 static void btrfs_endio_direct_read(struct bio *bio, int err)
5658 {
5659         struct btrfs_dio_private *dip = bio->bi_private;
5660         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5661         struct bio_vec *bvec = bio->bi_io_vec;
5662         struct inode *inode = dip->inode;
5663         struct btrfs_root *root = BTRFS_I(inode)->root;
5664         u64 start;
5665         u32 *private = dip->csums;
5666
5667         start = dip->logical_offset;
5668         do {
5669                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5670                         struct page *page = bvec->bv_page;
5671                         char *kaddr;
5672                         u32 csum = ~(u32)0;
5673                         unsigned long flags;
5674
5675                         local_irq_save(flags);
5676                         kaddr = kmap_atomic(page, KM_IRQ0);
5677                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5678                                                csum, bvec->bv_len);
5679                         btrfs_csum_final(csum, (char *)&csum);
5680                         kunmap_atomic(kaddr, KM_IRQ0);
5681                         local_irq_restore(flags);
5682
5683                         flush_dcache_page(bvec->bv_page);
5684                         if (csum != *private) {
5685                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5686                                       " %llu csum %u private %u\n",
5687                                       (unsigned long long)btrfs_ino(inode),
5688                                       (unsigned long long)start,
5689                                       csum, *private);
5690                                 err = -EIO;
5691                         }
5692                 }
5693
5694                 start += bvec->bv_len;
5695                 private++;
5696                 bvec++;
5697         } while (bvec <= bvec_end);
5698
5699         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5700                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5701         bio->bi_private = dip->private;
5702
5703         kfree(dip->csums);
5704         kfree(dip);
5705
5706         /* If we had a csum failure make sure to clear the uptodate flag */
5707         if (err)
5708                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5709         dio_end_io(bio, err);
5710 }
5711
5712 static void btrfs_endio_direct_write(struct bio *bio, int err)
5713 {
5714         struct btrfs_dio_private *dip = bio->bi_private;
5715         struct inode *inode = dip->inode;
5716         struct btrfs_root *root = BTRFS_I(inode)->root;
5717         struct btrfs_trans_handle *trans;
5718         struct btrfs_ordered_extent *ordered = NULL;
5719         struct extent_state *cached_state = NULL;
5720         u64 ordered_offset = dip->logical_offset;
5721         u64 ordered_bytes = dip->bytes;
5722         int ret;
5723
5724         if (err)
5725                 goto out_done;
5726 again:
5727         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5728                                                    &ordered_offset,
5729                                                    ordered_bytes);
5730         if (!ret)
5731                 goto out_test;
5732
5733         BUG_ON(!ordered);
5734
5735         trans = btrfs_join_transaction(root);
5736         if (IS_ERR(trans)) {
5737                 err = -ENOMEM;
5738                 goto out;
5739         }
5740         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5741
5742         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5743                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5744                 if (!ret)
5745                         ret = btrfs_update_inode(trans, root, inode);
5746                 err = ret;
5747                 goto out;
5748         }
5749
5750         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5751                          ordered->file_offset + ordered->len - 1, 0,
5752                          &cached_state, GFP_NOFS);
5753
5754         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5755                 ret = btrfs_mark_extent_written(trans, inode,
5756                                                 ordered->file_offset,
5757                                                 ordered->file_offset +
5758                                                 ordered->len);
5759                 if (ret) {
5760                         err = ret;
5761                         goto out_unlock;
5762                 }
5763         } else {
5764                 ret = insert_reserved_file_extent(trans, inode,
5765                                                   ordered->file_offset,
5766                                                   ordered->start,
5767                                                   ordered->disk_len,
5768                                                   ordered->len,
5769                                                   ordered->len,
5770                                                   0, 0, 0,
5771                                                   BTRFS_FILE_EXTENT_REG);
5772                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5773                                    ordered->file_offset, ordered->len);
5774                 if (ret) {
5775                         err = ret;
5776                         WARN_ON(1);
5777                         goto out_unlock;
5778                 }
5779         }
5780
5781         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5782         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5783         if (!ret)
5784                 btrfs_update_inode(trans, root, inode);
5785         ret = 0;
5786 out_unlock:
5787         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5788                              ordered->file_offset + ordered->len - 1,
5789                              &cached_state, GFP_NOFS);
5790 out:
5791         btrfs_delalloc_release_metadata(inode, ordered->len);
5792         btrfs_end_transaction(trans, root);
5793         ordered_offset = ordered->file_offset + ordered->len;
5794         btrfs_put_ordered_extent(ordered);
5795         btrfs_put_ordered_extent(ordered);
5796
5797 out_test:
5798         /*
5799          * our bio might span multiple ordered extents.  If we haven't
5800          * completed the accounting for the whole dio, go back and try again
5801          */
5802         if (ordered_offset < dip->logical_offset + dip->bytes) {
5803                 ordered_bytes = dip->logical_offset + dip->bytes -
5804                         ordered_offset;
5805                 goto again;
5806         }
5807 out_done:
5808         bio->bi_private = dip->private;
5809
5810         kfree(dip->csums);
5811         kfree(dip);
5812
5813         /* If we had an error make sure to clear the uptodate flag */
5814         if (err)
5815                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5816         dio_end_io(bio, err);
5817 }
5818
5819 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5820                                     struct bio *bio, int mirror_num,
5821                                     unsigned long bio_flags, u64 offset)
5822 {
5823         int ret;
5824         struct btrfs_root *root = BTRFS_I(inode)->root;
5825         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5826         BUG_ON(ret);
5827         return 0;
5828 }
5829
5830 static void btrfs_end_dio_bio(struct bio *bio, int err)
5831 {
5832         struct btrfs_dio_private *dip = bio->bi_private;
5833
5834         if (err) {
5835                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5836                       "sector %#Lx len %u err no %d\n",
5837                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5838                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5839                 dip->errors = 1;
5840
5841                 /*
5842                  * before atomic variable goto zero, we must make sure
5843                  * dip->errors is perceived to be set.
5844                  */
5845                 smp_mb__before_atomic_dec();
5846         }
5847
5848         /* if there are more bios still pending for this dio, just exit */
5849         if (!atomic_dec_and_test(&dip->pending_bios))
5850                 goto out;
5851
5852         if (dip->errors)
5853                 bio_io_error(dip->orig_bio);
5854         else {
5855                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5856                 bio_endio(dip->orig_bio, 0);
5857         }
5858 out:
5859         bio_put(bio);
5860 }
5861
5862 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5863                                        u64 first_sector, gfp_t gfp_flags)
5864 {
5865         int nr_vecs = bio_get_nr_vecs(bdev);
5866         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5867 }
5868
5869 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5870                                          int rw, u64 file_offset, int skip_sum,
5871                                          u32 *csums, int async_submit)
5872 {
5873         int write = rw & REQ_WRITE;
5874         struct btrfs_root *root = BTRFS_I(inode)->root;
5875         int ret;
5876
5877         bio_get(bio);
5878         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5879         if (ret)
5880                 goto err;
5881
5882         if (skip_sum)
5883                 goto map;
5884
5885         if (write && async_submit) {
5886                 ret = btrfs_wq_submit_bio(root->fs_info,
5887                                    inode, rw, bio, 0, 0,
5888                                    file_offset,
5889                                    __btrfs_submit_bio_start_direct_io,
5890                                    __btrfs_submit_bio_done);
5891                 goto err;
5892         } else if (write) {
5893                 /*
5894                  * If we aren't doing async submit, calculate the csum of the
5895                  * bio now.
5896                  */
5897                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5898                 if (ret)
5899                         goto err;
5900         } else if (!skip_sum) {
5901                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5902                                           file_offset, csums);
5903                 if (ret)
5904                         goto err;
5905         }
5906
5907 map:
5908         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5909 err:
5910         bio_put(bio);
5911         return ret;
5912 }
5913
5914 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5915                                     int skip_sum)
5916 {
5917         struct inode *inode = dip->inode;
5918         struct btrfs_root *root = BTRFS_I(inode)->root;
5919         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5920         struct bio *bio;
5921         struct bio *orig_bio = dip->orig_bio;
5922         struct bio_vec *bvec = orig_bio->bi_io_vec;
5923         u64 start_sector = orig_bio->bi_sector;
5924         u64 file_offset = dip->logical_offset;
5925         u64 submit_len = 0;
5926         u64 map_length;
5927         int nr_pages = 0;
5928         u32 *csums = dip->csums;
5929         int ret = 0;
5930         int async_submit = 0;
5931         int write = rw & REQ_WRITE;
5932
5933         map_length = orig_bio->bi_size;
5934         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5935                               &map_length, NULL, 0);
5936         if (ret) {
5937                 bio_put(orig_bio);
5938                 return -EIO;
5939         }
5940
5941         if (map_length >= orig_bio->bi_size) {
5942                 bio = orig_bio;
5943                 goto submit;
5944         }
5945
5946         async_submit = 1;
5947         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5948         if (!bio)
5949                 return -ENOMEM;
5950         bio->bi_private = dip;
5951         bio->bi_end_io = btrfs_end_dio_bio;
5952         atomic_inc(&dip->pending_bios);
5953
5954         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5955                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5956                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5957                                  bvec->bv_offset) < bvec->bv_len)) {
5958                         /*
5959                          * inc the count before we submit the bio so
5960                          * we know the end IO handler won't happen before
5961                          * we inc the count. Otherwise, the dip might get freed
5962                          * before we're done setting it up
5963                          */
5964                         atomic_inc(&dip->pending_bios);
5965                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5966                                                      file_offset, skip_sum,
5967                                                      csums, async_submit);
5968                         if (ret) {
5969                                 bio_put(bio);
5970                                 atomic_dec(&dip->pending_bios);
5971                                 goto out_err;
5972                         }
5973
5974                         /* Write's use the ordered csums */
5975                         if (!write && !skip_sum)
5976                                 csums = csums + nr_pages;
5977                         start_sector += submit_len >> 9;
5978                         file_offset += submit_len;
5979
5980                         submit_len = 0;
5981                         nr_pages = 0;
5982
5983                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5984                                                   start_sector, GFP_NOFS);
5985                         if (!bio)
5986                                 goto out_err;
5987                         bio->bi_private = dip;
5988                         bio->bi_end_io = btrfs_end_dio_bio;
5989
5990                         map_length = orig_bio->bi_size;
5991                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5992                                               &map_length, NULL, 0);
5993                         if (ret) {
5994                                 bio_put(bio);
5995                                 goto out_err;
5996                         }
5997                 } else {
5998                         submit_len += bvec->bv_len;
5999                         nr_pages ++;
6000                         bvec++;
6001                 }
6002         }
6003
6004 submit:
6005         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6006                                      csums, async_submit);
6007         if (!ret)
6008                 return 0;
6009
6010         bio_put(bio);
6011 out_err:
6012         dip->errors = 1;
6013         /*
6014          * before atomic variable goto zero, we must
6015          * make sure dip->errors is perceived to be set.
6016          */
6017         smp_mb__before_atomic_dec();
6018         if (atomic_dec_and_test(&dip->pending_bios))
6019                 bio_io_error(dip->orig_bio);
6020
6021         /* bio_end_io() will handle error, so we needn't return it */
6022         return 0;
6023 }
6024
6025 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6026                                 loff_t file_offset)
6027 {
6028         struct btrfs_root *root = BTRFS_I(inode)->root;
6029         struct btrfs_dio_private *dip;
6030         struct bio_vec *bvec = bio->bi_io_vec;
6031         int skip_sum;
6032         int write = rw & REQ_WRITE;
6033         int ret = 0;
6034
6035         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6036
6037         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6038         if (!dip) {
6039                 ret = -ENOMEM;
6040                 goto free_ordered;
6041         }
6042         dip->csums = NULL;
6043
6044         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6045         if (!write && !skip_sum) {
6046                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6047                 if (!dip->csums) {
6048                         kfree(dip);
6049                         ret = -ENOMEM;
6050                         goto free_ordered;
6051                 }
6052         }
6053
6054         dip->private = bio->bi_private;
6055         dip->inode = inode;
6056         dip->logical_offset = file_offset;
6057
6058         dip->bytes = 0;
6059         do {
6060                 dip->bytes += bvec->bv_len;
6061                 bvec++;
6062         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6063
6064         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6065         bio->bi_private = dip;
6066         dip->errors = 0;
6067         dip->orig_bio = bio;
6068         atomic_set(&dip->pending_bios, 0);
6069
6070         if (write)
6071                 bio->bi_end_io = btrfs_endio_direct_write;
6072         else
6073                 bio->bi_end_io = btrfs_endio_direct_read;
6074
6075         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6076         if (!ret)
6077                 return;
6078 free_ordered:
6079         /*
6080          * If this is a write, we need to clean up the reserved space and kill
6081          * the ordered extent.
6082          */
6083         if (write) {
6084                 struct btrfs_ordered_extent *ordered;
6085                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6086                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6087                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6088                         btrfs_free_reserved_extent(root, ordered->start,
6089                                                    ordered->disk_len);
6090                 btrfs_put_ordered_extent(ordered);
6091                 btrfs_put_ordered_extent(ordered);
6092         }
6093         bio_endio(bio, ret);
6094 }
6095
6096 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6097                         const struct iovec *iov, loff_t offset,
6098                         unsigned long nr_segs)
6099 {
6100         int seg;
6101         int i;
6102         size_t size;
6103         unsigned long addr;
6104         unsigned blocksize_mask = root->sectorsize - 1;
6105         ssize_t retval = -EINVAL;
6106         loff_t end = offset;
6107
6108         if (offset & blocksize_mask)
6109                 goto out;
6110
6111         /* Check the memory alignment.  Blocks cannot straddle pages */
6112         for (seg = 0; seg < nr_segs; seg++) {
6113                 addr = (unsigned long)iov[seg].iov_base;
6114                 size = iov[seg].iov_len;
6115                 end += size;
6116                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6117                         goto out;
6118
6119                 /* If this is a write we don't need to check anymore */
6120                 if (rw & WRITE)
6121                         continue;
6122
6123                 /*
6124                  * Check to make sure we don't have duplicate iov_base's in this
6125                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6126                  * when reading back.
6127                  */
6128                 for (i = seg + 1; i < nr_segs; i++) {
6129                         if (iov[seg].iov_base == iov[i].iov_base)
6130                                 goto out;
6131                 }
6132         }
6133         retval = 0;
6134 out:
6135         return retval;
6136 }
6137 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6138                         const struct iovec *iov, loff_t offset,
6139                         unsigned long nr_segs)
6140 {
6141         struct file *file = iocb->ki_filp;
6142         struct inode *inode = file->f_mapping->host;
6143         struct btrfs_ordered_extent *ordered;
6144         struct extent_state *cached_state = NULL;
6145         u64 lockstart, lockend;
6146         ssize_t ret;
6147         int writing = rw & WRITE;
6148         int write_bits = 0;
6149         size_t count = iov_length(iov, nr_segs);
6150
6151         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6152                             offset, nr_segs)) {
6153                 return 0;
6154         }
6155
6156         lockstart = offset;
6157         lockend = offset + count - 1;
6158
6159         if (writing) {
6160                 ret = btrfs_delalloc_reserve_space(inode, count);
6161                 if (ret)
6162                         goto out;
6163         }
6164
6165         while (1) {
6166                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6167                                  0, &cached_state, GFP_NOFS);
6168                 /*
6169                  * We're concerned with the entire range that we're going to be
6170                  * doing DIO to, so we need to make sure theres no ordered
6171                  * extents in this range.
6172                  */
6173                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6174                                                      lockend - lockstart + 1);
6175                 if (!ordered)
6176                         break;
6177                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6178                                      &cached_state, GFP_NOFS);
6179                 btrfs_start_ordered_extent(inode, ordered, 1);
6180                 btrfs_put_ordered_extent(ordered);
6181                 cond_resched();
6182         }
6183
6184         /*
6185          * we don't use btrfs_set_extent_delalloc because we don't want
6186          * the dirty or uptodate bits
6187          */
6188         if (writing) {
6189                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6190                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6191                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6192                                      GFP_NOFS);
6193                 if (ret) {
6194                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6195                                          lockend, EXTENT_LOCKED | write_bits,
6196                                          1, 0, &cached_state, GFP_NOFS);
6197                         goto out;
6198                 }
6199         }
6200
6201         free_extent_state(cached_state);
6202         cached_state = NULL;
6203
6204         ret = __blockdev_direct_IO(rw, iocb, inode,
6205                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6206                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6207                    btrfs_submit_direct, 0);
6208
6209         if (ret < 0 && ret != -EIOCBQUEUED) {
6210                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6211                               offset + iov_length(iov, nr_segs) - 1,
6212                               EXTENT_LOCKED | write_bits, 1, 0,
6213                               &cached_state, GFP_NOFS);
6214         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6215                 /*
6216                  * We're falling back to buffered, unlock the section we didn't
6217                  * do IO on.
6218                  */
6219                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6220                               offset + iov_length(iov, nr_segs) - 1,
6221                               EXTENT_LOCKED | write_bits, 1, 0,
6222                               &cached_state, GFP_NOFS);
6223         }
6224 out:
6225         free_extent_state(cached_state);
6226         return ret;
6227 }
6228
6229 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6230                 __u64 start, __u64 len)
6231 {
6232         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6233 }
6234
6235 int btrfs_readpage(struct file *file, struct page *page)
6236 {
6237         struct extent_io_tree *tree;
6238         tree = &BTRFS_I(page->mapping->host)->io_tree;
6239         return extent_read_full_page(tree, page, btrfs_get_extent);
6240 }
6241
6242 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6243 {
6244         struct extent_io_tree *tree;
6245
6246
6247         if (current->flags & PF_MEMALLOC) {
6248                 redirty_page_for_writepage(wbc, page);
6249                 unlock_page(page);
6250                 return 0;
6251         }
6252         tree = &BTRFS_I(page->mapping->host)->io_tree;
6253         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6254 }
6255
6256 int btrfs_writepages(struct address_space *mapping,
6257                      struct writeback_control *wbc)
6258 {
6259         struct extent_io_tree *tree;
6260
6261         tree = &BTRFS_I(mapping->host)->io_tree;
6262         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6263 }
6264
6265 static int
6266 btrfs_readpages(struct file *file, struct address_space *mapping,
6267                 struct list_head *pages, unsigned nr_pages)
6268 {
6269         struct extent_io_tree *tree;
6270         tree = &BTRFS_I(mapping->host)->io_tree;
6271         return extent_readpages(tree, mapping, pages, nr_pages,
6272                                 btrfs_get_extent);
6273 }
6274 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6275 {
6276         struct extent_io_tree *tree;
6277         struct extent_map_tree *map;
6278         int ret;
6279
6280         tree = &BTRFS_I(page->mapping->host)->io_tree;
6281         map = &BTRFS_I(page->mapping->host)->extent_tree;
6282         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6283         if (ret == 1) {
6284                 ClearPagePrivate(page);
6285                 set_page_private(page, 0);
6286                 page_cache_release(page);
6287         }
6288         return ret;
6289 }
6290
6291 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6292 {
6293         if (PageWriteback(page) || PageDirty(page))
6294                 return 0;
6295         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6296 }
6297
6298 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6299 {
6300         struct extent_io_tree *tree;
6301         struct btrfs_ordered_extent *ordered;
6302         struct extent_state *cached_state = NULL;
6303         u64 page_start = page_offset(page);
6304         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6305
6306
6307         /*
6308          * we have the page locked, so new writeback can't start,
6309          * and the dirty bit won't be cleared while we are here.
6310          *
6311          * Wait for IO on this page so that we can safely clear
6312          * the PagePrivate2 bit and do ordered accounting
6313          */
6314         wait_on_page_writeback(page);
6315
6316         tree = &BTRFS_I(page->mapping->host)->io_tree;
6317         if (offset) {
6318                 btrfs_releasepage(page, GFP_NOFS);
6319                 return;
6320         }
6321         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6322                          GFP_NOFS);
6323         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6324                                            page_offset(page));
6325         if (ordered) {
6326                 /*
6327                  * IO on this page will never be started, so we need
6328                  * to account for any ordered extents now
6329                  */
6330                 clear_extent_bit(tree, page_start, page_end,
6331                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6332                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6333                                  &cached_state, GFP_NOFS);
6334                 /*
6335                  * whoever cleared the private bit is responsible
6336                  * for the finish_ordered_io
6337                  */
6338                 if (TestClearPagePrivate2(page)) {
6339                         btrfs_finish_ordered_io(page->mapping->host,
6340                                                 page_start, page_end);
6341                 }
6342                 btrfs_put_ordered_extent(ordered);
6343                 cached_state = NULL;
6344                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6345                                  GFP_NOFS);
6346         }
6347         clear_extent_bit(tree, page_start, page_end,
6348                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6349                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6350         __btrfs_releasepage(page, GFP_NOFS);
6351
6352         ClearPageChecked(page);
6353         if (PagePrivate(page)) {
6354                 ClearPagePrivate(page);
6355                 set_page_private(page, 0);
6356                 page_cache_release(page);
6357         }
6358 }
6359
6360 /*
6361  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6362  * called from a page fault handler when a page is first dirtied. Hence we must
6363  * be careful to check for EOF conditions here. We set the page up correctly
6364  * for a written page which means we get ENOSPC checking when writing into
6365  * holes and correct delalloc and unwritten extent mapping on filesystems that
6366  * support these features.
6367  *
6368  * We are not allowed to take the i_mutex here so we have to play games to
6369  * protect against truncate races as the page could now be beyond EOF.  Because
6370  * vmtruncate() writes the inode size before removing pages, once we have the
6371  * page lock we can determine safely if the page is beyond EOF. If it is not
6372  * beyond EOF, then the page is guaranteed safe against truncation until we
6373  * unlock the page.
6374  */
6375 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6376 {
6377         struct page *page = vmf->page;
6378         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6379         struct btrfs_root *root = BTRFS_I(inode)->root;
6380         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6381         struct btrfs_ordered_extent *ordered;
6382         struct extent_state *cached_state = NULL;
6383         char *kaddr;
6384         unsigned long zero_start;
6385         loff_t size;
6386         int ret;
6387         u64 page_start;
6388         u64 page_end;
6389
6390         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6391         if (ret) {
6392                 if (ret == -ENOMEM)
6393                         ret = VM_FAULT_OOM;
6394                 else /* -ENOSPC, -EIO, etc */
6395                         ret = VM_FAULT_SIGBUS;
6396                 goto out;
6397         }
6398
6399         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6400 again:
6401         lock_page(page);
6402         size = i_size_read(inode);
6403         page_start = page_offset(page);
6404         page_end = page_start + PAGE_CACHE_SIZE - 1;
6405
6406         if ((page->mapping != inode->i_mapping) ||
6407             (page_start >= size)) {
6408                 /* page got truncated out from underneath us */
6409                 goto out_unlock;
6410         }
6411         wait_on_page_writeback(page);
6412
6413         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6414                          GFP_NOFS);
6415         set_page_extent_mapped(page);
6416
6417         /*
6418          * we can't set the delalloc bits if there are pending ordered
6419          * extents.  Drop our locks and wait for them to finish
6420          */
6421         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6422         if (ordered) {
6423                 unlock_extent_cached(io_tree, page_start, page_end,
6424                                      &cached_state, GFP_NOFS);
6425                 unlock_page(page);
6426                 btrfs_start_ordered_extent(inode, ordered, 1);
6427                 btrfs_put_ordered_extent(ordered);
6428                 goto again;
6429         }
6430
6431         /*
6432          * XXX - page_mkwrite gets called every time the page is dirtied, even
6433          * if it was already dirty, so for space accounting reasons we need to
6434          * clear any delalloc bits for the range we are fixing to save.  There
6435          * is probably a better way to do this, but for now keep consistent with
6436          * prepare_pages in the normal write path.
6437          */
6438         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6439                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6440                           0, 0, &cached_state, GFP_NOFS);
6441
6442         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6443                                         &cached_state);
6444         if (ret) {
6445                 unlock_extent_cached(io_tree, page_start, page_end,
6446                                      &cached_state, GFP_NOFS);
6447                 ret = VM_FAULT_SIGBUS;
6448                 goto out_unlock;
6449         }
6450         ret = 0;
6451
6452         /* page is wholly or partially inside EOF */
6453         if (page_start + PAGE_CACHE_SIZE > size)
6454                 zero_start = size & ~PAGE_CACHE_MASK;
6455         else
6456                 zero_start = PAGE_CACHE_SIZE;
6457
6458         if (zero_start != PAGE_CACHE_SIZE) {
6459                 kaddr = kmap(page);
6460                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6461                 flush_dcache_page(page);
6462                 kunmap(page);
6463         }
6464         ClearPageChecked(page);
6465         set_page_dirty(page);
6466         SetPageUptodate(page);
6467
6468         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6469         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6470
6471         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6472
6473 out_unlock:
6474         if (!ret)
6475                 return VM_FAULT_LOCKED;
6476         unlock_page(page);
6477         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6478 out:
6479         return ret;
6480 }
6481
6482 static int btrfs_truncate(struct inode *inode)
6483 {
6484         struct btrfs_root *root = BTRFS_I(inode)->root;
6485         struct btrfs_block_rsv *rsv;
6486         int ret;
6487         int err = 0;
6488         struct btrfs_trans_handle *trans;
6489         unsigned long nr;
6490         u64 mask = root->sectorsize - 1;
6491
6492         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6493         if (ret)
6494                 return ret;
6495
6496         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6497         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6498
6499         /*
6500          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6501          * 3 things going on here
6502          *
6503          * 1) We need to reserve space for our orphan item and the space to
6504          * delete our orphan item.  Lord knows we don't want to have a dangling
6505          * orphan item because we didn't reserve space to remove it.
6506          *
6507          * 2) We need to reserve space to update our inode.
6508          *
6509          * 3) We need to have something to cache all the space that is going to
6510          * be free'd up by the truncate operation, but also have some slack
6511          * space reserved in case it uses space during the truncate (thank you
6512          * very much snapshotting).
6513          *
6514          * And we need these to all be seperate.  The fact is we can use alot of
6515          * space doing the truncate, and we have no earthly idea how much space
6516          * we will use, so we need the truncate reservation to be seperate so it
6517          * doesn't end up using space reserved for updating the inode or
6518          * removing the orphan item.  We also need to be able to stop the
6519          * transaction and start a new one, which means we need to be able to
6520          * update the inode several times, and we have no idea of knowing how
6521          * many times that will be, so we can't just reserve 1 item for the
6522          * entirety of the opration, so that has to be done seperately as well.
6523          * Then there is the orphan item, which does indeed need to be held on
6524          * to for the whole operation, and we need nobody to touch this reserved
6525          * space except the orphan code.
6526          *
6527          * So that leaves us with
6528          *
6529          * 1) root->orphan_block_rsv - for the orphan deletion.
6530          * 2) rsv - for the truncate reservation, which we will steal from the
6531          * transaction reservation.
6532          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6533          * updating the inode.
6534          */
6535         rsv = btrfs_alloc_block_rsv(root);
6536         if (!rsv)
6537                 return -ENOMEM;
6538         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6539
6540         trans = btrfs_start_transaction(root, 4);
6541         if (IS_ERR(trans)) {
6542                 err = PTR_ERR(trans);
6543                 goto out;
6544         }
6545
6546         /*
6547          * Reserve space for the truncate process.  Truncate should be adding
6548          * space, but if there are snapshots it may end up using space.
6549          */
6550         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6551         BUG_ON(ret);
6552
6553         ret = btrfs_orphan_add(trans, inode);
6554         if (ret) {
6555                 btrfs_end_transaction(trans, root);
6556                 goto out;
6557         }
6558
6559         nr = trans->blocks_used;
6560         btrfs_end_transaction(trans, root);
6561         btrfs_btree_balance_dirty(root, nr);
6562
6563         /*
6564          * Ok so we've already migrated our bytes over for the truncate, so here
6565          * just reserve the one slot we need for updating the inode.
6566          */
6567         trans = btrfs_start_transaction(root, 1);
6568         if (IS_ERR(trans)) {
6569                 err = PTR_ERR(trans);
6570                 goto out;
6571         }
6572         trans->block_rsv = rsv;
6573
6574         /*
6575          * setattr is responsible for setting the ordered_data_close flag,
6576          * but that is only tested during the last file release.  That
6577          * could happen well after the next commit, leaving a great big
6578          * window where new writes may get lost if someone chooses to write
6579          * to this file after truncating to zero
6580          *
6581          * The inode doesn't have any dirty data here, and so if we commit
6582          * this is a noop.  If someone immediately starts writing to the inode
6583          * it is very likely we'll catch some of their writes in this
6584          * transaction, and the commit will find this file on the ordered
6585          * data list with good things to send down.
6586          *
6587          * This is a best effort solution, there is still a window where
6588          * using truncate to replace the contents of the file will
6589          * end up with a zero length file after a crash.
6590          */
6591         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6592                 btrfs_add_ordered_operation(trans, root, inode);
6593
6594         while (1) {
6595                 if (!trans) {
6596                         trans = btrfs_start_transaction(root, 3);
6597                         if (IS_ERR(trans)) {
6598                                 err = PTR_ERR(trans);
6599                                 goto out;
6600                         }
6601
6602                         ret = btrfs_truncate_reserve_metadata(trans, root,
6603                                                               rsv);
6604                         BUG_ON(ret);
6605
6606                         trans->block_rsv = rsv;
6607                 }
6608
6609                 ret = btrfs_truncate_inode_items(trans, root, inode,
6610                                                  inode->i_size,
6611                                                  BTRFS_EXTENT_DATA_KEY);
6612                 if (ret != -EAGAIN) {
6613                         err = ret;
6614                         break;
6615                 }
6616
6617                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6618                 ret = btrfs_update_inode(trans, root, inode);
6619                 if (ret) {
6620                         err = ret;
6621                         break;
6622                 }
6623
6624                 nr = trans->blocks_used;
6625                 btrfs_end_transaction(trans, root);
6626                 trans = NULL;
6627                 btrfs_btree_balance_dirty(root, nr);
6628         }
6629
6630         if (ret == 0 && inode->i_nlink > 0) {
6631                 trans->block_rsv = root->orphan_block_rsv;
6632                 ret = btrfs_orphan_del(trans, inode);
6633                 if (ret)
6634                         err = ret;
6635         } else if (ret && inode->i_nlink > 0) {
6636                 /*
6637                  * Failed to do the truncate, remove us from the in memory
6638                  * orphan list.
6639                  */
6640                 ret = btrfs_orphan_del(NULL, inode);
6641         }
6642
6643         trans->block_rsv = &root->fs_info->trans_block_rsv;
6644         ret = btrfs_update_inode(trans, root, inode);
6645         if (ret && !err)
6646                 err = ret;
6647
6648         nr = trans->blocks_used;
6649         ret = btrfs_end_transaction_throttle(trans, root);
6650         btrfs_btree_balance_dirty(root, nr);
6651
6652 out:
6653         btrfs_free_block_rsv(root, rsv);
6654
6655         if (ret && !err)
6656                 err = ret;
6657
6658         return err;
6659 }
6660
6661 /*
6662  * create a new subvolume directory/inode (helper for the ioctl).
6663  */
6664 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6665                              struct btrfs_root *new_root, u64 new_dirid)
6666 {
6667         struct inode *inode;
6668         int err;
6669         u64 index = 0;
6670
6671         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6672                                 new_dirid, S_IFDIR | 0700, &index);
6673         if (IS_ERR(inode))
6674                 return PTR_ERR(inode);
6675         inode->i_op = &btrfs_dir_inode_operations;
6676         inode->i_fop = &btrfs_dir_file_operations;
6677
6678         inode->i_nlink = 1;
6679         btrfs_i_size_write(inode, 0);
6680
6681         err = btrfs_update_inode(trans, new_root, inode);
6682         BUG_ON(err);
6683
6684         iput(inode);
6685         return 0;
6686 }
6687
6688 /* helper function for file defrag and space balancing.  This
6689  * forces readahead on a given range of bytes in an inode
6690  */
6691 unsigned long btrfs_force_ra(struct address_space *mapping,
6692                               struct file_ra_state *ra, struct file *file,
6693                               pgoff_t offset, pgoff_t last_index)
6694 {
6695         pgoff_t req_size = last_index - offset + 1;
6696
6697         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6698         return offset + req_size;
6699 }
6700
6701 struct inode *btrfs_alloc_inode(struct super_block *sb)
6702 {
6703         struct btrfs_inode *ei;
6704         struct inode *inode;
6705
6706         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6707         if (!ei)
6708                 return NULL;
6709
6710         ei->root = NULL;
6711         ei->space_info = NULL;
6712         ei->generation = 0;
6713         ei->sequence = 0;
6714         ei->last_trans = 0;
6715         ei->last_sub_trans = 0;
6716         ei->logged_trans = 0;
6717         ei->delalloc_bytes = 0;
6718         ei->reserved_bytes = 0;
6719         ei->disk_i_size = 0;
6720         ei->flags = 0;
6721         ei->index_cnt = (u64)-1;
6722         ei->last_unlink_trans = 0;
6723
6724         spin_lock_init(&ei->lock);
6725         ei->outstanding_extents = 0;
6726         ei->reserved_extents = 0;
6727
6728         ei->ordered_data_close = 0;
6729         ei->orphan_meta_reserved = 0;
6730         ei->dummy_inode = 0;
6731         ei->in_defrag = 0;
6732         ei->force_compress = BTRFS_COMPRESS_NONE;
6733
6734         ei->delayed_node = NULL;
6735
6736         inode = &ei->vfs_inode;
6737         extent_map_tree_init(&ei->extent_tree);
6738         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6739         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6740         mutex_init(&ei->log_mutex);
6741         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6742         INIT_LIST_HEAD(&ei->i_orphan);
6743         INIT_LIST_HEAD(&ei->delalloc_inodes);
6744         INIT_LIST_HEAD(&ei->ordered_operations);
6745         RB_CLEAR_NODE(&ei->rb_node);
6746
6747         return inode;
6748 }
6749
6750 static void btrfs_i_callback(struct rcu_head *head)
6751 {
6752         struct inode *inode = container_of(head, struct inode, i_rcu);
6753         INIT_LIST_HEAD(&inode->i_dentry);
6754         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6755 }
6756
6757 void btrfs_destroy_inode(struct inode *inode)
6758 {
6759         struct btrfs_ordered_extent *ordered;
6760         struct btrfs_root *root = BTRFS_I(inode)->root;
6761
6762         WARN_ON(!list_empty(&inode->i_dentry));
6763         WARN_ON(inode->i_data.nrpages);
6764         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6765         WARN_ON(BTRFS_I(inode)->reserved_extents);
6766
6767         /*
6768          * This can happen where we create an inode, but somebody else also
6769          * created the same inode and we need to destroy the one we already
6770          * created.
6771          */
6772         if (!root)
6773                 goto free;
6774
6775         /*
6776          * Make sure we're properly removed from the ordered operation
6777          * lists.
6778          */
6779         smp_mb();
6780         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6781                 spin_lock(&root->fs_info->ordered_extent_lock);
6782                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6783                 spin_unlock(&root->fs_info->ordered_extent_lock);
6784         }
6785
6786         spin_lock(&root->orphan_lock);
6787         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6788                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6789                        (unsigned long long)btrfs_ino(inode));
6790                 list_del_init(&BTRFS_I(inode)->i_orphan);
6791         }
6792         spin_unlock(&root->orphan_lock);
6793
6794         while (1) {
6795                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6796                 if (!ordered)
6797                         break;
6798                 else {
6799                         printk(KERN_ERR "btrfs found ordered "
6800                                "extent %llu %llu on inode cleanup\n",
6801                                (unsigned long long)ordered->file_offset,
6802                                (unsigned long long)ordered->len);
6803                         btrfs_remove_ordered_extent(inode, ordered);
6804                         btrfs_put_ordered_extent(ordered);
6805                         btrfs_put_ordered_extent(ordered);
6806                 }
6807         }
6808         inode_tree_del(inode);
6809         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6810 free:
6811         btrfs_remove_delayed_node(inode);
6812         call_rcu(&inode->i_rcu, btrfs_i_callback);
6813 }
6814
6815 int btrfs_drop_inode(struct inode *inode)
6816 {
6817         struct btrfs_root *root = BTRFS_I(inode)->root;
6818
6819         if (btrfs_root_refs(&root->root_item) == 0 &&
6820             !is_free_space_inode(root, inode))
6821                 return 1;
6822         else
6823                 return generic_drop_inode(inode);
6824 }
6825
6826 static void init_once(void *foo)
6827 {
6828         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6829
6830         inode_init_once(&ei->vfs_inode);
6831 }
6832
6833 void btrfs_destroy_cachep(void)
6834 {
6835         if (btrfs_inode_cachep)
6836                 kmem_cache_destroy(btrfs_inode_cachep);
6837         if (btrfs_trans_handle_cachep)
6838                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6839         if (btrfs_transaction_cachep)
6840                 kmem_cache_destroy(btrfs_transaction_cachep);
6841         if (btrfs_path_cachep)
6842                 kmem_cache_destroy(btrfs_path_cachep);
6843         if (btrfs_free_space_cachep)
6844                 kmem_cache_destroy(btrfs_free_space_cachep);
6845 }
6846
6847 int btrfs_init_cachep(void)
6848 {
6849         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6850                         sizeof(struct btrfs_inode), 0,
6851                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6852         if (!btrfs_inode_cachep)
6853                 goto fail;
6854
6855         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6856                         sizeof(struct btrfs_trans_handle), 0,
6857                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6858         if (!btrfs_trans_handle_cachep)
6859                 goto fail;
6860
6861         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6862                         sizeof(struct btrfs_transaction), 0,
6863                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6864         if (!btrfs_transaction_cachep)
6865                 goto fail;
6866
6867         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6868                         sizeof(struct btrfs_path), 0,
6869                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6870         if (!btrfs_path_cachep)
6871                 goto fail;
6872
6873         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6874                         sizeof(struct btrfs_free_space), 0,
6875                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6876         if (!btrfs_free_space_cachep)
6877                 goto fail;
6878
6879         return 0;
6880 fail:
6881         btrfs_destroy_cachep();
6882         return -ENOMEM;
6883 }
6884
6885 static int btrfs_getattr(struct vfsmount *mnt,
6886                          struct dentry *dentry, struct kstat *stat)
6887 {
6888         struct inode *inode = dentry->d_inode;
6889         generic_fillattr(inode, stat);
6890         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6891         stat->blksize = PAGE_CACHE_SIZE;
6892         stat->blocks = (inode_get_bytes(inode) +
6893                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6894         return 0;
6895 }
6896
6897 /*
6898  * If a file is moved, it will inherit the cow and compression flags of the new
6899  * directory.
6900  */
6901 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6902 {
6903         struct btrfs_inode *b_dir = BTRFS_I(dir);
6904         struct btrfs_inode *b_inode = BTRFS_I(inode);
6905
6906         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6907                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6908         else
6909                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6910
6911         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6912                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6913         else
6914                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6915 }
6916
6917 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6918                            struct inode *new_dir, struct dentry *new_dentry)
6919 {
6920         struct btrfs_trans_handle *trans;
6921         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6922         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6923         struct inode *new_inode = new_dentry->d_inode;
6924         struct inode *old_inode = old_dentry->d_inode;
6925         struct timespec ctime = CURRENT_TIME;
6926         u64 index = 0;
6927         u64 root_objectid;
6928         int ret;
6929         u64 old_ino = btrfs_ino(old_inode);
6930
6931         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6932                 return -EPERM;
6933
6934         /* we only allow rename subvolume link between subvolumes */
6935         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6936                 return -EXDEV;
6937
6938         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6939             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6940                 return -ENOTEMPTY;
6941
6942         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6943             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6944                 return -ENOTEMPTY;
6945         /*
6946          * we're using rename to replace one file with another.
6947          * and the replacement file is large.  Start IO on it now so
6948          * we don't add too much work to the end of the transaction
6949          */
6950         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6951             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6952                 filemap_flush(old_inode->i_mapping);
6953
6954         /* close the racy window with snapshot create/destroy ioctl */
6955         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6956                 down_read(&root->fs_info->subvol_sem);
6957         /*
6958          * We want to reserve the absolute worst case amount of items.  So if
6959          * both inodes are subvols and we need to unlink them then that would
6960          * require 4 item modifications, but if they are both normal inodes it
6961          * would require 5 item modifications, so we'll assume their normal
6962          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6963          * should cover the worst case number of items we'll modify.
6964          */
6965         trans = btrfs_start_transaction(root, 20);
6966         if (IS_ERR(trans)) {
6967                 ret = PTR_ERR(trans);
6968                 goto out_notrans;
6969         }
6970
6971         if (dest != root)
6972                 btrfs_record_root_in_trans(trans, dest);
6973
6974         ret = btrfs_set_inode_index(new_dir, &index);
6975         if (ret)
6976                 goto out_fail;
6977
6978         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6979                 /* force full log commit if subvolume involved. */
6980                 root->fs_info->last_trans_log_full_commit = trans->transid;
6981         } else {
6982                 ret = btrfs_insert_inode_ref(trans, dest,
6983                                              new_dentry->d_name.name,
6984                                              new_dentry->d_name.len,
6985                                              old_ino,
6986                                              btrfs_ino(new_dir), index);
6987                 if (ret)
6988                         goto out_fail;
6989                 /*
6990                  * this is an ugly little race, but the rename is required
6991                  * to make sure that if we crash, the inode is either at the
6992                  * old name or the new one.  pinning the log transaction lets
6993                  * us make sure we don't allow a log commit to come in after
6994                  * we unlink the name but before we add the new name back in.
6995                  */
6996                 btrfs_pin_log_trans(root);
6997         }
6998         /*
6999          * make sure the inode gets flushed if it is replacing
7000          * something.
7001          */
7002         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7003                 btrfs_add_ordered_operation(trans, root, old_inode);
7004
7005         old_dir->i_ctime = old_dir->i_mtime = ctime;
7006         new_dir->i_ctime = new_dir->i_mtime = ctime;
7007         old_inode->i_ctime = ctime;
7008
7009         if (old_dentry->d_parent != new_dentry->d_parent)
7010                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7011
7012         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7013                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7014                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7015                                         old_dentry->d_name.name,
7016                                         old_dentry->d_name.len);
7017         } else {
7018                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7019                                         old_dentry->d_inode,
7020                                         old_dentry->d_name.name,
7021                                         old_dentry->d_name.len);
7022                 if (!ret)
7023                         ret = btrfs_update_inode(trans, root, old_inode);
7024         }
7025         BUG_ON(ret);
7026
7027         if (new_inode) {
7028                 new_inode->i_ctime = CURRENT_TIME;
7029                 if (unlikely(btrfs_ino(new_inode) ==
7030                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7031                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7032                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7033                                                 root_objectid,
7034                                                 new_dentry->d_name.name,
7035                                                 new_dentry->d_name.len);
7036                         BUG_ON(new_inode->i_nlink == 0);
7037                 } else {
7038                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7039                                                  new_dentry->d_inode,
7040                                                  new_dentry->d_name.name,
7041                                                  new_dentry->d_name.len);
7042                 }
7043                 BUG_ON(ret);
7044                 if (new_inode->i_nlink == 0) {
7045                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7046                         BUG_ON(ret);
7047                 }
7048         }
7049
7050         fixup_inode_flags(new_dir, old_inode);
7051
7052         ret = btrfs_add_link(trans, new_dir, old_inode,
7053                              new_dentry->d_name.name,
7054                              new_dentry->d_name.len, 0, index);
7055         BUG_ON(ret);
7056
7057         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7058                 struct dentry *parent = dget_parent(new_dentry);
7059                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7060                 dput(parent);
7061                 btrfs_end_log_trans(root);
7062         }
7063 out_fail:
7064         btrfs_end_transaction_throttle(trans, root);
7065 out_notrans:
7066         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7067                 up_read(&root->fs_info->subvol_sem);
7068
7069         return ret;
7070 }
7071
7072 /*
7073  * some fairly slow code that needs optimization. This walks the list
7074  * of all the inodes with pending delalloc and forces them to disk.
7075  */
7076 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7077 {
7078         struct list_head *head = &root->fs_info->delalloc_inodes;
7079         struct btrfs_inode *binode;
7080         struct inode *inode;
7081
7082         if (root->fs_info->sb->s_flags & MS_RDONLY)
7083                 return -EROFS;
7084
7085         spin_lock(&root->fs_info->delalloc_lock);
7086         while (!list_empty(head)) {
7087                 binode = list_entry(head->next, struct btrfs_inode,
7088                                     delalloc_inodes);
7089                 inode = igrab(&binode->vfs_inode);
7090                 if (!inode)
7091                         list_del_init(&binode->delalloc_inodes);
7092                 spin_unlock(&root->fs_info->delalloc_lock);
7093                 if (inode) {
7094                         filemap_flush(inode->i_mapping);
7095                         if (delay_iput)
7096                                 btrfs_add_delayed_iput(inode);
7097                         else
7098                                 iput(inode);
7099                 }
7100                 cond_resched();
7101                 spin_lock(&root->fs_info->delalloc_lock);
7102         }
7103         spin_unlock(&root->fs_info->delalloc_lock);
7104
7105         /* the filemap_flush will queue IO into the worker threads, but
7106          * we have to make sure the IO is actually started and that
7107          * ordered extents get created before we return
7108          */
7109         atomic_inc(&root->fs_info->async_submit_draining);
7110         while (atomic_read(&root->fs_info->nr_async_submits) ||
7111               atomic_read(&root->fs_info->async_delalloc_pages)) {
7112                 wait_event(root->fs_info->async_submit_wait,
7113                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7114                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7115         }
7116         atomic_dec(&root->fs_info->async_submit_draining);
7117         return 0;
7118 }
7119
7120 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7121                          const char *symname)
7122 {
7123         struct btrfs_trans_handle *trans;
7124         struct btrfs_root *root = BTRFS_I(dir)->root;
7125         struct btrfs_path *path;
7126         struct btrfs_key key;
7127         struct inode *inode = NULL;
7128         int err;
7129         int drop_inode = 0;
7130         u64 objectid;
7131         u64 index = 0 ;
7132         int name_len;
7133         int datasize;
7134         unsigned long ptr;
7135         struct btrfs_file_extent_item *ei;
7136         struct extent_buffer *leaf;
7137         unsigned long nr = 0;
7138
7139         name_len = strlen(symname) + 1;
7140         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7141                 return -ENAMETOOLONG;
7142
7143         /*
7144          * 2 items for inode item and ref
7145          * 2 items for dir items
7146          * 1 item for xattr if selinux is on
7147          */
7148         trans = btrfs_start_transaction(root, 5);
7149         if (IS_ERR(trans))
7150                 return PTR_ERR(trans);
7151
7152         err = btrfs_find_free_ino(root, &objectid);
7153         if (err)
7154                 goto out_unlock;
7155
7156         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7157                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7158                                 S_IFLNK|S_IRWXUGO, &index);
7159         if (IS_ERR(inode)) {
7160                 err = PTR_ERR(inode);
7161                 goto out_unlock;
7162         }
7163
7164         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7165         if (err) {
7166                 drop_inode = 1;
7167                 goto out_unlock;
7168         }
7169
7170         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7171         if (err)
7172                 drop_inode = 1;
7173         else {
7174                 inode->i_mapping->a_ops = &btrfs_aops;
7175                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7176                 inode->i_fop = &btrfs_file_operations;
7177                 inode->i_op = &btrfs_file_inode_operations;
7178                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7179         }
7180         if (drop_inode)
7181                 goto out_unlock;
7182
7183         path = btrfs_alloc_path();
7184         BUG_ON(!path);
7185         key.objectid = btrfs_ino(inode);
7186         key.offset = 0;
7187         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7188         datasize = btrfs_file_extent_calc_inline_size(name_len);
7189         err = btrfs_insert_empty_item(trans, root, path, &key,
7190                                       datasize);
7191         if (err) {
7192                 drop_inode = 1;
7193                 btrfs_free_path(path);
7194                 goto out_unlock;
7195         }
7196         leaf = path->nodes[0];
7197         ei = btrfs_item_ptr(leaf, path->slots[0],
7198                             struct btrfs_file_extent_item);
7199         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7200         btrfs_set_file_extent_type(leaf, ei,
7201                                    BTRFS_FILE_EXTENT_INLINE);
7202         btrfs_set_file_extent_encryption(leaf, ei, 0);
7203         btrfs_set_file_extent_compression(leaf, ei, 0);
7204         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7205         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7206
7207         ptr = btrfs_file_extent_inline_start(ei);
7208         write_extent_buffer(leaf, symname, ptr, name_len);
7209         btrfs_mark_buffer_dirty(leaf);
7210         btrfs_free_path(path);
7211
7212         inode->i_op = &btrfs_symlink_inode_operations;
7213         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7214         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7215         inode_set_bytes(inode, name_len);
7216         btrfs_i_size_write(inode, name_len - 1);
7217         err = btrfs_update_inode(trans, root, inode);
7218         if (err)
7219                 drop_inode = 1;
7220
7221 out_unlock:
7222         nr = trans->blocks_used;
7223         btrfs_end_transaction_throttle(trans, root);
7224         if (drop_inode) {
7225                 inode_dec_link_count(inode);
7226                 iput(inode);
7227         }
7228         btrfs_btree_balance_dirty(root, nr);
7229         return err;
7230 }
7231
7232 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7233                                        u64 start, u64 num_bytes, u64 min_size,
7234                                        loff_t actual_len, u64 *alloc_hint,
7235                                        struct btrfs_trans_handle *trans)
7236 {
7237         struct btrfs_root *root = BTRFS_I(inode)->root;
7238         struct btrfs_key ins;
7239         u64 cur_offset = start;
7240         u64 i_size;
7241         int ret = 0;
7242         bool own_trans = true;
7243
7244         if (trans)
7245                 own_trans = false;
7246         while (num_bytes > 0) {
7247                 if (own_trans) {
7248                         trans = btrfs_start_transaction(root, 3);
7249                         if (IS_ERR(trans)) {
7250                                 ret = PTR_ERR(trans);
7251                                 break;
7252                         }
7253                 }
7254
7255                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7256                                            0, *alloc_hint, (u64)-1, &ins, 1);
7257                 if (ret) {
7258                         if (own_trans)
7259                                 btrfs_end_transaction(trans, root);
7260                         break;
7261                 }
7262
7263                 ret = insert_reserved_file_extent(trans, inode,
7264                                                   cur_offset, ins.objectid,
7265                                                   ins.offset, ins.offset,
7266                                                   ins.offset, 0, 0, 0,
7267                                                   BTRFS_FILE_EXTENT_PREALLOC);
7268                 BUG_ON(ret);
7269                 btrfs_drop_extent_cache(inode, cur_offset,
7270                                         cur_offset + ins.offset -1, 0);
7271
7272                 num_bytes -= ins.offset;
7273                 cur_offset += ins.offset;
7274                 *alloc_hint = ins.objectid + ins.offset;
7275
7276                 inode->i_ctime = CURRENT_TIME;
7277                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7278                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7279                     (actual_len > inode->i_size) &&
7280                     (cur_offset > inode->i_size)) {
7281                         if (cur_offset > actual_len)
7282                                 i_size = actual_len;
7283                         else
7284                                 i_size = cur_offset;
7285                         i_size_write(inode, i_size);
7286                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7287                 }
7288
7289                 ret = btrfs_update_inode(trans, root, inode);
7290                 BUG_ON(ret);
7291
7292                 if (own_trans)
7293                         btrfs_end_transaction(trans, root);
7294         }
7295         return ret;
7296 }
7297
7298 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7299                               u64 start, u64 num_bytes, u64 min_size,
7300                               loff_t actual_len, u64 *alloc_hint)
7301 {
7302         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7303                                            min_size, actual_len, alloc_hint,
7304                                            NULL);
7305 }
7306
7307 int btrfs_prealloc_file_range_trans(struct inode *inode,
7308                                     struct btrfs_trans_handle *trans, int mode,
7309                                     u64 start, u64 num_bytes, u64 min_size,
7310                                     loff_t actual_len, u64 *alloc_hint)
7311 {
7312         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7313                                            min_size, actual_len, alloc_hint, trans);
7314 }
7315
7316 static int btrfs_set_page_dirty(struct page *page)
7317 {
7318         return __set_page_dirty_nobuffers(page);
7319 }
7320
7321 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7322 {
7323         struct btrfs_root *root = BTRFS_I(inode)->root;
7324
7325         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7326                 return -EROFS;
7327         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7328                 return -EACCES;
7329         return generic_permission(inode, mask, flags, btrfs_check_acl);
7330 }
7331
7332 static const struct inode_operations btrfs_dir_inode_operations = {
7333         .getattr        = btrfs_getattr,
7334         .lookup         = btrfs_lookup,
7335         .create         = btrfs_create,
7336         .unlink         = btrfs_unlink,
7337         .link           = btrfs_link,
7338         .mkdir          = btrfs_mkdir,
7339         .rmdir          = btrfs_rmdir,
7340         .rename         = btrfs_rename,
7341         .symlink        = btrfs_symlink,
7342         .setattr        = btrfs_setattr,
7343         .mknod          = btrfs_mknod,
7344         .setxattr       = btrfs_setxattr,
7345         .getxattr       = btrfs_getxattr,
7346         .listxattr      = btrfs_listxattr,
7347         .removexattr    = btrfs_removexattr,
7348         .permission     = btrfs_permission,
7349 };
7350 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7351         .lookup         = btrfs_lookup,
7352         .permission     = btrfs_permission,
7353 };
7354
7355 static const struct file_operations btrfs_dir_file_operations = {
7356         .llseek         = generic_file_llseek,
7357         .read           = generic_read_dir,
7358         .readdir        = btrfs_real_readdir,
7359         .unlocked_ioctl = btrfs_ioctl,
7360 #ifdef CONFIG_COMPAT
7361         .compat_ioctl   = btrfs_ioctl,
7362 #endif
7363         .release        = btrfs_release_file,
7364         .fsync          = btrfs_sync_file,
7365 };
7366
7367 static struct extent_io_ops btrfs_extent_io_ops = {
7368         .fill_delalloc = run_delalloc_range,
7369         .submit_bio_hook = btrfs_submit_bio_hook,
7370         .merge_bio_hook = btrfs_merge_bio_hook,
7371         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7372         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7373         .writepage_start_hook = btrfs_writepage_start_hook,
7374         .readpage_io_failed_hook = btrfs_io_failed_hook,
7375         .set_bit_hook = btrfs_set_bit_hook,
7376         .clear_bit_hook = btrfs_clear_bit_hook,
7377         .merge_extent_hook = btrfs_merge_extent_hook,
7378         .split_extent_hook = btrfs_split_extent_hook,
7379 };
7380
7381 /*
7382  * btrfs doesn't support the bmap operation because swapfiles
7383  * use bmap to make a mapping of extents in the file.  They assume
7384  * these extents won't change over the life of the file and they
7385  * use the bmap result to do IO directly to the drive.
7386  *
7387  * the btrfs bmap call would return logical addresses that aren't
7388  * suitable for IO and they also will change frequently as COW
7389  * operations happen.  So, swapfile + btrfs == corruption.
7390  *
7391  * For now we're avoiding this by dropping bmap.
7392  */
7393 static const struct address_space_operations btrfs_aops = {
7394         .readpage       = btrfs_readpage,
7395         .writepage      = btrfs_writepage,
7396         .writepages     = btrfs_writepages,
7397         .readpages      = btrfs_readpages,
7398         .direct_IO      = btrfs_direct_IO,
7399         .invalidatepage = btrfs_invalidatepage,
7400         .releasepage    = btrfs_releasepage,
7401         .set_page_dirty = btrfs_set_page_dirty,
7402         .error_remove_page = generic_error_remove_page,
7403 };
7404
7405 static const struct address_space_operations btrfs_symlink_aops = {
7406         .readpage       = btrfs_readpage,
7407         .writepage      = btrfs_writepage,
7408         .invalidatepage = btrfs_invalidatepage,
7409         .releasepage    = btrfs_releasepage,
7410 };
7411
7412 static const struct inode_operations btrfs_file_inode_operations = {
7413         .getattr        = btrfs_getattr,
7414         .setattr        = btrfs_setattr,
7415         .setxattr       = btrfs_setxattr,
7416         .getxattr       = btrfs_getxattr,
7417         .listxattr      = btrfs_listxattr,
7418         .removexattr    = btrfs_removexattr,
7419         .permission     = btrfs_permission,
7420         .fiemap         = btrfs_fiemap,
7421 };
7422 static const struct inode_operations btrfs_special_inode_operations = {
7423         .getattr        = btrfs_getattr,
7424         .setattr        = btrfs_setattr,
7425         .permission     = btrfs_permission,
7426         .setxattr       = btrfs_setxattr,
7427         .getxattr       = btrfs_getxattr,
7428         .listxattr      = btrfs_listxattr,
7429         .removexattr    = btrfs_removexattr,
7430 };
7431 static const struct inode_operations btrfs_symlink_inode_operations = {
7432         .readlink       = generic_readlink,
7433         .follow_link    = page_follow_link_light,
7434         .put_link       = page_put_link,
7435         .getattr        = btrfs_getattr,
7436         .permission     = btrfs_permission,
7437         .setxattr       = btrfs_setxattr,
7438         .getxattr       = btrfs_getxattr,
7439         .listxattr      = btrfs_listxattr,
7440         .removexattr    = btrfs_removexattr,
7441 };
7442
7443 const struct dentry_operations btrfs_dentry_operations = {
7444         .d_delete       = btrfs_dentry_delete,
7445 };