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) {
1790                 ret = btrfs_update_inode(trans, root, inode);
1791                 BUG_ON(ret);
1792         }
1793         ret = 0;
1794 out:
1795         if (nolock) {
1796                 if (trans)
1797                         btrfs_end_transaction_nolock(trans, root);
1798         } else {
1799                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
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 /*
2083  * calculate extra metadata reservation when snapshotting a subvolume
2084  * contains orphan files.
2085  */
2086 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2087                                 struct btrfs_pending_snapshot *pending,
2088                                 u64 *bytes_to_reserve)
2089 {
2090         struct btrfs_root *root;
2091         struct btrfs_block_rsv *block_rsv;
2092         u64 num_bytes;
2093         int index;
2094
2095         root = pending->root;
2096         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2097                 return;
2098
2099         block_rsv = root->orphan_block_rsv;
2100
2101         /* orphan block reservation for the snapshot */
2102         num_bytes = block_rsv->size;
2103
2104         /*
2105          * after the snapshot is created, COWing tree blocks may use more
2106          * space than it frees. So we should make sure there is enough
2107          * reserved space.
2108          */
2109         index = trans->transid & 0x1;
2110         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2111                 num_bytes += block_rsv->size -
2112                              (block_rsv->reserved + block_rsv->freed[index]);
2113         }
2114
2115         *bytes_to_reserve += num_bytes;
2116 }
2117
2118 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2119                                 struct btrfs_pending_snapshot *pending)
2120 {
2121         struct btrfs_root *root = pending->root;
2122         struct btrfs_root *snap = pending->snap;
2123         struct btrfs_block_rsv *block_rsv;
2124         u64 num_bytes;
2125         int index;
2126         int ret;
2127
2128         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2129                 return;
2130
2131         /* refill source subvolume's orphan block reservation */
2132         block_rsv = root->orphan_block_rsv;
2133         index = trans->transid & 0x1;
2134         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2135                 num_bytes = block_rsv->size -
2136                             (block_rsv->reserved + block_rsv->freed[index]);
2137                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2138                                               root->orphan_block_rsv,
2139                                               num_bytes);
2140                 BUG_ON(ret);
2141         }
2142
2143         /* setup orphan block reservation for the snapshot */
2144         block_rsv = btrfs_alloc_block_rsv(snap);
2145         BUG_ON(!block_rsv);
2146
2147         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2148         snap->orphan_block_rsv = block_rsv;
2149
2150         num_bytes = root->orphan_block_rsv->size;
2151         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2152                                       block_rsv, num_bytes);
2153         BUG_ON(ret);
2154
2155 #if 0
2156         /* insert orphan item for the snapshot */
2157         WARN_ON(!root->orphan_item_inserted);
2158         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2159                                        snap->root_key.objectid);
2160         BUG_ON(ret);
2161         snap->orphan_item_inserted = 1;
2162 #endif
2163 }
2164
2165 enum btrfs_orphan_cleanup_state {
2166         ORPHAN_CLEANUP_STARTED  = 1,
2167         ORPHAN_CLEANUP_DONE     = 2,
2168 };
2169
2170 /*
2171  * This is called in transaction commmit time. If there are no orphan
2172  * files in the subvolume, it removes orphan item and frees block_rsv
2173  * structure.
2174  */
2175 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2176                               struct btrfs_root *root)
2177 {
2178         int ret;
2179
2180         if (!list_empty(&root->orphan_list) ||
2181             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2182                 return;
2183
2184         if (root->orphan_item_inserted &&
2185             btrfs_root_refs(&root->root_item) > 0) {
2186                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187                                             root->root_key.objectid);
2188                 BUG_ON(ret);
2189                 root->orphan_item_inserted = 0;
2190         }
2191
2192         if (root->orphan_block_rsv) {
2193                 WARN_ON(root->orphan_block_rsv->size > 0);
2194                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2195                 root->orphan_block_rsv = NULL;
2196         }
2197 }
2198
2199 /*
2200  * This creates an orphan entry for the given inode in case something goes
2201  * wrong in the middle of an unlink/truncate.
2202  *
2203  * NOTE: caller of this function should reserve 5 units of metadata for
2204  *       this function.
2205  */
2206 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2207 {
2208         struct btrfs_root *root = BTRFS_I(inode)->root;
2209         struct btrfs_block_rsv *block_rsv = NULL;
2210         int reserve = 0;
2211         int insert = 0;
2212         int ret;
2213
2214         if (!root->orphan_block_rsv) {
2215                 block_rsv = btrfs_alloc_block_rsv(root);
2216                 if (!block_rsv)
2217                         return -ENOMEM;
2218         }
2219
2220         spin_lock(&root->orphan_lock);
2221         if (!root->orphan_block_rsv) {
2222                 root->orphan_block_rsv = block_rsv;
2223         } else if (block_rsv) {
2224                 btrfs_free_block_rsv(root, block_rsv);
2225                 block_rsv = NULL;
2226         }
2227
2228         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2229                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2230 #if 0
2231                 /*
2232                  * For proper ENOSPC handling, we should do orphan
2233                  * cleanup when mounting. But this introduces backward
2234                  * compatibility issue.
2235                  */
2236                 if (!xchg(&root->orphan_item_inserted, 1))
2237                         insert = 2;
2238                 else
2239                         insert = 1;
2240 #endif
2241                 insert = 1;
2242         }
2243
2244         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2245                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2246                 reserve = 1;
2247         }
2248         spin_unlock(&root->orphan_lock);
2249
2250         if (block_rsv)
2251                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2252
2253         /* grab metadata reservation from transaction handle */
2254         if (reserve) {
2255                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2256                 BUG_ON(ret);
2257         }
2258
2259         /* insert an orphan item to track this unlinked/truncated file */
2260         if (insert >= 1) {
2261                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2262                 BUG_ON(ret);
2263         }
2264
2265         /* insert an orphan item to track subvolume contains orphan files */
2266         if (insert >= 2) {
2267                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2268                                                root->root_key.objectid);
2269                 BUG_ON(ret);
2270         }
2271         return 0;
2272 }
2273
2274 /*
2275  * We have done the truncate/delete so we can go ahead and remove the orphan
2276  * item for this particular inode.
2277  */
2278 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2279 {
2280         struct btrfs_root *root = BTRFS_I(inode)->root;
2281         int delete_item = 0;
2282         int release_rsv = 0;
2283         int ret = 0;
2284
2285         spin_lock(&root->orphan_lock);
2286         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2287                 list_del_init(&BTRFS_I(inode)->i_orphan);
2288                 delete_item = 1;
2289         }
2290
2291         if (BTRFS_I(inode)->orphan_meta_reserved) {
2292                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2293                 release_rsv = 1;
2294         }
2295         spin_unlock(&root->orphan_lock);
2296
2297         if (trans && delete_item) {
2298                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2299                 BUG_ON(ret);
2300         }
2301
2302         if (release_rsv)
2303                 btrfs_orphan_release_metadata(inode);
2304
2305         return 0;
2306 }
2307
2308 /*
2309  * this cleans up any orphans that may be left on the list from the last use
2310  * of this root.
2311  */
2312 int btrfs_orphan_cleanup(struct btrfs_root *root)
2313 {
2314         struct btrfs_path *path;
2315         struct extent_buffer *leaf;
2316         struct btrfs_key key, found_key;
2317         struct btrfs_trans_handle *trans;
2318         struct inode *inode;
2319         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2320
2321         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2322                 return 0;
2323
2324         path = btrfs_alloc_path();
2325         if (!path) {
2326                 ret = -ENOMEM;
2327                 goto out;
2328         }
2329         path->reada = -1;
2330
2331         key.objectid = BTRFS_ORPHAN_OBJECTID;
2332         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2333         key.offset = (u64)-1;
2334
2335         while (1) {
2336                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2337                 if (ret < 0)
2338                         goto out;
2339
2340                 /*
2341                  * if ret == 0 means we found what we were searching for, which
2342                  * is weird, but possible, so only screw with path if we didn't
2343                  * find the key and see if we have stuff that matches
2344                  */
2345                 if (ret > 0) {
2346                         ret = 0;
2347                         if (path->slots[0] == 0)
2348                                 break;
2349                         path->slots[0]--;
2350                 }
2351
2352                 /* pull out the item */
2353                 leaf = path->nodes[0];
2354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2355
2356                 /* make sure the item matches what we want */
2357                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2358                         break;
2359                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2360                         break;
2361
2362                 /* release the path since we're done with it */
2363                 btrfs_release_path(path);
2364
2365                 /*
2366                  * this is where we are basically btrfs_lookup, without the
2367                  * crossing root thing.  we store the inode number in the
2368                  * offset of the orphan item.
2369                  */
2370                 found_key.objectid = found_key.offset;
2371                 found_key.type = BTRFS_INODE_ITEM_KEY;
2372                 found_key.offset = 0;
2373                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2374                 if (IS_ERR(inode)) {
2375                         ret = PTR_ERR(inode);
2376                         goto out;
2377                 }
2378
2379                 /*
2380                  * add this inode to the orphan list so btrfs_orphan_del does
2381                  * the proper thing when we hit it
2382                  */
2383                 spin_lock(&root->orphan_lock);
2384                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2385                 spin_unlock(&root->orphan_lock);
2386
2387                 /*
2388                  * if this is a bad inode, means we actually succeeded in
2389                  * removing the inode, but not the orphan record, which means
2390                  * we need to manually delete the orphan since iput will just
2391                  * do a destroy_inode
2392                  */
2393                 if (is_bad_inode(inode)) {
2394                         trans = btrfs_start_transaction(root, 0);
2395                         if (IS_ERR(trans)) {
2396                                 ret = PTR_ERR(trans);
2397                                 goto out;
2398                         }
2399                         btrfs_orphan_del(trans, inode);
2400                         btrfs_end_transaction(trans, root);
2401                         iput(inode);
2402                         continue;
2403                 }
2404
2405                 /* if we have links, this was a truncate, lets do that */
2406                 if (inode->i_nlink) {
2407                         if (!S_ISREG(inode->i_mode)) {
2408                                 WARN_ON(1);
2409                                 iput(inode);
2410                                 continue;
2411                         }
2412                         nr_truncate++;
2413                         ret = btrfs_truncate(inode);
2414                 } else {
2415                         nr_unlink++;
2416                 }
2417
2418                 /* this will do delete_inode and everything for us */
2419                 iput(inode);
2420                 if (ret)
2421                         goto out;
2422         }
2423         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2424
2425         if (root->orphan_block_rsv)
2426                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2427                                         (u64)-1);
2428
2429         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2430                 trans = btrfs_join_transaction(root);
2431                 if (!IS_ERR(trans))
2432                         btrfs_end_transaction(trans, root);
2433         }
2434
2435         if (nr_unlink)
2436                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2437         if (nr_truncate)
2438                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2439
2440 out:
2441         if (ret)
2442                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2443         btrfs_free_path(path);
2444         return ret;
2445 }
2446
2447 /*
2448  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2449  * don't find any xattrs, we know there can't be any acls.
2450  *
2451  * slot is the slot the inode is in, objectid is the objectid of the inode
2452  */
2453 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2454                                           int slot, u64 objectid)
2455 {
2456         u32 nritems = btrfs_header_nritems(leaf);
2457         struct btrfs_key found_key;
2458         int scanned = 0;
2459
2460         slot++;
2461         while (slot < nritems) {
2462                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2463
2464                 /* we found a different objectid, there must not be acls */
2465                 if (found_key.objectid != objectid)
2466                         return 0;
2467
2468                 /* we found an xattr, assume we've got an acl */
2469                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2470                         return 1;
2471
2472                 /*
2473                  * we found a key greater than an xattr key, there can't
2474                  * be any acls later on
2475                  */
2476                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2477                         return 0;
2478
2479                 slot++;
2480                 scanned++;
2481
2482                 /*
2483                  * it goes inode, inode backrefs, xattrs, extents,
2484                  * so if there are a ton of hard links to an inode there can
2485                  * be a lot of backrefs.  Don't waste time searching too hard,
2486                  * this is just an optimization
2487                  */
2488                 if (scanned >= 8)
2489                         break;
2490         }
2491         /* we hit the end of the leaf before we found an xattr or
2492          * something larger than an xattr.  We have to assume the inode
2493          * has acls
2494          */
2495         return 1;
2496 }
2497
2498 /*
2499  * read an inode from the btree into the in-memory inode
2500  */
2501 static void btrfs_read_locked_inode(struct inode *inode)
2502 {
2503         struct btrfs_path *path;
2504         struct extent_buffer *leaf;
2505         struct btrfs_inode_item *inode_item;
2506         struct btrfs_timespec *tspec;
2507         struct btrfs_root *root = BTRFS_I(inode)->root;
2508         struct btrfs_key location;
2509         int maybe_acls;
2510         u32 rdev;
2511         int ret;
2512         bool filled = false;
2513
2514         ret = btrfs_fill_inode(inode, &rdev);
2515         if (!ret)
2516                 filled = true;
2517
2518         path = btrfs_alloc_path();
2519         if (!path)
2520                 goto make_bad;
2521
2522         path->leave_spinning = 1;
2523         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2524
2525         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2526         if (ret)
2527                 goto make_bad;
2528
2529         leaf = path->nodes[0];
2530
2531         if (filled)
2532                 goto cache_acl;
2533
2534         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2535                                     struct btrfs_inode_item);
2536         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2537         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2538         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2539         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2540         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2541
2542         tspec = btrfs_inode_atime(inode_item);
2543         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2544         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2545
2546         tspec = btrfs_inode_mtime(inode_item);
2547         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2548         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2549
2550         tspec = btrfs_inode_ctime(inode_item);
2551         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2552         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2553
2554         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2555         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2556         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2557         inode->i_generation = BTRFS_I(inode)->generation;
2558         inode->i_rdev = 0;
2559         rdev = btrfs_inode_rdev(leaf, inode_item);
2560
2561         BTRFS_I(inode)->index_cnt = (u64)-1;
2562         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2563 cache_acl:
2564         /*
2565          * try to precache a NULL acl entry for files that don't have
2566          * any xattrs or acls
2567          */
2568         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2569                                            btrfs_ino(inode));
2570         if (!maybe_acls)
2571                 cache_no_acl(inode);
2572
2573         btrfs_free_path(path);
2574
2575         switch (inode->i_mode & S_IFMT) {
2576         case S_IFREG:
2577                 inode->i_mapping->a_ops = &btrfs_aops;
2578                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2579                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2580                 inode->i_fop = &btrfs_file_operations;
2581                 inode->i_op = &btrfs_file_inode_operations;
2582                 break;
2583         case S_IFDIR:
2584                 inode->i_fop = &btrfs_dir_file_operations;
2585                 if (root == root->fs_info->tree_root)
2586                         inode->i_op = &btrfs_dir_ro_inode_operations;
2587                 else
2588                         inode->i_op = &btrfs_dir_inode_operations;
2589                 break;
2590         case S_IFLNK:
2591                 inode->i_op = &btrfs_symlink_inode_operations;
2592                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2593                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2594                 break;
2595         default:
2596                 inode->i_op = &btrfs_special_inode_operations;
2597                 init_special_inode(inode, inode->i_mode, rdev);
2598                 break;
2599         }
2600
2601         btrfs_update_iflags(inode);
2602         return;
2603
2604 make_bad:
2605         btrfs_free_path(path);
2606         make_bad_inode(inode);
2607 }
2608
2609 /*
2610  * given a leaf and an inode, copy the inode fields into the leaf
2611  */
2612 static void fill_inode_item(struct btrfs_trans_handle *trans,
2613                             struct extent_buffer *leaf,
2614                             struct btrfs_inode_item *item,
2615                             struct inode *inode)
2616 {
2617         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2618         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2619         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2620         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2621         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2622
2623         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2624                                inode->i_atime.tv_sec);
2625         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2626                                 inode->i_atime.tv_nsec);
2627
2628         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2629                                inode->i_mtime.tv_sec);
2630         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2631                                 inode->i_mtime.tv_nsec);
2632
2633         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2634                                inode->i_ctime.tv_sec);
2635         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2636                                 inode->i_ctime.tv_nsec);
2637
2638         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2639         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2640         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2641         btrfs_set_inode_transid(leaf, item, trans->transid);
2642         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2643         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2644         btrfs_set_inode_block_group(leaf, item, 0);
2645 }
2646
2647 /*
2648  * copy everything in the in-memory inode into the btree.
2649  */
2650 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2651                                 struct btrfs_root *root, struct inode *inode)
2652 {
2653         struct btrfs_inode_item *inode_item;
2654         struct btrfs_path *path;
2655         struct extent_buffer *leaf;
2656         int ret;
2657
2658         /*
2659          * If the inode is a free space inode, we can deadlock during commit
2660          * if we put it into the delayed code.
2661          *
2662          * The data relocation inode should also be directly updated
2663          * without delay
2664          */
2665         if (!btrfs_is_free_space_inode(root, inode)
2666             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2667                 ret = btrfs_delayed_update_inode(trans, root, inode);
2668                 if (!ret)
2669                         btrfs_set_inode_last_trans(trans, inode);
2670                 return ret;
2671         }
2672
2673         path = btrfs_alloc_path();
2674         if (!path)
2675                 return -ENOMEM;
2676
2677         path->leave_spinning = 1;
2678         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2679                                  1);
2680         if (ret) {
2681                 if (ret > 0)
2682                         ret = -ENOENT;
2683                 goto failed;
2684         }
2685
2686         btrfs_unlock_up_safe(path, 1);
2687         leaf = path->nodes[0];
2688         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2689                                     struct btrfs_inode_item);
2690
2691         fill_inode_item(trans, leaf, inode_item, inode);
2692         btrfs_mark_buffer_dirty(leaf);
2693         btrfs_set_inode_last_trans(trans, inode);
2694         ret = 0;
2695 failed:
2696         btrfs_free_path(path);
2697         return ret;
2698 }
2699
2700 /*
2701  * unlink helper that gets used here in inode.c and in the tree logging
2702  * recovery code.  It remove a link in a directory with a given name, and
2703  * also drops the back refs in the inode to the directory
2704  */
2705 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2706                                 struct btrfs_root *root,
2707                                 struct inode *dir, struct inode *inode,
2708                                 const char *name, int name_len)
2709 {
2710         struct btrfs_path *path;
2711         int ret = 0;
2712         struct extent_buffer *leaf;
2713         struct btrfs_dir_item *di;
2714         struct btrfs_key key;
2715         u64 index;
2716         u64 ino = btrfs_ino(inode);
2717         u64 dir_ino = btrfs_ino(dir);
2718
2719         path = btrfs_alloc_path();
2720         if (!path) {
2721                 ret = -ENOMEM;
2722                 goto out;
2723         }
2724
2725         path->leave_spinning = 1;
2726         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2727                                     name, name_len, -1);
2728         if (IS_ERR(di)) {
2729                 ret = PTR_ERR(di);
2730                 goto err;
2731         }
2732         if (!di) {
2733                 ret = -ENOENT;
2734                 goto err;
2735         }
2736         leaf = path->nodes[0];
2737         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2738         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2739         if (ret)
2740                 goto err;
2741         btrfs_release_path(path);
2742
2743         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2744                                   dir_ino, &index);
2745         if (ret) {
2746                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2747                        "inode %llu parent %llu\n", name_len, name,
2748                        (unsigned long long)ino, (unsigned long long)dir_ino);
2749                 goto err;
2750         }
2751
2752         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2753         if (ret)
2754                 goto err;
2755
2756         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2757                                          inode, dir_ino);
2758         BUG_ON(ret != 0 && ret != -ENOENT);
2759
2760         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2761                                            dir, index);
2762         if (ret == -ENOENT)
2763                 ret = 0;
2764 err:
2765         btrfs_free_path(path);
2766         if (ret)
2767                 goto out;
2768
2769         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2770         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2771         btrfs_update_inode(trans, root, dir);
2772 out:
2773         return ret;
2774 }
2775
2776 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2777                        struct btrfs_root *root,
2778                        struct inode *dir, struct inode *inode,
2779                        const char *name, int name_len)
2780 {
2781         int ret;
2782         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2783         if (!ret) {
2784                 btrfs_drop_nlink(inode);
2785                 ret = btrfs_update_inode(trans, root, inode);
2786         }
2787         return ret;
2788 }
2789
2790
2791 /* helper to check if there is any shared block in the path */
2792 static int check_path_shared(struct btrfs_root *root,
2793                              struct btrfs_path *path)
2794 {
2795         struct extent_buffer *eb;
2796         int level;
2797         u64 refs = 1;
2798
2799         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2800                 int ret;
2801
2802                 if (!path->nodes[level])
2803                         break;
2804                 eb = path->nodes[level];
2805                 if (!btrfs_block_can_be_shared(root, eb))
2806                         continue;
2807                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2808                                                &refs, NULL);
2809                 if (refs > 1)
2810                         return 1;
2811         }
2812         return 0;
2813 }
2814
2815 /*
2816  * helper to start transaction for unlink and rmdir.
2817  *
2818  * unlink and rmdir are special in btrfs, they do not always free space.
2819  * so in enospc case, we should make sure they will free space before
2820  * allowing them to use the global metadata reservation.
2821  */
2822 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2823                                                        struct dentry *dentry)
2824 {
2825         struct btrfs_trans_handle *trans;
2826         struct btrfs_root *root = BTRFS_I(dir)->root;
2827         struct btrfs_path *path;
2828         struct btrfs_inode_ref *ref;
2829         struct btrfs_dir_item *di;
2830         struct inode *inode = dentry->d_inode;
2831         u64 index;
2832         int check_link = 1;
2833         int err = -ENOSPC;
2834         int ret;
2835         u64 ino = btrfs_ino(inode);
2836         u64 dir_ino = btrfs_ino(dir);
2837
2838         trans = btrfs_start_transaction(root, 10);
2839         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2840                 return trans;
2841
2842         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2843                 return ERR_PTR(-ENOSPC);
2844
2845         /* check if there is someone else holds reference */
2846         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2847                 return ERR_PTR(-ENOSPC);
2848
2849         if (atomic_read(&inode->i_count) > 2)
2850                 return ERR_PTR(-ENOSPC);
2851
2852         if (xchg(&root->fs_info->enospc_unlink, 1))
2853                 return ERR_PTR(-ENOSPC);
2854
2855         path = btrfs_alloc_path();
2856         if (!path) {
2857                 root->fs_info->enospc_unlink = 0;
2858                 return ERR_PTR(-ENOMEM);
2859         }
2860
2861         trans = btrfs_start_transaction(root, 0);
2862         if (IS_ERR(trans)) {
2863                 btrfs_free_path(path);
2864                 root->fs_info->enospc_unlink = 0;
2865                 return trans;
2866         }
2867
2868         path->skip_locking = 1;
2869         path->search_commit_root = 1;
2870
2871         ret = btrfs_lookup_inode(trans, root, path,
2872                                 &BTRFS_I(dir)->location, 0);
2873         if (ret < 0) {
2874                 err = ret;
2875                 goto out;
2876         }
2877         if (ret == 0) {
2878                 if (check_path_shared(root, path))
2879                         goto out;
2880         } else {
2881                 check_link = 0;
2882         }
2883         btrfs_release_path(path);
2884
2885         ret = btrfs_lookup_inode(trans, root, path,
2886                                 &BTRFS_I(inode)->location, 0);
2887         if (ret < 0) {
2888                 err = ret;
2889                 goto out;
2890         }
2891         if (ret == 0) {
2892                 if (check_path_shared(root, path))
2893                         goto out;
2894         } else {
2895                 check_link = 0;
2896         }
2897         btrfs_release_path(path);
2898
2899         if (ret == 0 && S_ISREG(inode->i_mode)) {
2900                 ret = btrfs_lookup_file_extent(trans, root, path,
2901                                                ino, (u64)-1, 0);
2902                 if (ret < 0) {
2903                         err = ret;
2904                         goto out;
2905                 }
2906                 BUG_ON(ret == 0);
2907                 if (check_path_shared(root, path))
2908                         goto out;
2909                 btrfs_release_path(path);
2910         }
2911
2912         if (!check_link) {
2913                 err = 0;
2914                 goto out;
2915         }
2916
2917         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2918                                 dentry->d_name.name, dentry->d_name.len, 0);
2919         if (IS_ERR(di)) {
2920                 err = PTR_ERR(di);
2921                 goto out;
2922         }
2923         if (di) {
2924                 if (check_path_shared(root, path))
2925                         goto out;
2926         } else {
2927                 err = 0;
2928                 goto out;
2929         }
2930         btrfs_release_path(path);
2931
2932         ref = btrfs_lookup_inode_ref(trans, root, path,
2933                                 dentry->d_name.name, dentry->d_name.len,
2934                                 ino, dir_ino, 0);
2935         if (IS_ERR(ref)) {
2936                 err = PTR_ERR(ref);
2937                 goto out;
2938         }
2939         BUG_ON(!ref);
2940         if (check_path_shared(root, path))
2941                 goto out;
2942         index = btrfs_inode_ref_index(path->nodes[0], ref);
2943         btrfs_release_path(path);
2944
2945         /*
2946          * This is a commit root search, if we can lookup inode item and other
2947          * relative items in the commit root, it means the transaction of
2948          * dir/file creation has been committed, and the dir index item that we
2949          * delay to insert has also been inserted into the commit root. So
2950          * we needn't worry about the delayed insertion of the dir index item
2951          * here.
2952          */
2953         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2954                                 dentry->d_name.name, dentry->d_name.len, 0);
2955         if (IS_ERR(di)) {
2956                 err = PTR_ERR(di);
2957                 goto out;
2958         }
2959         BUG_ON(ret == -ENOENT);
2960         if (check_path_shared(root, path))
2961                 goto out;
2962
2963         err = 0;
2964 out:
2965         btrfs_free_path(path);
2966         if (err) {
2967                 btrfs_end_transaction(trans, root);
2968                 root->fs_info->enospc_unlink = 0;
2969                 return ERR_PTR(err);
2970         }
2971
2972         trans->block_rsv = &root->fs_info->global_block_rsv;
2973         return trans;
2974 }
2975
2976 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2977                                struct btrfs_root *root)
2978 {
2979         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2980                 BUG_ON(!root->fs_info->enospc_unlink);
2981                 root->fs_info->enospc_unlink = 0;
2982         }
2983         btrfs_end_transaction_throttle(trans, root);
2984 }
2985
2986 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2987 {
2988         struct btrfs_root *root = BTRFS_I(dir)->root;
2989         struct btrfs_trans_handle *trans;
2990         struct inode *inode = dentry->d_inode;
2991         int ret;
2992         unsigned long nr = 0;
2993
2994         trans = __unlink_start_trans(dir, dentry);
2995         if (IS_ERR(trans))
2996                 return PTR_ERR(trans);
2997
2998         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2999
3000         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3001                                  dentry->d_name.name, dentry->d_name.len);
3002         if (ret)
3003                 goto out;
3004
3005         if (inode->i_nlink == 0) {
3006                 ret = btrfs_orphan_add(trans, inode);
3007                 if (ret)
3008                         goto out;
3009         }
3010
3011 out:
3012         nr = trans->blocks_used;
3013         __unlink_end_trans(trans, root);
3014         btrfs_btree_balance_dirty(root, nr);
3015         return ret;
3016 }
3017
3018 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3019                         struct btrfs_root *root,
3020                         struct inode *dir, u64 objectid,
3021                         const char *name, int name_len)
3022 {
3023         struct btrfs_path *path;
3024         struct extent_buffer *leaf;
3025         struct btrfs_dir_item *di;
3026         struct btrfs_key key;
3027         u64 index;
3028         int ret;
3029         u64 dir_ino = btrfs_ino(dir);
3030
3031         path = btrfs_alloc_path();
3032         if (!path)
3033                 return -ENOMEM;
3034
3035         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3036                                    name, name_len, -1);
3037         BUG_ON(IS_ERR_OR_NULL(di));
3038
3039         leaf = path->nodes[0];
3040         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3041         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3042         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3043         BUG_ON(ret);
3044         btrfs_release_path(path);
3045
3046         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3047                                  objectid, root->root_key.objectid,
3048                                  dir_ino, &index, name, name_len);
3049         if (ret < 0) {
3050                 BUG_ON(ret != -ENOENT);
3051                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3052                                                  name, name_len);
3053                 BUG_ON(IS_ERR_OR_NULL(di));
3054
3055                 leaf = path->nodes[0];
3056                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3057                 btrfs_release_path(path);
3058                 index = key.offset;
3059         }
3060         btrfs_release_path(path);
3061
3062         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3063         BUG_ON(ret);
3064
3065         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3066         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3067         ret = btrfs_update_inode(trans, root, dir);
3068         BUG_ON(ret);
3069
3070         btrfs_free_path(path);
3071         return 0;
3072 }
3073
3074 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3075 {
3076         struct inode *inode = dentry->d_inode;
3077         int err = 0;
3078         struct btrfs_root *root = BTRFS_I(dir)->root;
3079         struct btrfs_trans_handle *trans;
3080         unsigned long nr = 0;
3081
3082         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3083             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3084                 return -ENOTEMPTY;
3085
3086         trans = __unlink_start_trans(dir, dentry);
3087         if (IS_ERR(trans))
3088                 return PTR_ERR(trans);
3089
3090         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3091                 err = btrfs_unlink_subvol(trans, root, dir,
3092                                           BTRFS_I(inode)->location.objectid,
3093                                           dentry->d_name.name,
3094                                           dentry->d_name.len);
3095                 goto out;
3096         }
3097
3098         err = btrfs_orphan_add(trans, inode);
3099         if (err)
3100                 goto out;
3101
3102         /* now the directory is empty */
3103         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3104                                  dentry->d_name.name, dentry->d_name.len);
3105         if (!err)
3106                 btrfs_i_size_write(inode, 0);
3107 out:
3108         nr = trans->blocks_used;
3109         __unlink_end_trans(trans, root);
3110         btrfs_btree_balance_dirty(root, nr);
3111
3112         return err;
3113 }
3114
3115 /*
3116  * this can truncate away extent items, csum items and directory items.
3117  * It starts at a high offset and removes keys until it can't find
3118  * any higher than new_size
3119  *
3120  * csum items that cross the new i_size are truncated to the new size
3121  * as well.
3122  *
3123  * min_type is the minimum key type to truncate down to.  If set to 0, this
3124  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3125  */
3126 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3127                                struct btrfs_root *root,
3128                                struct inode *inode,
3129                                u64 new_size, u32 min_type)
3130 {
3131         struct btrfs_path *path;
3132         struct extent_buffer *leaf;
3133         struct btrfs_file_extent_item *fi;
3134         struct btrfs_key key;
3135         struct btrfs_key found_key;
3136         u64 extent_start = 0;
3137         u64 extent_num_bytes = 0;
3138         u64 extent_offset = 0;
3139         u64 item_end = 0;
3140         u64 mask = root->sectorsize - 1;
3141         u32 found_type = (u8)-1;
3142         int found_extent;
3143         int del_item;
3144         int pending_del_nr = 0;
3145         int pending_del_slot = 0;
3146         int extent_type = -1;
3147         int encoding;
3148         int ret;
3149         int err = 0;
3150         u64 ino = btrfs_ino(inode);
3151
3152         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3153
3154         path = btrfs_alloc_path();
3155         if (!path)
3156                 return -ENOMEM;
3157         path->reada = -1;
3158
3159         if (root->ref_cows || root == root->fs_info->tree_root)
3160                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3161
3162         /*
3163          * This function is also used to drop the items in the log tree before
3164          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3165          * it is used to drop the loged items. So we shouldn't kill the delayed
3166          * items.
3167          */
3168         if (min_type == 0 && root == BTRFS_I(inode)->root)
3169                 btrfs_kill_delayed_inode_items(inode);
3170
3171         key.objectid = ino;
3172         key.offset = (u64)-1;
3173         key.type = (u8)-1;
3174
3175 search_again:
3176         path->leave_spinning = 1;
3177         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3178         if (ret < 0) {
3179                 err = ret;
3180                 goto out;
3181         }
3182
3183         if (ret > 0) {
3184                 /* there are no items in the tree for us to truncate, we're
3185                  * done
3186                  */
3187                 if (path->slots[0] == 0)
3188                         goto out;
3189                 path->slots[0]--;
3190         }
3191
3192         while (1) {
3193                 fi = NULL;
3194                 leaf = path->nodes[0];
3195                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3196                 found_type = btrfs_key_type(&found_key);
3197                 encoding = 0;
3198
3199                 if (found_key.objectid != ino)
3200                         break;
3201
3202                 if (found_type < min_type)
3203                         break;
3204
3205                 item_end = found_key.offset;
3206                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3207                         fi = btrfs_item_ptr(leaf, path->slots[0],
3208                                             struct btrfs_file_extent_item);
3209                         extent_type = btrfs_file_extent_type(leaf, fi);
3210                         encoding = btrfs_file_extent_compression(leaf, fi);
3211                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3212                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3213
3214                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3215                                 item_end +=
3216                                     btrfs_file_extent_num_bytes(leaf, fi);
3217                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3218                                 item_end += btrfs_file_extent_inline_len(leaf,
3219                                                                          fi);
3220                         }
3221                         item_end--;
3222                 }
3223                 if (found_type > min_type) {
3224                         del_item = 1;
3225                 } else {
3226                         if (item_end < new_size)
3227                                 break;
3228                         if (found_key.offset >= new_size)
3229                                 del_item = 1;
3230                         else
3231                                 del_item = 0;
3232                 }
3233                 found_extent = 0;
3234                 /* FIXME, shrink the extent if the ref count is only 1 */
3235                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3236                         goto delete;
3237
3238                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3239                         u64 num_dec;
3240                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3241                         if (!del_item && !encoding) {
3242                                 u64 orig_num_bytes =
3243                                         btrfs_file_extent_num_bytes(leaf, fi);
3244                                 extent_num_bytes = new_size -
3245                                         found_key.offset + root->sectorsize - 1;
3246                                 extent_num_bytes = extent_num_bytes &
3247                                         ~((u64)root->sectorsize - 1);
3248                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3249                                                          extent_num_bytes);
3250                                 num_dec = (orig_num_bytes -
3251                                            extent_num_bytes);
3252                                 if (root->ref_cows && extent_start != 0)
3253                                         inode_sub_bytes(inode, num_dec);
3254                                 btrfs_mark_buffer_dirty(leaf);
3255                         } else {
3256                                 extent_num_bytes =
3257                                         btrfs_file_extent_disk_num_bytes(leaf,
3258                                                                          fi);
3259                                 extent_offset = found_key.offset -
3260                                         btrfs_file_extent_offset(leaf, fi);
3261
3262                                 /* FIXME blocksize != 4096 */
3263                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3264                                 if (extent_start != 0) {
3265                                         found_extent = 1;
3266                                         if (root->ref_cows)
3267                                                 inode_sub_bytes(inode, num_dec);
3268                                 }
3269                         }
3270                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3271                         /*
3272                          * we can't truncate inline items that have had
3273                          * special encodings
3274                          */
3275                         if (!del_item &&
3276                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3277                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3278                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3279                                 u32 size = new_size - found_key.offset;
3280
3281                                 if (root->ref_cows) {
3282                                         inode_sub_bytes(inode, item_end + 1 -
3283                                                         new_size);
3284                                 }
3285                                 size =
3286                                     btrfs_file_extent_calc_inline_size(size);
3287                                 ret = btrfs_truncate_item(trans, root, path,
3288                                                           size, 1);
3289                         } else if (root->ref_cows) {
3290                                 inode_sub_bytes(inode, item_end + 1 -
3291                                                 found_key.offset);
3292                         }
3293                 }
3294 delete:
3295                 if (del_item) {
3296                         if (!pending_del_nr) {
3297                                 /* no pending yet, add ourselves */
3298                                 pending_del_slot = path->slots[0];
3299                                 pending_del_nr = 1;
3300                         } else if (pending_del_nr &&
3301                                    path->slots[0] + 1 == pending_del_slot) {
3302                                 /* hop on the pending chunk */
3303                                 pending_del_nr++;
3304                                 pending_del_slot = path->slots[0];
3305                         } else {
3306                                 BUG();
3307                         }
3308                 } else {
3309                         break;
3310                 }
3311                 if (found_extent && (root->ref_cows ||
3312                                      root == root->fs_info->tree_root)) {
3313                         btrfs_set_path_blocking(path);
3314                         ret = btrfs_free_extent(trans, root, extent_start,
3315                                                 extent_num_bytes, 0,
3316                                                 btrfs_header_owner(leaf),
3317                                                 ino, extent_offset);
3318                         BUG_ON(ret);
3319                 }
3320
3321                 if (found_type == BTRFS_INODE_ITEM_KEY)
3322                         break;
3323
3324                 if (path->slots[0] == 0 ||
3325                     path->slots[0] != pending_del_slot) {
3326                         if (root->ref_cows &&
3327                             BTRFS_I(inode)->location.objectid !=
3328                                                 BTRFS_FREE_INO_OBJECTID) {
3329                                 err = -EAGAIN;
3330                                 goto out;
3331                         }
3332                         if (pending_del_nr) {
3333                                 ret = btrfs_del_items(trans, root, path,
3334                                                 pending_del_slot,
3335                                                 pending_del_nr);
3336                                 BUG_ON(ret);
3337                                 pending_del_nr = 0;
3338                         }
3339                         btrfs_release_path(path);
3340                         goto search_again;
3341                 } else {
3342                         path->slots[0]--;
3343                 }
3344         }
3345 out:
3346         if (pending_del_nr) {
3347                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3348                                       pending_del_nr);
3349                 BUG_ON(ret);
3350         }
3351         btrfs_free_path(path);
3352         return err;
3353 }
3354
3355 /*
3356  * taken from block_truncate_page, but does cow as it zeros out
3357  * any bytes left in the last page in the file.
3358  */
3359 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3360 {
3361         struct inode *inode = mapping->host;
3362         struct btrfs_root *root = BTRFS_I(inode)->root;
3363         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3364         struct btrfs_ordered_extent *ordered;
3365         struct extent_state *cached_state = NULL;
3366         char *kaddr;
3367         u32 blocksize = root->sectorsize;
3368         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3369         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3370         struct page *page;
3371         int ret = 0;
3372         u64 page_start;
3373         u64 page_end;
3374
3375         if ((offset & (blocksize - 1)) == 0)
3376                 goto out;
3377         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3378         if (ret)
3379                 goto out;
3380
3381         ret = -ENOMEM;
3382 again:
3383         page = find_or_create_page(mapping, index, GFP_NOFS);
3384         if (!page) {
3385                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3386                 goto out;
3387         }
3388
3389         page_start = page_offset(page);
3390         page_end = page_start + PAGE_CACHE_SIZE - 1;
3391
3392         if (!PageUptodate(page)) {
3393                 ret = btrfs_readpage(NULL, page);
3394                 lock_page(page);
3395                 if (page->mapping != mapping) {
3396                         unlock_page(page);
3397                         page_cache_release(page);
3398                         goto again;
3399                 }
3400                 if (!PageUptodate(page)) {
3401                         ret = -EIO;
3402                         goto out_unlock;
3403                 }
3404         }
3405         wait_on_page_writeback(page);
3406
3407         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3408                          GFP_NOFS);
3409         set_page_extent_mapped(page);
3410
3411         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3412         if (ordered) {
3413                 unlock_extent_cached(io_tree, page_start, page_end,
3414                                      &cached_state, GFP_NOFS);
3415                 unlock_page(page);
3416                 page_cache_release(page);
3417                 btrfs_start_ordered_extent(inode, ordered, 1);
3418                 btrfs_put_ordered_extent(ordered);
3419                 goto again;
3420         }
3421
3422         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3423                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3424                           0, 0, &cached_state, GFP_NOFS);
3425
3426         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3427                                         &cached_state);
3428         if (ret) {
3429                 unlock_extent_cached(io_tree, page_start, page_end,
3430                                      &cached_state, GFP_NOFS);
3431                 goto out_unlock;
3432         }
3433
3434         ret = 0;
3435         if (offset != PAGE_CACHE_SIZE) {
3436                 kaddr = kmap(page);
3437                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3438                 flush_dcache_page(page);
3439                 kunmap(page);
3440         }
3441         ClearPageChecked(page);
3442         set_page_dirty(page);
3443         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3444                              GFP_NOFS);
3445
3446 out_unlock:
3447         if (ret)
3448                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3449         unlock_page(page);
3450         page_cache_release(page);
3451 out:
3452         return ret;
3453 }
3454
3455 /*
3456  * This function puts in dummy file extents for the area we're creating a hole
3457  * for.  So if we are truncating this file to a larger size we need to insert
3458  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3459  * the range between oldsize and size
3460  */
3461 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3462 {
3463         struct btrfs_trans_handle *trans;
3464         struct btrfs_root *root = BTRFS_I(inode)->root;
3465         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3466         struct extent_map *em = NULL;
3467         struct extent_state *cached_state = NULL;
3468         u64 mask = root->sectorsize - 1;
3469         u64 hole_start = (oldsize + mask) & ~mask;
3470         u64 block_end = (size + mask) & ~mask;
3471         u64 last_byte;
3472         u64 cur_offset;
3473         u64 hole_size;
3474         int err = 0;
3475
3476         if (size <= hole_start)
3477                 return 0;
3478
3479         while (1) {
3480                 struct btrfs_ordered_extent *ordered;
3481                 btrfs_wait_ordered_range(inode, hole_start,
3482                                          block_end - hole_start);
3483                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3484                                  &cached_state, GFP_NOFS);
3485                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3486                 if (!ordered)
3487                         break;
3488                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3489                                      &cached_state, GFP_NOFS);
3490                 btrfs_put_ordered_extent(ordered);
3491         }
3492
3493         cur_offset = hole_start;
3494         while (1) {
3495                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3496                                 block_end - cur_offset, 0);
3497                 BUG_ON(IS_ERR_OR_NULL(em));
3498                 last_byte = min(extent_map_end(em), block_end);
3499                 last_byte = (last_byte + mask) & ~mask;
3500                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3501                         u64 hint_byte = 0;
3502                         hole_size = last_byte - cur_offset;
3503
3504                         trans = btrfs_start_transaction(root, 2);
3505                         if (IS_ERR(trans)) {
3506                                 err = PTR_ERR(trans);
3507                                 break;
3508                         }
3509
3510                         err = btrfs_drop_extents(trans, inode, cur_offset,
3511                                                  cur_offset + hole_size,
3512                                                  &hint_byte, 1);
3513                         if (err) {
3514                                 btrfs_end_transaction(trans, root);
3515                                 break;
3516                         }
3517
3518                         err = btrfs_insert_file_extent(trans, root,
3519                                         btrfs_ino(inode), cur_offset, 0,
3520                                         0, hole_size, 0, hole_size,
3521                                         0, 0, 0);
3522                         if (err) {
3523                                 btrfs_end_transaction(trans, root);
3524                                 break;
3525                         }
3526
3527                         btrfs_drop_extent_cache(inode, hole_start,
3528                                         last_byte - 1, 0);
3529
3530                         btrfs_end_transaction(trans, root);
3531                 }
3532                 free_extent_map(em);
3533                 em = NULL;
3534                 cur_offset = last_byte;
3535                 if (cur_offset >= block_end)
3536                         break;
3537         }
3538
3539         free_extent_map(em);
3540         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3541                              GFP_NOFS);
3542         return err;
3543 }
3544
3545 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3546 {
3547         loff_t oldsize = i_size_read(inode);
3548         int ret;
3549
3550         if (newsize == oldsize)
3551                 return 0;
3552
3553         if (newsize > oldsize) {
3554                 i_size_write(inode, newsize);
3555                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3556                 truncate_pagecache(inode, oldsize, newsize);
3557                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3558                 if (ret) {
3559                         btrfs_setsize(inode, oldsize);
3560                         return ret;
3561                 }
3562
3563                 mark_inode_dirty(inode);
3564         } else {
3565
3566                 /*
3567                  * We're truncating a file that used to have good data down to
3568                  * zero. Make sure it gets into the ordered flush list so that
3569                  * any new writes get down to disk quickly.
3570                  */
3571                 if (newsize == 0)
3572                         BTRFS_I(inode)->ordered_data_close = 1;
3573
3574                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3575                 truncate_setsize(inode, newsize);
3576                 ret = btrfs_truncate(inode);
3577         }
3578
3579         return ret;
3580 }
3581
3582 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3583 {
3584         struct inode *inode = dentry->d_inode;
3585         struct btrfs_root *root = BTRFS_I(inode)->root;
3586         int err;
3587
3588         if (btrfs_root_readonly(root))
3589                 return -EROFS;
3590
3591         err = inode_change_ok(inode, attr);
3592         if (err)
3593                 return err;
3594
3595         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3596                 err = btrfs_setsize(inode, attr->ia_size);
3597                 if (err)
3598                         return err;
3599         }
3600
3601         if (attr->ia_valid) {
3602                 setattr_copy(inode, attr);
3603                 mark_inode_dirty(inode);
3604
3605                 if (attr->ia_valid & ATTR_MODE)
3606                         err = btrfs_acl_chmod(inode);
3607         }
3608
3609         return err;
3610 }
3611
3612 void btrfs_evict_inode(struct inode *inode)
3613 {
3614         struct btrfs_trans_handle *trans;
3615         struct btrfs_root *root = BTRFS_I(inode)->root;
3616         unsigned long nr;
3617         int ret;
3618
3619         trace_btrfs_inode_evict(inode);
3620
3621         truncate_inode_pages(&inode->i_data, 0);
3622         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3623                                btrfs_is_free_space_inode(root, inode)))
3624                 goto no_delete;
3625
3626         if (is_bad_inode(inode)) {
3627                 btrfs_orphan_del(NULL, inode);
3628                 goto no_delete;
3629         }
3630         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3631         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3632
3633         if (root->fs_info->log_root_recovering) {
3634                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3635                 goto no_delete;
3636         }
3637
3638         if (inode->i_nlink > 0) {
3639                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3640                 goto no_delete;
3641         }
3642
3643         btrfs_i_size_write(inode, 0);
3644
3645         while (1) {
3646                 trans = btrfs_join_transaction(root);
3647                 BUG_ON(IS_ERR(trans));
3648                 trans->block_rsv = root->orphan_block_rsv;
3649
3650                 ret = btrfs_block_rsv_check(trans, root,
3651                                             root->orphan_block_rsv, 0, 5);
3652                 if (ret) {
3653                         BUG_ON(ret != -EAGAIN);
3654                         ret = btrfs_commit_transaction(trans, root);
3655                         BUG_ON(ret);
3656                         continue;
3657                 }
3658
3659                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3660                 if (ret != -EAGAIN)
3661                         break;
3662
3663                 nr = trans->blocks_used;
3664                 btrfs_end_transaction(trans, root);
3665                 trans = NULL;
3666                 btrfs_btree_balance_dirty(root, nr);
3667
3668         }
3669
3670         if (ret == 0) {
3671                 ret = btrfs_orphan_del(trans, inode);
3672                 BUG_ON(ret);
3673         }
3674
3675         if (!(root == root->fs_info->tree_root ||
3676               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3677                 btrfs_return_ino(root, btrfs_ino(inode));
3678
3679         nr = trans->blocks_used;
3680         btrfs_end_transaction(trans, root);
3681         btrfs_btree_balance_dirty(root, nr);
3682 no_delete:
3683         end_writeback(inode);
3684         return;
3685 }
3686
3687 /*
3688  * this returns the key found in the dir entry in the location pointer.
3689  * If no dir entries were found, location->objectid is 0.
3690  */
3691 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3692                                struct btrfs_key *location)
3693 {
3694         const char *name = dentry->d_name.name;
3695         int namelen = dentry->d_name.len;
3696         struct btrfs_dir_item *di;
3697         struct btrfs_path *path;
3698         struct btrfs_root *root = BTRFS_I(dir)->root;
3699         int ret = 0;
3700
3701         path = btrfs_alloc_path();
3702         if (!path)
3703                 return -ENOMEM;
3704
3705         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3706                                     namelen, 0);
3707         if (IS_ERR(di))
3708                 ret = PTR_ERR(di);
3709
3710         if (IS_ERR_OR_NULL(di))
3711                 goto out_err;
3712
3713         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3714 out:
3715         btrfs_free_path(path);
3716         return ret;
3717 out_err:
3718         location->objectid = 0;
3719         goto out;
3720 }
3721
3722 /*
3723  * when we hit a tree root in a directory, the btrfs part of the inode
3724  * needs to be changed to reflect the root directory of the tree root.  This
3725  * is kind of like crossing a mount point.
3726  */
3727 static int fixup_tree_root_location(struct btrfs_root *root,
3728                                     struct inode *dir,
3729                                     struct dentry *dentry,
3730                                     struct btrfs_key *location,
3731                                     struct btrfs_root **sub_root)
3732 {
3733         struct btrfs_path *path;
3734         struct btrfs_root *new_root;
3735         struct btrfs_root_ref *ref;
3736         struct extent_buffer *leaf;
3737         int ret;
3738         int err = 0;
3739
3740         path = btrfs_alloc_path();
3741         if (!path) {
3742                 err = -ENOMEM;
3743                 goto out;
3744         }
3745
3746         err = -ENOENT;
3747         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3748                                   BTRFS_I(dir)->root->root_key.objectid,
3749                                   location->objectid);
3750         if (ret) {
3751                 if (ret < 0)
3752                         err = ret;
3753                 goto out;
3754         }
3755
3756         leaf = path->nodes[0];
3757         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3758         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3759             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3760                 goto out;
3761
3762         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3763                                    (unsigned long)(ref + 1),
3764                                    dentry->d_name.len);
3765         if (ret)
3766                 goto out;
3767
3768         btrfs_release_path(path);
3769
3770         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3771         if (IS_ERR(new_root)) {
3772                 err = PTR_ERR(new_root);
3773                 goto out;
3774         }
3775
3776         if (btrfs_root_refs(&new_root->root_item) == 0) {
3777                 err = -ENOENT;
3778                 goto out;
3779         }
3780
3781         *sub_root = new_root;
3782         location->objectid = btrfs_root_dirid(&new_root->root_item);
3783         location->type = BTRFS_INODE_ITEM_KEY;
3784         location->offset = 0;
3785         err = 0;
3786 out:
3787         btrfs_free_path(path);
3788         return err;
3789 }
3790
3791 static void inode_tree_add(struct inode *inode)
3792 {
3793         struct btrfs_root *root = BTRFS_I(inode)->root;
3794         struct btrfs_inode *entry;
3795         struct rb_node **p;
3796         struct rb_node *parent;
3797         u64 ino = btrfs_ino(inode);
3798 again:
3799         p = &root->inode_tree.rb_node;
3800         parent = NULL;
3801
3802         if (inode_unhashed(inode))
3803                 return;
3804
3805         spin_lock(&root->inode_lock);
3806         while (*p) {
3807                 parent = *p;
3808                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3809
3810                 if (ino < btrfs_ino(&entry->vfs_inode))
3811                         p = &parent->rb_left;
3812                 else if (ino > btrfs_ino(&entry->vfs_inode))
3813                         p = &parent->rb_right;
3814                 else {
3815                         WARN_ON(!(entry->vfs_inode.i_state &
3816                                   (I_WILL_FREE | I_FREEING)));
3817                         rb_erase(parent, &root->inode_tree);
3818                         RB_CLEAR_NODE(parent);
3819                         spin_unlock(&root->inode_lock);
3820                         goto again;
3821                 }
3822         }
3823         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3824         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3825         spin_unlock(&root->inode_lock);
3826 }
3827
3828 static void inode_tree_del(struct inode *inode)
3829 {
3830         struct btrfs_root *root = BTRFS_I(inode)->root;
3831         int empty = 0;
3832
3833         spin_lock(&root->inode_lock);
3834         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3835                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3836                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3837                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3838         }
3839         spin_unlock(&root->inode_lock);
3840
3841         /*
3842          * Free space cache has inodes in the tree root, but the tree root has a
3843          * root_refs of 0, so this could end up dropping the tree root as a
3844          * snapshot, so we need the extra !root->fs_info->tree_root check to
3845          * make sure we don't drop it.
3846          */
3847         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3848             root != root->fs_info->tree_root) {
3849                 synchronize_srcu(&root->fs_info->subvol_srcu);
3850                 spin_lock(&root->inode_lock);
3851                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3852                 spin_unlock(&root->inode_lock);
3853                 if (empty)
3854                         btrfs_add_dead_root(root);
3855         }
3856 }
3857
3858 int btrfs_invalidate_inodes(struct btrfs_root *root)
3859 {
3860         struct rb_node *node;
3861         struct rb_node *prev;
3862         struct btrfs_inode *entry;
3863         struct inode *inode;
3864         u64 objectid = 0;
3865
3866         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3867
3868         spin_lock(&root->inode_lock);
3869 again:
3870         node = root->inode_tree.rb_node;
3871         prev = NULL;
3872         while (node) {
3873                 prev = node;
3874                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3875
3876                 if (objectid < btrfs_ino(&entry->vfs_inode))
3877                         node = node->rb_left;
3878                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3879                         node = node->rb_right;
3880                 else
3881                         break;
3882         }
3883         if (!node) {
3884                 while (prev) {
3885                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3886                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3887                                 node = prev;
3888                                 break;
3889                         }
3890                         prev = rb_next(prev);
3891                 }
3892         }
3893         while (node) {
3894                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3895                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3896                 inode = igrab(&entry->vfs_inode);
3897                 if (inode) {
3898                         spin_unlock(&root->inode_lock);
3899                         if (atomic_read(&inode->i_count) > 1)
3900                                 d_prune_aliases(inode);
3901                         /*
3902                          * btrfs_drop_inode will have it removed from
3903                          * the inode cache when its usage count
3904                          * hits zero.
3905                          */
3906                         iput(inode);
3907                         cond_resched();
3908                         spin_lock(&root->inode_lock);
3909                         goto again;
3910                 }
3911
3912                 if (cond_resched_lock(&root->inode_lock))
3913                         goto again;
3914
3915                 node = rb_next(node);
3916         }
3917         spin_unlock(&root->inode_lock);
3918         return 0;
3919 }
3920
3921 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3922 {
3923         struct btrfs_iget_args *args = p;
3924         inode->i_ino = args->ino;
3925         BTRFS_I(inode)->root = args->root;
3926         btrfs_set_inode_space_info(args->root, inode);
3927         return 0;
3928 }
3929
3930 static int btrfs_find_actor(struct inode *inode, void *opaque)
3931 {
3932         struct btrfs_iget_args *args = opaque;
3933         return args->ino == btrfs_ino(inode) &&
3934                 args->root == BTRFS_I(inode)->root;
3935 }
3936
3937 static struct inode *btrfs_iget_locked(struct super_block *s,
3938                                        u64 objectid,
3939                                        struct btrfs_root *root)
3940 {
3941         struct inode *inode;
3942         struct btrfs_iget_args args;
3943         args.ino = objectid;
3944         args.root = root;
3945
3946         inode = iget5_locked(s, objectid, btrfs_find_actor,
3947                              btrfs_init_locked_inode,
3948                              (void *)&args);
3949         return inode;
3950 }
3951
3952 /* Get an inode object given its location and corresponding root.
3953  * Returns in *is_new if the inode was read from disk
3954  */
3955 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3956                          struct btrfs_root *root, int *new)
3957 {
3958         struct inode *inode;
3959
3960         inode = btrfs_iget_locked(s, location->objectid, root);
3961         if (!inode)
3962                 return ERR_PTR(-ENOMEM);
3963
3964         if (inode->i_state & I_NEW) {
3965                 BTRFS_I(inode)->root = root;
3966                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3967                 btrfs_read_locked_inode(inode);
3968                 if (!is_bad_inode(inode)) {
3969                         inode_tree_add(inode);
3970                         unlock_new_inode(inode);
3971                         if (new)
3972                                 *new = 1;
3973                 } else {
3974                         unlock_new_inode(inode);
3975                         iput(inode);
3976                         inode = ERR_PTR(-ESTALE);
3977                 }
3978         }
3979
3980         return inode;
3981 }
3982
3983 static struct inode *new_simple_dir(struct super_block *s,
3984                                     struct btrfs_key *key,
3985                                     struct btrfs_root *root)
3986 {
3987         struct inode *inode = new_inode(s);
3988
3989         if (!inode)
3990                 return ERR_PTR(-ENOMEM);
3991
3992         BTRFS_I(inode)->root = root;
3993         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3994         BTRFS_I(inode)->dummy_inode = 1;
3995
3996         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3997         inode->i_op = &simple_dir_inode_operations;
3998         inode->i_fop = &simple_dir_operations;
3999         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4000         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4001
4002         return inode;
4003 }
4004
4005 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4006 {
4007         struct inode *inode;
4008         struct btrfs_root *root = BTRFS_I(dir)->root;
4009         struct btrfs_root *sub_root = root;
4010         struct btrfs_key location;
4011         int index;
4012         int ret;
4013
4014         if (dentry->d_name.len > BTRFS_NAME_LEN)
4015                 return ERR_PTR(-ENAMETOOLONG);
4016
4017         ret = btrfs_inode_by_name(dir, dentry, &location);
4018
4019         if (ret < 0)
4020                 return ERR_PTR(ret);
4021
4022         if (location.objectid == 0)
4023                 return NULL;
4024
4025         if (location.type == BTRFS_INODE_ITEM_KEY) {
4026                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4027                 return inode;
4028         }
4029
4030         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4031
4032         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4033         ret = fixup_tree_root_location(root, dir, dentry,
4034                                        &location, &sub_root);
4035         if (ret < 0) {
4036                 if (ret != -ENOENT)
4037                         inode = ERR_PTR(ret);
4038                 else
4039                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4040         } else {
4041                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4042         }
4043         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4044
4045         if (!IS_ERR(inode) && root != sub_root) {
4046                 down_read(&root->fs_info->cleanup_work_sem);
4047                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4048                         ret = btrfs_orphan_cleanup(sub_root);
4049                 up_read(&root->fs_info->cleanup_work_sem);
4050                 if (ret)
4051                         inode = ERR_PTR(ret);
4052         }
4053
4054         return inode;
4055 }
4056
4057 static int btrfs_dentry_delete(const struct dentry *dentry)
4058 {
4059         struct btrfs_root *root;
4060
4061         if (!dentry->d_inode && !IS_ROOT(dentry))
4062                 dentry = dentry->d_parent;
4063
4064         if (dentry->d_inode) {
4065                 root = BTRFS_I(dentry->d_inode)->root;
4066                 if (btrfs_root_refs(&root->root_item) == 0)
4067                         return 1;
4068         }
4069         return 0;
4070 }
4071
4072 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4073                                    struct nameidata *nd)
4074 {
4075         struct inode *inode;
4076
4077         inode = btrfs_lookup_dentry(dir, dentry);
4078         if (IS_ERR(inode))
4079                 return ERR_CAST(inode);
4080
4081         return d_splice_alias(inode, dentry);
4082 }
4083
4084 unsigned char btrfs_filetype_table[] = {
4085         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4086 };
4087
4088 static int btrfs_real_readdir(struct file *filp, void *dirent,
4089                               filldir_t filldir)
4090 {
4091         struct inode *inode = filp->f_dentry->d_inode;
4092         struct btrfs_root *root = BTRFS_I(inode)->root;
4093         struct btrfs_item *item;
4094         struct btrfs_dir_item *di;
4095         struct btrfs_key key;
4096         struct btrfs_key found_key;
4097         struct btrfs_path *path;
4098         struct list_head ins_list;
4099         struct list_head del_list;
4100         int ret;
4101         struct extent_buffer *leaf;
4102         int slot;
4103         unsigned char d_type;
4104         int over = 0;
4105         u32 di_cur;
4106         u32 di_total;
4107         u32 di_len;
4108         int key_type = BTRFS_DIR_INDEX_KEY;
4109         char tmp_name[32];
4110         char *name_ptr;
4111         int name_len;
4112         int is_curr = 0;        /* filp->f_pos points to the current index? */
4113
4114         /* FIXME, use a real flag for deciding about the key type */
4115         if (root->fs_info->tree_root == root)
4116                 key_type = BTRFS_DIR_ITEM_KEY;
4117
4118         /* special case for "." */
4119         if (filp->f_pos == 0) {
4120                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4121                 if (over)
4122                         return 0;
4123                 filp->f_pos = 1;
4124         }
4125         /* special case for .., just use the back ref */
4126         if (filp->f_pos == 1) {
4127                 u64 pino = parent_ino(filp->f_path.dentry);
4128                 over = filldir(dirent, "..", 2,
4129                                2, pino, DT_DIR);
4130                 if (over)
4131                         return 0;
4132                 filp->f_pos = 2;
4133         }
4134         path = btrfs_alloc_path();
4135         if (!path)
4136                 return -ENOMEM;
4137
4138         path->reada = 1;
4139
4140         if (key_type == BTRFS_DIR_INDEX_KEY) {
4141                 INIT_LIST_HEAD(&ins_list);
4142                 INIT_LIST_HEAD(&del_list);
4143                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4144         }
4145
4146         btrfs_set_key_type(&key, key_type);
4147         key.offset = filp->f_pos;
4148         key.objectid = btrfs_ino(inode);
4149
4150         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4151         if (ret < 0)
4152                 goto err;
4153
4154         while (1) {
4155                 leaf = path->nodes[0];
4156                 slot = path->slots[0];
4157                 if (slot >= btrfs_header_nritems(leaf)) {
4158                         ret = btrfs_next_leaf(root, path);
4159                         if (ret < 0)
4160                                 goto err;
4161                         else if (ret > 0)
4162                                 break;
4163                         continue;
4164                 }
4165
4166                 item = btrfs_item_nr(leaf, slot);
4167                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4168
4169                 if (found_key.objectid != key.objectid)
4170                         break;
4171                 if (btrfs_key_type(&found_key) != key_type)
4172                         break;
4173                 if (found_key.offset < filp->f_pos)
4174                         goto next;
4175                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4176                     btrfs_should_delete_dir_index(&del_list,
4177                                                   found_key.offset))
4178                         goto next;
4179
4180                 filp->f_pos = found_key.offset;
4181                 is_curr = 1;
4182
4183                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4184                 di_cur = 0;
4185                 di_total = btrfs_item_size(leaf, item);
4186
4187                 while (di_cur < di_total) {
4188                         struct btrfs_key location;
4189
4190                         if (verify_dir_item(root, leaf, di))
4191                                 break;
4192
4193                         name_len = btrfs_dir_name_len(leaf, di);
4194                         if (name_len <= sizeof(tmp_name)) {
4195                                 name_ptr = tmp_name;
4196                         } else {
4197                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4198                                 if (!name_ptr) {
4199                                         ret = -ENOMEM;
4200                                         goto err;
4201                                 }
4202                         }
4203                         read_extent_buffer(leaf, name_ptr,
4204                                            (unsigned long)(di + 1), name_len);
4205
4206                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4207                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4208
4209                         /* is this a reference to our own snapshot? If so
4210                          * skip it
4211                          */
4212                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4213                             location.objectid == root->root_key.objectid) {
4214                                 over = 0;
4215                                 goto skip;
4216                         }
4217                         over = filldir(dirent, name_ptr, name_len,
4218                                        found_key.offset, location.objectid,
4219                                        d_type);
4220
4221 skip:
4222                         if (name_ptr != tmp_name)
4223                                 kfree(name_ptr);
4224
4225                         if (over)
4226                                 goto nopos;
4227                         di_len = btrfs_dir_name_len(leaf, di) +
4228                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4229                         di_cur += di_len;
4230                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4231                 }
4232 next:
4233                 path->slots[0]++;
4234         }
4235
4236         if (key_type == BTRFS_DIR_INDEX_KEY) {
4237                 if (is_curr)
4238                         filp->f_pos++;
4239                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4240                                                       &ins_list);
4241                 if (ret)
4242                         goto nopos;
4243         }
4244
4245         /* Reached end of directory/root. Bump pos past the last item. */
4246         if (key_type == BTRFS_DIR_INDEX_KEY)
4247                 /*
4248                  * 32-bit glibc will use getdents64, but then strtol -
4249                  * so the last number we can serve is this.
4250                  */
4251                 filp->f_pos = 0x7fffffff;
4252         else
4253                 filp->f_pos++;
4254 nopos:
4255         ret = 0;
4256 err:
4257         if (key_type == BTRFS_DIR_INDEX_KEY)
4258                 btrfs_put_delayed_items(&ins_list, &del_list);
4259         btrfs_free_path(path);
4260         return ret;
4261 }
4262
4263 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4264 {
4265         struct btrfs_root *root = BTRFS_I(inode)->root;
4266         struct btrfs_trans_handle *trans;
4267         int ret = 0;
4268         bool nolock = false;
4269
4270         if (BTRFS_I(inode)->dummy_inode)
4271                 return 0;
4272
4273         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4274                 nolock = true;
4275
4276         if (wbc->sync_mode == WB_SYNC_ALL) {
4277                 if (nolock)
4278                         trans = btrfs_join_transaction_nolock(root);
4279                 else
4280                         trans = btrfs_join_transaction(root);
4281                 if (IS_ERR(trans))
4282                         return PTR_ERR(trans);
4283                 if (nolock)
4284                         ret = btrfs_end_transaction_nolock(trans, root);
4285                 else
4286                         ret = btrfs_commit_transaction(trans, root);
4287         }
4288         return ret;
4289 }
4290
4291 /*
4292  * This is somewhat expensive, updating the tree every time the
4293  * inode changes.  But, it is most likely to find the inode in cache.
4294  * FIXME, needs more benchmarking...there are no reasons other than performance
4295  * to keep or drop this code.
4296  */
4297 void btrfs_dirty_inode(struct inode *inode, int flags)
4298 {
4299         struct btrfs_root *root = BTRFS_I(inode)->root;
4300         struct btrfs_trans_handle *trans;
4301         int ret;
4302
4303         if (BTRFS_I(inode)->dummy_inode)
4304                 return;
4305
4306         trans = btrfs_join_transaction(root);
4307         BUG_ON(IS_ERR(trans));
4308
4309         ret = btrfs_update_inode(trans, root, inode);
4310         if (ret && ret == -ENOSPC) {
4311                 /* whoops, lets try again with the full transaction */
4312                 btrfs_end_transaction(trans, root);
4313                 trans = btrfs_start_transaction(root, 1);
4314                 if (IS_ERR(trans)) {
4315                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4316                                        "dirty  inode %llu error %ld\n",
4317                                        (unsigned long long)btrfs_ino(inode),
4318                                        PTR_ERR(trans));
4319                         return;
4320                 }
4321
4322                 ret = btrfs_update_inode(trans, root, inode);
4323                 if (ret) {
4324                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4325                                        "dirty  inode %llu error %d\n",
4326                                        (unsigned long long)btrfs_ino(inode),
4327                                        ret);
4328                 }
4329         }
4330         btrfs_end_transaction(trans, root);
4331         if (BTRFS_I(inode)->delayed_node)
4332                 btrfs_balance_delayed_items(root);
4333 }
4334
4335 /*
4336  * find the highest existing sequence number in a directory
4337  * and then set the in-memory index_cnt variable to reflect
4338  * free sequence numbers
4339  */
4340 static int btrfs_set_inode_index_count(struct inode *inode)
4341 {
4342         struct btrfs_root *root = BTRFS_I(inode)->root;
4343         struct btrfs_key key, found_key;
4344         struct btrfs_path *path;
4345         struct extent_buffer *leaf;
4346         int ret;
4347
4348         key.objectid = btrfs_ino(inode);
4349         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4350         key.offset = (u64)-1;
4351
4352         path = btrfs_alloc_path();
4353         if (!path)
4354                 return -ENOMEM;
4355
4356         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4357         if (ret < 0)
4358                 goto out;
4359         /* FIXME: we should be able to handle this */
4360         if (ret == 0)
4361                 goto out;
4362         ret = 0;
4363
4364         /*
4365          * MAGIC NUMBER EXPLANATION:
4366          * since we search a directory based on f_pos we have to start at 2
4367          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4368          * else has to start at 2
4369          */
4370         if (path->slots[0] == 0) {
4371                 BTRFS_I(inode)->index_cnt = 2;
4372                 goto out;
4373         }
4374
4375         path->slots[0]--;
4376
4377         leaf = path->nodes[0];
4378         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4379
4380         if (found_key.objectid != btrfs_ino(inode) ||
4381             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4382                 BTRFS_I(inode)->index_cnt = 2;
4383                 goto out;
4384         }
4385
4386         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4387 out:
4388         btrfs_free_path(path);
4389         return ret;
4390 }
4391
4392 /*
4393  * helper to find a free sequence number in a given directory.  This current
4394  * code is very simple, later versions will do smarter things in the btree
4395  */
4396 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4397 {
4398         int ret = 0;
4399
4400         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4401                 ret = btrfs_inode_delayed_dir_index_count(dir);
4402                 if (ret) {
4403                         ret = btrfs_set_inode_index_count(dir);
4404                         if (ret)
4405                                 return ret;
4406                 }
4407         }
4408
4409         *index = BTRFS_I(dir)->index_cnt;
4410         BTRFS_I(dir)->index_cnt++;
4411
4412         return ret;
4413 }
4414
4415 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4416                                      struct btrfs_root *root,
4417                                      struct inode *dir,
4418                                      const char *name, int name_len,
4419                                      u64 ref_objectid, u64 objectid, int mode,
4420                                      u64 *index)
4421 {
4422         struct inode *inode;
4423         struct btrfs_inode_item *inode_item;
4424         struct btrfs_key *location;
4425         struct btrfs_path *path;
4426         struct btrfs_inode_ref *ref;
4427         struct btrfs_key key[2];
4428         u32 sizes[2];
4429         unsigned long ptr;
4430         int ret;
4431         int owner;
4432
4433         path = btrfs_alloc_path();
4434         if (!path)
4435                 return ERR_PTR(-ENOMEM);
4436
4437         inode = new_inode(root->fs_info->sb);
4438         if (!inode) {
4439                 btrfs_free_path(path);
4440                 return ERR_PTR(-ENOMEM);
4441         }
4442
4443         /*
4444          * we have to initialize this early, so we can reclaim the inode
4445          * number if we fail afterwards in this function.
4446          */
4447         inode->i_ino = objectid;
4448
4449         if (dir) {
4450                 trace_btrfs_inode_request(dir);
4451
4452                 ret = btrfs_set_inode_index(dir, index);
4453                 if (ret) {
4454                         btrfs_free_path(path);
4455                         iput(inode);
4456                         return ERR_PTR(ret);
4457                 }
4458         }
4459         /*
4460          * index_cnt is ignored for everything but a dir,
4461          * btrfs_get_inode_index_count has an explanation for the magic
4462          * number
4463          */
4464         BTRFS_I(inode)->index_cnt = 2;
4465         BTRFS_I(inode)->root = root;
4466         BTRFS_I(inode)->generation = trans->transid;
4467         inode->i_generation = BTRFS_I(inode)->generation;
4468         btrfs_set_inode_space_info(root, inode);
4469
4470         if (mode & S_IFDIR)
4471                 owner = 0;
4472         else
4473                 owner = 1;
4474
4475         key[0].objectid = objectid;
4476         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4477         key[0].offset = 0;
4478
4479         key[1].objectid = objectid;
4480         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4481         key[1].offset = ref_objectid;
4482
4483         sizes[0] = sizeof(struct btrfs_inode_item);
4484         sizes[1] = name_len + sizeof(*ref);
4485
4486         path->leave_spinning = 1;
4487         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4488         if (ret != 0)
4489                 goto fail;
4490
4491         inode_init_owner(inode, dir, mode);
4492         inode_set_bytes(inode, 0);
4493         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4494         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4495                                   struct btrfs_inode_item);
4496         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4497
4498         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4499                              struct btrfs_inode_ref);
4500         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4501         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4502         ptr = (unsigned long)(ref + 1);
4503         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4504
4505         btrfs_mark_buffer_dirty(path->nodes[0]);
4506         btrfs_free_path(path);
4507
4508         location = &BTRFS_I(inode)->location;
4509         location->objectid = objectid;
4510         location->offset = 0;
4511         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4512
4513         btrfs_inherit_iflags(inode, dir);
4514
4515         if ((mode & S_IFREG)) {
4516                 if (btrfs_test_opt(root, NODATASUM))
4517                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4518                 if (btrfs_test_opt(root, NODATACOW) ||
4519                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4520                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4521         }
4522
4523         insert_inode_hash(inode);
4524         inode_tree_add(inode);
4525
4526         trace_btrfs_inode_new(inode);
4527         btrfs_set_inode_last_trans(trans, inode);
4528
4529         return inode;
4530 fail:
4531         if (dir)
4532                 BTRFS_I(dir)->index_cnt--;
4533         btrfs_free_path(path);
4534         iput(inode);
4535         return ERR_PTR(ret);
4536 }
4537
4538 static inline u8 btrfs_inode_type(struct inode *inode)
4539 {
4540         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4541 }
4542
4543 /*
4544  * utility function to add 'inode' into 'parent_inode' with
4545  * a give name and a given sequence number.
4546  * if 'add_backref' is true, also insert a backref from the
4547  * inode to the parent directory.
4548  */
4549 int btrfs_add_link(struct btrfs_trans_handle *trans,
4550                    struct inode *parent_inode, struct inode *inode,
4551                    const char *name, int name_len, int add_backref, u64 index)
4552 {
4553         int ret = 0;
4554         struct btrfs_key key;
4555         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4556         u64 ino = btrfs_ino(inode);
4557         u64 parent_ino = btrfs_ino(parent_inode);
4558
4559         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4560                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4561         } else {
4562                 key.objectid = ino;
4563                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4564                 key.offset = 0;
4565         }
4566
4567         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4568                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4569                                          key.objectid, root->root_key.objectid,
4570                                          parent_ino, index, name, name_len);
4571         } else if (add_backref) {
4572                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4573                                              parent_ino, index);
4574         }
4575
4576         if (ret == 0) {
4577                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4578                                             parent_inode, &key,
4579                                             btrfs_inode_type(inode), index);
4580                 BUG_ON(ret);
4581
4582                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4583                                    name_len * 2);
4584                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4585                 ret = btrfs_update_inode(trans, root, parent_inode);
4586         }
4587         return ret;
4588 }
4589
4590 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4591                             struct inode *dir, struct dentry *dentry,
4592                             struct inode *inode, int backref, u64 index)
4593 {
4594         int err = btrfs_add_link(trans, dir, inode,
4595                                  dentry->d_name.name, dentry->d_name.len,
4596                                  backref, index);
4597         if (!err) {
4598                 d_instantiate(dentry, inode);
4599                 return 0;
4600         }
4601         if (err > 0)
4602                 err = -EEXIST;
4603         return err;
4604 }
4605
4606 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4607                         int mode, dev_t rdev)
4608 {
4609         struct btrfs_trans_handle *trans;
4610         struct btrfs_root *root = BTRFS_I(dir)->root;
4611         struct inode *inode = NULL;
4612         int err;
4613         int drop_inode = 0;
4614         u64 objectid;
4615         unsigned long nr = 0;
4616         u64 index = 0;
4617
4618         if (!new_valid_dev(rdev))
4619                 return -EINVAL;
4620
4621         /*
4622          * 2 for inode item and ref
4623          * 2 for dir items
4624          * 1 for xattr if selinux is on
4625          */
4626         trans = btrfs_start_transaction(root, 5);
4627         if (IS_ERR(trans))
4628                 return PTR_ERR(trans);
4629
4630         err = btrfs_find_free_ino(root, &objectid);
4631         if (err)
4632                 goto out_unlock;
4633
4634         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4635                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4636                                 mode, &index);
4637         if (IS_ERR(inode)) {
4638                 err = PTR_ERR(inode);
4639                 goto out_unlock;
4640         }
4641
4642         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4643         if (err) {
4644                 drop_inode = 1;
4645                 goto out_unlock;
4646         }
4647
4648         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4649         if (err)
4650                 drop_inode = 1;
4651         else {
4652                 inode->i_op = &btrfs_special_inode_operations;
4653                 init_special_inode(inode, inode->i_mode, rdev);
4654                 btrfs_update_inode(trans, root, inode);
4655         }
4656 out_unlock:
4657         nr = trans->blocks_used;
4658         btrfs_end_transaction_throttle(trans, root);
4659         btrfs_btree_balance_dirty(root, nr);
4660         if (drop_inode) {
4661                 inode_dec_link_count(inode);
4662                 iput(inode);
4663         }
4664         return err;
4665 }
4666
4667 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4668                         int mode, struct nameidata *nd)
4669 {
4670         struct btrfs_trans_handle *trans;
4671         struct btrfs_root *root = BTRFS_I(dir)->root;
4672         struct inode *inode = NULL;
4673         int drop_inode = 0;
4674         int err;
4675         unsigned long nr = 0;
4676         u64 objectid;
4677         u64 index = 0;
4678
4679         /*
4680          * 2 for inode item and ref
4681          * 2 for dir items
4682          * 1 for xattr if selinux is on
4683          */
4684         trans = btrfs_start_transaction(root, 5);
4685         if (IS_ERR(trans))
4686                 return PTR_ERR(trans);
4687
4688         err = btrfs_find_free_ino(root, &objectid);
4689         if (err)
4690                 goto out_unlock;
4691
4692         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4693                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4694                                 mode, &index);
4695         if (IS_ERR(inode)) {
4696                 err = PTR_ERR(inode);
4697                 goto out_unlock;
4698         }
4699
4700         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4701         if (err) {
4702                 drop_inode = 1;
4703                 goto out_unlock;
4704         }
4705
4706         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4707         if (err)
4708                 drop_inode = 1;
4709         else {
4710                 inode->i_mapping->a_ops = &btrfs_aops;
4711                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4712                 inode->i_fop = &btrfs_file_operations;
4713                 inode->i_op = &btrfs_file_inode_operations;
4714                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4715         }
4716 out_unlock:
4717         nr = trans->blocks_used;
4718         btrfs_end_transaction_throttle(trans, root);
4719         if (drop_inode) {
4720                 inode_dec_link_count(inode);
4721                 iput(inode);
4722         }
4723         btrfs_btree_balance_dirty(root, nr);
4724         return err;
4725 }
4726
4727 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4728                       struct dentry *dentry)
4729 {
4730         struct btrfs_trans_handle *trans;
4731         struct btrfs_root *root = BTRFS_I(dir)->root;
4732         struct inode *inode = old_dentry->d_inode;
4733         u64 index;
4734         unsigned long nr = 0;
4735         int err;
4736         int drop_inode = 0;
4737
4738         /* do not allow sys_link's with other subvols of the same device */
4739         if (root->objectid != BTRFS_I(inode)->root->objectid)
4740                 return -EXDEV;
4741
4742         if (inode->i_nlink == ~0U)
4743                 return -EMLINK;
4744
4745         err = btrfs_set_inode_index(dir, &index);
4746         if (err)
4747                 goto fail;
4748
4749         /*
4750          * 2 items for inode and inode ref
4751          * 2 items for dir items
4752          * 1 item for parent inode
4753          */
4754         trans = btrfs_start_transaction(root, 5);
4755         if (IS_ERR(trans)) {
4756                 err = PTR_ERR(trans);
4757                 goto fail;
4758         }
4759
4760         btrfs_inc_nlink(inode);
4761         inode->i_ctime = CURRENT_TIME;
4762         ihold(inode);
4763
4764         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4765
4766         if (err) {
4767                 drop_inode = 1;
4768         } else {
4769                 struct dentry *parent = dget_parent(dentry);
4770                 err = btrfs_update_inode(trans, root, inode);
4771                 BUG_ON(err);
4772                 btrfs_log_new_name(trans, inode, NULL, parent);
4773                 dput(parent);
4774         }
4775
4776         nr = trans->blocks_used;
4777         btrfs_end_transaction_throttle(trans, root);
4778 fail:
4779         if (drop_inode) {
4780                 inode_dec_link_count(inode);
4781                 iput(inode);
4782         }
4783         btrfs_btree_balance_dirty(root, nr);
4784         return err;
4785 }
4786
4787 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4788 {
4789         struct inode *inode = NULL;
4790         struct btrfs_trans_handle *trans;
4791         struct btrfs_root *root = BTRFS_I(dir)->root;
4792         int err = 0;
4793         int drop_on_err = 0;
4794         u64 objectid = 0;
4795         u64 index = 0;
4796         unsigned long nr = 1;
4797
4798         /*
4799          * 2 items for inode and ref
4800          * 2 items for dir items
4801          * 1 for xattr if selinux is on
4802          */
4803         trans = btrfs_start_transaction(root, 5);
4804         if (IS_ERR(trans))
4805                 return PTR_ERR(trans);
4806
4807         err = btrfs_find_free_ino(root, &objectid);
4808         if (err)
4809                 goto out_fail;
4810
4811         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4812                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4813                                 S_IFDIR | mode, &index);
4814         if (IS_ERR(inode)) {
4815                 err = PTR_ERR(inode);
4816                 goto out_fail;
4817         }
4818
4819         drop_on_err = 1;
4820
4821         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4822         if (err)
4823                 goto out_fail;
4824
4825         inode->i_op = &btrfs_dir_inode_operations;
4826         inode->i_fop = &btrfs_dir_file_operations;
4827
4828         btrfs_i_size_write(inode, 0);
4829         err = btrfs_update_inode(trans, root, inode);
4830         if (err)
4831                 goto out_fail;
4832
4833         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4834                              dentry->d_name.len, 0, index);
4835         if (err)
4836                 goto out_fail;
4837
4838         d_instantiate(dentry, inode);
4839         drop_on_err = 0;
4840
4841 out_fail:
4842         nr = trans->blocks_used;
4843         btrfs_end_transaction_throttle(trans, root);
4844         if (drop_on_err)
4845                 iput(inode);
4846         btrfs_btree_balance_dirty(root, nr);
4847         return err;
4848 }
4849
4850 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4851  * and an extent that you want to insert, deal with overlap and insert
4852  * the new extent into the tree.
4853  */
4854 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4855                                 struct extent_map *existing,
4856                                 struct extent_map *em,
4857                                 u64 map_start, u64 map_len)
4858 {
4859         u64 start_diff;
4860
4861         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4862         start_diff = map_start - em->start;
4863         em->start = map_start;
4864         em->len = map_len;
4865         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4866             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4867                 em->block_start += start_diff;
4868                 em->block_len -= start_diff;
4869         }
4870         return add_extent_mapping(em_tree, em);
4871 }
4872
4873 static noinline int uncompress_inline(struct btrfs_path *path,
4874                                       struct inode *inode, struct page *page,
4875                                       size_t pg_offset, u64 extent_offset,
4876                                       struct btrfs_file_extent_item *item)
4877 {
4878         int ret;
4879         struct extent_buffer *leaf = path->nodes[0];
4880         char *tmp;
4881         size_t max_size;
4882         unsigned long inline_size;
4883         unsigned long ptr;
4884         int compress_type;
4885
4886         WARN_ON(pg_offset != 0);
4887         compress_type = btrfs_file_extent_compression(leaf, item);
4888         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4889         inline_size = btrfs_file_extent_inline_item_len(leaf,
4890                                         btrfs_item_nr(leaf, path->slots[0]));
4891         tmp = kmalloc(inline_size, GFP_NOFS);
4892         if (!tmp)
4893                 return -ENOMEM;
4894         ptr = btrfs_file_extent_inline_start(item);
4895
4896         read_extent_buffer(leaf, tmp, ptr, inline_size);
4897
4898         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4899         ret = btrfs_decompress(compress_type, tmp, page,
4900                                extent_offset, inline_size, max_size);
4901         if (ret) {
4902                 char *kaddr = kmap_atomic(page, KM_USER0);
4903                 unsigned long copy_size = min_t(u64,
4904                                   PAGE_CACHE_SIZE - pg_offset,
4905                                   max_size - extent_offset);
4906                 memset(kaddr + pg_offset, 0, copy_size);
4907                 kunmap_atomic(kaddr, KM_USER0);
4908         }
4909         kfree(tmp);
4910         return 0;
4911 }
4912
4913 /*
4914  * a bit scary, this does extent mapping from logical file offset to the disk.
4915  * the ugly parts come from merging extents from the disk with the in-ram
4916  * representation.  This gets more complex because of the data=ordered code,
4917  * where the in-ram extents might be locked pending data=ordered completion.
4918  *
4919  * This also copies inline extents directly into the page.
4920  */
4921
4922 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4923                                     size_t pg_offset, u64 start, u64 len,
4924                                     int create)
4925 {
4926         int ret;
4927         int err = 0;
4928         u64 bytenr;
4929         u64 extent_start = 0;
4930         u64 extent_end = 0;
4931         u64 objectid = btrfs_ino(inode);
4932         u32 found_type;
4933         struct btrfs_path *path = NULL;
4934         struct btrfs_root *root = BTRFS_I(inode)->root;
4935         struct btrfs_file_extent_item *item;
4936         struct extent_buffer *leaf;
4937         struct btrfs_key found_key;
4938         struct extent_map *em = NULL;
4939         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4940         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4941         struct btrfs_trans_handle *trans = NULL;
4942         int compress_type;
4943
4944 again:
4945         read_lock(&em_tree->lock);
4946         em = lookup_extent_mapping(em_tree, start, len);
4947         if (em)
4948                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4949         read_unlock(&em_tree->lock);
4950
4951         if (em) {
4952                 if (em->start > start || em->start + em->len <= start)
4953                         free_extent_map(em);
4954                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4955                         free_extent_map(em);
4956                 else
4957                         goto out;
4958         }
4959         em = alloc_extent_map();
4960         if (!em) {
4961                 err = -ENOMEM;
4962                 goto out;
4963         }
4964         em->bdev = root->fs_info->fs_devices->latest_bdev;
4965         em->start = EXTENT_MAP_HOLE;
4966         em->orig_start = EXTENT_MAP_HOLE;
4967         em->len = (u64)-1;
4968         em->block_len = (u64)-1;
4969
4970         if (!path) {
4971                 path = btrfs_alloc_path();
4972                 if (!path) {
4973                         err = -ENOMEM;
4974                         goto out;
4975                 }
4976                 /*
4977                  * Chances are we'll be called again, so go ahead and do
4978                  * readahead
4979                  */
4980                 path->reada = 1;
4981         }
4982
4983         ret = btrfs_lookup_file_extent(trans, root, path,
4984                                        objectid, start, trans != NULL);
4985         if (ret < 0) {
4986                 err = ret;
4987                 goto out;
4988         }
4989
4990         if (ret != 0) {
4991                 if (path->slots[0] == 0)
4992                         goto not_found;
4993                 path->slots[0]--;
4994         }
4995
4996         leaf = path->nodes[0];
4997         item = btrfs_item_ptr(leaf, path->slots[0],
4998                               struct btrfs_file_extent_item);
4999         /* are we inside the extent that was found? */
5000         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5001         found_type = btrfs_key_type(&found_key);
5002         if (found_key.objectid != objectid ||
5003             found_type != BTRFS_EXTENT_DATA_KEY) {
5004                 goto not_found;
5005         }
5006
5007         found_type = btrfs_file_extent_type(leaf, item);
5008         extent_start = found_key.offset;
5009         compress_type = btrfs_file_extent_compression(leaf, item);
5010         if (found_type == BTRFS_FILE_EXTENT_REG ||
5011             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5012                 extent_end = extent_start +
5013                        btrfs_file_extent_num_bytes(leaf, item);
5014         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5015                 size_t size;
5016                 size = btrfs_file_extent_inline_len(leaf, item);
5017                 extent_end = (extent_start + size + root->sectorsize - 1) &
5018                         ~((u64)root->sectorsize - 1);
5019         }
5020
5021         if (start >= extent_end) {
5022                 path->slots[0]++;
5023                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5024                         ret = btrfs_next_leaf(root, path);
5025                         if (ret < 0) {
5026                                 err = ret;
5027                                 goto out;
5028                         }
5029                         if (ret > 0)
5030                                 goto not_found;
5031                         leaf = path->nodes[0];
5032                 }
5033                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5034                 if (found_key.objectid != objectid ||
5035                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5036                         goto not_found;
5037                 if (start + len <= found_key.offset)
5038                         goto not_found;
5039                 em->start = start;
5040                 em->len = found_key.offset - start;
5041                 goto not_found_em;
5042         }
5043
5044         if (found_type == BTRFS_FILE_EXTENT_REG ||
5045             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5046                 em->start = extent_start;
5047                 em->len = extent_end - extent_start;
5048                 em->orig_start = extent_start -
5049                                  btrfs_file_extent_offset(leaf, item);
5050                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5051                 if (bytenr == 0) {
5052                         em->block_start = EXTENT_MAP_HOLE;
5053                         goto insert;
5054                 }
5055                 if (compress_type != BTRFS_COMPRESS_NONE) {
5056                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5057                         em->compress_type = compress_type;
5058                         em->block_start = bytenr;
5059                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5060                                                                          item);
5061                 } else {
5062                         bytenr += btrfs_file_extent_offset(leaf, item);
5063                         em->block_start = bytenr;
5064                         em->block_len = em->len;
5065                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5066                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5067                 }
5068                 goto insert;
5069         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5070                 unsigned long ptr;
5071                 char *map;
5072                 size_t size;
5073                 size_t extent_offset;
5074                 size_t copy_size;
5075
5076                 em->block_start = EXTENT_MAP_INLINE;
5077                 if (!page || create) {
5078                         em->start = extent_start;
5079                         em->len = extent_end - extent_start;
5080                         goto out;
5081                 }
5082
5083                 size = btrfs_file_extent_inline_len(leaf, item);
5084                 extent_offset = page_offset(page) + pg_offset - extent_start;
5085                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5086                                 size - extent_offset);
5087                 em->start = extent_start + extent_offset;
5088                 em->len = (copy_size + root->sectorsize - 1) &
5089                         ~((u64)root->sectorsize - 1);
5090                 em->orig_start = EXTENT_MAP_INLINE;
5091                 if (compress_type) {
5092                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5093                         em->compress_type = compress_type;
5094                 }
5095                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5096                 if (create == 0 && !PageUptodate(page)) {
5097                         if (btrfs_file_extent_compression(leaf, item) !=
5098                             BTRFS_COMPRESS_NONE) {
5099                                 ret = uncompress_inline(path, inode, page,
5100                                                         pg_offset,
5101                                                         extent_offset, item);
5102                                 BUG_ON(ret);
5103                         } else {
5104                                 map = kmap(page);
5105                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5106                                                    copy_size);
5107                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5108                                         memset(map + pg_offset + copy_size, 0,
5109                                                PAGE_CACHE_SIZE - pg_offset -
5110                                                copy_size);
5111                                 }
5112                                 kunmap(page);
5113                         }
5114                         flush_dcache_page(page);
5115                 } else if (create && PageUptodate(page)) {
5116                         WARN_ON(1);
5117                         if (!trans) {
5118                                 kunmap(page);
5119                                 free_extent_map(em);
5120                                 em = NULL;
5121
5122                                 btrfs_release_path(path);
5123                                 trans = btrfs_join_transaction(root);
5124
5125                                 if (IS_ERR(trans))
5126                                         return ERR_CAST(trans);
5127                                 goto again;
5128                         }
5129                         map = kmap(page);
5130                         write_extent_buffer(leaf, map + pg_offset, ptr,
5131                                             copy_size);
5132                         kunmap(page);
5133                         btrfs_mark_buffer_dirty(leaf);
5134                 }
5135                 set_extent_uptodate(io_tree, em->start,
5136                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5137                 goto insert;
5138         } else {
5139                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5140                 WARN_ON(1);
5141         }
5142 not_found:
5143         em->start = start;
5144         em->len = len;
5145 not_found_em:
5146         em->block_start = EXTENT_MAP_HOLE;
5147         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5148 insert:
5149         btrfs_release_path(path);
5150         if (em->start > start || extent_map_end(em) <= start) {
5151                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5152                        "[%llu %llu]\n", (unsigned long long)em->start,
5153                        (unsigned long long)em->len,
5154                        (unsigned long long)start,
5155                        (unsigned long long)len);
5156                 err = -EIO;
5157                 goto out;
5158         }
5159
5160         err = 0;
5161         write_lock(&em_tree->lock);
5162         ret = add_extent_mapping(em_tree, em);
5163         /* it is possible that someone inserted the extent into the tree
5164          * while we had the lock dropped.  It is also possible that
5165          * an overlapping map exists in the tree
5166          */
5167         if (ret == -EEXIST) {
5168                 struct extent_map *existing;
5169
5170                 ret = 0;
5171
5172                 existing = lookup_extent_mapping(em_tree, start, len);
5173                 if (existing && (existing->start > start ||
5174                     existing->start + existing->len <= start)) {
5175                         free_extent_map(existing);
5176                         existing = NULL;
5177                 }
5178                 if (!existing) {
5179                         existing = lookup_extent_mapping(em_tree, em->start,
5180                                                          em->len);
5181                         if (existing) {
5182                                 err = merge_extent_mapping(em_tree, existing,
5183                                                            em, start,
5184                                                            root->sectorsize);
5185                                 free_extent_map(existing);
5186                                 if (err) {
5187                                         free_extent_map(em);
5188                                         em = NULL;
5189                                 }
5190                         } else {
5191                                 err = -EIO;
5192                                 free_extent_map(em);
5193                                 em = NULL;
5194                         }
5195                 } else {
5196                         free_extent_map(em);
5197                         em = existing;
5198                         err = 0;
5199                 }
5200         }
5201         write_unlock(&em_tree->lock);
5202 out:
5203
5204         trace_btrfs_get_extent(root, em);
5205
5206         if (path)
5207                 btrfs_free_path(path);
5208         if (trans) {
5209                 ret = btrfs_end_transaction(trans, root);
5210                 if (!err)
5211                         err = ret;
5212         }
5213         if (err) {
5214                 free_extent_map(em);
5215                 return ERR_PTR(err);
5216         }
5217         return em;
5218 }
5219
5220 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5221                                            size_t pg_offset, u64 start, u64 len,
5222                                            int create)
5223 {
5224         struct extent_map *em;
5225         struct extent_map *hole_em = NULL;
5226         u64 range_start = start;
5227         u64 end;
5228         u64 found;
5229         u64 found_end;
5230         int err = 0;
5231
5232         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5233         if (IS_ERR(em))
5234                 return em;
5235         if (em) {
5236                 /*
5237                  * if our em maps to a hole, there might
5238                  * actually be delalloc bytes behind it
5239                  */
5240                 if (em->block_start != EXTENT_MAP_HOLE)
5241                         return em;
5242                 else
5243                         hole_em = em;
5244         }
5245
5246         /* check to see if we've wrapped (len == -1 or similar) */
5247         end = start + len;
5248         if (end < start)
5249                 end = (u64)-1;
5250         else
5251                 end -= 1;
5252
5253         em = NULL;
5254
5255         /* ok, we didn't find anything, lets look for delalloc */
5256         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5257                                  end, len, EXTENT_DELALLOC, 1);
5258         found_end = range_start + found;
5259         if (found_end < range_start)
5260                 found_end = (u64)-1;
5261
5262         /*
5263          * we didn't find anything useful, return
5264          * the original results from get_extent()
5265          */
5266         if (range_start > end || found_end <= start) {
5267                 em = hole_em;
5268                 hole_em = NULL;
5269                 goto out;
5270         }
5271
5272         /* adjust the range_start to make sure it doesn't
5273          * go backwards from the start they passed in
5274          */
5275         range_start = max(start,range_start);
5276         found = found_end - range_start;
5277
5278         if (found > 0) {
5279                 u64 hole_start = start;
5280                 u64 hole_len = len;
5281
5282                 em = alloc_extent_map();
5283                 if (!em) {
5284                         err = -ENOMEM;
5285                         goto out;
5286                 }
5287                 /*
5288                  * when btrfs_get_extent can't find anything it
5289                  * returns one huge hole
5290                  *
5291                  * make sure what it found really fits our range, and
5292                  * adjust to make sure it is based on the start from
5293                  * the caller
5294                  */
5295                 if (hole_em) {
5296                         u64 calc_end = extent_map_end(hole_em);
5297
5298                         if (calc_end <= start || (hole_em->start > end)) {
5299                                 free_extent_map(hole_em);
5300                                 hole_em = NULL;
5301                         } else {
5302                                 hole_start = max(hole_em->start, start);
5303                                 hole_len = calc_end - hole_start;
5304                         }
5305                 }
5306                 em->bdev = NULL;
5307                 if (hole_em && range_start > hole_start) {
5308                         /* our hole starts before our delalloc, so we
5309                          * have to return just the parts of the hole
5310                          * that go until  the delalloc starts
5311                          */
5312                         em->len = min(hole_len,
5313                                       range_start - hole_start);
5314                         em->start = hole_start;
5315                         em->orig_start = hole_start;
5316                         /*
5317                          * don't adjust block start at all,
5318                          * it is fixed at EXTENT_MAP_HOLE
5319                          */
5320                         em->block_start = hole_em->block_start;
5321                         em->block_len = hole_len;
5322                 } else {
5323                         em->start = range_start;
5324                         em->len = found;
5325                         em->orig_start = range_start;
5326                         em->block_start = EXTENT_MAP_DELALLOC;
5327                         em->block_len = found;
5328                 }
5329         } else if (hole_em) {
5330                 return hole_em;
5331         }
5332 out:
5333
5334         free_extent_map(hole_em);
5335         if (err) {
5336                 free_extent_map(em);
5337                 return ERR_PTR(err);
5338         }
5339         return em;
5340 }
5341
5342 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5343                                                   struct extent_map *em,
5344                                                   u64 start, u64 len)
5345 {
5346         struct btrfs_root *root = BTRFS_I(inode)->root;
5347         struct btrfs_trans_handle *trans;
5348         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5349         struct btrfs_key ins;
5350         u64 alloc_hint;
5351         int ret;
5352         bool insert = false;
5353
5354         /*
5355          * Ok if the extent map we looked up is a hole and is for the exact
5356          * range we want, there is no reason to allocate a new one, however if
5357          * it is not right then we need to free this one and drop the cache for
5358          * our range.
5359          */
5360         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5361             em->len != len) {
5362                 free_extent_map(em);
5363                 em = NULL;
5364                 insert = true;
5365                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5366         }
5367
5368         trans = btrfs_join_transaction(root);
5369         if (IS_ERR(trans))
5370                 return ERR_CAST(trans);
5371
5372         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5373                 btrfs_add_inode_defrag(trans, inode);
5374
5375         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5376
5377         alloc_hint = get_extent_allocation_hint(inode, start, len);
5378         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5379                                    alloc_hint, (u64)-1, &ins, 1);
5380         if (ret) {
5381                 em = ERR_PTR(ret);
5382                 goto out;
5383         }
5384
5385         if (!em) {
5386                 em = alloc_extent_map();
5387                 if (!em) {
5388                         em = ERR_PTR(-ENOMEM);
5389                         goto out;
5390                 }
5391         }
5392
5393         em->start = start;
5394         em->orig_start = em->start;
5395         em->len = ins.offset;
5396
5397         em->block_start = ins.objectid;
5398         em->block_len = ins.offset;
5399         em->bdev = root->fs_info->fs_devices->latest_bdev;
5400
5401         /*
5402          * We need to do this because if we're using the original em we searched
5403          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5404          */
5405         em->flags = 0;
5406         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5407
5408         while (insert) {
5409                 write_lock(&em_tree->lock);
5410                 ret = add_extent_mapping(em_tree, em);
5411                 write_unlock(&em_tree->lock);
5412                 if (ret != -EEXIST)
5413                         break;
5414                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5415         }
5416
5417         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5418                                            ins.offset, ins.offset, 0);
5419         if (ret) {
5420                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5421                 em = ERR_PTR(ret);
5422         }
5423 out:
5424         btrfs_end_transaction(trans, root);
5425         return em;
5426 }
5427
5428 /*
5429  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5430  * block must be cow'd
5431  */
5432 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5433                                       struct inode *inode, u64 offset, u64 len)
5434 {
5435         struct btrfs_path *path;
5436         int ret;
5437         struct extent_buffer *leaf;
5438         struct btrfs_root *root = BTRFS_I(inode)->root;
5439         struct btrfs_file_extent_item *fi;
5440         struct btrfs_key key;
5441         u64 disk_bytenr;
5442         u64 backref_offset;
5443         u64 extent_end;
5444         u64 num_bytes;
5445         int slot;
5446         int found_type;
5447
5448         path = btrfs_alloc_path();
5449         if (!path)
5450                 return -ENOMEM;
5451
5452         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5453                                        offset, 0);
5454         if (ret < 0)
5455                 goto out;
5456
5457         slot = path->slots[0];
5458         if (ret == 1) {
5459                 if (slot == 0) {
5460                         /* can't find the item, must cow */
5461                         ret = 0;
5462                         goto out;
5463                 }
5464                 slot--;
5465         }
5466         ret = 0;
5467         leaf = path->nodes[0];
5468         btrfs_item_key_to_cpu(leaf, &key, slot);
5469         if (key.objectid != btrfs_ino(inode) ||
5470             key.type != BTRFS_EXTENT_DATA_KEY) {
5471                 /* not our file or wrong item type, must cow */
5472                 goto out;
5473         }
5474
5475         if (key.offset > offset) {
5476                 /* Wrong offset, must cow */
5477                 goto out;
5478         }
5479
5480         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5481         found_type = btrfs_file_extent_type(leaf, fi);
5482         if (found_type != BTRFS_FILE_EXTENT_REG &&
5483             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5484                 /* not a regular extent, must cow */
5485                 goto out;
5486         }
5487         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5488         backref_offset = btrfs_file_extent_offset(leaf, fi);
5489
5490         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5491         if (extent_end < offset + len) {
5492                 /* extent doesn't include our full range, must cow */
5493                 goto out;
5494         }
5495
5496         if (btrfs_extent_readonly(root, disk_bytenr))
5497                 goto out;
5498
5499         /*
5500          * look for other files referencing this extent, if we
5501          * find any we must cow
5502          */
5503         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5504                                   key.offset - backref_offset, disk_bytenr))
5505                 goto out;
5506
5507         /*
5508          * adjust disk_bytenr and num_bytes to cover just the bytes
5509          * in this extent we are about to write.  If there
5510          * are any csums in that range we have to cow in order
5511          * to keep the csums correct
5512          */
5513         disk_bytenr += backref_offset;
5514         disk_bytenr += offset - key.offset;
5515         num_bytes = min(offset + len, extent_end) - offset;
5516         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5517                                 goto out;
5518         /*
5519          * all of the above have passed, it is safe to overwrite this extent
5520          * without cow
5521          */
5522         ret = 1;
5523 out:
5524         btrfs_free_path(path);
5525         return ret;
5526 }
5527
5528 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5529                                    struct buffer_head *bh_result, int create)
5530 {
5531         struct extent_map *em;
5532         struct btrfs_root *root = BTRFS_I(inode)->root;
5533         u64 start = iblock << inode->i_blkbits;
5534         u64 len = bh_result->b_size;
5535         struct btrfs_trans_handle *trans;
5536
5537         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5538         if (IS_ERR(em))
5539                 return PTR_ERR(em);
5540
5541         /*
5542          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5543          * io.  INLINE is special, and we could probably kludge it in here, but
5544          * it's still buffered so for safety lets just fall back to the generic
5545          * buffered path.
5546          *
5547          * For COMPRESSED we _have_ to read the entire extent in so we can
5548          * decompress it, so there will be buffering required no matter what we
5549          * do, so go ahead and fallback to buffered.
5550          *
5551          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5552          * to buffered IO.  Don't blame me, this is the price we pay for using
5553          * the generic code.
5554          */
5555         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5556             em->block_start == EXTENT_MAP_INLINE) {
5557                 free_extent_map(em);
5558                 return -ENOTBLK;
5559         }
5560
5561         /* Just a good old fashioned hole, return */
5562         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5563                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5564                 free_extent_map(em);
5565                 /* DIO will do one hole at a time, so just unlock a sector */
5566                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5567                               start + root->sectorsize - 1, GFP_NOFS);
5568                 return 0;
5569         }
5570
5571         /*
5572          * We don't allocate a new extent in the following cases
5573          *
5574          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5575          * existing extent.
5576          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5577          * just use the extent.
5578          *
5579          */
5580         if (!create) {
5581                 len = em->len - (start - em->start);
5582                 goto map;
5583         }
5584
5585         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5586             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5587              em->block_start != EXTENT_MAP_HOLE)) {
5588                 int type;
5589                 int ret;
5590                 u64 block_start;
5591
5592                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5593                         type = BTRFS_ORDERED_PREALLOC;
5594                 else
5595                         type = BTRFS_ORDERED_NOCOW;
5596                 len = min(len, em->len - (start - em->start));
5597                 block_start = em->block_start + (start - em->start);
5598
5599                 /*
5600                  * we're not going to log anything, but we do need
5601                  * to make sure the current transaction stays open
5602                  * while we look for nocow cross refs
5603                  */
5604                 trans = btrfs_join_transaction(root);
5605                 if (IS_ERR(trans))
5606                         goto must_cow;
5607
5608                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5609                         ret = btrfs_add_ordered_extent_dio(inode, start,
5610                                            block_start, len, len, type);
5611                         btrfs_end_transaction(trans, root);
5612                         if (ret) {
5613                                 free_extent_map(em);
5614                                 return ret;
5615                         }
5616                         goto unlock;
5617                 }
5618                 btrfs_end_transaction(trans, root);
5619         }
5620 must_cow:
5621         /*
5622          * this will cow the extent, reset the len in case we changed
5623          * it above
5624          */
5625         len = bh_result->b_size;
5626         em = btrfs_new_extent_direct(inode, em, start, len);
5627         if (IS_ERR(em))
5628                 return PTR_ERR(em);
5629         len = min(len, em->len - (start - em->start));
5630 unlock:
5631         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5632                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5633                           0, NULL, GFP_NOFS);
5634 map:
5635         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5636                 inode->i_blkbits;
5637         bh_result->b_size = len;
5638         bh_result->b_bdev = em->bdev;
5639         set_buffer_mapped(bh_result);
5640         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5641                 set_buffer_new(bh_result);
5642
5643         free_extent_map(em);
5644
5645         return 0;
5646 }
5647
5648 struct btrfs_dio_private {
5649         struct inode *inode;
5650         u64 logical_offset;
5651         u64 disk_bytenr;
5652         u64 bytes;
5653         u32 *csums;
5654         void *private;
5655
5656         /* number of bios pending for this dio */
5657         atomic_t pending_bios;
5658
5659         /* IO errors */
5660         int errors;
5661
5662         struct bio *orig_bio;
5663 };
5664
5665 static void btrfs_endio_direct_read(struct bio *bio, int err)
5666 {
5667         struct btrfs_dio_private *dip = bio->bi_private;
5668         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5669         struct bio_vec *bvec = bio->bi_io_vec;
5670         struct inode *inode = dip->inode;
5671         struct btrfs_root *root = BTRFS_I(inode)->root;
5672         u64 start;
5673         u32 *private = dip->csums;
5674
5675         start = dip->logical_offset;
5676         do {
5677                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5678                         struct page *page = bvec->bv_page;
5679                         char *kaddr;
5680                         u32 csum = ~(u32)0;
5681                         unsigned long flags;
5682
5683                         local_irq_save(flags);
5684                         kaddr = kmap_atomic(page, KM_IRQ0);
5685                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5686                                                csum, bvec->bv_len);
5687                         btrfs_csum_final(csum, (char *)&csum);
5688                         kunmap_atomic(kaddr, KM_IRQ0);
5689                         local_irq_restore(flags);
5690
5691                         flush_dcache_page(bvec->bv_page);
5692                         if (csum != *private) {
5693                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5694                                       " %llu csum %u private %u\n",
5695                                       (unsigned long long)btrfs_ino(inode),
5696                                       (unsigned long long)start,
5697                                       csum, *private);
5698                                 err = -EIO;
5699                         }
5700                 }
5701
5702                 start += bvec->bv_len;
5703                 private++;
5704                 bvec++;
5705         } while (bvec <= bvec_end);
5706
5707         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5708                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5709         bio->bi_private = dip->private;
5710
5711         kfree(dip->csums);
5712         kfree(dip);
5713
5714         /* If we had a csum failure make sure to clear the uptodate flag */
5715         if (err)
5716                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5717         dio_end_io(bio, err);
5718 }
5719
5720 static void btrfs_endio_direct_write(struct bio *bio, int err)
5721 {
5722         struct btrfs_dio_private *dip = bio->bi_private;
5723         struct inode *inode = dip->inode;
5724         struct btrfs_root *root = BTRFS_I(inode)->root;
5725         struct btrfs_trans_handle *trans;
5726         struct btrfs_ordered_extent *ordered = NULL;
5727         struct extent_state *cached_state = NULL;
5728         u64 ordered_offset = dip->logical_offset;
5729         u64 ordered_bytes = dip->bytes;
5730         int ret;
5731
5732         if (err)
5733                 goto out_done;
5734 again:
5735         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5736                                                    &ordered_offset,
5737                                                    ordered_bytes);
5738         if (!ret)
5739                 goto out_test;
5740
5741         BUG_ON(!ordered);
5742
5743         trans = btrfs_join_transaction(root);
5744         if (IS_ERR(trans)) {
5745                 err = -ENOMEM;
5746                 goto out;
5747         }
5748         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5749
5750         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5751                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5752                 if (!ret)
5753                         ret = btrfs_update_inode(trans, root, inode);
5754                 err = ret;
5755                 goto out;
5756         }
5757
5758         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5759                          ordered->file_offset + ordered->len - 1, 0,
5760                          &cached_state, GFP_NOFS);
5761
5762         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5763                 ret = btrfs_mark_extent_written(trans, inode,
5764                                                 ordered->file_offset,
5765                                                 ordered->file_offset +
5766                                                 ordered->len);
5767                 if (ret) {
5768                         err = ret;
5769                         goto out_unlock;
5770                 }
5771         } else {
5772                 ret = insert_reserved_file_extent(trans, inode,
5773                                                   ordered->file_offset,
5774                                                   ordered->start,
5775                                                   ordered->disk_len,
5776                                                   ordered->len,
5777                                                   ordered->len,
5778                                                   0, 0, 0,
5779                                                   BTRFS_FILE_EXTENT_REG);
5780                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5781                                    ordered->file_offset, ordered->len);
5782                 if (ret) {
5783                         err = ret;
5784                         WARN_ON(1);
5785                         goto out_unlock;
5786                 }
5787         }
5788
5789         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5790         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5791         if (!ret)
5792                 btrfs_update_inode(trans, root, inode);
5793         ret = 0;
5794 out_unlock:
5795         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5796                              ordered->file_offset + ordered->len - 1,
5797                              &cached_state, GFP_NOFS);
5798 out:
5799         btrfs_delalloc_release_metadata(inode, ordered->len);
5800         btrfs_end_transaction(trans, root);
5801         ordered_offset = ordered->file_offset + ordered->len;
5802         btrfs_put_ordered_extent(ordered);
5803         btrfs_put_ordered_extent(ordered);
5804
5805 out_test:
5806         /*
5807          * our bio might span multiple ordered extents.  If we haven't
5808          * completed the accounting for the whole dio, go back and try again
5809          */
5810         if (ordered_offset < dip->logical_offset + dip->bytes) {
5811                 ordered_bytes = dip->logical_offset + dip->bytes -
5812                         ordered_offset;
5813                 goto again;
5814         }
5815 out_done:
5816         bio->bi_private = dip->private;
5817
5818         kfree(dip->csums);
5819         kfree(dip);
5820
5821         /* If we had an error make sure to clear the uptodate flag */
5822         if (err)
5823                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5824         dio_end_io(bio, err);
5825 }
5826
5827 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5828                                     struct bio *bio, int mirror_num,
5829                                     unsigned long bio_flags, u64 offset)
5830 {
5831         int ret;
5832         struct btrfs_root *root = BTRFS_I(inode)->root;
5833         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5834         BUG_ON(ret);
5835         return 0;
5836 }
5837
5838 static void btrfs_end_dio_bio(struct bio *bio, int err)
5839 {
5840         struct btrfs_dio_private *dip = bio->bi_private;
5841
5842         if (err) {
5843                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5844                       "sector %#Lx len %u err no %d\n",
5845                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5846                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5847                 dip->errors = 1;
5848
5849                 /*
5850                  * before atomic variable goto zero, we must make sure
5851                  * dip->errors is perceived to be set.
5852                  */
5853                 smp_mb__before_atomic_dec();
5854         }
5855
5856         /* if there are more bios still pending for this dio, just exit */
5857         if (!atomic_dec_and_test(&dip->pending_bios))
5858                 goto out;
5859
5860         if (dip->errors)
5861                 bio_io_error(dip->orig_bio);
5862         else {
5863                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5864                 bio_endio(dip->orig_bio, 0);
5865         }
5866 out:
5867         bio_put(bio);
5868 }
5869
5870 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5871                                        u64 first_sector, gfp_t gfp_flags)
5872 {
5873         int nr_vecs = bio_get_nr_vecs(bdev);
5874         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5875 }
5876
5877 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5878                                          int rw, u64 file_offset, int skip_sum,
5879                                          u32 *csums, int async_submit)
5880 {
5881         int write = rw & REQ_WRITE;
5882         struct btrfs_root *root = BTRFS_I(inode)->root;
5883         int ret;
5884
5885         bio_get(bio);
5886         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5887         if (ret)
5888                 goto err;
5889
5890         if (skip_sum)
5891                 goto map;
5892
5893         if (write && async_submit) {
5894                 ret = btrfs_wq_submit_bio(root->fs_info,
5895                                    inode, rw, bio, 0, 0,
5896                                    file_offset,
5897                                    __btrfs_submit_bio_start_direct_io,
5898                                    __btrfs_submit_bio_done);
5899                 goto err;
5900         } else if (write) {
5901                 /*
5902                  * If we aren't doing async submit, calculate the csum of the
5903                  * bio now.
5904                  */
5905                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5906                 if (ret)
5907                         goto err;
5908         } else if (!skip_sum) {
5909                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5910                                           file_offset, csums);
5911                 if (ret)
5912                         goto err;
5913         }
5914
5915 map:
5916         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5917 err:
5918         bio_put(bio);
5919         return ret;
5920 }
5921
5922 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5923                                     int skip_sum)
5924 {
5925         struct inode *inode = dip->inode;
5926         struct btrfs_root *root = BTRFS_I(inode)->root;
5927         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5928         struct bio *bio;
5929         struct bio *orig_bio = dip->orig_bio;
5930         struct bio_vec *bvec = orig_bio->bi_io_vec;
5931         u64 start_sector = orig_bio->bi_sector;
5932         u64 file_offset = dip->logical_offset;
5933         u64 submit_len = 0;
5934         u64 map_length;
5935         int nr_pages = 0;
5936         u32 *csums = dip->csums;
5937         int ret = 0;
5938         int async_submit = 0;
5939         int write = rw & REQ_WRITE;
5940
5941         map_length = orig_bio->bi_size;
5942         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5943                               &map_length, NULL, 0);
5944         if (ret) {
5945                 bio_put(orig_bio);
5946                 return -EIO;
5947         }
5948
5949         if (map_length >= orig_bio->bi_size) {
5950                 bio = orig_bio;
5951                 goto submit;
5952         }
5953
5954         async_submit = 1;
5955         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5956         if (!bio)
5957                 return -ENOMEM;
5958         bio->bi_private = dip;
5959         bio->bi_end_io = btrfs_end_dio_bio;
5960         atomic_inc(&dip->pending_bios);
5961
5962         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5963                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5964                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5965                                  bvec->bv_offset) < bvec->bv_len)) {
5966                         /*
5967                          * inc the count before we submit the bio so
5968                          * we know the end IO handler won't happen before
5969                          * we inc the count. Otherwise, the dip might get freed
5970                          * before we're done setting it up
5971                          */
5972                         atomic_inc(&dip->pending_bios);
5973                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5974                                                      file_offset, skip_sum,
5975                                                      csums, async_submit);
5976                         if (ret) {
5977                                 bio_put(bio);
5978                                 atomic_dec(&dip->pending_bios);
5979                                 goto out_err;
5980                         }
5981
5982                         /* Write's use the ordered csums */
5983                         if (!write && !skip_sum)
5984                                 csums = csums + nr_pages;
5985                         start_sector += submit_len >> 9;
5986                         file_offset += submit_len;
5987
5988                         submit_len = 0;
5989                         nr_pages = 0;
5990
5991                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5992                                                   start_sector, GFP_NOFS);
5993                         if (!bio)
5994                                 goto out_err;
5995                         bio->bi_private = dip;
5996                         bio->bi_end_io = btrfs_end_dio_bio;
5997
5998                         map_length = orig_bio->bi_size;
5999                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6000                                               &map_length, NULL, 0);
6001                         if (ret) {
6002                                 bio_put(bio);
6003                                 goto out_err;
6004                         }
6005                 } else {
6006                         submit_len += bvec->bv_len;
6007                         nr_pages ++;
6008                         bvec++;
6009                 }
6010         }
6011
6012 submit:
6013         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6014                                      csums, async_submit);
6015         if (!ret)
6016                 return 0;
6017
6018         bio_put(bio);
6019 out_err:
6020         dip->errors = 1;
6021         /*
6022          * before atomic variable goto zero, we must
6023          * make sure dip->errors is perceived to be set.
6024          */
6025         smp_mb__before_atomic_dec();
6026         if (atomic_dec_and_test(&dip->pending_bios))
6027                 bio_io_error(dip->orig_bio);
6028
6029         /* bio_end_io() will handle error, so we needn't return it */
6030         return 0;
6031 }
6032
6033 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6034                                 loff_t file_offset)
6035 {
6036         struct btrfs_root *root = BTRFS_I(inode)->root;
6037         struct btrfs_dio_private *dip;
6038         struct bio_vec *bvec = bio->bi_io_vec;
6039         int skip_sum;
6040         int write = rw & REQ_WRITE;
6041         int ret = 0;
6042
6043         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6044
6045         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6046         if (!dip) {
6047                 ret = -ENOMEM;
6048                 goto free_ordered;
6049         }
6050         dip->csums = NULL;
6051
6052         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6053         if (!write && !skip_sum) {
6054                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6055                 if (!dip->csums) {
6056                         kfree(dip);
6057                         ret = -ENOMEM;
6058                         goto free_ordered;
6059                 }
6060         }
6061
6062         dip->private = bio->bi_private;
6063         dip->inode = inode;
6064         dip->logical_offset = file_offset;
6065
6066         dip->bytes = 0;
6067         do {
6068                 dip->bytes += bvec->bv_len;
6069                 bvec++;
6070         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6071
6072         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6073         bio->bi_private = dip;
6074         dip->errors = 0;
6075         dip->orig_bio = bio;
6076         atomic_set(&dip->pending_bios, 0);
6077
6078         if (write)
6079                 bio->bi_end_io = btrfs_endio_direct_write;
6080         else
6081                 bio->bi_end_io = btrfs_endio_direct_read;
6082
6083         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6084         if (!ret)
6085                 return;
6086 free_ordered:
6087         /*
6088          * If this is a write, we need to clean up the reserved space and kill
6089          * the ordered extent.
6090          */
6091         if (write) {
6092                 struct btrfs_ordered_extent *ordered;
6093                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6094                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6095                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6096                         btrfs_free_reserved_extent(root, ordered->start,
6097                                                    ordered->disk_len);
6098                 btrfs_put_ordered_extent(ordered);
6099                 btrfs_put_ordered_extent(ordered);
6100         }
6101         bio_endio(bio, ret);
6102 }
6103
6104 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6105                         const struct iovec *iov, loff_t offset,
6106                         unsigned long nr_segs)
6107 {
6108         int seg;
6109         int i;
6110         size_t size;
6111         unsigned long addr;
6112         unsigned blocksize_mask = root->sectorsize - 1;
6113         ssize_t retval = -EINVAL;
6114         loff_t end = offset;
6115
6116         if (offset & blocksize_mask)
6117                 goto out;
6118
6119         /* Check the memory alignment.  Blocks cannot straddle pages */
6120         for (seg = 0; seg < nr_segs; seg++) {
6121                 addr = (unsigned long)iov[seg].iov_base;
6122                 size = iov[seg].iov_len;
6123                 end += size;
6124                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6125                         goto out;
6126
6127                 /* If this is a write we don't need to check anymore */
6128                 if (rw & WRITE)
6129                         continue;
6130
6131                 /*
6132                  * Check to make sure we don't have duplicate iov_base's in this
6133                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6134                  * when reading back.
6135                  */
6136                 for (i = seg + 1; i < nr_segs; i++) {
6137                         if (iov[seg].iov_base == iov[i].iov_base)
6138                                 goto out;
6139                 }
6140         }
6141         retval = 0;
6142 out:
6143         return retval;
6144 }
6145 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6146                         const struct iovec *iov, loff_t offset,
6147                         unsigned long nr_segs)
6148 {
6149         struct file *file = iocb->ki_filp;
6150         struct inode *inode = file->f_mapping->host;
6151         struct btrfs_ordered_extent *ordered;
6152         struct extent_state *cached_state = NULL;
6153         u64 lockstart, lockend;
6154         ssize_t ret;
6155         int writing = rw & WRITE;
6156         int write_bits = 0;
6157         size_t count = iov_length(iov, nr_segs);
6158
6159         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6160                             offset, nr_segs)) {
6161                 return 0;
6162         }
6163
6164         lockstart = offset;
6165         lockend = offset + count - 1;
6166
6167         if (writing) {
6168                 ret = btrfs_delalloc_reserve_space(inode, count);
6169                 if (ret)
6170                         goto out;
6171         }
6172
6173         while (1) {
6174                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6175                                  0, &cached_state, GFP_NOFS);
6176                 /*
6177                  * We're concerned with the entire range that we're going to be
6178                  * doing DIO to, so we need to make sure theres no ordered
6179                  * extents in this range.
6180                  */
6181                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6182                                                      lockend - lockstart + 1);
6183                 if (!ordered)
6184                         break;
6185                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186                                      &cached_state, GFP_NOFS);
6187                 btrfs_start_ordered_extent(inode, ordered, 1);
6188                 btrfs_put_ordered_extent(ordered);
6189                 cond_resched();
6190         }
6191
6192         /*
6193          * we don't use btrfs_set_extent_delalloc because we don't want
6194          * the dirty or uptodate bits
6195          */
6196         if (writing) {
6197                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6198                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6199                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6200                                      GFP_NOFS);
6201                 if (ret) {
6202                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6203                                          lockend, EXTENT_LOCKED | write_bits,
6204                                          1, 0, &cached_state, GFP_NOFS);
6205                         goto out;
6206                 }
6207         }
6208
6209         free_extent_state(cached_state);
6210         cached_state = NULL;
6211
6212         ret = __blockdev_direct_IO(rw, iocb, inode,
6213                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6214                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6215                    btrfs_submit_direct, 0);
6216
6217         if (ret < 0 && ret != -EIOCBQUEUED) {
6218                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6219                               offset + iov_length(iov, nr_segs) - 1,
6220                               EXTENT_LOCKED | write_bits, 1, 0,
6221                               &cached_state, GFP_NOFS);
6222         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6223                 /*
6224                  * We're falling back to buffered, unlock the section we didn't
6225                  * do IO on.
6226                  */
6227                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6228                               offset + iov_length(iov, nr_segs) - 1,
6229                               EXTENT_LOCKED | write_bits, 1, 0,
6230                               &cached_state, GFP_NOFS);
6231         }
6232 out:
6233         free_extent_state(cached_state);
6234         return ret;
6235 }
6236
6237 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6238                 __u64 start, __u64 len)
6239 {
6240         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6241 }
6242
6243 int btrfs_readpage(struct file *file, struct page *page)
6244 {
6245         struct extent_io_tree *tree;
6246         tree = &BTRFS_I(page->mapping->host)->io_tree;
6247         return extent_read_full_page(tree, page, btrfs_get_extent);
6248 }
6249
6250 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6251 {
6252         struct extent_io_tree *tree;
6253
6254
6255         if (current->flags & PF_MEMALLOC) {
6256                 redirty_page_for_writepage(wbc, page);
6257                 unlock_page(page);
6258                 return 0;
6259         }
6260         tree = &BTRFS_I(page->mapping->host)->io_tree;
6261         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6262 }
6263
6264 int btrfs_writepages(struct address_space *mapping,
6265                      struct writeback_control *wbc)
6266 {
6267         struct extent_io_tree *tree;
6268
6269         tree = &BTRFS_I(mapping->host)->io_tree;
6270         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6271 }
6272
6273 static int
6274 btrfs_readpages(struct file *file, struct address_space *mapping,
6275                 struct list_head *pages, unsigned nr_pages)
6276 {
6277         struct extent_io_tree *tree;
6278         tree = &BTRFS_I(mapping->host)->io_tree;
6279         return extent_readpages(tree, mapping, pages, nr_pages,
6280                                 btrfs_get_extent);
6281 }
6282 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6283 {
6284         struct extent_io_tree *tree;
6285         struct extent_map_tree *map;
6286         int ret;
6287
6288         tree = &BTRFS_I(page->mapping->host)->io_tree;
6289         map = &BTRFS_I(page->mapping->host)->extent_tree;
6290         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6291         if (ret == 1) {
6292                 ClearPagePrivate(page);
6293                 set_page_private(page, 0);
6294                 page_cache_release(page);
6295         }
6296         return ret;
6297 }
6298
6299 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6300 {
6301         if (PageWriteback(page) || PageDirty(page))
6302                 return 0;
6303         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6304 }
6305
6306 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6307 {
6308         struct extent_io_tree *tree;
6309         struct btrfs_ordered_extent *ordered;
6310         struct extent_state *cached_state = NULL;
6311         u64 page_start = page_offset(page);
6312         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6313
6314
6315         /*
6316          * we have the page locked, so new writeback can't start,
6317          * and the dirty bit won't be cleared while we are here.
6318          *
6319          * Wait for IO on this page so that we can safely clear
6320          * the PagePrivate2 bit and do ordered accounting
6321          */
6322         wait_on_page_writeback(page);
6323
6324         tree = &BTRFS_I(page->mapping->host)->io_tree;
6325         if (offset) {
6326                 btrfs_releasepage(page, GFP_NOFS);
6327                 return;
6328         }
6329         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6330                          GFP_NOFS);
6331         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6332                                            page_offset(page));
6333         if (ordered) {
6334                 /*
6335                  * IO on this page will never be started, so we need
6336                  * to account for any ordered extents now
6337                  */
6338                 clear_extent_bit(tree, page_start, page_end,
6339                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6340                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6341                                  &cached_state, GFP_NOFS);
6342                 /*
6343                  * whoever cleared the private bit is responsible
6344                  * for the finish_ordered_io
6345                  */
6346                 if (TestClearPagePrivate2(page)) {
6347                         btrfs_finish_ordered_io(page->mapping->host,
6348                                                 page_start, page_end);
6349                 }
6350                 btrfs_put_ordered_extent(ordered);
6351                 cached_state = NULL;
6352                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6353                                  GFP_NOFS);
6354         }
6355         clear_extent_bit(tree, page_start, page_end,
6356                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6357                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6358         __btrfs_releasepage(page, GFP_NOFS);
6359
6360         ClearPageChecked(page);
6361         if (PagePrivate(page)) {
6362                 ClearPagePrivate(page);
6363                 set_page_private(page, 0);
6364                 page_cache_release(page);
6365         }
6366 }
6367
6368 /*
6369  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6370  * called from a page fault handler when a page is first dirtied. Hence we must
6371  * be careful to check for EOF conditions here. We set the page up correctly
6372  * for a written page which means we get ENOSPC checking when writing into
6373  * holes and correct delalloc and unwritten extent mapping on filesystems that
6374  * support these features.
6375  *
6376  * We are not allowed to take the i_mutex here so we have to play games to
6377  * protect against truncate races as the page could now be beyond EOF.  Because
6378  * vmtruncate() writes the inode size before removing pages, once we have the
6379  * page lock we can determine safely if the page is beyond EOF. If it is not
6380  * beyond EOF, then the page is guaranteed safe against truncation until we
6381  * unlock the page.
6382  */
6383 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6384 {
6385         struct page *page = vmf->page;
6386         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6387         struct btrfs_root *root = BTRFS_I(inode)->root;
6388         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6389         struct btrfs_ordered_extent *ordered;
6390         struct extent_state *cached_state = NULL;
6391         char *kaddr;
6392         unsigned long zero_start;
6393         loff_t size;
6394         int ret;
6395         u64 page_start;
6396         u64 page_end;
6397
6398         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6399         if (ret) {
6400                 if (ret == -ENOMEM)
6401                         ret = VM_FAULT_OOM;
6402                 else /* -ENOSPC, -EIO, etc */
6403                         ret = VM_FAULT_SIGBUS;
6404                 goto out;
6405         }
6406
6407         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6408 again:
6409         lock_page(page);
6410         size = i_size_read(inode);
6411         page_start = page_offset(page);
6412         page_end = page_start + PAGE_CACHE_SIZE - 1;
6413
6414         if ((page->mapping != inode->i_mapping) ||
6415             (page_start >= size)) {
6416                 /* page got truncated out from underneath us */
6417                 goto out_unlock;
6418         }
6419         wait_on_page_writeback(page);
6420
6421         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6422                          GFP_NOFS);
6423         set_page_extent_mapped(page);
6424
6425         /*
6426          * we can't set the delalloc bits if there are pending ordered
6427          * extents.  Drop our locks and wait for them to finish
6428          */
6429         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6430         if (ordered) {
6431                 unlock_extent_cached(io_tree, page_start, page_end,
6432                                      &cached_state, GFP_NOFS);
6433                 unlock_page(page);
6434                 btrfs_start_ordered_extent(inode, ordered, 1);
6435                 btrfs_put_ordered_extent(ordered);
6436                 goto again;
6437         }
6438
6439         /*
6440          * XXX - page_mkwrite gets called every time the page is dirtied, even
6441          * if it was already dirty, so for space accounting reasons we need to
6442          * clear any delalloc bits for the range we are fixing to save.  There
6443          * is probably a better way to do this, but for now keep consistent with
6444          * prepare_pages in the normal write path.
6445          */
6446         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6447                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6448                           0, 0, &cached_state, GFP_NOFS);
6449
6450         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6451                                         &cached_state);
6452         if (ret) {
6453                 unlock_extent_cached(io_tree, page_start, page_end,
6454                                      &cached_state, GFP_NOFS);
6455                 ret = VM_FAULT_SIGBUS;
6456                 goto out_unlock;
6457         }
6458         ret = 0;
6459
6460         /* page is wholly or partially inside EOF */
6461         if (page_start + PAGE_CACHE_SIZE > size)
6462                 zero_start = size & ~PAGE_CACHE_MASK;
6463         else
6464                 zero_start = PAGE_CACHE_SIZE;
6465
6466         if (zero_start != PAGE_CACHE_SIZE) {
6467                 kaddr = kmap(page);
6468                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6469                 flush_dcache_page(page);
6470                 kunmap(page);
6471         }
6472         ClearPageChecked(page);
6473         set_page_dirty(page);
6474         SetPageUptodate(page);
6475
6476         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6477         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6478
6479         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6480
6481 out_unlock:
6482         if (!ret)
6483                 return VM_FAULT_LOCKED;
6484         unlock_page(page);
6485         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6486 out:
6487         return ret;
6488 }
6489
6490 static int btrfs_truncate(struct inode *inode)
6491 {
6492         struct btrfs_root *root = BTRFS_I(inode)->root;
6493         struct btrfs_block_rsv *rsv;
6494         int ret;
6495         int err = 0;
6496         struct btrfs_trans_handle *trans;
6497         unsigned long nr;
6498         u64 mask = root->sectorsize - 1;
6499
6500         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6501         if (ret)
6502                 return ret;
6503
6504         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6505         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6506
6507         /*
6508          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6509          * 3 things going on here
6510          *
6511          * 1) We need to reserve space for our orphan item and the space to
6512          * delete our orphan item.  Lord knows we don't want to have a dangling
6513          * orphan item because we didn't reserve space to remove it.
6514          *
6515          * 2) We need to reserve space to update our inode.
6516          *
6517          * 3) We need to have something to cache all the space that is going to
6518          * be free'd up by the truncate operation, but also have some slack
6519          * space reserved in case it uses space during the truncate (thank you
6520          * very much snapshotting).
6521          *
6522          * And we need these to all be seperate.  The fact is we can use alot of
6523          * space doing the truncate, and we have no earthly idea how much space
6524          * we will use, so we need the truncate reservation to be seperate so it
6525          * doesn't end up using space reserved for updating the inode or
6526          * removing the orphan item.  We also need to be able to stop the
6527          * transaction and start a new one, which means we need to be able to
6528          * update the inode several times, and we have no idea of knowing how
6529          * many times that will be, so we can't just reserve 1 item for the
6530          * entirety of the opration, so that has to be done seperately as well.
6531          * Then there is the orphan item, which does indeed need to be held on
6532          * to for the whole operation, and we need nobody to touch this reserved
6533          * space except the orphan code.
6534          *
6535          * So that leaves us with
6536          *
6537          * 1) root->orphan_block_rsv - for the orphan deletion.
6538          * 2) rsv - for the truncate reservation, which we will steal from the
6539          * transaction reservation.
6540          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6541          * updating the inode.
6542          */
6543         rsv = btrfs_alloc_block_rsv(root);
6544         if (!rsv)
6545                 return -ENOMEM;
6546         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6547
6548         trans = btrfs_start_transaction(root, 4);
6549         if (IS_ERR(trans)) {
6550                 err = PTR_ERR(trans);
6551                 goto out;
6552         }
6553
6554         /*
6555          * Reserve space for the truncate process.  Truncate should be adding
6556          * space, but if there are snapshots it may end up using space.
6557          */
6558         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6559         BUG_ON(ret);
6560
6561         ret = btrfs_orphan_add(trans, inode);
6562         if (ret) {
6563                 btrfs_end_transaction(trans, root);
6564                 goto out;
6565         }
6566
6567         nr = trans->blocks_used;
6568         btrfs_end_transaction(trans, root);
6569         btrfs_btree_balance_dirty(root, nr);
6570
6571         /*
6572          * Ok so we've already migrated our bytes over for the truncate, so here
6573          * just reserve the one slot we need for updating the inode.
6574          */
6575         trans = btrfs_start_transaction(root, 1);
6576         if (IS_ERR(trans)) {
6577                 err = PTR_ERR(trans);
6578                 goto out;
6579         }
6580         trans->block_rsv = rsv;
6581
6582         /*
6583          * setattr is responsible for setting the ordered_data_close flag,
6584          * but that is only tested during the last file release.  That
6585          * could happen well after the next commit, leaving a great big
6586          * window where new writes may get lost if someone chooses to write
6587          * to this file after truncating to zero
6588          *
6589          * The inode doesn't have any dirty data here, and so if we commit
6590          * this is a noop.  If someone immediately starts writing to the inode
6591          * it is very likely we'll catch some of their writes in this
6592          * transaction, and the commit will find this file on the ordered
6593          * data list with good things to send down.
6594          *
6595          * This is a best effort solution, there is still a window where
6596          * using truncate to replace the contents of the file will
6597          * end up with a zero length file after a crash.
6598          */
6599         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6600                 btrfs_add_ordered_operation(trans, root, inode);
6601
6602         while (1) {
6603                 if (!trans) {
6604                         trans = btrfs_start_transaction(root, 3);
6605                         if (IS_ERR(trans)) {
6606                                 err = PTR_ERR(trans);
6607                                 goto out;
6608                         }
6609
6610                         ret = btrfs_truncate_reserve_metadata(trans, root,
6611                                                               rsv);
6612                         BUG_ON(ret);
6613
6614                         trans->block_rsv = rsv;
6615                 }
6616
6617                 ret = btrfs_truncate_inode_items(trans, root, inode,
6618                                                  inode->i_size,
6619                                                  BTRFS_EXTENT_DATA_KEY);
6620                 if (ret != -EAGAIN) {
6621                         err = ret;
6622                         break;
6623                 }
6624
6625                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6626                 ret = btrfs_update_inode(trans, root, inode);
6627                 if (ret) {
6628                         err = ret;
6629                         break;
6630                 }
6631
6632                 nr = trans->blocks_used;
6633                 btrfs_end_transaction(trans, root);
6634                 trans = NULL;
6635                 btrfs_btree_balance_dirty(root, nr);
6636         }
6637
6638         if (ret == 0 && inode->i_nlink > 0) {
6639                 trans->block_rsv = root->orphan_block_rsv;
6640                 ret = btrfs_orphan_del(trans, inode);
6641                 if (ret)
6642                         err = ret;
6643         } else if (ret && inode->i_nlink > 0) {
6644                 /*
6645                  * Failed to do the truncate, remove us from the in memory
6646                  * orphan list.
6647                  */
6648                 ret = btrfs_orphan_del(NULL, inode);
6649         }
6650
6651         trans->block_rsv = &root->fs_info->trans_block_rsv;
6652         ret = btrfs_update_inode(trans, root, inode);
6653         if (ret && !err)
6654                 err = ret;
6655
6656         nr = trans->blocks_used;
6657         ret = btrfs_end_transaction_throttle(trans, root);
6658         btrfs_btree_balance_dirty(root, nr);
6659
6660 out:
6661         btrfs_free_block_rsv(root, rsv);
6662
6663         if (ret && !err)
6664                 err = ret;
6665
6666         return err;
6667 }
6668
6669 /*
6670  * create a new subvolume directory/inode (helper for the ioctl).
6671  */
6672 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6673                              struct btrfs_root *new_root, u64 new_dirid)
6674 {
6675         struct inode *inode;
6676         int err;
6677         u64 index = 0;
6678
6679         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6680                                 new_dirid, S_IFDIR | 0700, &index);
6681         if (IS_ERR(inode))
6682                 return PTR_ERR(inode);
6683         inode->i_op = &btrfs_dir_inode_operations;
6684         inode->i_fop = &btrfs_dir_file_operations;
6685
6686         inode->i_nlink = 1;
6687         btrfs_i_size_write(inode, 0);
6688
6689         err = btrfs_update_inode(trans, new_root, inode);
6690         BUG_ON(err);
6691
6692         iput(inode);
6693         return 0;
6694 }
6695
6696 struct inode *btrfs_alloc_inode(struct super_block *sb)
6697 {
6698         struct btrfs_inode *ei;
6699         struct inode *inode;
6700
6701         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6702         if (!ei)
6703                 return NULL;
6704
6705         ei->root = NULL;
6706         ei->space_info = NULL;
6707         ei->generation = 0;
6708         ei->sequence = 0;
6709         ei->last_trans = 0;
6710         ei->last_sub_trans = 0;
6711         ei->logged_trans = 0;
6712         ei->delalloc_bytes = 0;
6713         ei->reserved_bytes = 0;
6714         ei->disk_i_size = 0;
6715         ei->flags = 0;
6716         ei->index_cnt = (u64)-1;
6717         ei->last_unlink_trans = 0;
6718
6719         spin_lock_init(&ei->lock);
6720         ei->outstanding_extents = 0;
6721         ei->reserved_extents = 0;
6722
6723         ei->ordered_data_close = 0;
6724         ei->orphan_meta_reserved = 0;
6725         ei->dummy_inode = 0;
6726         ei->in_defrag = 0;
6727         ei->force_compress = BTRFS_COMPRESS_NONE;
6728
6729         ei->delayed_node = NULL;
6730
6731         inode = &ei->vfs_inode;
6732         extent_map_tree_init(&ei->extent_tree);
6733         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6734         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6735         mutex_init(&ei->log_mutex);
6736         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6737         INIT_LIST_HEAD(&ei->i_orphan);
6738         INIT_LIST_HEAD(&ei->delalloc_inodes);
6739         INIT_LIST_HEAD(&ei->ordered_operations);
6740         RB_CLEAR_NODE(&ei->rb_node);
6741
6742         return inode;
6743 }
6744
6745 static void btrfs_i_callback(struct rcu_head *head)
6746 {
6747         struct inode *inode = container_of(head, struct inode, i_rcu);
6748         INIT_LIST_HEAD(&inode->i_dentry);
6749         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6750 }
6751
6752 void btrfs_destroy_inode(struct inode *inode)
6753 {
6754         struct btrfs_ordered_extent *ordered;
6755         struct btrfs_root *root = BTRFS_I(inode)->root;
6756
6757         WARN_ON(!list_empty(&inode->i_dentry));
6758         WARN_ON(inode->i_data.nrpages);
6759         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6760         WARN_ON(BTRFS_I(inode)->reserved_extents);
6761
6762         /*
6763          * This can happen where we create an inode, but somebody else also
6764          * created the same inode and we need to destroy the one we already
6765          * created.
6766          */
6767         if (!root)
6768                 goto free;
6769
6770         /*
6771          * Make sure we're properly removed from the ordered operation
6772          * lists.
6773          */
6774         smp_mb();
6775         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6776                 spin_lock(&root->fs_info->ordered_extent_lock);
6777                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6778                 spin_unlock(&root->fs_info->ordered_extent_lock);
6779         }
6780
6781         spin_lock(&root->orphan_lock);
6782         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6783                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6784                        (unsigned long long)btrfs_ino(inode));
6785                 list_del_init(&BTRFS_I(inode)->i_orphan);
6786         }
6787         spin_unlock(&root->orphan_lock);
6788
6789         while (1) {
6790                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6791                 if (!ordered)
6792                         break;
6793                 else {
6794                         printk(KERN_ERR "btrfs found ordered "
6795                                "extent %llu %llu on inode cleanup\n",
6796                                (unsigned long long)ordered->file_offset,
6797                                (unsigned long long)ordered->len);
6798                         btrfs_remove_ordered_extent(inode, ordered);
6799                         btrfs_put_ordered_extent(ordered);
6800                         btrfs_put_ordered_extent(ordered);
6801                 }
6802         }
6803         inode_tree_del(inode);
6804         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6805 free:
6806         btrfs_remove_delayed_node(inode);
6807         call_rcu(&inode->i_rcu, btrfs_i_callback);
6808 }
6809
6810 int btrfs_drop_inode(struct inode *inode)
6811 {
6812         struct btrfs_root *root = BTRFS_I(inode)->root;
6813
6814         if (btrfs_root_refs(&root->root_item) == 0 &&
6815             !btrfs_is_free_space_inode(root, inode))
6816                 return 1;
6817         else
6818                 return generic_drop_inode(inode);
6819 }
6820
6821 static void init_once(void *foo)
6822 {
6823         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6824
6825         inode_init_once(&ei->vfs_inode);
6826 }
6827
6828 void btrfs_destroy_cachep(void)
6829 {
6830         if (btrfs_inode_cachep)
6831                 kmem_cache_destroy(btrfs_inode_cachep);
6832         if (btrfs_trans_handle_cachep)
6833                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6834         if (btrfs_transaction_cachep)
6835                 kmem_cache_destroy(btrfs_transaction_cachep);
6836         if (btrfs_path_cachep)
6837                 kmem_cache_destroy(btrfs_path_cachep);
6838         if (btrfs_free_space_cachep)
6839                 kmem_cache_destroy(btrfs_free_space_cachep);
6840 }
6841
6842 int btrfs_init_cachep(void)
6843 {
6844         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6845                         sizeof(struct btrfs_inode), 0,
6846                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6847         if (!btrfs_inode_cachep)
6848                 goto fail;
6849
6850         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6851                         sizeof(struct btrfs_trans_handle), 0,
6852                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6853         if (!btrfs_trans_handle_cachep)
6854                 goto fail;
6855
6856         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6857                         sizeof(struct btrfs_transaction), 0,
6858                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6859         if (!btrfs_transaction_cachep)
6860                 goto fail;
6861
6862         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6863                         sizeof(struct btrfs_path), 0,
6864                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6865         if (!btrfs_path_cachep)
6866                 goto fail;
6867
6868         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6869                         sizeof(struct btrfs_free_space), 0,
6870                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6871         if (!btrfs_free_space_cachep)
6872                 goto fail;
6873
6874         return 0;
6875 fail:
6876         btrfs_destroy_cachep();
6877         return -ENOMEM;
6878 }
6879
6880 static int btrfs_getattr(struct vfsmount *mnt,
6881                          struct dentry *dentry, struct kstat *stat)
6882 {
6883         struct inode *inode = dentry->d_inode;
6884         generic_fillattr(inode, stat);
6885         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6886         stat->blksize = PAGE_CACHE_SIZE;
6887         stat->blocks = (inode_get_bytes(inode) +
6888                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6889         return 0;
6890 }
6891
6892 /*
6893  * If a file is moved, it will inherit the cow and compression flags of the new
6894  * directory.
6895  */
6896 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6897 {
6898         struct btrfs_inode *b_dir = BTRFS_I(dir);
6899         struct btrfs_inode *b_inode = BTRFS_I(inode);
6900
6901         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6902                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6903         else
6904                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6905
6906         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6907                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6908         else
6909                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6910 }
6911
6912 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6913                            struct inode *new_dir, struct dentry *new_dentry)
6914 {
6915         struct btrfs_trans_handle *trans;
6916         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6917         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6918         struct inode *new_inode = new_dentry->d_inode;
6919         struct inode *old_inode = old_dentry->d_inode;
6920         struct timespec ctime = CURRENT_TIME;
6921         u64 index = 0;
6922         u64 root_objectid;
6923         int ret;
6924         u64 old_ino = btrfs_ino(old_inode);
6925
6926         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6927                 return -EPERM;
6928
6929         /* we only allow rename subvolume link between subvolumes */
6930         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6931                 return -EXDEV;
6932
6933         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6934             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6935                 return -ENOTEMPTY;
6936
6937         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6938             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6939                 return -ENOTEMPTY;
6940         /*
6941          * we're using rename to replace one file with another.
6942          * and the replacement file is large.  Start IO on it now so
6943          * we don't add too much work to the end of the transaction
6944          */
6945         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6946             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6947                 filemap_flush(old_inode->i_mapping);
6948
6949         /* close the racy window with snapshot create/destroy ioctl */
6950         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6951                 down_read(&root->fs_info->subvol_sem);
6952         /*
6953          * We want to reserve the absolute worst case amount of items.  So if
6954          * both inodes are subvols and we need to unlink them then that would
6955          * require 4 item modifications, but if they are both normal inodes it
6956          * would require 5 item modifications, so we'll assume their normal
6957          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6958          * should cover the worst case number of items we'll modify.
6959          */
6960         trans = btrfs_start_transaction(root, 20);
6961         if (IS_ERR(trans)) {
6962                 ret = PTR_ERR(trans);
6963                 goto out_notrans;
6964         }
6965
6966         if (dest != root)
6967                 btrfs_record_root_in_trans(trans, dest);
6968
6969         ret = btrfs_set_inode_index(new_dir, &index);
6970         if (ret)
6971                 goto out_fail;
6972
6973         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6974                 /* force full log commit if subvolume involved. */
6975                 root->fs_info->last_trans_log_full_commit = trans->transid;
6976         } else {
6977                 ret = btrfs_insert_inode_ref(trans, dest,
6978                                              new_dentry->d_name.name,
6979                                              new_dentry->d_name.len,
6980                                              old_ino,
6981                                              btrfs_ino(new_dir), index);
6982                 if (ret)
6983                         goto out_fail;
6984                 /*
6985                  * this is an ugly little race, but the rename is required
6986                  * to make sure that if we crash, the inode is either at the
6987                  * old name or the new one.  pinning the log transaction lets
6988                  * us make sure we don't allow a log commit to come in after
6989                  * we unlink the name but before we add the new name back in.
6990                  */
6991                 btrfs_pin_log_trans(root);
6992         }
6993         /*
6994          * make sure the inode gets flushed if it is replacing
6995          * something.
6996          */
6997         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6998                 btrfs_add_ordered_operation(trans, root, old_inode);
6999
7000         old_dir->i_ctime = old_dir->i_mtime = ctime;
7001         new_dir->i_ctime = new_dir->i_mtime = ctime;
7002         old_inode->i_ctime = ctime;
7003
7004         if (old_dentry->d_parent != new_dentry->d_parent)
7005                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7006
7007         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7008                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7009                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7010                                         old_dentry->d_name.name,
7011                                         old_dentry->d_name.len);
7012         } else {
7013                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7014                                         old_dentry->d_inode,
7015                                         old_dentry->d_name.name,
7016                                         old_dentry->d_name.len);
7017                 if (!ret)
7018                         ret = btrfs_update_inode(trans, root, old_inode);
7019         }
7020         BUG_ON(ret);
7021
7022         if (new_inode) {
7023                 new_inode->i_ctime = CURRENT_TIME;
7024                 if (unlikely(btrfs_ino(new_inode) ==
7025                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7026                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7027                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7028                                                 root_objectid,
7029                                                 new_dentry->d_name.name,
7030                                                 new_dentry->d_name.len);
7031                         BUG_ON(new_inode->i_nlink == 0);
7032                 } else {
7033                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7034                                                  new_dentry->d_inode,
7035                                                  new_dentry->d_name.name,
7036                                                  new_dentry->d_name.len);
7037                 }
7038                 BUG_ON(ret);
7039                 if (new_inode->i_nlink == 0) {
7040                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7041                         BUG_ON(ret);
7042                 }
7043         }
7044
7045         fixup_inode_flags(new_dir, old_inode);
7046
7047         ret = btrfs_add_link(trans, new_dir, old_inode,
7048                              new_dentry->d_name.name,
7049                              new_dentry->d_name.len, 0, index);
7050         BUG_ON(ret);
7051
7052         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7053                 struct dentry *parent = dget_parent(new_dentry);
7054                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7055                 dput(parent);
7056                 btrfs_end_log_trans(root);
7057         }
7058 out_fail:
7059         btrfs_end_transaction_throttle(trans, root);
7060 out_notrans:
7061         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7062                 up_read(&root->fs_info->subvol_sem);
7063
7064         return ret;
7065 }
7066
7067 /*
7068  * some fairly slow code that needs optimization. This walks the list
7069  * of all the inodes with pending delalloc and forces them to disk.
7070  */
7071 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7072 {
7073         struct list_head *head = &root->fs_info->delalloc_inodes;
7074         struct btrfs_inode *binode;
7075         struct inode *inode;
7076
7077         if (root->fs_info->sb->s_flags & MS_RDONLY)
7078                 return -EROFS;
7079
7080         spin_lock(&root->fs_info->delalloc_lock);
7081         while (!list_empty(head)) {
7082                 binode = list_entry(head->next, struct btrfs_inode,
7083                                     delalloc_inodes);
7084                 inode = igrab(&binode->vfs_inode);
7085                 if (!inode)
7086                         list_del_init(&binode->delalloc_inodes);
7087                 spin_unlock(&root->fs_info->delalloc_lock);
7088                 if (inode) {
7089                         filemap_flush(inode->i_mapping);
7090                         if (delay_iput)
7091                                 btrfs_add_delayed_iput(inode);
7092                         else
7093                                 iput(inode);
7094                 }
7095                 cond_resched();
7096                 spin_lock(&root->fs_info->delalloc_lock);
7097         }
7098         spin_unlock(&root->fs_info->delalloc_lock);
7099
7100         /* the filemap_flush will queue IO into the worker threads, but
7101          * we have to make sure the IO is actually started and that
7102          * ordered extents get created before we return
7103          */
7104         atomic_inc(&root->fs_info->async_submit_draining);
7105         while (atomic_read(&root->fs_info->nr_async_submits) ||
7106               atomic_read(&root->fs_info->async_delalloc_pages)) {
7107                 wait_event(root->fs_info->async_submit_wait,
7108                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7109                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7110         }
7111         atomic_dec(&root->fs_info->async_submit_draining);
7112         return 0;
7113 }
7114
7115 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7116                          const char *symname)
7117 {
7118         struct btrfs_trans_handle *trans;
7119         struct btrfs_root *root = BTRFS_I(dir)->root;
7120         struct btrfs_path *path;
7121         struct btrfs_key key;
7122         struct inode *inode = NULL;
7123         int err;
7124         int drop_inode = 0;
7125         u64 objectid;
7126         u64 index = 0 ;
7127         int name_len;
7128         int datasize;
7129         unsigned long ptr;
7130         struct btrfs_file_extent_item *ei;
7131         struct extent_buffer *leaf;
7132         unsigned long nr = 0;
7133
7134         name_len = strlen(symname) + 1;
7135         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7136                 return -ENAMETOOLONG;
7137
7138         /*
7139          * 2 items for inode item and ref
7140          * 2 items for dir items
7141          * 1 item for xattr if selinux is on
7142          */
7143         trans = btrfs_start_transaction(root, 5);
7144         if (IS_ERR(trans))
7145                 return PTR_ERR(trans);
7146
7147         err = btrfs_find_free_ino(root, &objectid);
7148         if (err)
7149                 goto out_unlock;
7150
7151         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7152                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7153                                 S_IFLNK|S_IRWXUGO, &index);
7154         if (IS_ERR(inode)) {
7155                 err = PTR_ERR(inode);
7156                 goto out_unlock;
7157         }
7158
7159         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7160         if (err) {
7161                 drop_inode = 1;
7162                 goto out_unlock;
7163         }
7164
7165         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7166         if (err)
7167                 drop_inode = 1;
7168         else {
7169                 inode->i_mapping->a_ops = &btrfs_aops;
7170                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7171                 inode->i_fop = &btrfs_file_operations;
7172                 inode->i_op = &btrfs_file_inode_operations;
7173                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7174         }
7175         if (drop_inode)
7176                 goto out_unlock;
7177
7178         path = btrfs_alloc_path();
7179         if (!path) {
7180                 err = -ENOMEM;
7181                 drop_inode = 1;
7182                 goto out_unlock;
7183         }
7184         key.objectid = btrfs_ino(inode);
7185         key.offset = 0;
7186         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7187         datasize = btrfs_file_extent_calc_inline_size(name_len);
7188         err = btrfs_insert_empty_item(trans, root, path, &key,
7189                                       datasize);
7190         if (err) {
7191                 drop_inode = 1;
7192                 btrfs_free_path(path);
7193                 goto out_unlock;
7194         }
7195         leaf = path->nodes[0];
7196         ei = btrfs_item_ptr(leaf, path->slots[0],
7197                             struct btrfs_file_extent_item);
7198         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7199         btrfs_set_file_extent_type(leaf, ei,
7200                                    BTRFS_FILE_EXTENT_INLINE);
7201         btrfs_set_file_extent_encryption(leaf, ei, 0);
7202         btrfs_set_file_extent_compression(leaf, ei, 0);
7203         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7204         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7205
7206         ptr = btrfs_file_extent_inline_start(ei);
7207         write_extent_buffer(leaf, symname, ptr, name_len);
7208         btrfs_mark_buffer_dirty(leaf);
7209         btrfs_free_path(path);
7210
7211         inode->i_op = &btrfs_symlink_inode_operations;
7212         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7213         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7214         inode_set_bytes(inode, name_len);
7215         btrfs_i_size_write(inode, name_len - 1);
7216         err = btrfs_update_inode(trans, root, inode);
7217         if (err)
7218                 drop_inode = 1;
7219
7220 out_unlock:
7221         nr = trans->blocks_used;
7222         btrfs_end_transaction_throttle(trans, root);
7223         if (drop_inode) {
7224                 inode_dec_link_count(inode);
7225                 iput(inode);
7226         }
7227         btrfs_btree_balance_dirty(root, nr);
7228         return err;
7229 }
7230
7231 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7232                                        u64 start, u64 num_bytes, u64 min_size,
7233                                        loff_t actual_len, u64 *alloc_hint,
7234                                        struct btrfs_trans_handle *trans)
7235 {
7236         struct btrfs_root *root = BTRFS_I(inode)->root;
7237         struct btrfs_key ins;
7238         u64 cur_offset = start;
7239         u64 i_size;
7240         int ret = 0;
7241         bool own_trans = true;
7242
7243         if (trans)
7244                 own_trans = false;
7245         while (num_bytes > 0) {
7246                 if (own_trans) {
7247                         trans = btrfs_start_transaction(root, 3);
7248                         if (IS_ERR(trans)) {
7249                                 ret = PTR_ERR(trans);
7250                                 break;
7251                         }
7252                 }
7253
7254                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7255                                            0, *alloc_hint, (u64)-1, &ins, 1);
7256                 if (ret) {
7257                         if (own_trans)
7258                                 btrfs_end_transaction(trans, root);
7259                         break;
7260                 }
7261
7262                 ret = insert_reserved_file_extent(trans, inode,
7263                                                   cur_offset, ins.objectid,
7264                                                   ins.offset, ins.offset,
7265                                                   ins.offset, 0, 0, 0,
7266                                                   BTRFS_FILE_EXTENT_PREALLOC);
7267                 BUG_ON(ret);
7268                 btrfs_drop_extent_cache(inode, cur_offset,
7269                                         cur_offset + ins.offset -1, 0);
7270
7271                 num_bytes -= ins.offset;
7272                 cur_offset += ins.offset;
7273                 *alloc_hint = ins.objectid + ins.offset;
7274
7275                 inode->i_ctime = CURRENT_TIME;
7276                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7277                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7278                     (actual_len > inode->i_size) &&
7279                     (cur_offset > inode->i_size)) {
7280                         if (cur_offset > actual_len)
7281                                 i_size = actual_len;
7282                         else
7283                                 i_size = cur_offset;
7284                         i_size_write(inode, i_size);
7285                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7286                 }
7287
7288                 ret = btrfs_update_inode(trans, root, inode);
7289                 BUG_ON(ret);
7290
7291                 if (own_trans)
7292                         btrfs_end_transaction(trans, root);
7293         }
7294         return ret;
7295 }
7296
7297 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7298                               u64 start, u64 num_bytes, u64 min_size,
7299                               loff_t actual_len, u64 *alloc_hint)
7300 {
7301         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7302                                            min_size, actual_len, alloc_hint,
7303                                            NULL);
7304 }
7305
7306 int btrfs_prealloc_file_range_trans(struct inode *inode,
7307                                     struct btrfs_trans_handle *trans, int mode,
7308                                     u64 start, u64 num_bytes, u64 min_size,
7309                                     loff_t actual_len, u64 *alloc_hint)
7310 {
7311         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7312                                            min_size, actual_len, alloc_hint, trans);
7313 }
7314
7315 static int btrfs_set_page_dirty(struct page *page)
7316 {
7317         return __set_page_dirty_nobuffers(page);
7318 }
7319
7320 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7321 {
7322         struct btrfs_root *root = BTRFS_I(inode)->root;
7323
7324         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7325                 return -EROFS;
7326         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7327                 return -EACCES;
7328         return generic_permission(inode, mask, flags, btrfs_check_acl);
7329 }
7330
7331 static const struct inode_operations btrfs_dir_inode_operations = {
7332         .getattr        = btrfs_getattr,
7333         .lookup         = btrfs_lookup,
7334         .create         = btrfs_create,
7335         .unlink         = btrfs_unlink,
7336         .link           = btrfs_link,
7337         .mkdir          = btrfs_mkdir,
7338         .rmdir          = btrfs_rmdir,
7339         .rename         = btrfs_rename,
7340         .symlink        = btrfs_symlink,
7341         .setattr        = btrfs_setattr,
7342         .mknod          = btrfs_mknod,
7343         .setxattr       = btrfs_setxattr,
7344         .getxattr       = btrfs_getxattr,
7345         .listxattr      = btrfs_listxattr,
7346         .removexattr    = btrfs_removexattr,
7347         .permission     = btrfs_permission,
7348 };
7349 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7350         .lookup         = btrfs_lookup,
7351         .permission     = btrfs_permission,
7352 };
7353
7354 static const struct file_operations btrfs_dir_file_operations = {
7355         .llseek         = generic_file_llseek,
7356         .read           = generic_read_dir,
7357         .readdir        = btrfs_real_readdir,
7358         .unlocked_ioctl = btrfs_ioctl,
7359 #ifdef CONFIG_COMPAT
7360         .compat_ioctl   = btrfs_ioctl,
7361 #endif
7362         .release        = btrfs_release_file,
7363         .fsync          = btrfs_sync_file,
7364 };
7365
7366 static struct extent_io_ops btrfs_extent_io_ops = {
7367         .fill_delalloc = run_delalloc_range,
7368         .submit_bio_hook = btrfs_submit_bio_hook,
7369         .merge_bio_hook = btrfs_merge_bio_hook,
7370         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7371         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7372         .writepage_start_hook = btrfs_writepage_start_hook,
7373         .readpage_io_failed_hook = btrfs_io_failed_hook,
7374         .set_bit_hook = btrfs_set_bit_hook,
7375         .clear_bit_hook = btrfs_clear_bit_hook,
7376         .merge_extent_hook = btrfs_merge_extent_hook,
7377         .split_extent_hook = btrfs_split_extent_hook,
7378 };
7379
7380 /*
7381  * btrfs doesn't support the bmap operation because swapfiles
7382  * use bmap to make a mapping of extents in the file.  They assume
7383  * these extents won't change over the life of the file and they
7384  * use the bmap result to do IO directly to the drive.
7385  *
7386  * the btrfs bmap call would return logical addresses that aren't
7387  * suitable for IO and they also will change frequently as COW
7388  * operations happen.  So, swapfile + btrfs == corruption.
7389  *
7390  * For now we're avoiding this by dropping bmap.
7391  */
7392 static const struct address_space_operations btrfs_aops = {
7393         .readpage       = btrfs_readpage,
7394         .writepage      = btrfs_writepage,
7395         .writepages     = btrfs_writepages,
7396         .readpages      = btrfs_readpages,
7397         .direct_IO      = btrfs_direct_IO,
7398         .invalidatepage = btrfs_invalidatepage,
7399         .releasepage    = btrfs_releasepage,
7400         .set_page_dirty = btrfs_set_page_dirty,
7401         .error_remove_page = generic_error_remove_page,
7402 };
7403
7404 static const struct address_space_operations btrfs_symlink_aops = {
7405         .readpage       = btrfs_readpage,
7406         .writepage      = btrfs_writepage,
7407         .invalidatepage = btrfs_invalidatepage,
7408         .releasepage    = btrfs_releasepage,
7409 };
7410
7411 static const struct inode_operations btrfs_file_inode_operations = {
7412         .getattr        = btrfs_getattr,
7413         .setattr        = btrfs_setattr,
7414         .setxattr       = btrfs_setxattr,
7415         .getxattr       = btrfs_getxattr,
7416         .listxattr      = btrfs_listxattr,
7417         .removexattr    = btrfs_removexattr,
7418         .permission     = btrfs_permission,
7419         .fiemap         = btrfs_fiemap,
7420 };
7421 static const struct inode_operations btrfs_special_inode_operations = {
7422         .getattr        = btrfs_getattr,
7423         .setattr        = btrfs_setattr,
7424         .permission     = btrfs_permission,
7425         .setxattr       = btrfs_setxattr,
7426         .getxattr       = btrfs_getxattr,
7427         .listxattr      = btrfs_listxattr,
7428         .removexattr    = btrfs_removexattr,
7429 };
7430 static const struct inode_operations btrfs_symlink_inode_operations = {
7431         .readlink       = generic_readlink,
7432         .follow_link    = page_follow_link_light,
7433         .put_link       = page_put_link,
7434         .getattr        = btrfs_getattr,
7435         .permission     = btrfs_permission,
7436         .setxattr       = btrfs_setxattr,
7437         .getxattr       = btrfs_getxattr,
7438         .listxattr      = btrfs_listxattr,
7439         .removexattr    = btrfs_removexattr,
7440 };
7441
7442 const struct dentry_operations btrfs_dentry_operations = {
7443         .d_delete       = btrfs_dentry_delete,
7444 };