fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 BUG_ON(!pages);
 397
 398                 if (BTRFS_I(inode)->force_compress)
 399                         compress_type = BTRFS_I(inode)->force_compress;
 400
 401                 ret = btrfs_compress_pages(compress_type,
 402                                            inode->i_mapping, start,
 403                                            total_compressed, pages,
 404                                            nr_pages, &nr_pages_ret,
 405                                            &total_in,
 406                                            &total_compressed,
 407                                            max_compressed);
 408
 409                 if (!ret) {
 410                         unsigned long offset = total_compressed &
 411                                 (PAGE_CACHE_SIZE - 1);
 412                         struct page *page = pages[nr_pages_ret - 1];
 413                         char *kaddr;
 414
 415                         /* zero the tail end of the last page, we might be
 416                          * sending it down to disk
 417                          */
 418                         if (offset) {
 419                                 kaddr = kmap_atomic(page, KM_USER0);
 420                                 memset(kaddr + offset, 0,
 421                                        PAGE_CACHE_SIZE - offset);
 422                                 kunmap_atomic(kaddr, KM_USER0);
 423                         }
 424                         will_compress = 1;
 425                 }
 426         }
 427         if (start == 0) {
 428                 trans = btrfs_join_transaction(root);
 429                 BUG_ON(IS_ERR(trans));
 430                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 431
 432                 /* lets try to make an inline extent */
 433                 if (ret || total_in < (actual_end - start)) {
 434                         /* we didn't compress the entire range, try
 435                          * to make an uncompressed inline extent.
 436                          */
 437                         ret = cow_file_range_inline(trans, root, inode,
 438                                                     start, end, 0, 0, NULL);
 439                 } else {
 440                         /* try making a compressed inline extent */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end,
 443                                                     total_compressed,
 444                                                     compress_type, pages);
 445                 }
 446                 if (ret == 0) {
 447                         /*
 448                          * inline extent creation worked, we don't need
 449                          * to create any more async work items.  Unlock
 450                          * and free up our temp pages.
 451                          */
 452                         extent_clear_unlock_delalloc(inode,
 453                              &BTRFS_I(inode)->io_tree,
 454                              start, end, NULL,
 455                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 456                              EXTENT_CLEAR_DELALLOC |
 457                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 458
 459                         btrfs_end_transaction(trans, root);
 460                         goto free_pages_out;
 461                 }
 462                 btrfs_end_transaction(trans, root);
 463         }
 464
 465         if (will_compress) {
 466                 /*
 467                  * we aren't doing an inline extent round the compressed size
 468                  * up to a block size boundary so the allocator does sane
 469                  * things
 470                  */
 471                 total_compressed = (total_compressed + blocksize - 1) &
 472                         ~(blocksize - 1);
 473
 474                 /*
 475                  * one last check to make sure the compression is really a
 476                  * win, compare the page count read with the blocks on disk
 477                  */
 478                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 479                         ~(PAGE_CACHE_SIZE - 1);
 480                 if (total_compressed >= total_in) {
 481                         will_compress = 0;
 482                 } else {
 483                         num_bytes = total_in;
 484                 }
 485         }
 486         if (!will_compress && pages) {
 487                 /*
 488                  * the compression code ran but failed to make things smaller,
 489                  * free any pages it allocated and our page pointer array
 490                  */
 491                 for (i = 0; i < nr_pages_ret; i++) {
 492                         WARN_ON(pages[i]->mapping);
 493                         page_cache_release(pages[i]);
 494                 }
 495                 kfree(pages);
 496                 pages = NULL;
 497                 total_compressed = 0;
 498                 nr_pages_ret = 0;
 499
 500                 /* flag the file so we don't compress in the future */
 501                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 502                     !(BTRFS_I(inode)->force_compress)) {
 503                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 504                 }
 505         }
 506         if (will_compress) {
 507                 *num_added += 1;
 508
 509                 /* the async work queues will take care of doing actual
 510                  * allocation on disk for these compressed pages,
 511                  * and will submit them to the elevator.
 512                  */
 513                 add_async_extent(async_cow, start, num_bytes,
 514                                  total_compressed, pages, nr_pages_ret,
 515                                  compress_type);
 516
 517                 if (start + num_bytes < end) {
 518                         start += num_bytes;
 519                         pages = NULL;
 520                         cond_resched();
 521                         goto again;
 522                 }
 523         } else {
 524 cleanup_and_bail_uncompressed:
 525                 /*
 526                  * No compression, but we still need to write the pages in
 527                  * the file we've been given so far.  redirty the locked
 528                  * page if it corresponds to our extent and set things up
 529                  * for the async work queue to run cow_file_range to do
 530                  * the normal delalloc dance
 531                  */
 532                 if (page_offset(locked_page) >= start &&
 533                     page_offset(locked_page) <= end) {
 534                         __set_page_dirty_nobuffers(locked_page);
 535                         /* unlocked later on in the async handlers */
 536                 }
 537                 add_async_extent(async_cow, start, end - start + 1,
 538                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 539                 *num_added += 1;
 540         }
 541
 542 out:
 543         return 0;
 544
 545 free_pages_out:
 546         for (i = 0; i < nr_pages_ret; i++) {
 547                 WARN_ON(pages[i]->mapping);
 548                 page_cache_release(pages[i]);
 549         }
 550         kfree(pages);
 551
 552         goto out;
 553 }
 554
 555 /*
 556  * phase two of compressed writeback.  This is the ordered portion
 557  * of the code, which only gets called in the order the work was
 558  * queued.  We walk all the async extents created by compress_file_range
 559  * and send them down to the disk.
 560  */
 561 static noinline int submit_compressed_extents(struct inode *inode,
 562                                               struct async_cow *async_cow)
 563 {
 564         struct async_extent *async_extent;
 565         u64 alloc_hint = 0;
 566         struct btrfs_trans_handle *trans;
 567         struct btrfs_key ins;
 568         struct extent_map *em;
 569         struct btrfs_root *root = BTRFS_I(inode)->root;
 570         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 571         struct extent_io_tree *io_tree;
 572         int ret = 0;
 573
 574         if (list_empty(&async_cow->extents))
 575                 return 0;
 576
 577
 578         while (!list_empty(&async_cow->extents)) {
 579                 async_extent = list_entry(async_cow->extents.next,
 580                                           struct async_extent, list);
 581                 list_del(&async_extent->list);
 582
 583                 io_tree = &BTRFS_I(inode)->io_tree;
 584
 585 retry:
 586                 /* did the compression code fall back to uncompressed IO? */
 587                 if (!async_extent->pages) {
 588                         int page_started = 0;
 589                         unsigned long nr_written = 0;
 590
 591                         lock_extent(io_tree, async_extent->start,
 592                                          async_extent->start +
 593                                          async_extent->ram_size - 1, GFP_NOFS);
 594
 595                         /* allocate blocks */
 596                         ret = cow_file_range(inode, async_cow->locked_page,
 597                                              async_extent->start,
 598                                              async_extent->start +
 599                                              async_extent->ram_size - 1,
 600                                              &page_started, &nr_written, 0);
 601
 602                         /*
 603                          * if page_started, cow_file_range inserted an
 604                          * inline extent and took care of all the unlocking
 605                          * and IO for us.  Otherwise, we need to submit
 606                          * all those pages down to the drive.
 607                          */
 608                         if (!page_started && !ret)
 609                                 extent_write_locked_range(io_tree,
 610                                                   inode, async_extent->start,
 611                                                   async_extent->start +
 612                                                   async_extent->ram_size - 1,
 613                                                   btrfs_get_extent,
 614                                                   WB_SYNC_ALL);
 615                         kfree(async_extent);
 616                         cond_resched();
 617                         continue;
 618                 }
 619
 620                 lock_extent(io_tree, async_extent->start,
 621                             async_extent->start + async_extent->ram_size - 1,
 622                             GFP_NOFS);
 623
 624                 trans = btrfs_join_transaction(root);
 625                 BUG_ON(IS_ERR(trans));
 626                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 627                 ret = btrfs_reserve_extent(trans, root,
 628                                            async_extent->compressed_size,
 629                                            async_extent->compressed_size,
 630                                            0, alloc_hint,
 631                                            (u64)-1, &ins, 1);
 632                 btrfs_end_transaction(trans, root);
 633
 634                 if (ret) {
 635                         int i;
 636                         for (i = 0; i < async_extent->nr_pages; i++) {
 637                                 WARN_ON(async_extent->pages[i]->mapping);
 638                                 page_cache_release(async_extent->pages[i]);
 639                         }
 640                         kfree(async_extent->pages);
 641                         async_extent->nr_pages = 0;
 642                         async_extent->pages = NULL;
 643                         unlock_extent(io_tree, async_extent->start,
 644                                       async_extent->start +
 645                                       async_extent->ram_size - 1, GFP_NOFS);
 646                         goto retry;
 647                 }
 648
 649                 /*
 650                  * here we're doing allocation and writeback of the
 651                  * compressed pages
 652                  */
 653                 btrfs_drop_extent_cache(inode, async_extent->start,
 654                                         async_extent->start +
 655                                         async_extent->ram_size - 1, 0);
 656
 657                 em = alloc_extent_map();
 658                 BUG_ON(!em);
 659                 em->start = async_extent->start;
 660                 em->len = async_extent->ram_size;
 661                 em->orig_start = em->start;
 662
 663                 em->block_start = ins.objectid;
 664                 em->block_len = ins.offset;
 665                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 666                 em->compress_type = async_extent->compress_type;
 667                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 668                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 669
 670                 while (1) {
 671                         write_lock(&em_tree->lock);
 672                         ret = add_extent_mapping(em_tree, em);
 673                         write_unlock(&em_tree->lock);
 674                         if (ret != -EEXIST) {
 675                                 free_extent_map(em);
 676                                 break;
 677                         }
 678                         btrfs_drop_extent_cache(inode, async_extent->start,
 679                                                 async_extent->start +
 680                                                 async_extent->ram_size - 1, 0);
 681                 }
 682
 683                 ret = btrfs_add_ordered_extent_compress(inode,
 684                                                 async_extent->start,
 685                                                 ins.objectid,
 686                                                 async_extent->ram_size,
 687                                                 ins.offset,
 688                                                 BTRFS_ORDERED_COMPRESSED,
 689                                                 async_extent->compress_type);
 690                 BUG_ON(ret);
 691
 692                 /*
 693                  * clear dirty, set writeback and unlock the pages.
 694                  */
 695                 extent_clear_unlock_delalloc(inode,
 696                                 &BTRFS_I(inode)->io_tree,
 697                                 async_extent->start,
 698                                 async_extent->start +
 699                                 async_extent->ram_size - 1,
 700                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 701                                 EXTENT_CLEAR_UNLOCK |
 702                                 EXTENT_CLEAR_DELALLOC |
 703                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 704
 705                 ret = btrfs_submit_compressed_write(inode,
 706                                     async_extent->start,
 707                                     async_extent->ram_size,
 708                                     ins.objectid,
 709                                     ins.offset, async_extent->pages,
 710                                     async_extent->nr_pages);
 711
 712                 BUG_ON(ret);
 713                 alloc_hint = ins.objectid + ins.offset;
 714                 kfree(async_extent);
 715                 cond_resched();
 716         }
 717
 718         return 0;
 719 }
 720
 721 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 722                                       u64 num_bytes)
 723 {
 724         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 725         struct extent_map *em;
 726         u64 alloc_hint = 0;
 727
 728         read_lock(&em_tree->lock);
 729         em = search_extent_mapping(em_tree, start, num_bytes);
 730         if (em) {
 731                 /*
 732                  * if block start isn't an actual block number then find the
 733                  * first block in this inode and use that as a hint.  If that
 734                  * block is also bogus then just don't worry about it.
 735                  */
 736                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 737                         free_extent_map(em);
 738                         em = search_extent_mapping(em_tree, 0, 0);
 739                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 740                                 alloc_hint = em->block_start;
 741                         if (em)
 742                                 free_extent_map(em);
 743                 } else {
 744                         alloc_hint = em->block_start;
 745                         free_extent_map(em);
 746                 }
 747         }
 748         read_unlock(&em_tree->lock);
 749
 750         return alloc_hint;
 751 }
 752
 753 /*
 754  * when extent_io.c finds a delayed allocation range in the file,
 755  * the call backs end up in this code.  The basic idea is to
 756  * allocate extents on disk for the range, and create ordered data structs
 757  * in ram to track those extents.
 758  *
 759  * locked_page is the page that writepage had locked already.  We use
 760  * it to make sure we don't do extra locks or unlocks.
 761  *
 762  * *page_started is set to one if we unlock locked_page and do everything
 763  * required to start IO on it.  It may be clean and already done with
 764  * IO when we return.
 765  */
 766 static noinline int cow_file_range(struct inode *inode,
 767                                    struct page *locked_page,
 768                                    u64 start, u64 end, int *page_started,
 769                                    unsigned long *nr_written,
 770                                    int unlock)
 771 {
 772         struct btrfs_root *root = BTRFS_I(inode)->root;
 773         struct btrfs_trans_handle *trans;
 774         u64 alloc_hint = 0;
 775         u64 num_bytes;
 776         unsigned long ram_size;
 777         u64 disk_num_bytes;
 778         u64 cur_alloc_size;
 779         u64 blocksize = root->sectorsize;
 780         struct btrfs_key ins;
 781         struct extent_map *em;
 782         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 783         int ret = 0;
 784
 785         BUG_ON(btrfs_is_free_space_inode(root, inode));
 786         trans = btrfs_join_transaction(root);
 787         BUG_ON(IS_ERR(trans));
 788         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 789
 790         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 791         num_bytes = max(blocksize,  num_bytes);
 792         disk_num_bytes = num_bytes;
 793         ret = 0;
 794
 795         /* if this is a small write inside eof, kick off defrag */
 796         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 797                 btrfs_add_inode_defrag(trans, inode);
 798
 799         if (start == 0) {
 800                 /* lets try to make an inline extent */
 801                 ret = cow_file_range_inline(trans, root, inode,
 802                                             start, end, 0, 0, NULL);
 803                 if (ret == 0) {
 804                         extent_clear_unlock_delalloc(inode,
 805                                      &BTRFS_I(inode)->io_tree,
 806                                      start, end, NULL,
 807                                      EXTENT_CLEAR_UNLOCK_PAGE |
 808                                      EXTENT_CLEAR_UNLOCK |
 809                                      EXTENT_CLEAR_DELALLOC |
 810                                      EXTENT_CLEAR_DIRTY |
 811                                      EXTENT_SET_WRITEBACK |
 812                                      EXTENT_END_WRITEBACK);
 813
 814                         *nr_written = *nr_written +
 815                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 816                         *page_started = 1;
 817                         ret = 0;
 818                         goto out;
 819                 }
 820         }
 821
 822         BUG_ON(disk_num_bytes >
 823                btrfs_super_total_bytes(&root->fs_info->super_copy));
 824
 825         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 826         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 827
 828         while (disk_num_bytes > 0) {
 829                 unsigned long op;
 830
 831                 cur_alloc_size = disk_num_bytes;
 832                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 833                                            root->sectorsize, 0, alloc_hint,
 834                                            (u64)-1, &ins, 1);
 835                 BUG_ON(ret);
 836
 837                 em = alloc_extent_map();
 838                 BUG_ON(!em);
 839                 em->start = start;
 840                 em->orig_start = em->start;
 841                 ram_size = ins.offset;
 842                 em->len = ins.offset;
 843
 844                 em->block_start = ins.objectid;
 845                 em->block_len = ins.offset;
 846                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 847                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 848
 849                 while (1) {
 850                         write_lock(&em_tree->lock);
 851                         ret = add_extent_mapping(em_tree, em);
 852                         write_unlock(&em_tree->lock);
 853                         if (ret != -EEXIST) {
 854                                 free_extent_map(em);
 855                                 break;
 856                         }
 857                         btrfs_drop_extent_cache(inode, start,
 858                                                 start + ram_size - 1, 0);
 859                 }
 860
 861                 cur_alloc_size = ins.offset;
 862                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 863                                                ram_size, cur_alloc_size, 0);
 864                 BUG_ON(ret);
 865
 866                 if (root->root_key.objectid ==
 867                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 868                         ret = btrfs_reloc_clone_csums(inode, start,
 869                                                       cur_alloc_size);
 870                         BUG_ON(ret);
 871                 }
 872
 873                 if (disk_num_bytes < cur_alloc_size)
 874                         break;
 875
 876                 /* we're not doing compressed IO, don't unlock the first
 877                  * page (which the caller expects to stay locked), don't
 878                  * clear any dirty bits and don't set any writeback bits
 879                  *
 880                  * Do set the Private2 bit so we know this page was properly
 881                  * setup for writepage
 882                  */
 883                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 884                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 885                         EXTENT_SET_PRIVATE2;
 886
 887                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 888                                              start, start + ram_size - 1,
 889                                              locked_page, op);
 890                 disk_num_bytes -= cur_alloc_size;
 891                 num_bytes -= cur_alloc_size;
 892                 alloc_hint = ins.objectid + ins.offset;
 893                 start += cur_alloc_size;
 894         }
 895 out:
 896         ret = 0;
 897         btrfs_end_transaction(trans, root);
 898
 899         return ret;
 900 }
 901
 902 /*
 903  * work queue call back to started compression on a file and pages
 904  */
 905 static noinline void async_cow_start(struct btrfs_work *work)
 906 {
 907         struct async_cow *async_cow;
 908         int num_added = 0;
 909         async_cow = container_of(work, struct async_cow, work);
 910
 911         compress_file_range(async_cow->inode, async_cow->locked_page,
 912                             async_cow->start, async_cow->end, async_cow,
 913                             &num_added);
 914         if (num_added == 0)
 915                 async_cow->inode = NULL;
 916 }
 917
 918 /*
 919  * work queue call back to submit previously compressed pages
 920  */
 921 static noinline void async_cow_submit(struct btrfs_work *work)
 922 {
 923         struct async_cow *async_cow;
 924         struct btrfs_root *root;
 925         unsigned long nr_pages;
 926
 927         async_cow = container_of(work, struct async_cow, work);
 928
 929         root = async_cow->root;
 930         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 931                 PAGE_CACHE_SHIFT;
 932
 933         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 934
 935         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 936             5 * 1042 * 1024 &&
 937             waitqueue_active(&root->fs_info->async_submit_wait))
 938                 wake_up(&root->fs_info->async_submit_wait);
 939
 940         if (async_cow->inode)
 941                 submit_compressed_extents(async_cow->inode, async_cow);
 942 }
 943
 944 static noinline void async_cow_free(struct btrfs_work *work)
 945 {
 946         struct async_cow *async_cow;
 947         async_cow = container_of(work, struct async_cow, work);
 948         kfree(async_cow);
 949 }
 950
 951 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 952                                 u64 start, u64 end, int *page_started,
 953                                 unsigned long *nr_written)
 954 {
 955         struct async_cow *async_cow;
 956         struct btrfs_root *root = BTRFS_I(inode)->root;
 957         unsigned long nr_pages;
 958         u64 cur_end;
 959         int limit = 10 * 1024 * 1042;
 960
 961         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 962                          1, 0, NULL, GFP_NOFS);
 963         while (start < end) {
 964                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 965                 BUG_ON(!async_cow);
 966                 async_cow->inode = inode;
 967                 async_cow->root = root;
 968                 async_cow->locked_page = locked_page;
 969                 async_cow->start = start;
 970
 971                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 972                         cur_end = end;
 973                 else
 974                         cur_end = min(end, start + 512 * 1024 - 1);
 975
 976                 async_cow->end = cur_end;
 977                 INIT_LIST_HEAD(&async_cow->extents);
 978
 979                 async_cow->work.func = async_cow_start;
 980                 async_cow->work.ordered_func = async_cow_submit;
 981                 async_cow->work.ordered_free = async_cow_free;
 982                 async_cow->work.flags = 0;
 983
 984                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 985                         PAGE_CACHE_SHIFT;
 986                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 987
 988                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 989                                    &async_cow->work);
 990
 991                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 992                         wait_event(root->fs_info->async_submit_wait,
 993                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 994                             limit));
 995                 }
 996
 997                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 998                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1001                            0));
1002                 }
1003
1004                 *nr_written += nr_pages;
1005                 start = cur_end + 1;
1006         }
1007         *page_started = 1;
1008         return 0;
1009 }
1010
1011 static noinline int csum_exist_in_range(struct btrfs_root *root,
1012                                         u64 bytenr, u64 num_bytes)
1013 {
1014         int ret;
1015         struct btrfs_ordered_sum *sums;
1016         LIST_HEAD(list);
1017
1018         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1019                                        bytenr + num_bytes - 1, &list, 0);
1020         if (ret == 0 && list_empty(&list))
1021                 return 0;
1022
1023         while (!list_empty(&list)) {
1024                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1025                 list_del(&sums->list);
1026                 kfree(sums);
1027         }
1028         return 1;
1029 }
1030
1031 /*
1032  * when nowcow writeback call back.  This checks for snapshots or COW copies
1033  * of the extents that exist in the file, and COWs the file as required.
1034  *
1035  * If no cow copies or snapshots exist, we write directly to the existing
1036  * blocks on disk
1037  */
1038 static noinline int run_delalloc_nocow(struct inode *inode,
1039                                        struct page *locked_page,
1040                               u64 start, u64 end, int *page_started, int force,
1041                               unsigned long *nr_written)
1042 {
1043         struct btrfs_root *root = BTRFS_I(inode)->root;
1044         struct btrfs_trans_handle *trans;
1045         struct extent_buffer *leaf;
1046         struct btrfs_path *path;
1047         struct btrfs_file_extent_item *fi;
1048         struct btrfs_key found_key;
1049         u64 cow_start;
1050         u64 cur_offset;
1051         u64 extent_end;
1052         u64 extent_offset;
1053         u64 disk_bytenr;
1054         u64 num_bytes;
1055         int extent_type;
1056         int ret;
1057         int type;
1058         int nocow;
1059         int check_prev = 1;
1060         bool nolock;
1061         u64 ino = btrfs_ino(inode);
1062
1063         path = btrfs_alloc_path();
1064         if (!path)
1065                 return -ENOMEM;
1066
1067         nolock = btrfs_is_free_space_inode(root, inode);
1068
1069         if (nolock)
1070                 trans = btrfs_join_transaction_nolock(root);
1071         else
1072                 trans = btrfs_join_transaction(root);
1073
1074         BUG_ON(IS_ERR(trans));
1075         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1076
1077         cow_start = (u64)-1;
1078         cur_offset = start;
1079         while (1) {
1080                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1081                                                cur_offset, 0);
1082                 BUG_ON(ret < 0);
1083                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084                         leaf = path->nodes[0];
1085                         btrfs_item_key_to_cpu(leaf, &found_key,
1086                                               path->slots[0] - 1);
1087                         if (found_key.objectid == ino &&
1088                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1089                                 path->slots[0]--;
1090                 }
1091                 check_prev = 0;
1092 next_slot:
1093                 leaf = path->nodes[0];
1094                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095                         ret = btrfs_next_leaf(root, path);
1096                         if (ret < 0)
1097                                 BUG_ON(1);
1098                         if (ret > 0)
1099                                 break;
1100                         leaf = path->nodes[0];
1101                 }
1102
1103                 nocow = 0;
1104                 disk_bytenr = 0;
1105                 num_bytes = 0;
1106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
1108                 if (found_key.objectid > ino ||
1109                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110                     found_key.offset > end)
1111                         break;
1112
1113                 if (found_key.offset > cur_offset) {
1114                         extent_end = found_key.offset;
1115                         extent_type = 0;
1116                         goto out_check;
1117                 }
1118
1119                 fi = btrfs_item_ptr(leaf, path->slots[0],
1120                                     struct btrfs_file_extent_item);
1121                 extent_type = btrfs_file_extent_type(leaf, fi);
1122
1123                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1127                         extent_end = found_key.offset +
1128                                 btrfs_file_extent_num_bytes(leaf, fi);
1129                         if (extent_end <= start) {
1130                                 path->slots[0]++;
1131                                 goto next_slot;
1132                         }
1133                         if (disk_bytenr == 0)
1134                                 goto out_check;
1135                         if (btrfs_file_extent_compression(leaf, fi) ||
1136                             btrfs_file_extent_encryption(leaf, fi) ||
1137                             btrfs_file_extent_other_encoding(leaf, fi))
1138                                 goto out_check;
1139                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140                                 goto out_check;
1141                         if (btrfs_extent_readonly(root, disk_bytenr))
1142                                 goto out_check;
1143                         if (btrfs_cross_ref_exist(trans, root, ino,
1144                                                   found_key.offset -
1145                                                   extent_offset, disk_bytenr))
1146                                 goto out_check;
1147                         disk_bytenr += extent_offset;
1148                         disk_bytenr += cur_offset - found_key.offset;
1149                         num_bytes = min(end + 1, extent_end) - cur_offset;
1150                         /*
1151                          * force cow if csum exists in the range.
1152                          * this ensure that csum for a given extent are
1153                          * either valid or do not exist.
1154                          */
1155                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156                                 goto out_check;
1157                         nocow = 1;
1158                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159                         extent_end = found_key.offset +
1160                                 btrfs_file_extent_inline_len(leaf, fi);
1161                         extent_end = ALIGN(extent_end, root->sectorsize);
1162                 } else {
1163                         BUG_ON(1);
1164                 }
1165 out_check:
1166                 if (extent_end <= start) {
1167                         path->slots[0]++;
1168                         goto next_slot;
1169                 }
1170                 if (!nocow) {
1171                         if (cow_start == (u64)-1)
1172                                 cow_start = cur_offset;
1173                         cur_offset = extent_end;
1174                         if (cur_offset > end)
1175                                 break;
1176                         path->slots[0]++;
1177                         goto next_slot;
1178                 }
1179
1180                 btrfs_release_path(path);
1181                 if (cow_start != (u64)-1) {
1182                         ret = cow_file_range(inode, locked_page, cow_start,
1183                                         found_key.offset - 1, page_started,
1184                                         nr_written, 1);
1185                         BUG_ON(ret);
1186                         cow_start = (u64)-1;
1187                 }
1188
1189                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190                         struct extent_map *em;
1191                         struct extent_map_tree *em_tree;
1192                         em_tree = &BTRFS_I(inode)->extent_tree;
1193                         em = alloc_extent_map();
1194                         BUG_ON(!em);
1195                         em->start = cur_offset;
1196                         em->orig_start = em->start;
1197                         em->len = num_bytes;
1198                         em->block_len = num_bytes;
1199                         em->block_start = disk_bytenr;
1200                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1201                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202                         while (1) {
1203                                 write_lock(&em_tree->lock);
1204                                 ret = add_extent_mapping(em_tree, em);
1205                                 write_unlock(&em_tree->lock);
1206                                 if (ret != -EEXIST) {
1207                                         free_extent_map(em);
1208                                         break;
1209                                 }
1210                                 btrfs_drop_extent_cache(inode, em->start,
1211                                                 em->start + em->len - 1, 0);
1212                         }
1213                         type = BTRFS_ORDERED_PREALLOC;
1214                 } else {
1215                         type = BTRFS_ORDERED_NOCOW;
1216                 }
1217
1218                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219                                                num_bytes, num_bytes, type);
1220                 BUG_ON(ret);
1221
1222                 if (root->root_key.objectid ==
1223                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225                                                       num_bytes);
1226                         BUG_ON(ret);
1227                 }
1228
1229                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230                                 cur_offset, cur_offset + num_bytes - 1,
1231                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233                                 EXTENT_SET_PRIVATE2);
1234                 cur_offset = extent_end;
1235                 if (cur_offset > end)
1236                         break;
1237         }
1238         btrfs_release_path(path);
1239
1240         if (cur_offset <= end && cow_start == (u64)-1)
1241                 cow_start = cur_offset;
1242         if (cow_start != (u64)-1) {
1243                 ret = cow_file_range(inode, locked_page, cow_start, end,
1244                                      page_started, nr_written, 1);
1245                 BUG_ON(ret);
1246         }
1247
1248         if (nolock) {
1249                 ret = btrfs_end_transaction_nolock(trans, root);
1250                 BUG_ON(ret);
1251         } else {
1252                 ret = btrfs_end_transaction(trans, root);
1253                 BUG_ON(ret);
1254         }
1255         btrfs_free_path(path);
1256         return 0;
1257 }
1258
1259 /*
1260  * extent_io.c call back to do delayed allocation processing
1261  */
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263                               u64 start, u64 end, int *page_started,
1264                               unsigned long *nr_written)
1265 {
1266         int ret;
1267         struct btrfs_root *root = BTRFS_I(inode)->root;
1268
1269         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271                                          page_started, 1, nr_written);
1272         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274                                          page_started, 0, nr_written);
1275         else if (!btrfs_test_opt(root, COMPRESS) &&
1276                  !(BTRFS_I(inode)->force_compress) &&
1277                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278                 ret = cow_file_range(inode, locked_page, start, end,
1279                                       page_started, nr_written, 1);
1280         else
1281                 ret = cow_file_range_async(inode, locked_page, start, end,
1282                                            page_started, nr_written);
1283         return ret;
1284 }
1285
1286 static void btrfs_split_extent_hook(struct inode *inode,
1287                                     struct extent_state *orig, u64 split)
1288 {
1289         /* not delalloc, ignore it */
1290         if (!(orig->state & EXTENT_DELALLOC))
1291                 return;
1292
1293         spin_lock(&BTRFS_I(inode)->lock);
1294         BTRFS_I(inode)->outstanding_extents++;
1295         spin_unlock(&BTRFS_I(inode)->lock);
1296 }
1297
1298 /*
1299  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1300  * extents so we can keep track of new extents that are just merged onto old
1301  * extents, such as when we are doing sequential writes, so we can properly
1302  * account for the metadata space we'll need.
1303  */
1304 static void btrfs_merge_extent_hook(struct inode *inode,
1305                                     struct extent_state *new,
1306                                     struct extent_state *other)
1307 {
1308         /* not delalloc, ignore it */
1309         if (!(other->state & EXTENT_DELALLOC))
1310                 return;
1311
1312         spin_lock(&BTRFS_I(inode)->lock);
1313         BTRFS_I(inode)->outstanding_extents--;
1314         spin_unlock(&BTRFS_I(inode)->lock);
1315 }
1316
1317 /*
1318  * extent_io.c set_bit_hook, used to track delayed allocation
1319  * bytes in this file, and to maintain the list of inodes that
1320  * have pending delalloc work to be done.
1321  */
1322 static void btrfs_set_bit_hook(struct inode *inode,
1323                                struct extent_state *state, int *bits)
1324 {
1325
1326         /*
1327          * set_bit and clear bit hooks normally require _irqsave/restore
1328          * but in this case, we are only testing for the DELALLOC
1329          * bit, which is only set or cleared with irqs on
1330          */
1331         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1332                 struct btrfs_root *root = BTRFS_I(inode)->root;
1333                 u64 len = state->end + 1 - state->start;
1334                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1335
1336                 if (*bits & EXTENT_FIRST_DELALLOC) {
1337                         *bits &= ~EXTENT_FIRST_DELALLOC;
1338                 } else {
1339                         spin_lock(&BTRFS_I(inode)->lock);
1340                         BTRFS_I(inode)->outstanding_extents++;
1341                         spin_unlock(&BTRFS_I(inode)->lock);
1342                 }
1343
1344                 spin_lock(&root->fs_info->delalloc_lock);
1345                 BTRFS_I(inode)->delalloc_bytes += len;
1346                 root->fs_info->delalloc_bytes += len;
1347                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1348                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1349                                       &root->fs_info->delalloc_inodes);
1350                 }
1351                 spin_unlock(&root->fs_info->delalloc_lock);
1352         }
1353 }
1354
1355 /*
1356  * extent_io.c clear_bit_hook, see set_bit_hook for why
1357  */
1358 static void btrfs_clear_bit_hook(struct inode *inode,
1359                                  struct extent_state *state, int *bits)
1360 {
1361         /*
1362          * set_bit and clear bit hooks normally require _irqsave/restore
1363          * but in this case, we are only testing for the DELALLOC
1364          * bit, which is only set or cleared with irqs on
1365          */
1366         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1367                 struct btrfs_root *root = BTRFS_I(inode)->root;
1368                 u64 len = state->end + 1 - state->start;
1369                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1370
1371                 if (*bits & EXTENT_FIRST_DELALLOC) {
1372                         *bits &= ~EXTENT_FIRST_DELALLOC;
1373                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1374                         spin_lock(&BTRFS_I(inode)->lock);
1375                         BTRFS_I(inode)->outstanding_extents--;
1376                         spin_unlock(&BTRFS_I(inode)->lock);
1377                 }
1378
1379                 if (*bits & EXTENT_DO_ACCOUNTING)
1380                         btrfs_delalloc_release_metadata(inode, len);
1381
1382                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1383                     && do_list)
1384                         btrfs_free_reserved_data_space(inode, len);
1385
1386                 spin_lock(&root->fs_info->delalloc_lock);
1387                 root->fs_info->delalloc_bytes -= len;
1388                 BTRFS_I(inode)->delalloc_bytes -= len;
1389
1390                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1391                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1392                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1393                 }
1394                 spin_unlock(&root->fs_info->delalloc_lock);
1395         }
1396 }
1397
1398 /*
1399  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1400  * we don't create bios that span stripes or chunks
1401  */
1402 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1403                          size_t size, struct bio *bio,
1404                          unsigned long bio_flags)
1405 {
1406         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1407         struct btrfs_mapping_tree *map_tree;
1408         u64 logical = (u64)bio->bi_sector << 9;
1409         u64 length = 0;
1410         u64 map_length;
1411         int ret;
1412
1413         if (bio_flags & EXTENT_BIO_COMPRESSED)
1414                 return 0;
1415
1416         length = bio->bi_size;
1417         map_tree = &root->fs_info->mapping_tree;
1418         map_length = length;
1419         ret = btrfs_map_block(map_tree, READ, logical,
1420                               &map_length, NULL, 0);
1421
1422         if (map_length < length + size)
1423                 return 1;
1424         return ret;
1425 }
1426
1427 /*
1428  * in order to insert checksums into the metadata in large chunks,
1429  * we wait until bio submission time.   All the pages in the bio are
1430  * checksummed and sums are attached onto the ordered extent record.
1431  *
1432  * At IO completion time the cums attached on the ordered extent record
1433  * are inserted into the btree
1434  */
1435 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1436                                     struct bio *bio, int mirror_num,
1437                                     unsigned long bio_flags,
1438                                     u64 bio_offset)
1439 {
1440         struct btrfs_root *root = BTRFS_I(inode)->root;
1441         int ret = 0;
1442
1443         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1444         BUG_ON(ret);
1445         return 0;
1446 }
1447
1448 /*
1449  * in order to insert checksums into the metadata in large chunks,
1450  * we wait until bio submission time.   All the pages in the bio are
1451  * checksummed and sums are attached onto the ordered extent record.
1452  *
1453  * At IO completion time the cums attached on the ordered extent record
1454  * are inserted into the btree
1455  */
1456 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1457                           int mirror_num, unsigned long bio_flags,
1458                           u64 bio_offset)
1459 {
1460         struct btrfs_root *root = BTRFS_I(inode)->root;
1461         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1462 }
1463
1464 /*
1465  * extent_io.c submission hook. This does the right thing for csum calculation
1466  * on write, or reading the csums from the tree before a read
1467  */
1468 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1469                           int mirror_num, unsigned long bio_flags,
1470                           u64 bio_offset)
1471 {
1472         struct btrfs_root *root = BTRFS_I(inode)->root;
1473         int ret = 0;
1474         int skip_sum;
1475
1476         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1477
1478         if (btrfs_is_free_space_inode(root, inode))
1479                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1480         else
1481                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1482         BUG_ON(ret);
1483
1484         if (!(rw & REQ_WRITE)) {
1485                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1486                         return btrfs_submit_compressed_read(inode, bio,
1487                                                     mirror_num, bio_flags);
1488                 } else if (!skip_sum) {
1489                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1490                         if (ret)
1491                                 return ret;
1492                 }
1493                 goto mapit;
1494         } else if (!skip_sum) {
1495                 /* csum items have already been cloned */
1496                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1497                         goto mapit;
1498                 /* we're doing a write, do the async checksumming */
1499                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1500                                    inode, rw, bio, mirror_num,
1501                                    bio_flags, bio_offset,
1502                                    __btrfs_submit_bio_start,
1503                                    __btrfs_submit_bio_done);
1504         }
1505
1506 mapit:
1507         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1508 }
1509
1510 /*
1511  * given a list of ordered sums record them in the inode.  This happens
1512  * at IO completion time based on sums calculated at bio submission time.
1513  */
1514 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1515                              struct inode *inode, u64 file_offset,
1516                              struct list_head *list)
1517 {
1518         struct btrfs_ordered_sum *sum;
1519
1520         list_for_each_entry(sum, list, list) {
1521                 btrfs_csum_file_blocks(trans,
1522                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1523         }
1524         return 0;
1525 }
1526
1527 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1528                               struct extent_state **cached_state)
1529 {
1530         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1531                 WARN_ON(1);
1532         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1533                                    cached_state, GFP_NOFS);
1534 }
1535
1536 /* see btrfs_writepage_start_hook for details on why this is required */
1537 struct btrfs_writepage_fixup {
1538         struct page *page;
1539         struct btrfs_work work;
1540 };
1541
1542 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1543 {
1544         struct btrfs_writepage_fixup *fixup;
1545         struct btrfs_ordered_extent *ordered;
1546         struct extent_state *cached_state = NULL;
1547         struct page *page;
1548         struct inode *inode;
1549         u64 page_start;
1550         u64 page_end;
1551
1552         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1553         page = fixup->page;
1554 again:
1555         lock_page(page);
1556         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1557                 ClearPageChecked(page);
1558                 goto out_page;
1559         }
1560
1561         inode = page->mapping->host;
1562         page_start = page_offset(page);
1563         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1564
1565         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1566                          &cached_state, GFP_NOFS);
1567
1568         /* already ordered? We're done */
1569         if (PagePrivate2(page))
1570                 goto out;
1571
1572         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1573         if (ordered) {
1574                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1575                                      page_end, &cached_state, GFP_NOFS);
1576                 unlock_page(page);
1577                 btrfs_start_ordered_extent(inode, ordered, 1);
1578                 goto again;
1579         }
1580
1581         BUG();
1582         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1583         ClearPageChecked(page);
1584 out:
1585         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1586                              &cached_state, GFP_NOFS);
1587 out_page:
1588         unlock_page(page);
1589         page_cache_release(page);
1590         kfree(fixup);
1591 }
1592
1593 /*
1594  * There are a few paths in the higher layers of the kernel that directly
1595  * set the page dirty bit without asking the filesystem if it is a
1596  * good idea.  This causes problems because we want to make sure COW
1597  * properly happens and the data=ordered rules are followed.
1598  *
1599  * In our case any range that doesn't have the ORDERED bit set
1600  * hasn't been properly setup for IO.  We kick off an async process
1601  * to fix it up.  The async helper will wait for ordered extents, set
1602  * the delalloc bit and make it safe to write the page.
1603  */
1604 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1605 {
1606         struct inode *inode = page->mapping->host;
1607         struct btrfs_writepage_fixup *fixup;
1608         struct btrfs_root *root = BTRFS_I(inode)->root;
1609
1610         /* this page is properly in the ordered list */
1611         if (TestClearPagePrivate2(page))
1612                 return 0;
1613
1614         if (PageChecked(page))
1615                 return -EAGAIN;
1616
1617         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1618         if (!fixup)
1619                 return -EAGAIN;
1620
1621         SetPageChecked(page);
1622         page_cache_get(page);
1623         fixup->work.func = btrfs_writepage_fixup_worker;
1624         fixup->page = page;
1625         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1626         return -EAGAIN;
1627 }
1628
1629 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1630                                        struct inode *inode, u64 file_pos,
1631                                        u64 disk_bytenr, u64 disk_num_bytes,
1632                                        u64 num_bytes, u64 ram_bytes,
1633                                        u8 compression, u8 encryption,
1634                                        u16 other_encoding, int extent_type)
1635 {
1636         struct btrfs_root *root = BTRFS_I(inode)->root;
1637         struct btrfs_file_extent_item *fi;
1638         struct btrfs_path *path;
1639         struct extent_buffer *leaf;
1640         struct btrfs_key ins;
1641         u64 hint;
1642         int ret;
1643
1644         path = btrfs_alloc_path();
1645         if (!path)
1646                 return -ENOMEM;
1647
1648         path->leave_spinning = 1;
1649
1650         /*
1651          * we may be replacing one extent in the tree with another.
1652          * The new extent is pinned in the extent map, and we don't want
1653          * to drop it from the cache until it is completely in the btree.
1654          *
1655          * So, tell btrfs_drop_extents to leave this extent in the cache.
1656          * the caller is expected to unpin it and allow it to be merged
1657          * with the others.
1658          */
1659         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1660                                  &hint, 0);
1661         BUG_ON(ret);
1662
1663         ins.objectid = btrfs_ino(inode);
1664         ins.offset = file_pos;
1665         ins.type = BTRFS_EXTENT_DATA_KEY;
1666         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1667         BUG_ON(ret);
1668         leaf = path->nodes[0];
1669         fi = btrfs_item_ptr(leaf, path->slots[0],
1670                             struct btrfs_file_extent_item);
1671         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1672         btrfs_set_file_extent_type(leaf, fi, extent_type);
1673         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1674         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1675         btrfs_set_file_extent_offset(leaf, fi, 0);
1676         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1677         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1678         btrfs_set_file_extent_compression(leaf, fi, compression);
1679         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1680         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1681
1682         btrfs_unlock_up_safe(path, 1);
1683         btrfs_set_lock_blocking(leaf);
1684
1685         btrfs_mark_buffer_dirty(leaf);
1686
1687         inode_add_bytes(inode, num_bytes);
1688
1689         ins.objectid = disk_bytenr;
1690         ins.offset = disk_num_bytes;
1691         ins.type = BTRFS_EXTENT_ITEM_KEY;
1692         ret = btrfs_alloc_reserved_file_extent(trans, root,
1693                                         root->root_key.objectid,
1694                                         btrfs_ino(inode), file_pos, &ins);
1695         BUG_ON(ret);
1696         btrfs_free_path(path);
1697
1698         return 0;
1699 }
1700
1701 /*
1702  * helper function for btrfs_finish_ordered_io, this
1703  * just reads in some of the csum leaves to prime them into ram
1704  * before we start the transaction.  It limits the amount of btree
1705  * reads required while inside the transaction.
1706  */
1707 /* as ordered data IO finishes, this gets called so we can finish
1708  * an ordered extent if the range of bytes in the file it covers are
1709  * fully written.
1710  */
1711 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1712 {
1713         struct btrfs_root *root = BTRFS_I(inode)->root;
1714         struct btrfs_trans_handle *trans = NULL;
1715         struct btrfs_ordered_extent *ordered_extent = NULL;
1716         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1717         struct extent_state *cached_state = NULL;
1718         int compress_type = 0;
1719         int ret;
1720         bool nolock;
1721
1722         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1723                                              end - start + 1);
1724         if (!ret)
1725                 return 0;
1726         BUG_ON(!ordered_extent);
1727
1728         nolock = btrfs_is_free_space_inode(root, inode);
1729
1730         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1731                 BUG_ON(!list_empty(&ordered_extent->list));
1732                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1733                 if (!ret) {
1734                         if (nolock)
1735                                 trans = btrfs_join_transaction_nolock(root);
1736                         else
1737                                 trans = btrfs_join_transaction(root);
1738                         BUG_ON(IS_ERR(trans));
1739                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740                         ret = btrfs_update_inode(trans, root, inode);
1741                         BUG_ON(ret);
1742                 }
1743                 goto out;
1744         }
1745
1746         lock_extent_bits(io_tree, ordered_extent->file_offset,
1747                          ordered_extent->file_offset + ordered_extent->len - 1,
1748                          0, &cached_state, GFP_NOFS);
1749
1750         if (nolock)
1751                 trans = btrfs_join_transaction_nolock(root);
1752         else
1753                 trans = btrfs_join_transaction(root);
1754         BUG_ON(IS_ERR(trans));
1755         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756
1757         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1758                 compress_type = ordered_extent->compress_type;
1759         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1760                 BUG_ON(compress_type);
1761                 ret = btrfs_mark_extent_written(trans, inode,
1762                                                 ordered_extent->file_offset,
1763                                                 ordered_extent->file_offset +
1764                                                 ordered_extent->len);
1765                 BUG_ON(ret);
1766         } else {
1767                 BUG_ON(root == root->fs_info->tree_root);
1768                 ret = insert_reserved_file_extent(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->start,
1771                                                 ordered_extent->disk_len,
1772                                                 ordered_extent->len,
1773                                                 ordered_extent->len,
1774                                                 compress_type, 0, 0,
1775                                                 BTRFS_FILE_EXTENT_REG);
1776                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1777                                    ordered_extent->file_offset,
1778                                    ordered_extent->len);
1779                 BUG_ON(ret);
1780         }
1781         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1782                              ordered_extent->file_offset +
1783                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1784
1785         add_pending_csums(trans, inode, ordered_extent->file_offset,
1786                           &ordered_extent->list);
1787
1788         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1789         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1790                 ret = btrfs_update_inode(trans, root, inode);
1791                 BUG_ON(ret);
1792         }
1793         ret = 0;
1794 out:
1795         btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1796         if (nolock) {
1797                 if (trans)
1798                         btrfs_end_transaction_nolock(trans, root);
1799         } else {
1800                 if (trans)
1801                         btrfs_end_transaction(trans, root);
1802         }
1803
1804         /* once for us */
1805         btrfs_put_ordered_extent(ordered_extent);
1806         /* once for the tree */
1807         btrfs_put_ordered_extent(ordered_extent);
1808
1809         return 0;
1810 }
1811
1812 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1813                                 struct extent_state *state, int uptodate)
1814 {
1815         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1816
1817         ClearPagePrivate2(page);
1818         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1819 }
1820
1821 /*
1822  * When IO fails, either with EIO or csum verification fails, we
1823  * try other mirrors that might have a good copy of the data.  This
1824  * io_failure_record is used to record state as we go through all the
1825  * mirrors.  If another mirror has good data, the page is set up to date
1826  * and things continue.  If a good mirror can't be found, the original
1827  * bio end_io callback is called to indicate things have failed.
1828  */
1829 struct io_failure_record {
1830         struct page *page;
1831         u64 start;
1832         u64 len;
1833         u64 logical;
1834         unsigned long bio_flags;
1835         int last_mirror;
1836 };
1837
1838 static int btrfs_io_failed_hook(struct bio *failed_bio,
1839                          struct page *page, u64 start, u64 end,
1840                          struct extent_state *state)
1841 {
1842         struct io_failure_record *failrec = NULL;
1843         u64 private;
1844         struct extent_map *em;
1845         struct inode *inode = page->mapping->host;
1846         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1847         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1848         struct bio *bio;
1849         int num_copies;
1850         int ret;
1851         int rw;
1852         u64 logical;
1853
1854         ret = get_state_private(failure_tree, start, &private);
1855         if (ret) {
1856                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1857                 if (!failrec)
1858                         return -ENOMEM;
1859                 failrec->start = start;
1860                 failrec->len = end - start + 1;
1861                 failrec->last_mirror = 0;
1862                 failrec->bio_flags = 0;
1863
1864                 read_lock(&em_tree->lock);
1865                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1866                 if (em->start > start || em->start + em->len < start) {
1867                         free_extent_map(em);
1868                         em = NULL;
1869                 }
1870                 read_unlock(&em_tree->lock);
1871
1872                 if (IS_ERR_OR_NULL(em)) {
1873                         kfree(failrec);
1874                         return -EIO;
1875                 }
1876                 logical = start - em->start;
1877                 logical = em->block_start + logical;
1878                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1879                         logical = em->block_start;
1880                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1881                         extent_set_compress_type(&failrec->bio_flags,
1882                                                  em->compress_type);
1883                 }
1884                 failrec->logical = logical;
1885                 free_extent_map(em);
1886                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1887                                 EXTENT_DIRTY, GFP_NOFS);
1888                 set_state_private(failure_tree, start,
1889                                  (u64)(unsigned long)failrec);
1890         } else {
1891                 failrec = (struct io_failure_record *)(unsigned long)private;
1892         }
1893         num_copies = btrfs_num_copies(
1894                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1895                               failrec->logical, failrec->len);
1896         failrec->last_mirror++;
1897         if (!state) {
1898                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1899                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1900                                                     failrec->start,
1901                                                     EXTENT_LOCKED);
1902                 if (state && state->start != failrec->start)
1903                         state = NULL;
1904                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1905         }
1906         if (!state || failrec->last_mirror > num_copies) {
1907                 set_state_private(failure_tree, failrec->start, 0);
1908                 clear_extent_bits(failure_tree, failrec->start,
1909                                   failrec->start + failrec->len - 1,
1910                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1911                 kfree(failrec);
1912                 return -EIO;
1913         }
1914         bio = bio_alloc(GFP_NOFS, 1);
1915         bio->bi_private = state;
1916         bio->bi_end_io = failed_bio->bi_end_io;
1917         bio->bi_sector = failrec->logical >> 9;
1918         bio->bi_bdev = failed_bio->bi_bdev;
1919         bio->bi_size = 0;
1920
1921         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1922         if (failed_bio->bi_rw & REQ_WRITE)
1923                 rw = WRITE;
1924         else
1925                 rw = READ;
1926
1927         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1928                                                       failrec->last_mirror,
1929                                                       failrec->bio_flags, 0);
1930         return ret;
1931 }
1932
1933 /*
1934  * each time an IO finishes, we do a fast check in the IO failure tree
1935  * to see if we need to process or clean up an io_failure_record
1936  */
1937 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1938 {
1939         u64 private;
1940         u64 private_failure;
1941         struct io_failure_record *failure;
1942         int ret;
1943
1944         private = 0;
1945         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1946                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1947                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1948                                         start, &private_failure);
1949                 if (ret == 0) {
1950                         failure = (struct io_failure_record *)(unsigned long)
1951                                    private_failure;
1952                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1953                                           failure->start, 0);
1954                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1955                                           failure->start,
1956                                           failure->start + failure->len - 1,
1957                                           EXTENT_DIRTY | EXTENT_LOCKED,
1958                                           GFP_NOFS);
1959                         kfree(failure);
1960                 }
1961         }
1962         return 0;
1963 }
1964
1965 /*
1966  * when reads are done, we need to check csums to verify the data is correct
1967  * if there's a match, we allow the bio to finish.  If not, we go through
1968  * the io_failure_record routines to find good copies
1969  */
1970 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1971                                struct extent_state *state)
1972 {
1973         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1974         struct inode *inode = page->mapping->host;
1975         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1976         char *kaddr;
1977         u64 private = ~(u32)0;
1978         int ret;
1979         struct btrfs_root *root = BTRFS_I(inode)->root;
1980         u32 csum = ~(u32)0;
1981
1982         if (PageChecked(page)) {
1983                 ClearPageChecked(page);
1984                 goto good;
1985         }
1986
1987         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1988                 goto good;
1989
1990         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1991             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1992                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1993                                   GFP_NOFS);
1994                 return 0;
1995         }
1996
1997         if (state && state->start == start) {
1998                 private = state->private;
1999                 ret = 0;
2000         } else {
2001                 ret = get_state_private(io_tree, start, &private);
2002         }
2003         kaddr = kmap_atomic(page, KM_USER0);
2004         if (ret)
2005                 goto zeroit;
2006
2007         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2008         btrfs_csum_final(csum, (char *)&csum);
2009         if (csum != private)
2010                 goto zeroit;
2011
2012         kunmap_atomic(kaddr, KM_USER0);
2013 good:
2014         /* if the io failure tree for this inode is non-empty,
2015          * check to see if we've recovered from a failed IO
2016          */
2017         btrfs_clean_io_failures(inode, start);
2018         return 0;
2019
2020 zeroit:
2021         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2022                        "private %llu\n",
2023                        (unsigned long long)btrfs_ino(page->mapping->host),
2024                        (unsigned long long)start, csum,
2025                        (unsigned long long)private);
2026         memset(kaddr + offset, 1, end - start + 1);
2027         flush_dcache_page(page);
2028         kunmap_atomic(kaddr, KM_USER0);
2029         if (private == 0)
2030                 return 0;
2031         return -EIO;
2032 }
2033
2034 struct delayed_iput {
2035         struct list_head list;
2036         struct inode *inode;
2037 };
2038
2039 void btrfs_add_delayed_iput(struct inode *inode)
2040 {
2041         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2042         struct delayed_iput *delayed;
2043
2044         if (atomic_add_unless(&inode->i_count, -1, 1))
2045                 return;
2046
2047         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2048         delayed->inode = inode;
2049
2050         spin_lock(&fs_info->delayed_iput_lock);
2051         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2052         spin_unlock(&fs_info->delayed_iput_lock);
2053 }
2054
2055 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2056 {
2057         LIST_HEAD(list);
2058         struct btrfs_fs_info *fs_info = root->fs_info;
2059         struct delayed_iput *delayed;
2060         int empty;
2061
2062         spin_lock(&fs_info->delayed_iput_lock);
2063         empty = list_empty(&fs_info->delayed_iputs);
2064         spin_unlock(&fs_info->delayed_iput_lock);
2065         if (empty)
2066                 return;
2067
2068         down_read(&root->fs_info->cleanup_work_sem);
2069         spin_lock(&fs_info->delayed_iput_lock);
2070         list_splice_init(&fs_info->delayed_iputs, &list);
2071         spin_unlock(&fs_info->delayed_iput_lock);
2072
2073         while (!list_empty(&list)) {
2074                 delayed = list_entry(list.next, struct delayed_iput, list);
2075                 list_del(&delayed->list);
2076                 iput(delayed->inode);
2077                 kfree(delayed);
2078         }
2079         up_read(&root->fs_info->cleanup_work_sem);
2080 }
2081
2082 enum btrfs_orphan_cleanup_state {
2083         ORPHAN_CLEANUP_STARTED  = 1,
2084         ORPHAN_CLEANUP_DONE     = 2,
2085 };
2086
2087 /*
2088  * This is called in transaction commmit time. If there are no orphan
2089  * files in the subvolume, it removes orphan item and frees block_rsv
2090  * structure.
2091  */
2092 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2093                               struct btrfs_root *root)
2094 {
2095         int ret;
2096
2097         if (!list_empty(&root->orphan_list) ||
2098             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2099                 return;
2100
2101         if (root->orphan_item_inserted &&
2102             btrfs_root_refs(&root->root_item) > 0) {
2103                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2104                                             root->root_key.objectid);
2105                 BUG_ON(ret);
2106                 root->orphan_item_inserted = 0;
2107         }
2108
2109         if (root->orphan_block_rsv) {
2110                 WARN_ON(root->orphan_block_rsv->size > 0);
2111                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2112                 root->orphan_block_rsv = NULL;
2113         }
2114 }
2115
2116 /*
2117  * This creates an orphan entry for the given inode in case something goes
2118  * wrong in the middle of an unlink/truncate.
2119  *
2120  * NOTE: caller of this function should reserve 5 units of metadata for
2121  *       this function.
2122  */
2123 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2124 {
2125         struct btrfs_root *root = BTRFS_I(inode)->root;
2126         struct btrfs_block_rsv *block_rsv = NULL;
2127         int reserve = 0;
2128         int insert = 0;
2129         int ret;
2130
2131         if (!root->orphan_block_rsv) {
2132                 block_rsv = btrfs_alloc_block_rsv(root);
2133                 if (!block_rsv)
2134                         return -ENOMEM;
2135         }
2136
2137         spin_lock(&root->orphan_lock);
2138         if (!root->orphan_block_rsv) {
2139                 root->orphan_block_rsv = block_rsv;
2140         } else if (block_rsv) {
2141                 btrfs_free_block_rsv(root, block_rsv);
2142                 block_rsv = NULL;
2143         }
2144
2145         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2146                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2147 #if 0
2148                 /*
2149                  * For proper ENOSPC handling, we should do orphan
2150                  * cleanup when mounting. But this introduces backward
2151                  * compatibility issue.
2152                  */
2153                 if (!xchg(&root->orphan_item_inserted, 1))
2154                         insert = 2;
2155                 else
2156                         insert = 1;
2157 #endif
2158                 insert = 1;
2159         }
2160
2161         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2162                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2163                 reserve = 1;
2164         }
2165         spin_unlock(&root->orphan_lock);
2166
2167         /* grab metadata reservation from transaction handle */
2168         if (reserve) {
2169                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2170                 BUG_ON(ret);
2171         }
2172
2173         /* insert an orphan item to track this unlinked/truncated file */
2174         if (insert >= 1) {
2175                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2176                 BUG_ON(ret);
2177         }
2178
2179         /* insert an orphan item to track subvolume contains orphan files */
2180         if (insert >= 2) {
2181                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2182                                                root->root_key.objectid);
2183                 BUG_ON(ret);
2184         }
2185         return 0;
2186 }
2187
2188 /*
2189  * We have done the truncate/delete so we can go ahead and remove the orphan
2190  * item for this particular inode.
2191  */
2192 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2193 {
2194         struct btrfs_root *root = BTRFS_I(inode)->root;
2195         int delete_item = 0;
2196         int release_rsv = 0;
2197         int ret = 0;
2198
2199         spin_lock(&root->orphan_lock);
2200         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2201                 list_del_init(&BTRFS_I(inode)->i_orphan);
2202                 delete_item = 1;
2203         }
2204
2205         if (BTRFS_I(inode)->orphan_meta_reserved) {
2206                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2207                 release_rsv = 1;
2208         }
2209         spin_unlock(&root->orphan_lock);
2210
2211         if (trans && delete_item) {
2212                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2213                 BUG_ON(ret);
2214         }
2215
2216         if (release_rsv)
2217                 btrfs_orphan_release_metadata(inode);
2218
2219         return 0;
2220 }
2221
2222 /*
2223  * this cleans up any orphans that may be left on the list from the last use
2224  * of this root.
2225  */
2226 int btrfs_orphan_cleanup(struct btrfs_root *root)
2227 {
2228         struct btrfs_path *path;
2229         struct extent_buffer *leaf;
2230         struct btrfs_key key, found_key;
2231         struct btrfs_trans_handle *trans;
2232         struct inode *inode;
2233         u64 last_objectid = 0;
2234         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2235
2236         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2237                 return 0;
2238
2239         path = btrfs_alloc_path();
2240         if (!path) {
2241                 ret = -ENOMEM;
2242                 goto out;
2243         }
2244         path->reada = -1;
2245
2246         key.objectid = BTRFS_ORPHAN_OBJECTID;
2247         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2248         key.offset = (u64)-1;
2249
2250         while (1) {
2251                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2252                 if (ret < 0)
2253                         goto out;
2254
2255                 /*
2256                  * if ret == 0 means we found what we were searching for, which
2257                  * is weird, but possible, so only screw with path if we didn't
2258                  * find the key and see if we have stuff that matches
2259                  */
2260                 if (ret > 0) {
2261                         ret = 0;
2262                         if (path->slots[0] == 0)
2263                                 break;
2264                         path->slots[0]--;
2265                 }
2266
2267                 /* pull out the item */
2268                 leaf = path->nodes[0];
2269                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2270
2271                 /* make sure the item matches what we want */
2272                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2273                         break;
2274                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2275                         break;
2276
2277                 /* release the path since we're done with it */
2278                 btrfs_release_path(path);
2279
2280                 /*
2281                  * this is where we are basically btrfs_lookup, without the
2282                  * crossing root thing.  we store the inode number in the
2283                  * offset of the orphan item.
2284                  */
2285
2286                 if (found_key.offset == last_objectid) {
2287                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2288                                "stopping orphan cleanup\n");
2289                         ret = -EINVAL;
2290                         goto out;
2291                 }
2292
2293                 last_objectid = found_key.offset;
2294
2295                 found_key.objectid = found_key.offset;
2296                 found_key.type = BTRFS_INODE_ITEM_KEY;
2297                 found_key.offset = 0;
2298                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2299                 ret = PTR_RET(inode);
2300                 if (ret && ret != -ESTALE)
2301                         goto out;
2302
2303                 /*
2304                  * Inode is already gone but the orphan item is still there,
2305                  * kill the orphan item.
2306                  */
2307                 if (ret == -ESTALE) {
2308                         trans = btrfs_start_transaction(root, 1);
2309                         if (IS_ERR(trans)) {
2310                                 ret = PTR_ERR(trans);
2311                                 goto out;
2312                         }
2313                         ret = btrfs_del_orphan_item(trans, root,
2314                                                     found_key.objectid);
2315                         BUG_ON(ret);
2316                         btrfs_end_transaction(trans, root);
2317                         continue;
2318                 }
2319
2320                 /*
2321                  * add this inode to the orphan list so btrfs_orphan_del does
2322                  * the proper thing when we hit it
2323                  */
2324                 spin_lock(&root->orphan_lock);
2325                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2326                 spin_unlock(&root->orphan_lock);
2327
2328                 /* if we have links, this was a truncate, lets do that */
2329                 if (inode->i_nlink) {
2330                         if (!S_ISREG(inode->i_mode)) {
2331                                 WARN_ON(1);
2332                                 iput(inode);
2333                                 continue;
2334                         }
2335                         nr_truncate++;
2336                         ret = btrfs_truncate(inode);
2337                 } else {
2338                         nr_unlink++;
2339                 }
2340
2341                 /* this will do delete_inode and everything for us */
2342                 iput(inode);
2343                 if (ret)
2344                         goto out;
2345         }
2346         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2347
2348         if (root->orphan_block_rsv)
2349                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2350                                         (u64)-1);
2351
2352         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2353                 trans = btrfs_join_transaction(root);
2354                 if (!IS_ERR(trans))
2355                         btrfs_end_transaction(trans, root);
2356         }
2357
2358         if (nr_unlink)
2359                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2360         if (nr_truncate)
2361                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2362
2363 out:
2364         if (ret)
2365                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2366         btrfs_free_path(path);
2367         return ret;
2368 }
2369
2370 /*
2371  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2372  * don't find any xattrs, we know there can't be any acls.
2373  *
2374  * slot is the slot the inode is in, objectid is the objectid of the inode
2375  */
2376 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2377                                           int slot, u64 objectid)
2378 {
2379         u32 nritems = btrfs_header_nritems(leaf);
2380         struct btrfs_key found_key;
2381         int scanned = 0;
2382
2383         slot++;
2384         while (slot < nritems) {
2385                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2386
2387                 /* we found a different objectid, there must not be acls */
2388                 if (found_key.objectid != objectid)
2389                         return 0;
2390
2391                 /* we found an xattr, assume we've got an acl */
2392                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2393                         return 1;
2394
2395                 /*
2396                  * we found a key greater than an xattr key, there can't
2397                  * be any acls later on
2398                  */
2399                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2400                         return 0;
2401
2402                 slot++;
2403                 scanned++;
2404
2405                 /*
2406                  * it goes inode, inode backrefs, xattrs, extents,
2407                  * so if there are a ton of hard links to an inode there can
2408                  * be a lot of backrefs.  Don't waste time searching too hard,
2409                  * this is just an optimization
2410                  */
2411                 if (scanned >= 8)
2412                         break;
2413         }
2414         /* we hit the end of the leaf before we found an xattr or
2415          * something larger than an xattr.  We have to assume the inode
2416          * has acls
2417          */
2418         return 1;
2419 }
2420
2421 /*
2422  * read an inode from the btree into the in-memory inode
2423  */
2424 static void btrfs_read_locked_inode(struct inode *inode)
2425 {
2426         struct btrfs_path *path;
2427         struct extent_buffer *leaf;
2428         struct btrfs_inode_item *inode_item;
2429         struct btrfs_timespec *tspec;
2430         struct btrfs_root *root = BTRFS_I(inode)->root;
2431         struct btrfs_key location;
2432         int maybe_acls;
2433         u32 rdev;
2434         int ret;
2435         bool filled = false;
2436
2437         ret = btrfs_fill_inode(inode, &rdev);
2438         if (!ret)
2439                 filled = true;
2440
2441         path = btrfs_alloc_path();
2442         if (!path)
2443                 goto make_bad;
2444
2445         path->leave_spinning = 1;
2446         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2447
2448         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2449         if (ret)
2450                 goto make_bad;
2451
2452         leaf = path->nodes[0];
2453
2454         if (filled)
2455                 goto cache_acl;
2456
2457         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2458                                     struct btrfs_inode_item);
2459         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2460         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2461         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2462         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2463         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2464
2465         tspec = btrfs_inode_atime(inode_item);
2466         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2467         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2468
2469         tspec = btrfs_inode_mtime(inode_item);
2470         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2472
2473         tspec = btrfs_inode_ctime(inode_item);
2474         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2476
2477         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2478         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2479         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2480         inode->i_generation = BTRFS_I(inode)->generation;
2481         inode->i_rdev = 0;
2482         rdev = btrfs_inode_rdev(leaf, inode_item);
2483
2484         BTRFS_I(inode)->index_cnt = (u64)-1;
2485         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2486 cache_acl:
2487         /*
2488          * try to precache a NULL acl entry for files that don't have
2489          * any xattrs or acls
2490          */
2491         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2492                                            btrfs_ino(inode));
2493         if (!maybe_acls)
2494                 cache_no_acl(inode);
2495
2496         btrfs_free_path(path);
2497
2498         switch (inode->i_mode & S_IFMT) {
2499         case S_IFREG:
2500                 inode->i_mapping->a_ops = &btrfs_aops;
2501                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2502                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2503                 inode->i_fop = &btrfs_file_operations;
2504                 inode->i_op = &btrfs_file_inode_operations;
2505                 break;
2506         case S_IFDIR:
2507                 inode->i_fop = &btrfs_dir_file_operations;
2508                 if (root == root->fs_info->tree_root)
2509                         inode->i_op = &btrfs_dir_ro_inode_operations;
2510                 else
2511                         inode->i_op = &btrfs_dir_inode_operations;
2512                 break;
2513         case S_IFLNK:
2514                 inode->i_op = &btrfs_symlink_inode_operations;
2515                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2516                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2517                 break;
2518         default:
2519                 inode->i_op = &btrfs_special_inode_operations;
2520                 init_special_inode(inode, inode->i_mode, rdev);
2521                 break;
2522         }
2523
2524         btrfs_update_iflags(inode);
2525         return;
2526
2527 make_bad:
2528         btrfs_free_path(path);
2529         make_bad_inode(inode);
2530 }
2531
2532 /*
2533  * given a leaf and an inode, copy the inode fields into the leaf
2534  */
2535 static void fill_inode_item(struct btrfs_trans_handle *trans,
2536                             struct extent_buffer *leaf,
2537                             struct btrfs_inode_item *item,
2538                             struct inode *inode)
2539 {
2540         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2541         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2542         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2543         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2544         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2545
2546         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2547                                inode->i_atime.tv_sec);
2548         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2549                                 inode->i_atime.tv_nsec);
2550
2551         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2552                                inode->i_mtime.tv_sec);
2553         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2554                                 inode->i_mtime.tv_nsec);
2555
2556         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2557                                inode->i_ctime.tv_sec);
2558         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2559                                 inode->i_ctime.tv_nsec);
2560
2561         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2562         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2563         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2564         btrfs_set_inode_transid(leaf, item, trans->transid);
2565         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2566         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2567         btrfs_set_inode_block_group(leaf, item, 0);
2568 }
2569
2570 /*
2571  * copy everything in the in-memory inode into the btree.
2572  */
2573 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2574                                 struct btrfs_root *root, struct inode *inode)
2575 {
2576         struct btrfs_inode_item *inode_item;
2577         struct btrfs_path *path;
2578         struct extent_buffer *leaf;
2579         int ret;
2580
2581         /*
2582          * If the inode is a free space inode, we can deadlock during commit
2583          * if we put it into the delayed code.
2584          *
2585          * The data relocation inode should also be directly updated
2586          * without delay
2587          */
2588         if (!btrfs_is_free_space_inode(root, inode)
2589             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2590                 ret = btrfs_delayed_update_inode(trans, root, inode);
2591                 if (!ret)
2592                         btrfs_set_inode_last_trans(trans, inode);
2593                 return ret;
2594         }
2595
2596         path = btrfs_alloc_path();
2597         if (!path)
2598                 return -ENOMEM;
2599
2600         path->leave_spinning = 1;
2601         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2602                                  1);
2603         if (ret) {
2604                 if (ret > 0)
2605                         ret = -ENOENT;
2606                 goto failed;
2607         }
2608
2609         btrfs_unlock_up_safe(path, 1);
2610         leaf = path->nodes[0];
2611         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2612                                     struct btrfs_inode_item);
2613
2614         fill_inode_item(trans, leaf, inode_item, inode);
2615         btrfs_mark_buffer_dirty(leaf);
2616         btrfs_set_inode_last_trans(trans, inode);
2617         ret = 0;
2618 failed:
2619         btrfs_free_path(path);
2620         return ret;
2621 }
2622
2623 /*
2624  * unlink helper that gets used here in inode.c and in the tree logging
2625  * recovery code.  It remove a link in a directory with a given name, and
2626  * also drops the back refs in the inode to the directory
2627  */
2628 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2629                                 struct btrfs_root *root,
2630                                 struct inode *dir, struct inode *inode,
2631                                 const char *name, int name_len)
2632 {
2633         struct btrfs_path *path;
2634         int ret = 0;
2635         struct extent_buffer *leaf;
2636         struct btrfs_dir_item *di;
2637         struct btrfs_key key;
2638         u64 index;
2639         u64 ino = btrfs_ino(inode);
2640         u64 dir_ino = btrfs_ino(dir);
2641
2642         path = btrfs_alloc_path();
2643         if (!path) {
2644                 ret = -ENOMEM;
2645                 goto out;
2646         }
2647
2648         path->leave_spinning = 1;
2649         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2650                                     name, name_len, -1);
2651         if (IS_ERR(di)) {
2652                 ret = PTR_ERR(di);
2653                 goto err;
2654         }
2655         if (!di) {
2656                 ret = -ENOENT;
2657                 goto err;
2658         }
2659         leaf = path->nodes[0];
2660         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2661         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2662         if (ret)
2663                 goto err;
2664         btrfs_release_path(path);
2665
2666         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2667                                   dir_ino, &index);
2668         if (ret) {
2669                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2670                        "inode %llu parent %llu\n", name_len, name,
2671                        (unsigned long long)ino, (unsigned long long)dir_ino);
2672                 goto err;
2673         }
2674
2675         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2676         if (ret)
2677                 goto err;
2678
2679         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2680                                          inode, dir_ino);
2681         BUG_ON(ret != 0 && ret != -ENOENT);
2682
2683         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2684                                            dir, index);
2685         if (ret == -ENOENT)
2686                 ret = 0;
2687 err:
2688         btrfs_free_path(path);
2689         if (ret)
2690                 goto out;
2691
2692         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2693         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2694         btrfs_update_inode(trans, root, dir);
2695 out:
2696         return ret;
2697 }
2698
2699 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2700                        struct btrfs_root *root,
2701                        struct inode *dir, struct inode *inode,
2702                        const char *name, int name_len)
2703 {
2704         int ret;
2705         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2706         if (!ret) {
2707                 btrfs_drop_nlink(inode);
2708                 ret = btrfs_update_inode(trans, root, inode);
2709         }
2710         return ret;
2711 }
2712
2713
2714 /* helper to check if there is any shared block in the path */
2715 static int check_path_shared(struct btrfs_root *root,
2716                              struct btrfs_path *path)
2717 {
2718         struct extent_buffer *eb;
2719         int level;
2720         u64 refs = 1;
2721
2722         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2723                 int ret;
2724
2725                 if (!path->nodes[level])
2726                         break;
2727                 eb = path->nodes[level];
2728                 if (!btrfs_block_can_be_shared(root, eb))
2729                         continue;
2730                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2731                                                &refs, NULL);
2732                 if (refs > 1)
2733                         return 1;
2734         }
2735         return 0;
2736 }
2737
2738 /*
2739  * helper to start transaction for unlink and rmdir.
2740  *
2741  * unlink and rmdir are special in btrfs, they do not always free space.
2742  * so in enospc case, we should make sure they will free space before
2743  * allowing them to use the global metadata reservation.
2744  */
2745 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2746                                                        struct dentry *dentry)
2747 {
2748         struct btrfs_trans_handle *trans;
2749         struct btrfs_root *root = BTRFS_I(dir)->root;
2750         struct btrfs_path *path;
2751         struct btrfs_inode_ref *ref;
2752         struct btrfs_dir_item *di;
2753         struct inode *inode = dentry->d_inode;
2754         u64 index;
2755         int check_link = 1;
2756         int err = -ENOSPC;
2757         int ret;
2758         u64 ino = btrfs_ino(inode);
2759         u64 dir_ino = btrfs_ino(dir);
2760
2761         trans = btrfs_start_transaction(root, 10);
2762         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2763                 return trans;
2764
2765         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2766                 return ERR_PTR(-ENOSPC);
2767
2768         /* check if there is someone else holds reference */
2769         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2770                 return ERR_PTR(-ENOSPC);
2771
2772         if (atomic_read(&inode->i_count) > 2)
2773                 return ERR_PTR(-ENOSPC);
2774
2775         if (xchg(&root->fs_info->enospc_unlink, 1))
2776                 return ERR_PTR(-ENOSPC);
2777
2778         path = btrfs_alloc_path();
2779         if (!path) {
2780                 root->fs_info->enospc_unlink = 0;
2781                 return ERR_PTR(-ENOMEM);
2782         }
2783
2784         trans = btrfs_start_transaction(root, 0);
2785         if (IS_ERR(trans)) {
2786                 btrfs_free_path(path);
2787                 root->fs_info->enospc_unlink = 0;
2788                 return trans;
2789         }
2790
2791         path->skip_locking = 1;
2792         path->search_commit_root = 1;
2793
2794         ret = btrfs_lookup_inode(trans, root, path,
2795                                 &BTRFS_I(dir)->location, 0);
2796         if (ret < 0) {
2797                 err = ret;
2798                 goto out;
2799         }
2800         if (ret == 0) {
2801                 if (check_path_shared(root, path))
2802                         goto out;
2803         } else {
2804                 check_link = 0;
2805         }
2806         btrfs_release_path(path);
2807
2808         ret = btrfs_lookup_inode(trans, root, path,
2809                                 &BTRFS_I(inode)->location, 0);
2810         if (ret < 0) {
2811                 err = ret;
2812                 goto out;
2813         }
2814         if (ret == 0) {
2815                 if (check_path_shared(root, path))
2816                         goto out;
2817         } else {
2818                 check_link = 0;
2819         }
2820         btrfs_release_path(path);
2821
2822         if (ret == 0 && S_ISREG(inode->i_mode)) {
2823                 ret = btrfs_lookup_file_extent(trans, root, path,
2824                                                ino, (u64)-1, 0);
2825                 if (ret < 0) {
2826                         err = ret;
2827                         goto out;
2828                 }
2829                 BUG_ON(ret == 0);
2830                 if (check_path_shared(root, path))
2831                         goto out;
2832                 btrfs_release_path(path);
2833         }
2834
2835         if (!check_link) {
2836                 err = 0;
2837                 goto out;
2838         }
2839
2840         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2841                                 dentry->d_name.name, dentry->d_name.len, 0);
2842         if (IS_ERR(di)) {
2843                 err = PTR_ERR(di);
2844                 goto out;
2845         }
2846         if (di) {
2847                 if (check_path_shared(root, path))
2848                         goto out;
2849         } else {
2850                 err = 0;
2851                 goto out;
2852         }
2853         btrfs_release_path(path);
2854
2855         ref = btrfs_lookup_inode_ref(trans, root, path,
2856                                 dentry->d_name.name, dentry->d_name.len,
2857                                 ino, dir_ino, 0);
2858         if (IS_ERR(ref)) {
2859                 err = PTR_ERR(ref);
2860                 goto out;
2861         }
2862         BUG_ON(!ref);
2863         if (check_path_shared(root, path))
2864                 goto out;
2865         index = btrfs_inode_ref_index(path->nodes[0], ref);
2866         btrfs_release_path(path);
2867
2868         /*
2869          * This is a commit root search, if we can lookup inode item and other
2870          * relative items in the commit root, it means the transaction of
2871          * dir/file creation has been committed, and the dir index item that we
2872          * delay to insert has also been inserted into the commit root. So
2873          * we needn't worry about the delayed insertion of the dir index item
2874          * here.
2875          */
2876         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2877                                 dentry->d_name.name, dentry->d_name.len, 0);
2878         if (IS_ERR(di)) {
2879                 err = PTR_ERR(di);
2880                 goto out;
2881         }
2882         BUG_ON(ret == -ENOENT);
2883         if (check_path_shared(root, path))
2884                 goto out;
2885
2886         err = 0;
2887 out:
2888         btrfs_free_path(path);
2889         if (err) {
2890                 btrfs_end_transaction(trans, root);
2891                 root->fs_info->enospc_unlink = 0;
2892                 return ERR_PTR(err);
2893         }
2894
2895         trans->block_rsv = &root->fs_info->global_block_rsv;
2896         return trans;
2897 }
2898
2899 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2900                                struct btrfs_root *root)
2901 {
2902         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2903                 BUG_ON(!root->fs_info->enospc_unlink);
2904                 root->fs_info->enospc_unlink = 0;
2905         }
2906         btrfs_end_transaction_throttle(trans, root);
2907 }
2908
2909 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2910 {
2911         struct btrfs_root *root = BTRFS_I(dir)->root;
2912         struct btrfs_trans_handle *trans;
2913         struct inode *inode = dentry->d_inode;
2914         int ret;
2915         unsigned long nr = 0;
2916
2917         trans = __unlink_start_trans(dir, dentry);
2918         if (IS_ERR(trans))
2919                 return PTR_ERR(trans);
2920
2921         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2922
2923         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2924                                  dentry->d_name.name, dentry->d_name.len);
2925         if (ret)
2926                 goto out;
2927
2928         if (inode->i_nlink == 0) {
2929                 ret = btrfs_orphan_add(trans, inode);
2930                 if (ret)
2931                         goto out;
2932         }
2933
2934 out:
2935         nr = trans->blocks_used;
2936         __unlink_end_trans(trans, root);
2937         btrfs_btree_balance_dirty(root, nr);
2938         return ret;
2939 }
2940
2941 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2942                         struct btrfs_root *root,
2943                         struct inode *dir, u64 objectid,
2944                         const char *name, int name_len)
2945 {
2946         struct btrfs_path *path;
2947         struct extent_buffer *leaf;
2948         struct btrfs_dir_item *di;
2949         struct btrfs_key key;
2950         u64 index;
2951         int ret;
2952         u64 dir_ino = btrfs_ino(dir);
2953
2954         path = btrfs_alloc_path();
2955         if (!path)
2956                 return -ENOMEM;
2957
2958         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2959                                    name, name_len, -1);
2960         BUG_ON(IS_ERR_OR_NULL(di));
2961
2962         leaf = path->nodes[0];
2963         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2964         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2965         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2966         BUG_ON(ret);
2967         btrfs_release_path(path);
2968
2969         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2970                                  objectid, root->root_key.objectid,
2971                                  dir_ino, &index, name, name_len);
2972         if (ret < 0) {
2973                 BUG_ON(ret != -ENOENT);
2974                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2975                                                  name, name_len);
2976                 BUG_ON(IS_ERR_OR_NULL(di));
2977
2978                 leaf = path->nodes[0];
2979                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2980                 btrfs_release_path(path);
2981                 index = key.offset;
2982         }
2983         btrfs_release_path(path);
2984
2985         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2986         BUG_ON(ret);
2987
2988         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2989         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2990         ret = btrfs_update_inode(trans, root, dir);
2991         BUG_ON(ret);
2992
2993         btrfs_free_path(path);
2994         return 0;
2995 }
2996
2997 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2998 {
2999         struct inode *inode = dentry->d_inode;
3000         int err = 0;
3001         struct btrfs_root *root = BTRFS_I(dir)->root;
3002         struct btrfs_trans_handle *trans;
3003         unsigned long nr = 0;
3004
3005         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3006             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3007                 return -ENOTEMPTY;
3008
3009         trans = __unlink_start_trans(dir, dentry);
3010         if (IS_ERR(trans))
3011                 return PTR_ERR(trans);
3012
3013         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3014                 err = btrfs_unlink_subvol(trans, root, dir,
3015                                           BTRFS_I(inode)->location.objectid,
3016                                           dentry->d_name.name,
3017                                           dentry->d_name.len);
3018                 goto out;
3019         }
3020
3021         err = btrfs_orphan_add(trans, inode);
3022         if (err)
3023                 goto out;
3024
3025         /* now the directory is empty */
3026         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3027                                  dentry->d_name.name, dentry->d_name.len);
3028         if (!err)
3029                 btrfs_i_size_write(inode, 0);
3030 out:
3031         nr = trans->blocks_used;
3032         __unlink_end_trans(trans, root);
3033         btrfs_btree_balance_dirty(root, nr);
3034
3035         return err;
3036 }
3037
3038 /*
3039  * this can truncate away extent items, csum items and directory items.
3040  * It starts at a high offset and removes keys until it can't find
3041  * any higher than new_size
3042  *
3043  * csum items that cross the new i_size are truncated to the new size
3044  * as well.
3045  *
3046  * min_type is the minimum key type to truncate down to.  If set to 0, this
3047  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3048  */
3049 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3050                                struct btrfs_root *root,
3051                                struct inode *inode,
3052                                u64 new_size, u32 min_type)
3053 {
3054         struct btrfs_path *path;
3055         struct extent_buffer *leaf;
3056         struct btrfs_file_extent_item *fi;
3057         struct btrfs_key key;
3058         struct btrfs_key found_key;
3059         u64 extent_start = 0;
3060         u64 extent_num_bytes = 0;
3061         u64 extent_offset = 0;
3062         u64 item_end = 0;
3063         u64 mask = root->sectorsize - 1;
3064         u32 found_type = (u8)-1;
3065         int found_extent;
3066         int del_item;
3067         int pending_del_nr = 0;
3068         int pending_del_slot = 0;
3069         int extent_type = -1;
3070         int encoding;
3071         int ret;
3072         int err = 0;
3073         u64 ino = btrfs_ino(inode);
3074
3075         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3076
3077         path = btrfs_alloc_path();
3078         if (!path)
3079                 return -ENOMEM;
3080         path->reada = -1;
3081
3082         if (root->ref_cows || root == root->fs_info->tree_root)
3083                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3084
3085         /*
3086          * This function is also used to drop the items in the log tree before
3087          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3088          * it is used to drop the loged items. So we shouldn't kill the delayed
3089          * items.
3090          */
3091         if (min_type == 0 && root == BTRFS_I(inode)->root)
3092                 btrfs_kill_delayed_inode_items(inode);
3093
3094         key.objectid = ino;
3095         key.offset = (u64)-1;
3096         key.type = (u8)-1;
3097
3098 search_again:
3099         path->leave_spinning = 1;
3100         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3101         if (ret < 0) {
3102                 err = ret;
3103                 goto out;
3104         }
3105
3106         if (ret > 0) {
3107                 /* there are no items in the tree for us to truncate, we're
3108                  * done
3109                  */
3110                 if (path->slots[0] == 0)
3111                         goto out;
3112                 path->slots[0]--;
3113         }
3114
3115         while (1) {
3116                 fi = NULL;
3117                 leaf = path->nodes[0];
3118                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3119                 found_type = btrfs_key_type(&found_key);
3120                 encoding = 0;
3121
3122                 if (found_key.objectid != ino)
3123                         break;
3124
3125                 if (found_type < min_type)
3126                         break;
3127
3128                 item_end = found_key.offset;
3129                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3130                         fi = btrfs_item_ptr(leaf, path->slots[0],
3131                                             struct btrfs_file_extent_item);
3132                         extent_type = btrfs_file_extent_type(leaf, fi);
3133                         encoding = btrfs_file_extent_compression(leaf, fi);
3134                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3135                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3136
3137                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3138                                 item_end +=
3139                                     btrfs_file_extent_num_bytes(leaf, fi);
3140                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3141                                 item_end += btrfs_file_extent_inline_len(leaf,
3142                                                                          fi);
3143                         }
3144                         item_end--;
3145                 }
3146                 if (found_type > min_type) {
3147                         del_item = 1;
3148                 } else {
3149                         if (item_end < new_size)
3150                                 break;
3151                         if (found_key.offset >= new_size)
3152                                 del_item = 1;
3153                         else
3154                                 del_item = 0;
3155                 }
3156                 found_extent = 0;
3157                 /* FIXME, shrink the extent if the ref count is only 1 */
3158                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3159                         goto delete;
3160
3161                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3162                         u64 num_dec;
3163                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3164                         if (!del_item && !encoding) {
3165                                 u64 orig_num_bytes =
3166                                         btrfs_file_extent_num_bytes(leaf, fi);
3167                                 extent_num_bytes = new_size -
3168                                         found_key.offset + root->sectorsize - 1;
3169                                 extent_num_bytes = extent_num_bytes &
3170                                         ~((u64)root->sectorsize - 1);
3171                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3172                                                          extent_num_bytes);
3173                                 num_dec = (orig_num_bytes -
3174                                            extent_num_bytes);
3175                                 if (root->ref_cows && extent_start != 0)
3176                                         inode_sub_bytes(inode, num_dec);
3177                                 btrfs_mark_buffer_dirty(leaf);
3178                         } else {
3179                                 extent_num_bytes =
3180                                         btrfs_file_extent_disk_num_bytes(leaf,
3181                                                                          fi);
3182                                 extent_offset = found_key.offset -
3183                                         btrfs_file_extent_offset(leaf, fi);
3184
3185                                 /* FIXME blocksize != 4096 */
3186                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3187                                 if (extent_start != 0) {
3188                                         found_extent = 1;
3189                                         if (root->ref_cows)
3190                                                 inode_sub_bytes(inode, num_dec);
3191                                 }
3192                         }
3193                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3194                         /*
3195                          * we can't truncate inline items that have had
3196                          * special encodings
3197                          */
3198                         if (!del_item &&
3199                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3200                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3201                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3202                                 u32 size = new_size - found_key.offset;
3203
3204                                 if (root->ref_cows) {
3205                                         inode_sub_bytes(inode, item_end + 1 -
3206                                                         new_size);
3207                                 }
3208                                 size =
3209                                     btrfs_file_extent_calc_inline_size(size);
3210                                 ret = btrfs_truncate_item(trans, root, path,
3211                                                           size, 1);
3212                         } else if (root->ref_cows) {
3213                                 inode_sub_bytes(inode, item_end + 1 -
3214                                                 found_key.offset);
3215                         }
3216                 }
3217 delete:
3218                 if (del_item) {
3219                         if (!pending_del_nr) {
3220                                 /* no pending yet, add ourselves */
3221                                 pending_del_slot = path->slots[0];
3222                                 pending_del_nr = 1;
3223                         } else if (pending_del_nr &&
3224                                    path->slots[0] + 1 == pending_del_slot) {
3225                                 /* hop on the pending chunk */
3226                                 pending_del_nr++;
3227                                 pending_del_slot = path->slots[0];
3228                         } else {
3229                                 BUG();
3230                         }
3231                 } else {
3232                         break;
3233                 }
3234                 if (found_extent && (root->ref_cows ||
3235                                      root == root->fs_info->tree_root)) {
3236                         btrfs_set_path_blocking(path);
3237                         ret = btrfs_free_extent(trans, root, extent_start,
3238                                                 extent_num_bytes, 0,
3239                                                 btrfs_header_owner(leaf),
3240                                                 ino, extent_offset);
3241                         BUG_ON(ret);
3242                 }
3243
3244                 if (found_type == BTRFS_INODE_ITEM_KEY)
3245                         break;
3246
3247                 if (path->slots[0] == 0 ||
3248                     path->slots[0] != pending_del_slot) {
3249                         if (root->ref_cows &&
3250                             BTRFS_I(inode)->location.objectid !=
3251                                                 BTRFS_FREE_INO_OBJECTID) {
3252                                 err = -EAGAIN;
3253                                 goto out;
3254                         }
3255                         if (pending_del_nr) {
3256                                 ret = btrfs_del_items(trans, root, path,
3257                                                 pending_del_slot,
3258                                                 pending_del_nr);
3259                                 BUG_ON(ret);
3260                                 pending_del_nr = 0;
3261                         }
3262                         btrfs_release_path(path);
3263                         goto search_again;
3264                 } else {
3265                         path->slots[0]--;
3266                 }
3267         }
3268 out:
3269         if (pending_del_nr) {
3270                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3271                                       pending_del_nr);
3272                 BUG_ON(ret);
3273         }
3274         btrfs_free_path(path);
3275         return err;
3276 }
3277
3278 /*
3279  * taken from block_truncate_page, but does cow as it zeros out
3280  * any bytes left in the last page in the file.
3281  */
3282 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3283 {
3284         struct inode *inode = mapping->host;
3285         struct btrfs_root *root = BTRFS_I(inode)->root;
3286         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3287         struct btrfs_ordered_extent *ordered;
3288         struct extent_state *cached_state = NULL;
3289         char *kaddr;
3290         u32 blocksize = root->sectorsize;
3291         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3292         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3293         struct page *page;
3294         gfp_t mask = btrfs_alloc_write_mask(mapping);
3295         int ret = 0;
3296         u64 page_start;
3297         u64 page_end;
3298
3299         if ((offset & (blocksize - 1)) == 0)
3300                 goto out;
3301         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3302         if (ret)
3303                 goto out;
3304
3305         ret = -ENOMEM;
3306 again:
3307         page = find_or_create_page(mapping, index, mask);
3308         if (!page) {
3309                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3310                 goto out;
3311         }
3312
3313         page_start = page_offset(page);
3314         page_end = page_start + PAGE_CACHE_SIZE - 1;
3315
3316         if (!PageUptodate(page)) {
3317                 ret = btrfs_readpage(NULL, page);
3318                 lock_page(page);
3319                 if (page->mapping != mapping) {
3320                         unlock_page(page);
3321                         page_cache_release(page);
3322                         goto again;
3323                 }
3324                 if (!PageUptodate(page)) {
3325                         ret = -EIO;
3326                         goto out_unlock;
3327                 }
3328         }
3329         wait_on_page_writeback(page);
3330
3331         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3332                          GFP_NOFS);
3333         set_page_extent_mapped(page);
3334
3335         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3336         if (ordered) {
3337                 unlock_extent_cached(io_tree, page_start, page_end,
3338                                      &cached_state, GFP_NOFS);
3339                 unlock_page(page);
3340                 page_cache_release(page);
3341                 btrfs_start_ordered_extent(inode, ordered, 1);
3342                 btrfs_put_ordered_extent(ordered);
3343                 goto again;
3344         }
3345
3346         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3347                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3348                           0, 0, &cached_state, GFP_NOFS);
3349
3350         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3351                                         &cached_state);
3352         if (ret) {
3353                 unlock_extent_cached(io_tree, page_start, page_end,
3354                                      &cached_state, GFP_NOFS);
3355                 goto out_unlock;
3356         }
3357
3358         ret = 0;
3359         if (offset != PAGE_CACHE_SIZE) {
3360                 kaddr = kmap(page);
3361                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3362                 flush_dcache_page(page);
3363                 kunmap(page);
3364         }
3365         ClearPageChecked(page);
3366         set_page_dirty(page);
3367         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3368                              GFP_NOFS);
3369
3370 out_unlock:
3371         if (ret)
3372                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3373         unlock_page(page);
3374         page_cache_release(page);
3375 out:
3376         return ret;
3377 }
3378
3379 /*
3380  * This function puts in dummy file extents for the area we're creating a hole
3381  * for.  So if we are truncating this file to a larger size we need to insert
3382  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3383  * the range between oldsize and size
3384  */
3385 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3386 {
3387         struct btrfs_trans_handle *trans;
3388         struct btrfs_root *root = BTRFS_I(inode)->root;
3389         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3390         struct extent_map *em = NULL;
3391         struct extent_state *cached_state = NULL;
3392         u64 mask = root->sectorsize - 1;
3393         u64 hole_start = (oldsize + mask) & ~mask;
3394         u64 block_end = (size + mask) & ~mask;
3395         u64 last_byte;
3396         u64 cur_offset;
3397         u64 hole_size;
3398         int err = 0;
3399
3400         if (size <= hole_start)
3401                 return 0;
3402
3403         while (1) {
3404                 struct btrfs_ordered_extent *ordered;
3405                 btrfs_wait_ordered_range(inode, hole_start,
3406                                          block_end - hole_start);
3407                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3408                                  &cached_state, GFP_NOFS);
3409                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3410                 if (!ordered)
3411                         break;
3412                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3413                                      &cached_state, GFP_NOFS);
3414                 btrfs_put_ordered_extent(ordered);
3415         }
3416
3417         cur_offset = hole_start;
3418         while (1) {
3419                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3420                                 block_end - cur_offset, 0);
3421                 BUG_ON(IS_ERR_OR_NULL(em));
3422                 last_byte = min(extent_map_end(em), block_end);
3423                 last_byte = (last_byte + mask) & ~mask;
3424                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3425                         u64 hint_byte = 0;
3426                         hole_size = last_byte - cur_offset;
3427
3428                         trans = btrfs_start_transaction(root, 2);
3429                         if (IS_ERR(trans)) {
3430                                 err = PTR_ERR(trans);
3431                                 break;
3432                         }
3433
3434                         err = btrfs_drop_extents(trans, inode, cur_offset,
3435                                                  cur_offset + hole_size,
3436                                                  &hint_byte, 1);
3437                         if (err) {
3438                                 btrfs_end_transaction(trans, root);
3439                                 break;
3440                         }
3441
3442                         err = btrfs_insert_file_extent(trans, root,
3443                                         btrfs_ino(inode), cur_offset, 0,
3444                                         0, hole_size, 0, hole_size,
3445                                         0, 0, 0);
3446                         if (err) {
3447                                 btrfs_end_transaction(trans, root);
3448                                 break;
3449                         }
3450
3451                         btrfs_drop_extent_cache(inode, hole_start,
3452                                         last_byte - 1, 0);
3453
3454                         btrfs_end_transaction(trans, root);
3455                 }
3456                 free_extent_map(em);
3457                 em = NULL;
3458                 cur_offset = last_byte;
3459                 if (cur_offset >= block_end)
3460                         break;
3461         }
3462
3463         free_extent_map(em);
3464         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3465                              GFP_NOFS);
3466         return err;
3467 }
3468
3469 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3470 {
3471         loff_t oldsize = i_size_read(inode);
3472         int ret;
3473
3474         if (newsize == oldsize)
3475                 return 0;
3476
3477         if (newsize > oldsize) {
3478                 i_size_write(inode, newsize);
3479                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3480                 truncate_pagecache(inode, oldsize, newsize);
3481                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3482                 if (ret) {
3483                         btrfs_setsize(inode, oldsize);
3484                         return ret;
3485                 }
3486
3487                 mark_inode_dirty(inode);
3488         } else {
3489
3490                 /*
3491                  * We're truncating a file that used to have good data down to
3492                  * zero. Make sure it gets into the ordered flush list so that
3493                  * any new writes get down to disk quickly.
3494                  */
3495                 if (newsize == 0)
3496                         BTRFS_I(inode)->ordered_data_close = 1;
3497
3498                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3499                 truncate_setsize(inode, newsize);
3500                 ret = btrfs_truncate(inode);
3501         }
3502
3503         return ret;
3504 }
3505
3506 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3507 {
3508         struct inode *inode = dentry->d_inode;
3509         struct btrfs_root *root = BTRFS_I(inode)->root;
3510         int err;
3511
3512         if (btrfs_root_readonly(root))
3513                 return -EROFS;
3514
3515         err = inode_change_ok(inode, attr);
3516         if (err)
3517                 return err;
3518
3519         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3520                 err = btrfs_setsize(inode, attr->ia_size);
3521                 if (err)
3522                         return err;
3523         }
3524
3525         if (attr->ia_valid) {
3526                 setattr_copy(inode, attr);
3527                 mark_inode_dirty(inode);
3528
3529                 if (attr->ia_valid & ATTR_MODE)
3530                         err = btrfs_acl_chmod(inode);
3531         }
3532
3533         return err;
3534 }
3535
3536 void btrfs_evict_inode(struct inode *inode)
3537 {
3538         struct btrfs_trans_handle *trans;
3539         struct btrfs_root *root = BTRFS_I(inode)->root;
3540         struct btrfs_block_rsv *rsv, *global_rsv;
3541         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3542         unsigned long nr;
3543         int ret;
3544
3545         trace_btrfs_inode_evict(inode);
3546
3547         truncate_inode_pages(&inode->i_data, 0);
3548         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3549                                btrfs_is_free_space_inode(root, inode)))
3550                 goto no_delete;
3551
3552         if (is_bad_inode(inode)) {
3553                 btrfs_orphan_del(NULL, inode);
3554                 goto no_delete;
3555         }
3556         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3557         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3558
3559         if (root->fs_info->log_root_recovering) {
3560                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3561                 goto no_delete;
3562         }
3563
3564         if (inode->i_nlink > 0) {
3565                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3566                 goto no_delete;
3567         }
3568
3569         rsv = btrfs_alloc_block_rsv(root);
3570         if (!rsv) {
3571                 btrfs_orphan_del(NULL, inode);
3572                 goto no_delete;
3573         }
3574         rsv->size = min_size;
3575         global_rsv = &root->fs_info->global_block_rsv;
3576
3577         btrfs_i_size_write(inode, 0);
3578
3579         /*
3580          * This is a bit simpler than btrfs_truncate since
3581          *
3582          * 1) We've already reserved our space for our orphan item in the
3583          *    unlink.
3584          * 2) We're going to delete the inode item, so we don't need to update
3585          *    it at all.
3586          *
3587          * So we just need to reserve some slack space in case we add bytes when
3588          * doing the truncate.
3589          */
3590         while (1) {
3591                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
3592
3593                 /*
3594                  * Try and steal from the global reserve since we will
3595                  * likely not use this space anyway, we want to try as
3596                  * hard as possible to get this to work.
3597                  */
3598                 if (ret)
3599                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3600
3601                 if (ret) {
3602                         printk(KERN_WARNING "Could not get space for a "
3603                                "delete, will truncate on mount %d\n", ret);
3604                         btrfs_orphan_del(NULL, inode);
3605                         btrfs_free_block_rsv(root, rsv);
3606                         goto no_delete;
3607                 }
3608
3609                 trans = btrfs_start_transaction(root, 0);
3610                 if (IS_ERR(trans)) {
3611                         btrfs_orphan_del(NULL, inode);
3612                         btrfs_free_block_rsv(root, rsv);
3613                         goto no_delete;
3614                 }
3615
3616                 trans->block_rsv = rsv;
3617
3618                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3619                 if (ret != -EAGAIN)
3620                         break;
3621
3622                 nr = trans->blocks_used;
3623                 btrfs_end_transaction(trans, root);
3624                 trans = NULL;
3625                 btrfs_btree_balance_dirty(root, nr);
3626         }
3627
3628         btrfs_free_block_rsv(root, rsv);
3629
3630         if (ret == 0) {
3631                 trans->block_rsv = root->orphan_block_rsv;
3632                 ret = btrfs_orphan_del(trans, inode);
3633                 BUG_ON(ret);
3634         }
3635
3636         trans->block_rsv = &root->fs_info->trans_block_rsv;
3637         if (!(root == root->fs_info->tree_root ||
3638               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3639                 btrfs_return_ino(root, btrfs_ino(inode));
3640
3641         nr = trans->blocks_used;
3642         btrfs_end_transaction(trans, root);
3643         btrfs_btree_balance_dirty(root, nr);
3644 no_delete:
3645         end_writeback(inode);
3646         return;
3647 }
3648
3649 /*
3650  * this returns the key found in the dir entry in the location pointer.
3651  * If no dir entries were found, location->objectid is 0.
3652  */
3653 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3654                                struct btrfs_key *location)
3655 {
3656         const char *name = dentry->d_name.name;
3657         int namelen = dentry->d_name.len;
3658         struct btrfs_dir_item *di;
3659         struct btrfs_path *path;
3660         struct btrfs_root *root = BTRFS_I(dir)->root;
3661         int ret = 0;
3662
3663         path = btrfs_alloc_path();
3664         if (!path)
3665                 return -ENOMEM;
3666
3667         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3668                                     namelen, 0);
3669         if (IS_ERR(di))
3670                 ret = PTR_ERR(di);
3671
3672         if (IS_ERR_OR_NULL(di))
3673                 goto out_err;
3674
3675         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3676 out:
3677         btrfs_free_path(path);
3678         return ret;
3679 out_err:
3680         location->objectid = 0;
3681         goto out;
3682 }
3683
3684 /*
3685  * when we hit a tree root in a directory, the btrfs part of the inode
3686  * needs to be changed to reflect the root directory of the tree root.  This
3687  * is kind of like crossing a mount point.
3688  */
3689 static int fixup_tree_root_location(struct btrfs_root *root,
3690                                     struct inode *dir,
3691                                     struct dentry *dentry,
3692                                     struct btrfs_key *location,
3693                                     struct btrfs_root **sub_root)
3694 {
3695         struct btrfs_path *path;
3696         struct btrfs_root *new_root;
3697         struct btrfs_root_ref *ref;
3698         struct extent_buffer *leaf;
3699         int ret;
3700         int err = 0;
3701
3702         path = btrfs_alloc_path();
3703         if (!path) {
3704                 err = -ENOMEM;
3705                 goto out;
3706         }
3707
3708         err = -ENOENT;
3709         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3710                                   BTRFS_I(dir)->root->root_key.objectid,
3711                                   location->objectid);
3712         if (ret) {
3713                 if (ret < 0)
3714                         err = ret;
3715                 goto out;
3716         }
3717
3718         leaf = path->nodes[0];
3719         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3720         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3721             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3722                 goto out;
3723
3724         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3725                                    (unsigned long)(ref + 1),
3726                                    dentry->d_name.len);
3727         if (ret)
3728                 goto out;
3729
3730         btrfs_release_path(path);
3731
3732         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3733         if (IS_ERR(new_root)) {
3734                 err = PTR_ERR(new_root);
3735                 goto out;
3736         }
3737
3738         if (btrfs_root_refs(&new_root->root_item) == 0) {
3739                 err = -ENOENT;
3740                 goto out;
3741         }
3742
3743         *sub_root = new_root;
3744         location->objectid = btrfs_root_dirid(&new_root->root_item);
3745         location->type = BTRFS_INODE_ITEM_KEY;
3746         location->offset = 0;
3747         err = 0;
3748 out:
3749         btrfs_free_path(path);
3750         return err;
3751 }
3752
3753 static void inode_tree_add(struct inode *inode)
3754 {
3755         struct btrfs_root *root = BTRFS_I(inode)->root;
3756         struct btrfs_inode *entry;
3757         struct rb_node **p;
3758         struct rb_node *parent;
3759         u64 ino = btrfs_ino(inode);
3760 again:
3761         p = &root->inode_tree.rb_node;
3762         parent = NULL;
3763
3764         if (inode_unhashed(inode))
3765                 return;
3766
3767         spin_lock(&root->inode_lock);
3768         while (*p) {
3769                 parent = *p;
3770                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3771
3772                 if (ino < btrfs_ino(&entry->vfs_inode))
3773                         p = &parent->rb_left;
3774                 else if (ino > btrfs_ino(&entry->vfs_inode))
3775                         p = &parent->rb_right;
3776                 else {
3777                         WARN_ON(!(entry->vfs_inode.i_state &
3778                                   (I_WILL_FREE | I_FREEING)));
3779                         rb_erase(parent, &root->inode_tree);
3780                         RB_CLEAR_NODE(parent);
3781                         spin_unlock(&root->inode_lock);
3782                         goto again;
3783                 }
3784         }
3785         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3786         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3787         spin_unlock(&root->inode_lock);
3788 }
3789
3790 static void inode_tree_del(struct inode *inode)
3791 {
3792         struct btrfs_root *root = BTRFS_I(inode)->root;
3793         int empty = 0;
3794
3795         spin_lock(&root->inode_lock);
3796         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3797                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3798                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3799                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3800         }
3801         spin_unlock(&root->inode_lock);
3802
3803         /*
3804          * Free space cache has inodes in the tree root, but the tree root has a
3805          * root_refs of 0, so this could end up dropping the tree root as a
3806          * snapshot, so we need the extra !root->fs_info->tree_root check to
3807          * make sure we don't drop it.
3808          */
3809         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3810             root != root->fs_info->tree_root) {
3811                 synchronize_srcu(&root->fs_info->subvol_srcu);
3812                 spin_lock(&root->inode_lock);
3813                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3814                 spin_unlock(&root->inode_lock);
3815                 if (empty)
3816                         btrfs_add_dead_root(root);
3817         }
3818 }
3819
3820 int btrfs_invalidate_inodes(struct btrfs_root *root)
3821 {
3822         struct rb_node *node;
3823         struct rb_node *prev;
3824         struct btrfs_inode *entry;
3825         struct inode *inode;
3826         u64 objectid = 0;
3827
3828         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3829
3830         spin_lock(&root->inode_lock);
3831 again:
3832         node = root->inode_tree.rb_node;
3833         prev = NULL;
3834         while (node) {
3835                 prev = node;
3836                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3837
3838                 if (objectid < btrfs_ino(&entry->vfs_inode))
3839                         node = node->rb_left;
3840                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3841                         node = node->rb_right;
3842                 else
3843                         break;
3844         }
3845         if (!node) {
3846                 while (prev) {
3847                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3848                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3849                                 node = prev;
3850                                 break;
3851                         }
3852                         prev = rb_next(prev);
3853                 }
3854         }
3855         while (node) {
3856                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3857                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3858                 inode = igrab(&entry->vfs_inode);
3859                 if (inode) {
3860                         spin_unlock(&root->inode_lock);
3861                         if (atomic_read(&inode->i_count) > 1)
3862                                 d_prune_aliases(inode);
3863                         /*
3864                          * btrfs_drop_inode will have it removed from
3865                          * the inode cache when its usage count
3866                          * hits zero.
3867                          */
3868                         iput(inode);
3869                         cond_resched();
3870                         spin_lock(&root->inode_lock);
3871                         goto again;
3872                 }
3873
3874                 if (cond_resched_lock(&root->inode_lock))
3875                         goto again;
3876
3877                 node = rb_next(node);
3878         }
3879         spin_unlock(&root->inode_lock);
3880         return 0;
3881 }
3882
3883 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3884 {
3885         struct btrfs_iget_args *args = p;
3886         inode->i_ino = args->ino;
3887         BTRFS_I(inode)->root = args->root;
3888         btrfs_set_inode_space_info(args->root, inode);
3889         return 0;
3890 }
3891
3892 static int btrfs_find_actor(struct inode *inode, void *opaque)
3893 {
3894         struct btrfs_iget_args *args = opaque;
3895         return args->ino == btrfs_ino(inode) &&
3896                 args->root == BTRFS_I(inode)->root;
3897 }
3898
3899 static struct inode *btrfs_iget_locked(struct super_block *s,
3900                                        u64 objectid,
3901                                        struct btrfs_root *root)
3902 {
3903         struct inode *inode;
3904         struct btrfs_iget_args args;
3905         args.ino = objectid;
3906         args.root = root;
3907
3908         inode = iget5_locked(s, objectid, btrfs_find_actor,
3909                              btrfs_init_locked_inode,
3910                              (void *)&args);
3911         return inode;
3912 }
3913
3914 /* Get an inode object given its location and corresponding root.
3915  * Returns in *is_new if the inode was read from disk
3916  */
3917 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3918                          struct btrfs_root *root, int *new)
3919 {
3920         struct inode *inode;
3921
3922         inode = btrfs_iget_locked(s, location->objectid, root);
3923         if (!inode)
3924                 return ERR_PTR(-ENOMEM);
3925
3926         if (inode->i_state & I_NEW) {
3927                 BTRFS_I(inode)->root = root;
3928                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3929                 btrfs_read_locked_inode(inode);
3930                 if (!is_bad_inode(inode)) {
3931                         inode_tree_add(inode);
3932                         unlock_new_inode(inode);
3933                         if (new)
3934                                 *new = 1;
3935                 } else {
3936                         unlock_new_inode(inode);
3937                         iput(inode);
3938                         inode = ERR_PTR(-ESTALE);
3939                 }
3940         }
3941
3942         return inode;
3943 }
3944
3945 static struct inode *new_simple_dir(struct super_block *s,
3946                                     struct btrfs_key *key,
3947                                     struct btrfs_root *root)
3948 {
3949         struct inode *inode = new_inode(s);
3950
3951         if (!inode)
3952                 return ERR_PTR(-ENOMEM);
3953
3954         BTRFS_I(inode)->root = root;
3955         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3956         BTRFS_I(inode)->dummy_inode = 1;
3957
3958         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3959         inode->i_op = &simple_dir_inode_operations;
3960         inode->i_fop = &simple_dir_operations;
3961         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3962         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3963
3964         return inode;
3965 }
3966
3967 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3968 {
3969         struct inode *inode;
3970         struct btrfs_root *root = BTRFS_I(dir)->root;
3971         struct btrfs_root *sub_root = root;
3972         struct btrfs_key location;
3973         int index;
3974         int ret = 0;
3975
3976         if (dentry->d_name.len > BTRFS_NAME_LEN)
3977                 return ERR_PTR(-ENAMETOOLONG);
3978
3979         if (unlikely(d_need_lookup(dentry))) {
3980                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3981                 kfree(dentry->d_fsdata);
3982                 dentry->d_fsdata = NULL;
3983                 /* This thing is hashed, drop it for now */
3984                 d_drop(dentry);
3985         } else {
3986                 ret = btrfs_inode_by_name(dir, dentry, &location);
3987         }
3988
3989         if (ret < 0)
3990                 return ERR_PTR(ret);
3991
3992         if (location.objectid == 0)
3993                 return NULL;
3994
3995         if (location.type == BTRFS_INODE_ITEM_KEY) {
3996                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3997                 return inode;
3998         }
3999
4000         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4001
4002         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4003         ret = fixup_tree_root_location(root, dir, dentry,
4004                                        &location, &sub_root);
4005         if (ret < 0) {
4006                 if (ret != -ENOENT)
4007                         inode = ERR_PTR(ret);
4008                 else
4009                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4010         } else {
4011                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4012         }
4013         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4014
4015         if (!IS_ERR(inode) && root != sub_root) {
4016                 down_read(&root->fs_info->cleanup_work_sem);
4017                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4018                         ret = btrfs_orphan_cleanup(sub_root);
4019                 up_read(&root->fs_info->cleanup_work_sem);
4020                 if (ret)
4021                         inode = ERR_PTR(ret);
4022         }
4023
4024         return inode;
4025 }
4026
4027 static int btrfs_dentry_delete(const struct dentry *dentry)
4028 {
4029         struct btrfs_root *root;
4030
4031         if (!dentry->d_inode && !IS_ROOT(dentry))
4032                 dentry = dentry->d_parent;
4033
4034         if (dentry->d_inode) {
4035                 root = BTRFS_I(dentry->d_inode)->root;
4036                 if (btrfs_root_refs(&root->root_item) == 0)
4037                         return 1;
4038         }
4039         return 0;
4040 }
4041
4042 static void btrfs_dentry_release(struct dentry *dentry)
4043 {
4044         if (dentry->d_fsdata)
4045                 kfree(dentry->d_fsdata);
4046 }
4047
4048 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4049                                    struct nameidata *nd)
4050 {
4051         struct dentry *ret;
4052
4053         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4054         if (unlikely(d_need_lookup(dentry))) {
4055                 spin_lock(&dentry->d_lock);
4056                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4057                 spin_unlock(&dentry->d_lock);
4058         }
4059         return ret;
4060 }
4061
4062 unsigned char btrfs_filetype_table[] = {
4063         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4064 };
4065
4066 static int btrfs_real_readdir(struct file *filp, void *dirent,
4067                               filldir_t filldir)
4068 {
4069         struct inode *inode = filp->f_dentry->d_inode;
4070         struct btrfs_root *root = BTRFS_I(inode)->root;
4071         struct btrfs_item *item;
4072         struct btrfs_dir_item *di;
4073         struct btrfs_key key;
4074         struct btrfs_key found_key;
4075         struct btrfs_path *path;
4076         struct list_head ins_list;
4077         struct list_head del_list;
4078         struct qstr q;
4079         int ret;
4080         struct extent_buffer *leaf;
4081         int slot;
4082         unsigned char d_type;
4083         int over = 0;
4084         u32 di_cur;
4085         u32 di_total;
4086         u32 di_len;
4087         int key_type = BTRFS_DIR_INDEX_KEY;
4088         char tmp_name[32];
4089         char *name_ptr;
4090         int name_len;
4091         int is_curr = 0;        /* filp->f_pos points to the current index? */
4092
4093         /* FIXME, use a real flag for deciding about the key type */
4094         if (root->fs_info->tree_root == root)
4095                 key_type = BTRFS_DIR_ITEM_KEY;
4096
4097         /* special case for "." */
4098         if (filp->f_pos == 0) {
4099                 over = filldir(dirent, ".", 1,
4100                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4101                 if (over)
4102                         return 0;
4103                 filp->f_pos = 1;
4104         }
4105         /* special case for .., just use the back ref */
4106         if (filp->f_pos == 1) {
4107                 u64 pino = parent_ino(filp->f_path.dentry);
4108                 over = filldir(dirent, "..", 2,
4109                                filp->f_pos, pino, DT_DIR);
4110                 if (over)
4111                         return 0;
4112                 filp->f_pos = 2;
4113         }
4114         path = btrfs_alloc_path();
4115         if (!path)
4116                 return -ENOMEM;
4117
4118         path->reada = 1;
4119
4120         if (key_type == BTRFS_DIR_INDEX_KEY) {
4121                 INIT_LIST_HEAD(&ins_list);
4122                 INIT_LIST_HEAD(&del_list);
4123                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4124         }
4125
4126         btrfs_set_key_type(&key, key_type);
4127         key.offset = filp->f_pos;
4128         key.objectid = btrfs_ino(inode);
4129
4130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4131         if (ret < 0)
4132                 goto err;
4133
4134         while (1) {
4135                 leaf = path->nodes[0];
4136                 slot = path->slots[0];
4137                 if (slot >= btrfs_header_nritems(leaf)) {
4138                         ret = btrfs_next_leaf(root, path);
4139                         if (ret < 0)
4140                                 goto err;
4141                         else if (ret > 0)
4142                                 break;
4143                         continue;
4144                 }
4145
4146                 item = btrfs_item_nr(leaf, slot);
4147                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4148
4149                 if (found_key.objectid != key.objectid)
4150                         break;
4151                 if (btrfs_key_type(&found_key) != key_type)
4152                         break;
4153                 if (found_key.offset < filp->f_pos)
4154                         goto next;
4155                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4156                     btrfs_should_delete_dir_index(&del_list,
4157                                                   found_key.offset))
4158                         goto next;
4159
4160                 filp->f_pos = found_key.offset;
4161                 is_curr = 1;
4162
4163                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4164                 di_cur = 0;
4165                 di_total = btrfs_item_size(leaf, item);
4166
4167                 while (di_cur < di_total) {
4168                         struct btrfs_key location;
4169                         struct dentry *tmp;
4170
4171                         if (verify_dir_item(root, leaf, di))
4172                                 break;
4173
4174                         name_len = btrfs_dir_name_len(leaf, di);
4175                         if (name_len <= sizeof(tmp_name)) {
4176                                 name_ptr = tmp_name;
4177                         } else {
4178                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4179                                 if (!name_ptr) {
4180                                         ret = -ENOMEM;
4181                                         goto err;
4182                                 }
4183                         }
4184                         read_extent_buffer(leaf, name_ptr,
4185                                            (unsigned long)(di + 1), name_len);
4186
4187                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4188                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4189
4190                         q.name = name_ptr;
4191                         q.len = name_len;
4192                         q.hash = full_name_hash(q.name, q.len);
4193                         tmp = d_lookup(filp->f_dentry, &q);
4194                         if (!tmp) {
4195                                 struct btrfs_key *newkey;
4196
4197                                 newkey = kzalloc(sizeof(struct btrfs_key),
4198                                                  GFP_NOFS);
4199                                 if (!newkey)
4200                                         goto no_dentry;
4201                                 tmp = d_alloc(filp->f_dentry, &q);
4202                                 if (!tmp) {
4203                                         kfree(newkey);
4204                                         dput(tmp);
4205                                         goto no_dentry;
4206                                 }
4207                                 memcpy(newkey, &location,
4208                                        sizeof(struct btrfs_key));
4209                                 tmp->d_fsdata = newkey;
4210                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4211                                 d_rehash(tmp);
4212                                 dput(tmp);
4213                         } else {
4214                                 dput(tmp);
4215                         }
4216 no_dentry:
4217                         /* is this a reference to our own snapshot? If so
4218                          * skip it
4219                          */
4220                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4221                             location.objectid == root->root_key.objectid) {
4222                                 over = 0;
4223                                 goto skip;
4224                         }
4225                         over = filldir(dirent, name_ptr, name_len,
4226                                        found_key.offset, location.objectid,
4227                                        d_type);
4228
4229 skip:
4230                         if (name_ptr != tmp_name)
4231                                 kfree(name_ptr);
4232
4233                         if (over)
4234                                 goto nopos;
4235                         di_len = btrfs_dir_name_len(leaf, di) +
4236                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4237                         di_cur += di_len;
4238                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4239                 }
4240 next:
4241                 path->slots[0]++;
4242         }
4243
4244         if (key_type == BTRFS_DIR_INDEX_KEY) {
4245                 if (is_curr)
4246                         filp->f_pos++;
4247                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4248                                                       &ins_list);
4249                 if (ret)
4250                         goto nopos;
4251         }
4252
4253         /* Reached end of directory/root. Bump pos past the last item. */
4254         if (key_type == BTRFS_DIR_INDEX_KEY)
4255                 /*
4256                  * 32-bit glibc will use getdents64, but then strtol -
4257                  * so the last number we can serve is this.
4258                  */
4259                 filp->f_pos = 0x7fffffff;
4260         else
4261                 filp->f_pos++;
4262 nopos:
4263         ret = 0;
4264 err:
4265         if (key_type == BTRFS_DIR_INDEX_KEY)
4266                 btrfs_put_delayed_items(&ins_list, &del_list);
4267         btrfs_free_path(path);
4268         return ret;
4269 }
4270
4271 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4272 {
4273         struct btrfs_root *root = BTRFS_I(inode)->root;
4274         struct btrfs_trans_handle *trans;
4275         int ret = 0;
4276         bool nolock = false;
4277
4278         if (BTRFS_I(inode)->dummy_inode)
4279                 return 0;
4280
4281         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4282                 nolock = true;
4283
4284         if (wbc->sync_mode == WB_SYNC_ALL) {
4285                 if (nolock)
4286                         trans = btrfs_join_transaction_nolock(root);
4287                 else
4288                         trans = btrfs_join_transaction(root);
4289                 if (IS_ERR(trans))
4290                         return PTR_ERR(trans);
4291                 if (nolock)
4292                         ret = btrfs_end_transaction_nolock(trans, root);
4293                 else
4294                         ret = btrfs_commit_transaction(trans, root);
4295         }
4296         return ret;
4297 }
4298
4299 /*
4300  * This is somewhat expensive, updating the tree every time the
4301  * inode changes.  But, it is most likely to find the inode in cache.
4302  * FIXME, needs more benchmarking...there are no reasons other than performance
4303  * to keep or drop this code.
4304  */
4305 void btrfs_dirty_inode(struct inode *inode, int flags)
4306 {
4307         struct btrfs_root *root = BTRFS_I(inode)->root;
4308         struct btrfs_trans_handle *trans;
4309         int ret;
4310
4311         if (BTRFS_I(inode)->dummy_inode)
4312                 return;
4313
4314         trans = btrfs_join_transaction(root);
4315         BUG_ON(IS_ERR(trans));
4316
4317         ret = btrfs_update_inode(trans, root, inode);
4318         if (ret && ret == -ENOSPC) {
4319                 /* whoops, lets try again with the full transaction */
4320                 btrfs_end_transaction(trans, root);
4321                 trans = btrfs_start_transaction(root, 1);
4322                 if (IS_ERR(trans)) {
4323                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4324                                        "dirty  inode %llu error %ld\n",
4325                                        (unsigned long long)btrfs_ino(inode),
4326                                        PTR_ERR(trans));
4327                         return;
4328                 }
4329
4330                 ret = btrfs_update_inode(trans, root, inode);
4331                 if (ret) {
4332                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4333                                        "dirty  inode %llu error %d\n",
4334                                        (unsigned long long)btrfs_ino(inode),
4335                                        ret);
4336                 }
4337         }
4338         btrfs_end_transaction(trans, root);
4339         if (BTRFS_I(inode)->delayed_node)
4340                 btrfs_balance_delayed_items(root);
4341 }
4342
4343 /*
4344  * find the highest existing sequence number in a directory
4345  * and then set the in-memory index_cnt variable to reflect
4346  * free sequence numbers
4347  */
4348 static int btrfs_set_inode_index_count(struct inode *inode)
4349 {
4350         struct btrfs_root *root = BTRFS_I(inode)->root;
4351         struct btrfs_key key, found_key;
4352         struct btrfs_path *path;
4353         struct extent_buffer *leaf;
4354         int ret;
4355
4356         key.objectid = btrfs_ino(inode);
4357         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4358         key.offset = (u64)-1;
4359
4360         path = btrfs_alloc_path();
4361         if (!path)
4362                 return -ENOMEM;
4363
4364         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4365         if (ret < 0)
4366                 goto out;
4367         /* FIXME: we should be able to handle this */
4368         if (ret == 0)
4369                 goto out;
4370         ret = 0;
4371
4372         /*
4373          * MAGIC NUMBER EXPLANATION:
4374          * since we search a directory based on f_pos we have to start at 2
4375          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4376          * else has to start at 2
4377          */
4378         if (path->slots[0] == 0) {
4379                 BTRFS_I(inode)->index_cnt = 2;
4380                 goto out;
4381         }
4382
4383         path->slots[0]--;
4384
4385         leaf = path->nodes[0];
4386         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4387
4388         if (found_key.objectid != btrfs_ino(inode) ||
4389             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4390                 BTRFS_I(inode)->index_cnt = 2;
4391                 goto out;
4392         }
4393
4394         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4395 out:
4396         btrfs_free_path(path);
4397         return ret;
4398 }
4399
4400 /*
4401  * helper to find a free sequence number in a given directory.  This current
4402  * code is very simple, later versions will do smarter things in the btree
4403  */
4404 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4405 {
4406         int ret = 0;
4407
4408         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4409                 ret = btrfs_inode_delayed_dir_index_count(dir);
4410                 if (ret) {
4411                         ret = btrfs_set_inode_index_count(dir);
4412                         if (ret)
4413                                 return ret;
4414                 }
4415         }
4416
4417         *index = BTRFS_I(dir)->index_cnt;
4418         BTRFS_I(dir)->index_cnt++;
4419
4420         return ret;
4421 }
4422
4423 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4424                                      struct btrfs_root *root,
4425                                      struct inode *dir,
4426                                      const char *name, int name_len,
4427                                      u64 ref_objectid, u64 objectid, int mode,
4428                                      u64 *index)
4429 {
4430         struct inode *inode;
4431         struct btrfs_inode_item *inode_item;
4432         struct btrfs_key *location;
4433         struct btrfs_path *path;
4434         struct btrfs_inode_ref *ref;
4435         struct btrfs_key key[2];
4436         u32 sizes[2];
4437         unsigned long ptr;
4438         int ret;
4439         int owner;
4440
4441         path = btrfs_alloc_path();
4442         if (!path)
4443                 return ERR_PTR(-ENOMEM);
4444
4445         inode = new_inode(root->fs_info->sb);
4446         if (!inode) {
4447                 btrfs_free_path(path);
4448                 return ERR_PTR(-ENOMEM);
4449         }
4450
4451         /*
4452          * we have to initialize this early, so we can reclaim the inode
4453          * number if we fail afterwards in this function.
4454          */
4455         inode->i_ino = objectid;
4456
4457         if (dir) {
4458                 trace_btrfs_inode_request(dir);
4459
4460                 ret = btrfs_set_inode_index(dir, index);
4461                 if (ret) {
4462                         btrfs_free_path(path);
4463                         iput(inode);
4464                         return ERR_PTR(ret);
4465                 }
4466         }
4467         /*
4468          * index_cnt is ignored for everything but a dir,
4469          * btrfs_get_inode_index_count has an explanation for the magic
4470          * number
4471          */
4472         BTRFS_I(inode)->index_cnt = 2;
4473         BTRFS_I(inode)->root = root;
4474         BTRFS_I(inode)->generation = trans->transid;
4475         inode->i_generation = BTRFS_I(inode)->generation;
4476         btrfs_set_inode_space_info(root, inode);
4477
4478         if (S_ISDIR(mode))
4479                 owner = 0;
4480         else
4481                 owner = 1;
4482
4483         key[0].objectid = objectid;
4484         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4485         key[0].offset = 0;
4486
4487         key[1].objectid = objectid;
4488         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4489         key[1].offset = ref_objectid;
4490
4491         sizes[0] = sizeof(struct btrfs_inode_item);
4492         sizes[1] = name_len + sizeof(*ref);
4493
4494         path->leave_spinning = 1;
4495         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4496         if (ret != 0)
4497                 goto fail;
4498
4499         inode_init_owner(inode, dir, mode);
4500         inode_set_bytes(inode, 0);
4501         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4502         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4503                                   struct btrfs_inode_item);
4504         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4505
4506         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4507                              struct btrfs_inode_ref);
4508         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4509         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4510         ptr = (unsigned long)(ref + 1);
4511         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4512
4513         btrfs_mark_buffer_dirty(path->nodes[0]);
4514         btrfs_free_path(path);
4515
4516         location = &BTRFS_I(inode)->location;
4517         location->objectid = objectid;
4518         location->offset = 0;
4519         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4520
4521         btrfs_inherit_iflags(inode, dir);
4522
4523         if (S_ISREG(mode)) {
4524                 if (btrfs_test_opt(root, NODATASUM))
4525                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4526                 if (btrfs_test_opt(root, NODATACOW) ||
4527                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4528                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4529         }
4530
4531         insert_inode_hash(inode);
4532         inode_tree_add(inode);
4533
4534         trace_btrfs_inode_new(inode);
4535         btrfs_set_inode_last_trans(trans, inode);
4536
4537         return inode;
4538 fail:
4539         if (dir)
4540                 BTRFS_I(dir)->index_cnt--;
4541         btrfs_free_path(path);
4542         iput(inode);
4543         return ERR_PTR(ret);
4544 }
4545
4546 static inline u8 btrfs_inode_type(struct inode *inode)
4547 {
4548         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4549 }
4550
4551 /*
4552  * utility function to add 'inode' into 'parent_inode' with
4553  * a give name and a given sequence number.
4554  * if 'add_backref' is true, also insert a backref from the
4555  * inode to the parent directory.
4556  */
4557 int btrfs_add_link(struct btrfs_trans_handle *trans,
4558                    struct inode *parent_inode, struct inode *inode,
4559                    const char *name, int name_len, int add_backref, u64 index)
4560 {
4561         int ret = 0;
4562         struct btrfs_key key;
4563         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4564         u64 ino = btrfs_ino(inode);
4565         u64 parent_ino = btrfs_ino(parent_inode);
4566
4567         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4568                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4569         } else {
4570                 key.objectid = ino;
4571                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4572                 key.offset = 0;
4573         }
4574
4575         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4576                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4577                                          key.objectid, root->root_key.objectid,
4578                                          parent_ino, index, name, name_len);
4579         } else if (add_backref) {
4580                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4581                                              parent_ino, index);
4582         }
4583
4584         if (ret == 0) {
4585                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4586                                             parent_inode, &key,
4587                                             btrfs_inode_type(inode), index);
4588                 BUG_ON(ret);
4589
4590                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4591                                    name_len * 2);
4592                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4593                 ret = btrfs_update_inode(trans, root, parent_inode);
4594         }
4595         return ret;
4596 }
4597
4598 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4599                             struct inode *dir, struct dentry *dentry,
4600                             struct inode *inode, int backref, u64 index)
4601 {
4602         int err = btrfs_add_link(trans, dir, inode,
4603                                  dentry->d_name.name, dentry->d_name.len,
4604                                  backref, index);
4605         if (!err) {
4606                 d_instantiate(dentry, inode);
4607                 return 0;
4608         }
4609         if (err > 0)
4610                 err = -EEXIST;
4611         return err;
4612 }
4613
4614 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4615                         int mode, dev_t rdev)
4616 {
4617         struct btrfs_trans_handle *trans;
4618         struct btrfs_root *root = BTRFS_I(dir)->root;
4619         struct inode *inode = NULL;
4620         int err;
4621         int drop_inode = 0;
4622         u64 objectid;
4623         unsigned long nr = 0;
4624         u64 index = 0;
4625
4626         if (!new_valid_dev(rdev))
4627                 return -EINVAL;
4628
4629         /*
4630          * 2 for inode item and ref
4631          * 2 for dir items
4632          * 1 for xattr if selinux is on
4633          */
4634         trans = btrfs_start_transaction(root, 5);
4635         if (IS_ERR(trans))
4636                 return PTR_ERR(trans);
4637
4638         err = btrfs_find_free_ino(root, &objectid);
4639         if (err)
4640                 goto out_unlock;
4641
4642         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4643                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4644                                 mode, &index);
4645         if (IS_ERR(inode)) {
4646                 err = PTR_ERR(inode);
4647                 goto out_unlock;
4648         }
4649
4650         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4651         if (err) {
4652                 drop_inode = 1;
4653                 goto out_unlock;
4654         }
4655
4656         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4657         if (err)
4658                 drop_inode = 1;
4659         else {
4660                 inode->i_op = &btrfs_special_inode_operations;
4661                 init_special_inode(inode, inode->i_mode, rdev);
4662                 btrfs_update_inode(trans, root, inode);
4663         }
4664 out_unlock:
4665         nr = trans->blocks_used;
4666         btrfs_end_transaction_throttle(trans, root);
4667         btrfs_btree_balance_dirty(root, nr);
4668         if (drop_inode) {
4669                 inode_dec_link_count(inode);
4670                 iput(inode);
4671         }
4672         return err;
4673 }
4674
4675 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4676                         int mode, struct nameidata *nd)
4677 {
4678         struct btrfs_trans_handle *trans;
4679         struct btrfs_root *root = BTRFS_I(dir)->root;
4680         struct inode *inode = NULL;
4681         int drop_inode = 0;
4682         int err;
4683         unsigned long nr = 0;
4684         u64 objectid;
4685         u64 index = 0;
4686
4687         /*
4688          * 2 for inode item and ref
4689          * 2 for dir items
4690          * 1 for xattr if selinux is on
4691          */
4692         trans = btrfs_start_transaction(root, 5);
4693         if (IS_ERR(trans))
4694                 return PTR_ERR(trans);
4695
4696         err = btrfs_find_free_ino(root, &objectid);
4697         if (err)
4698                 goto out_unlock;
4699
4700         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4701                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4702                                 mode, &index);
4703         if (IS_ERR(inode)) {
4704                 err = PTR_ERR(inode);
4705                 goto out_unlock;
4706         }
4707
4708         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4709         if (err) {
4710                 drop_inode = 1;
4711                 goto out_unlock;
4712         }
4713
4714         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4715         if (err)
4716                 drop_inode = 1;
4717         else {
4718                 inode->i_mapping->a_ops = &btrfs_aops;
4719                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4720                 inode->i_fop = &btrfs_file_operations;
4721                 inode->i_op = &btrfs_file_inode_operations;
4722                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4723         }
4724 out_unlock:
4725         nr = trans->blocks_used;
4726         btrfs_end_transaction_throttle(trans, root);
4727         if (drop_inode) {
4728                 inode_dec_link_count(inode);
4729                 iput(inode);
4730         }
4731         btrfs_btree_balance_dirty(root, nr);
4732         return err;
4733 }
4734
4735 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4736                       struct dentry *dentry)
4737 {
4738         struct btrfs_trans_handle *trans;
4739         struct btrfs_root *root = BTRFS_I(dir)->root;
4740         struct inode *inode = old_dentry->d_inode;
4741         u64 index;
4742         unsigned long nr = 0;
4743         int err;
4744         int drop_inode = 0;
4745
4746         /* do not allow sys_link's with other subvols of the same device */
4747         if (root->objectid != BTRFS_I(inode)->root->objectid)
4748                 return -EXDEV;
4749
4750         if (inode->i_nlink == ~0U)
4751                 return -EMLINK;
4752
4753         err = btrfs_set_inode_index(dir, &index);
4754         if (err)
4755                 goto fail;
4756
4757         /*
4758          * 2 items for inode and inode ref
4759          * 2 items for dir items
4760          * 1 item for parent inode
4761          */
4762         trans = btrfs_start_transaction(root, 5);
4763         if (IS_ERR(trans)) {
4764                 err = PTR_ERR(trans);
4765                 goto fail;
4766         }
4767
4768         btrfs_inc_nlink(inode);
4769         inode->i_ctime = CURRENT_TIME;
4770         ihold(inode);
4771
4772         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4773
4774         if (err) {
4775                 drop_inode = 1;
4776         } else {
4777                 struct dentry *parent = dentry->d_parent;
4778                 err = btrfs_update_inode(trans, root, inode);
4779                 BUG_ON(err);
4780                 btrfs_log_new_name(trans, inode, NULL, parent);
4781         }
4782
4783         nr = trans->blocks_used;
4784         btrfs_end_transaction_throttle(trans, root);
4785 fail:
4786         if (drop_inode) {
4787                 inode_dec_link_count(inode);
4788                 iput(inode);
4789         }
4790         btrfs_btree_balance_dirty(root, nr);
4791         return err;
4792 }
4793
4794 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4795 {
4796         struct inode *inode = NULL;
4797         struct btrfs_trans_handle *trans;
4798         struct btrfs_root *root = BTRFS_I(dir)->root;
4799         int err = 0;
4800         int drop_on_err = 0;
4801         u64 objectid = 0;
4802         u64 index = 0;
4803         unsigned long nr = 1;
4804
4805         /*
4806          * 2 items for inode and ref
4807          * 2 items for dir items
4808          * 1 for xattr if selinux is on
4809          */
4810         trans = btrfs_start_transaction(root, 5);
4811         if (IS_ERR(trans))
4812                 return PTR_ERR(trans);
4813
4814         err = btrfs_find_free_ino(root, &objectid);
4815         if (err)
4816                 goto out_fail;
4817
4818         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4819                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4820                                 S_IFDIR | mode, &index);
4821         if (IS_ERR(inode)) {
4822                 err = PTR_ERR(inode);
4823                 goto out_fail;
4824         }
4825
4826         drop_on_err = 1;
4827
4828         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4829         if (err)
4830                 goto out_fail;
4831
4832         inode->i_op = &btrfs_dir_inode_operations;
4833         inode->i_fop = &btrfs_dir_file_operations;
4834
4835         btrfs_i_size_write(inode, 0);
4836         err = btrfs_update_inode(trans, root, inode);
4837         if (err)
4838                 goto out_fail;
4839
4840         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4841                              dentry->d_name.len, 0, index);
4842         if (err)
4843                 goto out_fail;
4844
4845         d_instantiate(dentry, inode);
4846         drop_on_err = 0;
4847
4848 out_fail:
4849         nr = trans->blocks_used;
4850         btrfs_end_transaction_throttle(trans, root);
4851         if (drop_on_err)
4852                 iput(inode);
4853         btrfs_btree_balance_dirty(root, nr);
4854         return err;
4855 }
4856
4857 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4858  * and an extent that you want to insert, deal with overlap and insert
4859  * the new extent into the tree.
4860  */
4861 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4862                                 struct extent_map *existing,
4863                                 struct extent_map *em,
4864                                 u64 map_start, u64 map_len)
4865 {
4866         u64 start_diff;
4867
4868         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4869         start_diff = map_start - em->start;
4870         em->start = map_start;
4871         em->len = map_len;
4872         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4873             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4874                 em->block_start += start_diff;
4875                 em->block_len -= start_diff;
4876         }
4877         return add_extent_mapping(em_tree, em);
4878 }
4879
4880 static noinline int uncompress_inline(struct btrfs_path *path,
4881                                       struct inode *inode, struct page *page,
4882                                       size_t pg_offset, u64 extent_offset,
4883                                       struct btrfs_file_extent_item *item)
4884 {
4885         int ret;
4886         struct extent_buffer *leaf = path->nodes[0];
4887         char *tmp;
4888         size_t max_size;
4889         unsigned long inline_size;
4890         unsigned long ptr;
4891         int compress_type;
4892
4893         WARN_ON(pg_offset != 0);
4894         compress_type = btrfs_file_extent_compression(leaf, item);
4895         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4896         inline_size = btrfs_file_extent_inline_item_len(leaf,
4897                                         btrfs_item_nr(leaf, path->slots[0]));
4898         tmp = kmalloc(inline_size, GFP_NOFS);
4899         if (!tmp)
4900                 return -ENOMEM;
4901         ptr = btrfs_file_extent_inline_start(item);
4902
4903         read_extent_buffer(leaf, tmp, ptr, inline_size);
4904
4905         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4906         ret = btrfs_decompress(compress_type, tmp, page,
4907                                extent_offset, inline_size, max_size);
4908         if (ret) {
4909                 char *kaddr = kmap_atomic(page, KM_USER0);
4910                 unsigned long copy_size = min_t(u64,
4911                                   PAGE_CACHE_SIZE - pg_offset,
4912                                   max_size - extent_offset);
4913                 memset(kaddr + pg_offset, 0, copy_size);
4914                 kunmap_atomic(kaddr, KM_USER0);
4915         }
4916         kfree(tmp);
4917         return 0;
4918 }
4919
4920 /*
4921  * a bit scary, this does extent mapping from logical file offset to the disk.
4922  * the ugly parts come from merging extents from the disk with the in-ram
4923  * representation.  This gets more complex because of the data=ordered code,
4924  * where the in-ram extents might be locked pending data=ordered completion.
4925  *
4926  * This also copies inline extents directly into the page.
4927  */
4928
4929 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4930                                     size_t pg_offset, u64 start, u64 len,
4931                                     int create)
4932 {
4933         int ret;
4934         int err = 0;
4935         u64 bytenr;
4936         u64 extent_start = 0;
4937         u64 extent_end = 0;
4938         u64 objectid = btrfs_ino(inode);
4939         u32 found_type;
4940         struct btrfs_path *path = NULL;
4941         struct btrfs_root *root = BTRFS_I(inode)->root;
4942         struct btrfs_file_extent_item *item;
4943         struct extent_buffer *leaf;
4944         struct btrfs_key found_key;
4945         struct extent_map *em = NULL;
4946         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4947         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4948         struct btrfs_trans_handle *trans = NULL;
4949         int compress_type;
4950
4951 again:
4952         read_lock(&em_tree->lock);
4953         em = lookup_extent_mapping(em_tree, start, len);
4954         if (em)
4955                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4956         read_unlock(&em_tree->lock);
4957
4958         if (em) {
4959                 if (em->start > start || em->start + em->len <= start)
4960                         free_extent_map(em);
4961                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4962                         free_extent_map(em);
4963                 else
4964                         goto out;
4965         }
4966         em = alloc_extent_map();
4967         if (!em) {
4968                 err = -ENOMEM;
4969                 goto out;
4970         }
4971         em->bdev = root->fs_info->fs_devices->latest_bdev;
4972         em->start = EXTENT_MAP_HOLE;
4973         em->orig_start = EXTENT_MAP_HOLE;
4974         em->len = (u64)-1;
4975         em->block_len = (u64)-1;
4976
4977         if (!path) {
4978                 path = btrfs_alloc_path();
4979                 if (!path) {
4980                         err = -ENOMEM;
4981                         goto out;
4982                 }
4983                 /*
4984                  * Chances are we'll be called again, so go ahead and do
4985                  * readahead
4986                  */
4987                 path->reada = 1;
4988         }
4989
4990         ret = btrfs_lookup_file_extent(trans, root, path,
4991                                        objectid, start, trans != NULL);
4992         if (ret < 0) {
4993                 err = ret;
4994                 goto out;
4995         }
4996
4997         if (ret != 0) {
4998                 if (path->slots[0] == 0)
4999                         goto not_found;
5000                 path->slots[0]--;
5001         }
5002
5003         leaf = path->nodes[0];
5004         item = btrfs_item_ptr(leaf, path->slots[0],
5005                               struct btrfs_file_extent_item);
5006         /* are we inside the extent that was found? */
5007         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5008         found_type = btrfs_key_type(&found_key);
5009         if (found_key.objectid != objectid ||
5010             found_type != BTRFS_EXTENT_DATA_KEY) {
5011                 goto not_found;
5012         }
5013
5014         found_type = btrfs_file_extent_type(leaf, item);
5015         extent_start = found_key.offset;
5016         compress_type = btrfs_file_extent_compression(leaf, item);
5017         if (found_type == BTRFS_FILE_EXTENT_REG ||
5018             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5019                 extent_end = extent_start +
5020                        btrfs_file_extent_num_bytes(leaf, item);
5021         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5022                 size_t size;
5023                 size = btrfs_file_extent_inline_len(leaf, item);
5024                 extent_end = (extent_start + size + root->sectorsize - 1) &
5025                         ~((u64)root->sectorsize - 1);
5026         }
5027
5028         if (start >= extent_end) {
5029                 path->slots[0]++;
5030                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5031                         ret = btrfs_next_leaf(root, path);
5032                         if (ret < 0) {
5033                                 err = ret;
5034                                 goto out;
5035                         }
5036                         if (ret > 0)
5037                                 goto not_found;
5038                         leaf = path->nodes[0];
5039                 }
5040                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5041                 if (found_key.objectid != objectid ||
5042                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5043                         goto not_found;
5044                 if (start + len <= found_key.offset)
5045                         goto not_found;
5046                 em->start = start;
5047                 em->len = found_key.offset - start;
5048                 goto not_found_em;
5049         }
5050
5051         if (found_type == BTRFS_FILE_EXTENT_REG ||
5052             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5053                 em->start = extent_start;
5054                 em->len = extent_end - extent_start;
5055                 em->orig_start = extent_start -
5056                                  btrfs_file_extent_offset(leaf, item);
5057                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5058                 if (bytenr == 0) {
5059                         em->block_start = EXTENT_MAP_HOLE;
5060                         goto insert;
5061                 }
5062                 if (compress_type != BTRFS_COMPRESS_NONE) {
5063                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5064                         em->compress_type = compress_type;
5065                         em->block_start = bytenr;
5066                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5067                                                                          item);
5068                 } else {
5069                         bytenr += btrfs_file_extent_offset(leaf, item);
5070                         em->block_start = bytenr;
5071                         em->block_len = em->len;
5072                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5073                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5074                 }
5075                 goto insert;
5076         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5077                 unsigned long ptr;
5078                 char *map;
5079                 size_t size;
5080                 size_t extent_offset;
5081                 size_t copy_size;
5082
5083                 em->block_start = EXTENT_MAP_INLINE;
5084                 if (!page || create) {
5085                         em->start = extent_start;
5086                         em->len = extent_end - extent_start;
5087                         goto out;
5088                 }
5089
5090                 size = btrfs_file_extent_inline_len(leaf, item);
5091                 extent_offset = page_offset(page) + pg_offset - extent_start;
5092                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5093                                 size - extent_offset);
5094                 em->start = extent_start + extent_offset;
5095                 em->len = (copy_size + root->sectorsize - 1) &
5096                         ~((u64)root->sectorsize - 1);
5097                 em->orig_start = EXTENT_MAP_INLINE;
5098                 if (compress_type) {
5099                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5100                         em->compress_type = compress_type;
5101                 }
5102                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5103                 if (create == 0 && !PageUptodate(page)) {
5104                         if (btrfs_file_extent_compression(leaf, item) !=
5105                             BTRFS_COMPRESS_NONE) {
5106                                 ret = uncompress_inline(path, inode, page,
5107                                                         pg_offset,
5108                                                         extent_offset, item);
5109                                 BUG_ON(ret);
5110                         } else {
5111                                 map = kmap(page);
5112                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5113                                                    copy_size);
5114                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5115                                         memset(map + pg_offset + copy_size, 0,
5116                                                PAGE_CACHE_SIZE - pg_offset -
5117                                                copy_size);
5118                                 }
5119                                 kunmap(page);
5120                         }
5121                         flush_dcache_page(page);
5122                 } else if (create && PageUptodate(page)) {
5123                         WARN_ON(1);
5124                         if (!trans) {
5125                                 kunmap(page);
5126                                 free_extent_map(em);
5127                                 em = NULL;
5128
5129                                 btrfs_release_path(path);
5130                                 trans = btrfs_join_transaction(root);
5131
5132                                 if (IS_ERR(trans))
5133                                         return ERR_CAST(trans);
5134                                 goto again;
5135                         }
5136                         map = kmap(page);
5137                         write_extent_buffer(leaf, map + pg_offset, ptr,
5138                                             copy_size);
5139                         kunmap(page);
5140                         btrfs_mark_buffer_dirty(leaf);
5141                 }
5142                 set_extent_uptodate(io_tree, em->start,
5143                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5144                 goto insert;
5145         } else {
5146                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5147                 WARN_ON(1);
5148         }
5149 not_found:
5150         em->start = start;
5151         em->len = len;
5152 not_found_em:
5153         em->block_start = EXTENT_MAP_HOLE;
5154         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5155 insert:
5156         btrfs_release_path(path);
5157         if (em->start > start || extent_map_end(em) <= start) {
5158                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5159                        "[%llu %llu]\n", (unsigned long long)em->start,
5160                        (unsigned long long)em->len,
5161                        (unsigned long long)start,
5162                        (unsigned long long)len);
5163                 err = -EIO;
5164                 goto out;
5165         }
5166
5167         err = 0;
5168         write_lock(&em_tree->lock);
5169         ret = add_extent_mapping(em_tree, em);
5170         /* it is possible that someone inserted the extent into the tree
5171          * while we had the lock dropped.  It is also possible that
5172          * an overlapping map exists in the tree
5173          */
5174         if (ret == -EEXIST) {
5175                 struct extent_map *existing;
5176
5177                 ret = 0;
5178
5179                 existing = lookup_extent_mapping(em_tree, start, len);
5180                 if (existing && (existing->start > start ||
5181                     existing->start + existing->len <= start)) {
5182                         free_extent_map(existing);
5183                         existing = NULL;
5184                 }
5185                 if (!existing) {
5186                         existing = lookup_extent_mapping(em_tree, em->start,
5187                                                          em->len);
5188                         if (existing) {
5189                                 err = merge_extent_mapping(em_tree, existing,
5190                                                            em, start,
5191                                                            root->sectorsize);
5192                                 free_extent_map(existing);
5193                                 if (err) {
5194                                         free_extent_map(em);
5195                                         em = NULL;
5196                                 }
5197                         } else {
5198                                 err = -EIO;
5199                                 free_extent_map(em);
5200                                 em = NULL;
5201                         }
5202                 } else {
5203                         free_extent_map(em);
5204                         em = existing;
5205                         err = 0;
5206                 }
5207         }
5208         write_unlock(&em_tree->lock);
5209 out:
5210
5211         trace_btrfs_get_extent(root, em);
5212
5213         if (path)
5214                 btrfs_free_path(path);
5215         if (trans) {
5216                 ret = btrfs_end_transaction(trans, root);
5217                 if (!err)
5218                         err = ret;
5219         }
5220         if (err) {
5221                 free_extent_map(em);
5222                 return ERR_PTR(err);
5223         }
5224         return em;
5225 }
5226
5227 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5228                                            size_t pg_offset, u64 start, u64 len,
5229                                            int create)
5230 {
5231         struct extent_map *em;
5232         struct extent_map *hole_em = NULL;
5233         u64 range_start = start;
5234         u64 end;
5235         u64 found;
5236         u64 found_end;
5237         int err = 0;
5238
5239         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5240         if (IS_ERR(em))
5241                 return em;
5242         if (em) {
5243                 /*
5244                  * if our em maps to a hole, there might
5245                  * actually be delalloc bytes behind it
5246                  */
5247                 if (em->block_start != EXTENT_MAP_HOLE)
5248                         return em;
5249                 else
5250                         hole_em = em;
5251         }
5252
5253         /* check to see if we've wrapped (len == -1 or similar) */
5254         end = start + len;
5255         if (end < start)
5256                 end = (u64)-1;
5257         else
5258                 end -= 1;
5259
5260         em = NULL;
5261
5262         /* ok, we didn't find anything, lets look for delalloc */
5263         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5264                                  end, len, EXTENT_DELALLOC, 1);
5265         found_end = range_start + found;
5266         if (found_end < range_start)
5267                 found_end = (u64)-1;
5268
5269         /*
5270          * we didn't find anything useful, return
5271          * the original results from get_extent()
5272          */
5273         if (range_start > end || found_end <= start) {
5274                 em = hole_em;
5275                 hole_em = NULL;
5276                 goto out;
5277         }
5278
5279         /* adjust the range_start to make sure it doesn't
5280          * go backwards from the start they passed in
5281          */
5282         range_start = max(start,range_start);
5283         found = found_end - range_start;
5284
5285         if (found > 0) {
5286                 u64 hole_start = start;
5287                 u64 hole_len = len;
5288
5289                 em = alloc_extent_map();
5290                 if (!em) {
5291                         err = -ENOMEM;
5292                         goto out;
5293                 }
5294                 /*
5295                  * when btrfs_get_extent can't find anything it
5296                  * returns one huge hole
5297                  *
5298                  * make sure what it found really fits our range, and
5299                  * adjust to make sure it is based on the start from
5300                  * the caller
5301                  */
5302                 if (hole_em) {
5303                         u64 calc_end = extent_map_end(hole_em);
5304
5305                         if (calc_end <= start || (hole_em->start > end)) {
5306                                 free_extent_map(hole_em);
5307                                 hole_em = NULL;
5308                         } else {
5309                                 hole_start = max(hole_em->start, start);
5310                                 hole_len = calc_end - hole_start;
5311                         }
5312                 }
5313                 em->bdev = NULL;
5314                 if (hole_em && range_start > hole_start) {
5315                         /* our hole starts before our delalloc, so we
5316                          * have to return just the parts of the hole
5317                          * that go until  the delalloc starts
5318                          */
5319                         em->len = min(hole_len,
5320                                       range_start - hole_start);
5321                         em->start = hole_start;
5322                         em->orig_start = hole_start;
5323                         /*
5324                          * don't adjust block start at all,
5325                          * it is fixed at EXTENT_MAP_HOLE
5326                          */
5327                         em->block_start = hole_em->block_start;
5328                         em->block_len = hole_len;
5329                 } else {
5330                         em->start = range_start;
5331                         em->len = found;
5332                         em->orig_start = range_start;
5333                         em->block_start = EXTENT_MAP_DELALLOC;
5334                         em->block_len = found;
5335                 }
5336         } else if (hole_em) {
5337                 return hole_em;
5338         }
5339 out:
5340
5341         free_extent_map(hole_em);
5342         if (err) {
5343                 free_extent_map(em);
5344                 return ERR_PTR(err);
5345         }
5346         return em;
5347 }
5348
5349 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5350                                                   struct extent_map *em,
5351                                                   u64 start, u64 len)
5352 {
5353         struct btrfs_root *root = BTRFS_I(inode)->root;
5354         struct btrfs_trans_handle *trans;
5355         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5356         struct btrfs_key ins;
5357         u64 alloc_hint;
5358         int ret;
5359         bool insert = false;
5360
5361         /*
5362          * Ok if the extent map we looked up is a hole and is for the exact
5363          * range we want, there is no reason to allocate a new one, however if
5364          * it is not right then we need to free this one and drop the cache for
5365          * our range.
5366          */
5367         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5368             em->len != len) {
5369                 free_extent_map(em);
5370                 em = NULL;
5371                 insert = true;
5372                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5373         }
5374
5375         trans = btrfs_join_transaction(root);
5376         if (IS_ERR(trans))
5377                 return ERR_CAST(trans);
5378
5379         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5380                 btrfs_add_inode_defrag(trans, inode);
5381
5382         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5383
5384         alloc_hint = get_extent_allocation_hint(inode, start, len);
5385         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5386                                    alloc_hint, (u64)-1, &ins, 1);
5387         if (ret) {
5388                 em = ERR_PTR(ret);
5389                 goto out;
5390         }
5391
5392         if (!em) {
5393                 em = alloc_extent_map();
5394                 if (!em) {
5395                         em = ERR_PTR(-ENOMEM);
5396                         goto out;
5397                 }
5398         }
5399
5400         em->start = start;
5401         em->orig_start = em->start;
5402         em->len = ins.offset;
5403
5404         em->block_start = ins.objectid;
5405         em->block_len = ins.offset;
5406         em->bdev = root->fs_info->fs_devices->latest_bdev;
5407
5408         /*
5409          * We need to do this because if we're using the original em we searched
5410          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5411          */
5412         em->flags = 0;
5413         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5414
5415         while (insert) {
5416                 write_lock(&em_tree->lock);
5417                 ret = add_extent_mapping(em_tree, em);
5418                 write_unlock(&em_tree->lock);
5419                 if (ret != -EEXIST)
5420                         break;
5421                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5422         }
5423
5424         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5425                                            ins.offset, ins.offset, 0);
5426         if (ret) {
5427                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5428                 em = ERR_PTR(ret);
5429         }
5430 out:
5431         btrfs_end_transaction(trans, root);
5432         return em;
5433 }
5434
5435 /*
5436  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5437  * block must be cow'd
5438  */
5439 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5440                                       struct inode *inode, u64 offset, u64 len)
5441 {
5442         struct btrfs_path *path;
5443         int ret;
5444         struct extent_buffer *leaf;
5445         struct btrfs_root *root = BTRFS_I(inode)->root;
5446         struct btrfs_file_extent_item *fi;
5447         struct btrfs_key key;
5448         u64 disk_bytenr;
5449         u64 backref_offset;
5450         u64 extent_end;
5451         u64 num_bytes;
5452         int slot;
5453         int found_type;
5454
5455         path = btrfs_alloc_path();
5456         if (!path)
5457                 return -ENOMEM;
5458
5459         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5460                                        offset, 0);
5461         if (ret < 0)
5462                 goto out;
5463
5464         slot = path->slots[0];
5465         if (ret == 1) {
5466                 if (slot == 0) {
5467                         /* can't find the item, must cow */
5468                         ret = 0;
5469                         goto out;
5470                 }
5471                 slot--;
5472         }
5473         ret = 0;
5474         leaf = path->nodes[0];
5475         btrfs_item_key_to_cpu(leaf, &key, slot);
5476         if (key.objectid != btrfs_ino(inode) ||
5477             key.type != BTRFS_EXTENT_DATA_KEY) {
5478                 /* not our file or wrong item type, must cow */
5479                 goto out;
5480         }
5481
5482         if (key.offset > offset) {
5483                 /* Wrong offset, must cow */
5484                 goto out;
5485         }
5486
5487         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5488         found_type = btrfs_file_extent_type(leaf, fi);
5489         if (found_type != BTRFS_FILE_EXTENT_REG &&
5490             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5491                 /* not a regular extent, must cow */
5492                 goto out;
5493         }
5494         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5495         backref_offset = btrfs_file_extent_offset(leaf, fi);
5496
5497         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5498         if (extent_end < offset + len) {
5499                 /* extent doesn't include our full range, must cow */
5500                 goto out;
5501         }
5502
5503         if (btrfs_extent_readonly(root, disk_bytenr))
5504                 goto out;
5505
5506         /*
5507          * look for other files referencing this extent, if we
5508          * find any we must cow
5509          */
5510         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5511                                   key.offset - backref_offset, disk_bytenr))
5512                 goto out;
5513
5514         /*
5515          * adjust disk_bytenr and num_bytes to cover just the bytes
5516          * in this extent we are about to write.  If there
5517          * are any csums in that range we have to cow in order
5518          * to keep the csums correct
5519          */
5520         disk_bytenr += backref_offset;
5521         disk_bytenr += offset - key.offset;
5522         num_bytes = min(offset + len, extent_end) - offset;
5523         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5524                                 goto out;
5525         /*
5526          * all of the above have passed, it is safe to overwrite this extent
5527          * without cow
5528          */
5529         ret = 1;
5530 out:
5531         btrfs_free_path(path);
5532         return ret;
5533 }
5534
5535 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5536                                    struct buffer_head *bh_result, int create)
5537 {
5538         struct extent_map *em;
5539         struct btrfs_root *root = BTRFS_I(inode)->root;
5540         u64 start = iblock << inode->i_blkbits;
5541         u64 len = bh_result->b_size;
5542         struct btrfs_trans_handle *trans;
5543
5544         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5545         if (IS_ERR(em))
5546                 return PTR_ERR(em);
5547
5548         /*
5549          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5550          * io.  INLINE is special, and we could probably kludge it in here, but
5551          * it's still buffered so for safety lets just fall back to the generic
5552          * buffered path.
5553          *
5554          * For COMPRESSED we _have_ to read the entire extent in so we can
5555          * decompress it, so there will be buffering required no matter what we
5556          * do, so go ahead and fallback to buffered.
5557          *
5558          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5559          * to buffered IO.  Don't blame me, this is the price we pay for using
5560          * the generic code.
5561          */
5562         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5563             em->block_start == EXTENT_MAP_INLINE) {
5564                 free_extent_map(em);
5565                 return -ENOTBLK;
5566         }
5567
5568         /* Just a good old fashioned hole, return */
5569         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5570                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5571                 free_extent_map(em);
5572                 /* DIO will do one hole at a time, so just unlock a sector */
5573                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5574                               start + root->sectorsize - 1, GFP_NOFS);
5575                 return 0;
5576         }
5577
5578         /*
5579          * We don't allocate a new extent in the following cases
5580          *
5581          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5582          * existing extent.
5583          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5584          * just use the extent.
5585          *
5586          */
5587         if (!create) {
5588                 len = em->len - (start - em->start);
5589                 goto map;
5590         }
5591
5592         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5593             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5594              em->block_start != EXTENT_MAP_HOLE)) {
5595                 int type;
5596                 int ret;
5597                 u64 block_start;
5598
5599                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5600                         type = BTRFS_ORDERED_PREALLOC;
5601                 else
5602                         type = BTRFS_ORDERED_NOCOW;
5603                 len = min(len, em->len - (start - em->start));
5604                 block_start = em->block_start + (start - em->start);
5605
5606                 /*
5607                  * we're not going to log anything, but we do need
5608                  * to make sure the current transaction stays open
5609                  * while we look for nocow cross refs
5610                  */
5611                 trans = btrfs_join_transaction(root);
5612                 if (IS_ERR(trans))
5613                         goto must_cow;
5614
5615                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5616                         ret = btrfs_add_ordered_extent_dio(inode, start,
5617                                            block_start, len, len, type);
5618                         btrfs_end_transaction(trans, root);
5619                         if (ret) {
5620                                 free_extent_map(em);
5621                                 return ret;
5622                         }
5623                         goto unlock;
5624                 }
5625                 btrfs_end_transaction(trans, root);
5626         }
5627 must_cow:
5628         /*
5629          * this will cow the extent, reset the len in case we changed
5630          * it above
5631          */
5632         len = bh_result->b_size;
5633         em = btrfs_new_extent_direct(inode, em, start, len);
5634         if (IS_ERR(em))
5635                 return PTR_ERR(em);
5636         len = min(len, em->len - (start - em->start));
5637 unlock:
5638         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5639                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5640                           0, NULL, GFP_NOFS);
5641 map:
5642         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5643                 inode->i_blkbits;
5644         bh_result->b_size = len;
5645         bh_result->b_bdev = em->bdev;
5646         set_buffer_mapped(bh_result);
5647         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5648                 set_buffer_new(bh_result);
5649
5650         free_extent_map(em);
5651
5652         return 0;
5653 }
5654
5655 struct btrfs_dio_private {
5656         struct inode *inode;
5657         u64 logical_offset;
5658         u64 disk_bytenr;
5659         u64 bytes;
5660         u32 *csums;
5661         void *private;
5662
5663         /* number of bios pending for this dio */
5664         atomic_t pending_bios;
5665
5666         /* IO errors */
5667         int errors;
5668
5669         struct bio *orig_bio;
5670 };
5671
5672 static void btrfs_endio_direct_read(struct bio *bio, int err)
5673 {
5674         struct btrfs_dio_private *dip = bio->bi_private;
5675         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5676         struct bio_vec *bvec = bio->bi_io_vec;
5677         struct inode *inode = dip->inode;
5678         struct btrfs_root *root = BTRFS_I(inode)->root;
5679         u64 start;
5680         u32 *private = dip->csums;
5681
5682         start = dip->logical_offset;
5683         do {
5684                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5685                         struct page *page = bvec->bv_page;
5686                         char *kaddr;
5687                         u32 csum = ~(u32)0;
5688                         unsigned long flags;
5689
5690                         local_irq_save(flags);
5691                         kaddr = kmap_atomic(page, KM_IRQ0);
5692                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5693                                                csum, bvec->bv_len);
5694                         btrfs_csum_final(csum, (char *)&csum);
5695                         kunmap_atomic(kaddr, KM_IRQ0);
5696                         local_irq_restore(flags);
5697
5698                         flush_dcache_page(bvec->bv_page);
5699                         if (csum != *private) {
5700                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5701                                       " %llu csum %u private %u\n",
5702                                       (unsigned long long)btrfs_ino(inode),
5703                                       (unsigned long long)start,
5704                                       csum, *private);
5705                                 err = -EIO;
5706                         }
5707                 }
5708
5709                 start += bvec->bv_len;
5710                 private++;
5711                 bvec++;
5712         } while (bvec <= bvec_end);
5713
5714         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5715                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5716         bio->bi_private = dip->private;
5717
5718         kfree(dip->csums);
5719         kfree(dip);
5720
5721         /* If we had a csum failure make sure to clear the uptodate flag */
5722         if (err)
5723                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5724         dio_end_io(bio, err);
5725 }
5726
5727 static void btrfs_endio_direct_write(struct bio *bio, int err)
5728 {
5729         struct btrfs_dio_private *dip = bio->bi_private;
5730         struct inode *inode = dip->inode;
5731         struct btrfs_root *root = BTRFS_I(inode)->root;
5732         struct btrfs_trans_handle *trans;
5733         struct btrfs_ordered_extent *ordered = NULL;
5734         struct extent_state *cached_state = NULL;
5735         u64 ordered_offset = dip->logical_offset;
5736         u64 ordered_bytes = dip->bytes;
5737         int ret;
5738
5739         if (err)
5740                 goto out_done;
5741 again:
5742         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5743                                                    &ordered_offset,
5744                                                    ordered_bytes);
5745         if (!ret)
5746                 goto out_test;
5747
5748         BUG_ON(!ordered);
5749
5750         trans = btrfs_join_transaction(root);
5751         if (IS_ERR(trans)) {
5752                 err = -ENOMEM;
5753                 goto out;
5754         }
5755         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5756
5757         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5758                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5759                 if (!ret)
5760                         ret = btrfs_update_inode(trans, root, inode);
5761                 err = ret;
5762                 goto out;
5763         }
5764
5765         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5766                          ordered->file_offset + ordered->len - 1, 0,
5767                          &cached_state, GFP_NOFS);
5768
5769         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5770                 ret = btrfs_mark_extent_written(trans, inode,
5771                                                 ordered->file_offset,
5772                                                 ordered->file_offset +
5773                                                 ordered->len);
5774                 if (ret) {
5775                         err = ret;
5776                         goto out_unlock;
5777                 }
5778         } else {
5779                 ret = insert_reserved_file_extent(trans, inode,
5780                                                   ordered->file_offset,
5781                                                   ordered->start,
5782                                                   ordered->disk_len,
5783                                                   ordered->len,
5784                                                   ordered->len,
5785                                                   0, 0, 0,
5786                                                   BTRFS_FILE_EXTENT_REG);
5787                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5788                                    ordered->file_offset, ordered->len);
5789                 if (ret) {
5790                         err = ret;
5791                         WARN_ON(1);
5792                         goto out_unlock;
5793                 }
5794         }
5795
5796         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5797         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5798         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5799                 btrfs_update_inode(trans, root, inode);
5800         ret = 0;
5801 out_unlock:
5802         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5803                              ordered->file_offset + ordered->len - 1,
5804                              &cached_state, GFP_NOFS);
5805 out:
5806         btrfs_delalloc_release_metadata(inode, ordered->len);
5807         btrfs_end_transaction(trans, root);
5808         ordered_offset = ordered->file_offset + ordered->len;
5809         btrfs_put_ordered_extent(ordered);
5810         btrfs_put_ordered_extent(ordered);
5811
5812 out_test:
5813         /*
5814          * our bio might span multiple ordered extents.  If we haven't
5815          * completed the accounting for the whole dio, go back and try again
5816          */
5817         if (ordered_offset < dip->logical_offset + dip->bytes) {
5818                 ordered_bytes = dip->logical_offset + dip->bytes -
5819                         ordered_offset;
5820                 goto again;
5821         }
5822 out_done:
5823         bio->bi_private = dip->private;
5824
5825         kfree(dip->csums);
5826         kfree(dip);
5827
5828         /* If we had an error make sure to clear the uptodate flag */
5829         if (err)
5830                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5831         dio_end_io(bio, err);
5832 }
5833
5834 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5835                                     struct bio *bio, int mirror_num,
5836                                     unsigned long bio_flags, u64 offset)
5837 {
5838         int ret;
5839         struct btrfs_root *root = BTRFS_I(inode)->root;
5840         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5841         BUG_ON(ret);
5842         return 0;
5843 }
5844
5845 static void btrfs_end_dio_bio(struct bio *bio, int err)
5846 {
5847         struct btrfs_dio_private *dip = bio->bi_private;
5848
5849         if (err) {
5850                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5851                       "sector %#Lx len %u err no %d\n",
5852                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5853                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5854                 dip->errors = 1;
5855
5856                 /*
5857                  * before atomic variable goto zero, we must make sure
5858                  * dip->errors is perceived to be set.
5859                  */
5860                 smp_mb__before_atomic_dec();
5861         }
5862
5863         /* if there are more bios still pending for this dio, just exit */
5864         if (!atomic_dec_and_test(&dip->pending_bios))
5865                 goto out;
5866
5867         if (dip->errors)
5868                 bio_io_error(dip->orig_bio);
5869         else {
5870                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5871                 bio_endio(dip->orig_bio, 0);
5872         }
5873 out:
5874         bio_put(bio);
5875 }
5876
5877 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5878                                        u64 first_sector, gfp_t gfp_flags)
5879 {
5880         int nr_vecs = bio_get_nr_vecs(bdev);
5881         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5882 }
5883
5884 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5885                                          int rw, u64 file_offset, int skip_sum,
5886                                          u32 *csums, int async_submit)
5887 {
5888         int write = rw & REQ_WRITE;
5889         struct btrfs_root *root = BTRFS_I(inode)->root;
5890         int ret;
5891
5892         bio_get(bio);
5893         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5894         if (ret)
5895                 goto err;
5896
5897         if (skip_sum)
5898                 goto map;
5899
5900         if (write && async_submit) {
5901                 ret = btrfs_wq_submit_bio(root->fs_info,
5902                                    inode, rw, bio, 0, 0,
5903                                    file_offset,
5904                                    __btrfs_submit_bio_start_direct_io,
5905                                    __btrfs_submit_bio_done);
5906                 goto err;
5907         } else if (write) {
5908                 /*
5909                  * If we aren't doing async submit, calculate the csum of the
5910                  * bio now.
5911                  */
5912                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5913                 if (ret)
5914                         goto err;
5915         } else if (!skip_sum) {
5916                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5917                                           file_offset, csums);
5918                 if (ret)
5919                         goto err;
5920         }
5921
5922 map:
5923         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5924 err:
5925         bio_put(bio);
5926         return ret;
5927 }
5928
5929 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5930                                     int skip_sum)
5931 {
5932         struct inode *inode = dip->inode;
5933         struct btrfs_root *root = BTRFS_I(inode)->root;
5934         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5935         struct bio *bio;
5936         struct bio *orig_bio = dip->orig_bio;
5937         struct bio_vec *bvec = orig_bio->bi_io_vec;
5938         u64 start_sector = orig_bio->bi_sector;
5939         u64 file_offset = dip->logical_offset;
5940         u64 submit_len = 0;
5941         u64 map_length;
5942         int nr_pages = 0;
5943         u32 *csums = dip->csums;
5944         int ret = 0;
5945         int async_submit = 0;
5946         int write = rw & REQ_WRITE;
5947
5948         map_length = orig_bio->bi_size;
5949         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5950                               &map_length, NULL, 0);
5951         if (ret) {
5952                 bio_put(orig_bio);
5953                 return -EIO;
5954         }
5955
5956         if (map_length >= orig_bio->bi_size) {
5957                 bio = orig_bio;
5958                 goto submit;
5959         }
5960
5961         async_submit = 1;
5962         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5963         if (!bio)
5964                 return -ENOMEM;
5965         bio->bi_private = dip;
5966         bio->bi_end_io = btrfs_end_dio_bio;
5967         atomic_inc(&dip->pending_bios);
5968
5969         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5970                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5971                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5972                                  bvec->bv_offset) < bvec->bv_len)) {
5973                         /*
5974                          * inc the count before we submit the bio so
5975                          * we know the end IO handler won't happen before
5976                          * we inc the count. Otherwise, the dip might get freed
5977                          * before we're done setting it up
5978                          */
5979                         atomic_inc(&dip->pending_bios);
5980                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5981                                                      file_offset, skip_sum,
5982                                                      csums, async_submit);
5983                         if (ret) {
5984                                 bio_put(bio);
5985                                 atomic_dec(&dip->pending_bios);
5986                                 goto out_err;
5987                         }
5988
5989                         /* Write's use the ordered csums */
5990                         if (!write && !skip_sum)
5991                                 csums = csums + nr_pages;
5992                         start_sector += submit_len >> 9;
5993                         file_offset += submit_len;
5994
5995                         submit_len = 0;
5996                         nr_pages = 0;
5997
5998                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5999                                                   start_sector, GFP_NOFS);
6000                         if (!bio)
6001                                 goto out_err;
6002                         bio->bi_private = dip;
6003                         bio->bi_end_io = btrfs_end_dio_bio;
6004
6005                         map_length = orig_bio->bi_size;
6006                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6007                                               &map_length, NULL, 0);
6008                         if (ret) {
6009                                 bio_put(bio);
6010                                 goto out_err;
6011                         }
6012                 } else {
6013                         submit_len += bvec->bv_len;
6014                         nr_pages ++;
6015                         bvec++;
6016                 }
6017         }
6018
6019 submit:
6020         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6021                                      csums, async_submit);
6022         if (!ret)
6023                 return 0;
6024
6025         bio_put(bio);
6026 out_err:
6027         dip->errors = 1;
6028         /*
6029          * before atomic variable goto zero, we must
6030          * make sure dip->errors is perceived to be set.
6031          */
6032         smp_mb__before_atomic_dec();
6033         if (atomic_dec_and_test(&dip->pending_bios))
6034                 bio_io_error(dip->orig_bio);
6035
6036         /* bio_end_io() will handle error, so we needn't return it */
6037         return 0;
6038 }
6039
6040 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6041                                 loff_t file_offset)
6042 {
6043         struct btrfs_root *root = BTRFS_I(inode)->root;
6044         struct btrfs_dio_private *dip;
6045         struct bio_vec *bvec = bio->bi_io_vec;
6046         int skip_sum;
6047         int write = rw & REQ_WRITE;
6048         int ret = 0;
6049
6050         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6051
6052         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6053         if (!dip) {
6054                 ret = -ENOMEM;
6055                 goto free_ordered;
6056         }
6057         dip->csums = NULL;
6058
6059         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6060         if (!write && !skip_sum) {
6061                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6062                 if (!dip->csums) {
6063                         kfree(dip);
6064                         ret = -ENOMEM;
6065                         goto free_ordered;
6066                 }
6067         }
6068
6069         dip->private = bio->bi_private;
6070         dip->inode = inode;
6071         dip->logical_offset = file_offset;
6072
6073         dip->bytes = 0;
6074         do {
6075                 dip->bytes += bvec->bv_len;
6076                 bvec++;
6077         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6078
6079         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6080         bio->bi_private = dip;
6081         dip->errors = 0;
6082         dip->orig_bio = bio;
6083         atomic_set(&dip->pending_bios, 0);
6084
6085         if (write)
6086                 bio->bi_end_io = btrfs_endio_direct_write;
6087         else
6088                 bio->bi_end_io = btrfs_endio_direct_read;
6089
6090         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6091         if (!ret)
6092                 return;
6093 free_ordered:
6094         /*
6095          * If this is a write, we need to clean up the reserved space and kill
6096          * the ordered extent.
6097          */
6098         if (write) {
6099                 struct btrfs_ordered_extent *ordered;
6100                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6101                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6102                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6103                         btrfs_free_reserved_extent(root, ordered->start,
6104                                                    ordered->disk_len);
6105                 btrfs_put_ordered_extent(ordered);
6106                 btrfs_put_ordered_extent(ordered);
6107         }
6108         bio_endio(bio, ret);
6109 }
6110
6111 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6112                         const struct iovec *iov, loff_t offset,
6113                         unsigned long nr_segs)
6114 {
6115         int seg;
6116         int i;
6117         size_t size;
6118         unsigned long addr;
6119         unsigned blocksize_mask = root->sectorsize - 1;
6120         ssize_t retval = -EINVAL;
6121         loff_t end = offset;
6122
6123         if (offset & blocksize_mask)
6124                 goto out;
6125
6126         /* Check the memory alignment.  Blocks cannot straddle pages */
6127         for (seg = 0; seg < nr_segs; seg++) {
6128                 addr = (unsigned long)iov[seg].iov_base;
6129                 size = iov[seg].iov_len;
6130                 end += size;
6131                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6132                         goto out;
6133
6134                 /* If this is a write we don't need to check anymore */
6135                 if (rw & WRITE)
6136                         continue;
6137
6138                 /*
6139                  * Check to make sure we don't have duplicate iov_base's in this
6140                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6141                  * when reading back.
6142                  */
6143                 for (i = seg + 1; i < nr_segs; i++) {
6144                         if (iov[seg].iov_base == iov[i].iov_base)
6145                                 goto out;
6146                 }
6147         }
6148         retval = 0;
6149 out:
6150         return retval;
6151 }
6152 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6153                         const struct iovec *iov, loff_t offset,
6154                         unsigned long nr_segs)
6155 {
6156         struct file *file = iocb->ki_filp;
6157         struct inode *inode = file->f_mapping->host;
6158         struct btrfs_ordered_extent *ordered;
6159         struct extent_state *cached_state = NULL;
6160         u64 lockstart, lockend;
6161         ssize_t ret;
6162         int writing = rw & WRITE;
6163         int write_bits = 0;
6164         size_t count = iov_length(iov, nr_segs);
6165
6166         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6167                             offset, nr_segs)) {
6168                 return 0;
6169         }
6170
6171         lockstart = offset;
6172         lockend = offset + count - 1;
6173
6174         if (writing) {
6175                 ret = btrfs_delalloc_reserve_space(inode, count);
6176                 if (ret)
6177                         goto out;
6178         }
6179
6180         while (1) {
6181                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6182                                  0, &cached_state, GFP_NOFS);
6183                 /*
6184                  * We're concerned with the entire range that we're going to be
6185                  * doing DIO to, so we need to make sure theres no ordered
6186                  * extents in this range.
6187                  */
6188                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6189                                                      lockend - lockstart + 1);
6190                 if (!ordered)
6191                         break;
6192                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6193                                      &cached_state, GFP_NOFS);
6194                 btrfs_start_ordered_extent(inode, ordered, 1);
6195                 btrfs_put_ordered_extent(ordered);
6196                 cond_resched();
6197         }
6198
6199         /*
6200          * we don't use btrfs_set_extent_delalloc because we don't want
6201          * the dirty or uptodate bits
6202          */
6203         if (writing) {
6204                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6205                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6206                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6207                                      GFP_NOFS);
6208                 if (ret) {
6209                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6210                                          lockend, EXTENT_LOCKED | write_bits,
6211                                          1, 0, &cached_state, GFP_NOFS);
6212                         goto out;
6213                 }
6214         }
6215
6216         free_extent_state(cached_state);
6217         cached_state = NULL;
6218
6219         ret = __blockdev_direct_IO(rw, iocb, inode,
6220                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6221                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6222                    btrfs_submit_direct, 0);
6223
6224         if (ret < 0 && ret != -EIOCBQUEUED) {
6225                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6226                               offset + iov_length(iov, nr_segs) - 1,
6227                               EXTENT_LOCKED | write_bits, 1, 0,
6228                               &cached_state, GFP_NOFS);
6229         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6230                 /*
6231                  * We're falling back to buffered, unlock the section we didn't
6232                  * do IO on.
6233                  */
6234                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6235                               offset + iov_length(iov, nr_segs) - 1,
6236                               EXTENT_LOCKED | write_bits, 1, 0,
6237                               &cached_state, GFP_NOFS);
6238         }
6239 out:
6240         free_extent_state(cached_state);
6241         return ret;
6242 }
6243
6244 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6245                 __u64 start, __u64 len)
6246 {
6247         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6248 }
6249
6250 int btrfs_readpage(struct file *file, struct page *page)
6251 {
6252         struct extent_io_tree *tree;
6253         tree = &BTRFS_I(page->mapping->host)->io_tree;
6254         return extent_read_full_page(tree, page, btrfs_get_extent);
6255 }
6256
6257 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6258 {
6259         struct extent_io_tree *tree;
6260
6261
6262         if (current->flags & PF_MEMALLOC) {
6263                 redirty_page_for_writepage(wbc, page);
6264                 unlock_page(page);
6265                 return 0;
6266         }
6267         tree = &BTRFS_I(page->mapping->host)->io_tree;
6268         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6269 }
6270
6271 int btrfs_writepages(struct address_space *mapping,
6272                      struct writeback_control *wbc)
6273 {
6274         struct extent_io_tree *tree;
6275
6276         tree = &BTRFS_I(mapping->host)->io_tree;
6277         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6278 }
6279
6280 static int
6281 btrfs_readpages(struct file *file, struct address_space *mapping,
6282                 struct list_head *pages, unsigned nr_pages)
6283 {
6284         struct extent_io_tree *tree;
6285         tree = &BTRFS_I(mapping->host)->io_tree;
6286         return extent_readpages(tree, mapping, pages, nr_pages,
6287                                 btrfs_get_extent);
6288 }
6289 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6290 {
6291         struct extent_io_tree *tree;
6292         struct extent_map_tree *map;
6293         int ret;
6294
6295         tree = &BTRFS_I(page->mapping->host)->io_tree;
6296         map = &BTRFS_I(page->mapping->host)->extent_tree;
6297         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6298         if (ret == 1) {
6299                 ClearPagePrivate(page);
6300                 set_page_private(page, 0);
6301                 page_cache_release(page);
6302         }
6303         return ret;
6304 }
6305
6306 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6307 {
6308         if (PageWriteback(page) || PageDirty(page))
6309                 return 0;
6310         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6311 }
6312
6313 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6314 {
6315         struct extent_io_tree *tree;
6316         struct btrfs_ordered_extent *ordered;
6317         struct extent_state *cached_state = NULL;
6318         u64 page_start = page_offset(page);
6319         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6320
6321
6322         /*
6323          * we have the page locked, so new writeback can't start,
6324          * and the dirty bit won't be cleared while we are here.
6325          *
6326          * Wait for IO on this page so that we can safely clear
6327          * the PagePrivate2 bit and do ordered accounting
6328          */
6329         wait_on_page_writeback(page);
6330
6331         tree = &BTRFS_I(page->mapping->host)->io_tree;
6332         if (offset) {
6333                 btrfs_releasepage(page, GFP_NOFS);
6334                 return;
6335         }
6336         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6337                          GFP_NOFS);
6338         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6339                                            page_offset(page));
6340         if (ordered) {
6341                 /*
6342                  * IO on this page will never be started, so we need
6343                  * to account for any ordered extents now
6344                  */
6345                 clear_extent_bit(tree, page_start, page_end,
6346                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6347                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6348                                  &cached_state, GFP_NOFS);
6349                 /*
6350                  * whoever cleared the private bit is responsible
6351                  * for the finish_ordered_io
6352                  */
6353                 if (TestClearPagePrivate2(page)) {
6354                         btrfs_finish_ordered_io(page->mapping->host,
6355                                                 page_start, page_end);
6356                 }
6357                 btrfs_put_ordered_extent(ordered);
6358                 cached_state = NULL;
6359                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6360                                  GFP_NOFS);
6361         }
6362         clear_extent_bit(tree, page_start, page_end,
6363                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6364                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6365         __btrfs_releasepage(page, GFP_NOFS);
6366
6367         ClearPageChecked(page);
6368         if (PagePrivate(page)) {
6369                 ClearPagePrivate(page);
6370                 set_page_private(page, 0);
6371                 page_cache_release(page);
6372         }
6373 }
6374
6375 /*
6376  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6377  * called from a page fault handler when a page is first dirtied. Hence we must
6378  * be careful to check for EOF conditions here. We set the page up correctly
6379  * for a written page which means we get ENOSPC checking when writing into
6380  * holes and correct delalloc and unwritten extent mapping on filesystems that
6381  * support these features.
6382  *
6383  * We are not allowed to take the i_mutex here so we have to play games to
6384  * protect against truncate races as the page could now be beyond EOF.  Because
6385  * vmtruncate() writes the inode size before removing pages, once we have the
6386  * page lock we can determine safely if the page is beyond EOF. If it is not
6387  * beyond EOF, then the page is guaranteed safe against truncation until we
6388  * unlock the page.
6389  */
6390 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6391 {
6392         struct page *page = vmf->page;
6393         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6394         struct btrfs_root *root = BTRFS_I(inode)->root;
6395         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6396         struct btrfs_ordered_extent *ordered;
6397         struct extent_state *cached_state = NULL;
6398         char *kaddr;
6399         unsigned long zero_start;
6400         loff_t size;
6401         int ret;
6402         u64 page_start;
6403         u64 page_end;
6404
6405         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6406         if (ret) {
6407                 if (ret == -ENOMEM)
6408                         ret = VM_FAULT_OOM;
6409                 else /* -ENOSPC, -EIO, etc */
6410                         ret = VM_FAULT_SIGBUS;
6411                 goto out;
6412         }
6413
6414         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6415 again:
6416         lock_page(page);
6417         size = i_size_read(inode);
6418         page_start = page_offset(page);
6419         page_end = page_start + PAGE_CACHE_SIZE - 1;
6420
6421         if ((page->mapping != inode->i_mapping) ||
6422             (page_start >= size)) {
6423                 /* page got truncated out from underneath us */
6424                 goto out_unlock;
6425         }
6426         wait_on_page_writeback(page);
6427
6428         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6429                          GFP_NOFS);
6430         set_page_extent_mapped(page);
6431
6432         /*
6433          * we can't set the delalloc bits if there are pending ordered
6434          * extents.  Drop our locks and wait for them to finish
6435          */
6436         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6437         if (ordered) {
6438                 unlock_extent_cached(io_tree, page_start, page_end,
6439                                      &cached_state, GFP_NOFS);
6440                 unlock_page(page);
6441                 btrfs_start_ordered_extent(inode, ordered, 1);
6442                 btrfs_put_ordered_extent(ordered);
6443                 goto again;
6444         }
6445
6446         /*
6447          * XXX - page_mkwrite gets called every time the page is dirtied, even
6448          * if it was already dirty, so for space accounting reasons we need to
6449          * clear any delalloc bits for the range we are fixing to save.  There
6450          * is probably a better way to do this, but for now keep consistent with
6451          * prepare_pages in the normal write path.
6452          */
6453         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6454                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6455                           0, 0, &cached_state, GFP_NOFS);
6456
6457         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6458                                         &cached_state);
6459         if (ret) {
6460                 unlock_extent_cached(io_tree, page_start, page_end,
6461                                      &cached_state, GFP_NOFS);
6462                 ret = VM_FAULT_SIGBUS;
6463                 goto out_unlock;
6464         }
6465         ret = 0;
6466
6467         /* page is wholly or partially inside EOF */
6468         if (page_start + PAGE_CACHE_SIZE > size)
6469                 zero_start = size & ~PAGE_CACHE_MASK;
6470         else
6471                 zero_start = PAGE_CACHE_SIZE;
6472
6473         if (zero_start != PAGE_CACHE_SIZE) {
6474                 kaddr = kmap(page);
6475                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6476                 flush_dcache_page(page);
6477                 kunmap(page);
6478         }
6479         ClearPageChecked(page);
6480         set_page_dirty(page);
6481         SetPageUptodate(page);
6482
6483         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6484         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6485
6486         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6487
6488 out_unlock:
6489         if (!ret)
6490                 return VM_FAULT_LOCKED;
6491         unlock_page(page);
6492         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6493 out:
6494         return ret;
6495 }
6496
6497 static int btrfs_truncate(struct inode *inode)
6498 {
6499         struct btrfs_root *root = BTRFS_I(inode)->root;
6500         struct btrfs_block_rsv *rsv;
6501         int ret;
6502         int err = 0;
6503         struct btrfs_trans_handle *trans;
6504         unsigned long nr;
6505         u64 mask = root->sectorsize - 1;
6506         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6507
6508         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6509         if (ret)
6510                 return ret;
6511
6512         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6513         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6514
6515         /*
6516          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6517          * 3 things going on here
6518          *
6519          * 1) We need to reserve space for our orphan item and the space to
6520          * delete our orphan item.  Lord knows we don't want to have a dangling
6521          * orphan item because we didn't reserve space to remove it.
6522          *
6523          * 2) We need to reserve space to update our inode.
6524          *
6525          * 3) We need to have something to cache all the space that is going to
6526          * be free'd up by the truncate operation, but also have some slack
6527          * space reserved in case it uses space during the truncate (thank you
6528          * very much snapshotting).
6529          *
6530          * And we need these to all be seperate.  The fact is we can use alot of
6531          * space doing the truncate, and we have no earthly idea how much space
6532          * we will use, so we need the truncate reservation to be seperate so it
6533          * doesn't end up using space reserved for updating the inode or
6534          * removing the orphan item.  We also need to be able to stop the
6535          * transaction and start a new one, which means we need to be able to
6536          * update the inode several times, and we have no idea of knowing how
6537          * many times that will be, so we can't just reserve 1 item for the
6538          * entirety of the opration, so that has to be done seperately as well.
6539          * Then there is the orphan item, which does indeed need to be held on
6540          * to for the whole operation, and we need nobody to touch this reserved
6541          * space except the orphan code.
6542          *
6543          * So that leaves us with
6544          *
6545          * 1) root->orphan_block_rsv - for the orphan deletion.
6546          * 2) rsv - for the truncate reservation, which we will steal from the
6547          * transaction reservation.
6548          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6549          * updating the inode.
6550          */
6551         rsv = btrfs_alloc_block_rsv(root);
6552         if (!rsv)
6553                 return -ENOMEM;
6554         rsv->size = min_size;
6555
6556         /*
6557          * 1 for the truncate slack space
6558          * 1 for the orphan item we're going to add
6559          * 1 for the orphan item deletion
6560          * 1 for updating the inode.
6561          */
6562         trans = btrfs_start_transaction(root, 4);
6563         if (IS_ERR(trans)) {
6564                 err = PTR_ERR(trans);
6565                 goto out;
6566         }
6567
6568         /* Migrate the slack space for the truncate to our reserve */
6569         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6570                                       min_size);
6571         BUG_ON(ret);
6572
6573         ret = btrfs_orphan_add(trans, inode);
6574         if (ret) {
6575                 btrfs_end_transaction(trans, root);
6576                 goto out;
6577         }
6578
6579         /*
6580          * setattr is responsible for setting the ordered_data_close flag,
6581          * but that is only tested during the last file release.  That
6582          * could happen well after the next commit, leaving a great big
6583          * window where new writes may get lost if someone chooses to write
6584          * to this file after truncating to zero
6585          *
6586          * The inode doesn't have any dirty data here, and so if we commit
6587          * this is a noop.  If someone immediately starts writing to the inode
6588          * it is very likely we'll catch some of their writes in this
6589          * transaction, and the commit will find this file on the ordered
6590          * data list with good things to send down.
6591          *
6592          * This is a best effort solution, there is still a window where
6593          * using truncate to replace the contents of the file will
6594          * end up with a zero length file after a crash.
6595          */
6596         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6597                 btrfs_add_ordered_operation(trans, root, inode);
6598
6599         while (1) {
6600                 ret = btrfs_block_rsv_check(root, rsv, min_size, 0, 1);
6601                 if (ret) {
6602                         /*
6603                          * This can only happen with the original transaction we
6604                          * started above, every other time we shouldn't have a
6605                          * transaction started yet.
6606                          */
6607                         if (ret == -EAGAIN)
6608                                 goto end_trans;
6609                         err = ret;
6610                         break;
6611                 }
6612
6613                 if (!trans) {
6614                         /* Just need the 1 for updating the inode */
6615                         trans = btrfs_start_transaction(root, 1);
6616                         if (IS_ERR(trans)) {
6617                                 err = PTR_ERR(trans);
6618                                 goto out;
6619                         }
6620                 }
6621
6622                 trans->block_rsv = rsv;
6623
6624                 ret = btrfs_truncate_inode_items(trans, root, inode,
6625                                                  inode->i_size,
6626                                                  BTRFS_EXTENT_DATA_KEY);
6627                 if (ret != -EAGAIN) {
6628                         err = ret;
6629                         break;
6630                 }
6631
6632                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6633                 ret = btrfs_update_inode(trans, root, inode);
6634                 if (ret) {
6635                         err = ret;
6636                         break;
6637                 }
6638 end_trans:
6639                 nr = trans->blocks_used;
6640                 btrfs_end_transaction(trans, root);
6641                 trans = NULL;
6642                 btrfs_btree_balance_dirty(root, nr);
6643         }
6644
6645         if (ret == 0 && inode->i_nlink > 0) {
6646                 trans->block_rsv = root->orphan_block_rsv;
6647                 ret = btrfs_orphan_del(trans, inode);
6648                 if (ret)
6649                         err = ret;
6650         } else if (ret && inode->i_nlink > 0) {
6651                 /*
6652                  * Failed to do the truncate, remove us from the in memory
6653                  * orphan list.
6654                  */
6655                 ret = btrfs_orphan_del(NULL, inode);
6656         }
6657
6658         trans->block_rsv = &root->fs_info->trans_block_rsv;
6659         ret = btrfs_update_inode(trans, root, inode);
6660         if (ret && !err)
6661                 err = ret;
6662
6663         nr = trans->blocks_used;
6664         ret = btrfs_end_transaction_throttle(trans, root);
6665         btrfs_btree_balance_dirty(root, nr);
6666
6667 out:
6668         btrfs_free_block_rsv(root, rsv);
6669
6670         if (ret && !err)
6671                 err = ret;
6672
6673         return err;
6674 }
6675
6676 /*
6677  * create a new subvolume directory/inode (helper for the ioctl).
6678  */
6679 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6680                              struct btrfs_root *new_root, u64 new_dirid)
6681 {
6682         struct inode *inode;
6683         int err;
6684         u64 index = 0;
6685
6686         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6687                                 new_dirid, S_IFDIR | 0700, &index);
6688         if (IS_ERR(inode))
6689                 return PTR_ERR(inode);
6690         inode->i_op = &btrfs_dir_inode_operations;
6691         inode->i_fop = &btrfs_dir_file_operations;
6692
6693         inode->i_nlink = 1;
6694         btrfs_i_size_write(inode, 0);
6695
6696         err = btrfs_update_inode(trans, new_root, inode);
6697         BUG_ON(err);
6698
6699         iput(inode);
6700         return 0;
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->disk_i_size = 0;
6721         ei->flags = 0;
6722         ei->csum_bytes = 0;
6723         ei->index_cnt = (u64)-1;
6724         ei->last_unlink_trans = 0;
6725
6726         spin_lock_init(&ei->lock);
6727         ei->outstanding_extents = 0;
6728         ei->reserved_extents = 0;
6729
6730         ei->ordered_data_close = 0;
6731         ei->orphan_meta_reserved = 0;
6732         ei->dummy_inode = 0;
6733         ei->in_defrag = 0;
6734         ei->force_compress = BTRFS_COMPRESS_NONE;
6735
6736         ei->delayed_node = NULL;
6737
6738         inode = &ei->vfs_inode;
6739         extent_map_tree_init(&ei->extent_tree);
6740         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6741         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6742         mutex_init(&ei->log_mutex);
6743         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6744         INIT_LIST_HEAD(&ei->i_orphan);
6745         INIT_LIST_HEAD(&ei->delalloc_inodes);
6746         INIT_LIST_HEAD(&ei->ordered_operations);
6747         RB_CLEAR_NODE(&ei->rb_node);
6748
6749         return inode;
6750 }
6751
6752 static void btrfs_i_callback(struct rcu_head *head)
6753 {
6754         struct inode *inode = container_of(head, struct inode, i_rcu);
6755         INIT_LIST_HEAD(&inode->i_dentry);
6756         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6757 }
6758
6759 void btrfs_destroy_inode(struct inode *inode)
6760 {
6761         struct btrfs_ordered_extent *ordered;
6762         struct btrfs_root *root = BTRFS_I(inode)->root;
6763
6764         WARN_ON(!list_empty(&inode->i_dentry));
6765         WARN_ON(inode->i_data.nrpages);
6766         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6767         WARN_ON(BTRFS_I(inode)->reserved_extents);
6768         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6769         WARN_ON(BTRFS_I(inode)->csum_bytes);
6770
6771         /*
6772          * This can happen where we create an inode, but somebody else also
6773          * created the same inode and we need to destroy the one we already
6774          * created.
6775          */
6776         if (!root)
6777                 goto free;
6778
6779         /*
6780          * Make sure we're properly removed from the ordered operation
6781          * lists.
6782          */
6783         smp_mb();
6784         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6785                 spin_lock(&root->fs_info->ordered_extent_lock);
6786                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6787                 spin_unlock(&root->fs_info->ordered_extent_lock);
6788         }
6789
6790         spin_lock(&root->orphan_lock);
6791         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6792                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6793                        (unsigned long long)btrfs_ino(inode));
6794                 list_del_init(&BTRFS_I(inode)->i_orphan);
6795         }
6796         spin_unlock(&root->orphan_lock);
6797
6798         while (1) {
6799                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6800                 if (!ordered)
6801                         break;
6802                 else {
6803                         printk(KERN_ERR "btrfs found ordered "
6804                                "extent %llu %llu on inode cleanup\n",
6805                                (unsigned long long)ordered->file_offset,
6806                                (unsigned long long)ordered->len);
6807                         btrfs_remove_ordered_extent(inode, ordered);
6808                         btrfs_put_ordered_extent(ordered);
6809                         btrfs_put_ordered_extent(ordered);
6810                 }
6811         }
6812         inode_tree_del(inode);
6813         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6814 free:
6815         btrfs_remove_delayed_node(inode);
6816         call_rcu(&inode->i_rcu, btrfs_i_callback);
6817 }
6818
6819 int btrfs_drop_inode(struct inode *inode)
6820 {
6821         struct btrfs_root *root = BTRFS_I(inode)->root;
6822
6823         if (btrfs_root_refs(&root->root_item) == 0 &&
6824             !btrfs_is_free_space_inode(root, inode))
6825                 return 1;
6826         else
6827                 return generic_drop_inode(inode);
6828 }
6829
6830 static void init_once(void *foo)
6831 {
6832         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6833
6834         inode_init_once(&ei->vfs_inode);
6835 }
6836
6837 void btrfs_destroy_cachep(void)
6838 {
6839         if (btrfs_inode_cachep)
6840                 kmem_cache_destroy(btrfs_inode_cachep);
6841         if (btrfs_trans_handle_cachep)
6842                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6843         if (btrfs_transaction_cachep)
6844                 kmem_cache_destroy(btrfs_transaction_cachep);
6845         if (btrfs_path_cachep)
6846                 kmem_cache_destroy(btrfs_path_cachep);
6847         if (btrfs_free_space_cachep)
6848                 kmem_cache_destroy(btrfs_free_space_cachep);
6849 }
6850
6851 int btrfs_init_cachep(void)
6852 {
6853         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6854                         sizeof(struct btrfs_inode), 0,
6855                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6856         if (!btrfs_inode_cachep)
6857                 goto fail;
6858
6859         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6860                         sizeof(struct btrfs_trans_handle), 0,
6861                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6862         if (!btrfs_trans_handle_cachep)
6863                 goto fail;
6864
6865         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6866                         sizeof(struct btrfs_transaction), 0,
6867                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6868         if (!btrfs_transaction_cachep)
6869                 goto fail;
6870
6871         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6872                         sizeof(struct btrfs_path), 0,
6873                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6874         if (!btrfs_path_cachep)
6875                 goto fail;
6876
6877         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6878                         sizeof(struct btrfs_free_space), 0,
6879                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6880         if (!btrfs_free_space_cachep)
6881                 goto fail;
6882
6883         return 0;
6884 fail:
6885         btrfs_destroy_cachep();
6886         return -ENOMEM;
6887 }
6888
6889 static int btrfs_getattr(struct vfsmount *mnt,
6890                          struct dentry *dentry, struct kstat *stat)
6891 {
6892         struct inode *inode = dentry->d_inode;
6893         generic_fillattr(inode, stat);
6894         stat->dev = BTRFS_I(inode)->root->anon_dev;
6895         stat->blksize = PAGE_CACHE_SIZE;
6896         stat->blocks = (inode_get_bytes(inode) +
6897                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6898         return 0;
6899 }
6900
6901 /*
6902  * If a file is moved, it will inherit the cow and compression flags of the new
6903  * directory.
6904  */
6905 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6906 {
6907         struct btrfs_inode *b_dir = BTRFS_I(dir);
6908         struct btrfs_inode *b_inode = BTRFS_I(inode);
6909
6910         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6911                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6912         else
6913                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6914
6915         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6916                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6917         else
6918                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6919 }
6920
6921 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6922                            struct inode *new_dir, struct dentry *new_dentry)
6923 {
6924         struct btrfs_trans_handle *trans;
6925         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6926         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6927         struct inode *new_inode = new_dentry->d_inode;
6928         struct inode *old_inode = old_dentry->d_inode;
6929         struct timespec ctime = CURRENT_TIME;
6930         u64 index = 0;
6931         u64 root_objectid;
6932         int ret;
6933         u64 old_ino = btrfs_ino(old_inode);
6934
6935         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6936                 return -EPERM;
6937
6938         /* we only allow rename subvolume link between subvolumes */
6939         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6940                 return -EXDEV;
6941
6942         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6943             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6944                 return -ENOTEMPTY;
6945
6946         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6947             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6948                 return -ENOTEMPTY;
6949         /*
6950          * we're using rename to replace one file with another.
6951          * and the replacement file is large.  Start IO on it now so
6952          * we don't add too much work to the end of the transaction
6953          */
6954         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6955             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6956                 filemap_flush(old_inode->i_mapping);
6957
6958         /* close the racy window with snapshot create/destroy ioctl */
6959         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6960                 down_read(&root->fs_info->subvol_sem);
6961         /*
6962          * We want to reserve the absolute worst case amount of items.  So if
6963          * both inodes are subvols and we need to unlink them then that would
6964          * require 4 item modifications, but if they are both normal inodes it
6965          * would require 5 item modifications, so we'll assume their normal
6966          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6967          * should cover the worst case number of items we'll modify.
6968          */
6969         trans = btrfs_start_transaction(root, 20);
6970         if (IS_ERR(trans)) {
6971                 ret = PTR_ERR(trans);
6972                 goto out_notrans;
6973         }
6974
6975         if (dest != root)
6976                 btrfs_record_root_in_trans(trans, dest);
6977
6978         ret = btrfs_set_inode_index(new_dir, &index);
6979         if (ret)
6980                 goto out_fail;
6981
6982         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6983                 /* force full log commit if subvolume involved. */
6984                 root->fs_info->last_trans_log_full_commit = trans->transid;
6985         } else {
6986                 ret = btrfs_insert_inode_ref(trans, dest,
6987                                              new_dentry->d_name.name,
6988                                              new_dentry->d_name.len,
6989                                              old_ino,
6990                                              btrfs_ino(new_dir), index);
6991                 if (ret)
6992                         goto out_fail;
6993                 /*
6994                  * this is an ugly little race, but the rename is required
6995                  * to make sure that if we crash, the inode is either at the
6996                  * old name or the new one.  pinning the log transaction lets
6997                  * us make sure we don't allow a log commit to come in after
6998                  * we unlink the name but before we add the new name back in.
6999                  */
7000                 btrfs_pin_log_trans(root);
7001         }
7002         /*
7003          * make sure the inode gets flushed if it is replacing
7004          * something.
7005          */
7006         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7007                 btrfs_add_ordered_operation(trans, root, old_inode);
7008
7009         old_dir->i_ctime = old_dir->i_mtime = ctime;
7010         new_dir->i_ctime = new_dir->i_mtime = ctime;
7011         old_inode->i_ctime = ctime;
7012
7013         if (old_dentry->d_parent != new_dentry->d_parent)
7014                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7015
7016         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7017                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7018                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7019                                         old_dentry->d_name.name,
7020                                         old_dentry->d_name.len);
7021         } else {
7022                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7023                                         old_dentry->d_inode,
7024                                         old_dentry->d_name.name,
7025                                         old_dentry->d_name.len);
7026                 if (!ret)
7027                         ret = btrfs_update_inode(trans, root, old_inode);
7028         }
7029         BUG_ON(ret);
7030
7031         if (new_inode) {
7032                 new_inode->i_ctime = CURRENT_TIME;
7033                 if (unlikely(btrfs_ino(new_inode) ==
7034                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7035                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7036                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7037                                                 root_objectid,
7038                                                 new_dentry->d_name.name,
7039                                                 new_dentry->d_name.len);
7040                         BUG_ON(new_inode->i_nlink == 0);
7041                 } else {
7042                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7043                                                  new_dentry->d_inode,
7044                                                  new_dentry->d_name.name,
7045                                                  new_dentry->d_name.len);
7046                 }
7047                 BUG_ON(ret);
7048                 if (new_inode->i_nlink == 0) {
7049                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7050                         BUG_ON(ret);
7051                 }
7052         }
7053
7054         fixup_inode_flags(new_dir, old_inode);
7055
7056         ret = btrfs_add_link(trans, new_dir, old_inode,
7057                              new_dentry->d_name.name,
7058                              new_dentry->d_name.len, 0, index);
7059         BUG_ON(ret);
7060
7061         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7062                 struct dentry *parent = new_dentry->d_parent;
7063                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7064                 btrfs_end_log_trans(root);
7065         }
7066 out_fail:
7067         btrfs_end_transaction_throttle(trans, root);
7068 out_notrans:
7069         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7070                 up_read(&root->fs_info->subvol_sem);
7071
7072         return ret;
7073 }
7074
7075 /*
7076  * some fairly slow code that needs optimization. This walks the list
7077  * of all the inodes with pending delalloc and forces them to disk.
7078  */
7079 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7080 {
7081         struct list_head *head = &root->fs_info->delalloc_inodes;
7082         struct btrfs_inode *binode;
7083         struct inode *inode;
7084
7085         if (root->fs_info->sb->s_flags & MS_RDONLY)
7086                 return -EROFS;
7087
7088         spin_lock(&root->fs_info->delalloc_lock);
7089         while (!list_empty(head)) {
7090                 binode = list_entry(head->next, struct btrfs_inode,
7091                                     delalloc_inodes);
7092                 inode = igrab(&binode->vfs_inode);
7093                 if (!inode)
7094                         list_del_init(&binode->delalloc_inodes);
7095                 spin_unlock(&root->fs_info->delalloc_lock);
7096                 if (inode) {
7097                         filemap_flush(inode->i_mapping);
7098                         if (delay_iput)
7099                                 btrfs_add_delayed_iput(inode);
7100                         else
7101                                 iput(inode);
7102                 }
7103                 cond_resched();
7104                 spin_lock(&root->fs_info->delalloc_lock);
7105         }
7106         spin_unlock(&root->fs_info->delalloc_lock);
7107
7108         /* the filemap_flush will queue IO into the worker threads, but
7109          * we have to make sure the IO is actually started and that
7110          * ordered extents get created before we return
7111          */
7112         atomic_inc(&root->fs_info->async_submit_draining);
7113         while (atomic_read(&root->fs_info->nr_async_submits) ||
7114               atomic_read(&root->fs_info->async_delalloc_pages)) {
7115                 wait_event(root->fs_info->async_submit_wait,
7116                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7117                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7118         }
7119         atomic_dec(&root->fs_info->async_submit_draining);
7120         return 0;
7121 }
7122
7123 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7124                          const char *symname)
7125 {
7126         struct btrfs_trans_handle *trans;
7127         struct btrfs_root *root = BTRFS_I(dir)->root;
7128         struct btrfs_path *path;
7129         struct btrfs_key key;
7130         struct inode *inode = NULL;
7131         int err;
7132         int drop_inode = 0;
7133         u64 objectid;
7134         u64 index = 0 ;
7135         int name_len;
7136         int datasize;
7137         unsigned long ptr;
7138         struct btrfs_file_extent_item *ei;
7139         struct extent_buffer *leaf;
7140         unsigned long nr = 0;
7141
7142         name_len = strlen(symname) + 1;
7143         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7144                 return -ENAMETOOLONG;
7145
7146         /*
7147          * 2 items for inode item and ref
7148          * 2 items for dir items
7149          * 1 item for xattr if selinux is on
7150          */
7151         trans = btrfs_start_transaction(root, 5);
7152         if (IS_ERR(trans))
7153                 return PTR_ERR(trans);
7154
7155         err = btrfs_find_free_ino(root, &objectid);
7156         if (err)
7157                 goto out_unlock;
7158
7159         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7160                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7161                                 S_IFLNK|S_IRWXUGO, &index);
7162         if (IS_ERR(inode)) {
7163                 err = PTR_ERR(inode);
7164                 goto out_unlock;
7165         }
7166
7167         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7168         if (err) {
7169                 drop_inode = 1;
7170                 goto out_unlock;
7171         }
7172
7173         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7174         if (err)
7175                 drop_inode = 1;
7176         else {
7177                 inode->i_mapping->a_ops = &btrfs_aops;
7178                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7179                 inode->i_fop = &btrfs_file_operations;
7180                 inode->i_op = &btrfs_file_inode_operations;
7181                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7182         }
7183         if (drop_inode)
7184                 goto out_unlock;
7185
7186         path = btrfs_alloc_path();
7187         if (!path) {
7188                 err = -ENOMEM;
7189                 drop_inode = 1;
7190                 goto out_unlock;
7191         }
7192         key.objectid = btrfs_ino(inode);
7193         key.offset = 0;
7194         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7195         datasize = btrfs_file_extent_calc_inline_size(name_len);
7196         err = btrfs_insert_empty_item(trans, root, path, &key,
7197                                       datasize);
7198         if (err) {
7199                 drop_inode = 1;
7200                 btrfs_free_path(path);
7201                 goto out_unlock;
7202         }
7203         leaf = path->nodes[0];
7204         ei = btrfs_item_ptr(leaf, path->slots[0],
7205                             struct btrfs_file_extent_item);
7206         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7207         btrfs_set_file_extent_type(leaf, ei,
7208                                    BTRFS_FILE_EXTENT_INLINE);
7209         btrfs_set_file_extent_encryption(leaf, ei, 0);
7210         btrfs_set_file_extent_compression(leaf, ei, 0);
7211         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7212         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7213
7214         ptr = btrfs_file_extent_inline_start(ei);
7215         write_extent_buffer(leaf, symname, ptr, name_len);
7216         btrfs_mark_buffer_dirty(leaf);
7217         btrfs_free_path(path);
7218
7219         inode->i_op = &btrfs_symlink_inode_operations;
7220         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7221         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7222         inode_set_bytes(inode, name_len);
7223         btrfs_i_size_write(inode, name_len - 1);
7224         err = btrfs_update_inode(trans, root, inode);
7225         if (err)
7226                 drop_inode = 1;
7227
7228 out_unlock:
7229         nr = trans->blocks_used;
7230         btrfs_end_transaction_throttle(trans, root);
7231         if (drop_inode) {
7232                 inode_dec_link_count(inode);
7233                 iput(inode);
7234         }
7235         btrfs_btree_balance_dirty(root, nr);
7236         return err;
7237 }
7238
7239 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7240                                        u64 start, u64 num_bytes, u64 min_size,
7241                                        loff_t actual_len, u64 *alloc_hint,
7242                                        struct btrfs_trans_handle *trans)
7243 {
7244         struct btrfs_root *root = BTRFS_I(inode)->root;
7245         struct btrfs_key ins;
7246         u64 cur_offset = start;
7247         u64 i_size;
7248         int ret = 0;
7249         bool own_trans = true;
7250
7251         if (trans)
7252                 own_trans = false;
7253         while (num_bytes > 0) {
7254                 if (own_trans) {
7255                         trans = btrfs_start_transaction(root, 3);
7256                         if (IS_ERR(trans)) {
7257                                 ret = PTR_ERR(trans);
7258                                 break;
7259                         }
7260                 }
7261
7262                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7263                                            0, *alloc_hint, (u64)-1, &ins, 1);
7264                 if (ret) {
7265                         if (own_trans)
7266                                 btrfs_end_transaction(trans, root);
7267                         break;
7268                 }
7269
7270                 ret = insert_reserved_file_extent(trans, inode,
7271                                                   cur_offset, ins.objectid,
7272                                                   ins.offset, ins.offset,
7273                                                   ins.offset, 0, 0, 0,
7274                                                   BTRFS_FILE_EXTENT_PREALLOC);
7275                 BUG_ON(ret);
7276                 btrfs_drop_extent_cache(inode, cur_offset,
7277                                         cur_offset + ins.offset -1, 0);
7278
7279                 num_bytes -= ins.offset;
7280                 cur_offset += ins.offset;
7281                 *alloc_hint = ins.objectid + ins.offset;
7282
7283                 inode->i_ctime = CURRENT_TIME;
7284                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7285                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7286                     (actual_len > inode->i_size) &&
7287                     (cur_offset > inode->i_size)) {
7288                         if (cur_offset > actual_len)
7289                                 i_size = actual_len;
7290                         else
7291                                 i_size = cur_offset;
7292                         i_size_write(inode, i_size);
7293                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7294                 }
7295
7296                 ret = btrfs_update_inode(trans, root, inode);
7297                 BUG_ON(ret);
7298
7299                 if (own_trans)
7300                         btrfs_end_transaction(trans, root);
7301         }
7302         return ret;
7303 }
7304
7305 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7306                               u64 start, u64 num_bytes, u64 min_size,
7307                               loff_t actual_len, u64 *alloc_hint)
7308 {
7309         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7310                                            min_size, actual_len, alloc_hint,
7311                                            NULL);
7312 }
7313
7314 int btrfs_prealloc_file_range_trans(struct inode *inode,
7315                                     struct btrfs_trans_handle *trans, int mode,
7316                                     u64 start, u64 num_bytes, u64 min_size,
7317                                     loff_t actual_len, u64 *alloc_hint)
7318 {
7319         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7320                                            min_size, actual_len, alloc_hint, trans);
7321 }
7322
7323 static int btrfs_set_page_dirty(struct page *page)
7324 {
7325         return __set_page_dirty_nobuffers(page);
7326 }
7327
7328 static int btrfs_permission(struct inode *inode, int mask)
7329 {
7330         struct btrfs_root *root = BTRFS_I(inode)->root;
7331         umode_t mode = inode->i_mode;
7332
7333         if (mask & MAY_WRITE &&
7334             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7335                 if (btrfs_root_readonly(root))
7336                         return -EROFS;
7337                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7338                         return -EACCES;
7339         }
7340         return generic_permission(inode, mask);
7341 }
7342
7343 static const struct inode_operations btrfs_dir_inode_operations = {
7344         .getattr        = btrfs_getattr,
7345         .lookup         = btrfs_lookup,
7346         .create         = btrfs_create,
7347         .unlink         = btrfs_unlink,
7348         .link           = btrfs_link,
7349         .mkdir          = btrfs_mkdir,
7350         .rmdir          = btrfs_rmdir,
7351         .rename         = btrfs_rename,
7352         .symlink        = btrfs_symlink,
7353         .setattr        = btrfs_setattr,
7354         .mknod          = btrfs_mknod,
7355         .setxattr       = btrfs_setxattr,
7356         .getxattr       = btrfs_getxattr,
7357         .listxattr      = btrfs_listxattr,
7358         .removexattr    = btrfs_removexattr,
7359         .permission     = btrfs_permission,
7360         .get_acl        = btrfs_get_acl,
7361 };
7362 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7363         .lookup         = btrfs_lookup,
7364         .permission     = btrfs_permission,
7365         .get_acl        = btrfs_get_acl,
7366 };
7367
7368 static const struct file_operations btrfs_dir_file_operations = {
7369         .llseek         = generic_file_llseek,
7370         .read           = generic_read_dir,
7371         .readdir        = btrfs_real_readdir,
7372         .unlocked_ioctl = btrfs_ioctl,
7373 #ifdef CONFIG_COMPAT
7374         .compat_ioctl   = btrfs_ioctl,
7375 #endif
7376         .release        = btrfs_release_file,
7377         .fsync          = btrfs_sync_file,
7378 };
7379
7380 static struct extent_io_ops btrfs_extent_io_ops = {
7381         .fill_delalloc = run_delalloc_range,
7382         .submit_bio_hook = btrfs_submit_bio_hook,
7383         .merge_bio_hook = btrfs_merge_bio_hook,
7384         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7385         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7386         .writepage_start_hook = btrfs_writepage_start_hook,
7387         .readpage_io_failed_hook = btrfs_io_failed_hook,
7388         .set_bit_hook = btrfs_set_bit_hook,
7389         .clear_bit_hook = btrfs_clear_bit_hook,
7390         .merge_extent_hook = btrfs_merge_extent_hook,
7391         .split_extent_hook = btrfs_split_extent_hook,
7392 };
7393
7394 /*
7395  * btrfs doesn't support the bmap operation because swapfiles
7396  * use bmap to make a mapping of extents in the file.  They assume
7397  * these extents won't change over the life of the file and they
7398  * use the bmap result to do IO directly to the drive.
7399  *
7400  * the btrfs bmap call would return logical addresses that aren't
7401  * suitable for IO and they also will change frequently as COW
7402  * operations happen.  So, swapfile + btrfs == corruption.
7403  *
7404  * For now we're avoiding this by dropping bmap.
7405  */
7406 static const struct address_space_operations btrfs_aops = {
7407         .readpage       = btrfs_readpage,
7408         .writepage      = btrfs_writepage,
7409         .writepages     = btrfs_writepages,
7410         .readpages      = btrfs_readpages,
7411         .direct_IO      = btrfs_direct_IO,
7412         .invalidatepage = btrfs_invalidatepage,
7413         .releasepage    = btrfs_releasepage,
7414         .set_page_dirty = btrfs_set_page_dirty,
7415         .error_remove_page = generic_error_remove_page,
7416 };
7417
7418 static const struct address_space_operations btrfs_symlink_aops = {
7419         .readpage       = btrfs_readpage,
7420         .writepage      = btrfs_writepage,
7421         .invalidatepage = btrfs_invalidatepage,
7422         .releasepage    = btrfs_releasepage,
7423 };
7424
7425 static const struct inode_operations btrfs_file_inode_operations = {
7426         .getattr        = btrfs_getattr,
7427         .setattr        = btrfs_setattr,
7428         .setxattr       = btrfs_setxattr,
7429         .getxattr       = btrfs_getxattr,
7430         .listxattr      = btrfs_listxattr,
7431         .removexattr    = btrfs_removexattr,
7432         .permission     = btrfs_permission,
7433         .fiemap         = btrfs_fiemap,
7434         .get_acl        = btrfs_get_acl,
7435 };
7436 static const struct inode_operations btrfs_special_inode_operations = {
7437         .getattr        = btrfs_getattr,
7438         .setattr        = btrfs_setattr,
7439         .permission     = btrfs_permission,
7440         .setxattr       = btrfs_setxattr,
7441         .getxattr       = btrfs_getxattr,
7442         .listxattr      = btrfs_listxattr,
7443         .removexattr    = btrfs_removexattr,
7444         .get_acl        = btrfs_get_acl,
7445 };
7446 static const struct inode_operations btrfs_symlink_inode_operations = {
7447         .readlink       = generic_readlink,
7448         .follow_link    = page_follow_link_light,
7449         .put_link       = page_put_link,
7450         .getattr        = btrfs_getattr,
7451         .permission     = btrfs_permission,
7452         .setxattr       = btrfs_setxattr,
7453         .getxattr       = btrfs_getxattr,
7454         .listxattr      = btrfs_listxattr,
7455         .removexattr    = btrfs_removexattr,
7456         .get_acl        = btrfs_get_acl,
7457 };
7458
7459 const struct dentry_operations btrfs_dentry_operations = {
7460         .d_delete       = btrfs_dentry_delete,
7461         .d_release      = btrfs_dentry_release,
7462 };