fs/btrfs/inode.c

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