fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 static inline bool is_free_space_inode(struct btrfs_root *root,
 754                                        struct inode *inode)
 755 {
 756         if (root == root->fs_info->tree_root ||
 757             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 758                 return true;
 759         return false;
 760 }
 761
 762 /*
 763  * when extent_io.c finds a delayed allocation range in the file,
 764  * the call backs end up in this code.  The basic idea is to
 765  * allocate extents on disk for the range, and create ordered data structs
 766  * in ram to track those extents.
 767  *
 768  * locked_page is the page that writepage had locked already.  We use
 769  * it to make sure we don't do extra locks or unlocks.
 770  *
 771  * *page_started is set to one if we unlock locked_page and do everything
 772  * required to start IO on it.  It may be clean and already done with
 773  * IO when we return.
 774  */
 775 static noinline int cow_file_range(struct inode *inode,
 776                                    struct page *locked_page,
 777                                    u64 start, u64 end, int *page_started,
 778                                    unsigned long *nr_written,
 779                                    int unlock)
 780 {
 781         struct btrfs_root *root = BTRFS_I(inode)->root;
 782         struct btrfs_trans_handle *trans;
 783         u64 alloc_hint = 0;
 784         u64 num_bytes;
 785         unsigned long ram_size;
 786         u64 disk_num_bytes;
 787         u64 cur_alloc_size;
 788         u64 blocksize = root->sectorsize;
 789         struct btrfs_key ins;
 790         struct extent_map *em;
 791         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 792         int ret = 0;
 793
 794         BUG_ON(is_free_space_inode(root, inode));
 795         trans = btrfs_join_transaction(root);
 796         BUG_ON(IS_ERR(trans));
 797         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 798
 799         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 800         num_bytes = max(blocksize,  num_bytes);
 801         disk_num_bytes = num_bytes;
 802         ret = 0;
 803
 804         /* if this is a small write inside eof, kick off defrag */
 805         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 806                 btrfs_add_inode_defrag(trans, inode);
 807
 808         if (start == 0) {
 809                 /* lets try to make an inline extent */
 810                 ret = cow_file_range_inline(trans, root, inode,
 811                                             start, end, 0, 0, NULL);
 812                 if (ret == 0) {
 813                         extent_clear_unlock_delalloc(inode,
 814                                      &BTRFS_I(inode)->io_tree,
 815                                      start, end, NULL,
 816                                      EXTENT_CLEAR_UNLOCK_PAGE |
 817                                      EXTENT_CLEAR_UNLOCK |
 818                                      EXTENT_CLEAR_DELALLOC |
 819                                      EXTENT_CLEAR_DIRTY |
 820                                      EXTENT_SET_WRITEBACK |
 821                                      EXTENT_END_WRITEBACK);
 822
 823                         *nr_written = *nr_written +
 824                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 825                         *page_started = 1;
 826                         ret = 0;
 827                         goto out;
 828                 }
 829         }
 830
 831         BUG_ON(disk_num_bytes >
 832                btrfs_super_total_bytes(&root->fs_info->super_copy));
 833
 834         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 835         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 836
 837         while (disk_num_bytes > 0) {
 838                 unsigned long op;
 839
 840                 cur_alloc_size = disk_num_bytes;
 841                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 842                                            root->sectorsize, 0, alloc_hint,
 843                                            (u64)-1, &ins, 1);
 844                 BUG_ON(ret);
 845
 846                 em = alloc_extent_map();
 847                 BUG_ON(!em);
 848                 em->start = start;
 849                 em->orig_start = em->start;
 850                 ram_size = ins.offset;
 851                 em->len = ins.offset;
 852
 853                 em->block_start = ins.objectid;
 854                 em->block_len = ins.offset;
 855                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 856                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 857
 858                 while (1) {
 859                         write_lock(&em_tree->lock);
 860                         ret = add_extent_mapping(em_tree, em);
 861                         write_unlock(&em_tree->lock);
 862                         if (ret != -EEXIST) {
 863                                 free_extent_map(em);
 864                                 break;
 865                         }
 866                         btrfs_drop_extent_cache(inode, start,
 867                                                 start + ram_size - 1, 0);
 868                 }
 869
 870                 cur_alloc_size = ins.offset;
 871                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 872                                                ram_size, cur_alloc_size, 0);
 873                 BUG_ON(ret);
 874
 875                 if (root->root_key.objectid ==
 876                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 877                         ret = btrfs_reloc_clone_csums(inode, start,
 878                                                       cur_alloc_size);
 879                         BUG_ON(ret);
 880                 }
 881
 882                 if (disk_num_bytes < cur_alloc_size)
 883                         break;
 884
 885                 /* we're not doing compressed IO, don't unlock the first
 886                  * page (which the caller expects to stay locked), don't
 887                  * clear any dirty bits and don't set any writeback bits
 888                  *
 889                  * Do set the Private2 bit so we know this page was properly
 890                  * setup for writepage
 891                  */
 892                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 893                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 894                         EXTENT_SET_PRIVATE2;
 895
 896                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 897                                              start, start + ram_size - 1,
 898                                              locked_page, op);
 899                 disk_num_bytes -= cur_alloc_size;
 900                 num_bytes -= cur_alloc_size;
 901                 alloc_hint = ins.objectid + ins.offset;
 902                 start += cur_alloc_size;
 903         }
 904 out:
 905         ret = 0;
 906         btrfs_end_transaction(trans, root);
 907
 908         return ret;
 909 }
 910
 911 /*
 912  * work queue call back to started compression on a file and pages
 913  */
 914 static noinline void async_cow_start(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         int num_added = 0;
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         compress_file_range(async_cow->inode, async_cow->locked_page,
 921                             async_cow->start, async_cow->end, async_cow,
 922                             &num_added);
 923         if (num_added == 0)
 924                 async_cow->inode = NULL;
 925 }
 926
 927 /*
 928  * work queue call back to submit previously compressed pages
 929  */
 930 static noinline void async_cow_submit(struct btrfs_work *work)
 931 {
 932         struct async_cow *async_cow;
 933         struct btrfs_root *root;
 934         unsigned long nr_pages;
 935
 936         async_cow = container_of(work, struct async_cow, work);
 937
 938         root = async_cow->root;
 939         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 940                 PAGE_CACHE_SHIFT;
 941
 942         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 943
 944         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 945             5 * 1042 * 1024 &&
 946             waitqueue_active(&root->fs_info->async_submit_wait))
 947                 wake_up(&root->fs_info->async_submit_wait);
 948
 949         if (async_cow->inode)
 950                 submit_compressed_extents(async_cow->inode, async_cow);
 951 }
 952
 953 static noinline void async_cow_free(struct btrfs_work *work)
 954 {
 955         struct async_cow *async_cow;
 956         async_cow = container_of(work, struct async_cow, work);
 957         kfree(async_cow);
 958 }
 959
 960 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 961                                 u64 start, u64 end, int *page_started,
 962                                 unsigned long *nr_written)
 963 {
 964         struct async_cow *async_cow;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         unsigned long nr_pages;
 967         u64 cur_end;
 968         int limit = 10 * 1024 * 1042;
 969
 970         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 971                          1, 0, NULL, GFP_NOFS);
 972         while (start < end) {
 973                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 974                 BUG_ON(!async_cow);
 975                 async_cow->inode = inode;
 976                 async_cow->root = root;
 977                 async_cow->locked_page = locked_page;
 978                 async_cow->start = start;
 979
 980                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 981                         cur_end = end;
 982                 else
 983                         cur_end = min(end, start + 512 * 1024 - 1);
 984
 985                 async_cow->end = cur_end;
 986                 INIT_LIST_HEAD(&async_cow->extents);
 987
 988                 async_cow->work.func = async_cow_start;
 989                 async_cow->work.ordered_func = async_cow_submit;
 990                 async_cow->work.ordered_free = async_cow_free;
 991                 async_cow->work.flags = 0;
 992
 993                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 994                         PAGE_CACHE_SHIFT;
 995                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 996
 997                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 998                                    &async_cow->work);
 999
1000                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1001                         wait_event(root->fs_info->async_submit_wait,
1002                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1003                             limit));
1004                 }
1005
1006                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1007                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1008                         wait_event(root->fs_info->async_submit_wait,
1009                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1010                            0));
1011                 }
1012
1013                 *nr_written += nr_pages;
1014                 start = cur_end + 1;
1015         }
1016         *page_started = 1;
1017         return 0;
1018 }
1019
1020 static noinline int csum_exist_in_range(struct btrfs_root *root,
1021                                         u64 bytenr, u64 num_bytes)
1022 {
1023         int ret;
1024         struct btrfs_ordered_sum *sums;
1025         LIST_HEAD(list);
1026
1027         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1028                                        bytenr + num_bytes - 1, &list, 0);
1029         if (ret == 0 && list_empty(&list))
1030                 return 0;
1031
1032         while (!list_empty(&list)) {
1033                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1034                 list_del(&sums->list);
1035                 kfree(sums);
1036         }
1037         return 1;
1038 }
1039
1040 /*
1041  * when nowcow writeback call back.  This checks for snapshots or COW copies
1042  * of the extents that exist in the file, and COWs the file as required.
1043  *
1044  * If no cow copies or snapshots exist, we write directly to the existing
1045  * blocks on disk
1046  */
1047 static noinline int run_delalloc_nocow(struct inode *inode,
1048                                        struct page *locked_page,
1049                               u64 start, u64 end, int *page_started, int force,
1050                               unsigned long *nr_written)
1051 {
1052         struct btrfs_root *root = BTRFS_I(inode)->root;
1053         struct btrfs_trans_handle *trans;
1054         struct extent_buffer *leaf;
1055         struct btrfs_path *path;
1056         struct btrfs_file_extent_item *fi;
1057         struct btrfs_key found_key;
1058         u64 cow_start;
1059         u64 cur_offset;
1060         u64 extent_end;
1061         u64 extent_offset;
1062         u64 disk_bytenr;
1063         u64 num_bytes;
1064         int extent_type;
1065         int ret;
1066         int type;
1067         int nocow;
1068         int check_prev = 1;
1069         bool nolock;
1070         u64 ino = btrfs_ino(inode);
1071
1072         path = btrfs_alloc_path();
1073         BUG_ON(!path);
1074
1075         nolock = is_free_space_inode(root, inode);
1076
1077         if (nolock)
1078                 trans = btrfs_join_transaction_nolock(root);
1079         else
1080                 trans = btrfs_join_transaction(root);
1081
1082         BUG_ON(IS_ERR(trans));
1083         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1084
1085         cow_start = (u64)-1;
1086         cur_offset = start;
1087         while (1) {
1088                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1089                                                cur_offset, 0);
1090                 BUG_ON(ret < 0);
1091                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1092                         leaf = path->nodes[0];
1093                         btrfs_item_key_to_cpu(leaf, &found_key,
1094                                               path->slots[0] - 1);
1095                         if (found_key.objectid == ino &&
1096                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1097                                 path->slots[0]--;
1098                 }
1099                 check_prev = 0;
1100 next_slot:
1101                 leaf = path->nodes[0];
1102                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1103                         ret = btrfs_next_leaf(root, path);
1104                         if (ret < 0)
1105                                 BUG_ON(1);
1106                         if (ret > 0)
1107                                 break;
1108                         leaf = path->nodes[0];
1109                 }
1110
1111                 nocow = 0;
1112                 disk_bytenr = 0;
1113                 num_bytes = 0;
1114                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1115
1116                 if (found_key.objectid > ino ||
1117                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1118                     found_key.offset > end)
1119                         break;
1120
1121                 if (found_key.offset > cur_offset) {
1122                         extent_end = found_key.offset;
1123                         extent_type = 0;
1124                         goto out_check;
1125                 }
1126
1127                 fi = btrfs_item_ptr(leaf, path->slots[0],
1128                                     struct btrfs_file_extent_item);
1129                 extent_type = btrfs_file_extent_type(leaf, fi);
1130
1131                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1132                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1133                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1134                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1135                         extent_end = found_key.offset +
1136                                 btrfs_file_extent_num_bytes(leaf, fi);
1137                         if (extent_end <= start) {
1138                                 path->slots[0]++;
1139                                 goto next_slot;
1140                         }
1141                         if (disk_bytenr == 0)
1142                                 goto out_check;
1143                         if (btrfs_file_extent_compression(leaf, fi) ||
1144                             btrfs_file_extent_encryption(leaf, fi) ||
1145                             btrfs_file_extent_other_encoding(leaf, fi))
1146                                 goto out_check;
1147                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1148                                 goto out_check;
1149                         if (btrfs_extent_readonly(root, disk_bytenr))
1150                                 goto out_check;
1151                         if (btrfs_cross_ref_exist(trans, root, ino,
1152                                                   found_key.offset -
1153                                                   extent_offset, disk_bytenr))
1154                                 goto out_check;
1155                         disk_bytenr += extent_offset;
1156                         disk_bytenr += cur_offset - found_key.offset;
1157                         num_bytes = min(end + 1, extent_end) - cur_offset;
1158                         /*
1159                          * force cow if csum exists in the range.
1160                          * this ensure that csum for a given extent are
1161                          * either valid or do not exist.
1162                          */
1163                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1164                                 goto out_check;
1165                         nocow = 1;
1166                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1167                         extent_end = found_key.offset +
1168                                 btrfs_file_extent_inline_len(leaf, fi);
1169                         extent_end = ALIGN(extent_end, root->sectorsize);
1170                 } else {
1171                         BUG_ON(1);
1172                 }
1173 out_check:
1174                 if (extent_end <= start) {
1175                         path->slots[0]++;
1176                         goto next_slot;
1177                 }
1178                 if (!nocow) {
1179                         if (cow_start == (u64)-1)
1180                                 cow_start = cur_offset;
1181                         cur_offset = extent_end;
1182                         if (cur_offset > end)
1183                                 break;
1184                         path->slots[0]++;
1185                         goto next_slot;
1186                 }
1187
1188                 btrfs_release_path(path);
1189                 if (cow_start != (u64)-1) {
1190                         ret = cow_file_range(inode, locked_page, cow_start,
1191                                         found_key.offset - 1, page_started,
1192                                         nr_written, 1);
1193                         BUG_ON(ret);
1194                         cow_start = (u64)-1;
1195                 }
1196
1197                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1198                         struct extent_map *em;
1199                         struct extent_map_tree *em_tree;
1200                         em_tree = &BTRFS_I(inode)->extent_tree;
1201                         em = alloc_extent_map();
1202                         BUG_ON(!em);
1203                         em->start = cur_offset;
1204                         em->orig_start = em->start;
1205                         em->len = num_bytes;
1206                         em->block_len = num_bytes;
1207                         em->block_start = disk_bytenr;
1208                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1209                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1210                         while (1) {
1211                                 write_lock(&em_tree->lock);
1212                                 ret = add_extent_mapping(em_tree, em);
1213                                 write_unlock(&em_tree->lock);
1214                                 if (ret != -EEXIST) {
1215                                         free_extent_map(em);
1216                                         break;
1217                                 }
1218                                 btrfs_drop_extent_cache(inode, em->start,
1219                                                 em->start + em->len - 1, 0);
1220                         }
1221                         type = BTRFS_ORDERED_PREALLOC;
1222                 } else {
1223                         type = BTRFS_ORDERED_NOCOW;
1224                 }
1225
1226                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1227                                                num_bytes, num_bytes, type);
1228                 BUG_ON(ret);
1229
1230                 if (root->root_key.objectid ==
1231                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1232                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1233                                                       num_bytes);
1234                         BUG_ON(ret);
1235                 }
1236
1237                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1238                                 cur_offset, cur_offset + num_bytes - 1,
1239                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1240                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1241                                 EXTENT_SET_PRIVATE2);
1242                 cur_offset = extent_end;
1243                 if (cur_offset > end)
1244                         break;
1245         }
1246         btrfs_release_path(path);
1247
1248         if (cur_offset <= end && cow_start == (u64)-1)
1249                 cow_start = cur_offset;
1250         if (cow_start != (u64)-1) {
1251                 ret = cow_file_range(inode, locked_page, cow_start, end,
1252                                      page_started, nr_written, 1);
1253                 BUG_ON(ret);
1254         }
1255
1256         if (nolock) {
1257                 ret = btrfs_end_transaction_nolock(trans, root);
1258                 BUG_ON(ret);
1259         } else {
1260                 ret = btrfs_end_transaction(trans, root);
1261                 BUG_ON(ret);
1262         }
1263         btrfs_free_path(path);
1264         return 0;
1265 }
1266
1267 /*
1268  * extent_io.c call back to do delayed allocation processing
1269  */
1270 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1271                               u64 start, u64 end, int *page_started,
1272                               unsigned long *nr_written)
1273 {
1274         int ret;
1275         struct btrfs_root *root = BTRFS_I(inode)->root;
1276
1277         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1278                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1279                                          page_started, 1, nr_written);
1280         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1281                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1282                                          page_started, 0, nr_written);
1283         else if (!btrfs_test_opt(root, COMPRESS) &&
1284                  !(BTRFS_I(inode)->force_compress) &&
1285                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1286                 ret = cow_file_range(inode, locked_page, start, end,
1287                                       page_started, nr_written, 1);
1288         else
1289                 ret = cow_file_range_async(inode, locked_page, start, end,
1290                                            page_started, nr_written);
1291         return ret;
1292 }
1293
1294 static int btrfs_split_extent_hook(struct inode *inode,
1295                                    struct extent_state *orig, u64 split)
1296 {
1297         /* not delalloc, ignore it */
1298         if (!(orig->state & EXTENT_DELALLOC))
1299                 return 0;
1300
1301         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1302         return 0;
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static int btrfs_merge_extent_hook(struct inode *inode,
1312                                    struct extent_state *new,
1313                                    struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return 0;
1318
1319         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1320         return 0;
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static int btrfs_set_bit_hook(struct inode *inode,
1329                               struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC)
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 else
1345                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1346
1347                 spin_lock(&root->fs_info->delalloc_lock);
1348                 BTRFS_I(inode)->delalloc_bytes += len;
1349                 root->fs_info->delalloc_bytes += len;
1350                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1351                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1352                                       &root->fs_info->delalloc_inodes);
1353                 }
1354                 spin_unlock(&root->fs_info->delalloc_lock);
1355         }
1356         return 0;
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static int btrfs_clear_bit_hook(struct inode *inode,
1363                                 struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC)
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1378                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1379
1380                 if (*bits & EXTENT_DO_ACCOUNTING)
1381                         btrfs_delalloc_release_metadata(inode, len);
1382
1383                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1384                     && do_list)
1385                         btrfs_free_reserved_data_space(inode, len);
1386
1387                 spin_lock(&root->fs_info->delalloc_lock);
1388                 root->fs_info->delalloc_bytes -= len;
1389                 BTRFS_I(inode)->delalloc_bytes -= len;
1390
1391                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1392                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1393                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1394                 }
1395                 spin_unlock(&root->fs_info->delalloc_lock);
1396         }
1397         return 0;
1398 }
1399
1400 /*
1401  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1402  * we don't create bios that span stripes or chunks
1403  */
1404 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1405                          size_t size, struct bio *bio,
1406                          unsigned long bio_flags)
1407 {
1408         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1409         struct btrfs_mapping_tree *map_tree;
1410         u64 logical = (u64)bio->bi_sector << 9;
1411         u64 length = 0;
1412         u64 map_length;
1413         int ret;
1414
1415         if (bio_flags & EXTENT_BIO_COMPRESSED)
1416                 return 0;
1417
1418         length = bio->bi_size;
1419         map_tree = &root->fs_info->mapping_tree;
1420         map_length = length;
1421         ret = btrfs_map_block(map_tree, READ, logical,
1422                               &map_length, NULL, 0);
1423
1424         if (map_length < length + size)
1425                 return 1;
1426         return ret;
1427 }
1428
1429 /*
1430  * in order to insert checksums into the metadata in large chunks,
1431  * we wait until bio submission time.   All the pages in the bio are
1432  * checksummed and sums are attached onto the ordered extent record.
1433  *
1434  * At IO completion time the cums attached on the ordered extent record
1435  * are inserted into the btree
1436  */
1437 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1438                                     struct bio *bio, int mirror_num,
1439                                     unsigned long bio_flags,
1440                                     u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         int ret = 0;
1444
1445         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1446         BUG_ON(ret);
1447         return 0;
1448 }
1449
1450 /*
1451  * in order to insert checksums into the metadata in large chunks,
1452  * we wait until bio submission time.   All the pages in the bio are
1453  * checksummed and sums are attached onto the ordered extent record.
1454  *
1455  * At IO completion time the cums attached on the ordered extent record
1456  * are inserted into the btree
1457  */
1458 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1459                           int mirror_num, unsigned long bio_flags,
1460                           u64 bio_offset)
1461 {
1462         struct btrfs_root *root = BTRFS_I(inode)->root;
1463         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1464 }
1465
1466 /*
1467  * extent_io.c submission hook. This does the right thing for csum calculation
1468  * on write, or reading the csums from the tree before a read
1469  */
1470 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1471                           int mirror_num, unsigned long bio_flags,
1472                           u64 bio_offset)
1473 {
1474         struct btrfs_root *root = BTRFS_I(inode)->root;
1475         int ret = 0;
1476         int skip_sum;
1477
1478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1479
1480         if (is_free_space_inode(root, inode))
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1482         else
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1484         BUG_ON(ret);
1485
1486         if (!(rw & REQ_WRITE)) {
1487                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1488                         return btrfs_submit_compressed_read(inode, bio,
1489                                                     mirror_num, bio_flags);
1490                 } else if (!skip_sum) {
1491                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1492                         if (ret)
1493                                 return ret;
1494                 }
1495                 goto mapit;
1496         } else if (!skip_sum) {
1497                 /* csum items have already been cloned */
1498                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1499                         goto mapit;
1500                 /* we're doing a write, do the async checksumming */
1501                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1502                                    inode, rw, bio, mirror_num,
1503                                    bio_flags, bio_offset,
1504                                    __btrfs_submit_bio_start,
1505                                    __btrfs_submit_bio_done);
1506         }
1507
1508 mapit:
1509         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1510 }
1511
1512 /*
1513  * given a list of ordered sums record them in the inode.  This happens
1514  * at IO completion time based on sums calculated at bio submission time.
1515  */
1516 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1517                              struct inode *inode, u64 file_offset,
1518                              struct list_head *list)
1519 {
1520         struct btrfs_ordered_sum *sum;
1521
1522         list_for_each_entry(sum, list, list) {
1523                 btrfs_csum_file_blocks(trans,
1524                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1525         }
1526         return 0;
1527 }
1528
1529 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1530                               struct extent_state **cached_state)
1531 {
1532         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1533                 WARN_ON(1);
1534         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1535                                    cached_state, GFP_NOFS);
1536 }
1537
1538 /* see btrfs_writepage_start_hook for details on why this is required */
1539 struct btrfs_writepage_fixup {
1540         struct page *page;
1541         struct btrfs_work work;
1542 };
1543
1544 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1545 {
1546         struct btrfs_writepage_fixup *fixup;
1547         struct btrfs_ordered_extent *ordered;
1548         struct extent_state *cached_state = NULL;
1549         struct page *page;
1550         struct inode *inode;
1551         u64 page_start;
1552         u64 page_end;
1553
1554         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1555         page = fixup->page;
1556 again:
1557         lock_page(page);
1558         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1559                 ClearPageChecked(page);
1560                 goto out_page;
1561         }
1562
1563         inode = page->mapping->host;
1564         page_start = page_offset(page);
1565         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1566
1567         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1568                          &cached_state, GFP_NOFS);
1569
1570         /* already ordered? We're done */
1571         if (PagePrivate2(page))
1572                 goto out;
1573
1574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1575         if (ordered) {
1576                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1577                                      page_end, &cached_state, GFP_NOFS);
1578                 unlock_page(page);
1579                 btrfs_start_ordered_extent(inode, ordered, 1);
1580                 goto again;
1581         }
1582
1583         BUG();
1584         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1585         ClearPageChecked(page);
1586 out:
1587         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1588                              &cached_state, GFP_NOFS);
1589 out_page:
1590         unlock_page(page);
1591         page_cache_release(page);
1592         kfree(fixup);
1593 }
1594
1595 /*
1596  * There are a few paths in the higher layers of the kernel that directly
1597  * set the page dirty bit without asking the filesystem if it is a
1598  * good idea.  This causes problems because we want to make sure COW
1599  * properly happens and the data=ordered rules are followed.
1600  *
1601  * In our case any range that doesn't have the ORDERED bit set
1602  * hasn't been properly setup for IO.  We kick off an async process
1603  * to fix it up.  The async helper will wait for ordered extents, set
1604  * the delalloc bit and make it safe to write the page.
1605  */
1606 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1607 {
1608         struct inode *inode = page->mapping->host;
1609         struct btrfs_writepage_fixup *fixup;
1610         struct btrfs_root *root = BTRFS_I(inode)->root;
1611
1612         /* this page is properly in the ordered list */
1613         if (TestClearPagePrivate2(page))
1614                 return 0;
1615
1616         if (PageChecked(page))
1617                 return -EAGAIN;
1618
1619         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1620         if (!fixup)
1621                 return -EAGAIN;
1622
1623         SetPageChecked(page);
1624         page_cache_get(page);
1625         fixup->work.func = btrfs_writepage_fixup_worker;
1626         fixup->page = page;
1627         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1628         return -EAGAIN;
1629 }
1630
1631 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1632                                        struct inode *inode, u64 file_pos,
1633                                        u64 disk_bytenr, u64 disk_num_bytes,
1634                                        u64 num_bytes, u64 ram_bytes,
1635                                        u8 compression, u8 encryption,
1636                                        u16 other_encoding, int extent_type)
1637 {
1638         struct btrfs_root *root = BTRFS_I(inode)->root;
1639         struct btrfs_file_extent_item *fi;
1640         struct btrfs_path *path;
1641         struct extent_buffer *leaf;
1642         struct btrfs_key ins;
1643         u64 hint;
1644         int ret;
1645
1646         path = btrfs_alloc_path();
1647         BUG_ON(!path);
1648
1649         path->leave_spinning = 1;
1650
1651         /*
1652          * we may be replacing one extent in the tree with another.
1653          * The new extent is pinned in the extent map, and we don't want
1654          * to drop it from the cache until it is completely in the btree.
1655          *
1656          * So, tell btrfs_drop_extents to leave this extent in the cache.
1657          * the caller is expected to unpin it and allow it to be merged
1658          * with the others.
1659          */
1660         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1661                                  &hint, 0);
1662         BUG_ON(ret);
1663
1664         ins.objectid = btrfs_ino(inode);
1665         ins.offset = file_pos;
1666         ins.type = BTRFS_EXTENT_DATA_KEY;
1667         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1668         BUG_ON(ret);
1669         leaf = path->nodes[0];
1670         fi = btrfs_item_ptr(leaf, path->slots[0],
1671                             struct btrfs_file_extent_item);
1672         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1673         btrfs_set_file_extent_type(leaf, fi, extent_type);
1674         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1675         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1676         btrfs_set_file_extent_offset(leaf, fi, 0);
1677         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1678         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1679         btrfs_set_file_extent_compression(leaf, fi, compression);
1680         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1681         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1682
1683         btrfs_unlock_up_safe(path, 1);
1684         btrfs_set_lock_blocking(leaf);
1685
1686         btrfs_mark_buffer_dirty(leaf);
1687
1688         inode_add_bytes(inode, num_bytes);
1689
1690         ins.objectid = disk_bytenr;
1691         ins.offset = disk_num_bytes;
1692         ins.type = BTRFS_EXTENT_ITEM_KEY;
1693         ret = btrfs_alloc_reserved_file_extent(trans, root,
1694                                         root->root_key.objectid,
1695                                         btrfs_ino(inode), file_pos, &ins);
1696         BUG_ON(ret);
1697         btrfs_free_path(path);
1698
1699         return 0;
1700 }
1701
1702 /*
1703  * helper function for btrfs_finish_ordered_io, this
1704  * just reads in some of the csum leaves to prime them into ram
1705  * before we start the transaction.  It limits the amount of btree
1706  * reads required while inside the transaction.
1707  */
1708 /* as ordered data IO finishes, this gets called so we can finish
1709  * an ordered extent if the range of bytes in the file it covers are
1710  * fully written.
1711  */
1712 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1713 {
1714         struct btrfs_root *root = BTRFS_I(inode)->root;
1715         struct btrfs_trans_handle *trans = NULL;
1716         struct btrfs_ordered_extent *ordered_extent = NULL;
1717         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1718         struct extent_state *cached_state = NULL;
1719         int compress_type = 0;
1720         int ret;
1721         bool nolock;
1722
1723         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1724                                              end - start + 1);
1725         if (!ret)
1726                 return 0;
1727         BUG_ON(!ordered_extent);
1728
1729         nolock = is_free_space_inode(root, inode);
1730
1731         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1732                 BUG_ON(!list_empty(&ordered_extent->list));
1733                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1734                 if (!ret) {
1735                         if (nolock)
1736                                 trans = btrfs_join_transaction_nolock(root);
1737                         else
1738                                 trans = btrfs_join_transaction(root);
1739                         BUG_ON(IS_ERR(trans));
1740                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741                         ret = btrfs_update_inode(trans, root, inode);
1742                         BUG_ON(ret);
1743                 }
1744                 goto out;
1745         }
1746
1747         lock_extent_bits(io_tree, ordered_extent->file_offset,
1748                          ordered_extent->file_offset + ordered_extent->len - 1,
1749                          0, &cached_state, GFP_NOFS);
1750
1751         if (nolock)
1752                 trans = btrfs_join_transaction_nolock(root);
1753         else
1754                 trans = btrfs_join_transaction(root);
1755         BUG_ON(IS_ERR(trans));
1756         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1757
1758         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1759                 compress_type = ordered_extent->compress_type;
1760         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1761                 BUG_ON(compress_type);
1762                 ret = btrfs_mark_extent_written(trans, inode,
1763                                                 ordered_extent->file_offset,
1764                                                 ordered_extent->file_offset +
1765                                                 ordered_extent->len);
1766                 BUG_ON(ret);
1767         } else {
1768                 BUG_ON(root == root->fs_info->tree_root);
1769                 ret = insert_reserved_file_extent(trans, inode,
1770                                                 ordered_extent->file_offset,
1771                                                 ordered_extent->start,
1772                                                 ordered_extent->disk_len,
1773                                                 ordered_extent->len,
1774                                                 ordered_extent->len,
1775                                                 compress_type, 0, 0,
1776                                                 BTRFS_FILE_EXTENT_REG);
1777                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1778                                    ordered_extent->file_offset,
1779                                    ordered_extent->len);
1780                 BUG_ON(ret);
1781         }
1782         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1783                              ordered_extent->file_offset +
1784                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1785
1786         add_pending_csums(trans, inode, ordered_extent->file_offset,
1787                           &ordered_extent->list);
1788
1789         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1790         if (!ret) {
1791                 ret = btrfs_update_inode(trans, root, inode);
1792                 BUG_ON(ret);
1793         }
1794         ret = 0;
1795 out:
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801                 if (trans)
1802                         btrfs_end_transaction(trans, root);
1803         }
1804
1805         /* once for us */
1806         btrfs_put_ordered_extent(ordered_extent);
1807         /* once for the tree */
1808         btrfs_put_ordered_extent(ordered_extent);
1809
1810         return 0;
1811 }
1812
1813 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1814                                 struct extent_state *state, int uptodate)
1815 {
1816         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1817
1818         ClearPagePrivate2(page);
1819         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1820 }
1821
1822 /*
1823  * When IO fails, either with EIO or csum verification fails, we
1824  * try other mirrors that might have a good copy of the data.  This
1825  * io_failure_record is used to record state as we go through all the
1826  * mirrors.  If another mirror has good data, the page is set up to date
1827  * and things continue.  If a good mirror can't be found, the original
1828  * bio end_io callback is called to indicate things have failed.
1829  */
1830 struct io_failure_record {
1831         struct page *page;
1832         u64 start;
1833         u64 len;
1834         u64 logical;
1835         unsigned long bio_flags;
1836         int last_mirror;
1837 };
1838
1839 static int btrfs_io_failed_hook(struct bio *failed_bio,
1840                          struct page *page, u64 start, u64 end,
1841                          struct extent_state *state)
1842 {
1843         struct io_failure_record *failrec = NULL;
1844         u64 private;
1845         struct extent_map *em;
1846         struct inode *inode = page->mapping->host;
1847         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1848         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1849         struct bio *bio;
1850         int num_copies;
1851         int ret;
1852         int rw;
1853         u64 logical;
1854
1855         ret = get_state_private(failure_tree, start, &private);
1856         if (ret) {
1857                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1858                 if (!failrec)
1859                         return -ENOMEM;
1860                 failrec->start = start;
1861                 failrec->len = end - start + 1;
1862                 failrec->last_mirror = 0;
1863                 failrec->bio_flags = 0;
1864
1865                 read_lock(&em_tree->lock);
1866                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1867                 if (em->start > start || em->start + em->len < start) {
1868                         free_extent_map(em);
1869                         em = NULL;
1870                 }
1871                 read_unlock(&em_tree->lock);
1872
1873                 if (IS_ERR_OR_NULL(em)) {
1874                         kfree(failrec);
1875                         return -EIO;
1876                 }
1877                 logical = start - em->start;
1878                 logical = em->block_start + logical;
1879                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1880                         logical = em->block_start;
1881                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1882                         extent_set_compress_type(&failrec->bio_flags,
1883                                                  em->compress_type);
1884                 }
1885                 failrec->logical = logical;
1886                 free_extent_map(em);
1887                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1888                                 EXTENT_DIRTY, GFP_NOFS);
1889                 set_state_private(failure_tree, start,
1890                                  (u64)(unsigned long)failrec);
1891         } else {
1892                 failrec = (struct io_failure_record *)(unsigned long)private;
1893         }
1894         num_copies = btrfs_num_copies(
1895                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1896                               failrec->logical, failrec->len);
1897         failrec->last_mirror++;
1898         if (!state) {
1899                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1900                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1901                                                     failrec->start,
1902                                                     EXTENT_LOCKED);
1903                 if (state && state->start != failrec->start)
1904                         state = NULL;
1905                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1906         }
1907         if (!state || failrec->last_mirror > num_copies) {
1908                 set_state_private(failure_tree, failrec->start, 0);
1909                 clear_extent_bits(failure_tree, failrec->start,
1910                                   failrec->start + failrec->len - 1,
1911                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1912                 kfree(failrec);
1913                 return -EIO;
1914         }
1915         bio = bio_alloc(GFP_NOFS, 1);
1916         bio->bi_private = state;
1917         bio->bi_end_io = failed_bio->bi_end_io;
1918         bio->bi_sector = failrec->logical >> 9;
1919         bio->bi_bdev = failed_bio->bi_bdev;
1920         bio->bi_size = 0;
1921
1922         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1923         if (failed_bio->bi_rw & REQ_WRITE)
1924                 rw = WRITE;
1925         else
1926                 rw = READ;
1927
1928         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1929                                                       failrec->last_mirror,
1930                                                       failrec->bio_flags, 0);
1931         return ret;
1932 }
1933
1934 /*
1935  * each time an IO finishes, we do a fast check in the IO failure tree
1936  * to see if we need to process or clean up an io_failure_record
1937  */
1938 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1939 {
1940         u64 private;
1941         u64 private_failure;
1942         struct io_failure_record *failure;
1943         int ret;
1944
1945         private = 0;
1946         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1947                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1948                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1949                                         start, &private_failure);
1950                 if (ret == 0) {
1951                         failure = (struct io_failure_record *)(unsigned long)
1952                                    private_failure;
1953                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1954                                           failure->start, 0);
1955                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1956                                           failure->start,
1957                                           failure->start + failure->len - 1,
1958                                           EXTENT_DIRTY | EXTENT_LOCKED,
1959                                           GFP_NOFS);
1960                         kfree(failure);
1961                 }
1962         }
1963         return 0;
1964 }
1965
1966 /*
1967  * when reads are done, we need to check csums to verify the data is correct
1968  * if there's a match, we allow the bio to finish.  If not, we go through
1969  * the io_failure_record routines to find good copies
1970  */
1971 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1972                                struct extent_state *state)
1973 {
1974         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1975         struct inode *inode = page->mapping->host;
1976         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1977         char *kaddr;
1978         u64 private = ~(u32)0;
1979         int ret;
1980         struct btrfs_root *root = BTRFS_I(inode)->root;
1981         u32 csum = ~(u32)0;
1982
1983         if (PageChecked(page)) {
1984                 ClearPageChecked(page);
1985                 goto good;
1986         }
1987
1988         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1989                 goto good;
1990
1991         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1992             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1993                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1994                                   GFP_NOFS);
1995                 return 0;
1996         }
1997
1998         if (state && state->start == start) {
1999                 private = state->private;
2000                 ret = 0;
2001         } else {
2002                 ret = get_state_private(io_tree, start, &private);
2003         }
2004         kaddr = kmap_atomic(page, KM_USER0);
2005         if (ret)
2006                 goto zeroit;
2007
2008         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2009         btrfs_csum_final(csum, (char *)&csum);
2010         if (csum != private)
2011                 goto zeroit;
2012
2013         kunmap_atomic(kaddr, KM_USER0);
2014 good:
2015         /* if the io failure tree for this inode is non-empty,
2016          * check to see if we've recovered from a failed IO
2017          */
2018         btrfs_clean_io_failures(inode, start);
2019         return 0;
2020
2021 zeroit:
2022         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2023                        "private %llu\n",
2024                        (unsigned long long)btrfs_ino(page->mapping->host),
2025                        (unsigned long long)start, csum,
2026                        (unsigned long long)private);
2027         memset(kaddr + offset, 1, end - start + 1);
2028         flush_dcache_page(page);
2029         kunmap_atomic(kaddr, KM_USER0);
2030         if (private == 0)
2031                 return 0;
2032         return -EIO;
2033 }
2034
2035 struct delayed_iput {
2036         struct list_head list;
2037         struct inode *inode;
2038 };
2039
2040 void btrfs_add_delayed_iput(struct inode *inode)
2041 {
2042         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2043         struct delayed_iput *delayed;
2044
2045         if (atomic_add_unless(&inode->i_count, -1, 1))
2046                 return;
2047
2048         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2049         delayed->inode = inode;
2050
2051         spin_lock(&fs_info->delayed_iput_lock);
2052         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2053         spin_unlock(&fs_info->delayed_iput_lock);
2054 }
2055
2056 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2057 {
2058         LIST_HEAD(list);
2059         struct btrfs_fs_info *fs_info = root->fs_info;
2060         struct delayed_iput *delayed;
2061         int empty;
2062
2063         spin_lock(&fs_info->delayed_iput_lock);
2064         empty = list_empty(&fs_info->delayed_iputs);
2065         spin_unlock(&fs_info->delayed_iput_lock);
2066         if (empty)
2067                 return;
2068
2069         down_read(&root->fs_info->cleanup_work_sem);
2070         spin_lock(&fs_info->delayed_iput_lock);
2071         list_splice_init(&fs_info->delayed_iputs, &list);
2072         spin_unlock(&fs_info->delayed_iput_lock);
2073
2074         while (!list_empty(&list)) {
2075                 delayed = list_entry(list.next, struct delayed_iput, list);
2076                 list_del(&delayed->list);
2077                 iput(delayed->inode);
2078                 kfree(delayed);
2079         }
2080         up_read(&root->fs_info->cleanup_work_sem);
2081 }
2082
2083 /*
2084  * calculate extra metadata reservation when snapshotting a subvolume
2085  * contains orphan files.
2086  */
2087 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2088                                 struct btrfs_pending_snapshot *pending,
2089                                 u64 *bytes_to_reserve)
2090 {
2091         struct btrfs_root *root;
2092         struct btrfs_block_rsv *block_rsv;
2093         u64 num_bytes;
2094         int index;
2095
2096         root = pending->root;
2097         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2098                 return;
2099
2100         block_rsv = root->orphan_block_rsv;
2101
2102         /* orphan block reservation for the snapshot */
2103         num_bytes = block_rsv->size;
2104
2105         /*
2106          * after the snapshot is created, COWing tree blocks may use more
2107          * space than it frees. So we should make sure there is enough
2108          * reserved space.
2109          */
2110         index = trans->transid & 0x1;
2111         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2112                 num_bytes += block_rsv->size -
2113                              (block_rsv->reserved + block_rsv->freed[index]);
2114         }
2115
2116         *bytes_to_reserve += num_bytes;
2117 }
2118
2119 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2120                                 struct btrfs_pending_snapshot *pending)
2121 {
2122         struct btrfs_root *root = pending->root;
2123         struct btrfs_root *snap = pending->snap;
2124         struct btrfs_block_rsv *block_rsv;
2125         u64 num_bytes;
2126         int index;
2127         int ret;
2128
2129         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2130                 return;
2131
2132         /* refill source subvolume's orphan block reservation */
2133         block_rsv = root->orphan_block_rsv;
2134         index = trans->transid & 0x1;
2135         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2136                 num_bytes = block_rsv->size -
2137                             (block_rsv->reserved + block_rsv->freed[index]);
2138                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2139                                               root->orphan_block_rsv,
2140                                               num_bytes);
2141                 BUG_ON(ret);
2142         }
2143
2144         /* setup orphan block reservation for the snapshot */
2145         block_rsv = btrfs_alloc_block_rsv(snap);
2146         BUG_ON(!block_rsv);
2147
2148         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2149         snap->orphan_block_rsv = block_rsv;
2150
2151         num_bytes = root->orphan_block_rsv->size;
2152         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2153                                       block_rsv, num_bytes);
2154         BUG_ON(ret);
2155
2156 #if 0
2157         /* insert orphan item for the snapshot */
2158         WARN_ON(!root->orphan_item_inserted);
2159         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2160                                        snap->root_key.objectid);
2161         BUG_ON(ret);
2162         snap->orphan_item_inserted = 1;
2163 #endif
2164 }
2165
2166 enum btrfs_orphan_cleanup_state {
2167         ORPHAN_CLEANUP_STARTED  = 1,
2168         ORPHAN_CLEANUP_DONE     = 2,
2169 };
2170
2171 /*
2172  * This is called in transaction commmit time. If there are no orphan
2173  * files in the subvolume, it removes orphan item and frees block_rsv
2174  * structure.
2175  */
2176 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2177                               struct btrfs_root *root)
2178 {
2179         int ret;
2180
2181         if (!list_empty(&root->orphan_list) ||
2182             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2183                 return;
2184
2185         if (root->orphan_item_inserted &&
2186             btrfs_root_refs(&root->root_item) > 0) {
2187                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2188                                             root->root_key.objectid);
2189                 BUG_ON(ret);
2190                 root->orphan_item_inserted = 0;
2191         }
2192
2193         if (root->orphan_block_rsv) {
2194                 WARN_ON(root->orphan_block_rsv->size > 0);
2195                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2196                 root->orphan_block_rsv = NULL;
2197         }
2198 }
2199
2200 /*
2201  * This creates an orphan entry for the given inode in case something goes
2202  * wrong in the middle of an unlink/truncate.
2203  *
2204  * NOTE: caller of this function should reserve 5 units of metadata for
2205  *       this function.
2206  */
2207 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2208 {
2209         struct btrfs_root *root = BTRFS_I(inode)->root;
2210         struct btrfs_block_rsv *block_rsv = NULL;
2211         int reserve = 0;
2212         int insert = 0;
2213         int ret;
2214
2215         if (!root->orphan_block_rsv) {
2216                 block_rsv = btrfs_alloc_block_rsv(root);
2217                 BUG_ON(!block_rsv);
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
2513         path = btrfs_alloc_path();
2514         BUG_ON(!path);
2515         path->leave_spinning = 1;
2516         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2517
2518         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2519         if (ret)
2520                 goto make_bad;
2521
2522         leaf = path->nodes[0];
2523         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2524                                     struct btrfs_inode_item);
2525         if (!leaf->map_token)
2526                 map_private_extent_buffer(leaf, (unsigned long)inode_item,
2527                                           sizeof(struct btrfs_inode_item),
2528                                           &leaf->map_token, &leaf->kaddr,
2529                                           &leaf->map_start, &leaf->map_len,
2530                                           KM_USER1);
2531
2532         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2533         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2534         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2535         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2536         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2537
2538         tspec = btrfs_inode_atime(inode_item);
2539         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2540         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2541
2542         tspec = btrfs_inode_mtime(inode_item);
2543         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2544         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2545
2546         tspec = btrfs_inode_ctime(inode_item);
2547         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2548         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2549
2550         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2551         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2552         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2553         inode->i_generation = BTRFS_I(inode)->generation;
2554         inode->i_rdev = 0;
2555         rdev = btrfs_inode_rdev(leaf, inode_item);
2556
2557         BTRFS_I(inode)->index_cnt = (u64)-1;
2558         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2559
2560         /*
2561          * try to precache a NULL acl entry for files that don't have
2562          * any xattrs or acls
2563          */
2564         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2565                                            btrfs_ino(inode));
2566         if (!maybe_acls)
2567                 cache_no_acl(inode);
2568
2569         if (leaf->map_token) {
2570                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2571                 leaf->map_token = NULL;
2572         }
2573
2574         btrfs_free_path(path);
2575         inode_item = NULL;
2576
2577         switch (inode->i_mode & S_IFMT) {
2578         case S_IFREG:
2579                 inode->i_mapping->a_ops = &btrfs_aops;
2580                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2581                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2582                 inode->i_fop = &btrfs_file_operations;
2583                 inode->i_op = &btrfs_file_inode_operations;
2584                 break;
2585         case S_IFDIR:
2586                 inode->i_fop = &btrfs_dir_file_operations;
2587                 if (root == root->fs_info->tree_root)
2588                         inode->i_op = &btrfs_dir_ro_inode_operations;
2589                 else
2590                         inode->i_op = &btrfs_dir_inode_operations;
2591                 break;
2592         case S_IFLNK:
2593                 inode->i_op = &btrfs_symlink_inode_operations;
2594                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2595                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2596                 break;
2597         default:
2598                 inode->i_op = &btrfs_special_inode_operations;
2599                 init_special_inode(inode, inode->i_mode, rdev);
2600                 break;
2601         }
2602
2603         btrfs_update_iflags(inode);
2604         return;
2605
2606 make_bad:
2607         btrfs_free_path(path);
2608         make_bad_inode(inode);
2609 }
2610
2611 /*
2612  * given a leaf and an inode, copy the inode fields into the leaf
2613  */
2614 static void fill_inode_item(struct btrfs_trans_handle *trans,
2615                             struct extent_buffer *leaf,
2616                             struct btrfs_inode_item *item,
2617                             struct inode *inode)
2618 {
2619         if (!leaf->map_token)
2620                 map_private_extent_buffer(leaf, (unsigned long)item,
2621                                           sizeof(struct btrfs_inode_item),
2622                                           &leaf->map_token, &leaf->kaddr,
2623                                           &leaf->map_start, &leaf->map_len,
2624                                           KM_USER1);
2625
2626         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2627         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2628         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2629         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2630         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2631
2632         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2633                                inode->i_atime.tv_sec);
2634         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2635                                 inode->i_atime.tv_nsec);
2636
2637         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2638                                inode->i_mtime.tv_sec);
2639         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2640                                 inode->i_mtime.tv_nsec);
2641
2642         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2643                                inode->i_ctime.tv_sec);
2644         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2645                                 inode->i_ctime.tv_nsec);
2646
2647         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2648         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2649         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2650         btrfs_set_inode_transid(leaf, item, trans->transid);
2651         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2652         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2653         btrfs_set_inode_block_group(leaf, item, 0);
2654
2655         if (leaf->map_token) {
2656                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2657                 leaf->map_token = NULL;
2658         }
2659 }
2660
2661 /*
2662  * copy everything in the in-memory inode into the btree.
2663  */
2664 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2665                                 struct btrfs_root *root, struct inode *inode)
2666 {
2667         struct btrfs_inode_item *inode_item;
2668         struct btrfs_path *path;
2669         struct extent_buffer *leaf;
2670         int ret;
2671
2672         /*
2673          * If root is tree root, it means this inode is used to
2674          * store free space information. And these inodes are updated
2675          * when committing the transaction, so they needn't delaye to
2676          * be updated, or deadlock will occured.
2677          */
2678         if (!is_free_space_inode(root, inode)) {
2679                 ret = btrfs_delayed_update_inode(trans, root, inode);
2680                 if (!ret)
2681                         btrfs_set_inode_last_trans(trans, inode);
2682                 return ret;
2683         }
2684
2685         path = btrfs_alloc_path();
2686         if (!path)
2687                 return -ENOMEM;
2688
2689         path->leave_spinning = 1;
2690         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2691                                  1);
2692         if (ret) {
2693                 if (ret > 0)
2694                         ret = -ENOENT;
2695                 goto failed;
2696         }
2697
2698         btrfs_unlock_up_safe(path, 1);
2699         leaf = path->nodes[0];
2700         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2701                                     struct btrfs_inode_item);
2702
2703         fill_inode_item(trans, leaf, inode_item, inode);
2704         btrfs_mark_buffer_dirty(leaf);
2705         btrfs_set_inode_last_trans(trans, inode);
2706         ret = 0;
2707 failed:
2708         btrfs_free_path(path);
2709         return ret;
2710 }
2711
2712 /*
2713  * unlink helper that gets used here in inode.c and in the tree logging
2714  * recovery code.  It remove a link in a directory with a given name, and
2715  * also drops the back refs in the inode to the directory
2716  */
2717 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2718                                 struct btrfs_root *root,
2719                                 struct inode *dir, struct inode *inode,
2720                                 const char *name, int name_len)
2721 {
2722         struct btrfs_path *path;
2723         int ret = 0;
2724         struct extent_buffer *leaf;
2725         struct btrfs_dir_item *di;
2726         struct btrfs_key key;
2727         u64 index;
2728         u64 ino = btrfs_ino(inode);
2729         u64 dir_ino = btrfs_ino(dir);
2730
2731         path = btrfs_alloc_path();
2732         if (!path) {
2733                 ret = -ENOMEM;
2734                 goto out;
2735         }
2736
2737         path->leave_spinning = 1;
2738         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2739                                     name, name_len, -1);
2740         if (IS_ERR(di)) {
2741                 ret = PTR_ERR(di);
2742                 goto err;
2743         }
2744         if (!di) {
2745                 ret = -ENOENT;
2746                 goto err;
2747         }
2748         leaf = path->nodes[0];
2749         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2750         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2751         if (ret)
2752                 goto err;
2753         btrfs_release_path(path);
2754
2755         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2756                                   dir_ino, &index);
2757         if (ret) {
2758                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2759                        "inode %llu parent %llu\n", name_len, name,
2760                        (unsigned long long)ino, (unsigned long long)dir_ino);
2761                 goto err;
2762         }
2763
2764         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2765         if (ret)
2766                 goto err;
2767
2768         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2769                                          inode, dir_ino);
2770         BUG_ON(ret != 0 && ret != -ENOENT);
2771
2772         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2773                                            dir, index);
2774         if (ret == -ENOENT)
2775                 ret = 0;
2776 err:
2777         btrfs_free_path(path);
2778         if (ret)
2779                 goto out;
2780
2781         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2782         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2783         btrfs_update_inode(trans, root, dir);
2784 out:
2785         return ret;
2786 }
2787
2788 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2789                        struct btrfs_root *root,
2790                        struct inode *dir, struct inode *inode,
2791                        const char *name, int name_len)
2792 {
2793         int ret;
2794         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2795         if (!ret) {
2796                 btrfs_drop_nlink(inode);
2797                 ret = btrfs_update_inode(trans, root, inode);
2798         }
2799         return ret;
2800 }
2801
2802
2803 /* helper to check if there is any shared block in the path */
2804 static int check_path_shared(struct btrfs_root *root,
2805                              struct btrfs_path *path)
2806 {
2807         struct extent_buffer *eb;
2808         int level;
2809         u64 refs = 1;
2810
2811         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2812                 int ret;
2813
2814                 if (!path->nodes[level])
2815                         break;
2816                 eb = path->nodes[level];
2817                 if (!btrfs_block_can_be_shared(root, eb))
2818                         continue;
2819                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2820                                                &refs, NULL);
2821                 if (refs > 1)
2822                         return 1;
2823         }
2824         return 0;
2825 }
2826
2827 /*
2828  * helper to start transaction for unlink and rmdir.
2829  *
2830  * unlink and rmdir are special in btrfs, they do not always free space.
2831  * so in enospc case, we should make sure they will free space before
2832  * allowing them to use the global metadata reservation.
2833  */
2834 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2835                                                        struct dentry *dentry)
2836 {
2837         struct btrfs_trans_handle *trans;
2838         struct btrfs_root *root = BTRFS_I(dir)->root;
2839         struct btrfs_path *path;
2840         struct btrfs_inode_ref *ref;
2841         struct btrfs_dir_item *di;
2842         struct inode *inode = dentry->d_inode;
2843         u64 index;
2844         int check_link = 1;
2845         int err = -ENOSPC;
2846         int ret;
2847         u64 ino = btrfs_ino(inode);
2848         u64 dir_ino = btrfs_ino(dir);
2849
2850         trans = btrfs_start_transaction(root, 10);
2851         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2852                 return trans;
2853
2854         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2855                 return ERR_PTR(-ENOSPC);
2856
2857         /* check if there is someone else holds reference */
2858         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2859                 return ERR_PTR(-ENOSPC);
2860
2861         if (atomic_read(&inode->i_count) > 2)
2862                 return ERR_PTR(-ENOSPC);
2863
2864         if (xchg(&root->fs_info->enospc_unlink, 1))
2865                 return ERR_PTR(-ENOSPC);
2866
2867         path = btrfs_alloc_path();
2868         if (!path) {
2869                 root->fs_info->enospc_unlink = 0;
2870                 return ERR_PTR(-ENOMEM);
2871         }
2872
2873         trans = btrfs_start_transaction(root, 0);
2874         if (IS_ERR(trans)) {
2875                 btrfs_free_path(path);
2876                 root->fs_info->enospc_unlink = 0;
2877                 return trans;
2878         }
2879
2880         path->skip_locking = 1;
2881         path->search_commit_root = 1;
2882
2883         ret = btrfs_lookup_inode(trans, root, path,
2884                                 &BTRFS_I(dir)->location, 0);
2885         if (ret < 0) {
2886                 err = ret;
2887                 goto out;
2888         }
2889         if (ret == 0) {
2890                 if (check_path_shared(root, path))
2891                         goto out;
2892         } else {
2893                 check_link = 0;
2894         }
2895         btrfs_release_path(path);
2896
2897         ret = btrfs_lookup_inode(trans, root, path,
2898                                 &BTRFS_I(inode)->location, 0);
2899         if (ret < 0) {
2900                 err = ret;
2901                 goto out;
2902         }
2903         if (ret == 0) {
2904                 if (check_path_shared(root, path))
2905                         goto out;
2906         } else {
2907                 check_link = 0;
2908         }
2909         btrfs_release_path(path);
2910
2911         if (ret == 0 && S_ISREG(inode->i_mode)) {
2912                 ret = btrfs_lookup_file_extent(trans, root, path,
2913                                                ino, (u64)-1, 0);
2914                 if (ret < 0) {
2915                         err = ret;
2916                         goto out;
2917                 }
2918                 BUG_ON(ret == 0);
2919                 if (check_path_shared(root, path))
2920                         goto out;
2921                 btrfs_release_path(path);
2922         }
2923
2924         if (!check_link) {
2925                 err = 0;
2926                 goto out;
2927         }
2928
2929         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2930                                 dentry->d_name.name, dentry->d_name.len, 0);
2931         if (IS_ERR(di)) {
2932                 err = PTR_ERR(di);
2933                 goto out;
2934         }
2935         if (di) {
2936                 if (check_path_shared(root, path))
2937                         goto out;
2938         } else {
2939                 err = 0;
2940                 goto out;
2941         }
2942         btrfs_release_path(path);
2943
2944         ref = btrfs_lookup_inode_ref(trans, root, path,
2945                                 dentry->d_name.name, dentry->d_name.len,
2946                                 ino, dir_ino, 0);
2947         if (IS_ERR(ref)) {
2948                 err = PTR_ERR(ref);
2949                 goto out;
2950         }
2951         BUG_ON(!ref);
2952         if (check_path_shared(root, path))
2953                 goto out;
2954         index = btrfs_inode_ref_index(path->nodes[0], ref);
2955         btrfs_release_path(path);
2956
2957         /*
2958          * This is a commit root search, if we can lookup inode item and other
2959          * relative items in the commit root, it means the transaction of
2960          * dir/file creation has been committed, and the dir index item that we
2961          * delay to insert has also been inserted into the commit root. So
2962          * we needn't worry about the delayed insertion of the dir index item
2963          * here.
2964          */
2965         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2966                                 dentry->d_name.name, dentry->d_name.len, 0);
2967         if (IS_ERR(di)) {
2968                 err = PTR_ERR(di);
2969                 goto out;
2970         }
2971         BUG_ON(ret == -ENOENT);
2972         if (check_path_shared(root, path))
2973                 goto out;
2974
2975         err = 0;
2976 out:
2977         btrfs_free_path(path);
2978         if (err) {
2979                 btrfs_end_transaction(trans, root);
2980                 root->fs_info->enospc_unlink = 0;
2981                 return ERR_PTR(err);
2982         }
2983
2984         trans->block_rsv = &root->fs_info->global_block_rsv;
2985         return trans;
2986 }
2987
2988 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2989                                struct btrfs_root *root)
2990 {
2991         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2992                 BUG_ON(!root->fs_info->enospc_unlink);
2993                 root->fs_info->enospc_unlink = 0;
2994         }
2995         btrfs_end_transaction_throttle(trans, root);
2996 }
2997
2998 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2999 {
3000         struct btrfs_root *root = BTRFS_I(dir)->root;
3001         struct btrfs_trans_handle *trans;
3002         struct inode *inode = dentry->d_inode;
3003         int ret;
3004         unsigned long nr = 0;
3005
3006         trans = __unlink_start_trans(dir, dentry);
3007         if (IS_ERR(trans))
3008                 return PTR_ERR(trans);
3009
3010         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3011
3012         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3013                                  dentry->d_name.name, dentry->d_name.len);
3014         BUG_ON(ret);
3015
3016         if (inode->i_nlink == 0) {
3017                 ret = btrfs_orphan_add(trans, inode);
3018                 BUG_ON(ret);
3019         }
3020
3021         nr = trans->blocks_used;
3022         __unlink_end_trans(trans, root);
3023         btrfs_btree_balance_dirty(root, nr);
3024         return ret;
3025 }
3026
3027 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3028                         struct btrfs_root *root,
3029                         struct inode *dir, u64 objectid,
3030                         const char *name, int name_len)
3031 {
3032         struct btrfs_path *path;
3033         struct extent_buffer *leaf;
3034         struct btrfs_dir_item *di;
3035         struct btrfs_key key;
3036         u64 index;
3037         int ret;
3038         u64 dir_ino = btrfs_ino(dir);
3039
3040         path = btrfs_alloc_path();
3041         if (!path)
3042                 return -ENOMEM;
3043
3044         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3045                                    name, name_len, -1);
3046         BUG_ON(IS_ERR_OR_NULL(di));
3047
3048         leaf = path->nodes[0];
3049         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3050         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3051         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3052         BUG_ON(ret);
3053         btrfs_release_path(path);
3054
3055         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3056                                  objectid, root->root_key.objectid,
3057                                  dir_ino, &index, name, name_len);
3058         if (ret < 0) {
3059                 BUG_ON(ret != -ENOENT);
3060                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3061                                                  name, name_len);
3062                 BUG_ON(IS_ERR_OR_NULL(di));
3063
3064                 leaf = path->nodes[0];
3065                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3066                 btrfs_release_path(path);
3067                 index = key.offset;
3068         }
3069         btrfs_release_path(path);
3070
3071         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3072         BUG_ON(ret);
3073
3074         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3075         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3076         ret = btrfs_update_inode(trans, root, dir);
3077         BUG_ON(ret);
3078
3079         return 0;
3080 }
3081
3082 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3083 {
3084         struct inode *inode = dentry->d_inode;
3085         int err = 0;
3086         struct btrfs_root *root = BTRFS_I(dir)->root;
3087         struct btrfs_trans_handle *trans;
3088         unsigned long nr = 0;
3089
3090         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3091             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3092                 return -ENOTEMPTY;
3093
3094         trans = __unlink_start_trans(dir, dentry);
3095         if (IS_ERR(trans))
3096                 return PTR_ERR(trans);
3097
3098         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3099                 err = btrfs_unlink_subvol(trans, root, dir,
3100                                           BTRFS_I(inode)->location.objectid,
3101                                           dentry->d_name.name,
3102                                           dentry->d_name.len);
3103                 goto out;
3104         }
3105
3106         err = btrfs_orphan_add(trans, inode);
3107         if (err)
3108                 goto out;
3109
3110         /* now the directory is empty */
3111         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3112                                  dentry->d_name.name, dentry->d_name.len);
3113         if (!err)
3114                 btrfs_i_size_write(inode, 0);
3115 out:
3116         nr = trans->blocks_used;
3117         __unlink_end_trans(trans, root);
3118         btrfs_btree_balance_dirty(root, nr);
3119
3120         return err;
3121 }
3122
3123 /*
3124  * this can truncate away extent items, csum items and directory items.
3125  * It starts at a high offset and removes keys until it can't find
3126  * any higher than new_size
3127  *
3128  * csum items that cross the new i_size are truncated to the new size
3129  * as well.
3130  *
3131  * min_type is the minimum key type to truncate down to.  If set to 0, this
3132  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3133  */
3134 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3135                                struct btrfs_root *root,
3136                                struct inode *inode,
3137                                u64 new_size, u32 min_type)
3138 {
3139         struct btrfs_path *path;
3140         struct extent_buffer *leaf;
3141         struct btrfs_file_extent_item *fi;
3142         struct btrfs_key key;
3143         struct btrfs_key found_key;
3144         u64 extent_start = 0;
3145         u64 extent_num_bytes = 0;
3146         u64 extent_offset = 0;
3147         u64 item_end = 0;
3148         u64 mask = root->sectorsize - 1;
3149         u32 found_type = (u8)-1;
3150         int found_extent;
3151         int del_item;
3152         int pending_del_nr = 0;
3153         int pending_del_slot = 0;
3154         int extent_type = -1;
3155         int encoding;
3156         int ret;
3157         int err = 0;
3158         u64 ino = btrfs_ino(inode);
3159
3160         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3161
3162         if (root->ref_cows || root == root->fs_info->tree_root)
3163                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3164
3165         /*
3166          * This function is also used to drop the items in the log tree before
3167          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3168          * it is used to drop the loged items. So we shouldn't kill the delayed
3169          * items.
3170          */
3171         if (min_type == 0 && root == BTRFS_I(inode)->root)
3172                 btrfs_kill_delayed_inode_items(inode);
3173
3174         path = btrfs_alloc_path();
3175         BUG_ON(!path);
3176         path->reada = -1;
3177
3178         key.objectid = ino;
3179         key.offset = (u64)-1;
3180         key.type = (u8)-1;
3181
3182 search_again:
3183         path->leave_spinning = 1;
3184         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3185         if (ret < 0) {
3186                 err = ret;
3187                 goto out;
3188         }
3189
3190         if (ret > 0) {
3191                 /* there are no items in the tree for us to truncate, we're
3192                  * done
3193                  */
3194                 if (path->slots[0] == 0)
3195                         goto out;
3196                 path->slots[0]--;
3197         }
3198
3199         while (1) {
3200                 fi = NULL;
3201                 leaf = path->nodes[0];
3202                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3203                 found_type = btrfs_key_type(&found_key);
3204                 encoding = 0;
3205
3206                 if (found_key.objectid != ino)
3207                         break;
3208
3209                 if (found_type < min_type)
3210                         break;
3211
3212                 item_end = found_key.offset;
3213                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3214                         fi = btrfs_item_ptr(leaf, path->slots[0],
3215                                             struct btrfs_file_extent_item);
3216                         extent_type = btrfs_file_extent_type(leaf, fi);
3217                         encoding = btrfs_file_extent_compression(leaf, fi);
3218                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3219                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3220
3221                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3222                                 item_end +=
3223                                     btrfs_file_extent_num_bytes(leaf, fi);
3224                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3225                                 item_end += btrfs_file_extent_inline_len(leaf,
3226                                                                          fi);
3227                         }
3228                         item_end--;
3229                 }
3230                 if (found_type > min_type) {
3231                         del_item = 1;
3232                 } else {
3233                         if (item_end < new_size)
3234                                 break;
3235                         if (found_key.offset >= new_size)
3236                                 del_item = 1;
3237                         else
3238                                 del_item = 0;
3239                 }
3240                 found_extent = 0;
3241                 /* FIXME, shrink the extent if the ref count is only 1 */
3242                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3243                         goto delete;
3244
3245                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3246                         u64 num_dec;
3247                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3248                         if (!del_item && !encoding) {
3249                                 u64 orig_num_bytes =
3250                                         btrfs_file_extent_num_bytes(leaf, fi);
3251                                 extent_num_bytes = new_size -
3252                                         found_key.offset + root->sectorsize - 1;
3253                                 extent_num_bytes = extent_num_bytes &
3254                                         ~((u64)root->sectorsize - 1);
3255                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3256                                                          extent_num_bytes);
3257                                 num_dec = (orig_num_bytes -
3258                                            extent_num_bytes);
3259                                 if (root->ref_cows && extent_start != 0)
3260                                         inode_sub_bytes(inode, num_dec);
3261                                 btrfs_mark_buffer_dirty(leaf);
3262                         } else {
3263                                 extent_num_bytes =
3264                                         btrfs_file_extent_disk_num_bytes(leaf,
3265                                                                          fi);
3266                                 extent_offset = found_key.offset -
3267                                         btrfs_file_extent_offset(leaf, fi);
3268
3269                                 /* FIXME blocksize != 4096 */
3270                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3271                                 if (extent_start != 0) {
3272                                         found_extent = 1;
3273                                         if (root->ref_cows)
3274                                                 inode_sub_bytes(inode, num_dec);
3275                                 }
3276                         }
3277                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3278                         /*
3279                          * we can't truncate inline items that have had
3280                          * special encodings
3281                          */
3282                         if (!del_item &&
3283                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3284                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3285                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3286                                 u32 size = new_size - found_key.offset;
3287
3288                                 if (root->ref_cows) {
3289                                         inode_sub_bytes(inode, item_end + 1 -
3290                                                         new_size);
3291                                 }
3292                                 size =
3293                                     btrfs_file_extent_calc_inline_size(size);
3294                                 ret = btrfs_truncate_item(trans, root, path,
3295                                                           size, 1);
3296                         } else if (root->ref_cows) {
3297                                 inode_sub_bytes(inode, item_end + 1 -
3298                                                 found_key.offset);
3299                         }
3300                 }
3301 delete:
3302                 if (del_item) {
3303                         if (!pending_del_nr) {
3304                                 /* no pending yet, add ourselves */
3305                                 pending_del_slot = path->slots[0];
3306                                 pending_del_nr = 1;
3307                         } else if (pending_del_nr &&
3308                                    path->slots[0] + 1 == pending_del_slot) {
3309                                 /* hop on the pending chunk */
3310                                 pending_del_nr++;
3311                                 pending_del_slot = path->slots[0];
3312                         } else {
3313                                 BUG();
3314                         }
3315                 } else {
3316                         break;
3317                 }
3318                 if (found_extent && (root->ref_cows ||
3319                                      root == root->fs_info->tree_root)) {
3320                         btrfs_set_path_blocking(path);
3321                         ret = btrfs_free_extent(trans, root, extent_start,
3322                                                 extent_num_bytes, 0,
3323                                                 btrfs_header_owner(leaf),
3324                                                 ino, extent_offset);
3325                         BUG_ON(ret);
3326                 }
3327
3328                 if (found_type == BTRFS_INODE_ITEM_KEY)
3329                         break;
3330
3331                 if (path->slots[0] == 0 ||
3332                     path->slots[0] != pending_del_slot) {
3333                         if (root->ref_cows &&
3334                             BTRFS_I(inode)->location.objectid !=
3335                                                 BTRFS_FREE_INO_OBJECTID) {
3336                                 err = -EAGAIN;
3337                                 goto out;
3338                         }
3339                         if (pending_del_nr) {
3340                                 ret = btrfs_del_items(trans, root, path,
3341                                                 pending_del_slot,
3342                                                 pending_del_nr);
3343                                 BUG_ON(ret);
3344                                 pending_del_nr = 0;
3345                         }
3346                         btrfs_release_path(path);
3347                         goto search_again;
3348                 } else {
3349                         path->slots[0]--;
3350                 }
3351         }
3352 out:
3353         if (pending_del_nr) {
3354                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3355                                       pending_del_nr);
3356                 BUG_ON(ret);
3357         }
3358         btrfs_free_path(path);
3359         return err;
3360 }
3361
3362 /*
3363  * taken from block_truncate_page, but does cow as it zeros out
3364  * any bytes left in the last page in the file.
3365  */
3366 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3367 {
3368         struct inode *inode = mapping->host;
3369         struct btrfs_root *root = BTRFS_I(inode)->root;
3370         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3371         struct btrfs_ordered_extent *ordered;
3372         struct extent_state *cached_state = NULL;
3373         char *kaddr;
3374         u32 blocksize = root->sectorsize;
3375         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3376         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3377         struct page *page;
3378         int ret = 0;
3379         u64 page_start;
3380         u64 page_end;
3381
3382         if ((offset & (blocksize - 1)) == 0)
3383                 goto out;
3384         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3385         if (ret)
3386                 goto out;
3387
3388         ret = -ENOMEM;
3389 again:
3390         page = grab_cache_page(mapping, index);
3391         if (!page) {
3392                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3393                 goto out;
3394         }
3395
3396         page_start = page_offset(page);
3397         page_end = page_start + PAGE_CACHE_SIZE - 1;
3398
3399         if (!PageUptodate(page)) {
3400                 ret = btrfs_readpage(NULL, page);
3401                 lock_page(page);
3402                 if (page->mapping != mapping) {
3403                         unlock_page(page);
3404                         page_cache_release(page);
3405                         goto again;
3406                 }
3407                 if (!PageUptodate(page)) {
3408                         ret = -EIO;
3409                         goto out_unlock;
3410                 }
3411         }
3412         wait_on_page_writeback(page);
3413
3414         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3415                          GFP_NOFS);
3416         set_page_extent_mapped(page);
3417
3418         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3419         if (ordered) {
3420                 unlock_extent_cached(io_tree, page_start, page_end,
3421                                      &cached_state, GFP_NOFS);
3422                 unlock_page(page);
3423                 page_cache_release(page);
3424                 btrfs_start_ordered_extent(inode, ordered, 1);
3425                 btrfs_put_ordered_extent(ordered);
3426                 goto again;
3427         }
3428
3429         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3430                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3431                           0, 0, &cached_state, GFP_NOFS);
3432
3433         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3434                                         &cached_state);
3435         if (ret) {
3436                 unlock_extent_cached(io_tree, page_start, page_end,
3437                                      &cached_state, GFP_NOFS);
3438                 goto out_unlock;
3439         }
3440
3441         ret = 0;
3442         if (offset != PAGE_CACHE_SIZE) {
3443                 kaddr = kmap(page);
3444                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3445                 flush_dcache_page(page);
3446                 kunmap(page);
3447         }
3448         ClearPageChecked(page);
3449         set_page_dirty(page);
3450         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3451                              GFP_NOFS);
3452
3453 out_unlock:
3454         if (ret)
3455                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3456         unlock_page(page);
3457         page_cache_release(page);
3458 out:
3459         return ret;
3460 }
3461
3462 /*
3463  * This function puts in dummy file extents for the area we're creating a hole
3464  * for.  So if we are truncating this file to a larger size we need to insert
3465  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3466  * the range between oldsize and size
3467  */
3468 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3469 {
3470         struct btrfs_trans_handle *trans;
3471         struct btrfs_root *root = BTRFS_I(inode)->root;
3472         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3473         struct extent_map *em = NULL;
3474         struct extent_state *cached_state = NULL;
3475         u64 mask = root->sectorsize - 1;
3476         u64 hole_start = (oldsize + mask) & ~mask;
3477         u64 block_end = (size + mask) & ~mask;
3478         u64 last_byte;
3479         u64 cur_offset;
3480         u64 hole_size;
3481         int err = 0;
3482
3483         if (size <= hole_start)
3484                 return 0;
3485
3486         while (1) {
3487                 struct btrfs_ordered_extent *ordered;
3488                 btrfs_wait_ordered_range(inode, hole_start,
3489                                          block_end - hole_start);
3490                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3491                                  &cached_state, GFP_NOFS);
3492                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3493                 if (!ordered)
3494                         break;
3495                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3496                                      &cached_state, GFP_NOFS);
3497                 btrfs_put_ordered_extent(ordered);
3498         }
3499
3500         cur_offset = hole_start;
3501         while (1) {
3502                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3503                                 block_end - cur_offset, 0);
3504                 BUG_ON(IS_ERR_OR_NULL(em));
3505                 last_byte = min(extent_map_end(em), block_end);
3506                 last_byte = (last_byte + mask) & ~mask;
3507                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3508                         u64 hint_byte = 0;
3509                         hole_size = last_byte - cur_offset;
3510
3511                         trans = btrfs_start_transaction(root, 2);
3512                         if (IS_ERR(trans)) {
3513                                 err = PTR_ERR(trans);
3514                                 break;
3515                         }
3516
3517                         err = btrfs_drop_extents(trans, inode, cur_offset,
3518                                                  cur_offset + hole_size,
3519                                                  &hint_byte, 1);
3520                         if (err)
3521                                 break;
3522
3523                         err = btrfs_insert_file_extent(trans, root,
3524                                         btrfs_ino(inode), cur_offset, 0,
3525                                         0, hole_size, 0, hole_size,
3526                                         0, 0, 0);
3527                         if (err)
3528                                 break;
3529
3530                         btrfs_drop_extent_cache(inode, hole_start,
3531                                         last_byte - 1, 0);
3532
3533                         btrfs_end_transaction(trans, root);
3534                 }
3535                 free_extent_map(em);
3536                 em = NULL;
3537                 cur_offset = last_byte;
3538                 if (cur_offset >= block_end)
3539                         break;
3540         }
3541
3542         free_extent_map(em);
3543         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3544                              GFP_NOFS);
3545         return err;
3546 }
3547
3548 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3549 {
3550         loff_t oldsize = i_size_read(inode);
3551         int ret;
3552
3553         if (newsize == oldsize)
3554                 return 0;
3555
3556         if (newsize > oldsize) {
3557                 i_size_write(inode, newsize);
3558                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3559                 truncate_pagecache(inode, oldsize, newsize);
3560                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3561                 if (ret) {
3562                         btrfs_setsize(inode, oldsize);
3563                         return ret;
3564                 }
3565
3566                 mark_inode_dirty(inode);
3567         } else {
3568
3569                 /*
3570                  * We're truncating a file that used to have good data down to
3571                  * zero. Make sure it gets into the ordered flush list so that
3572                  * any new writes get down to disk quickly.
3573                  */
3574                 if (newsize == 0)
3575                         BTRFS_I(inode)->ordered_data_close = 1;
3576
3577                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3578                 truncate_setsize(inode, newsize);
3579                 ret = btrfs_truncate(inode);
3580         }
3581
3582         return ret;
3583 }
3584
3585 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3586 {
3587         struct inode *inode = dentry->d_inode;
3588         struct btrfs_root *root = BTRFS_I(inode)->root;
3589         int err;
3590
3591         if (btrfs_root_readonly(root))
3592                 return -EROFS;
3593
3594         err = inode_change_ok(inode, attr);
3595         if (err)
3596                 return err;
3597
3598         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3599                 err = btrfs_setsize(inode, attr->ia_size);
3600                 if (err)
3601                         return err;
3602         }
3603
3604         if (attr->ia_valid) {
3605                 setattr_copy(inode, attr);
3606                 mark_inode_dirty(inode);
3607
3608                 if (attr->ia_valid & ATTR_MODE)
3609                         err = btrfs_acl_chmod(inode);
3610         }
3611
3612         return err;
3613 }
3614
3615 void btrfs_evict_inode(struct inode *inode)
3616 {
3617         struct btrfs_trans_handle *trans;
3618         struct btrfs_root *root = BTRFS_I(inode)->root;
3619         unsigned long nr;
3620         int ret;
3621
3622         trace_btrfs_inode_evict(inode);
3623
3624         truncate_inode_pages(&inode->i_data, 0);
3625         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3626                                is_free_space_inode(root, inode)))
3627                 goto no_delete;
3628
3629         if (is_bad_inode(inode)) {
3630                 btrfs_orphan_del(NULL, inode);
3631                 goto no_delete;
3632         }
3633         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3634         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3635
3636         if (root->fs_info->log_root_recovering) {
3637                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3638                 goto no_delete;
3639         }
3640
3641         if (inode->i_nlink > 0) {
3642                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3643                 goto no_delete;
3644         }
3645
3646         btrfs_i_size_write(inode, 0);
3647
3648         while (1) {
3649                 trans = btrfs_join_transaction(root);
3650                 BUG_ON(IS_ERR(trans));
3651                 trans->block_rsv = root->orphan_block_rsv;
3652
3653                 ret = btrfs_block_rsv_check(trans, root,
3654                                             root->orphan_block_rsv, 0, 5);
3655                 if (ret) {
3656                         BUG_ON(ret != -EAGAIN);
3657                         ret = btrfs_commit_transaction(trans, root);
3658                         BUG_ON(ret);
3659                         continue;
3660                 }
3661
3662                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3663                 if (ret != -EAGAIN)
3664                         break;
3665
3666                 nr = trans->blocks_used;
3667                 btrfs_end_transaction(trans, root);
3668                 trans = NULL;
3669                 btrfs_btree_balance_dirty(root, nr);
3670
3671         }
3672
3673         if (ret == 0) {
3674                 ret = btrfs_orphan_del(trans, inode);
3675                 BUG_ON(ret);
3676         }
3677
3678         if (!(root == root->fs_info->tree_root ||
3679               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3680                 btrfs_return_ino(root, btrfs_ino(inode));
3681
3682         nr = trans->blocks_used;
3683         btrfs_end_transaction(trans, root);
3684         btrfs_btree_balance_dirty(root, nr);
3685 no_delete:
3686         end_writeback(inode);
3687         return;
3688 }
3689
3690 /*
3691  * this returns the key found in the dir entry in the location pointer.
3692  * If no dir entries were found, location->objectid is 0.
3693  */
3694 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3695                                struct btrfs_key *location)
3696 {
3697         const char *name = dentry->d_name.name;
3698         int namelen = dentry->d_name.len;
3699         struct btrfs_dir_item *di;
3700         struct btrfs_path *path;
3701         struct btrfs_root *root = BTRFS_I(dir)->root;
3702         int ret = 0;
3703
3704         path = btrfs_alloc_path();
3705         BUG_ON(!path);
3706
3707         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3708                                     namelen, 0);
3709         if (IS_ERR(di))
3710                 ret = PTR_ERR(di);
3711
3712         if (IS_ERR_OR_NULL(di))
3713                 goto out_err;
3714
3715         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3716 out:
3717         btrfs_free_path(path);
3718         return ret;
3719 out_err:
3720         location->objectid = 0;
3721         goto out;
3722 }
3723
3724 /*
3725  * when we hit a tree root in a directory, the btrfs part of the inode
3726  * needs to be changed to reflect the root directory of the tree root.  This
3727  * is kind of like crossing a mount point.
3728  */
3729 static int fixup_tree_root_location(struct btrfs_root *root,
3730                                     struct inode *dir,
3731                                     struct dentry *dentry,
3732                                     struct btrfs_key *location,
3733                                     struct btrfs_root **sub_root)
3734 {
3735         struct btrfs_path *path;
3736         struct btrfs_root *new_root;
3737         struct btrfs_root_ref *ref;
3738         struct extent_buffer *leaf;
3739         int ret;
3740         int err = 0;
3741
3742         path = btrfs_alloc_path();
3743         if (!path) {
3744                 err = -ENOMEM;
3745                 goto out;
3746         }
3747
3748         err = -ENOENT;
3749         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3750                                   BTRFS_I(dir)->root->root_key.objectid,
3751                                   location->objectid);
3752         if (ret) {
3753                 if (ret < 0)
3754                         err = ret;
3755                 goto out;
3756         }
3757
3758         leaf = path->nodes[0];
3759         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3760         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3761             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3762                 goto out;
3763
3764         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3765                                    (unsigned long)(ref + 1),
3766                                    dentry->d_name.len);
3767         if (ret)
3768                 goto out;
3769
3770         btrfs_release_path(path);
3771
3772         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3773         if (IS_ERR(new_root)) {
3774                 err = PTR_ERR(new_root);
3775                 goto out;
3776         }
3777
3778         if (btrfs_root_refs(&new_root->root_item) == 0) {
3779                 err = -ENOENT;
3780                 goto out;
3781         }
3782
3783         *sub_root = new_root;
3784         location->objectid = btrfs_root_dirid(&new_root->root_item);
3785         location->type = BTRFS_INODE_ITEM_KEY;
3786         location->offset = 0;
3787         err = 0;
3788 out:
3789         btrfs_free_path(path);
3790         return err;
3791 }
3792
3793 static void inode_tree_add(struct inode *inode)
3794 {
3795         struct btrfs_root *root = BTRFS_I(inode)->root;
3796         struct btrfs_inode *entry;
3797         struct rb_node **p;
3798         struct rb_node *parent;
3799         u64 ino = btrfs_ino(inode);
3800 again:
3801         p = &root->inode_tree.rb_node;
3802         parent = NULL;
3803
3804         if (inode_unhashed(inode))
3805                 return;
3806
3807         spin_lock(&root->inode_lock);
3808         while (*p) {
3809                 parent = *p;
3810                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3811
3812                 if (ino < btrfs_ino(&entry->vfs_inode))
3813                         p = &parent->rb_left;
3814                 else if (ino > btrfs_ino(&entry->vfs_inode))
3815                         p = &parent->rb_right;
3816                 else {
3817                         WARN_ON(!(entry->vfs_inode.i_state &
3818                                   (I_WILL_FREE | I_FREEING)));
3819                         rb_erase(parent, &root->inode_tree);
3820                         RB_CLEAR_NODE(parent);
3821                         spin_unlock(&root->inode_lock);
3822                         goto again;
3823                 }
3824         }
3825         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3826         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3827         spin_unlock(&root->inode_lock);
3828 }
3829
3830 static void inode_tree_del(struct inode *inode)
3831 {
3832         struct btrfs_root *root = BTRFS_I(inode)->root;
3833         int empty = 0;
3834
3835         spin_lock(&root->inode_lock);
3836         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3837                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3838                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3839                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3840         }
3841         spin_unlock(&root->inode_lock);
3842
3843         /*
3844          * Free space cache has inodes in the tree root, but the tree root has a
3845          * root_refs of 0, so this could end up dropping the tree root as a
3846          * snapshot, so we need the extra !root->fs_info->tree_root check to
3847          * make sure we don't drop it.
3848          */
3849         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3850             root != root->fs_info->tree_root) {
3851                 synchronize_srcu(&root->fs_info->subvol_srcu);
3852                 spin_lock(&root->inode_lock);
3853                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3854                 spin_unlock(&root->inode_lock);
3855                 if (empty)
3856                         btrfs_add_dead_root(root);
3857         }
3858 }
3859
3860 int btrfs_invalidate_inodes(struct btrfs_root *root)
3861 {
3862         struct rb_node *node;
3863         struct rb_node *prev;
3864         struct btrfs_inode *entry;
3865         struct inode *inode;
3866         u64 objectid = 0;
3867
3868         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3869
3870         spin_lock(&root->inode_lock);
3871 again:
3872         node = root->inode_tree.rb_node;
3873         prev = NULL;
3874         while (node) {
3875                 prev = node;
3876                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3877
3878                 if (objectid < btrfs_ino(&entry->vfs_inode))
3879                         node = node->rb_left;
3880                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3881                         node = node->rb_right;
3882                 else
3883                         break;
3884         }
3885         if (!node) {
3886                 while (prev) {
3887                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3888                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3889                                 node = prev;
3890                                 break;
3891                         }
3892                         prev = rb_next(prev);
3893                 }
3894         }
3895         while (node) {
3896                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3897                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3898                 inode = igrab(&entry->vfs_inode);
3899                 if (inode) {
3900                         spin_unlock(&root->inode_lock);
3901                         if (atomic_read(&inode->i_count) > 1)
3902                                 d_prune_aliases(inode);
3903                         /*
3904                          * btrfs_drop_inode will have it removed from
3905                          * the inode cache when its usage count
3906                          * hits zero.
3907                          */
3908                         iput(inode);
3909                         cond_resched();
3910                         spin_lock(&root->inode_lock);
3911                         goto again;
3912                 }
3913
3914                 if (cond_resched_lock(&root->inode_lock))
3915                         goto again;
3916
3917                 node = rb_next(node);
3918         }
3919         spin_unlock(&root->inode_lock);
3920         return 0;
3921 }
3922
3923 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3924 {
3925         struct btrfs_iget_args *args = p;
3926         inode->i_ino = args->ino;
3927         BTRFS_I(inode)->root = args->root;
3928         btrfs_set_inode_space_info(args->root, inode);
3929         return 0;
3930 }
3931
3932 static int btrfs_find_actor(struct inode *inode, void *opaque)
3933 {
3934         struct btrfs_iget_args *args = opaque;
3935         return args->ino == btrfs_ino(inode) &&
3936                 args->root == BTRFS_I(inode)->root;
3937 }
3938
3939 static struct inode *btrfs_iget_locked(struct super_block *s,
3940                                        u64 objectid,
3941                                        struct btrfs_root *root)
3942 {
3943         struct inode *inode;
3944         struct btrfs_iget_args args;
3945         args.ino = objectid;
3946         args.root = root;
3947
3948         inode = iget5_locked(s, objectid, btrfs_find_actor,
3949                              btrfs_init_locked_inode,
3950                              (void *)&args);
3951         return inode;
3952 }
3953
3954 /* Get an inode object given its location and corresponding root.
3955  * Returns in *is_new if the inode was read from disk
3956  */
3957 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3958                          struct btrfs_root *root, int *new)
3959 {
3960         struct inode *inode;
3961
3962         inode = btrfs_iget_locked(s, location->objectid, root);
3963         if (!inode)
3964                 return ERR_PTR(-ENOMEM);
3965
3966         if (inode->i_state & I_NEW) {
3967                 BTRFS_I(inode)->root = root;
3968                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3969                 btrfs_read_locked_inode(inode);
3970                 inode_tree_add(inode);
3971                 unlock_new_inode(inode);
3972                 if (new)
3973                         *new = 1;
3974         }
3975
3976         return inode;
3977 }
3978
3979 static struct inode *new_simple_dir(struct super_block *s,
3980                                     struct btrfs_key *key,
3981                                     struct btrfs_root *root)
3982 {
3983         struct inode *inode = new_inode(s);
3984
3985         if (!inode)
3986                 return ERR_PTR(-ENOMEM);
3987
3988         BTRFS_I(inode)->root = root;
3989         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3990         BTRFS_I(inode)->dummy_inode = 1;
3991
3992         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3993         inode->i_op = &simple_dir_inode_operations;
3994         inode->i_fop = &simple_dir_operations;
3995         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3996         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3997
3998         return inode;
3999 }
4000
4001 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4002 {
4003         struct inode *inode;
4004         struct btrfs_root *root = BTRFS_I(dir)->root;
4005         struct btrfs_root *sub_root = root;
4006         struct btrfs_key location;
4007         int index;
4008         int ret;
4009
4010         if (dentry->d_name.len > BTRFS_NAME_LEN)
4011                 return ERR_PTR(-ENAMETOOLONG);
4012
4013         ret = btrfs_inode_by_name(dir, dentry, &location);
4014
4015         if (ret < 0)
4016                 return ERR_PTR(ret);
4017
4018         if (location.objectid == 0)
4019                 return NULL;
4020
4021         if (location.type == BTRFS_INODE_ITEM_KEY) {
4022                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4023                 return inode;
4024         }
4025
4026         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4027
4028         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4029         ret = fixup_tree_root_location(root, dir, dentry,
4030                                        &location, &sub_root);
4031         if (ret < 0) {
4032                 if (ret != -ENOENT)
4033                         inode = ERR_PTR(ret);
4034                 else
4035                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4036         } else {
4037                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4038         }
4039         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4040
4041         if (!IS_ERR(inode) && root != sub_root) {
4042                 down_read(&root->fs_info->cleanup_work_sem);
4043                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4044                         ret = btrfs_orphan_cleanup(sub_root);
4045                 up_read(&root->fs_info->cleanup_work_sem);
4046                 if (ret)
4047                         inode = ERR_PTR(ret);
4048         }
4049
4050         return inode;
4051 }
4052
4053 static int btrfs_dentry_delete(const struct dentry *dentry)
4054 {
4055         struct btrfs_root *root;
4056
4057         if (!dentry->d_inode && !IS_ROOT(dentry))
4058                 dentry = dentry->d_parent;
4059
4060         if (dentry->d_inode) {
4061                 root = BTRFS_I(dentry->d_inode)->root;
4062                 if (btrfs_root_refs(&root->root_item) == 0)
4063                         return 1;
4064         }
4065         return 0;
4066 }
4067
4068 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4069                                    struct nameidata *nd)
4070 {
4071         struct inode *inode;
4072
4073         inode = btrfs_lookup_dentry(dir, dentry);
4074         if (IS_ERR(inode))
4075                 return ERR_CAST(inode);
4076
4077         return d_splice_alias(inode, dentry);
4078 }
4079
4080 unsigned char btrfs_filetype_table[] = {
4081         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4082 };
4083
4084 static int btrfs_real_readdir(struct file *filp, void *dirent,
4085                               filldir_t filldir)
4086 {
4087         struct inode *inode = filp->f_dentry->d_inode;
4088         struct btrfs_root *root = BTRFS_I(inode)->root;
4089         struct btrfs_item *item;
4090         struct btrfs_dir_item *di;
4091         struct btrfs_key key;
4092         struct btrfs_key found_key;
4093         struct btrfs_path *path;
4094         struct list_head ins_list;
4095         struct list_head del_list;
4096         int ret;
4097         struct extent_buffer *leaf;
4098         int slot;
4099         unsigned char d_type;
4100         int over = 0;
4101         u32 di_cur;
4102         u32 di_total;
4103         u32 di_len;
4104         int key_type = BTRFS_DIR_INDEX_KEY;
4105         char tmp_name[32];
4106         char *name_ptr;
4107         int name_len;
4108         int is_curr = 0;        /* filp->f_pos points to the current index? */
4109
4110         /* FIXME, use a real flag for deciding about the key type */
4111         if (root->fs_info->tree_root == root)
4112                 key_type = BTRFS_DIR_ITEM_KEY;
4113
4114         /* special case for "." */
4115         if (filp->f_pos == 0) {
4116                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4117                 if (over)
4118                         return 0;
4119                 filp->f_pos = 1;
4120         }
4121         /* special case for .., just use the back ref */
4122         if (filp->f_pos == 1) {
4123                 u64 pino = parent_ino(filp->f_path.dentry);
4124                 over = filldir(dirent, "..", 2,
4125                                2, pino, DT_DIR);
4126                 if (over)
4127                         return 0;
4128                 filp->f_pos = 2;
4129         }
4130         path = btrfs_alloc_path();
4131         if (!path)
4132                 return -ENOMEM;
4133
4134         path->reada = 1;
4135
4136         if (key_type == BTRFS_DIR_INDEX_KEY) {
4137                 INIT_LIST_HEAD(&ins_list);
4138                 INIT_LIST_HEAD(&del_list);
4139                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4140         }
4141
4142         btrfs_set_key_type(&key, key_type);
4143         key.offset = filp->f_pos;
4144         key.objectid = btrfs_ino(inode);
4145
4146         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4147         if (ret < 0)
4148                 goto err;
4149
4150         while (1) {
4151                 leaf = path->nodes[0];
4152                 slot = path->slots[0];
4153                 if (slot >= btrfs_header_nritems(leaf)) {
4154                         ret = btrfs_next_leaf(root, path);
4155                         if (ret < 0)
4156                                 goto err;
4157                         else if (ret > 0)
4158                                 break;
4159                         continue;
4160                 }
4161
4162                 item = btrfs_item_nr(leaf, slot);
4163                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4164
4165                 if (found_key.objectid != key.objectid)
4166                         break;
4167                 if (btrfs_key_type(&found_key) != key_type)
4168                         break;
4169                 if (found_key.offset < filp->f_pos)
4170                         goto next;
4171                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4172                     btrfs_should_delete_dir_index(&del_list,
4173                                                   found_key.offset))
4174                         goto next;
4175
4176                 filp->f_pos = found_key.offset;
4177                 is_curr = 1;
4178
4179                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4180                 di_cur = 0;
4181                 di_total = btrfs_item_size(leaf, item);
4182
4183                 while (di_cur < di_total) {
4184                         struct btrfs_key location;
4185
4186                         if (verify_dir_item(root, leaf, di))
4187                                 break;
4188
4189                         name_len = btrfs_dir_name_len(leaf, di);
4190                         if (name_len <= sizeof(tmp_name)) {
4191                                 name_ptr = tmp_name;
4192                         } else {
4193                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4194                                 if (!name_ptr) {
4195                                         ret = -ENOMEM;
4196                                         goto err;
4197                                 }
4198                         }
4199                         read_extent_buffer(leaf, name_ptr,
4200                                            (unsigned long)(di + 1), name_len);
4201
4202                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4203                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4204
4205                         /* is this a reference to our own snapshot? If so
4206                          * skip it
4207                          */
4208                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4209                             location.objectid == root->root_key.objectid) {
4210                                 over = 0;
4211                                 goto skip;
4212                         }
4213                         over = filldir(dirent, name_ptr, name_len,
4214                                        found_key.offset, location.objectid,
4215                                        d_type);
4216
4217 skip:
4218                         if (name_ptr != tmp_name)
4219                                 kfree(name_ptr);
4220
4221                         if (over)
4222                                 goto nopos;
4223                         di_len = btrfs_dir_name_len(leaf, di) +
4224                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4225                         di_cur += di_len;
4226                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4227                 }
4228 next:
4229                 path->slots[0]++;
4230         }
4231
4232         if (key_type == BTRFS_DIR_INDEX_KEY) {
4233                 if (is_curr)
4234                         filp->f_pos++;
4235                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4236                                                       &ins_list);
4237                 if (ret)
4238                         goto nopos;
4239         }
4240
4241         /* Reached end of directory/root. Bump pos past the last item. */
4242         if (key_type == BTRFS_DIR_INDEX_KEY)
4243                 /*
4244                  * 32-bit glibc will use getdents64, but then strtol -
4245                  * so the last number we can serve is this.
4246                  */
4247                 filp->f_pos = 0x7fffffff;
4248         else
4249                 filp->f_pos++;
4250 nopos:
4251         ret = 0;
4252 err:
4253         if (key_type == BTRFS_DIR_INDEX_KEY)
4254                 btrfs_put_delayed_items(&ins_list, &del_list);
4255         btrfs_free_path(path);
4256         return ret;
4257 }
4258
4259 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4260 {
4261         struct btrfs_root *root = BTRFS_I(inode)->root;
4262         struct btrfs_trans_handle *trans;
4263         int ret = 0;
4264         bool nolock = false;
4265
4266         if (BTRFS_I(inode)->dummy_inode)
4267                 return 0;
4268
4269         if (btrfs_fs_closing(root->fs_info) && is_free_space_inode(root, inode))
4270                 nolock = true;
4271
4272         if (wbc->sync_mode == WB_SYNC_ALL) {
4273                 if (nolock)
4274                         trans = btrfs_join_transaction_nolock(root);
4275                 else
4276                         trans = btrfs_join_transaction(root);
4277                 if (IS_ERR(trans))
4278                         return PTR_ERR(trans);
4279                 if (nolock)
4280                         ret = btrfs_end_transaction_nolock(trans, root);
4281                 else
4282                         ret = btrfs_commit_transaction(trans, root);
4283         }
4284         return ret;
4285 }
4286
4287 /*
4288  * This is somewhat expensive, updating the tree every time the
4289  * inode changes.  But, it is most likely to find the inode in cache.
4290  * FIXME, needs more benchmarking...there are no reasons other than performance
4291  * to keep or drop this code.
4292  */
4293 void btrfs_dirty_inode(struct inode *inode, int flags)
4294 {
4295         struct btrfs_root *root = BTRFS_I(inode)->root;
4296         struct btrfs_trans_handle *trans;
4297         int ret;
4298
4299         if (BTRFS_I(inode)->dummy_inode)
4300                 return;
4301
4302         trans = btrfs_join_transaction(root);
4303         BUG_ON(IS_ERR(trans));
4304
4305         ret = btrfs_update_inode(trans, root, inode);
4306         if (ret && ret == -ENOSPC) {
4307                 /* whoops, lets try again with the full transaction */
4308                 btrfs_end_transaction(trans, root);
4309                 trans = btrfs_start_transaction(root, 1);
4310                 if (IS_ERR(trans)) {
4311                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4312                                        "dirty  inode %llu error %ld\n",
4313                                        (unsigned long long)btrfs_ino(inode),
4314                                        PTR_ERR(trans));
4315                         return;
4316                 }
4317
4318                 ret = btrfs_update_inode(trans, root, inode);
4319                 if (ret) {
4320                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4321                                        "dirty  inode %llu error %d\n",
4322                                        (unsigned long long)btrfs_ino(inode),
4323                                        ret);
4324                 }
4325         }
4326         btrfs_end_transaction(trans, root);
4327         if (BTRFS_I(inode)->delayed_node)
4328                 btrfs_balance_delayed_items(root);
4329 }
4330
4331 /*
4332  * find the highest existing sequence number in a directory
4333  * and then set the in-memory index_cnt variable to reflect
4334  * free sequence numbers
4335  */
4336 static int btrfs_set_inode_index_count(struct inode *inode)
4337 {
4338         struct btrfs_root *root = BTRFS_I(inode)->root;
4339         struct btrfs_key key, found_key;
4340         struct btrfs_path *path;
4341         struct extent_buffer *leaf;
4342         int ret;
4343
4344         key.objectid = btrfs_ino(inode);
4345         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4346         key.offset = (u64)-1;
4347
4348         path = btrfs_alloc_path();
4349         if (!path)
4350                 return -ENOMEM;
4351
4352         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4353         if (ret < 0)
4354                 goto out;
4355         /* FIXME: we should be able to handle this */
4356         if (ret == 0)
4357                 goto out;
4358         ret = 0;
4359
4360         /*
4361          * MAGIC NUMBER EXPLANATION:
4362          * since we search a directory based on f_pos we have to start at 2
4363          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4364          * else has to start at 2
4365          */
4366         if (path->slots[0] == 0) {
4367                 BTRFS_I(inode)->index_cnt = 2;
4368                 goto out;
4369         }
4370
4371         path->slots[0]--;
4372
4373         leaf = path->nodes[0];
4374         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4375
4376         if (found_key.objectid != btrfs_ino(inode) ||
4377             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4378                 BTRFS_I(inode)->index_cnt = 2;
4379                 goto out;
4380         }
4381
4382         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4383 out:
4384         btrfs_free_path(path);
4385         return ret;
4386 }
4387
4388 /*
4389  * helper to find a free sequence number in a given directory.  This current
4390  * code is very simple, later versions will do smarter things in the btree
4391  */
4392 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4393 {
4394         int ret = 0;
4395
4396         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4397                 ret = btrfs_inode_delayed_dir_index_count(dir);
4398                 if (ret) {
4399                         ret = btrfs_set_inode_index_count(dir);
4400                         if (ret)
4401                                 return ret;
4402                 }
4403         }
4404
4405         *index = BTRFS_I(dir)->index_cnt;
4406         BTRFS_I(dir)->index_cnt++;
4407
4408         return ret;
4409 }
4410
4411 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4412                                      struct btrfs_root *root,
4413                                      struct inode *dir,
4414                                      const char *name, int name_len,
4415                                      u64 ref_objectid, u64 objectid, int mode,
4416                                      u64 *index)
4417 {
4418         struct inode *inode;
4419         struct btrfs_inode_item *inode_item;
4420         struct btrfs_key *location;
4421         struct btrfs_path *path;
4422         struct btrfs_inode_ref *ref;
4423         struct btrfs_key key[2];
4424         u32 sizes[2];
4425         unsigned long ptr;
4426         int ret;
4427         int owner;
4428
4429         path = btrfs_alloc_path();
4430         BUG_ON(!path);
4431
4432         inode = new_inode(root->fs_info->sb);
4433         if (!inode) {
4434                 btrfs_free_path(path);
4435                 return ERR_PTR(-ENOMEM);
4436         }
4437
4438         /*
4439          * we have to initialize this early, so we can reclaim the inode
4440          * number if we fail afterwards in this function.
4441          */
4442         inode->i_ino = objectid;
4443
4444         if (dir) {
4445                 trace_btrfs_inode_request(dir);
4446
4447                 ret = btrfs_set_inode_index(dir, index);
4448                 if (ret) {
4449                         btrfs_free_path(path);
4450                         iput(inode);
4451                         return ERR_PTR(ret);
4452                 }
4453         }
4454         /*
4455          * index_cnt is ignored for everything but a dir,
4456          * btrfs_get_inode_index_count has an explanation for the magic
4457          * number
4458          */
4459         BTRFS_I(inode)->index_cnt = 2;
4460         BTRFS_I(inode)->root = root;
4461         BTRFS_I(inode)->generation = trans->transid;
4462         inode->i_generation = BTRFS_I(inode)->generation;
4463         btrfs_set_inode_space_info(root, inode);
4464
4465         if (mode & S_IFDIR)
4466                 owner = 0;
4467         else
4468                 owner = 1;
4469
4470         key[0].objectid = objectid;
4471         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4472         key[0].offset = 0;
4473
4474         key[1].objectid = objectid;
4475         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4476         key[1].offset = ref_objectid;
4477
4478         sizes[0] = sizeof(struct btrfs_inode_item);
4479         sizes[1] = name_len + sizeof(*ref);
4480
4481         path->leave_spinning = 1;
4482         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4483         if (ret != 0)
4484                 goto fail;
4485
4486         inode_init_owner(inode, dir, mode);
4487         inode_set_bytes(inode, 0);
4488         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4489         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4490                                   struct btrfs_inode_item);
4491         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4492
4493         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4494                              struct btrfs_inode_ref);
4495         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4496         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4497         ptr = (unsigned long)(ref + 1);
4498         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4499
4500         btrfs_mark_buffer_dirty(path->nodes[0]);
4501         btrfs_free_path(path);
4502
4503         location = &BTRFS_I(inode)->location;
4504         location->objectid = objectid;
4505         location->offset = 0;
4506         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4507
4508         btrfs_inherit_iflags(inode, dir);
4509
4510         if ((mode & S_IFREG)) {
4511                 if (btrfs_test_opt(root, NODATASUM))
4512                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4513                 if (btrfs_test_opt(root, NODATACOW) ||
4514                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4515                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4516         }
4517
4518         insert_inode_hash(inode);
4519         inode_tree_add(inode);
4520
4521         trace_btrfs_inode_new(inode);
4522
4523         return inode;
4524 fail:
4525         if (dir)
4526                 BTRFS_I(dir)->index_cnt--;
4527         btrfs_free_path(path);
4528         iput(inode);
4529         return ERR_PTR(ret);
4530 }
4531
4532 static inline u8 btrfs_inode_type(struct inode *inode)
4533 {
4534         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4535 }
4536
4537 /*
4538  * utility function to add 'inode' into 'parent_inode' with
4539  * a give name and a given sequence number.
4540  * if 'add_backref' is true, also insert a backref from the
4541  * inode to the parent directory.
4542  */
4543 int btrfs_add_link(struct btrfs_trans_handle *trans,
4544                    struct inode *parent_inode, struct inode *inode,
4545                    const char *name, int name_len, int add_backref, u64 index)
4546 {
4547         int ret = 0;
4548         struct btrfs_key key;
4549         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4550         u64 ino = btrfs_ino(inode);
4551         u64 parent_ino = btrfs_ino(parent_inode);
4552
4553         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4554                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4555         } else {
4556                 key.objectid = ino;
4557                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4558                 key.offset = 0;
4559         }
4560
4561         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4562                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4563                                          key.objectid, root->root_key.objectid,
4564                                          parent_ino, index, name, name_len);
4565         } else if (add_backref) {
4566                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4567                                              parent_ino, index);
4568         }
4569
4570         if (ret == 0) {
4571                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4572                                             parent_inode, &key,
4573                                             btrfs_inode_type(inode), index);
4574                 BUG_ON(ret);
4575
4576                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4577                                    name_len * 2);
4578                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4579                 ret = btrfs_update_inode(trans, root, parent_inode);
4580         }
4581         return ret;
4582 }
4583
4584 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4585                             struct inode *dir, struct dentry *dentry,
4586                             struct inode *inode, int backref, u64 index)
4587 {
4588         int err = btrfs_add_link(trans, dir, inode,
4589                                  dentry->d_name.name, dentry->d_name.len,
4590                                  backref, index);
4591         if (!err) {
4592                 d_instantiate(dentry, inode);
4593                 return 0;
4594         }
4595         if (err > 0)
4596                 err = -EEXIST;
4597         return err;
4598 }
4599
4600 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4601                         int mode, dev_t rdev)
4602 {
4603         struct btrfs_trans_handle *trans;
4604         struct btrfs_root *root = BTRFS_I(dir)->root;
4605         struct inode *inode = NULL;
4606         int err;
4607         int drop_inode = 0;
4608         u64 objectid;
4609         unsigned long nr = 0;
4610         u64 index = 0;
4611
4612         if (!new_valid_dev(rdev))
4613                 return -EINVAL;
4614
4615         /*
4616          * 2 for inode item and ref
4617          * 2 for dir items
4618          * 1 for xattr if selinux is on
4619          */
4620         trans = btrfs_start_transaction(root, 5);
4621         if (IS_ERR(trans))
4622                 return PTR_ERR(trans);
4623
4624         err = btrfs_find_free_ino(root, &objectid);
4625         if (err)
4626                 goto out_unlock;
4627
4628         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4629                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4630                                 mode, &index);
4631         if (IS_ERR(inode)) {
4632                 err = PTR_ERR(inode);
4633                 goto out_unlock;
4634         }
4635
4636         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4637         if (err) {
4638                 drop_inode = 1;
4639                 goto out_unlock;
4640         }
4641
4642         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4643         if (err)
4644                 drop_inode = 1;
4645         else {
4646                 inode->i_op = &btrfs_special_inode_operations;
4647                 init_special_inode(inode, inode->i_mode, rdev);
4648                 btrfs_update_inode(trans, root, inode);
4649         }
4650 out_unlock:
4651         nr = trans->blocks_used;
4652         btrfs_end_transaction_throttle(trans, root);
4653         btrfs_btree_balance_dirty(root, nr);
4654         if (drop_inode) {
4655                 inode_dec_link_count(inode);
4656                 iput(inode);
4657         }
4658         return err;
4659 }
4660
4661 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4662                         int mode, struct nameidata *nd)
4663 {
4664         struct btrfs_trans_handle *trans;
4665         struct btrfs_root *root = BTRFS_I(dir)->root;
4666         struct inode *inode = NULL;
4667         int drop_inode = 0;
4668         int err;
4669         unsigned long nr = 0;
4670         u64 objectid;
4671         u64 index = 0;
4672
4673         /*
4674          * 2 for inode item and ref
4675          * 2 for dir items
4676          * 1 for xattr if selinux is on
4677          */
4678         trans = btrfs_start_transaction(root, 5);
4679         if (IS_ERR(trans))
4680                 return PTR_ERR(trans);
4681
4682         err = btrfs_find_free_ino(root, &objectid);
4683         if (err)
4684                 goto out_unlock;
4685
4686         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4687                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4688                                 mode, &index);
4689         if (IS_ERR(inode)) {
4690                 err = PTR_ERR(inode);
4691                 goto out_unlock;
4692         }
4693
4694         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4695         if (err) {
4696                 drop_inode = 1;
4697                 goto out_unlock;
4698         }
4699
4700         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4701         if (err)
4702                 drop_inode = 1;
4703         else {
4704                 inode->i_mapping->a_ops = &btrfs_aops;
4705                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4706                 inode->i_fop = &btrfs_file_operations;
4707                 inode->i_op = &btrfs_file_inode_operations;
4708                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4709         }
4710 out_unlock:
4711         nr = trans->blocks_used;
4712         btrfs_end_transaction_throttle(trans, root);
4713         if (drop_inode) {
4714                 inode_dec_link_count(inode);
4715                 iput(inode);
4716         }
4717         btrfs_btree_balance_dirty(root, nr);
4718         return err;
4719 }
4720
4721 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4722                       struct dentry *dentry)
4723 {
4724         struct btrfs_trans_handle *trans;
4725         struct btrfs_root *root = BTRFS_I(dir)->root;
4726         struct inode *inode = old_dentry->d_inode;
4727         u64 index;
4728         unsigned long nr = 0;
4729         int err;
4730         int drop_inode = 0;
4731
4732         /* do not allow sys_link's with other subvols of the same device */
4733         if (root->objectid != BTRFS_I(inode)->root->objectid)
4734                 return -EXDEV;
4735
4736         if (inode->i_nlink == ~0U)
4737                 return -EMLINK;
4738
4739         err = btrfs_set_inode_index(dir, &index);
4740         if (err)
4741                 goto fail;
4742
4743         /*
4744          * 2 items for inode and inode ref
4745          * 2 items for dir items
4746          * 1 item for parent inode
4747          */
4748         trans = btrfs_start_transaction(root, 5);
4749         if (IS_ERR(trans)) {
4750                 err = PTR_ERR(trans);
4751                 goto fail;
4752         }
4753
4754         btrfs_inc_nlink(inode);
4755         inode->i_ctime = CURRENT_TIME;
4756         ihold(inode);
4757
4758         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4759
4760         if (err) {
4761                 drop_inode = 1;
4762         } else {
4763                 struct dentry *parent = dget_parent(dentry);
4764                 err = btrfs_update_inode(trans, root, inode);
4765                 BUG_ON(err);
4766                 btrfs_log_new_name(trans, inode, NULL, parent);
4767                 dput(parent);
4768         }
4769
4770         nr = trans->blocks_used;
4771         btrfs_end_transaction_throttle(trans, root);
4772 fail:
4773         if (drop_inode) {
4774                 inode_dec_link_count(inode);
4775                 iput(inode);
4776         }
4777         btrfs_btree_balance_dirty(root, nr);
4778         return err;
4779 }
4780
4781 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4782 {
4783         struct inode *inode = NULL;
4784         struct btrfs_trans_handle *trans;
4785         struct btrfs_root *root = BTRFS_I(dir)->root;
4786         int err = 0;
4787         int drop_on_err = 0;
4788         u64 objectid = 0;
4789         u64 index = 0;
4790         unsigned long nr = 1;
4791
4792         /*
4793          * 2 items for inode and ref
4794          * 2 items for dir items
4795          * 1 for xattr if selinux is on
4796          */
4797         trans = btrfs_start_transaction(root, 5);
4798         if (IS_ERR(trans))
4799                 return PTR_ERR(trans);
4800
4801         err = btrfs_find_free_ino(root, &objectid);
4802         if (err)
4803                 goto out_fail;
4804
4805         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4806                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4807                                 S_IFDIR | mode, &index);
4808         if (IS_ERR(inode)) {
4809                 err = PTR_ERR(inode);
4810                 goto out_fail;
4811         }
4812
4813         drop_on_err = 1;
4814
4815         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4816         if (err)
4817                 goto out_fail;
4818
4819         inode->i_op = &btrfs_dir_inode_operations;
4820         inode->i_fop = &btrfs_dir_file_operations;
4821
4822         btrfs_i_size_write(inode, 0);
4823         err = btrfs_update_inode(trans, root, inode);
4824         if (err)
4825                 goto out_fail;
4826
4827         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4828                              dentry->d_name.len, 0, index);
4829         if (err)
4830                 goto out_fail;
4831
4832         d_instantiate(dentry, inode);
4833         drop_on_err = 0;
4834
4835 out_fail:
4836         nr = trans->blocks_used;
4837         btrfs_end_transaction_throttle(trans, root);
4838         if (drop_on_err)
4839                 iput(inode);
4840         btrfs_btree_balance_dirty(root, nr);
4841         return err;
4842 }
4843
4844 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4845  * and an extent that you want to insert, deal with overlap and insert
4846  * the new extent into the tree.
4847  */
4848 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4849                                 struct extent_map *existing,
4850                                 struct extent_map *em,
4851                                 u64 map_start, u64 map_len)
4852 {
4853         u64 start_diff;
4854
4855         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4856         start_diff = map_start - em->start;
4857         em->start = map_start;
4858         em->len = map_len;
4859         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4860             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4861                 em->block_start += start_diff;
4862                 em->block_len -= start_diff;
4863         }
4864         return add_extent_mapping(em_tree, em);
4865 }
4866
4867 static noinline int uncompress_inline(struct btrfs_path *path,
4868                                       struct inode *inode, struct page *page,
4869                                       size_t pg_offset, u64 extent_offset,
4870                                       struct btrfs_file_extent_item *item)
4871 {
4872         int ret;
4873         struct extent_buffer *leaf = path->nodes[0];
4874         char *tmp;
4875         size_t max_size;
4876         unsigned long inline_size;
4877         unsigned long ptr;
4878         int compress_type;
4879
4880         WARN_ON(pg_offset != 0);
4881         compress_type = btrfs_file_extent_compression(leaf, item);
4882         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4883         inline_size = btrfs_file_extent_inline_item_len(leaf,
4884                                         btrfs_item_nr(leaf, path->slots[0]));
4885         tmp = kmalloc(inline_size, GFP_NOFS);
4886         if (!tmp)
4887                 return -ENOMEM;
4888         ptr = btrfs_file_extent_inline_start(item);
4889
4890         read_extent_buffer(leaf, tmp, ptr, inline_size);
4891
4892         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4893         ret = btrfs_decompress(compress_type, tmp, page,
4894                                extent_offset, inline_size, max_size);
4895         if (ret) {
4896                 char *kaddr = kmap_atomic(page, KM_USER0);
4897                 unsigned long copy_size = min_t(u64,
4898                                   PAGE_CACHE_SIZE - pg_offset,
4899                                   max_size - extent_offset);
4900                 memset(kaddr + pg_offset, 0, copy_size);
4901                 kunmap_atomic(kaddr, KM_USER0);
4902         }
4903         kfree(tmp);
4904         return 0;
4905 }
4906
4907 /*
4908  * a bit scary, this does extent mapping from logical file offset to the disk.
4909  * the ugly parts come from merging extents from the disk with the in-ram
4910  * representation.  This gets more complex because of the data=ordered code,
4911  * where the in-ram extents might be locked pending data=ordered completion.
4912  *
4913  * This also copies inline extents directly into the page.
4914  */
4915
4916 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4917                                     size_t pg_offset, u64 start, u64 len,
4918                                     int create)
4919 {
4920         int ret;
4921         int err = 0;
4922         u64 bytenr;
4923         u64 extent_start = 0;
4924         u64 extent_end = 0;
4925         u64 objectid = btrfs_ino(inode);
4926         u32 found_type;
4927         struct btrfs_path *path = NULL;
4928         struct btrfs_root *root = BTRFS_I(inode)->root;
4929         struct btrfs_file_extent_item *item;
4930         struct extent_buffer *leaf;
4931         struct btrfs_key found_key;
4932         struct extent_map *em = NULL;
4933         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4934         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4935         struct btrfs_trans_handle *trans = NULL;
4936         int compress_type;
4937
4938 again:
4939         read_lock(&em_tree->lock);
4940         em = lookup_extent_mapping(em_tree, start, len);
4941         if (em)
4942                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4943         read_unlock(&em_tree->lock);
4944
4945         if (em) {
4946                 if (em->start > start || em->start + em->len <= start)
4947                         free_extent_map(em);
4948                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4949                         free_extent_map(em);
4950                 else
4951                         goto out;
4952         }
4953         em = alloc_extent_map();
4954         if (!em) {
4955                 err = -ENOMEM;
4956                 goto out;
4957         }
4958         em->bdev = root->fs_info->fs_devices->latest_bdev;
4959         em->start = EXTENT_MAP_HOLE;
4960         em->orig_start = EXTENT_MAP_HOLE;
4961         em->len = (u64)-1;
4962         em->block_len = (u64)-1;
4963
4964         if (!path) {
4965                 path = btrfs_alloc_path();
4966                 if (!path) {
4967                         err = -ENOMEM;
4968                         goto out;
4969                 }
4970                 /*
4971                  * Chances are we'll be called again, so go ahead and do
4972                  * readahead
4973                  */
4974                 path->reada = 1;
4975         }
4976
4977         ret = btrfs_lookup_file_extent(trans, root, path,
4978                                        objectid, start, trans != NULL);
4979         if (ret < 0) {
4980                 err = ret;
4981                 goto out;
4982         }
4983
4984         if (ret != 0) {
4985                 if (path->slots[0] == 0)
4986                         goto not_found;
4987                 path->slots[0]--;
4988         }
4989
4990         leaf = path->nodes[0];
4991         item = btrfs_item_ptr(leaf, path->slots[0],
4992                               struct btrfs_file_extent_item);
4993         /* are we inside the extent that was found? */
4994         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4995         found_type = btrfs_key_type(&found_key);
4996         if (found_key.objectid != objectid ||
4997             found_type != BTRFS_EXTENT_DATA_KEY) {
4998                 goto not_found;
4999         }
5000
5001         found_type = btrfs_file_extent_type(leaf, item);
5002         extent_start = found_key.offset;
5003         compress_type = btrfs_file_extent_compression(leaf, item);
5004         if (found_type == BTRFS_FILE_EXTENT_REG ||
5005             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5006                 extent_end = extent_start +
5007                        btrfs_file_extent_num_bytes(leaf, item);
5008         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5009                 size_t size;
5010                 size = btrfs_file_extent_inline_len(leaf, item);
5011                 extent_end = (extent_start + size + root->sectorsize - 1) &
5012                         ~((u64)root->sectorsize - 1);
5013         }
5014
5015         if (start >= extent_end) {
5016                 path->slots[0]++;
5017                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5018                         ret = btrfs_next_leaf(root, path);
5019                         if (ret < 0) {
5020                                 err = ret;
5021                                 goto out;
5022                         }
5023                         if (ret > 0)
5024                                 goto not_found;
5025                         leaf = path->nodes[0];
5026                 }
5027                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5028                 if (found_key.objectid != objectid ||
5029                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5030                         goto not_found;
5031                 if (start + len <= found_key.offset)
5032                         goto not_found;
5033                 em->start = start;
5034                 em->len = found_key.offset - start;
5035                 goto not_found_em;
5036         }
5037
5038         if (found_type == BTRFS_FILE_EXTENT_REG ||
5039             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5040                 em->start = extent_start;
5041                 em->len = extent_end - extent_start;
5042                 em->orig_start = extent_start -
5043                                  btrfs_file_extent_offset(leaf, item);
5044                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5045                 if (bytenr == 0) {
5046                         em->block_start = EXTENT_MAP_HOLE;
5047                         goto insert;
5048                 }
5049                 if (compress_type != BTRFS_COMPRESS_NONE) {
5050                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5051                         em->compress_type = compress_type;
5052                         em->block_start = bytenr;
5053                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5054                                                                          item);
5055                 } else {
5056                         bytenr += btrfs_file_extent_offset(leaf, item);
5057                         em->block_start = bytenr;
5058                         em->block_len = em->len;
5059                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5060                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5061                 }
5062                 goto insert;
5063         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5064                 unsigned long ptr;
5065                 char *map;
5066                 size_t size;
5067                 size_t extent_offset;
5068                 size_t copy_size;
5069
5070                 em->block_start = EXTENT_MAP_INLINE;
5071                 if (!page || create) {
5072                         em->start = extent_start;
5073                         em->len = extent_end - extent_start;
5074                         goto out;
5075                 }
5076
5077                 size = btrfs_file_extent_inline_len(leaf, item);
5078                 extent_offset = page_offset(page) + pg_offset - extent_start;
5079                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5080                                 size - extent_offset);
5081                 em->start = extent_start + extent_offset;
5082                 em->len = (copy_size + root->sectorsize - 1) &
5083                         ~((u64)root->sectorsize - 1);
5084                 em->orig_start = EXTENT_MAP_INLINE;
5085                 if (compress_type) {
5086                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5087                         em->compress_type = compress_type;
5088                 }
5089                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5090                 if (create == 0 && !PageUptodate(page)) {
5091                         if (btrfs_file_extent_compression(leaf, item) !=
5092                             BTRFS_COMPRESS_NONE) {
5093                                 ret = uncompress_inline(path, inode, page,
5094                                                         pg_offset,
5095                                                         extent_offset, item);
5096                                 BUG_ON(ret);
5097                         } else {
5098                                 map = kmap(page);
5099                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5100                                                    copy_size);
5101                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5102                                         memset(map + pg_offset + copy_size, 0,
5103                                                PAGE_CACHE_SIZE - pg_offset -
5104                                                copy_size);
5105                                 }
5106                                 kunmap(page);
5107                         }
5108                         flush_dcache_page(page);
5109                 } else if (create && PageUptodate(page)) {
5110                         WARN_ON(1);
5111                         if (!trans) {
5112                                 kunmap(page);
5113                                 free_extent_map(em);
5114                                 em = NULL;
5115
5116                                 btrfs_release_path(path);
5117                                 trans = btrfs_join_transaction(root);
5118
5119                                 if (IS_ERR(trans))
5120                                         return ERR_CAST(trans);
5121                                 goto again;
5122                         }
5123                         map = kmap(page);
5124                         write_extent_buffer(leaf, map + pg_offset, ptr,
5125                                             copy_size);
5126                         kunmap(page);
5127                         btrfs_mark_buffer_dirty(leaf);
5128                 }
5129                 set_extent_uptodate(io_tree, em->start,
5130                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5131                 goto insert;
5132         } else {
5133                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5134                 WARN_ON(1);
5135         }
5136 not_found:
5137         em->start = start;
5138         em->len = len;
5139 not_found_em:
5140         em->block_start = EXTENT_MAP_HOLE;
5141         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5142 insert:
5143         btrfs_release_path(path);
5144         if (em->start > start || extent_map_end(em) <= start) {
5145                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5146                        "[%llu %llu]\n", (unsigned long long)em->start,
5147                        (unsigned long long)em->len,
5148                        (unsigned long long)start,
5149                        (unsigned long long)len);
5150                 err = -EIO;
5151                 goto out;
5152         }
5153
5154         err = 0;
5155         write_lock(&em_tree->lock);
5156         ret = add_extent_mapping(em_tree, em);
5157         /* it is possible that someone inserted the extent into the tree
5158          * while we had the lock dropped.  It is also possible that
5159          * an overlapping map exists in the tree
5160          */
5161         if (ret == -EEXIST) {
5162                 struct extent_map *existing;
5163
5164                 ret = 0;
5165
5166                 existing = lookup_extent_mapping(em_tree, start, len);
5167                 if (existing && (existing->start > start ||
5168                     existing->start + existing->len <= start)) {
5169                         free_extent_map(existing);
5170                         existing = NULL;
5171                 }
5172                 if (!existing) {
5173                         existing = lookup_extent_mapping(em_tree, em->start,
5174                                                          em->len);
5175                         if (existing) {
5176                                 err = merge_extent_mapping(em_tree, existing,
5177                                                            em, start,
5178                                                            root->sectorsize);
5179                                 free_extent_map(existing);
5180                                 if (err) {
5181                                         free_extent_map(em);
5182                                         em = NULL;
5183                                 }
5184                         } else {
5185                                 err = -EIO;
5186                                 free_extent_map(em);
5187                                 em = NULL;
5188                         }
5189                 } else {
5190                         free_extent_map(em);
5191                         em = existing;
5192                         err = 0;
5193                 }
5194         }
5195         write_unlock(&em_tree->lock);
5196 out:
5197
5198         trace_btrfs_get_extent(root, em);
5199
5200         if (path)
5201                 btrfs_free_path(path);
5202         if (trans) {
5203                 ret = btrfs_end_transaction(trans, root);
5204                 if (!err)
5205                         err = ret;
5206         }
5207         if (err) {
5208                 free_extent_map(em);
5209                 return ERR_PTR(err);
5210         }
5211         return em;
5212 }
5213
5214 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5215                                            size_t pg_offset, u64 start, u64 len,
5216                                            int create)
5217 {
5218         struct extent_map *em;
5219         struct extent_map *hole_em = NULL;
5220         u64 range_start = start;
5221         u64 end;
5222         u64 found;
5223         u64 found_end;
5224         int err = 0;
5225
5226         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5227         if (IS_ERR(em))
5228                 return em;
5229         if (em) {
5230                 /*
5231                  * if our em maps to a hole, there might
5232                  * actually be delalloc bytes behind it
5233                  */
5234                 if (em->block_start != EXTENT_MAP_HOLE)
5235                         return em;
5236                 else
5237                         hole_em = em;
5238         }
5239
5240         /* check to see if we've wrapped (len == -1 or similar) */
5241         end = start + len;
5242         if (end < start)
5243                 end = (u64)-1;
5244         else
5245                 end -= 1;
5246
5247         em = NULL;
5248
5249         /* ok, we didn't find anything, lets look for delalloc */
5250         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5251                                  end, len, EXTENT_DELALLOC, 1);
5252         found_end = range_start + found;
5253         if (found_end < range_start)
5254                 found_end = (u64)-1;
5255
5256         /*
5257          * we didn't find anything useful, return
5258          * the original results from get_extent()
5259          */
5260         if (range_start > end || found_end <= start) {
5261                 em = hole_em;
5262                 hole_em = NULL;
5263                 goto out;
5264         }
5265
5266         /* adjust the range_start to make sure it doesn't
5267          * go backwards from the start they passed in
5268          */
5269         range_start = max(start,range_start);
5270         found = found_end - range_start;
5271
5272         if (found > 0) {
5273                 u64 hole_start = start;
5274                 u64 hole_len = len;
5275
5276                 em = alloc_extent_map();
5277                 if (!em) {
5278                         err = -ENOMEM;
5279                         goto out;
5280                 }
5281                 /*
5282                  * when btrfs_get_extent can't find anything it
5283                  * returns one huge hole
5284                  *
5285                  * make sure what it found really fits our range, and
5286                  * adjust to make sure it is based on the start from
5287                  * the caller
5288                  */
5289                 if (hole_em) {
5290                         u64 calc_end = extent_map_end(hole_em);
5291
5292                         if (calc_end <= start || (hole_em->start > end)) {
5293                                 free_extent_map(hole_em);
5294                                 hole_em = NULL;
5295                         } else {
5296                                 hole_start = max(hole_em->start, start);
5297                                 hole_len = calc_end - hole_start;
5298                         }
5299                 }
5300                 em->bdev = NULL;
5301                 if (hole_em && range_start > hole_start) {
5302                         /* our hole starts before our delalloc, so we
5303                          * have to return just the parts of the hole
5304                          * that go until  the delalloc starts
5305                          */
5306                         em->len = min(hole_len,
5307                                       range_start - hole_start);
5308                         em->start = hole_start;
5309                         em->orig_start = hole_start;
5310                         /*
5311                          * don't adjust block start at all,
5312                          * it is fixed at EXTENT_MAP_HOLE
5313                          */
5314                         em->block_start = hole_em->block_start;
5315                         em->block_len = hole_len;
5316                 } else {
5317                         em->start = range_start;
5318                         em->len = found;
5319                         em->orig_start = range_start;
5320                         em->block_start = EXTENT_MAP_DELALLOC;
5321                         em->block_len = found;
5322                 }
5323         } else if (hole_em) {
5324                 return hole_em;
5325         }
5326 out:
5327
5328         free_extent_map(hole_em);
5329         if (err) {
5330                 free_extent_map(em);
5331                 return ERR_PTR(err);
5332         }
5333         return em;
5334 }
5335
5336 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5337                                                   struct extent_map *em,
5338                                                   u64 start, u64 len)
5339 {
5340         struct btrfs_root *root = BTRFS_I(inode)->root;
5341         struct btrfs_trans_handle *trans;
5342         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5343         struct btrfs_key ins;
5344         u64 alloc_hint;
5345         int ret;
5346         bool insert = false;
5347
5348         /*
5349          * Ok if the extent map we looked up is a hole and is for the exact
5350          * range we want, there is no reason to allocate a new one, however if
5351          * it is not right then we need to free this one and drop the cache for
5352          * our range.
5353          */
5354         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5355             em->len != len) {
5356                 free_extent_map(em);
5357                 em = NULL;
5358                 insert = true;
5359                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5360         }
5361
5362         trans = btrfs_join_transaction(root);
5363         if (IS_ERR(trans))
5364                 return ERR_CAST(trans);
5365
5366         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5367                 btrfs_add_inode_defrag(trans, inode);
5368
5369         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5370
5371         alloc_hint = get_extent_allocation_hint(inode, start, len);
5372         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5373                                    alloc_hint, (u64)-1, &ins, 1);
5374         if (ret) {
5375                 em = ERR_PTR(ret);
5376                 goto out;
5377         }
5378
5379         if (!em) {
5380                 em = alloc_extent_map();
5381                 if (!em) {
5382                         em = ERR_PTR(-ENOMEM);
5383                         goto out;
5384                 }
5385         }
5386
5387         em->start = start;
5388         em->orig_start = em->start;
5389         em->len = ins.offset;
5390
5391         em->block_start = ins.objectid;
5392         em->block_len = ins.offset;
5393         em->bdev = root->fs_info->fs_devices->latest_bdev;
5394
5395         /*
5396          * We need to do this because if we're using the original em we searched
5397          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5398          */
5399         em->flags = 0;
5400         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5401
5402         while (insert) {
5403                 write_lock(&em_tree->lock);
5404                 ret = add_extent_mapping(em_tree, em);
5405                 write_unlock(&em_tree->lock);
5406                 if (ret != -EEXIST)
5407                         break;
5408                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5409         }
5410
5411         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5412                                            ins.offset, ins.offset, 0);
5413         if (ret) {
5414                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5415                 em = ERR_PTR(ret);
5416         }
5417 out:
5418         btrfs_end_transaction(trans, root);
5419         return em;
5420 }
5421
5422 /*
5423  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5424  * block must be cow'd
5425  */
5426 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5427                                       struct inode *inode, u64 offset, u64 len)
5428 {
5429         struct btrfs_path *path;
5430         int ret;
5431         struct extent_buffer *leaf;
5432         struct btrfs_root *root = BTRFS_I(inode)->root;
5433         struct btrfs_file_extent_item *fi;
5434         struct btrfs_key key;
5435         u64 disk_bytenr;
5436         u64 backref_offset;
5437         u64 extent_end;
5438         u64 num_bytes;
5439         int slot;
5440         int found_type;
5441
5442         path = btrfs_alloc_path();
5443         if (!path)
5444                 return -ENOMEM;
5445
5446         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5447                                        offset, 0);
5448         if (ret < 0)
5449                 goto out;
5450
5451         slot = path->slots[0];
5452         if (ret == 1) {
5453                 if (slot == 0) {
5454                         /* can't find the item, must cow */
5455                         ret = 0;
5456                         goto out;
5457                 }
5458                 slot--;
5459         }
5460         ret = 0;
5461         leaf = path->nodes[0];
5462         btrfs_item_key_to_cpu(leaf, &key, slot);
5463         if (key.objectid != btrfs_ino(inode) ||
5464             key.type != BTRFS_EXTENT_DATA_KEY) {
5465                 /* not our file or wrong item type, must cow */
5466                 goto out;
5467         }
5468
5469         if (key.offset > offset) {
5470                 /* Wrong offset, must cow */
5471                 goto out;
5472         }
5473
5474         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5475         found_type = btrfs_file_extent_type(leaf, fi);
5476         if (found_type != BTRFS_FILE_EXTENT_REG &&
5477             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5478                 /* not a regular extent, must cow */
5479                 goto out;
5480         }
5481         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5482         backref_offset = btrfs_file_extent_offset(leaf, fi);
5483
5484         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5485         if (extent_end < offset + len) {
5486                 /* extent doesn't include our full range, must cow */
5487                 goto out;
5488         }
5489
5490         if (btrfs_extent_readonly(root, disk_bytenr))
5491                 goto out;
5492
5493         /*
5494          * look for other files referencing this extent, if we
5495          * find any we must cow
5496          */
5497         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5498                                   key.offset - backref_offset, disk_bytenr))
5499                 goto out;
5500
5501         /*
5502          * adjust disk_bytenr and num_bytes to cover just the bytes
5503          * in this extent we are about to write.  If there
5504          * are any csums in that range we have to cow in order
5505          * to keep the csums correct
5506          */
5507         disk_bytenr += backref_offset;
5508         disk_bytenr += offset - key.offset;
5509         num_bytes = min(offset + len, extent_end) - offset;
5510         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5511                                 goto out;
5512         /*
5513          * all of the above have passed, it is safe to overwrite this extent
5514          * without cow
5515          */
5516         ret = 1;
5517 out:
5518         btrfs_free_path(path);
5519         return ret;
5520 }
5521
5522 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5523                                    struct buffer_head *bh_result, int create)
5524 {
5525         struct extent_map *em;
5526         struct btrfs_root *root = BTRFS_I(inode)->root;
5527         u64 start = iblock << inode->i_blkbits;
5528         u64 len = bh_result->b_size;
5529         struct btrfs_trans_handle *trans;
5530
5531         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5532         if (IS_ERR(em))
5533                 return PTR_ERR(em);
5534
5535         /*
5536          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5537          * io.  INLINE is special, and we could probably kludge it in here, but
5538          * it's still buffered so for safety lets just fall back to the generic
5539          * buffered path.
5540          *
5541          * For COMPRESSED we _have_ to read the entire extent in so we can
5542          * decompress it, so there will be buffering required no matter what we
5543          * do, so go ahead and fallback to buffered.
5544          *
5545          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5546          * to buffered IO.  Don't blame me, this is the price we pay for using
5547          * the generic code.
5548          */
5549         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5550             em->block_start == EXTENT_MAP_INLINE) {
5551                 free_extent_map(em);
5552                 return -ENOTBLK;
5553         }
5554
5555         /* Just a good old fashioned hole, return */
5556         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5557                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5558                 free_extent_map(em);
5559                 /* DIO will do one hole at a time, so just unlock a sector */
5560                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5561                               start + root->sectorsize - 1, GFP_NOFS);
5562                 return 0;
5563         }
5564
5565         /*
5566          * We don't allocate a new extent in the following cases
5567          *
5568          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5569          * existing extent.
5570          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5571          * just use the extent.
5572          *
5573          */
5574         if (!create) {
5575                 len = em->len - (start - em->start);
5576                 goto map;
5577         }
5578
5579         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5580             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5581              em->block_start != EXTENT_MAP_HOLE)) {
5582                 int type;
5583                 int ret;
5584                 u64 block_start;
5585
5586                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5587                         type = BTRFS_ORDERED_PREALLOC;
5588                 else
5589                         type = BTRFS_ORDERED_NOCOW;
5590                 len = min(len, em->len - (start - em->start));
5591                 block_start = em->block_start + (start - em->start);
5592
5593                 /*
5594                  * we're not going to log anything, but we do need
5595                  * to make sure the current transaction stays open
5596                  * while we look for nocow cross refs
5597                  */
5598                 trans = btrfs_join_transaction(root);
5599                 if (IS_ERR(trans))
5600                         goto must_cow;
5601
5602                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5603                         ret = btrfs_add_ordered_extent_dio(inode, start,
5604                                            block_start, len, len, type);
5605                         btrfs_end_transaction(trans, root);
5606                         if (ret) {
5607                                 free_extent_map(em);
5608                                 return ret;
5609                         }
5610                         goto unlock;
5611                 }
5612                 btrfs_end_transaction(trans, root);
5613         }
5614 must_cow:
5615         /*
5616          * this will cow the extent, reset the len in case we changed
5617          * it above
5618          */
5619         len = bh_result->b_size;
5620         em = btrfs_new_extent_direct(inode, em, start, len);
5621         if (IS_ERR(em))
5622                 return PTR_ERR(em);
5623         len = min(len, em->len - (start - em->start));
5624 unlock:
5625         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5626                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5627                           0, NULL, GFP_NOFS);
5628 map:
5629         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5630                 inode->i_blkbits;
5631         bh_result->b_size = len;
5632         bh_result->b_bdev = em->bdev;
5633         set_buffer_mapped(bh_result);
5634         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5635                 set_buffer_new(bh_result);
5636
5637         free_extent_map(em);
5638
5639         return 0;
5640 }
5641
5642 struct btrfs_dio_private {
5643         struct inode *inode;
5644         u64 logical_offset;
5645         u64 disk_bytenr;
5646         u64 bytes;
5647         u32 *csums;
5648         void *private;
5649
5650         /* number of bios pending for this dio */
5651         atomic_t pending_bios;
5652
5653         /* IO errors */
5654         int errors;
5655
5656         struct bio *orig_bio;
5657 };
5658
5659 static void btrfs_endio_direct_read(struct bio *bio, int err)
5660 {
5661         struct btrfs_dio_private *dip = bio->bi_private;
5662         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5663         struct bio_vec *bvec = bio->bi_io_vec;
5664         struct inode *inode = dip->inode;
5665         struct btrfs_root *root = BTRFS_I(inode)->root;
5666         u64 start;
5667         u32 *private = dip->csums;
5668
5669         start = dip->logical_offset;
5670         do {
5671                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5672                         struct page *page = bvec->bv_page;
5673                         char *kaddr;
5674                         u32 csum = ~(u32)0;
5675                         unsigned long flags;
5676
5677                         local_irq_save(flags);
5678                         kaddr = kmap_atomic(page, KM_IRQ0);
5679                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5680                                                csum, bvec->bv_len);
5681                         btrfs_csum_final(csum, (char *)&csum);
5682                         kunmap_atomic(kaddr, KM_IRQ0);
5683                         local_irq_restore(flags);
5684
5685                         flush_dcache_page(bvec->bv_page);
5686                         if (csum != *private) {
5687                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5688                                       " %llu csum %u private %u\n",
5689                                       (unsigned long long)btrfs_ino(inode),
5690                                       (unsigned long long)start,
5691                                       csum, *private);
5692                                 err = -EIO;
5693                         }
5694                 }
5695
5696                 start += bvec->bv_len;
5697                 private++;
5698                 bvec++;
5699         } while (bvec <= bvec_end);
5700
5701         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5702                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5703         bio->bi_private = dip->private;
5704
5705         kfree(dip->csums);
5706         kfree(dip);
5707
5708         /* If we had a csum failure make sure to clear the uptodate flag */
5709         if (err)
5710                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5711         dio_end_io(bio, err);
5712 }
5713
5714 static void btrfs_endio_direct_write(struct bio *bio, int err)
5715 {
5716         struct btrfs_dio_private *dip = bio->bi_private;
5717         struct inode *inode = dip->inode;
5718         struct btrfs_root *root = BTRFS_I(inode)->root;
5719         struct btrfs_trans_handle *trans;
5720         struct btrfs_ordered_extent *ordered = NULL;
5721         struct extent_state *cached_state = NULL;
5722         u64 ordered_offset = dip->logical_offset;
5723         u64 ordered_bytes = dip->bytes;
5724         int ret;
5725
5726         if (err)
5727                 goto out_done;
5728 again:
5729         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5730                                                    &ordered_offset,
5731                                                    ordered_bytes);
5732         if (!ret)
5733                 goto out_test;
5734
5735         BUG_ON(!ordered);
5736
5737         trans = btrfs_join_transaction(root);
5738         if (IS_ERR(trans)) {
5739                 err = -ENOMEM;
5740                 goto out;
5741         }
5742         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5743
5744         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5745                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5746                 if (!ret)
5747                         ret = btrfs_update_inode(trans, root, inode);
5748                 err = ret;
5749                 goto out;
5750         }
5751
5752         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5753                          ordered->file_offset + ordered->len - 1, 0,
5754                          &cached_state, GFP_NOFS);
5755
5756         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5757                 ret = btrfs_mark_extent_written(trans, inode,
5758                                                 ordered->file_offset,
5759                                                 ordered->file_offset +
5760                                                 ordered->len);
5761                 if (ret) {
5762                         err = ret;
5763                         goto out_unlock;
5764                 }
5765         } else {
5766                 ret = insert_reserved_file_extent(trans, inode,
5767                                                   ordered->file_offset,
5768                                                   ordered->start,
5769                                                   ordered->disk_len,
5770                                                   ordered->len,
5771                                                   ordered->len,
5772                                                   0, 0, 0,
5773                                                   BTRFS_FILE_EXTENT_REG);
5774                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5775                                    ordered->file_offset, ordered->len);
5776                 if (ret) {
5777                         err = ret;
5778                         WARN_ON(1);
5779                         goto out_unlock;
5780                 }
5781         }
5782
5783         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5784         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5785         if (!ret)
5786                 btrfs_update_inode(trans, root, inode);
5787         ret = 0;
5788 out_unlock:
5789         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5790                              ordered->file_offset + ordered->len - 1,
5791                              &cached_state, GFP_NOFS);
5792 out:
5793         btrfs_delalloc_release_metadata(inode, ordered->len);
5794         btrfs_end_transaction(trans, root);
5795         ordered_offset = ordered->file_offset + ordered->len;
5796         btrfs_put_ordered_extent(ordered);
5797         btrfs_put_ordered_extent(ordered);
5798
5799 out_test:
5800         /*
5801          * our bio might span multiple ordered extents.  If we haven't
5802          * completed the accounting for the whole dio, go back and try again
5803          */
5804         if (ordered_offset < dip->logical_offset + dip->bytes) {
5805                 ordered_bytes = dip->logical_offset + dip->bytes -
5806                         ordered_offset;
5807                 goto again;
5808         }
5809 out_done:
5810         bio->bi_private = dip->private;
5811
5812         kfree(dip->csums);
5813         kfree(dip);
5814
5815         /* If we had an error make sure to clear the uptodate flag */
5816         if (err)
5817                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5818         dio_end_io(bio, err);
5819 }
5820
5821 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5822                                     struct bio *bio, int mirror_num,
5823                                     unsigned long bio_flags, u64 offset)
5824 {
5825         int ret;
5826         struct btrfs_root *root = BTRFS_I(inode)->root;
5827         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5828         BUG_ON(ret);
5829         return 0;
5830 }
5831
5832 static void btrfs_end_dio_bio(struct bio *bio, int err)
5833 {
5834         struct btrfs_dio_private *dip = bio->bi_private;
5835
5836         if (err) {
5837                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5838                       "sector %#Lx len %u err no %d\n",
5839                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5840                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5841                 dip->errors = 1;
5842
5843                 /*
5844                  * before atomic variable goto zero, we must make sure
5845                  * dip->errors is perceived to be set.
5846                  */
5847                 smp_mb__before_atomic_dec();
5848         }
5849
5850         /* if there are more bios still pending for this dio, just exit */
5851         if (!atomic_dec_and_test(&dip->pending_bios))
5852                 goto out;
5853
5854         if (dip->errors)
5855                 bio_io_error(dip->orig_bio);
5856         else {
5857                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5858                 bio_endio(dip->orig_bio, 0);
5859         }
5860 out:
5861         bio_put(bio);
5862 }
5863
5864 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5865                                        u64 first_sector, gfp_t gfp_flags)
5866 {
5867         int nr_vecs = bio_get_nr_vecs(bdev);
5868         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5869 }
5870
5871 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5872                                          int rw, u64 file_offset, int skip_sum,
5873                                          u32 *csums, int async_submit)
5874 {
5875         int write = rw & REQ_WRITE;
5876         struct btrfs_root *root = BTRFS_I(inode)->root;
5877         int ret;
5878
5879         bio_get(bio);
5880         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5881         if (ret)
5882                 goto err;
5883
5884         if (skip_sum)
5885                 goto map;
5886
5887         if (write && async_submit) {
5888                 ret = btrfs_wq_submit_bio(root->fs_info,
5889                                    inode, rw, bio, 0, 0,
5890                                    file_offset,
5891                                    __btrfs_submit_bio_start_direct_io,
5892                                    __btrfs_submit_bio_done);
5893                 goto err;
5894         } else if (write) {
5895                 /*
5896                  * If we aren't doing async submit, calculate the csum of the
5897                  * bio now.
5898                  */
5899                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5900                 if (ret)
5901                         goto err;
5902         } else if (!skip_sum) {
5903                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5904                                           file_offset, csums);
5905                 if (ret)
5906                         goto err;
5907         }
5908
5909 map:
5910         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5911 err:
5912         bio_put(bio);
5913         return ret;
5914 }
5915
5916 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5917                                     int skip_sum)
5918 {
5919         struct inode *inode = dip->inode;
5920         struct btrfs_root *root = BTRFS_I(inode)->root;
5921         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5922         struct bio *bio;
5923         struct bio *orig_bio = dip->orig_bio;
5924         struct bio_vec *bvec = orig_bio->bi_io_vec;
5925         u64 start_sector = orig_bio->bi_sector;
5926         u64 file_offset = dip->logical_offset;
5927         u64 submit_len = 0;
5928         u64 map_length;
5929         int nr_pages = 0;
5930         u32 *csums = dip->csums;
5931         int ret = 0;
5932         int async_submit = 0;
5933         int write = rw & REQ_WRITE;
5934
5935         map_length = orig_bio->bi_size;
5936         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5937                               &map_length, NULL, 0);
5938         if (ret) {
5939                 bio_put(orig_bio);
5940                 return -EIO;
5941         }
5942
5943         if (map_length >= orig_bio->bi_size) {
5944                 bio = orig_bio;
5945                 goto submit;
5946         }
5947
5948         async_submit = 1;
5949         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5950         if (!bio)
5951                 return -ENOMEM;
5952         bio->bi_private = dip;
5953         bio->bi_end_io = btrfs_end_dio_bio;
5954         atomic_inc(&dip->pending_bios);
5955
5956         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5957                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5958                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5959                                  bvec->bv_offset) < bvec->bv_len)) {
5960                         /*
5961                          * inc the count before we submit the bio so
5962                          * we know the end IO handler won't happen before
5963                          * we inc the count. Otherwise, the dip might get freed
5964                          * before we're done setting it up
5965                          */
5966                         atomic_inc(&dip->pending_bios);
5967                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5968                                                      file_offset, skip_sum,
5969                                                      csums, async_submit);
5970                         if (ret) {
5971                                 bio_put(bio);
5972                                 atomic_dec(&dip->pending_bios);
5973                                 goto out_err;
5974                         }
5975
5976                         /* Write's use the ordered csums */
5977                         if (!write && !skip_sum)
5978                                 csums = csums + nr_pages;
5979                         start_sector += submit_len >> 9;
5980                         file_offset += submit_len;
5981
5982                         submit_len = 0;
5983                         nr_pages = 0;
5984
5985                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5986                                                   start_sector, GFP_NOFS);
5987                         if (!bio)
5988                                 goto out_err;
5989                         bio->bi_private = dip;
5990                         bio->bi_end_io = btrfs_end_dio_bio;
5991
5992                         map_length = orig_bio->bi_size;
5993                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5994                                               &map_length, NULL, 0);
5995                         if (ret) {
5996                                 bio_put(bio);
5997                                 goto out_err;
5998                         }
5999                 } else {
6000                         submit_len += bvec->bv_len;
6001                         nr_pages ++;
6002                         bvec++;
6003                 }
6004         }
6005
6006 submit:
6007         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6008                                      csums, async_submit);
6009         if (!ret)
6010                 return 0;
6011
6012         bio_put(bio);
6013 out_err:
6014         dip->errors = 1;
6015         /*
6016          * before atomic variable goto zero, we must
6017          * make sure dip->errors is perceived to be set.
6018          */
6019         smp_mb__before_atomic_dec();
6020         if (atomic_dec_and_test(&dip->pending_bios))
6021                 bio_io_error(dip->orig_bio);
6022
6023         /* bio_end_io() will handle error, so we needn't return it */
6024         return 0;
6025 }
6026
6027 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6028                                 loff_t file_offset)
6029 {
6030         struct btrfs_root *root = BTRFS_I(inode)->root;
6031         struct btrfs_dio_private *dip;
6032         struct bio_vec *bvec = bio->bi_io_vec;
6033         int skip_sum;
6034         int write = rw & REQ_WRITE;
6035         int ret = 0;
6036
6037         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6038
6039         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6040         if (!dip) {
6041                 ret = -ENOMEM;
6042                 goto free_ordered;
6043         }
6044         dip->csums = NULL;
6045
6046         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6047         if (!write && !skip_sum) {
6048                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6049                 if (!dip->csums) {
6050                         kfree(dip);
6051                         ret = -ENOMEM;
6052                         goto free_ordered;
6053                 }
6054         }
6055
6056         dip->private = bio->bi_private;
6057         dip->inode = inode;
6058         dip->logical_offset = file_offset;
6059
6060         dip->bytes = 0;
6061         do {
6062                 dip->bytes += bvec->bv_len;
6063                 bvec++;
6064         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6065
6066         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6067         bio->bi_private = dip;
6068         dip->errors = 0;
6069         dip->orig_bio = bio;
6070         atomic_set(&dip->pending_bios, 0);
6071
6072         if (write)
6073                 bio->bi_end_io = btrfs_endio_direct_write;
6074         else
6075                 bio->bi_end_io = btrfs_endio_direct_read;
6076
6077         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6078         if (!ret)
6079                 return;
6080 free_ordered:
6081         /*
6082          * If this is a write, we need to clean up the reserved space and kill
6083          * the ordered extent.
6084          */
6085         if (write) {
6086                 struct btrfs_ordered_extent *ordered;
6087                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6088                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6089                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6090                         btrfs_free_reserved_extent(root, ordered->start,
6091                                                    ordered->disk_len);
6092                 btrfs_put_ordered_extent(ordered);
6093                 btrfs_put_ordered_extent(ordered);
6094         }
6095         bio_endio(bio, ret);
6096 }
6097
6098 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6099                         const struct iovec *iov, loff_t offset,
6100                         unsigned long nr_segs)
6101 {
6102         int seg;
6103         int i;
6104         size_t size;
6105         unsigned long addr;
6106         unsigned blocksize_mask = root->sectorsize - 1;
6107         ssize_t retval = -EINVAL;
6108         loff_t end = offset;
6109
6110         if (offset & blocksize_mask)
6111                 goto out;
6112
6113         /* Check the memory alignment.  Blocks cannot straddle pages */
6114         for (seg = 0; seg < nr_segs; seg++) {
6115                 addr = (unsigned long)iov[seg].iov_base;
6116                 size = iov[seg].iov_len;
6117                 end += size;
6118                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6119                         goto out;
6120
6121                 /* If this is a write we don't need to check anymore */
6122                 if (rw & WRITE)
6123                         continue;
6124
6125                 /*
6126                  * Check to make sure we don't have duplicate iov_base's in this
6127                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6128                  * when reading back.
6129                  */
6130                 for (i = seg + 1; i < nr_segs; i++) {
6131                         if (iov[seg].iov_base == iov[i].iov_base)
6132                                 goto out;
6133                 }
6134         }
6135         retval = 0;
6136 out:
6137         return retval;
6138 }
6139 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6140                         const struct iovec *iov, loff_t offset,
6141                         unsigned long nr_segs)
6142 {
6143         struct file *file = iocb->ki_filp;
6144         struct inode *inode = file->f_mapping->host;
6145         struct btrfs_ordered_extent *ordered;
6146         struct extent_state *cached_state = NULL;
6147         u64 lockstart, lockend;
6148         ssize_t ret;
6149         int writing = rw & WRITE;
6150         int write_bits = 0;
6151         size_t count = iov_length(iov, nr_segs);
6152
6153         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6154                             offset, nr_segs)) {
6155                 return 0;
6156         }
6157
6158         lockstart = offset;
6159         lockend = offset + count - 1;
6160
6161         if (writing) {
6162                 ret = btrfs_delalloc_reserve_space(inode, count);
6163                 if (ret)
6164                         goto out;
6165         }
6166
6167         while (1) {
6168                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6169                                  0, &cached_state, GFP_NOFS);
6170                 /*
6171                  * We're concerned with the entire range that we're going to be
6172                  * doing DIO to, so we need to make sure theres no ordered
6173                  * extents in this range.
6174                  */
6175                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6176                                                      lockend - lockstart + 1);
6177                 if (!ordered)
6178                         break;
6179                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6180                                      &cached_state, GFP_NOFS);
6181                 btrfs_start_ordered_extent(inode, ordered, 1);
6182                 btrfs_put_ordered_extent(ordered);
6183                 cond_resched();
6184         }
6185
6186         /*
6187          * we don't use btrfs_set_extent_delalloc because we don't want
6188          * the dirty or uptodate bits
6189          */
6190         if (writing) {
6191                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6192                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6193                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6194                                      GFP_NOFS);
6195                 if (ret) {
6196                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6197                                          lockend, EXTENT_LOCKED | write_bits,
6198                                          1, 0, &cached_state, GFP_NOFS);
6199                         goto out;
6200                 }
6201         }
6202
6203         free_extent_state(cached_state);
6204         cached_state = NULL;
6205
6206         ret = __blockdev_direct_IO(rw, iocb, inode,
6207                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6208                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6209                    btrfs_submit_direct, 0);
6210
6211         if (ret < 0 && ret != -EIOCBQUEUED) {
6212                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6213                               offset + iov_length(iov, nr_segs) - 1,
6214                               EXTENT_LOCKED | write_bits, 1, 0,
6215                               &cached_state, GFP_NOFS);
6216         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6217                 /*
6218                  * We're falling back to buffered, unlock the section we didn't
6219                  * do IO on.
6220                  */
6221                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6222                               offset + iov_length(iov, nr_segs) - 1,
6223                               EXTENT_LOCKED | write_bits, 1, 0,
6224                               &cached_state, GFP_NOFS);
6225         }
6226 out:
6227         free_extent_state(cached_state);
6228         return ret;
6229 }
6230
6231 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6232                 __u64 start, __u64 len)
6233 {
6234         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6235 }
6236
6237 int btrfs_readpage(struct file *file, struct page *page)
6238 {
6239         struct extent_io_tree *tree;
6240         tree = &BTRFS_I(page->mapping->host)->io_tree;
6241         return extent_read_full_page(tree, page, btrfs_get_extent);
6242 }
6243
6244 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6245 {
6246         struct extent_io_tree *tree;
6247
6248
6249         if (current->flags & PF_MEMALLOC) {
6250                 redirty_page_for_writepage(wbc, page);
6251                 unlock_page(page);
6252                 return 0;
6253         }
6254         tree = &BTRFS_I(page->mapping->host)->io_tree;
6255         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6256 }
6257
6258 int btrfs_writepages(struct address_space *mapping,
6259                      struct writeback_control *wbc)
6260 {
6261         struct extent_io_tree *tree;
6262
6263         tree = &BTRFS_I(mapping->host)->io_tree;
6264         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6265 }
6266
6267 static int
6268 btrfs_readpages(struct file *file, struct address_space *mapping,
6269                 struct list_head *pages, unsigned nr_pages)
6270 {
6271         struct extent_io_tree *tree;
6272         tree = &BTRFS_I(mapping->host)->io_tree;
6273         return extent_readpages(tree, mapping, pages, nr_pages,
6274                                 btrfs_get_extent);
6275 }
6276 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6277 {
6278         struct extent_io_tree *tree;
6279         struct extent_map_tree *map;
6280         int ret;
6281
6282         tree = &BTRFS_I(page->mapping->host)->io_tree;
6283         map = &BTRFS_I(page->mapping->host)->extent_tree;
6284         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6285         if (ret == 1) {
6286                 ClearPagePrivate(page);
6287                 set_page_private(page, 0);
6288                 page_cache_release(page);
6289         }
6290         return ret;
6291 }
6292
6293 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6294 {
6295         if (PageWriteback(page) || PageDirty(page))
6296                 return 0;
6297         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6298 }
6299
6300 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6301 {
6302         struct extent_io_tree *tree;
6303         struct btrfs_ordered_extent *ordered;
6304         struct extent_state *cached_state = NULL;
6305         u64 page_start = page_offset(page);
6306         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6307
6308
6309         /*
6310          * we have the page locked, so new writeback can't start,
6311          * and the dirty bit won't be cleared while we are here.
6312          *
6313          * Wait for IO on this page so that we can safely clear
6314          * the PagePrivate2 bit and do ordered accounting
6315          */
6316         wait_on_page_writeback(page);
6317
6318         tree = &BTRFS_I(page->mapping->host)->io_tree;
6319         if (offset) {
6320                 btrfs_releasepage(page, GFP_NOFS);
6321                 return;
6322         }
6323         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6324                          GFP_NOFS);
6325         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6326                                            page_offset(page));
6327         if (ordered) {
6328                 /*
6329                  * IO on this page will never be started, so we need
6330                  * to account for any ordered extents now
6331                  */
6332                 clear_extent_bit(tree, page_start, page_end,
6333                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6334                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6335                                  &cached_state, GFP_NOFS);
6336                 /*
6337                  * whoever cleared the private bit is responsible
6338                  * for the finish_ordered_io
6339                  */
6340                 if (TestClearPagePrivate2(page)) {
6341                         btrfs_finish_ordered_io(page->mapping->host,
6342                                                 page_start, page_end);
6343                 }
6344                 btrfs_put_ordered_extent(ordered);
6345                 cached_state = NULL;
6346                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6347                                  GFP_NOFS);
6348         }
6349         clear_extent_bit(tree, page_start, page_end,
6350                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6351                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6352         __btrfs_releasepage(page, GFP_NOFS);
6353
6354         ClearPageChecked(page);
6355         if (PagePrivate(page)) {
6356                 ClearPagePrivate(page);
6357                 set_page_private(page, 0);
6358                 page_cache_release(page);
6359         }
6360 }
6361
6362 /*
6363  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6364  * called from a page fault handler when a page is first dirtied. Hence we must
6365  * be careful to check for EOF conditions here. We set the page up correctly
6366  * for a written page which means we get ENOSPC checking when writing into
6367  * holes and correct delalloc and unwritten extent mapping on filesystems that
6368  * support these features.
6369  *
6370  * We are not allowed to take the i_mutex here so we have to play games to
6371  * protect against truncate races as the page could now be beyond EOF.  Because
6372  * vmtruncate() writes the inode size before removing pages, once we have the
6373  * page lock we can determine safely if the page is beyond EOF. If it is not
6374  * beyond EOF, then the page is guaranteed safe against truncation until we
6375  * unlock the page.
6376  */
6377 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6378 {
6379         struct page *page = vmf->page;
6380         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6381         struct btrfs_root *root = BTRFS_I(inode)->root;
6382         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6383         struct btrfs_ordered_extent *ordered;
6384         struct extent_state *cached_state = NULL;
6385         char *kaddr;
6386         unsigned long zero_start;
6387         loff_t size;
6388         int ret;
6389         u64 page_start;
6390         u64 page_end;
6391
6392         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6393         if (ret) {
6394                 if (ret == -ENOMEM)
6395                         ret = VM_FAULT_OOM;
6396                 else /* -ENOSPC, -EIO, etc */
6397                         ret = VM_FAULT_SIGBUS;
6398                 goto out;
6399         }
6400
6401         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6402 again:
6403         lock_page(page);
6404         size = i_size_read(inode);
6405         page_start = page_offset(page);
6406         page_end = page_start + PAGE_CACHE_SIZE - 1;
6407
6408         if ((page->mapping != inode->i_mapping) ||
6409             (page_start >= size)) {
6410                 /* page got truncated out from underneath us */
6411                 goto out_unlock;
6412         }
6413         wait_on_page_writeback(page);
6414
6415         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6416                          GFP_NOFS);
6417         set_page_extent_mapped(page);
6418
6419         /*
6420          * we can't set the delalloc bits if there are pending ordered
6421          * extents.  Drop our locks and wait for them to finish
6422          */
6423         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6424         if (ordered) {
6425                 unlock_extent_cached(io_tree, page_start, page_end,
6426                                      &cached_state, GFP_NOFS);
6427                 unlock_page(page);
6428                 btrfs_start_ordered_extent(inode, ordered, 1);
6429                 btrfs_put_ordered_extent(ordered);
6430                 goto again;
6431         }
6432
6433         /*
6434          * XXX - page_mkwrite gets called every time the page is dirtied, even
6435          * if it was already dirty, so for space accounting reasons we need to
6436          * clear any delalloc bits for the range we are fixing to save.  There
6437          * is probably a better way to do this, but for now keep consistent with
6438          * prepare_pages in the normal write path.
6439          */
6440         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6441                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6442                           0, 0, &cached_state, GFP_NOFS);
6443
6444         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6445                                         &cached_state);
6446         if (ret) {
6447                 unlock_extent_cached(io_tree, page_start, page_end,
6448                                      &cached_state, GFP_NOFS);
6449                 ret = VM_FAULT_SIGBUS;
6450                 goto out_unlock;
6451         }
6452         ret = 0;
6453
6454         /* page is wholly or partially inside EOF */
6455         if (page_start + PAGE_CACHE_SIZE > size)
6456                 zero_start = size & ~PAGE_CACHE_MASK;
6457         else
6458                 zero_start = PAGE_CACHE_SIZE;
6459
6460         if (zero_start != PAGE_CACHE_SIZE) {
6461                 kaddr = kmap(page);
6462                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6463                 flush_dcache_page(page);
6464                 kunmap(page);
6465         }
6466         ClearPageChecked(page);
6467         set_page_dirty(page);
6468         SetPageUptodate(page);
6469
6470         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6471         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6472
6473         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6474
6475 out_unlock:
6476         if (!ret)
6477                 return VM_FAULT_LOCKED;
6478         unlock_page(page);
6479         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6480 out:
6481         return ret;
6482 }
6483
6484 static int btrfs_truncate(struct inode *inode)
6485 {
6486         struct btrfs_root *root = BTRFS_I(inode)->root;
6487         struct btrfs_block_rsv *rsv;
6488         int ret;
6489         int err = 0;
6490         struct btrfs_trans_handle *trans;
6491         unsigned long nr;
6492         u64 mask = root->sectorsize - 1;
6493
6494         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6495         if (ret)
6496                 return ret;
6497
6498         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6499         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6500
6501         /*
6502          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6503          * 3 things going on here
6504          *
6505          * 1) We need to reserve space for our orphan item and the space to
6506          * delete our orphan item.  Lord knows we don't want to have a dangling
6507          * orphan item because we didn't reserve space to remove it.
6508          *
6509          * 2) We need to reserve space to update our inode.
6510          *
6511          * 3) We need to have something to cache all the space that is going to
6512          * be free'd up by the truncate operation, but also have some slack
6513          * space reserved in case it uses space during the truncate (thank you
6514          * very much snapshotting).
6515          *
6516          * And we need these to all be seperate.  The fact is we can use alot of
6517          * space doing the truncate, and we have no earthly idea how much space
6518          * we will use, so we need the truncate reservation to be seperate so it
6519          * doesn't end up using space reserved for updating the inode or
6520          * removing the orphan item.  We also need to be able to stop the
6521          * transaction and start a new one, which means we need to be able to
6522          * update the inode several times, and we have no idea of knowing how
6523          * many times that will be, so we can't just reserve 1 item for the
6524          * entirety of the opration, so that has to be done seperately as well.
6525          * Then there is the orphan item, which does indeed need to be held on
6526          * to for the whole operation, and we need nobody to touch this reserved
6527          * space except the orphan code.
6528          *
6529          * So that leaves us with
6530          *
6531          * 1) root->orphan_block_rsv - for the orphan deletion.
6532          * 2) rsv - for the truncate reservation, which we will steal from the
6533          * transaction reservation.
6534          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6535          * updating the inode.
6536          */
6537         rsv = btrfs_alloc_block_rsv(root);
6538         if (!rsv)
6539                 return -ENOMEM;
6540         btrfs_add_durable_block_rsv(root->fs_info, rsv);
6541
6542         trans = btrfs_start_transaction(root, 4);
6543         if (IS_ERR(trans)) {
6544                 err = PTR_ERR(trans);
6545                 goto out;
6546         }
6547
6548         /*
6549          * Reserve space for the truncate process.  Truncate should be adding
6550          * space, but if there are snapshots it may end up using space.
6551          */
6552         ret = btrfs_truncate_reserve_metadata(trans, root, rsv);
6553         BUG_ON(ret);
6554
6555         ret = btrfs_orphan_add(trans, inode);
6556         if (ret) {
6557                 btrfs_end_transaction(trans, root);
6558                 goto out;
6559         }
6560
6561         nr = trans->blocks_used;
6562         btrfs_end_transaction(trans, root);
6563         btrfs_btree_balance_dirty(root, nr);
6564
6565         /*
6566          * Ok so we've already migrated our bytes over for the truncate, so here
6567          * just reserve the one slot we need for updating the inode.
6568          */
6569         trans = btrfs_start_transaction(root, 1);
6570         if (IS_ERR(trans)) {
6571                 err = PTR_ERR(trans);
6572                 goto out;
6573         }
6574         trans->block_rsv = rsv;
6575
6576         /*
6577          * setattr is responsible for setting the ordered_data_close flag,
6578          * but that is only tested during the last file release.  That
6579          * could happen well after the next commit, leaving a great big
6580          * window where new writes may get lost if someone chooses to write
6581          * to this file after truncating to zero
6582          *
6583          * The inode doesn't have any dirty data here, and so if we commit
6584          * this is a noop.  If someone immediately starts writing to the inode
6585          * it is very likely we'll catch some of their writes in this
6586          * transaction, and the commit will find this file on the ordered
6587          * data list with good things to send down.
6588          *
6589          * This is a best effort solution, there is still a window where
6590          * using truncate to replace the contents of the file will
6591          * end up with a zero length file after a crash.
6592          */
6593         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6594                 btrfs_add_ordered_operation(trans, root, inode);
6595
6596         while (1) {
6597                 if (!trans) {
6598                         trans = btrfs_start_transaction(root, 3);
6599                         if (IS_ERR(trans)) {
6600                                 err = PTR_ERR(trans);
6601                                 goto out;
6602                         }
6603
6604                         ret = btrfs_truncate_reserve_metadata(trans, root,
6605                                                               rsv);
6606                         BUG_ON(ret);
6607
6608                         trans->block_rsv = rsv;
6609                 }
6610
6611                 ret = btrfs_truncate_inode_items(trans, root, inode,
6612                                                  inode->i_size,
6613                                                  BTRFS_EXTENT_DATA_KEY);
6614                 if (ret != -EAGAIN) {
6615                         err = ret;
6616                         break;
6617                 }
6618
6619                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6620                 ret = btrfs_update_inode(trans, root, inode);
6621                 if (ret) {
6622                         err = ret;
6623                         break;
6624                 }
6625
6626                 nr = trans->blocks_used;
6627                 btrfs_end_transaction(trans, root);
6628                 trans = NULL;
6629                 btrfs_btree_balance_dirty(root, nr);
6630         }
6631
6632         if (ret == 0 && inode->i_nlink > 0) {
6633                 trans->block_rsv = root->orphan_block_rsv;
6634                 ret = btrfs_orphan_del(trans, inode);
6635                 if (ret)
6636                         err = ret;
6637         } else if (ret && inode->i_nlink > 0) {
6638                 /*
6639                  * Failed to do the truncate, remove us from the in memory
6640                  * orphan list.
6641                  */
6642                 ret = btrfs_orphan_del(NULL, inode);
6643         }
6644
6645         trans->block_rsv = &root->fs_info->trans_block_rsv;
6646         ret = btrfs_update_inode(trans, root, inode);
6647         if (ret && !err)
6648                 err = ret;
6649
6650         nr = trans->blocks_used;
6651         ret = btrfs_end_transaction_throttle(trans, root);
6652         btrfs_btree_balance_dirty(root, nr);
6653
6654 out:
6655         btrfs_free_block_rsv(root, rsv);
6656
6657         if (ret && !err)
6658                 err = ret;
6659
6660         return err;
6661 }
6662
6663 /*
6664  * create a new subvolume directory/inode (helper for the ioctl).
6665  */
6666 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6667                              struct btrfs_root *new_root, u64 new_dirid)
6668 {
6669         struct inode *inode;
6670         int err;
6671         u64 index = 0;
6672
6673         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6674                                 new_dirid, S_IFDIR | 0700, &index);
6675         if (IS_ERR(inode))
6676                 return PTR_ERR(inode);
6677         inode->i_op = &btrfs_dir_inode_operations;
6678         inode->i_fop = &btrfs_dir_file_operations;
6679
6680         inode->i_nlink = 1;
6681         btrfs_i_size_write(inode, 0);
6682
6683         err = btrfs_update_inode(trans, new_root, inode);
6684         BUG_ON(err);
6685
6686         iput(inode);
6687         return 0;
6688 }
6689
6690 /* helper function for file defrag and space balancing.  This
6691  * forces readahead on a given range of bytes in an inode
6692  */
6693 unsigned long btrfs_force_ra(struct address_space *mapping,
6694                               struct file_ra_state *ra, struct file *file,
6695                               pgoff_t offset, pgoff_t last_index)
6696 {
6697         pgoff_t req_size = last_index - offset + 1;
6698
6699         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6700         return offset + req_size;
6701 }
6702
6703 struct inode *btrfs_alloc_inode(struct super_block *sb)
6704 {
6705         struct btrfs_inode *ei;
6706         struct inode *inode;
6707
6708         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6709         if (!ei)
6710                 return NULL;
6711
6712         ei->root = NULL;
6713         ei->space_info = NULL;
6714         ei->generation = 0;
6715         ei->sequence = 0;
6716         ei->last_trans = 0;
6717         ei->last_sub_trans = 0;
6718         ei->logged_trans = 0;
6719         ei->delalloc_bytes = 0;
6720         ei->reserved_bytes = 0;
6721         ei->disk_i_size = 0;
6722         ei->flags = 0;
6723         ei->index_cnt = (u64)-1;
6724         ei->last_unlink_trans = 0;
6725
6726         atomic_set(&ei->outstanding_extents, 0);
6727         atomic_set(&ei->reserved_extents, 0);
6728
6729         ei->ordered_data_close = 0;
6730         ei->orphan_meta_reserved = 0;
6731         ei->dummy_inode = 0;
6732         ei->in_defrag = 0;
6733         ei->force_compress = BTRFS_COMPRESS_NONE;
6734
6735         ei->delayed_node = NULL;
6736
6737         inode = &ei->vfs_inode;
6738         extent_map_tree_init(&ei->extent_tree);
6739         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6740         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6741         mutex_init(&ei->log_mutex);
6742         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6743         INIT_LIST_HEAD(&ei->i_orphan);
6744         INIT_LIST_HEAD(&ei->delalloc_inodes);
6745         INIT_LIST_HEAD(&ei->ordered_operations);
6746         RB_CLEAR_NODE(&ei->rb_node);
6747
6748         return inode;
6749 }
6750
6751 static void btrfs_i_callback(struct rcu_head *head)
6752 {
6753         struct inode *inode = container_of(head, struct inode, i_rcu);
6754         INIT_LIST_HEAD(&inode->i_dentry);
6755         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6756 }
6757
6758 void btrfs_destroy_inode(struct inode *inode)
6759 {
6760         struct btrfs_ordered_extent *ordered;
6761         struct btrfs_root *root = BTRFS_I(inode)->root;
6762
6763         WARN_ON(!list_empty(&inode->i_dentry));
6764         WARN_ON(inode->i_data.nrpages);
6765         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6766         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6767
6768         /*
6769          * This can happen where we create an inode, but somebody else also
6770          * created the same inode and we need to destroy the one we already
6771          * created.
6772          */
6773         if (!root)
6774                 goto free;
6775
6776         /*
6777          * Make sure we're properly removed from the ordered operation
6778          * lists.
6779          */
6780         smp_mb();
6781         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6782                 spin_lock(&root->fs_info->ordered_extent_lock);
6783                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6784                 spin_unlock(&root->fs_info->ordered_extent_lock);
6785         }
6786
6787         spin_lock(&root->orphan_lock);
6788         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6789                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6790                        (unsigned long long)btrfs_ino(inode));
6791                 list_del_init(&BTRFS_I(inode)->i_orphan);
6792         }
6793         spin_unlock(&root->orphan_lock);
6794
6795         while (1) {
6796                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6797                 if (!ordered)
6798                         break;
6799                 else {
6800                         printk(KERN_ERR "btrfs found ordered "
6801                                "extent %llu %llu on inode cleanup\n",
6802                                (unsigned long long)ordered->file_offset,
6803                                (unsigned long long)ordered->len);
6804                         btrfs_remove_ordered_extent(inode, ordered);
6805                         btrfs_put_ordered_extent(ordered);
6806                         btrfs_put_ordered_extent(ordered);
6807                 }
6808         }
6809         inode_tree_del(inode);
6810         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6811 free:
6812         btrfs_remove_delayed_node(inode);
6813         call_rcu(&inode->i_rcu, btrfs_i_callback);
6814 }
6815
6816 int btrfs_drop_inode(struct inode *inode)
6817 {
6818         struct btrfs_root *root = BTRFS_I(inode)->root;
6819
6820         if (btrfs_root_refs(&root->root_item) == 0 &&
6821             !is_free_space_inode(root, inode))
6822                 return 1;
6823         else
6824                 return generic_drop_inode(inode);
6825 }
6826
6827 static void init_once(void *foo)
6828 {
6829         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6830
6831         inode_init_once(&ei->vfs_inode);
6832 }
6833
6834 void btrfs_destroy_cachep(void)
6835 {
6836         if (btrfs_inode_cachep)
6837                 kmem_cache_destroy(btrfs_inode_cachep);
6838         if (btrfs_trans_handle_cachep)
6839                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6840         if (btrfs_transaction_cachep)
6841                 kmem_cache_destroy(btrfs_transaction_cachep);
6842         if (btrfs_path_cachep)
6843                 kmem_cache_destroy(btrfs_path_cachep);
6844         if (btrfs_free_space_cachep)
6845                 kmem_cache_destroy(btrfs_free_space_cachep);
6846 }
6847
6848 int btrfs_init_cachep(void)
6849 {
6850         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6851                         sizeof(struct btrfs_inode), 0,
6852                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6853         if (!btrfs_inode_cachep)
6854                 goto fail;
6855
6856         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6857                         sizeof(struct btrfs_trans_handle), 0,
6858                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6859         if (!btrfs_trans_handle_cachep)
6860                 goto fail;
6861
6862         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6863                         sizeof(struct btrfs_transaction), 0,
6864                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6865         if (!btrfs_transaction_cachep)
6866                 goto fail;
6867
6868         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6869                         sizeof(struct btrfs_path), 0,
6870                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6871         if (!btrfs_path_cachep)
6872                 goto fail;
6873
6874         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6875                         sizeof(struct btrfs_free_space), 0,
6876                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6877         if (!btrfs_free_space_cachep)
6878                 goto fail;
6879
6880         return 0;
6881 fail:
6882         btrfs_destroy_cachep();
6883         return -ENOMEM;
6884 }
6885
6886 static int btrfs_getattr(struct vfsmount *mnt,
6887                          struct dentry *dentry, struct kstat *stat)
6888 {
6889         struct inode *inode = dentry->d_inode;
6890         generic_fillattr(inode, stat);
6891         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6892         stat->blksize = PAGE_CACHE_SIZE;
6893         stat->blocks = (inode_get_bytes(inode) +
6894                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6895         return 0;
6896 }
6897
6898 /*
6899  * If a file is moved, it will inherit the cow and compression flags of the new
6900  * directory.
6901  */
6902 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6903 {
6904         struct btrfs_inode *b_dir = BTRFS_I(dir);
6905         struct btrfs_inode *b_inode = BTRFS_I(inode);
6906
6907         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6908                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6909         else
6910                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6911
6912         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6913                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6914         else
6915                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6916 }
6917
6918 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6919                            struct inode *new_dir, struct dentry *new_dentry)
6920 {
6921         struct btrfs_trans_handle *trans;
6922         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6923         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6924         struct inode *new_inode = new_dentry->d_inode;
6925         struct inode *old_inode = old_dentry->d_inode;
6926         struct timespec ctime = CURRENT_TIME;
6927         u64 index = 0;
6928         u64 root_objectid;
6929         int ret;
6930         u64 old_ino = btrfs_ino(old_inode);
6931
6932         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6933                 return -EPERM;
6934
6935         /* we only allow rename subvolume link between subvolumes */
6936         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6937                 return -EXDEV;
6938
6939         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6940             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6941                 return -ENOTEMPTY;
6942
6943         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6944             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6945                 return -ENOTEMPTY;
6946         /*
6947          * we're using rename to replace one file with another.
6948          * and the replacement file is large.  Start IO on it now so
6949          * we don't add too much work to the end of the transaction
6950          */
6951         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6952             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6953                 filemap_flush(old_inode->i_mapping);
6954
6955         /* close the racy window with snapshot create/destroy ioctl */
6956         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6957                 down_read(&root->fs_info->subvol_sem);
6958         /*
6959          * We want to reserve the absolute worst case amount of items.  So if
6960          * both inodes are subvols and we need to unlink them then that would
6961          * require 4 item modifications, but if they are both normal inodes it
6962          * would require 5 item modifications, so we'll assume their normal
6963          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6964          * should cover the worst case number of items we'll modify.
6965          */
6966         trans = btrfs_start_transaction(root, 20);
6967         if (IS_ERR(trans)) {
6968                 ret = PTR_ERR(trans);
6969                 goto out_notrans;
6970         }
6971
6972         if (dest != root)
6973                 btrfs_record_root_in_trans(trans, dest);
6974
6975         ret = btrfs_set_inode_index(new_dir, &index);
6976         if (ret)
6977                 goto out_fail;
6978
6979         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6980                 /* force full log commit if subvolume involved. */
6981                 root->fs_info->last_trans_log_full_commit = trans->transid;
6982         } else {
6983                 ret = btrfs_insert_inode_ref(trans, dest,
6984                                              new_dentry->d_name.name,
6985                                              new_dentry->d_name.len,
6986                                              old_ino,
6987                                              btrfs_ino(new_dir), index);
6988                 if (ret)
6989                         goto out_fail;
6990                 /*
6991                  * this is an ugly little race, but the rename is required
6992                  * to make sure that if we crash, the inode is either at the
6993                  * old name or the new one.  pinning the log transaction lets
6994                  * us make sure we don't allow a log commit to come in after
6995                  * we unlink the name but before we add the new name back in.
6996                  */
6997                 btrfs_pin_log_trans(root);
6998         }
6999         /*
7000          * make sure the inode gets flushed if it is replacing
7001          * something.
7002          */
7003         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7004                 btrfs_add_ordered_operation(trans, root, old_inode);
7005
7006         old_dir->i_ctime = old_dir->i_mtime = ctime;
7007         new_dir->i_ctime = new_dir->i_mtime = ctime;
7008         old_inode->i_ctime = ctime;
7009
7010         if (old_dentry->d_parent != new_dentry->d_parent)
7011                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7012
7013         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7014                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7015                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7016                                         old_dentry->d_name.name,
7017                                         old_dentry->d_name.len);
7018         } else {
7019                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7020                                         old_dentry->d_inode,
7021                                         old_dentry->d_name.name,
7022                                         old_dentry->d_name.len);
7023                 if (!ret)
7024                         ret = btrfs_update_inode(trans, root, old_inode);
7025         }
7026         BUG_ON(ret);
7027
7028         if (new_inode) {
7029                 new_inode->i_ctime = CURRENT_TIME;
7030                 if (unlikely(btrfs_ino(new_inode) ==
7031                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7032                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7033                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7034                                                 root_objectid,
7035                                                 new_dentry->d_name.name,
7036                                                 new_dentry->d_name.len);
7037                         BUG_ON(new_inode->i_nlink == 0);
7038                 } else {
7039                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7040                                                  new_dentry->d_inode,
7041                                                  new_dentry->d_name.name,
7042                                                  new_dentry->d_name.len);
7043                 }
7044                 BUG_ON(ret);
7045                 if (new_inode->i_nlink == 0) {
7046                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7047                         BUG_ON(ret);
7048                 }
7049         }
7050
7051         fixup_inode_flags(new_dir, old_inode);
7052
7053         ret = btrfs_add_link(trans, new_dir, old_inode,
7054                              new_dentry->d_name.name,
7055                              new_dentry->d_name.len, 0, index);
7056         BUG_ON(ret);
7057
7058         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7059                 struct dentry *parent = dget_parent(new_dentry);
7060                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7061                 dput(parent);
7062                 btrfs_end_log_trans(root);
7063         }
7064 out_fail:
7065         btrfs_end_transaction_throttle(trans, root);
7066 out_notrans:
7067         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7068                 up_read(&root->fs_info->subvol_sem);
7069
7070         return ret;
7071 }
7072
7073 /*
7074  * some fairly slow code that needs optimization. This walks the list
7075  * of all the inodes with pending delalloc and forces them to disk.
7076  */
7077 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7078 {
7079         struct list_head *head = &root->fs_info->delalloc_inodes;
7080         struct btrfs_inode *binode;
7081         struct inode *inode;
7082
7083         if (root->fs_info->sb->s_flags & MS_RDONLY)
7084                 return -EROFS;
7085
7086         spin_lock(&root->fs_info->delalloc_lock);
7087         while (!list_empty(head)) {
7088                 binode = list_entry(head->next, struct btrfs_inode,
7089                                     delalloc_inodes);
7090                 inode = igrab(&binode->vfs_inode);
7091                 if (!inode)
7092                         list_del_init(&binode->delalloc_inodes);
7093                 spin_unlock(&root->fs_info->delalloc_lock);
7094                 if (inode) {
7095                         filemap_flush(inode->i_mapping);
7096                         if (delay_iput)
7097                                 btrfs_add_delayed_iput(inode);
7098                         else
7099                                 iput(inode);
7100                 }
7101                 cond_resched();
7102                 spin_lock(&root->fs_info->delalloc_lock);
7103         }
7104         spin_unlock(&root->fs_info->delalloc_lock);
7105
7106         /* the filemap_flush will queue IO into the worker threads, but
7107          * we have to make sure the IO is actually started and that
7108          * ordered extents get created before we return
7109          */
7110         atomic_inc(&root->fs_info->async_submit_draining);
7111         while (atomic_read(&root->fs_info->nr_async_submits) ||
7112               atomic_read(&root->fs_info->async_delalloc_pages)) {
7113                 wait_event(root->fs_info->async_submit_wait,
7114                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7115                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7116         }
7117         atomic_dec(&root->fs_info->async_submit_draining);
7118         return 0;
7119 }
7120
7121 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7122                          const char *symname)
7123 {
7124         struct btrfs_trans_handle *trans;
7125         struct btrfs_root *root = BTRFS_I(dir)->root;
7126         struct btrfs_path *path;
7127         struct btrfs_key key;
7128         struct inode *inode = NULL;
7129         int err;
7130         int drop_inode = 0;
7131         u64 objectid;
7132         u64 index = 0 ;
7133         int name_len;
7134         int datasize;
7135         unsigned long ptr;
7136         struct btrfs_file_extent_item *ei;
7137         struct extent_buffer *leaf;
7138         unsigned long nr = 0;
7139
7140         name_len = strlen(symname) + 1;
7141         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7142                 return -ENAMETOOLONG;
7143
7144         /*
7145          * 2 items for inode item and ref
7146          * 2 items for dir items
7147          * 1 item for xattr if selinux is on
7148          */
7149         trans = btrfs_start_transaction(root, 5);
7150         if (IS_ERR(trans))
7151                 return PTR_ERR(trans);
7152
7153         err = btrfs_find_free_ino(root, &objectid);
7154         if (err)
7155                 goto out_unlock;
7156
7157         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7158                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7159                                 S_IFLNK|S_IRWXUGO, &index);
7160         if (IS_ERR(inode)) {
7161                 err = PTR_ERR(inode);
7162                 goto out_unlock;
7163         }
7164
7165         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7166         if (err) {
7167                 drop_inode = 1;
7168                 goto out_unlock;
7169         }
7170
7171         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7172         if (err)
7173                 drop_inode = 1;
7174         else {
7175                 inode->i_mapping->a_ops = &btrfs_aops;
7176                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7177                 inode->i_fop = &btrfs_file_operations;
7178                 inode->i_op = &btrfs_file_inode_operations;
7179                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7180         }
7181         if (drop_inode)
7182                 goto out_unlock;
7183
7184         path = btrfs_alloc_path();
7185         BUG_ON(!path);
7186         key.objectid = btrfs_ino(inode);
7187         key.offset = 0;
7188         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7189         datasize = btrfs_file_extent_calc_inline_size(name_len);
7190         err = btrfs_insert_empty_item(trans, root, path, &key,
7191                                       datasize);
7192         if (err) {
7193                 drop_inode = 1;
7194                 btrfs_free_path(path);
7195                 goto out_unlock;
7196         }
7197         leaf = path->nodes[0];
7198         ei = btrfs_item_ptr(leaf, path->slots[0],
7199                             struct btrfs_file_extent_item);
7200         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7201         btrfs_set_file_extent_type(leaf, ei,
7202                                    BTRFS_FILE_EXTENT_INLINE);
7203         btrfs_set_file_extent_encryption(leaf, ei, 0);
7204         btrfs_set_file_extent_compression(leaf, ei, 0);
7205         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7206         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7207
7208         ptr = btrfs_file_extent_inline_start(ei);
7209         write_extent_buffer(leaf, symname, ptr, name_len);
7210         btrfs_mark_buffer_dirty(leaf);
7211         btrfs_free_path(path);
7212
7213         inode->i_op = &btrfs_symlink_inode_operations;
7214         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7215         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7216         inode_set_bytes(inode, name_len);
7217         btrfs_i_size_write(inode, name_len - 1);
7218         err = btrfs_update_inode(trans, root, inode);
7219         if (err)
7220                 drop_inode = 1;
7221
7222 out_unlock:
7223         nr = trans->blocks_used;
7224         btrfs_end_transaction_throttle(trans, root);
7225         if (drop_inode) {
7226                 inode_dec_link_count(inode);
7227                 iput(inode);
7228         }
7229         btrfs_btree_balance_dirty(root, nr);
7230         return err;
7231 }
7232
7233 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7234                                        u64 start, u64 num_bytes, u64 min_size,
7235                                        loff_t actual_len, u64 *alloc_hint,
7236                                        struct btrfs_trans_handle *trans)
7237 {
7238         struct btrfs_root *root = BTRFS_I(inode)->root;
7239         struct btrfs_key ins;
7240         u64 cur_offset = start;
7241         u64 i_size;
7242         int ret = 0;
7243         bool own_trans = true;
7244
7245         if (trans)
7246                 own_trans = false;
7247         while (num_bytes > 0) {
7248                 if (own_trans) {
7249                         trans = btrfs_start_transaction(root, 3);
7250                         if (IS_ERR(trans)) {
7251                                 ret = PTR_ERR(trans);
7252                                 break;
7253                         }
7254                 }
7255
7256                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7257                                            0, *alloc_hint, (u64)-1, &ins, 1);
7258                 if (ret) {
7259                         if (own_trans)
7260                                 btrfs_end_transaction(trans, root);
7261                         break;
7262                 }
7263
7264                 ret = insert_reserved_file_extent(trans, inode,
7265                                                   cur_offset, ins.objectid,
7266                                                   ins.offset, ins.offset,
7267                                                   ins.offset, 0, 0, 0,
7268                                                   BTRFS_FILE_EXTENT_PREALLOC);
7269                 BUG_ON(ret);
7270                 btrfs_drop_extent_cache(inode, cur_offset,
7271                                         cur_offset + ins.offset -1, 0);
7272
7273                 num_bytes -= ins.offset;
7274                 cur_offset += ins.offset;
7275                 *alloc_hint = ins.objectid + ins.offset;
7276
7277                 inode->i_ctime = CURRENT_TIME;
7278                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7279                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7280                     (actual_len > inode->i_size) &&
7281                     (cur_offset > inode->i_size)) {
7282                         if (cur_offset > actual_len)
7283                                 i_size = actual_len;
7284                         else
7285                                 i_size = cur_offset;
7286                         i_size_write(inode, i_size);
7287                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7288                 }
7289
7290                 ret = btrfs_update_inode(trans, root, inode);
7291                 BUG_ON(ret);
7292
7293                 if (own_trans)
7294                         btrfs_end_transaction(trans, root);
7295         }
7296         return ret;
7297 }
7298
7299 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7300                               u64 start, u64 num_bytes, u64 min_size,
7301                               loff_t actual_len, u64 *alloc_hint)
7302 {
7303         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7304                                            min_size, actual_len, alloc_hint,
7305                                            NULL);
7306 }
7307
7308 int btrfs_prealloc_file_range_trans(struct inode *inode,
7309                                     struct btrfs_trans_handle *trans, int mode,
7310                                     u64 start, u64 num_bytes, u64 min_size,
7311                                     loff_t actual_len, u64 *alloc_hint)
7312 {
7313         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7314                                            min_size, actual_len, alloc_hint, trans);
7315 }
7316
7317 static int btrfs_set_page_dirty(struct page *page)
7318 {
7319         return __set_page_dirty_nobuffers(page);
7320 }
7321
7322 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7323 {
7324         struct btrfs_root *root = BTRFS_I(inode)->root;
7325
7326         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7327                 return -EROFS;
7328         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7329                 return -EACCES;
7330         return generic_permission(inode, mask, flags, btrfs_check_acl);
7331 }
7332
7333 static const struct inode_operations btrfs_dir_inode_operations = {
7334         .getattr        = btrfs_getattr,
7335         .lookup         = btrfs_lookup,
7336         .create         = btrfs_create,
7337         .unlink         = btrfs_unlink,
7338         .link           = btrfs_link,
7339         .mkdir          = btrfs_mkdir,
7340         .rmdir          = btrfs_rmdir,
7341         .rename         = btrfs_rename,
7342         .symlink        = btrfs_symlink,
7343         .setattr        = btrfs_setattr,
7344         .mknod          = btrfs_mknod,
7345         .setxattr       = btrfs_setxattr,
7346         .getxattr       = btrfs_getxattr,
7347         .listxattr      = btrfs_listxattr,
7348         .removexattr    = btrfs_removexattr,
7349         .permission     = btrfs_permission,
7350 };
7351 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7352         .lookup         = btrfs_lookup,
7353         .permission     = btrfs_permission,
7354 };
7355
7356 static const struct file_operations btrfs_dir_file_operations = {
7357         .llseek         = generic_file_llseek,
7358         .read           = generic_read_dir,
7359         .readdir        = btrfs_real_readdir,
7360         .unlocked_ioctl = btrfs_ioctl,
7361 #ifdef CONFIG_COMPAT
7362         .compat_ioctl   = btrfs_ioctl,
7363 #endif
7364         .release        = btrfs_release_file,
7365         .fsync          = btrfs_sync_file,
7366 };
7367
7368 static struct extent_io_ops btrfs_extent_io_ops = {
7369         .fill_delalloc = run_delalloc_range,
7370         .submit_bio_hook = btrfs_submit_bio_hook,
7371         .merge_bio_hook = btrfs_merge_bio_hook,
7372         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7373         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7374         .writepage_start_hook = btrfs_writepage_start_hook,
7375         .readpage_io_failed_hook = btrfs_io_failed_hook,
7376         .set_bit_hook = btrfs_set_bit_hook,
7377         .clear_bit_hook = btrfs_clear_bit_hook,
7378         .merge_extent_hook = btrfs_merge_extent_hook,
7379         .split_extent_hook = btrfs_split_extent_hook,
7380 };
7381
7382 /*
7383  * btrfs doesn't support the bmap operation because swapfiles
7384  * use bmap to make a mapping of extents in the file.  They assume
7385  * these extents won't change over the life of the file and they
7386  * use the bmap result to do IO directly to the drive.
7387  *
7388  * the btrfs bmap call would return logical addresses that aren't
7389  * suitable for IO and they also will change frequently as COW
7390  * operations happen.  So, swapfile + btrfs == corruption.
7391  *
7392  * For now we're avoiding this by dropping bmap.
7393  */
7394 static const struct address_space_operations btrfs_aops = {
7395         .readpage       = btrfs_readpage,
7396         .writepage      = btrfs_writepage,
7397         .writepages     = btrfs_writepages,
7398         .readpages      = btrfs_readpages,
7399         .direct_IO      = btrfs_direct_IO,
7400         .invalidatepage = btrfs_invalidatepage,
7401         .releasepage    = btrfs_releasepage,
7402         .set_page_dirty = btrfs_set_page_dirty,
7403         .error_remove_page = generic_error_remove_page,
7404 };
7405
7406 static const struct address_space_operations btrfs_symlink_aops = {
7407         .readpage       = btrfs_readpage,
7408         .writepage      = btrfs_writepage,
7409         .invalidatepage = btrfs_invalidatepage,
7410         .releasepage    = btrfs_releasepage,
7411 };
7412
7413 static const struct inode_operations btrfs_file_inode_operations = {
7414         .getattr        = btrfs_getattr,
7415         .setattr        = btrfs_setattr,
7416         .setxattr       = btrfs_setxattr,
7417         .getxattr       = btrfs_getxattr,
7418         .listxattr      = btrfs_listxattr,
7419         .removexattr    = btrfs_removexattr,
7420         .permission     = btrfs_permission,
7421         .fiemap         = btrfs_fiemap,
7422 };
7423 static const struct inode_operations btrfs_special_inode_operations = {
7424         .getattr        = btrfs_getattr,
7425         .setattr        = btrfs_setattr,
7426         .permission     = btrfs_permission,
7427         .setxattr       = btrfs_setxattr,
7428         .getxattr       = btrfs_getxattr,
7429         .listxattr      = btrfs_listxattr,
7430         .removexattr    = btrfs_removexattr,
7431 };
7432 static const struct inode_operations btrfs_symlink_inode_operations = {
7433         .readlink       = generic_readlink,
7434         .follow_link    = page_follow_link_light,
7435         .put_link       = page_put_link,
7436         .getattr        = btrfs_getattr,
7437         .permission     = btrfs_permission,
7438         .setxattr       = btrfs_setxattr,
7439         .getxattr       = btrfs_getxattr,
7440         .listxattr      = btrfs_listxattr,
7441         .removexattr    = btrfs_removexattr,
7442 };
7443
7444 const struct dentry_operations btrfs_dentry_operations = {
7445         .d_delete       = btrfs_dentry_delete,
7446 };