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 "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir)
  97 {
  98         int err;
  99
 100         err = btrfs_init_acl(trans, inode, dir);
 101         if (!err)
 102                 err = btrfs_xattr_security_init(trans, inode, dir);
 103         return err;
 104 }
 105
 106 /*
 107  * this does all the hard work for inserting an inline extent into
 108  * the btree.  The caller should have done a btrfs_drop_extents so that
 109  * no overlapping inline items exist in the btree
 110  */
 111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 112                                 struct btrfs_root *root, struct inode *inode,
 113                                 u64 start, size_t size, size_t compressed_size,
 114                                 struct page **compressed_pages)
 115 {
 116         struct btrfs_key key;
 117         struct btrfs_path *path;
 118         struct extent_buffer *leaf;
 119         struct page *page = NULL;
 120         char *kaddr;
 121         unsigned long ptr;
 122         struct btrfs_file_extent_item *ei;
 123         int err = 0;
 124         int ret;
 125         size_t cur_size = size;
 126         size_t datasize;
 127         unsigned long offset;
 128         int compress_type = BTRFS_COMPRESS_NONE;
 129
 130         if (compressed_size && compressed_pages) {
 131                 compress_type = root->fs_info->compress_type;
 132                 cur_size = compressed_size;
 133         }
 134
 135         path = btrfs_alloc_path();
 136         if (!path)
 137                 return -ENOMEM;
 138
 139         path->leave_spinning = 1;
 140         btrfs_set_trans_block_group(trans, inode);
 141
 142         key.objectid = inode->i_ino;
 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,
 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                                    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         actual_end = min_t(u64, isize, end + 1);
 345 again:
 346         will_compress = 0;
 347         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 348         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 349
 350         /*
 351          * we don't want to send crud past the end of i_size through
 352          * compression, that's just a waste of CPU time.  So, if the
 353          * end of the file is before the start of our current
 354          * requested range of bytes, we bail out to the uncompressed
 355          * cleanup code that can deal with all of this.
 356          *
 357          * It isn't really the fastest way to fix things, but this is a
 358          * very uncommon corner.
 359          */
 360         if (actual_end <= start)
 361                 goto cleanup_and_bail_uncompressed;
 362
 363         total_compressed = actual_end - start;
 364
 365         /* we want to make sure that amount of ram required to uncompress
 366          * an extent is reasonable, so we limit the total size in ram
 367          * of a compressed extent to 128k.  This is a crucial number
 368          * because it also controls how easily we can spread reads across
 369          * cpus for decompression.
 370          *
 371          * We also want to make sure the amount of IO required to do
 372          * a random read is reasonably small, so we limit the size of
 373          * a compressed extent to 128k.
 374          */
 375         total_compressed = min(total_compressed, max_uncompressed);
 376         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 377         num_bytes = max(blocksize,  num_bytes);
 378         total_in = 0;
 379         ret = 0;
 380
 381         /*
 382          * we do compression for mount -o compress and when the
 383          * inode has not been flagged as nocompress.  This flag can
 384          * change at any time if we discover bad compression ratios.
 385          */
 386         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 387             (btrfs_test_opt(root, COMPRESS) ||
 388              (BTRFS_I(inode)->force_compress) ||
 389              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 390                 WARN_ON(pages);
 391                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 392                 BUG_ON(!pages);
 393
 394                 if (BTRFS_I(inode)->force_compress)
 395                         compress_type = BTRFS_I(inode)->force_compress;
 396
 397                 ret = btrfs_compress_pages(compress_type,
 398                                            inode->i_mapping, start,
 399                                            total_compressed, pages,
 400                                            nr_pages, &nr_pages_ret,
 401                                            &total_in,
 402                                            &total_compressed,
 403                                            max_compressed);
 404
 405                 if (!ret) {
 406                         unsigned long offset = total_compressed &
 407                                 (PAGE_CACHE_SIZE - 1);
 408                         struct page *page = pages[nr_pages_ret - 1];
 409                         char *kaddr;
 410
 411                         /* zero the tail end of the last page, we might be
 412                          * sending it down to disk
 413                          */
 414                         if (offset) {
 415                                 kaddr = kmap_atomic(page, KM_USER0);
 416                                 memset(kaddr + offset, 0,
 417                                        PAGE_CACHE_SIZE - offset);
 418                                 kunmap_atomic(kaddr, KM_USER0);
 419                         }
 420                         will_compress = 1;
 421                 }
 422         }
 423         if (start == 0) {
 424                 trans = btrfs_join_transaction(root, 1);
 425                 BUG_ON(IS_ERR(trans));
 426                 btrfs_set_trans_block_group(trans, inode);
 427                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 428
 429                 /* lets try to make an inline extent */
 430                 if (ret || total_in < (actual_end - start)) {
 431                         /* we didn't compress the entire range, try
 432                          * to make an uncompressed inline extent.
 433                          */
 434                         ret = cow_file_range_inline(trans, root, inode,
 435                                                     start, end, 0, NULL);
 436                 } else {
 437                         /* try making a compressed inline extent */
 438                         ret = cow_file_range_inline(trans, root, inode,
 439                                                     start, end,
 440                                                     total_compressed, pages);
 441                 }
 442                 if (ret == 0) {
 443                         /*
 444                          * inline extent creation worked, we don't need
 445                          * to create any more async work items.  Unlock
 446                          * and free up our temp pages.
 447                          */
 448                         extent_clear_unlock_delalloc(inode,
 449                              &BTRFS_I(inode)->io_tree,
 450                              start, end, NULL,
 451                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 452                              EXTENT_CLEAR_DELALLOC |
 453                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 454
 455                         btrfs_end_transaction(trans, root);
 456                         goto free_pages_out;
 457                 }
 458                 btrfs_end_transaction(trans, root);
 459         }
 460
 461         if (will_compress) {
 462                 /*
 463                  * we aren't doing an inline extent round the compressed size
 464                  * up to a block size boundary so the allocator does sane
 465                  * things
 466                  */
 467                 total_compressed = (total_compressed + blocksize - 1) &
 468                         ~(blocksize - 1);
 469
 470                 /*
 471                  * one last check to make sure the compression is really a
 472                  * win, compare the page count read with the blocks on disk
 473                  */
 474                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 475                         ~(PAGE_CACHE_SIZE - 1);
 476                 if (total_compressed >= total_in) {
 477                         will_compress = 0;
 478                 } else {
 479                         num_bytes = total_in;
 480                 }
 481         }
 482         if (!will_compress && pages) {
 483                 /*
 484                  * the compression code ran but failed to make things smaller,
 485                  * free any pages it allocated and our page pointer array
 486                  */
 487                 for (i = 0; i < nr_pages_ret; i++) {
 488                         WARN_ON(pages[i]->mapping);
 489                         page_cache_release(pages[i]);
 490                 }
 491                 kfree(pages);
 492                 pages = NULL;
 493                 total_compressed = 0;
 494                 nr_pages_ret = 0;
 495
 496                 /* flag the file so we don't compress in the future */
 497                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 498                     !(BTRFS_I(inode)->force_compress)) {
 499                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 500                 }
 501         }
 502         if (will_compress) {
 503                 *num_added += 1;
 504
 505                 /* the async work queues will take care of doing actual
 506                  * allocation on disk for these compressed pages,
 507                  * and will submit them to the elevator.
 508                  */
 509                 add_async_extent(async_cow, start, num_bytes,
 510                                  total_compressed, pages, nr_pages_ret,
 511                                  compress_type);
 512
 513                 if (start + num_bytes < end) {
 514                         start += num_bytes;
 515                         pages = NULL;
 516                         cond_resched();
 517                         goto again;
 518                 }
 519         } else {
 520 cleanup_and_bail_uncompressed:
 521                 /*
 522                  * No compression, but we still need to write the pages in
 523                  * the file we've been given so far.  redirty the locked
 524                  * page if it corresponds to our extent and set things up
 525                  * for the async work queue to run cow_file_range to do
 526                  * the normal delalloc dance
 527                  */
 528                 if (page_offset(locked_page) >= start &&
 529                     page_offset(locked_page) <= end) {
 530                         __set_page_dirty_nobuffers(locked_page);
 531                         /* unlocked later on in the async handlers */
 532                 }
 533                 add_async_extent(async_cow, start, end - start + 1,
 534                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 535                 *num_added += 1;
 536         }
 537
 538 out:
 539         return 0;
 540
 541 free_pages_out:
 542         for (i = 0; i < nr_pages_ret; i++) {
 543                 WARN_ON(pages[i]->mapping);
 544                 page_cache_release(pages[i]);
 545         }
 546         kfree(pages);
 547
 548         goto out;
 549 }
 550
 551 /*
 552  * phase two of compressed writeback.  This is the ordered portion
 553  * of the code, which only gets called in the order the work was
 554  * queued.  We walk all the async extents created by compress_file_range
 555  * and send them down to the disk.
 556  */
 557 static noinline int submit_compressed_extents(struct inode *inode,
 558                                               struct async_cow *async_cow)
 559 {
 560         struct async_extent *async_extent;
 561         u64 alloc_hint = 0;
 562         struct btrfs_trans_handle *trans;
 563         struct btrfs_key ins;
 564         struct extent_map *em;
 565         struct btrfs_root *root = BTRFS_I(inode)->root;
 566         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 567         struct extent_io_tree *io_tree;
 568         int ret = 0;
 569
 570         if (list_empty(&async_cow->extents))
 571                 return 0;
 572
 573
 574         while (!list_empty(&async_cow->extents)) {
 575                 async_extent = list_entry(async_cow->extents.next,
 576                                           struct async_extent, list);
 577                 list_del(&async_extent->list);
 578
 579                 io_tree = &BTRFS_I(inode)->io_tree;
 580
 581 retry:
 582                 /* did the compression code fall back to uncompressed IO? */
 583                 if (!async_extent->pages) {
 584                         int page_started = 0;
 585                         unsigned long nr_written = 0;
 586
 587                         lock_extent(io_tree, async_extent->start,
 588                                          async_extent->start +
 589                                          async_extent->ram_size - 1, GFP_NOFS);
 590
 591                         /* allocate blocks */
 592                         ret = cow_file_range(inode, async_cow->locked_page,
 593                                              async_extent->start,
 594                                              async_extent->start +
 595                                              async_extent->ram_size - 1,
 596                                              &page_started, &nr_written, 0);
 597
 598                         /*
 599                          * if page_started, cow_file_range inserted an
 600                          * inline extent and took care of all the unlocking
 601                          * and IO for us.  Otherwise, we need to submit
 602                          * all those pages down to the drive.
 603                          */
 604                         if (!page_started && !ret)
 605                                 extent_write_locked_range(io_tree,
 606                                                   inode, async_extent->start,
 607                                                   async_extent->start +
 608                                                   async_extent->ram_size - 1,
 609                                                   btrfs_get_extent,
 610                                                   WB_SYNC_ALL);
 611                         kfree(async_extent);
 612                         cond_resched();
 613                         continue;
 614                 }
 615
 616                 lock_extent(io_tree, async_extent->start,
 617                             async_extent->start + async_extent->ram_size - 1,
 618                             GFP_NOFS);
 619
 620                 trans = btrfs_join_transaction(root, 1);
 621                 BUG_ON(IS_ERR(trans));
 622                 ret = btrfs_reserve_extent(trans, root,
 623                                            async_extent->compressed_size,
 624                                            async_extent->compressed_size,
 625                                            0, alloc_hint,
 626                                            (u64)-1, &ins, 1);
 627                 btrfs_end_transaction(trans, root);
 628
 629                 if (ret) {
 630                         int i;
 631                         for (i = 0; i < async_extent->nr_pages; i++) {
 632                                 WARN_ON(async_extent->pages[i]->mapping);
 633                                 page_cache_release(async_extent->pages[i]);
 634                         }
 635                         kfree(async_extent->pages);
 636                         async_extent->nr_pages = 0;
 637                         async_extent->pages = NULL;
 638                         unlock_extent(io_tree, async_extent->start,
 639                                       async_extent->start +
 640                                       async_extent->ram_size - 1, GFP_NOFS);
 641                         goto retry;
 642                 }
 643
 644                 /*
 645                  * here we're doing allocation and writeback of the
 646                  * compressed pages
 647                  */
 648                 btrfs_drop_extent_cache(inode, async_extent->start,
 649                                         async_extent->start +
 650                                         async_extent->ram_size - 1, 0);
 651
 652                 em = alloc_extent_map(GFP_NOFS);
 653                 BUG_ON(!em);
 654                 em->start = async_extent->start;
 655                 em->len = async_extent->ram_size;
 656                 em->orig_start = em->start;
 657
 658                 em->block_start = ins.objectid;
 659                 em->block_len = ins.offset;
 660                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 661                 em->compress_type = async_extent->compress_type;
 662                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 663                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 664
 665                 while (1) {
 666                         write_lock(&em_tree->lock);
 667                         ret = add_extent_mapping(em_tree, em);
 668                         write_unlock(&em_tree->lock);
 669                         if (ret != -EEXIST) {
 670                                 free_extent_map(em);
 671                                 break;
 672                         }
 673                         btrfs_drop_extent_cache(inode, async_extent->start,
 674                                                 async_extent->start +
 675                                                 async_extent->ram_size - 1, 0);
 676                 }
 677
 678                 ret = btrfs_add_ordered_extent_compress(inode,
 679                                                 async_extent->start,
 680                                                 ins.objectid,
 681                                                 async_extent->ram_size,
 682                                                 ins.offset,
 683                                                 BTRFS_ORDERED_COMPRESSED,
 684                                                 async_extent->compress_type);
 685                 BUG_ON(ret);
 686
 687                 /*
 688                  * clear dirty, set writeback and unlock the pages.
 689                  */
 690                 extent_clear_unlock_delalloc(inode,
 691                                 &BTRFS_I(inode)->io_tree,
 692                                 async_extent->start,
 693                                 async_extent->start +
 694                                 async_extent->ram_size - 1,
 695                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 696                                 EXTENT_CLEAR_UNLOCK |
 697                                 EXTENT_CLEAR_DELALLOC |
 698                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 699
 700                 ret = btrfs_submit_compressed_write(inode,
 701                                     async_extent->start,
 702                                     async_extent->ram_size,
 703                                     ins.objectid,
 704                                     ins.offset, async_extent->pages,
 705                                     async_extent->nr_pages);
 706
 707                 BUG_ON(ret);
 708                 alloc_hint = ins.objectid + ins.offset;
 709                 kfree(async_extent);
 710                 cond_resched();
 711         }
 712
 713         return 0;
 714 }
 715
 716 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 717                                       u64 num_bytes)
 718 {
 719         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 720         struct extent_map *em;
 721         u64 alloc_hint = 0;
 722
 723         read_lock(&em_tree->lock);
 724         em = search_extent_mapping(em_tree, start, num_bytes);
 725         if (em) {
 726                 /*
 727                  * if block start isn't an actual block number then find the
 728                  * first block in this inode and use that as a hint.  If that
 729                  * block is also bogus then just don't worry about it.
 730                  */
 731                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 732                         free_extent_map(em);
 733                         em = search_extent_mapping(em_tree, 0, 0);
 734                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 735                                 alloc_hint = em->block_start;
 736                         if (em)
 737                                 free_extent_map(em);
 738                 } else {
 739                         alloc_hint = em->block_start;
 740                         free_extent_map(em);
 741                 }
 742         }
 743         read_unlock(&em_tree->lock);
 744
 745         return alloc_hint;
 746 }
 747
 748 /*
 749  * when extent_io.c finds a delayed allocation range in the file,
 750  * the call backs end up in this code.  The basic idea is to
 751  * allocate extents on disk for the range, and create ordered data structs
 752  * in ram to track those extents.
 753  *
 754  * locked_page is the page that writepage had locked already.  We use
 755  * it to make sure we don't do extra locks or unlocks.
 756  *
 757  * *page_started is set to one if we unlock locked_page and do everything
 758  * required to start IO on it.  It may be clean and already done with
 759  * IO when we return.
 760  */
 761 static noinline int cow_file_range(struct inode *inode,
 762                                    struct page *locked_page,
 763                                    u64 start, u64 end, int *page_started,
 764                                    unsigned long *nr_written,
 765                                    int unlock)
 766 {
 767         struct btrfs_root *root = BTRFS_I(inode)->root;
 768         struct btrfs_trans_handle *trans;
 769         u64 alloc_hint = 0;
 770         u64 num_bytes;
 771         unsigned long ram_size;
 772         u64 disk_num_bytes;
 773         u64 cur_alloc_size;
 774         u64 blocksize = root->sectorsize;
 775         struct btrfs_key ins;
 776         struct extent_map *em;
 777         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 778         int ret = 0;
 779
 780         BUG_ON(root == root->fs_info->tree_root);
 781         trans = btrfs_join_transaction(root, 1);
 782         BUG_ON(IS_ERR(trans));
 783         btrfs_set_trans_block_group(trans, inode);
 784         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 785
 786         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 787         num_bytes = max(blocksize,  num_bytes);
 788         disk_num_bytes = num_bytes;
 789         ret = 0;
 790
 791         if (start == 0) {
 792                 /* lets try to make an inline extent */
 793                 ret = cow_file_range_inline(trans, root, inode,
 794                                             start, end, 0, NULL);
 795                 if (ret == 0) {
 796                         extent_clear_unlock_delalloc(inode,
 797                                      &BTRFS_I(inode)->io_tree,
 798                                      start, end, NULL,
 799                                      EXTENT_CLEAR_UNLOCK_PAGE |
 800                                      EXTENT_CLEAR_UNLOCK |
 801                                      EXTENT_CLEAR_DELALLOC |
 802                                      EXTENT_CLEAR_DIRTY |
 803                                      EXTENT_SET_WRITEBACK |
 804                                      EXTENT_END_WRITEBACK);
 805
 806                         *nr_written = *nr_written +
 807                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 808                         *page_started = 1;
 809                         ret = 0;
 810                         goto out;
 811                 }
 812         }
 813
 814         BUG_ON(disk_num_bytes >
 815                btrfs_super_total_bytes(&root->fs_info->super_copy));
 816
 817         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 818         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 819
 820         while (disk_num_bytes > 0) {
 821                 unsigned long op;
 822
 823                 cur_alloc_size = disk_num_bytes;
 824                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 825                                            root->sectorsize, 0, alloc_hint,
 826                                            (u64)-1, &ins, 1);
 827                 BUG_ON(ret);
 828
 829                 em = alloc_extent_map(GFP_NOFS);
 830                 BUG_ON(!em);
 831                 em->start = start;
 832                 em->orig_start = em->start;
 833                 ram_size = ins.offset;
 834                 em->len = ins.offset;
 835
 836                 em->block_start = ins.objectid;
 837                 em->block_len = ins.offset;
 838                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 839                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 840
 841                 while (1) {
 842                         write_lock(&em_tree->lock);
 843                         ret = add_extent_mapping(em_tree, em);
 844                         write_unlock(&em_tree->lock);
 845                         if (ret != -EEXIST) {
 846                                 free_extent_map(em);
 847                                 break;
 848                         }
 849                         btrfs_drop_extent_cache(inode, start,
 850                                                 start + ram_size - 1, 0);
 851                 }
 852
 853                 cur_alloc_size = ins.offset;
 854                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 855                                                ram_size, cur_alloc_size, 0);
 856                 BUG_ON(ret);
 857
 858                 if (root->root_key.objectid ==
 859                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 860                         ret = btrfs_reloc_clone_csums(inode, start,
 861                                                       cur_alloc_size);
 862                         BUG_ON(ret);
 863                 }
 864
 865                 if (disk_num_bytes < cur_alloc_size)
 866                         break;
 867
 868                 /* we're not doing compressed IO, don't unlock the first
 869                  * page (which the caller expects to stay locked), don't
 870                  * clear any dirty bits and don't set any writeback bits
 871                  *
 872                  * Do set the Private2 bit so we know this page was properly
 873                  * setup for writepage
 874                  */
 875                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 876                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 877                         EXTENT_SET_PRIVATE2;
 878
 879                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 880                                              start, start + ram_size - 1,
 881                                              locked_page, op);
 882                 disk_num_bytes -= cur_alloc_size;
 883                 num_bytes -= cur_alloc_size;
 884                 alloc_hint = ins.objectid + ins.offset;
 885                 start += cur_alloc_size;
 886         }
 887 out:
 888         ret = 0;
 889         btrfs_end_transaction(trans, root);
 890
 891         return ret;
 892 }
 893
 894 /*
 895  * work queue call back to started compression on a file and pages
 896  */
 897 static noinline void async_cow_start(struct btrfs_work *work)
 898 {
 899         struct async_cow *async_cow;
 900         int num_added = 0;
 901         async_cow = container_of(work, struct async_cow, work);
 902
 903         compress_file_range(async_cow->inode, async_cow->locked_page,
 904                             async_cow->start, async_cow->end, async_cow,
 905                             &num_added);
 906         if (num_added == 0)
 907                 async_cow->inode = NULL;
 908 }
 909
 910 /*
 911  * work queue call back to submit previously compressed pages
 912  */
 913 static noinline void async_cow_submit(struct btrfs_work *work)
 914 {
 915         struct async_cow *async_cow;
 916         struct btrfs_root *root;
 917         unsigned long nr_pages;
 918
 919         async_cow = container_of(work, struct async_cow, work);
 920
 921         root = async_cow->root;
 922         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 923                 PAGE_CACHE_SHIFT;
 924
 925         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 926
 927         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 928             5 * 1042 * 1024 &&
 929             waitqueue_active(&root->fs_info->async_submit_wait))
 930                 wake_up(&root->fs_info->async_submit_wait);
 931
 932         if (async_cow->inode)
 933                 submit_compressed_extents(async_cow->inode, async_cow);
 934 }
 935
 936 static noinline void async_cow_free(struct btrfs_work *work)
 937 {
 938         struct async_cow *async_cow;
 939         async_cow = container_of(work, struct async_cow, work);
 940         kfree(async_cow);
 941 }
 942
 943 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 944                                 u64 start, u64 end, int *page_started,
 945                                 unsigned long *nr_written)
 946 {
 947         struct async_cow *async_cow;
 948         struct btrfs_root *root = BTRFS_I(inode)->root;
 949         unsigned long nr_pages;
 950         u64 cur_end;
 951         int limit = 10 * 1024 * 1042;
 952
 953         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 954                          1, 0, NULL, GFP_NOFS);
 955         while (start < end) {
 956                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 957                 async_cow->inode = inode;
 958                 async_cow->root = root;
 959                 async_cow->locked_page = locked_page;
 960                 async_cow->start = start;
 961
 962                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 963                         cur_end = end;
 964                 else
 965                         cur_end = min(end, start + 512 * 1024 - 1);
 966
 967                 async_cow->end = cur_end;
 968                 INIT_LIST_HEAD(&async_cow->extents);
 969
 970                 async_cow->work.func = async_cow_start;
 971                 async_cow->work.ordered_func = async_cow_submit;
 972                 async_cow->work.ordered_free = async_cow_free;
 973                 async_cow->work.flags = 0;
 974
 975                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 976                         PAGE_CACHE_SHIFT;
 977                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 978
 979                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 980                                    &async_cow->work);
 981
 982                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 983                         wait_event(root->fs_info->async_submit_wait,
 984                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 985                             limit));
 986                 }
 987
 988                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 989                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 990                         wait_event(root->fs_info->async_submit_wait,
 991                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 992                            0));
 993                 }
 994
 995                 *nr_written += nr_pages;
 996                 start = cur_end + 1;
 997         }
 998         *page_started = 1;
 999         return 0;
1000 }
1001
1002 static noinline int csum_exist_in_range(struct btrfs_root *root,
1003                                         u64 bytenr, u64 num_bytes)
1004 {
1005         int ret;
1006         struct btrfs_ordered_sum *sums;
1007         LIST_HEAD(list);
1008
1009         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1010                                        bytenr + num_bytes - 1, &list);
1011         if (ret == 0 && list_empty(&list))
1012                 return 0;
1013
1014         while (!list_empty(&list)) {
1015                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1016                 list_del(&sums->list);
1017                 kfree(sums);
1018         }
1019         return 1;
1020 }
1021
1022 /*
1023  * when nowcow writeback call back.  This checks for snapshots or COW copies
1024  * of the extents that exist in the file, and COWs the file as required.
1025  *
1026  * If no cow copies or snapshots exist, we write directly to the existing
1027  * blocks on disk
1028  */
1029 static noinline int run_delalloc_nocow(struct inode *inode,
1030                                        struct page *locked_page,
1031                               u64 start, u64 end, int *page_started, int force,
1032                               unsigned long *nr_written)
1033 {
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         struct btrfs_trans_handle *trans;
1036         struct extent_buffer *leaf;
1037         struct btrfs_path *path;
1038         struct btrfs_file_extent_item *fi;
1039         struct btrfs_key found_key;
1040         u64 cow_start;
1041         u64 cur_offset;
1042         u64 extent_end;
1043         u64 extent_offset;
1044         u64 disk_bytenr;
1045         u64 num_bytes;
1046         int extent_type;
1047         int ret;
1048         int type;
1049         int nocow;
1050         int check_prev = 1;
1051         bool nolock = false;
1052
1053         path = btrfs_alloc_path();
1054         BUG_ON(!path);
1055         if (root == root->fs_info->tree_root) {
1056                 nolock = true;
1057                 trans = btrfs_join_transaction_nolock(root, 1);
1058         } else {
1059                 trans = btrfs_join_transaction(root, 1);
1060         }
1061         BUG_ON(IS_ERR(trans));
1062
1063         cow_start = (u64)-1;
1064         cur_offset = start;
1065         while (1) {
1066                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1067                                                cur_offset, 0);
1068                 BUG_ON(ret < 0);
1069                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1070                         leaf = path->nodes[0];
1071                         btrfs_item_key_to_cpu(leaf, &found_key,
1072                                               path->slots[0] - 1);
1073                         if (found_key.objectid == inode->i_ino &&
1074                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1075                                 path->slots[0]--;
1076                 }
1077                 check_prev = 0;
1078 next_slot:
1079                 leaf = path->nodes[0];
1080                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1081                         ret = btrfs_next_leaf(root, path);
1082                         if (ret < 0)
1083                                 BUG_ON(1);
1084                         if (ret > 0)
1085                                 break;
1086                         leaf = path->nodes[0];
1087                 }
1088
1089                 nocow = 0;
1090                 disk_bytenr = 0;
1091                 num_bytes = 0;
1092                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1093
1094                 if (found_key.objectid > inode->i_ino ||
1095                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1096                     found_key.offset > end)
1097                         break;
1098
1099                 if (found_key.offset > cur_offset) {
1100                         extent_end = found_key.offset;
1101                         extent_type = 0;
1102                         goto out_check;
1103                 }
1104
1105                 fi = btrfs_item_ptr(leaf, path->slots[0],
1106                                     struct btrfs_file_extent_item);
1107                 extent_type = btrfs_file_extent_type(leaf, fi);
1108
1109                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1110                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1111                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1112                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1113                         extent_end = found_key.offset +
1114                                 btrfs_file_extent_num_bytes(leaf, fi);
1115                         if (extent_end <= start) {
1116                                 path->slots[0]++;
1117                                 goto next_slot;
1118                         }
1119                         if (disk_bytenr == 0)
1120                                 goto out_check;
1121                         if (btrfs_file_extent_compression(leaf, fi) ||
1122                             btrfs_file_extent_encryption(leaf, fi) ||
1123                             btrfs_file_extent_other_encoding(leaf, fi))
1124                                 goto out_check;
1125                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1126                                 goto out_check;
1127                         if (btrfs_extent_readonly(root, disk_bytenr))
1128                                 goto out_check;
1129                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1130                                                   found_key.offset -
1131                                                   extent_offset, disk_bytenr))
1132                                 goto out_check;
1133                         disk_bytenr += extent_offset;
1134                         disk_bytenr += cur_offset - found_key.offset;
1135                         num_bytes = min(end + 1, extent_end) - cur_offset;
1136                         /*
1137                          * force cow if csum exists in the range.
1138                          * this ensure that csum for a given extent are
1139                          * either valid or do not exist.
1140                          */
1141                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1142                                 goto out_check;
1143                         nocow = 1;
1144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1145                         extent_end = found_key.offset +
1146                                 btrfs_file_extent_inline_len(leaf, fi);
1147                         extent_end = ALIGN(extent_end, root->sectorsize);
1148                 } else {
1149                         BUG_ON(1);
1150                 }
1151 out_check:
1152                 if (extent_end <= start) {
1153                         path->slots[0]++;
1154                         goto next_slot;
1155                 }
1156                 if (!nocow) {
1157                         if (cow_start == (u64)-1)
1158                                 cow_start = cur_offset;
1159                         cur_offset = extent_end;
1160                         if (cur_offset > end)
1161                                 break;
1162                         path->slots[0]++;
1163                         goto next_slot;
1164                 }
1165
1166                 btrfs_release_path(root, path);
1167                 if (cow_start != (u64)-1) {
1168                         ret = cow_file_range(inode, locked_page, cow_start,
1169                                         found_key.offset - 1, page_started,
1170                                         nr_written, 1);
1171                         BUG_ON(ret);
1172                         cow_start = (u64)-1;
1173                 }
1174
1175                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1176                         struct extent_map *em;
1177                         struct extent_map_tree *em_tree;
1178                         em_tree = &BTRFS_I(inode)->extent_tree;
1179                         em = alloc_extent_map(GFP_NOFS);
1180                         BUG_ON(!em);
1181                         em->start = cur_offset;
1182                         em->orig_start = em->start;
1183                         em->len = num_bytes;
1184                         em->block_len = num_bytes;
1185                         em->block_start = disk_bytenr;
1186                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1187                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1188                         while (1) {
1189                                 write_lock(&em_tree->lock);
1190                                 ret = add_extent_mapping(em_tree, em);
1191                                 write_unlock(&em_tree->lock);
1192                                 if (ret != -EEXIST) {
1193                                         free_extent_map(em);
1194                                         break;
1195                                 }
1196                                 btrfs_drop_extent_cache(inode, em->start,
1197                                                 em->start + em->len - 1, 0);
1198                         }
1199                         type = BTRFS_ORDERED_PREALLOC;
1200                 } else {
1201                         type = BTRFS_ORDERED_NOCOW;
1202                 }
1203
1204                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1205                                                num_bytes, num_bytes, type);
1206                 BUG_ON(ret);
1207
1208                 if (root->root_key.objectid ==
1209                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1210                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1211                                                       num_bytes);
1212                         BUG_ON(ret);
1213                 }
1214
1215                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1216                                 cur_offset, cur_offset + num_bytes - 1,
1217                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1218                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1219                                 EXTENT_SET_PRIVATE2);
1220                 cur_offset = extent_end;
1221                 if (cur_offset > end)
1222                         break;
1223         }
1224         btrfs_release_path(root, path);
1225
1226         if (cur_offset <= end && cow_start == (u64)-1)
1227                 cow_start = cur_offset;
1228         if (cow_start != (u64)-1) {
1229                 ret = cow_file_range(inode, locked_page, cow_start, end,
1230                                      page_started, nr_written, 1);
1231                 BUG_ON(ret);
1232         }
1233
1234         if (nolock) {
1235                 ret = btrfs_end_transaction_nolock(trans, root);
1236                 BUG_ON(ret);
1237         } else {
1238                 ret = btrfs_end_transaction(trans, root);
1239                 BUG_ON(ret);
1240         }
1241         btrfs_free_path(path);
1242         return 0;
1243 }
1244
1245 /*
1246  * extent_io.c call back to do delayed allocation processing
1247  */
1248 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1249                               u64 start, u64 end, int *page_started,
1250                               unsigned long *nr_written)
1251 {
1252         int ret;
1253         struct btrfs_root *root = BTRFS_I(inode)->root;
1254
1255         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1256                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1257                                          page_started, 1, nr_written);
1258         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1259                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1260                                          page_started, 0, nr_written);
1261         else if (!btrfs_test_opt(root, COMPRESS) &&
1262                  !(BTRFS_I(inode)->force_compress) &&
1263                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1264                 ret = cow_file_range(inode, locked_page, start, end,
1265                                       page_started, nr_written, 1);
1266         else
1267                 ret = cow_file_range_async(inode, locked_page, start, end,
1268                                            page_started, nr_written);
1269         return ret;
1270 }
1271
1272 static int btrfs_split_extent_hook(struct inode *inode,
1273                                    struct extent_state *orig, u64 split)
1274 {
1275         /* not delalloc, ignore it */
1276         if (!(orig->state & EXTENT_DELALLOC))
1277                 return 0;
1278
1279         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1280         return 0;
1281 }
1282
1283 /*
1284  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1285  * extents so we can keep track of new extents that are just merged onto old
1286  * extents, such as when we are doing sequential writes, so we can properly
1287  * account for the metadata space we'll need.
1288  */
1289 static int btrfs_merge_extent_hook(struct inode *inode,
1290                                    struct extent_state *new,
1291                                    struct extent_state *other)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(other->state & EXTENT_DELALLOC))
1295                 return 0;
1296
1297         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1298         return 0;
1299 }
1300
1301 /*
1302  * extent_io.c set_bit_hook, used to track delayed allocation
1303  * bytes in this file, and to maintain the list of inodes that
1304  * have pending delalloc work to be done.
1305  */
1306 static int btrfs_set_bit_hook(struct inode *inode,
1307                               struct extent_state *state, int *bits)
1308 {
1309
1310         /*
1311          * set_bit and clear bit hooks normally require _irqsave/restore
1312          * but in this case, we are only testeing for the DELALLOC
1313          * bit, which is only set or cleared with irqs on
1314          */
1315         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1316                 struct btrfs_root *root = BTRFS_I(inode)->root;
1317                 u64 len = state->end + 1 - state->start;
1318                 int do_list = (root->root_key.objectid !=
1319                                BTRFS_ROOT_TREE_OBJECTID);
1320
1321                 if (*bits & EXTENT_FIRST_DELALLOC)
1322                         *bits &= ~EXTENT_FIRST_DELALLOC;
1323                 else
1324                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1325
1326                 spin_lock(&root->fs_info->delalloc_lock);
1327                 BTRFS_I(inode)->delalloc_bytes += len;
1328                 root->fs_info->delalloc_bytes += len;
1329                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1330                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1331                                       &root->fs_info->delalloc_inodes);
1332                 }
1333                 spin_unlock(&root->fs_info->delalloc_lock);
1334         }
1335         return 0;
1336 }
1337
1338 /*
1339  * extent_io.c clear_bit_hook, see set_bit_hook for why
1340  */
1341 static int btrfs_clear_bit_hook(struct inode *inode,
1342                                 struct extent_state *state, int *bits)
1343 {
1344         /*
1345          * set_bit and clear bit hooks normally require _irqsave/restore
1346          * but in this case, we are only testeing for the DELALLOC
1347          * bit, which is only set or cleared with irqs on
1348          */
1349         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1350                 struct btrfs_root *root = BTRFS_I(inode)->root;
1351                 u64 len = state->end + 1 - state->start;
1352                 int do_list = (root->root_key.objectid !=
1353                                BTRFS_ROOT_TREE_OBJECTID);
1354
1355                 if (*bits & EXTENT_FIRST_DELALLOC)
1356                         *bits &= ~EXTENT_FIRST_DELALLOC;
1357                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1358                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1359
1360                 if (*bits & EXTENT_DO_ACCOUNTING)
1361                         btrfs_delalloc_release_metadata(inode, len);
1362
1363                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1364                     && do_list)
1365                         btrfs_free_reserved_data_space(inode, len);
1366
1367                 spin_lock(&root->fs_info->delalloc_lock);
1368                 root->fs_info->delalloc_bytes -= len;
1369                 BTRFS_I(inode)->delalloc_bytes -= len;
1370
1371                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1372                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1373                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1374                 }
1375                 spin_unlock(&root->fs_info->delalloc_lock);
1376         }
1377         return 0;
1378 }
1379
1380 /*
1381  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1382  * we don't create bios that span stripes or chunks
1383  */
1384 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1385                          size_t size, struct bio *bio,
1386                          unsigned long bio_flags)
1387 {
1388         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1389         struct btrfs_mapping_tree *map_tree;
1390         u64 logical = (u64)bio->bi_sector << 9;
1391         u64 length = 0;
1392         u64 map_length;
1393         int ret;
1394
1395         if (bio_flags & EXTENT_BIO_COMPRESSED)
1396                 return 0;
1397
1398         length = bio->bi_size;
1399         map_tree = &root->fs_info->mapping_tree;
1400         map_length = length;
1401         ret = btrfs_map_block(map_tree, READ, logical,
1402                               &map_length, NULL, 0);
1403
1404         if (map_length < length + size)
1405                 return 1;
1406         return ret;
1407 }
1408
1409 /*
1410  * in order to insert checksums into the metadata in large chunks,
1411  * we wait until bio submission time.   All the pages in the bio are
1412  * checksummed and sums are attached onto the ordered extent record.
1413  *
1414  * At IO completion time the cums attached on the ordered extent record
1415  * are inserted into the btree
1416  */
1417 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1418                                     struct bio *bio, int mirror_num,
1419                                     unsigned long bio_flags,
1420                                     u64 bio_offset)
1421 {
1422         struct btrfs_root *root = BTRFS_I(inode)->root;
1423         int ret = 0;
1424
1425         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1426         BUG_ON(ret);
1427         return 0;
1428 }
1429
1430 /*
1431  * in order to insert checksums into the metadata in large chunks,
1432  * we wait until bio submission time.   All the pages in the bio are
1433  * checksummed and sums are attached onto the ordered extent record.
1434  *
1435  * At IO completion time the cums attached on the ordered extent record
1436  * are inserted into the btree
1437  */
1438 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1439                           int mirror_num, unsigned long bio_flags,
1440                           u64 bio_offset)
1441 {
1442         struct btrfs_root *root = BTRFS_I(inode)->root;
1443         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1444 }
1445
1446 /*
1447  * extent_io.c submission hook. This does the right thing for csum calculation
1448  * on write, or reading the csums from the tree before a read
1449  */
1450 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1451                           int mirror_num, unsigned long bio_flags,
1452                           u64 bio_offset)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         int ret = 0;
1456         int skip_sum;
1457
1458         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1459
1460         if (root == root->fs_info->tree_root)
1461                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1462         else
1463                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1464         BUG_ON(ret);
1465
1466         if (!(rw & REQ_WRITE)) {
1467                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1468                         return btrfs_submit_compressed_read(inode, bio,
1469                                                     mirror_num, bio_flags);
1470                 } else if (!skip_sum)
1471                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1472                 goto mapit;
1473         } else if (!skip_sum) {
1474                 /* csum items have already been cloned */
1475                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1476                         goto mapit;
1477                 /* we're doing a write, do the async checksumming */
1478                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1479                                    inode, rw, bio, mirror_num,
1480                                    bio_flags, bio_offset,
1481                                    __btrfs_submit_bio_start,
1482                                    __btrfs_submit_bio_done);
1483         }
1484
1485 mapit:
1486         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1487 }
1488
1489 /*
1490  * given a list of ordered sums record them in the inode.  This happens
1491  * at IO completion time based on sums calculated at bio submission time.
1492  */
1493 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1494                              struct inode *inode, u64 file_offset,
1495                              struct list_head *list)
1496 {
1497         struct btrfs_ordered_sum *sum;
1498
1499         btrfs_set_trans_block_group(trans, inode);
1500
1501         list_for_each_entry(sum, list, list) {
1502                 btrfs_csum_file_blocks(trans,
1503                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1504         }
1505         return 0;
1506 }
1507
1508 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1509                               struct extent_state **cached_state)
1510 {
1511         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1512                 WARN_ON(1);
1513         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1514                                    cached_state, GFP_NOFS);
1515 }
1516
1517 /* see btrfs_writepage_start_hook for details on why this is required */
1518 struct btrfs_writepage_fixup {
1519         struct page *page;
1520         struct btrfs_work work;
1521 };
1522
1523 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1524 {
1525         struct btrfs_writepage_fixup *fixup;
1526         struct btrfs_ordered_extent *ordered;
1527         struct extent_state *cached_state = NULL;
1528         struct page *page;
1529         struct inode *inode;
1530         u64 page_start;
1531         u64 page_end;
1532
1533         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1534         page = fixup->page;
1535 again:
1536         lock_page(page);
1537         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1538                 ClearPageChecked(page);
1539                 goto out_page;
1540         }
1541
1542         inode = page->mapping->host;
1543         page_start = page_offset(page);
1544         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1545
1546         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1547                          &cached_state, GFP_NOFS);
1548
1549         /* already ordered? We're done */
1550         if (PagePrivate2(page))
1551                 goto out;
1552
1553         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1554         if (ordered) {
1555                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1556                                      page_end, &cached_state, GFP_NOFS);
1557                 unlock_page(page);
1558                 btrfs_start_ordered_extent(inode, ordered, 1);
1559                 goto again;
1560         }
1561
1562         BUG();
1563         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1564         ClearPageChecked(page);
1565 out:
1566         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1567                              &cached_state, GFP_NOFS);
1568 out_page:
1569         unlock_page(page);
1570         page_cache_release(page);
1571         kfree(fixup);
1572 }
1573
1574 /*
1575  * There are a few paths in the higher layers of the kernel that directly
1576  * set the page dirty bit without asking the filesystem if it is a
1577  * good idea.  This causes problems because we want to make sure COW
1578  * properly happens and the data=ordered rules are followed.
1579  *
1580  * In our case any range that doesn't have the ORDERED bit set
1581  * hasn't been properly setup for IO.  We kick off an async process
1582  * to fix it up.  The async helper will wait for ordered extents, set
1583  * the delalloc bit and make it safe to write the page.
1584  */
1585 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1586 {
1587         struct inode *inode = page->mapping->host;
1588         struct btrfs_writepage_fixup *fixup;
1589         struct btrfs_root *root = BTRFS_I(inode)->root;
1590
1591         /* this page is properly in the ordered list */
1592         if (TestClearPagePrivate2(page))
1593                 return 0;
1594
1595         if (PageChecked(page))
1596                 return -EAGAIN;
1597
1598         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1599         if (!fixup)
1600                 return -EAGAIN;
1601
1602         SetPageChecked(page);
1603         page_cache_get(page);
1604         fixup->work.func = btrfs_writepage_fixup_worker;
1605         fixup->page = page;
1606         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1607         return -EAGAIN;
1608 }
1609
1610 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1611                                        struct inode *inode, u64 file_pos,
1612                                        u64 disk_bytenr, u64 disk_num_bytes,
1613                                        u64 num_bytes, u64 ram_bytes,
1614                                        u8 compression, u8 encryption,
1615                                        u16 other_encoding, int extent_type)
1616 {
1617         struct btrfs_root *root = BTRFS_I(inode)->root;
1618         struct btrfs_file_extent_item *fi;
1619         struct btrfs_path *path;
1620         struct extent_buffer *leaf;
1621         struct btrfs_key ins;
1622         u64 hint;
1623         int ret;
1624
1625         path = btrfs_alloc_path();
1626         BUG_ON(!path);
1627
1628         path->leave_spinning = 1;
1629
1630         /*
1631          * we may be replacing one extent in the tree with another.
1632          * The new extent is pinned in the extent map, and we don't want
1633          * to drop it from the cache until it is completely in the btree.
1634          *
1635          * So, tell btrfs_drop_extents to leave this extent in the cache.
1636          * the caller is expected to unpin it and allow it to be merged
1637          * with the others.
1638          */
1639         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1640                                  &hint, 0);
1641         BUG_ON(ret);
1642
1643         ins.objectid = inode->i_ino;
1644         ins.offset = file_pos;
1645         ins.type = BTRFS_EXTENT_DATA_KEY;
1646         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1647         BUG_ON(ret);
1648         leaf = path->nodes[0];
1649         fi = btrfs_item_ptr(leaf, path->slots[0],
1650                             struct btrfs_file_extent_item);
1651         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1652         btrfs_set_file_extent_type(leaf, fi, extent_type);
1653         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1654         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1655         btrfs_set_file_extent_offset(leaf, fi, 0);
1656         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1657         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1658         btrfs_set_file_extent_compression(leaf, fi, compression);
1659         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1660         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1661
1662         btrfs_unlock_up_safe(path, 1);
1663         btrfs_set_lock_blocking(leaf);
1664
1665         btrfs_mark_buffer_dirty(leaf);
1666
1667         inode_add_bytes(inode, num_bytes);
1668
1669         ins.objectid = disk_bytenr;
1670         ins.offset = disk_num_bytes;
1671         ins.type = BTRFS_EXTENT_ITEM_KEY;
1672         ret = btrfs_alloc_reserved_file_extent(trans, root,
1673                                         root->root_key.objectid,
1674                                         inode->i_ino, file_pos, &ins);
1675         BUG_ON(ret);
1676         btrfs_free_path(path);
1677
1678         return 0;
1679 }
1680
1681 /*
1682  * helper function for btrfs_finish_ordered_io, this
1683  * just reads in some of the csum leaves to prime them into ram
1684  * before we start the transaction.  It limits the amount of btree
1685  * reads required while inside the transaction.
1686  */
1687 /* as ordered data IO finishes, this gets called so we can finish
1688  * an ordered extent if the range of bytes in the file it covers are
1689  * fully written.
1690  */
1691 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1692 {
1693         struct btrfs_root *root = BTRFS_I(inode)->root;
1694         struct btrfs_trans_handle *trans = NULL;
1695         struct btrfs_ordered_extent *ordered_extent = NULL;
1696         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1697         struct extent_state *cached_state = NULL;
1698         int compress_type = 0;
1699         int ret;
1700         bool nolock = false;
1701
1702         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1703                                              end - start + 1);
1704         if (!ret)
1705                 return 0;
1706         BUG_ON(!ordered_extent);
1707
1708         nolock = (root == root->fs_info->tree_root);
1709
1710         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1711                 BUG_ON(!list_empty(&ordered_extent->list));
1712                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1713                 if (!ret) {
1714                         if (nolock)
1715                                 trans = btrfs_join_transaction_nolock(root, 1);
1716                         else
1717                                 trans = btrfs_join_transaction(root, 1);
1718                         BUG_ON(IS_ERR(trans));
1719                         btrfs_set_trans_block_group(trans, inode);
1720                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1721                         ret = btrfs_update_inode(trans, root, inode);
1722                         BUG_ON(ret);
1723                 }
1724                 goto out;
1725         }
1726
1727         lock_extent_bits(io_tree, ordered_extent->file_offset,
1728                          ordered_extent->file_offset + ordered_extent->len - 1,
1729                          0, &cached_state, GFP_NOFS);
1730
1731         if (nolock)
1732                 trans = btrfs_join_transaction_nolock(root, 1);
1733         else
1734                 trans = btrfs_join_transaction(root, 1);
1735         BUG_ON(IS_ERR(trans));
1736         btrfs_set_trans_block_group(trans, inode);
1737         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1738
1739         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1740                 compress_type = ordered_extent->compress_type;
1741         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1742                 BUG_ON(compress_type);
1743                 ret = btrfs_mark_extent_written(trans, inode,
1744                                                 ordered_extent->file_offset,
1745                                                 ordered_extent->file_offset +
1746                                                 ordered_extent->len);
1747                 BUG_ON(ret);
1748         } else {
1749                 BUG_ON(root == root->fs_info->tree_root);
1750                 ret = insert_reserved_file_extent(trans, inode,
1751                                                 ordered_extent->file_offset,
1752                                                 ordered_extent->start,
1753                                                 ordered_extent->disk_len,
1754                                                 ordered_extent->len,
1755                                                 ordered_extent->len,
1756                                                 compress_type, 0, 0,
1757                                                 BTRFS_FILE_EXTENT_REG);
1758                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1759                                    ordered_extent->file_offset,
1760                                    ordered_extent->len);
1761                 BUG_ON(ret);
1762         }
1763         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1764                              ordered_extent->file_offset +
1765                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1766
1767         add_pending_csums(trans, inode, ordered_extent->file_offset,
1768                           &ordered_extent->list);
1769
1770         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1771         ret = btrfs_update_inode(trans, root, inode);
1772         BUG_ON(ret);
1773 out:
1774         if (nolock) {
1775                 if (trans)
1776                         btrfs_end_transaction_nolock(trans, root);
1777         } else {
1778                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1779                 if (trans)
1780                         btrfs_end_transaction(trans, root);
1781         }
1782
1783         /* once for us */
1784         btrfs_put_ordered_extent(ordered_extent);
1785         /* once for the tree */
1786         btrfs_put_ordered_extent(ordered_extent);
1787
1788         return 0;
1789 }
1790
1791 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1792                                 struct extent_state *state, int uptodate)
1793 {
1794         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1795
1796         ClearPagePrivate2(page);
1797         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1798 }
1799
1800 /*
1801  * When IO fails, either with EIO or csum verification fails, we
1802  * try other mirrors that might have a good copy of the data.  This
1803  * io_failure_record is used to record state as we go through all the
1804  * mirrors.  If another mirror has good data, the page is set up to date
1805  * and things continue.  If a good mirror can't be found, the original
1806  * bio end_io callback is called to indicate things have failed.
1807  */
1808 struct io_failure_record {
1809         struct page *page;
1810         u64 start;
1811         u64 len;
1812         u64 logical;
1813         unsigned long bio_flags;
1814         int last_mirror;
1815 };
1816
1817 static int btrfs_io_failed_hook(struct bio *failed_bio,
1818                          struct page *page, u64 start, u64 end,
1819                          struct extent_state *state)
1820 {
1821         struct io_failure_record *failrec = NULL;
1822         u64 private;
1823         struct extent_map *em;
1824         struct inode *inode = page->mapping->host;
1825         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1826         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1827         struct bio *bio;
1828         int num_copies;
1829         int ret;
1830         int rw;
1831         u64 logical;
1832
1833         ret = get_state_private(failure_tree, start, &private);
1834         if (ret) {
1835                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1836                 if (!failrec)
1837                         return -ENOMEM;
1838                 failrec->start = start;
1839                 failrec->len = end - start + 1;
1840                 failrec->last_mirror = 0;
1841                 failrec->bio_flags = 0;
1842
1843                 read_lock(&em_tree->lock);
1844                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1845                 if (em->start > start || em->start + em->len < start) {
1846                         free_extent_map(em);
1847                         em = NULL;
1848                 }
1849                 read_unlock(&em_tree->lock);
1850
1851                 if (!em || IS_ERR(em)) {
1852                         kfree(failrec);
1853                         return -EIO;
1854                 }
1855                 logical = start - em->start;
1856                 logical = em->block_start + logical;
1857                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1858                         logical = em->block_start;
1859                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1860                         extent_set_compress_type(&failrec->bio_flags,
1861                                                  em->compress_type);
1862                 }
1863                 failrec->logical = logical;
1864                 free_extent_map(em);
1865                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1866                                 EXTENT_DIRTY, GFP_NOFS);
1867                 set_state_private(failure_tree, start,
1868                                  (u64)(unsigned long)failrec);
1869         } else {
1870                 failrec = (struct io_failure_record *)(unsigned long)private;
1871         }
1872         num_copies = btrfs_num_copies(
1873                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1874                               failrec->logical, failrec->len);
1875         failrec->last_mirror++;
1876         if (!state) {
1877                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1878                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1879                                                     failrec->start,
1880                                                     EXTENT_LOCKED);
1881                 if (state && state->start != failrec->start)
1882                         state = NULL;
1883                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1884         }
1885         if (!state || failrec->last_mirror > num_copies) {
1886                 set_state_private(failure_tree, failrec->start, 0);
1887                 clear_extent_bits(failure_tree, failrec->start,
1888                                   failrec->start + failrec->len - 1,
1889                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1890                 kfree(failrec);
1891                 return -EIO;
1892         }
1893         bio = bio_alloc(GFP_NOFS, 1);
1894         bio->bi_private = state;
1895         bio->bi_end_io = failed_bio->bi_end_io;
1896         bio->bi_sector = failrec->logical >> 9;
1897         bio->bi_bdev = failed_bio->bi_bdev;
1898         bio->bi_size = 0;
1899
1900         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1901         if (failed_bio->bi_rw & REQ_WRITE)
1902                 rw = WRITE;
1903         else
1904                 rw = READ;
1905
1906         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1907                                                       failrec->last_mirror,
1908                                                       failrec->bio_flags, 0);
1909         return 0;
1910 }
1911
1912 /*
1913  * each time an IO finishes, we do a fast check in the IO failure tree
1914  * to see if we need to process or clean up an io_failure_record
1915  */
1916 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1917 {
1918         u64 private;
1919         u64 private_failure;
1920         struct io_failure_record *failure;
1921         int ret;
1922
1923         private = 0;
1924         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1925                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1926                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1927                                         start, &private_failure);
1928                 if (ret == 0) {
1929                         failure = (struct io_failure_record *)(unsigned long)
1930                                    private_failure;
1931                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1932                                           failure->start, 0);
1933                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1934                                           failure->start,
1935                                           failure->start + failure->len - 1,
1936                                           EXTENT_DIRTY | EXTENT_LOCKED,
1937                                           GFP_NOFS);
1938                         kfree(failure);
1939                 }
1940         }
1941         return 0;
1942 }
1943
1944 /*
1945  * when reads are done, we need to check csums to verify the data is correct
1946  * if there's a match, we allow the bio to finish.  If not, we go through
1947  * the io_failure_record routines to find good copies
1948  */
1949 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1950                                struct extent_state *state)
1951 {
1952         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1953         struct inode *inode = page->mapping->host;
1954         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1955         char *kaddr;
1956         u64 private = ~(u32)0;
1957         int ret;
1958         struct btrfs_root *root = BTRFS_I(inode)->root;
1959         u32 csum = ~(u32)0;
1960
1961         if (PageChecked(page)) {
1962                 ClearPageChecked(page);
1963                 goto good;
1964         }
1965
1966         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1967                 return 0;
1968
1969         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1970             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1971                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1972                                   GFP_NOFS);
1973                 return 0;
1974         }
1975
1976         if (state && state->start == start) {
1977                 private = state->private;
1978                 ret = 0;
1979         } else {
1980                 ret = get_state_private(io_tree, start, &private);
1981         }
1982         kaddr = kmap_atomic(page, KM_USER0);
1983         if (ret)
1984                 goto zeroit;
1985
1986         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1987         btrfs_csum_final(csum, (char *)&csum);
1988         if (csum != private)
1989                 goto zeroit;
1990
1991         kunmap_atomic(kaddr, KM_USER0);
1992 good:
1993         /* if the io failure tree for this inode is non-empty,
1994          * check to see if we've recovered from a failed IO
1995          */
1996         btrfs_clean_io_failures(inode, start);
1997         return 0;
1998
1999 zeroit:
2000         if (printk_ratelimit()) {
2001                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2002                        "private %llu\n", page->mapping->host->i_ino,
2003                        (unsigned long long)start, csum,
2004                        (unsigned long long)private);
2005         }
2006         memset(kaddr + offset, 1, end - start + 1);
2007         flush_dcache_page(page);
2008         kunmap_atomic(kaddr, KM_USER0);
2009         if (private == 0)
2010                 return 0;
2011         return -EIO;
2012 }
2013
2014 struct delayed_iput {
2015         struct list_head list;
2016         struct inode *inode;
2017 };
2018
2019 void btrfs_add_delayed_iput(struct inode *inode)
2020 {
2021         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2022         struct delayed_iput *delayed;
2023
2024         if (atomic_add_unless(&inode->i_count, -1, 1))
2025                 return;
2026
2027         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2028         delayed->inode = inode;
2029
2030         spin_lock(&fs_info->delayed_iput_lock);
2031         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2032         spin_unlock(&fs_info->delayed_iput_lock);
2033 }
2034
2035 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2036 {
2037         LIST_HEAD(list);
2038         struct btrfs_fs_info *fs_info = root->fs_info;
2039         struct delayed_iput *delayed;
2040         int empty;
2041
2042         spin_lock(&fs_info->delayed_iput_lock);
2043         empty = list_empty(&fs_info->delayed_iputs);
2044         spin_unlock(&fs_info->delayed_iput_lock);
2045         if (empty)
2046                 return;
2047
2048         down_read(&root->fs_info->cleanup_work_sem);
2049         spin_lock(&fs_info->delayed_iput_lock);
2050         list_splice_init(&fs_info->delayed_iputs, &list);
2051         spin_unlock(&fs_info->delayed_iput_lock);
2052
2053         while (!list_empty(&list)) {
2054                 delayed = list_entry(list.next, struct delayed_iput, list);
2055                 list_del(&delayed->list);
2056                 iput(delayed->inode);
2057                 kfree(delayed);
2058         }
2059         up_read(&root->fs_info->cleanup_work_sem);
2060 }
2061
2062 /*
2063  * calculate extra metadata reservation when snapshotting a subvolume
2064  * contains orphan files.
2065  */
2066 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2067                                 struct btrfs_pending_snapshot *pending,
2068                                 u64 *bytes_to_reserve)
2069 {
2070         struct btrfs_root *root;
2071         struct btrfs_block_rsv *block_rsv;
2072         u64 num_bytes;
2073         int index;
2074
2075         root = pending->root;
2076         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2077                 return;
2078
2079         block_rsv = root->orphan_block_rsv;
2080
2081         /* orphan block reservation for the snapshot */
2082         num_bytes = block_rsv->size;
2083
2084         /*
2085          * after the snapshot is created, COWing tree blocks may use more
2086          * space than it frees. So we should make sure there is enough
2087          * reserved space.
2088          */
2089         index = trans->transid & 0x1;
2090         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2091                 num_bytes += block_rsv->size -
2092                              (block_rsv->reserved + block_rsv->freed[index]);
2093         }
2094
2095         *bytes_to_reserve += num_bytes;
2096 }
2097
2098 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2099                                 struct btrfs_pending_snapshot *pending)
2100 {
2101         struct btrfs_root *root = pending->root;
2102         struct btrfs_root *snap = pending->snap;
2103         struct btrfs_block_rsv *block_rsv;
2104         u64 num_bytes;
2105         int index;
2106         int ret;
2107
2108         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2109                 return;
2110
2111         /* refill source subvolume's orphan block reservation */
2112         block_rsv = root->orphan_block_rsv;
2113         index = trans->transid & 0x1;
2114         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2115                 num_bytes = block_rsv->size -
2116                             (block_rsv->reserved + block_rsv->freed[index]);
2117                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2118                                               root->orphan_block_rsv,
2119                                               num_bytes);
2120                 BUG_ON(ret);
2121         }
2122
2123         /* setup orphan block reservation for the snapshot */
2124         block_rsv = btrfs_alloc_block_rsv(snap);
2125         BUG_ON(!block_rsv);
2126
2127         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2128         snap->orphan_block_rsv = block_rsv;
2129
2130         num_bytes = root->orphan_block_rsv->size;
2131         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2132                                       block_rsv, num_bytes);
2133         BUG_ON(ret);
2134
2135 #if 0
2136         /* insert orphan item for the snapshot */
2137         WARN_ON(!root->orphan_item_inserted);
2138         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2139                                        snap->root_key.objectid);
2140         BUG_ON(ret);
2141         snap->orphan_item_inserted = 1;
2142 #endif
2143 }
2144
2145 enum btrfs_orphan_cleanup_state {
2146         ORPHAN_CLEANUP_STARTED  = 1,
2147         ORPHAN_CLEANUP_DONE     = 2,
2148 };
2149
2150 /*
2151  * This is called in transaction commmit time. If there are no orphan
2152  * files in the subvolume, it removes orphan item and frees block_rsv
2153  * structure.
2154  */
2155 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2156                               struct btrfs_root *root)
2157 {
2158         int ret;
2159
2160         if (!list_empty(&root->orphan_list) ||
2161             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2162                 return;
2163
2164         if (root->orphan_item_inserted &&
2165             btrfs_root_refs(&root->root_item) > 0) {
2166                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2167                                             root->root_key.objectid);
2168                 BUG_ON(ret);
2169                 root->orphan_item_inserted = 0;
2170         }
2171
2172         if (root->orphan_block_rsv) {
2173                 WARN_ON(root->orphan_block_rsv->size > 0);
2174                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2175                 root->orphan_block_rsv = NULL;
2176         }
2177 }
2178
2179 /*
2180  * This creates an orphan entry for the given inode in case something goes
2181  * wrong in the middle of an unlink/truncate.
2182  *
2183  * NOTE: caller of this function should reserve 5 units of metadata for
2184  *       this function.
2185  */
2186 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2187 {
2188         struct btrfs_root *root = BTRFS_I(inode)->root;
2189         struct btrfs_block_rsv *block_rsv = NULL;
2190         int reserve = 0;
2191         int insert = 0;
2192         int ret;
2193
2194         if (!root->orphan_block_rsv) {
2195                 block_rsv = btrfs_alloc_block_rsv(root);
2196                 BUG_ON(!block_rsv);
2197         }
2198
2199         spin_lock(&root->orphan_lock);
2200         if (!root->orphan_block_rsv) {
2201                 root->orphan_block_rsv = block_rsv;
2202         } else if (block_rsv) {
2203                 btrfs_free_block_rsv(root, block_rsv);
2204                 block_rsv = NULL;
2205         }
2206
2207         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2208                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2209 #if 0
2210                 /*
2211                  * For proper ENOSPC handling, we should do orphan
2212                  * cleanup when mounting. But this introduces backward
2213                  * compatibility issue.
2214                  */
2215                 if (!xchg(&root->orphan_item_inserted, 1))
2216                         insert = 2;
2217                 else
2218                         insert = 1;
2219 #endif
2220                 insert = 1;
2221         } else {
2222                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2223         }
2224
2225         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2226                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2227                 reserve = 1;
2228         }
2229         spin_unlock(&root->orphan_lock);
2230
2231         if (block_rsv)
2232                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2233
2234         /* grab metadata reservation from transaction handle */
2235         if (reserve) {
2236                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2237                 BUG_ON(ret);
2238         }
2239
2240         /* insert an orphan item to track this unlinked/truncated file */
2241         if (insert >= 1) {
2242                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2243                 BUG_ON(ret);
2244         }
2245
2246         /* insert an orphan item to track subvolume contains orphan files */
2247         if (insert >= 2) {
2248                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2249                                                root->root_key.objectid);
2250                 BUG_ON(ret);
2251         }
2252         return 0;
2253 }
2254
2255 /*
2256  * We have done the truncate/delete so we can go ahead and remove the orphan
2257  * item for this particular inode.
2258  */
2259 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2260 {
2261         struct btrfs_root *root = BTRFS_I(inode)->root;
2262         int delete_item = 0;
2263         int release_rsv = 0;
2264         int ret = 0;
2265
2266         spin_lock(&root->orphan_lock);
2267         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2268                 list_del_init(&BTRFS_I(inode)->i_orphan);
2269                 delete_item = 1;
2270         }
2271
2272         if (BTRFS_I(inode)->orphan_meta_reserved) {
2273                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2274                 release_rsv = 1;
2275         }
2276         spin_unlock(&root->orphan_lock);
2277
2278         if (trans && delete_item) {
2279                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2280                 BUG_ON(ret);
2281         }
2282
2283         if (release_rsv)
2284                 btrfs_orphan_release_metadata(inode);
2285
2286         return 0;
2287 }
2288
2289 /*
2290  * this cleans up any orphans that may be left on the list from the last use
2291  * of this root.
2292  */
2293 int btrfs_orphan_cleanup(struct btrfs_root *root)
2294 {
2295         struct btrfs_path *path;
2296         struct extent_buffer *leaf;
2297         struct btrfs_key key, found_key;
2298         struct btrfs_trans_handle *trans;
2299         struct inode *inode;
2300         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2301
2302         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2303                 return 0;
2304
2305         path = btrfs_alloc_path();
2306         if (!path) {
2307                 ret = -ENOMEM;
2308                 goto out;
2309         }
2310         path->reada = -1;
2311
2312         key.objectid = BTRFS_ORPHAN_OBJECTID;
2313         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2314         key.offset = (u64)-1;
2315
2316         while (1) {
2317                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2318                 if (ret < 0)
2319                         goto out;
2320
2321                 /*
2322                  * if ret == 0 means we found what we were searching for, which
2323                  * is weird, but possible, so only screw with path if we didnt
2324                  * find the key and see if we have stuff that matches
2325                  */
2326                 if (ret > 0) {
2327                         ret = 0;
2328                         if (path->slots[0] == 0)
2329                                 break;
2330                         path->slots[0]--;
2331                 }
2332
2333                 /* pull out the item */
2334                 leaf = path->nodes[0];
2335                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2336
2337                 /* make sure the item matches what we want */
2338                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2339                         break;
2340                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2341                         break;
2342
2343                 /* release the path since we're done with it */
2344                 btrfs_release_path(root, path);
2345
2346                 /*
2347                  * this is where we are basically btrfs_lookup, without the
2348                  * crossing root thing.  we store the inode number in the
2349                  * offset of the orphan item.
2350                  */
2351                 found_key.objectid = found_key.offset;
2352                 found_key.type = BTRFS_INODE_ITEM_KEY;
2353                 found_key.offset = 0;
2354                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2355                 if (IS_ERR(inode)) {
2356                         ret = PTR_ERR(inode);
2357                         goto out;
2358                 }
2359
2360                 /*
2361                  * add this inode to the orphan list so btrfs_orphan_del does
2362                  * the proper thing when we hit it
2363                  */
2364                 spin_lock(&root->orphan_lock);
2365                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2366                 spin_unlock(&root->orphan_lock);
2367
2368                 /*
2369                  * if this is a bad inode, means we actually succeeded in
2370                  * removing the inode, but not the orphan record, which means
2371                  * we need to manually delete the orphan since iput will just
2372                  * do a destroy_inode
2373                  */
2374                 if (is_bad_inode(inode)) {
2375                         trans = btrfs_start_transaction(root, 0);
2376                         if (IS_ERR(trans)) {
2377                                 ret = PTR_ERR(trans);
2378                                 goto out;
2379                         }
2380                         btrfs_orphan_del(trans, inode);
2381                         btrfs_end_transaction(trans, root);
2382                         iput(inode);
2383                         continue;
2384                 }
2385
2386                 /* if we have links, this was a truncate, lets do that */
2387                 if (inode->i_nlink) {
2388                         if (!S_ISREG(inode->i_mode)) {
2389                                 WARN_ON(1);
2390                                 iput(inode);
2391                                 continue;
2392                         }
2393                         nr_truncate++;
2394                         ret = btrfs_truncate(inode);
2395                 } else {
2396                         nr_unlink++;
2397                 }
2398
2399                 /* this will do delete_inode and everything for us */
2400                 iput(inode);
2401                 if (ret)
2402                         goto out;
2403         }
2404         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2405
2406         if (root->orphan_block_rsv)
2407                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2408                                         (u64)-1);
2409
2410         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2411                 trans = btrfs_join_transaction(root, 1);
2412                 if (!IS_ERR(trans))
2413                         btrfs_end_transaction(trans, root);
2414         }
2415
2416         if (nr_unlink)
2417                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2418         if (nr_truncate)
2419                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2420
2421 out:
2422         if (ret)
2423                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2424         btrfs_free_path(path);
2425         return ret;
2426 }
2427
2428 /*
2429  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2430  * don't find any xattrs, we know there can't be any acls.
2431  *
2432  * slot is the slot the inode is in, objectid is the objectid of the inode
2433  */
2434 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2435                                           int slot, u64 objectid)
2436 {
2437         u32 nritems = btrfs_header_nritems(leaf);
2438         struct btrfs_key found_key;
2439         int scanned = 0;
2440
2441         slot++;
2442         while (slot < nritems) {
2443                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2444
2445                 /* we found a different objectid, there must not be acls */
2446                 if (found_key.objectid != objectid)
2447                         return 0;
2448
2449                 /* we found an xattr, assume we've got an acl */
2450                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2451                         return 1;
2452
2453                 /*
2454                  * we found a key greater than an xattr key, there can't
2455                  * be any acls later on
2456                  */
2457                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2458                         return 0;
2459
2460                 slot++;
2461                 scanned++;
2462
2463                 /*
2464                  * it goes inode, inode backrefs, xattrs, extents,
2465                  * so if there are a ton of hard links to an inode there can
2466                  * be a lot of backrefs.  Don't waste time searching too hard,
2467                  * this is just an optimization
2468                  */
2469                 if (scanned >= 8)
2470                         break;
2471         }
2472         /* we hit the end of the leaf before we found an xattr or
2473          * something larger than an xattr.  We have to assume the inode
2474          * has acls
2475          */
2476         return 1;
2477 }
2478
2479 /*
2480  * read an inode from the btree into the in-memory inode
2481  */
2482 static void btrfs_read_locked_inode(struct inode *inode)
2483 {
2484         struct btrfs_path *path;
2485         struct extent_buffer *leaf;
2486         struct btrfs_inode_item *inode_item;
2487         struct btrfs_timespec *tspec;
2488         struct btrfs_root *root = BTRFS_I(inode)->root;
2489         struct btrfs_key location;
2490         int maybe_acls;
2491         u64 alloc_group_block;
2492         u32 rdev;
2493         int ret;
2494
2495         path = btrfs_alloc_path();
2496         BUG_ON(!path);
2497         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2498
2499         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2500         if (ret)
2501                 goto make_bad;
2502
2503         leaf = path->nodes[0];
2504         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2505                                     struct btrfs_inode_item);
2506
2507         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2508         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2509         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2510         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2511         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2512
2513         tspec = btrfs_inode_atime(inode_item);
2514         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2515         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2516
2517         tspec = btrfs_inode_mtime(inode_item);
2518         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2519         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2520
2521         tspec = btrfs_inode_ctime(inode_item);
2522         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2523         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2524
2525         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2526         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2527         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2528         inode->i_generation = BTRFS_I(inode)->generation;
2529         inode->i_rdev = 0;
2530         rdev = btrfs_inode_rdev(leaf, inode_item);
2531
2532         BTRFS_I(inode)->index_cnt = (u64)-1;
2533         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2534
2535         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2536
2537         /*
2538          * try to precache a NULL acl entry for files that don't have
2539          * any xattrs or acls
2540          */
2541         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2542         if (!maybe_acls)
2543                 cache_no_acl(inode);
2544
2545         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2546                                                 alloc_group_block, 0);
2547         btrfs_free_path(path);
2548         inode_item = NULL;
2549
2550         switch (inode->i_mode & S_IFMT) {
2551         case S_IFREG:
2552                 inode->i_mapping->a_ops = &btrfs_aops;
2553                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2554                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2555                 inode->i_fop = &btrfs_file_operations;
2556                 inode->i_op = &btrfs_file_inode_operations;
2557                 break;
2558         case S_IFDIR:
2559                 inode->i_fop = &btrfs_dir_file_operations;
2560                 if (root == root->fs_info->tree_root)
2561                         inode->i_op = &btrfs_dir_ro_inode_operations;
2562                 else
2563                         inode->i_op = &btrfs_dir_inode_operations;
2564                 break;
2565         case S_IFLNK:
2566                 inode->i_op = &btrfs_symlink_inode_operations;
2567                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2568                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2569                 break;
2570         default:
2571                 inode->i_op = &btrfs_special_inode_operations;
2572                 init_special_inode(inode, inode->i_mode, rdev);
2573                 break;
2574         }
2575
2576         btrfs_update_iflags(inode);
2577         return;
2578
2579 make_bad:
2580         btrfs_free_path(path);
2581         make_bad_inode(inode);
2582 }
2583
2584 /*
2585  * given a leaf and an inode, copy the inode fields into the leaf
2586  */
2587 static void fill_inode_item(struct btrfs_trans_handle *trans,
2588                             struct extent_buffer *leaf,
2589                             struct btrfs_inode_item *item,
2590                             struct inode *inode)
2591 {
2592         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2593         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2594         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2595         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2596         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2597
2598         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2599                                inode->i_atime.tv_sec);
2600         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2601                                 inode->i_atime.tv_nsec);
2602
2603         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2604                                inode->i_mtime.tv_sec);
2605         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2606                                 inode->i_mtime.tv_nsec);
2607
2608         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2609                                inode->i_ctime.tv_sec);
2610         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2611                                 inode->i_ctime.tv_nsec);
2612
2613         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2614         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2615         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2616         btrfs_set_inode_transid(leaf, item, trans->transid);
2617         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2618         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2619         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2620 }
2621
2622 /*
2623  * copy everything in the in-memory inode into the btree.
2624  */
2625 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2626                                 struct btrfs_root *root, struct inode *inode)
2627 {
2628         struct btrfs_inode_item *inode_item;
2629         struct btrfs_path *path;
2630         struct extent_buffer *leaf;
2631         int ret;
2632
2633         path = btrfs_alloc_path();
2634         BUG_ON(!path);
2635         path->leave_spinning = 1;
2636         ret = btrfs_lookup_inode(trans, root, path,
2637                                  &BTRFS_I(inode)->location, 1);
2638         if (ret) {
2639                 if (ret > 0)
2640                         ret = -ENOENT;
2641                 goto failed;
2642         }
2643
2644         btrfs_unlock_up_safe(path, 1);
2645         leaf = path->nodes[0];
2646         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2647                                   struct btrfs_inode_item);
2648
2649         fill_inode_item(trans, leaf, inode_item, inode);
2650         btrfs_mark_buffer_dirty(leaf);
2651         btrfs_set_inode_last_trans(trans, inode);
2652         ret = 0;
2653 failed:
2654         btrfs_free_path(path);
2655         return ret;
2656 }
2657
2658
2659 /*
2660  * unlink helper that gets used here in inode.c and in the tree logging
2661  * recovery code.  It remove a link in a directory with a given name, and
2662  * also drops the back refs in the inode to the directory
2663  */
2664 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2665                        struct btrfs_root *root,
2666                        struct inode *dir, struct inode *inode,
2667                        const char *name, int name_len)
2668 {
2669         struct btrfs_path *path;
2670         int ret = 0;
2671         struct extent_buffer *leaf;
2672         struct btrfs_dir_item *di;
2673         struct btrfs_key key;
2674         u64 index;
2675
2676         path = btrfs_alloc_path();
2677         if (!path) {
2678                 ret = -ENOMEM;
2679                 goto out;
2680         }
2681
2682         path->leave_spinning = 1;
2683         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2684                                     name, name_len, -1);
2685         if (IS_ERR(di)) {
2686                 ret = PTR_ERR(di);
2687                 goto err;
2688         }
2689         if (!di) {
2690                 ret = -ENOENT;
2691                 goto err;
2692         }
2693         leaf = path->nodes[0];
2694         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2695         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2696         if (ret)
2697                 goto err;
2698         btrfs_release_path(root, path);
2699
2700         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2701                                   inode->i_ino,
2702                                   dir->i_ino, &index);
2703         if (ret) {
2704                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2705                        "inode %lu parent %lu\n", name_len, name,
2706                        inode->i_ino, dir->i_ino);
2707                 goto err;
2708         }
2709
2710         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2711                                          index, name, name_len, -1);
2712         if (IS_ERR(di)) {
2713                 ret = PTR_ERR(di);
2714                 goto err;
2715         }
2716         if (!di) {
2717                 ret = -ENOENT;
2718                 goto err;
2719         }
2720         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2721         btrfs_release_path(root, path);
2722
2723         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2724                                          inode, dir->i_ino);
2725         BUG_ON(ret != 0 && ret != -ENOENT);
2726
2727         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2728                                            dir, index);
2729         if (ret == -ENOENT)
2730                 ret = 0;
2731 err:
2732         btrfs_free_path(path);
2733         if (ret)
2734                 goto out;
2735
2736         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2737         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2738         btrfs_update_inode(trans, root, dir);
2739         btrfs_drop_nlink(inode);
2740         ret = btrfs_update_inode(trans, root, inode);
2741 out:
2742         return ret;
2743 }
2744
2745 /* helper to check if there is any shared block in the path */
2746 static int check_path_shared(struct btrfs_root *root,
2747                              struct btrfs_path *path)
2748 {
2749         struct extent_buffer *eb;
2750         int level;
2751         u64 refs = 1;
2752
2753         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2754                 int ret;
2755
2756                 if (!path->nodes[level])
2757                         break;
2758                 eb = path->nodes[level];
2759                 if (!btrfs_block_can_be_shared(root, eb))
2760                         continue;
2761                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2762                                                &refs, NULL);
2763                 if (refs > 1)
2764                         return 1;
2765         }
2766         return 0;
2767 }
2768
2769 /*
2770  * helper to start transaction for unlink and rmdir.
2771  *
2772  * unlink and rmdir are special in btrfs, they do not always free space.
2773  * so in enospc case, we should make sure they will free space before
2774  * allowing them to use the global metadata reservation.
2775  */
2776 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2777                                                        struct dentry *dentry)
2778 {
2779         struct btrfs_trans_handle *trans;
2780         struct btrfs_root *root = BTRFS_I(dir)->root;
2781         struct btrfs_path *path;
2782         struct btrfs_inode_ref *ref;
2783         struct btrfs_dir_item *di;
2784         struct inode *inode = dentry->d_inode;
2785         u64 index;
2786         int check_link = 1;
2787         int err = -ENOSPC;
2788         int ret;
2789
2790         trans = btrfs_start_transaction(root, 10);
2791         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2792                 return trans;
2793
2794         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2795                 return ERR_PTR(-ENOSPC);
2796
2797         /* check if there is someone else holds reference */
2798         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2799                 return ERR_PTR(-ENOSPC);
2800
2801         if (atomic_read(&inode->i_count) > 2)
2802                 return ERR_PTR(-ENOSPC);
2803
2804         if (xchg(&root->fs_info->enospc_unlink, 1))
2805                 return ERR_PTR(-ENOSPC);
2806
2807         path = btrfs_alloc_path();
2808         if (!path) {
2809                 root->fs_info->enospc_unlink = 0;
2810                 return ERR_PTR(-ENOMEM);
2811         }
2812
2813         trans = btrfs_start_transaction(root, 0);
2814         if (IS_ERR(trans)) {
2815                 btrfs_free_path(path);
2816                 root->fs_info->enospc_unlink = 0;
2817                 return trans;
2818         }
2819
2820         path->skip_locking = 1;
2821         path->search_commit_root = 1;
2822
2823         ret = btrfs_lookup_inode(trans, root, path,
2824                                 &BTRFS_I(dir)->location, 0);
2825         if (ret < 0) {
2826                 err = ret;
2827                 goto out;
2828         }
2829         if (ret == 0) {
2830                 if (check_path_shared(root, path))
2831                         goto out;
2832         } else {
2833                 check_link = 0;
2834         }
2835         btrfs_release_path(root, path);
2836
2837         ret = btrfs_lookup_inode(trans, root, path,
2838                                 &BTRFS_I(inode)->location, 0);
2839         if (ret < 0) {
2840                 err = ret;
2841                 goto out;
2842         }
2843         if (ret == 0) {
2844                 if (check_path_shared(root, path))
2845                         goto out;
2846         } else {
2847                 check_link = 0;
2848         }
2849         btrfs_release_path(root, path);
2850
2851         if (ret == 0 && S_ISREG(inode->i_mode)) {
2852                 ret = btrfs_lookup_file_extent(trans, root, path,
2853                                                inode->i_ino, (u64)-1, 0);
2854                 if (ret < 0) {
2855                         err = ret;
2856                         goto out;
2857                 }
2858                 BUG_ON(ret == 0);
2859                 if (check_path_shared(root, path))
2860                         goto out;
2861                 btrfs_release_path(root, path);
2862         }
2863
2864         if (!check_link) {
2865                 err = 0;
2866                 goto out;
2867         }
2868
2869         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2870                                 dentry->d_name.name, dentry->d_name.len, 0);
2871         if (IS_ERR(di)) {
2872                 err = PTR_ERR(di);
2873                 goto out;
2874         }
2875         if (di) {
2876                 if (check_path_shared(root, path))
2877                         goto out;
2878         } else {
2879                 err = 0;
2880                 goto out;
2881         }
2882         btrfs_release_path(root, path);
2883
2884         ref = btrfs_lookup_inode_ref(trans, root, path,
2885                                 dentry->d_name.name, dentry->d_name.len,
2886                                 inode->i_ino, dir->i_ino, 0);
2887         if (IS_ERR(ref)) {
2888                 err = PTR_ERR(ref);
2889                 goto out;
2890         }
2891         BUG_ON(!ref);
2892         if (check_path_shared(root, path))
2893                 goto out;
2894         index = btrfs_inode_ref_index(path->nodes[0], ref);
2895         btrfs_release_path(root, path);
2896
2897         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2898                                 dentry->d_name.name, dentry->d_name.len, 0);
2899         if (IS_ERR(di)) {
2900                 err = PTR_ERR(di);
2901                 goto out;
2902         }
2903         BUG_ON(ret == -ENOENT);
2904         if (check_path_shared(root, path))
2905                 goto out;
2906
2907         err = 0;
2908 out:
2909         btrfs_free_path(path);
2910         if (err) {
2911                 btrfs_end_transaction(trans, root);
2912                 root->fs_info->enospc_unlink = 0;
2913                 return ERR_PTR(err);
2914         }
2915
2916         trans->block_rsv = &root->fs_info->global_block_rsv;
2917         return trans;
2918 }
2919
2920 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2921                                struct btrfs_root *root)
2922 {
2923         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2924                 BUG_ON(!root->fs_info->enospc_unlink);
2925                 root->fs_info->enospc_unlink = 0;
2926         }
2927         btrfs_end_transaction_throttle(trans, root);
2928 }
2929
2930 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2931 {
2932         struct btrfs_root *root = BTRFS_I(dir)->root;
2933         struct btrfs_trans_handle *trans;
2934         struct inode *inode = dentry->d_inode;
2935         int ret;
2936         unsigned long nr = 0;
2937
2938         trans = __unlink_start_trans(dir, dentry);
2939         if (IS_ERR(trans))
2940                 return PTR_ERR(trans);
2941
2942         btrfs_set_trans_block_group(trans, dir);
2943
2944         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2945
2946         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2947                                  dentry->d_name.name, dentry->d_name.len);
2948         BUG_ON(ret);
2949
2950         if (inode->i_nlink == 0) {
2951                 ret = btrfs_orphan_add(trans, inode);
2952                 BUG_ON(ret);
2953         }
2954
2955         nr = trans->blocks_used;
2956         __unlink_end_trans(trans, root);
2957         btrfs_btree_balance_dirty(root, nr);
2958         return ret;
2959 }
2960
2961 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2962                         struct btrfs_root *root,
2963                         struct inode *dir, u64 objectid,
2964                         const char *name, int name_len)
2965 {
2966         struct btrfs_path *path;
2967         struct extent_buffer *leaf;
2968         struct btrfs_dir_item *di;
2969         struct btrfs_key key;
2970         u64 index;
2971         int ret;
2972
2973         path = btrfs_alloc_path();
2974         if (!path)
2975                 return -ENOMEM;
2976
2977         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2978                                    name, name_len, -1);
2979         BUG_ON(!di || IS_ERR(di));
2980
2981         leaf = path->nodes[0];
2982         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2983         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2984         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2985         BUG_ON(ret);
2986         btrfs_release_path(root, path);
2987
2988         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2989                                  objectid, root->root_key.objectid,
2990                                  dir->i_ino, &index, name, name_len);
2991         if (ret < 0) {
2992                 BUG_ON(ret != -ENOENT);
2993                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2994                                                  name, name_len);
2995                 BUG_ON(!di || IS_ERR(di));
2996
2997                 leaf = path->nodes[0];
2998                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2999                 btrfs_release_path(root, path);
3000                 index = key.offset;
3001         }
3002
3003         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3004                                          index, name, name_len, -1);
3005         BUG_ON(!di || IS_ERR(di));
3006
3007         leaf = path->nodes[0];
3008         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3009         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3010         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3011         BUG_ON(ret);
3012         btrfs_release_path(root, path);
3013
3014         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3015         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3016         ret = btrfs_update_inode(trans, root, dir);
3017         BUG_ON(ret);
3018
3019         btrfs_free_path(path);
3020         return 0;
3021 }
3022
3023 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3024 {
3025         struct inode *inode = dentry->d_inode;
3026         int err = 0;
3027         struct btrfs_root *root = BTRFS_I(dir)->root;
3028         struct btrfs_trans_handle *trans;
3029         unsigned long nr = 0;
3030
3031         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3032             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3033                 return -ENOTEMPTY;
3034
3035         trans = __unlink_start_trans(dir, dentry);
3036         if (IS_ERR(trans))
3037                 return PTR_ERR(trans);
3038
3039         btrfs_set_trans_block_group(trans, dir);
3040
3041         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3042                 err = btrfs_unlink_subvol(trans, root, dir,
3043                                           BTRFS_I(inode)->location.objectid,
3044                                           dentry->d_name.name,
3045                                           dentry->d_name.len);
3046                 goto out;
3047         }
3048
3049         err = btrfs_orphan_add(trans, inode);
3050         if (err)
3051                 goto out;
3052
3053         /* now the directory is empty */
3054         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3055                                  dentry->d_name.name, dentry->d_name.len);
3056         if (!err)
3057                 btrfs_i_size_write(inode, 0);
3058 out:
3059         nr = trans->blocks_used;
3060         __unlink_end_trans(trans, root);
3061         btrfs_btree_balance_dirty(root, nr);
3062
3063         return err;
3064 }
3065
3066 #if 0
3067 /*
3068  * when truncating bytes in a file, it is possible to avoid reading
3069  * the leaves that contain only checksum items.  This can be the
3070  * majority of the IO required to delete a large file, but it must
3071  * be done carefully.
3072  *
3073  * The keys in the level just above the leaves are checked to make sure
3074  * the lowest key in a given leaf is a csum key, and starts at an offset
3075  * after the new  size.
3076  *
3077  * Then the key for the next leaf is checked to make sure it also has
3078  * a checksum item for the same file.  If it does, we know our target leaf
3079  * contains only checksum items, and it can be safely freed without reading
3080  * it.
3081  *
3082  * This is just an optimization targeted at large files.  It may do
3083  * nothing.  It will return 0 unless things went badly.
3084  */
3085 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3086                                      struct btrfs_root *root,
3087                                      struct btrfs_path *path,
3088                                      struct inode *inode, u64 new_size)
3089 {
3090         struct btrfs_key key;
3091         int ret;
3092         int nritems;
3093         struct btrfs_key found_key;
3094         struct btrfs_key other_key;
3095         struct btrfs_leaf_ref *ref;
3096         u64 leaf_gen;
3097         u64 leaf_start;
3098
3099         path->lowest_level = 1;
3100         key.objectid = inode->i_ino;
3101         key.type = BTRFS_CSUM_ITEM_KEY;
3102         key.offset = new_size;
3103 again:
3104         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3105         if (ret < 0)
3106                 goto out;
3107
3108         if (path->nodes[1] == NULL) {
3109                 ret = 0;
3110                 goto out;
3111         }
3112         ret = 0;
3113         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3114         nritems = btrfs_header_nritems(path->nodes[1]);
3115
3116         if (!nritems)
3117                 goto out;
3118
3119         if (path->slots[1] >= nritems)
3120                 goto next_node;
3121
3122         /* did we find a key greater than anything we want to delete? */
3123         if (found_key.objectid > inode->i_ino ||
3124            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3125                 goto out;
3126
3127         /* we check the next key in the node to make sure the leave contains
3128          * only checksum items.  This comparison doesn't work if our
3129          * leaf is the last one in the node
3130          */
3131         if (path->slots[1] + 1 >= nritems) {
3132 next_node:
3133                 /* search forward from the last key in the node, this
3134                  * will bring us into the next node in the tree
3135                  */
3136                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3137
3138                 /* unlikely, but we inc below, so check to be safe */
3139                 if (found_key.offset == (u64)-1)
3140                         goto out;
3141
3142                 /* search_forward needs a path with locks held, do the
3143                  * search again for the original key.  It is possible
3144                  * this will race with a balance and return a path that
3145                  * we could modify, but this drop is just an optimization
3146                  * and is allowed to miss some leaves.
3147                  */
3148                 btrfs_release_path(root, path);
3149                 found_key.offset++;
3150
3151                 /* setup a max key for search_forward */
3152                 other_key.offset = (u64)-1;
3153                 other_key.type = key.type;
3154                 other_key.objectid = key.objectid;
3155
3156                 path->keep_locks = 1;
3157                 ret = btrfs_search_forward(root, &found_key, &other_key,
3158                                            path, 0, 0);
3159                 path->keep_locks = 0;
3160                 if (ret || found_key.objectid != key.objectid ||
3161                     found_key.type != key.type) {
3162                         ret = 0;
3163                         goto out;
3164                 }
3165
3166                 key.offset = found_key.offset;
3167                 btrfs_release_path(root, path);
3168                 cond_resched();
3169                 goto again;
3170         }
3171
3172         /* we know there's one more slot after us in the tree,
3173          * read that key so we can verify it is also a checksum item
3174          */
3175         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3176
3177         if (found_key.objectid < inode->i_ino)
3178                 goto next_key;
3179
3180         if (found_key.type != key.type || found_key.offset < new_size)
3181                 goto next_key;
3182
3183         /*
3184          * if the key for the next leaf isn't a csum key from this objectid,
3185          * we can't be sure there aren't good items inside this leaf.
3186          * Bail out
3187          */
3188         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3189                 goto out;
3190
3191         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3192         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3193         /*
3194          * it is safe to delete this leaf, it contains only
3195          * csum items from this inode at an offset >= new_size
3196          */
3197         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3198         BUG_ON(ret);
3199
3200         if (root->ref_cows && leaf_gen < trans->transid) {
3201                 ref = btrfs_alloc_leaf_ref(root, 0);
3202                 if (ref) {
3203                         ref->root_gen = root->root_key.offset;
3204                         ref->bytenr = leaf_start;
3205                         ref->owner = 0;
3206                         ref->generation = leaf_gen;
3207                         ref->nritems = 0;
3208
3209                         btrfs_sort_leaf_ref(ref);
3210
3211                         ret = btrfs_add_leaf_ref(root, ref, 0);
3212                         WARN_ON(ret);
3213                         btrfs_free_leaf_ref(root, ref);
3214                 } else {
3215                         WARN_ON(1);
3216                 }
3217         }
3218 next_key:
3219         btrfs_release_path(root, path);
3220
3221         if (other_key.objectid == inode->i_ino &&
3222             other_key.type == key.type && other_key.offset > key.offset) {
3223                 key.offset = other_key.offset;
3224                 cond_resched();
3225                 goto again;
3226         }
3227         ret = 0;
3228 out:
3229         /* fixup any changes we've made to the path */
3230         path->lowest_level = 0;
3231         path->keep_locks = 0;
3232         btrfs_release_path(root, path);
3233         return ret;
3234 }
3235
3236 #endif
3237
3238 /*
3239  * this can truncate away extent items, csum items and directory items.
3240  * It starts at a high offset and removes keys until it can't find
3241  * any higher than new_size
3242  *
3243  * csum items that cross the new i_size are truncated to the new size
3244  * as well.
3245  *
3246  * min_type is the minimum key type to truncate down to.  If set to 0, this
3247  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3248  */
3249 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3250                                struct btrfs_root *root,
3251                                struct inode *inode,
3252                                u64 new_size, u32 min_type)
3253 {
3254         struct btrfs_path *path;
3255         struct extent_buffer *leaf;
3256         struct btrfs_file_extent_item *fi;
3257         struct btrfs_key key;
3258         struct btrfs_key found_key;
3259         u64 extent_start = 0;
3260         u64 extent_num_bytes = 0;
3261         u64 extent_offset = 0;
3262         u64 item_end = 0;
3263         u64 mask = root->sectorsize - 1;
3264         u32 found_type = (u8)-1;
3265         int found_extent;
3266         int del_item;
3267         int pending_del_nr = 0;
3268         int pending_del_slot = 0;
3269         int extent_type = -1;
3270         int encoding;
3271         int ret;
3272         int err = 0;
3273
3274         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3275
3276         if (root->ref_cows || root == root->fs_info->tree_root)
3277                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3278
3279         path = btrfs_alloc_path();
3280         BUG_ON(!path);
3281         path->reada = -1;
3282
3283         key.objectid = inode->i_ino;
3284         key.offset = (u64)-1;
3285         key.type = (u8)-1;
3286
3287 search_again:
3288         path->leave_spinning = 1;
3289         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3290         if (ret < 0) {
3291                 err = ret;
3292                 goto out;
3293         }
3294
3295         if (ret > 0) {
3296                 /* there are no items in the tree for us to truncate, we're
3297                  * done
3298                  */
3299                 if (path->slots[0] == 0)
3300                         goto out;
3301                 path->slots[0]--;
3302         }
3303
3304         while (1) {
3305                 fi = NULL;
3306                 leaf = path->nodes[0];
3307                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3308                 found_type = btrfs_key_type(&found_key);
3309                 encoding = 0;
3310
3311                 if (found_key.objectid != inode->i_ino)
3312                         break;
3313
3314                 if (found_type < min_type)
3315                         break;
3316
3317                 item_end = found_key.offset;
3318                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3319                         fi = btrfs_item_ptr(leaf, path->slots[0],
3320                                             struct btrfs_file_extent_item);
3321                         extent_type = btrfs_file_extent_type(leaf, fi);
3322                         encoding = btrfs_file_extent_compression(leaf, fi);
3323                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3324                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3325
3326                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3327                                 item_end +=
3328                                     btrfs_file_extent_num_bytes(leaf, fi);
3329                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3330                                 item_end += btrfs_file_extent_inline_len(leaf,
3331                                                                          fi);
3332                         }
3333                         item_end--;
3334                 }
3335                 if (found_type > min_type) {
3336                         del_item = 1;
3337                 } else {
3338                         if (item_end < new_size)
3339                                 break;
3340                         if (found_key.offset >= new_size)
3341                                 del_item = 1;
3342                         else
3343                                 del_item = 0;
3344                 }
3345                 found_extent = 0;
3346                 /* FIXME, shrink the extent if the ref count is only 1 */
3347                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3348                         goto delete;
3349
3350                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3351                         u64 num_dec;
3352                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3353                         if (!del_item && !encoding) {
3354                                 u64 orig_num_bytes =
3355                                         btrfs_file_extent_num_bytes(leaf, fi);
3356                                 extent_num_bytes = new_size -
3357                                         found_key.offset + root->sectorsize - 1;
3358                                 extent_num_bytes = extent_num_bytes &
3359                                         ~((u64)root->sectorsize - 1);
3360                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3361                                                          extent_num_bytes);
3362                                 num_dec = (orig_num_bytes -
3363                                            extent_num_bytes);
3364                                 if (root->ref_cows && extent_start != 0)
3365                                         inode_sub_bytes(inode, num_dec);
3366                                 btrfs_mark_buffer_dirty(leaf);
3367                         } else {
3368                                 extent_num_bytes =
3369                                         btrfs_file_extent_disk_num_bytes(leaf,
3370                                                                          fi);
3371                                 extent_offset = found_key.offset -
3372                                         btrfs_file_extent_offset(leaf, fi);
3373
3374                                 /* FIXME blocksize != 4096 */
3375                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3376                                 if (extent_start != 0) {
3377                                         found_extent = 1;
3378                                         if (root->ref_cows)
3379                                                 inode_sub_bytes(inode, num_dec);
3380                                 }
3381                         }
3382                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3383                         /*
3384                          * we can't truncate inline items that have had
3385                          * special encodings
3386                          */
3387                         if (!del_item &&
3388                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3389                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3390                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3391                                 u32 size = new_size - found_key.offset;
3392
3393                                 if (root->ref_cows) {
3394                                         inode_sub_bytes(inode, item_end + 1 -
3395                                                         new_size);
3396                                 }
3397                                 size =
3398                                     btrfs_file_extent_calc_inline_size(size);
3399                                 ret = btrfs_truncate_item(trans, root, path,
3400                                                           size, 1);
3401                                 BUG_ON(ret);
3402                         } else if (root->ref_cows) {
3403                                 inode_sub_bytes(inode, item_end + 1 -
3404                                                 found_key.offset);
3405                         }
3406                 }
3407 delete:
3408                 if (del_item) {
3409                         if (!pending_del_nr) {
3410                                 /* no pending yet, add ourselves */
3411                                 pending_del_slot = path->slots[0];
3412                                 pending_del_nr = 1;
3413                         } else if (pending_del_nr &&
3414                                    path->slots[0] + 1 == pending_del_slot) {
3415                                 /* hop on the pending chunk */
3416                                 pending_del_nr++;
3417                                 pending_del_slot = path->slots[0];
3418                         } else {
3419                                 BUG();
3420                         }
3421                 } else {
3422                         break;
3423                 }
3424                 if (found_extent && (root->ref_cows ||
3425                                      root == root->fs_info->tree_root)) {
3426                         btrfs_set_path_blocking(path);
3427                         ret = btrfs_free_extent(trans, root, extent_start,
3428                                                 extent_num_bytes, 0,
3429                                                 btrfs_header_owner(leaf),
3430                                                 inode->i_ino, extent_offset);
3431                         BUG_ON(ret);
3432                 }
3433
3434                 if (found_type == BTRFS_INODE_ITEM_KEY)
3435                         break;
3436
3437                 if (path->slots[0] == 0 ||
3438                     path->slots[0] != pending_del_slot) {
3439                         if (root->ref_cows) {
3440                                 err = -EAGAIN;
3441                                 goto out;
3442                         }
3443                         if (pending_del_nr) {
3444                                 ret = btrfs_del_items(trans, root, path,
3445                                                 pending_del_slot,
3446                                                 pending_del_nr);
3447                                 BUG_ON(ret);
3448                                 pending_del_nr = 0;
3449                         }
3450                         btrfs_release_path(root, path);
3451                         goto search_again;
3452                 } else {
3453                         path->slots[0]--;
3454                 }
3455         }
3456 out:
3457         if (pending_del_nr) {
3458                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3459                                       pending_del_nr);
3460                 BUG_ON(ret);
3461         }
3462         btrfs_free_path(path);
3463         return err;
3464 }
3465
3466 /*
3467  * taken from block_truncate_page, but does cow as it zeros out
3468  * any bytes left in the last page in the file.
3469  */
3470 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3471 {
3472         struct inode *inode = mapping->host;
3473         struct btrfs_root *root = BTRFS_I(inode)->root;
3474         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3475         struct btrfs_ordered_extent *ordered;
3476         struct extent_state *cached_state = NULL;
3477         char *kaddr;
3478         u32 blocksize = root->sectorsize;
3479         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3480         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3481         struct page *page;
3482         int ret = 0;
3483         u64 page_start;
3484         u64 page_end;
3485
3486         if ((offset & (blocksize - 1)) == 0)
3487                 goto out;
3488         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3489         if (ret)
3490                 goto out;
3491
3492         ret = -ENOMEM;
3493 again:
3494         page = grab_cache_page(mapping, index);
3495         if (!page) {
3496                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3497                 goto out;
3498         }
3499
3500         page_start = page_offset(page);
3501         page_end = page_start + PAGE_CACHE_SIZE - 1;
3502
3503         if (!PageUptodate(page)) {
3504                 ret = btrfs_readpage(NULL, page);
3505                 lock_page(page);
3506                 if (page->mapping != mapping) {
3507                         unlock_page(page);
3508                         page_cache_release(page);
3509                         goto again;
3510                 }
3511                 if (!PageUptodate(page)) {
3512                         ret = -EIO;
3513                         goto out_unlock;
3514                 }
3515         }
3516         wait_on_page_writeback(page);
3517
3518         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3519                          GFP_NOFS);
3520         set_page_extent_mapped(page);
3521
3522         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3523         if (ordered) {
3524                 unlock_extent_cached(io_tree, page_start, page_end,
3525                                      &cached_state, GFP_NOFS);
3526                 unlock_page(page);
3527                 page_cache_release(page);
3528                 btrfs_start_ordered_extent(inode, ordered, 1);
3529                 btrfs_put_ordered_extent(ordered);
3530                 goto again;
3531         }
3532
3533         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3534                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3535                           0, 0, &cached_state, GFP_NOFS);
3536
3537         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3538                                         &cached_state);
3539         if (ret) {
3540                 unlock_extent_cached(io_tree, page_start, page_end,
3541                                      &cached_state, GFP_NOFS);
3542                 goto out_unlock;
3543         }
3544
3545         ret = 0;
3546         if (offset != PAGE_CACHE_SIZE) {
3547                 kaddr = kmap(page);
3548                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3549                 flush_dcache_page(page);
3550                 kunmap(page);
3551         }
3552         ClearPageChecked(page);
3553         set_page_dirty(page);
3554         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3555                              GFP_NOFS);
3556
3557 out_unlock:
3558         if (ret)
3559                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3560         unlock_page(page);
3561         page_cache_release(page);
3562 out:
3563         return ret;
3564 }
3565
3566 /*
3567  * This function puts in dummy file extents for the area we're creating a hole
3568  * for.  So if we are truncating this file to a larger size we need to insert
3569  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3570  * the range between oldsize and size
3571  */
3572 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3573 {
3574         struct btrfs_trans_handle *trans;
3575         struct btrfs_root *root = BTRFS_I(inode)->root;
3576         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3577         struct extent_map *em = NULL;
3578         struct extent_state *cached_state = NULL;
3579         u64 mask = root->sectorsize - 1;
3580         u64 hole_start = (oldsize + mask) & ~mask;
3581         u64 block_end = (size + mask) & ~mask;
3582         u64 last_byte;
3583         u64 cur_offset;
3584         u64 hole_size;
3585         int err = 0;
3586
3587         if (size <= hole_start)
3588                 return 0;
3589
3590         while (1) {
3591                 struct btrfs_ordered_extent *ordered;
3592                 btrfs_wait_ordered_range(inode, hole_start,
3593                                          block_end - hole_start);
3594                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3595                                  &cached_state, GFP_NOFS);
3596                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3597                 if (!ordered)
3598                         break;
3599                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3600                                      &cached_state, GFP_NOFS);
3601                 btrfs_put_ordered_extent(ordered);
3602         }
3603
3604         cur_offset = hole_start;
3605         while (1) {
3606                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3607                                 block_end - cur_offset, 0);
3608                 BUG_ON(IS_ERR(em) || !em);
3609                 last_byte = min(extent_map_end(em), block_end);
3610                 last_byte = (last_byte + mask) & ~mask;
3611                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3612                         u64 hint_byte = 0;
3613                         hole_size = last_byte - cur_offset;
3614
3615                         trans = btrfs_start_transaction(root, 2);
3616                         if (IS_ERR(trans)) {
3617                                 err = PTR_ERR(trans);
3618                                 break;
3619                         }
3620                         btrfs_set_trans_block_group(trans, inode);
3621
3622                         err = btrfs_drop_extents(trans, inode, cur_offset,
3623                                                  cur_offset + hole_size,
3624                                                  &hint_byte, 1);
3625                         if (err)
3626                                 break;
3627
3628                         err = btrfs_insert_file_extent(trans, root,
3629                                         inode->i_ino, cur_offset, 0,
3630                                         0, hole_size, 0, hole_size,
3631                                         0, 0, 0);
3632                         if (err)
3633                                 break;
3634
3635                         btrfs_drop_extent_cache(inode, hole_start,
3636                                         last_byte - 1, 0);
3637
3638                         btrfs_end_transaction(trans, root);
3639                 }
3640                 free_extent_map(em);
3641                 em = NULL;
3642                 cur_offset = last_byte;
3643                 if (cur_offset >= block_end)
3644                         break;
3645         }
3646
3647         free_extent_map(em);
3648         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3649                              GFP_NOFS);
3650         return err;
3651 }
3652
3653 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3654 {
3655         loff_t oldsize = i_size_read(inode);
3656         int ret;
3657
3658         if (newsize == oldsize)
3659                 return 0;
3660
3661         if (newsize > oldsize) {
3662                 i_size_write(inode, newsize);
3663                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3664                 truncate_pagecache(inode, oldsize, newsize);
3665                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3666                 if (ret) {
3667                         btrfs_setsize(inode, oldsize);
3668                         return ret;
3669                 }
3670
3671                 mark_inode_dirty(inode);
3672         } else {
3673
3674                 /*
3675                  * We're truncating a file that used to have good data down to
3676                  * zero. Make sure it gets into the ordered flush list so that
3677                  * any new writes get down to disk quickly.
3678                  */
3679                 if (newsize == 0)
3680                         BTRFS_I(inode)->ordered_data_close = 1;
3681
3682                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3683                 truncate_setsize(inode, newsize);
3684                 ret = btrfs_truncate(inode);
3685         }
3686
3687         return ret;
3688 }
3689
3690 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3691 {
3692         struct inode *inode = dentry->d_inode;
3693         struct btrfs_root *root = BTRFS_I(inode)->root;
3694         int err;
3695
3696         if (btrfs_root_readonly(root))
3697                 return -EROFS;
3698
3699         err = inode_change_ok(inode, attr);
3700         if (err)
3701                 return err;
3702
3703         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3704                 err = btrfs_setsize(inode, attr->ia_size);
3705                 if (err)
3706                         return err;
3707         }
3708
3709         if (attr->ia_valid) {
3710                 setattr_copy(inode, attr);
3711                 mark_inode_dirty(inode);
3712
3713                 if (attr->ia_valid & ATTR_MODE)
3714                         err = btrfs_acl_chmod(inode);
3715         }
3716
3717         return err;
3718 }
3719
3720 void btrfs_evict_inode(struct inode *inode)
3721 {
3722         struct btrfs_trans_handle *trans;
3723         struct btrfs_root *root = BTRFS_I(inode)->root;
3724         unsigned long nr;
3725         int ret;
3726
3727         trace_btrfs_inode_evict(inode);
3728
3729         truncate_inode_pages(&inode->i_data, 0);
3730         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3731                                root == root->fs_info->tree_root))
3732                 goto no_delete;
3733
3734         if (is_bad_inode(inode)) {
3735                 btrfs_orphan_del(NULL, inode);
3736                 goto no_delete;
3737         }
3738         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3739         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3740
3741         if (root->fs_info->log_root_recovering) {
3742                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3743                 goto no_delete;
3744         }
3745
3746         if (inode->i_nlink > 0) {
3747                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3748                 goto no_delete;
3749         }
3750
3751         btrfs_i_size_write(inode, 0);
3752
3753         while (1) {
3754                 trans = btrfs_start_transaction(root, 0);
3755                 BUG_ON(IS_ERR(trans));
3756                 btrfs_set_trans_block_group(trans, inode);
3757                 trans->block_rsv = root->orphan_block_rsv;
3758
3759                 ret = btrfs_block_rsv_check(trans, root,
3760                                             root->orphan_block_rsv, 0, 5);
3761                 if (ret) {
3762                         BUG_ON(ret != -EAGAIN);
3763                         ret = btrfs_commit_transaction(trans, root);
3764                         BUG_ON(ret);
3765                         continue;
3766                 }
3767
3768                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3769                 if (ret != -EAGAIN)
3770                         break;
3771
3772                 nr = trans->blocks_used;
3773                 btrfs_end_transaction(trans, root);
3774                 trans = NULL;
3775                 btrfs_btree_balance_dirty(root, nr);
3776
3777         }
3778
3779         if (ret == 0) {
3780                 ret = btrfs_orphan_del(trans, inode);
3781                 BUG_ON(ret);
3782         }
3783
3784         nr = trans->blocks_used;
3785         btrfs_end_transaction(trans, root);
3786         btrfs_btree_balance_dirty(root, nr);
3787 no_delete:
3788         end_writeback(inode);
3789         return;
3790 }
3791
3792 /*
3793  * this returns the key found in the dir entry in the location pointer.
3794  * If no dir entries were found, location->objectid is 0.
3795  */
3796 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3797                                struct btrfs_key *location)
3798 {
3799         const char *name = dentry->d_name.name;
3800         int namelen = dentry->d_name.len;
3801         struct btrfs_dir_item *di;
3802         struct btrfs_path *path;
3803         struct btrfs_root *root = BTRFS_I(dir)->root;
3804         int ret = 0;
3805
3806         path = btrfs_alloc_path();
3807         BUG_ON(!path);
3808
3809         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3810                                     namelen, 0);
3811         if (IS_ERR(di))
3812                 ret = PTR_ERR(di);
3813
3814         if (!di || IS_ERR(di))
3815                 goto out_err;
3816
3817         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3818 out:
3819         btrfs_free_path(path);
3820         return ret;
3821 out_err:
3822         location->objectid = 0;
3823         goto out;
3824 }
3825
3826 /*
3827  * when we hit a tree root in a directory, the btrfs part of the inode
3828  * needs to be changed to reflect the root directory of the tree root.  This
3829  * is kind of like crossing a mount point.
3830  */
3831 static int fixup_tree_root_location(struct btrfs_root *root,
3832                                     struct inode *dir,
3833                                     struct dentry *dentry,
3834                                     struct btrfs_key *location,
3835                                     struct btrfs_root **sub_root)
3836 {
3837         struct btrfs_path *path;
3838         struct btrfs_root *new_root;
3839         struct btrfs_root_ref *ref;
3840         struct extent_buffer *leaf;
3841         int ret;
3842         int err = 0;
3843
3844         path = btrfs_alloc_path();
3845         if (!path) {
3846                 err = -ENOMEM;
3847                 goto out;
3848         }
3849
3850         err = -ENOENT;
3851         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3852                                   BTRFS_I(dir)->root->root_key.objectid,
3853                                   location->objectid);
3854         if (ret) {
3855                 if (ret < 0)
3856                         err = ret;
3857                 goto out;
3858         }
3859
3860         leaf = path->nodes[0];
3861         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3862         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3863             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3864                 goto out;
3865
3866         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3867                                    (unsigned long)(ref + 1),
3868                                    dentry->d_name.len);
3869         if (ret)
3870                 goto out;
3871
3872         btrfs_release_path(root->fs_info->tree_root, path);
3873
3874         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3875         if (IS_ERR(new_root)) {
3876                 err = PTR_ERR(new_root);
3877                 goto out;
3878         }
3879
3880         if (btrfs_root_refs(&new_root->root_item) == 0) {
3881                 err = -ENOENT;
3882                 goto out;
3883         }
3884
3885         *sub_root = new_root;
3886         location->objectid = btrfs_root_dirid(&new_root->root_item);
3887         location->type = BTRFS_INODE_ITEM_KEY;
3888         location->offset = 0;
3889         err = 0;
3890 out:
3891         btrfs_free_path(path);
3892         return err;
3893 }
3894
3895 static void inode_tree_add(struct inode *inode)
3896 {
3897         struct btrfs_root *root = BTRFS_I(inode)->root;
3898         struct btrfs_inode *entry;
3899         struct rb_node **p;
3900         struct rb_node *parent;
3901 again:
3902         p = &root->inode_tree.rb_node;
3903         parent = NULL;
3904
3905         if (inode_unhashed(inode))
3906                 return;
3907
3908         spin_lock(&root->inode_lock);
3909         while (*p) {
3910                 parent = *p;
3911                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3912
3913                 if (inode->i_ino < entry->vfs_inode.i_ino)
3914                         p = &parent->rb_left;
3915                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3916                         p = &parent->rb_right;
3917                 else {
3918                         WARN_ON(!(entry->vfs_inode.i_state &
3919                                   (I_WILL_FREE | I_FREEING)));
3920                         rb_erase(parent, &root->inode_tree);
3921                         RB_CLEAR_NODE(parent);
3922                         spin_unlock(&root->inode_lock);
3923                         goto again;
3924                 }
3925         }
3926         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3927         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3928         spin_unlock(&root->inode_lock);
3929 }
3930
3931 static void inode_tree_del(struct inode *inode)
3932 {
3933         struct btrfs_root *root = BTRFS_I(inode)->root;
3934         int empty = 0;
3935
3936         spin_lock(&root->inode_lock);
3937         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3938                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3939                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3940                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3941         }
3942         spin_unlock(&root->inode_lock);
3943
3944         /*
3945          * Free space cache has inodes in the tree root, but the tree root has a
3946          * root_refs of 0, so this could end up dropping the tree root as a
3947          * snapshot, so we need the extra !root->fs_info->tree_root check to
3948          * make sure we don't drop it.
3949          */
3950         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3951             root != root->fs_info->tree_root) {
3952                 synchronize_srcu(&root->fs_info->subvol_srcu);
3953                 spin_lock(&root->inode_lock);
3954                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3955                 spin_unlock(&root->inode_lock);
3956                 if (empty)
3957                         btrfs_add_dead_root(root);
3958         }
3959 }
3960
3961 int btrfs_invalidate_inodes(struct btrfs_root *root)
3962 {
3963         struct rb_node *node;
3964         struct rb_node *prev;
3965         struct btrfs_inode *entry;
3966         struct inode *inode;
3967         u64 objectid = 0;
3968
3969         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3970
3971         spin_lock(&root->inode_lock);
3972 again:
3973         node = root->inode_tree.rb_node;
3974         prev = NULL;
3975         while (node) {
3976                 prev = node;
3977                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3978
3979                 if (objectid < entry->vfs_inode.i_ino)
3980                         node = node->rb_left;
3981                 else if (objectid > entry->vfs_inode.i_ino)
3982                         node = node->rb_right;
3983                 else
3984                         break;
3985         }
3986         if (!node) {
3987                 while (prev) {
3988                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3989                         if (objectid <= entry->vfs_inode.i_ino) {
3990                                 node = prev;
3991                                 break;
3992                         }
3993                         prev = rb_next(prev);
3994                 }
3995         }
3996         while (node) {
3997                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3998                 objectid = entry->vfs_inode.i_ino + 1;
3999                 inode = igrab(&entry->vfs_inode);
4000                 if (inode) {
4001                         spin_unlock(&root->inode_lock);
4002                         if (atomic_read(&inode->i_count) > 1)
4003                                 d_prune_aliases(inode);
4004                         /*
4005                          * btrfs_drop_inode will have it removed from
4006                          * the inode cache when its usage count
4007                          * hits zero.
4008                          */
4009                         iput(inode);
4010                         cond_resched();
4011                         spin_lock(&root->inode_lock);
4012                         goto again;
4013                 }
4014
4015                 if (cond_resched_lock(&root->inode_lock))
4016                         goto again;
4017
4018                 node = rb_next(node);
4019         }
4020         spin_unlock(&root->inode_lock);
4021         return 0;
4022 }
4023
4024 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4025 {
4026         struct btrfs_iget_args *args = p;
4027         inode->i_ino = args->ino;
4028         BTRFS_I(inode)->root = args->root;
4029         btrfs_set_inode_space_info(args->root, inode);
4030         return 0;
4031 }
4032
4033 static int btrfs_find_actor(struct inode *inode, void *opaque)
4034 {
4035         struct btrfs_iget_args *args = opaque;
4036         return args->ino == inode->i_ino &&
4037                 args->root == BTRFS_I(inode)->root;
4038 }
4039
4040 static struct inode *btrfs_iget_locked(struct super_block *s,
4041                                        u64 objectid,
4042                                        struct btrfs_root *root)
4043 {
4044         struct inode *inode;
4045         struct btrfs_iget_args args;
4046         args.ino = objectid;
4047         args.root = root;
4048
4049         inode = iget5_locked(s, objectid, btrfs_find_actor,
4050                              btrfs_init_locked_inode,
4051                              (void *)&args);
4052         return inode;
4053 }
4054
4055 /* Get an inode object given its location and corresponding root.
4056  * Returns in *is_new if the inode was read from disk
4057  */
4058 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4059                          struct btrfs_root *root, int *new)
4060 {
4061         struct inode *inode;
4062
4063         inode = btrfs_iget_locked(s, location->objectid, root);
4064         if (!inode)
4065                 return ERR_PTR(-ENOMEM);
4066
4067         if (inode->i_state & I_NEW) {
4068                 BTRFS_I(inode)->root = root;
4069                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4070                 btrfs_read_locked_inode(inode);
4071
4072                 inode_tree_add(inode);
4073                 unlock_new_inode(inode);
4074                 if (new)
4075                         *new = 1;
4076         }
4077
4078         return inode;
4079 }
4080
4081 static struct inode *new_simple_dir(struct super_block *s,
4082                                     struct btrfs_key *key,
4083                                     struct btrfs_root *root)
4084 {
4085         struct inode *inode = new_inode(s);
4086
4087         if (!inode)
4088                 return ERR_PTR(-ENOMEM);
4089
4090         BTRFS_I(inode)->root = root;
4091         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4092         BTRFS_I(inode)->dummy_inode = 1;
4093
4094         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4095         inode->i_op = &simple_dir_inode_operations;
4096         inode->i_fop = &simple_dir_operations;
4097         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4098         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4099
4100         return inode;
4101 }
4102
4103 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4104 {
4105         struct inode *inode;
4106         struct btrfs_root *root = BTRFS_I(dir)->root;
4107         struct btrfs_root *sub_root = root;
4108         struct btrfs_key location;
4109         int index;
4110         int ret;
4111
4112         if (dentry->d_name.len > BTRFS_NAME_LEN)
4113                 return ERR_PTR(-ENAMETOOLONG);
4114
4115         ret = btrfs_inode_by_name(dir, dentry, &location);
4116
4117         if (ret < 0)
4118                 return ERR_PTR(ret);
4119
4120         if (location.objectid == 0)
4121                 return NULL;
4122
4123         if (location.type == BTRFS_INODE_ITEM_KEY) {
4124                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4125                 return inode;
4126         }
4127
4128         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4129
4130         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4131         ret = fixup_tree_root_location(root, dir, dentry,
4132                                        &location, &sub_root);
4133         if (ret < 0) {
4134                 if (ret != -ENOENT)
4135                         inode = ERR_PTR(ret);
4136                 else
4137                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4138         } else {
4139                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4140         }
4141         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4142
4143         if (!IS_ERR(inode) && root != sub_root) {
4144                 down_read(&root->fs_info->cleanup_work_sem);
4145                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4146                         ret = btrfs_orphan_cleanup(sub_root);
4147                 up_read(&root->fs_info->cleanup_work_sem);
4148                 if (ret)
4149                         inode = ERR_PTR(ret);
4150         }
4151
4152         return inode;
4153 }
4154
4155 static int btrfs_dentry_delete(const struct dentry *dentry)
4156 {
4157         struct btrfs_root *root;
4158
4159         if (!dentry->d_inode && !IS_ROOT(dentry))
4160                 dentry = dentry->d_parent;
4161
4162         if (dentry->d_inode) {
4163                 root = BTRFS_I(dentry->d_inode)->root;
4164                 if (btrfs_root_refs(&root->root_item) == 0)
4165                         return 1;
4166         }
4167         return 0;
4168 }
4169
4170 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4171                                    struct nameidata *nd)
4172 {
4173         struct inode *inode;
4174
4175         inode = btrfs_lookup_dentry(dir, dentry);
4176         if (IS_ERR(inode))
4177                 return ERR_CAST(inode);
4178
4179         return d_splice_alias(inode, dentry);
4180 }
4181
4182 static unsigned char btrfs_filetype_table[] = {
4183         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4184 };
4185
4186 static int btrfs_real_readdir(struct file *filp, void *dirent,
4187                               filldir_t filldir)
4188 {
4189         struct inode *inode = filp->f_dentry->d_inode;
4190         struct btrfs_root *root = BTRFS_I(inode)->root;
4191         struct btrfs_item *item;
4192         struct btrfs_dir_item *di;
4193         struct btrfs_key key;
4194         struct btrfs_key found_key;
4195         struct btrfs_path *path;
4196         int ret;
4197         u32 nritems;
4198         struct extent_buffer *leaf;
4199         int slot;
4200         int advance;
4201         unsigned char d_type;
4202         int over = 0;
4203         u32 di_cur;
4204         u32 di_total;
4205         u32 di_len;
4206         int key_type = BTRFS_DIR_INDEX_KEY;
4207         char tmp_name[32];
4208         char *name_ptr;
4209         int name_len;
4210
4211         /* FIXME, use a real flag for deciding about the key type */
4212         if (root->fs_info->tree_root == root)
4213                 key_type = BTRFS_DIR_ITEM_KEY;
4214
4215         /* special case for "." */
4216         if (filp->f_pos == 0) {
4217                 over = filldir(dirent, ".", 1,
4218                                1, inode->i_ino,
4219                                DT_DIR);
4220                 if (over)
4221                         return 0;
4222                 filp->f_pos = 1;
4223         }
4224         /* special case for .., just use the back ref */
4225         if (filp->f_pos == 1) {
4226                 u64 pino = parent_ino(filp->f_path.dentry);
4227                 over = filldir(dirent, "..", 2,
4228                                2, pino, DT_DIR);
4229                 if (over)
4230                         return 0;
4231                 filp->f_pos = 2;
4232         }
4233         path = btrfs_alloc_path();
4234         path->reada = 2;
4235
4236         btrfs_set_key_type(&key, key_type);
4237         key.offset = filp->f_pos;
4238         key.objectid = inode->i_ino;
4239
4240         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4241         if (ret < 0)
4242                 goto err;
4243         advance = 0;
4244
4245         while (1) {
4246                 leaf = path->nodes[0];
4247                 nritems = btrfs_header_nritems(leaf);
4248                 slot = path->slots[0];
4249                 if (advance || slot >= nritems) {
4250                         if (slot >= nritems - 1) {
4251                                 ret = btrfs_next_leaf(root, path);
4252                                 if (ret)
4253                                         break;
4254                                 leaf = path->nodes[0];
4255                                 nritems = btrfs_header_nritems(leaf);
4256                                 slot = path->slots[0];
4257                         } else {
4258                                 slot++;
4259                                 path->slots[0]++;
4260                         }
4261                 }
4262
4263                 advance = 1;
4264                 item = btrfs_item_nr(leaf, slot);
4265                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4266
4267                 if (found_key.objectid != key.objectid)
4268                         break;
4269                 if (btrfs_key_type(&found_key) != key_type)
4270                         break;
4271                 if (found_key.offset < filp->f_pos)
4272                         continue;
4273
4274                 filp->f_pos = found_key.offset;
4275
4276                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4277                 di_cur = 0;
4278                 di_total = btrfs_item_size(leaf, item);
4279
4280                 while (di_cur < di_total) {
4281                         struct btrfs_key location;
4282
4283                         if (verify_dir_item(root, leaf, di))
4284                                 break;
4285
4286                         name_len = btrfs_dir_name_len(leaf, di);
4287                         if (name_len <= sizeof(tmp_name)) {
4288                                 name_ptr = tmp_name;
4289                         } else {
4290                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4291                                 if (!name_ptr) {
4292                                         ret = -ENOMEM;
4293                                         goto err;
4294                                 }
4295                         }
4296                         read_extent_buffer(leaf, name_ptr,
4297                                            (unsigned long)(di + 1), name_len);
4298
4299                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4300                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4301
4302                         /* is this a reference to our own snapshot? If so
4303                          * skip it
4304                          */
4305                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4306                             location.objectid == root->root_key.objectid) {
4307                                 over = 0;
4308                                 goto skip;
4309                         }
4310                         over = filldir(dirent, name_ptr, name_len,
4311                                        found_key.offset, location.objectid,
4312                                        d_type);
4313
4314 skip:
4315                         if (name_ptr != tmp_name)
4316                                 kfree(name_ptr);
4317
4318                         if (over)
4319                                 goto nopos;
4320                         di_len = btrfs_dir_name_len(leaf, di) +
4321                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4322                         di_cur += di_len;
4323                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4324                 }
4325         }
4326
4327         /* Reached end of directory/root. Bump pos past the last item. */
4328         if (key_type == BTRFS_DIR_INDEX_KEY)
4329                 /*
4330                  * 32-bit glibc will use getdents64, but then strtol -
4331                  * so the last number we can serve is this.
4332                  */
4333                 filp->f_pos = 0x7fffffff;
4334         else
4335                 filp->f_pos++;
4336 nopos:
4337         ret = 0;
4338 err:
4339         btrfs_free_path(path);
4340         return ret;
4341 }
4342
4343 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4344 {
4345         struct btrfs_root *root = BTRFS_I(inode)->root;
4346         struct btrfs_trans_handle *trans;
4347         int ret = 0;
4348         bool nolock = false;
4349
4350         if (BTRFS_I(inode)->dummy_inode)
4351                 return 0;
4352
4353         smp_mb();
4354         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4355
4356         if (wbc->sync_mode == WB_SYNC_ALL) {
4357                 if (nolock)
4358                         trans = btrfs_join_transaction_nolock(root, 1);
4359                 else
4360                         trans = btrfs_join_transaction(root, 1);
4361                 if (IS_ERR(trans))
4362                         return PTR_ERR(trans);
4363                 btrfs_set_trans_block_group(trans, inode);
4364                 if (nolock)
4365                         ret = btrfs_end_transaction_nolock(trans, root);
4366                 else
4367                         ret = btrfs_commit_transaction(trans, root);
4368         }
4369         return ret;
4370 }
4371
4372 /*
4373  * This is somewhat expensive, updating the tree every time the
4374  * inode changes.  But, it is most likely to find the inode in cache.
4375  * FIXME, needs more benchmarking...there are no reasons other than performance
4376  * to keep or drop this code.
4377  */
4378 void btrfs_dirty_inode(struct inode *inode)
4379 {
4380         struct btrfs_root *root = BTRFS_I(inode)->root;
4381         struct btrfs_trans_handle *trans;
4382         int ret;
4383
4384         if (BTRFS_I(inode)->dummy_inode)
4385                 return;
4386
4387         trans = btrfs_join_transaction(root, 1);
4388         BUG_ON(IS_ERR(trans));
4389         btrfs_set_trans_block_group(trans, inode);
4390
4391         ret = btrfs_update_inode(trans, root, inode);
4392         if (ret && ret == -ENOSPC) {
4393                 /* whoops, lets try again with the full transaction */
4394                 btrfs_end_transaction(trans, root);
4395                 trans = btrfs_start_transaction(root, 1);
4396                 if (IS_ERR(trans)) {
4397                         if (printk_ratelimit()) {
4398                                 printk(KERN_ERR "btrfs: fail to "
4399                                        "dirty  inode %lu error %ld\n",
4400                                        inode->i_ino, PTR_ERR(trans));
4401                         }
4402                         return;
4403                 }
4404                 btrfs_set_trans_block_group(trans, inode);
4405
4406                 ret = btrfs_update_inode(trans, root, inode);
4407                 if (ret) {
4408                         if (printk_ratelimit()) {
4409                                 printk(KERN_ERR "btrfs: fail to "
4410                                        "dirty  inode %lu error %d\n",
4411                                        inode->i_ino, ret);
4412                         }
4413                 }
4414         }
4415         btrfs_end_transaction(trans, root);
4416 }
4417
4418 /*
4419  * find the highest existing sequence number in a directory
4420  * and then set the in-memory index_cnt variable to reflect
4421  * free sequence numbers
4422  */
4423 static int btrfs_set_inode_index_count(struct inode *inode)
4424 {
4425         struct btrfs_root *root = BTRFS_I(inode)->root;
4426         struct btrfs_key key, found_key;
4427         struct btrfs_path *path;
4428         struct extent_buffer *leaf;
4429         int ret;
4430
4431         key.objectid = inode->i_ino;
4432         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4433         key.offset = (u64)-1;
4434
4435         path = btrfs_alloc_path();
4436         if (!path)
4437                 return -ENOMEM;
4438
4439         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4440         if (ret < 0)
4441                 goto out;
4442         /* FIXME: we should be able to handle this */
4443         if (ret == 0)
4444                 goto out;
4445         ret = 0;
4446
4447         /*
4448          * MAGIC NUMBER EXPLANATION:
4449          * since we search a directory based on f_pos we have to start at 2
4450          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4451          * else has to start at 2
4452          */
4453         if (path->slots[0] == 0) {
4454                 BTRFS_I(inode)->index_cnt = 2;
4455                 goto out;
4456         }
4457
4458         path->slots[0]--;
4459
4460         leaf = path->nodes[0];
4461         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4462
4463         if (found_key.objectid != inode->i_ino ||
4464             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4465                 BTRFS_I(inode)->index_cnt = 2;
4466                 goto out;
4467         }
4468
4469         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4470 out:
4471         btrfs_free_path(path);
4472         return ret;
4473 }
4474
4475 /*
4476  * helper to find a free sequence number in a given directory.  This current
4477  * code is very simple, later versions will do smarter things in the btree
4478  */
4479 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4480 {
4481         int ret = 0;
4482
4483         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4484                 ret = btrfs_set_inode_index_count(dir);
4485                 if (ret)
4486                         return ret;
4487         }
4488
4489         *index = BTRFS_I(dir)->index_cnt;
4490         BTRFS_I(dir)->index_cnt++;
4491
4492         return ret;
4493 }
4494
4495 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4496                                      struct btrfs_root *root,
4497                                      struct inode *dir,
4498                                      const char *name, int name_len,
4499                                      u64 ref_objectid, u64 objectid,
4500                                      u64 alloc_hint, int mode, u64 *index)
4501 {
4502         struct inode *inode;
4503         struct btrfs_inode_item *inode_item;
4504         struct btrfs_key *location;
4505         struct btrfs_path *path;
4506         struct btrfs_inode_ref *ref;
4507         struct btrfs_key key[2];
4508         u32 sizes[2];
4509         unsigned long ptr;
4510         int ret;
4511         int owner;
4512
4513         path = btrfs_alloc_path();
4514         BUG_ON(!path);
4515
4516         inode = new_inode(root->fs_info->sb);
4517         if (!inode)
4518                 return ERR_PTR(-ENOMEM);
4519
4520         if (dir) {
4521                 trace_btrfs_inode_request(dir);
4522
4523                 ret = btrfs_set_inode_index(dir, index);
4524                 if (ret) {
4525                         iput(inode);
4526                         return ERR_PTR(ret);
4527                 }
4528         }
4529         /*
4530          * index_cnt is ignored for everything but a dir,
4531          * btrfs_get_inode_index_count has an explanation for the magic
4532          * number
4533          */
4534         BTRFS_I(inode)->index_cnt = 2;
4535         BTRFS_I(inode)->root = root;
4536         BTRFS_I(inode)->generation = trans->transid;
4537         inode->i_generation = BTRFS_I(inode)->generation;
4538         btrfs_set_inode_space_info(root, inode);
4539
4540         if (mode & S_IFDIR)
4541                 owner = 0;
4542         else
4543                 owner = 1;
4544         BTRFS_I(inode)->block_group =
4545                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4546
4547         key[0].objectid = objectid;
4548         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4549         key[0].offset = 0;
4550
4551         key[1].objectid = objectid;
4552         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4553         key[1].offset = ref_objectid;
4554
4555         sizes[0] = sizeof(struct btrfs_inode_item);
4556         sizes[1] = name_len + sizeof(*ref);
4557
4558         path->leave_spinning = 1;
4559         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4560         if (ret != 0)
4561                 goto fail;
4562
4563         inode_init_owner(inode, dir, mode);
4564         inode->i_ino = objectid;
4565         inode_set_bytes(inode, 0);
4566         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4567         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4568                                   struct btrfs_inode_item);
4569         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4570
4571         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4572                              struct btrfs_inode_ref);
4573         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4574         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4575         ptr = (unsigned long)(ref + 1);
4576         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4577
4578         btrfs_mark_buffer_dirty(path->nodes[0]);
4579         btrfs_free_path(path);
4580
4581         location = &BTRFS_I(inode)->location;
4582         location->objectid = objectid;
4583         location->offset = 0;
4584         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4585
4586         btrfs_inherit_iflags(inode, dir);
4587
4588         if ((mode & S_IFREG)) {
4589                 if (btrfs_test_opt(root, NODATASUM))
4590                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4591                 if (btrfs_test_opt(root, NODATACOW) ||
4592                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4593                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4594         }
4595
4596         insert_inode_hash(inode);
4597         inode_tree_add(inode);
4598
4599         trace_btrfs_inode_new(inode);
4600
4601         return inode;
4602 fail:
4603         if (dir)
4604                 BTRFS_I(dir)->index_cnt--;
4605         btrfs_free_path(path);
4606         iput(inode);
4607         return ERR_PTR(ret);
4608 }
4609
4610 static inline u8 btrfs_inode_type(struct inode *inode)
4611 {
4612         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4613 }
4614
4615 /*
4616  * utility function to add 'inode' into 'parent_inode' with
4617  * a give name and a given sequence number.
4618  * if 'add_backref' is true, also insert a backref from the
4619  * inode to the parent directory.
4620  */
4621 int btrfs_add_link(struct btrfs_trans_handle *trans,
4622                    struct inode *parent_inode, struct inode *inode,
4623                    const char *name, int name_len, int add_backref, u64 index)
4624 {
4625         int ret = 0;
4626         struct btrfs_key key;
4627         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4628
4629         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4630                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4631         } else {
4632                 key.objectid = inode->i_ino;
4633                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4634                 key.offset = 0;
4635         }
4636
4637         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4638                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4639                                          key.objectid, root->root_key.objectid,
4640                                          parent_inode->i_ino,
4641                                          index, name, name_len);
4642         } else if (add_backref) {
4643                 ret = btrfs_insert_inode_ref(trans, root,
4644                                              name, name_len, inode->i_ino,
4645                                              parent_inode->i_ino, index);
4646         }
4647
4648         if (ret == 0) {
4649                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4650                                             parent_inode->i_ino, &key,
4651                                             btrfs_inode_type(inode), index);
4652                 BUG_ON(ret);
4653
4654                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4655                                    name_len * 2);
4656                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4657                 ret = btrfs_update_inode(trans, root, parent_inode);
4658         }
4659         return ret;
4660 }
4661
4662 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4663                             struct inode *dir, struct dentry *dentry,
4664                             struct inode *inode, int backref, u64 index)
4665 {
4666         int err = btrfs_add_link(trans, dir, inode,
4667                                  dentry->d_name.name, dentry->d_name.len,
4668                                  backref, index);
4669         if (!err) {
4670                 d_instantiate(dentry, inode);
4671                 return 0;
4672         }
4673         if (err > 0)
4674                 err = -EEXIST;
4675         return err;
4676 }
4677
4678 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4679                         int mode, dev_t rdev)
4680 {
4681         struct btrfs_trans_handle *trans;
4682         struct btrfs_root *root = BTRFS_I(dir)->root;
4683         struct inode *inode = NULL;
4684         int err;
4685         int drop_inode = 0;
4686         u64 objectid;
4687         unsigned long nr = 0;
4688         u64 index = 0;
4689
4690         if (!new_valid_dev(rdev))
4691                 return -EINVAL;
4692
4693         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4694         if (err)
4695                 return err;
4696
4697         /*
4698          * 2 for inode item and ref
4699          * 2 for dir items
4700          * 1 for xattr if selinux is on
4701          */
4702         trans = btrfs_start_transaction(root, 5);
4703         if (IS_ERR(trans))
4704                 return PTR_ERR(trans);
4705
4706         btrfs_set_trans_block_group(trans, dir);
4707
4708         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4709                                 dentry->d_name.len, dir->i_ino, objectid,
4710                                 BTRFS_I(dir)->block_group, mode, &index);
4711         err = PTR_ERR(inode);
4712         if (IS_ERR(inode))
4713                 goto out_unlock;
4714
4715         err = btrfs_init_inode_security(trans, inode, dir);
4716         if (err) {
4717                 drop_inode = 1;
4718                 goto out_unlock;
4719         }
4720
4721         btrfs_set_trans_block_group(trans, inode);
4722         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4723         if (err)
4724                 drop_inode = 1;
4725         else {
4726                 inode->i_op = &btrfs_special_inode_operations;
4727                 init_special_inode(inode, inode->i_mode, rdev);
4728                 btrfs_update_inode(trans, root, inode);
4729         }
4730         btrfs_update_inode_block_group(trans, inode);
4731         btrfs_update_inode_block_group(trans, dir);
4732 out_unlock:
4733         nr = trans->blocks_used;
4734         btrfs_end_transaction_throttle(trans, root);
4735         btrfs_btree_balance_dirty(root, nr);
4736         if (drop_inode) {
4737                 inode_dec_link_count(inode);
4738                 iput(inode);
4739         }
4740         return err;
4741 }
4742
4743 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4744                         int mode, struct nameidata *nd)
4745 {
4746         struct btrfs_trans_handle *trans;
4747         struct btrfs_root *root = BTRFS_I(dir)->root;
4748         struct inode *inode = NULL;
4749         int drop_inode = 0;
4750         int err;
4751         unsigned long nr = 0;
4752         u64 objectid;
4753         u64 index = 0;
4754
4755         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4756         if (err)
4757                 return err;
4758         /*
4759          * 2 for inode item and ref
4760          * 2 for dir items
4761          * 1 for xattr if selinux is on
4762          */
4763         trans = btrfs_start_transaction(root, 5);
4764         if (IS_ERR(trans))
4765                 return PTR_ERR(trans);
4766
4767         btrfs_set_trans_block_group(trans, dir);
4768
4769         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4770                                 dentry->d_name.len, dir->i_ino, objectid,
4771                                 BTRFS_I(dir)->block_group, mode, &index);
4772         err = PTR_ERR(inode);
4773         if (IS_ERR(inode))
4774                 goto out_unlock;
4775
4776         err = btrfs_init_inode_security(trans, inode, dir);
4777         if (err) {
4778                 drop_inode = 1;
4779                 goto out_unlock;
4780         }
4781
4782         btrfs_set_trans_block_group(trans, inode);
4783         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4784         if (err)
4785                 drop_inode = 1;
4786         else {
4787                 inode->i_mapping->a_ops = &btrfs_aops;
4788                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4789                 inode->i_fop = &btrfs_file_operations;
4790                 inode->i_op = &btrfs_file_inode_operations;
4791                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4792         }
4793         btrfs_update_inode_block_group(trans, inode);
4794         btrfs_update_inode_block_group(trans, dir);
4795 out_unlock:
4796         nr = trans->blocks_used;
4797         btrfs_end_transaction_throttle(trans, root);
4798         if (drop_inode) {
4799                 inode_dec_link_count(inode);
4800                 iput(inode);
4801         }
4802         btrfs_btree_balance_dirty(root, nr);
4803         return err;
4804 }
4805
4806 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4807                       struct dentry *dentry)
4808 {
4809         struct btrfs_trans_handle *trans;
4810         struct btrfs_root *root = BTRFS_I(dir)->root;
4811         struct inode *inode = old_dentry->d_inode;
4812         u64 index;
4813         unsigned long nr = 0;
4814         int err;
4815         int drop_inode = 0;
4816
4817         if (inode->i_nlink == 0)
4818                 return -ENOENT;
4819
4820         /* do not allow sys_link's with other subvols of the same device */
4821         if (root->objectid != BTRFS_I(inode)->root->objectid)
4822                 return -EXDEV;
4823
4824         btrfs_inc_nlink(inode);
4825         inode->i_ctime = CURRENT_TIME;
4826
4827         err = btrfs_set_inode_index(dir, &index);
4828         if (err)
4829                 goto fail;
4830
4831         /*
4832          * 2 items for inode and inode ref
4833          * 2 items for dir items
4834          * 1 item for parent inode
4835          */
4836         trans = btrfs_start_transaction(root, 5);
4837         if (IS_ERR(trans)) {
4838                 err = PTR_ERR(trans);
4839                 goto fail;
4840         }
4841
4842         btrfs_set_trans_block_group(trans, dir);
4843         ihold(inode);
4844
4845         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4846
4847         if (err) {
4848                 drop_inode = 1;
4849         } else {
4850                 struct dentry *parent = dget_parent(dentry);
4851                 btrfs_update_inode_block_group(trans, dir);
4852                 err = btrfs_update_inode(trans, root, inode);
4853                 BUG_ON(err);
4854                 btrfs_log_new_name(trans, inode, NULL, parent);
4855                 dput(parent);
4856         }
4857
4858         nr = trans->blocks_used;
4859         btrfs_end_transaction_throttle(trans, root);
4860 fail:
4861         if (drop_inode) {
4862                 inode_dec_link_count(inode);
4863                 iput(inode);
4864         }
4865         btrfs_btree_balance_dirty(root, nr);
4866         return err;
4867 }
4868
4869 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4870 {
4871         struct inode *inode = NULL;
4872         struct btrfs_trans_handle *trans;
4873         struct btrfs_root *root = BTRFS_I(dir)->root;
4874         int err = 0;
4875         int drop_on_err = 0;
4876         u64 objectid = 0;
4877         u64 index = 0;
4878         unsigned long nr = 1;
4879
4880         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4881         if (err)
4882                 return err;
4883
4884         /*
4885          * 2 items for inode and ref
4886          * 2 items for dir items
4887          * 1 for xattr if selinux is on
4888          */
4889         trans = btrfs_start_transaction(root, 5);
4890         if (IS_ERR(trans))
4891                 return PTR_ERR(trans);
4892         btrfs_set_trans_block_group(trans, dir);
4893
4894         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4895                                 dentry->d_name.len, dir->i_ino, objectid,
4896                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4897                                 &index);
4898         if (IS_ERR(inode)) {
4899                 err = PTR_ERR(inode);
4900                 goto out_fail;
4901         }
4902
4903         drop_on_err = 1;
4904
4905         err = btrfs_init_inode_security(trans, inode, dir);
4906         if (err)
4907                 goto out_fail;
4908
4909         inode->i_op = &btrfs_dir_inode_operations;
4910         inode->i_fop = &btrfs_dir_file_operations;
4911         btrfs_set_trans_block_group(trans, inode);
4912
4913         btrfs_i_size_write(inode, 0);
4914         err = btrfs_update_inode(trans, root, inode);
4915         if (err)
4916                 goto out_fail;
4917
4918         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4919                              dentry->d_name.len, 0, index);
4920         if (err)
4921                 goto out_fail;
4922
4923         d_instantiate(dentry, inode);
4924         drop_on_err = 0;
4925         btrfs_update_inode_block_group(trans, inode);
4926         btrfs_update_inode_block_group(trans, dir);
4927
4928 out_fail:
4929         nr = trans->blocks_used;
4930         btrfs_end_transaction_throttle(trans, root);
4931         if (drop_on_err)
4932                 iput(inode);
4933         btrfs_btree_balance_dirty(root, nr);
4934         return err;
4935 }
4936
4937 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4938  * and an extent that you want to insert, deal with overlap and insert
4939  * the new extent into the tree.
4940  */
4941 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4942                                 struct extent_map *existing,
4943                                 struct extent_map *em,
4944                                 u64 map_start, u64 map_len)
4945 {
4946         u64 start_diff;
4947
4948         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4949         start_diff = map_start - em->start;
4950         em->start = map_start;
4951         em->len = map_len;
4952         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4953             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4954                 em->block_start += start_diff;
4955                 em->block_len -= start_diff;
4956         }
4957         return add_extent_mapping(em_tree, em);
4958 }
4959
4960 static noinline int uncompress_inline(struct btrfs_path *path,
4961                                       struct inode *inode, struct page *page,
4962                                       size_t pg_offset, u64 extent_offset,
4963                                       struct btrfs_file_extent_item *item)
4964 {
4965         int ret;
4966         struct extent_buffer *leaf = path->nodes[0];
4967         char *tmp;
4968         size_t max_size;
4969         unsigned long inline_size;
4970         unsigned long ptr;
4971         int compress_type;
4972
4973         WARN_ON(pg_offset != 0);
4974         compress_type = btrfs_file_extent_compression(leaf, item);
4975         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4976         inline_size = btrfs_file_extent_inline_item_len(leaf,
4977                                         btrfs_item_nr(leaf, path->slots[0]));
4978         tmp = kmalloc(inline_size, GFP_NOFS);
4979         ptr = btrfs_file_extent_inline_start(item);
4980
4981         read_extent_buffer(leaf, tmp, ptr, inline_size);
4982
4983         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4984         ret = btrfs_decompress(compress_type, tmp, page,
4985                                extent_offset, inline_size, max_size);
4986         if (ret) {
4987                 char *kaddr = kmap_atomic(page, KM_USER0);
4988                 unsigned long copy_size = min_t(u64,
4989                                   PAGE_CACHE_SIZE - pg_offset,
4990                                   max_size - extent_offset);
4991                 memset(kaddr + pg_offset, 0, copy_size);
4992                 kunmap_atomic(kaddr, KM_USER0);
4993         }
4994         kfree(tmp);
4995         return 0;
4996 }
4997
4998 /*
4999  * a bit scary, this does extent mapping from logical file offset to the disk.
5000  * the ugly parts come from merging extents from the disk with the in-ram
5001  * representation.  This gets more complex because of the data=ordered code,
5002  * where the in-ram extents might be locked pending data=ordered completion.
5003  *
5004  * This also copies inline extents directly into the page.
5005  */
5006
5007 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5008                                     size_t pg_offset, u64 start, u64 len,
5009                                     int create)
5010 {
5011         int ret;
5012         int err = 0;
5013         u64 bytenr;
5014         u64 extent_start = 0;
5015         u64 extent_end = 0;
5016         u64 objectid = inode->i_ino;
5017         u32 found_type;
5018         struct btrfs_path *path = NULL;
5019         struct btrfs_root *root = BTRFS_I(inode)->root;
5020         struct btrfs_file_extent_item *item;
5021         struct extent_buffer *leaf;
5022         struct btrfs_key found_key;
5023         struct extent_map *em = NULL;
5024         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5025         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5026         struct btrfs_trans_handle *trans = NULL;
5027         int compress_type;
5028
5029 again:
5030         read_lock(&em_tree->lock);
5031         em = lookup_extent_mapping(em_tree, start, len);
5032         if (em)
5033                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5034         read_unlock(&em_tree->lock);
5035
5036         if (em) {
5037                 if (em->start > start || em->start + em->len <= start)
5038                         free_extent_map(em);
5039                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5040                         free_extent_map(em);
5041                 else
5042                         goto out;
5043         }
5044         em = alloc_extent_map(GFP_NOFS);
5045         if (!em) {
5046                 err = -ENOMEM;
5047                 goto out;
5048         }
5049         em->bdev = root->fs_info->fs_devices->latest_bdev;
5050         em->start = EXTENT_MAP_HOLE;
5051         em->orig_start = EXTENT_MAP_HOLE;
5052         em->len = (u64)-1;
5053         em->block_len = (u64)-1;
5054
5055         if (!path) {
5056                 path = btrfs_alloc_path();
5057                 BUG_ON(!path);
5058         }
5059
5060         ret = btrfs_lookup_file_extent(trans, root, path,
5061                                        objectid, start, trans != NULL);
5062         if (ret < 0) {
5063                 err = ret;
5064                 goto out;
5065         }
5066
5067         if (ret != 0) {
5068                 if (path->slots[0] == 0)
5069                         goto not_found;
5070                 path->slots[0]--;
5071         }
5072
5073         leaf = path->nodes[0];
5074         item = btrfs_item_ptr(leaf, path->slots[0],
5075                               struct btrfs_file_extent_item);
5076         /* are we inside the extent that was found? */
5077         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5078         found_type = btrfs_key_type(&found_key);
5079         if (found_key.objectid != objectid ||
5080             found_type != BTRFS_EXTENT_DATA_KEY) {
5081                 goto not_found;
5082         }
5083
5084         found_type = btrfs_file_extent_type(leaf, item);
5085         extent_start = found_key.offset;
5086         compress_type = btrfs_file_extent_compression(leaf, item);
5087         if (found_type == BTRFS_FILE_EXTENT_REG ||
5088             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5089                 extent_end = extent_start +
5090                        btrfs_file_extent_num_bytes(leaf, item);
5091         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5092                 size_t size;
5093                 size = btrfs_file_extent_inline_len(leaf, item);
5094                 extent_end = (extent_start + size + root->sectorsize - 1) &
5095                         ~((u64)root->sectorsize - 1);
5096         }
5097
5098         if (start >= extent_end) {
5099                 path->slots[0]++;
5100                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5101                         ret = btrfs_next_leaf(root, path);
5102                         if (ret < 0) {
5103                                 err = ret;
5104                                 goto out;
5105                         }
5106                         if (ret > 0)
5107                                 goto not_found;
5108                         leaf = path->nodes[0];
5109                 }
5110                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5111                 if (found_key.objectid != objectid ||
5112                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5113                         goto not_found;
5114                 if (start + len <= found_key.offset)
5115                         goto not_found;
5116                 em->start = start;
5117                 em->len = found_key.offset - start;
5118                 goto not_found_em;
5119         }
5120
5121         if (found_type == BTRFS_FILE_EXTENT_REG ||
5122             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5123                 em->start = extent_start;
5124                 em->len = extent_end - extent_start;
5125                 em->orig_start = extent_start -
5126                                  btrfs_file_extent_offset(leaf, item);
5127                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5128                 if (bytenr == 0) {
5129                         em->block_start = EXTENT_MAP_HOLE;
5130                         goto insert;
5131                 }
5132                 if (compress_type != BTRFS_COMPRESS_NONE) {
5133                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5134                         em->compress_type = compress_type;
5135                         em->block_start = bytenr;
5136                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5137                                                                          item);
5138                 } else {
5139                         bytenr += btrfs_file_extent_offset(leaf, item);
5140                         em->block_start = bytenr;
5141                         em->block_len = em->len;
5142                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5143                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5144                 }
5145                 goto insert;
5146         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5147                 unsigned long ptr;
5148                 char *map;
5149                 size_t size;
5150                 size_t extent_offset;
5151                 size_t copy_size;
5152
5153                 em->block_start = EXTENT_MAP_INLINE;
5154                 if (!page || create) {
5155                         em->start = extent_start;
5156                         em->len = extent_end - extent_start;
5157                         goto out;
5158                 }
5159
5160                 size = btrfs_file_extent_inline_len(leaf, item);
5161                 extent_offset = page_offset(page) + pg_offset - extent_start;
5162                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5163                                 size - extent_offset);
5164                 em->start = extent_start + extent_offset;
5165                 em->len = (copy_size + root->sectorsize - 1) &
5166                         ~((u64)root->sectorsize - 1);
5167                 em->orig_start = EXTENT_MAP_INLINE;
5168                 if (compress_type) {
5169                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5170                         em->compress_type = compress_type;
5171                 }
5172                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5173                 if (create == 0 && !PageUptodate(page)) {
5174                         if (btrfs_file_extent_compression(leaf, item) !=
5175                             BTRFS_COMPRESS_NONE) {
5176                                 ret = uncompress_inline(path, inode, page,
5177                                                         pg_offset,
5178                                                         extent_offset, item);
5179                                 BUG_ON(ret);
5180                         } else {
5181                                 map = kmap(page);
5182                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5183                                                    copy_size);
5184                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5185                                         memset(map + pg_offset + copy_size, 0,
5186                                                PAGE_CACHE_SIZE - pg_offset -
5187                                                copy_size);
5188                                 }
5189                                 kunmap(page);
5190                         }
5191                         flush_dcache_page(page);
5192                 } else if (create && PageUptodate(page)) {
5193                         WARN_ON(1);
5194                         if (!trans) {
5195                                 kunmap(page);
5196                                 free_extent_map(em);
5197                                 em = NULL;
5198                                 btrfs_release_path(root, path);
5199                                 trans = btrfs_join_transaction(root, 1);
5200                                 if (IS_ERR(trans))
5201                                         return ERR_CAST(trans);
5202                                 goto again;
5203                         }
5204                         map = kmap(page);
5205                         write_extent_buffer(leaf, map + pg_offset, ptr,
5206                                             copy_size);
5207                         kunmap(page);
5208                         btrfs_mark_buffer_dirty(leaf);
5209                 }
5210                 set_extent_uptodate(io_tree, em->start,
5211                                     extent_map_end(em) - 1, GFP_NOFS);
5212                 goto insert;
5213         } else {
5214                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5215                 WARN_ON(1);
5216         }
5217 not_found:
5218         em->start = start;
5219         em->len = len;
5220 not_found_em:
5221         em->block_start = EXTENT_MAP_HOLE;
5222         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5223 insert:
5224         btrfs_release_path(root, path);
5225         if (em->start > start || extent_map_end(em) <= start) {
5226                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5227                        "[%llu %llu]\n", (unsigned long long)em->start,
5228                        (unsigned long long)em->len,
5229                        (unsigned long long)start,
5230                        (unsigned long long)len);
5231                 err = -EIO;
5232                 goto out;
5233         }
5234
5235         err = 0;
5236         write_lock(&em_tree->lock);
5237         ret = add_extent_mapping(em_tree, em);
5238         /* it is possible that someone inserted the extent into the tree
5239          * while we had the lock dropped.  It is also possible that
5240          * an overlapping map exists in the tree
5241          */
5242         if (ret == -EEXIST) {
5243                 struct extent_map *existing;
5244
5245                 ret = 0;
5246
5247                 existing = lookup_extent_mapping(em_tree, start, len);
5248                 if (existing && (existing->start > start ||
5249                     existing->start + existing->len <= start)) {
5250                         free_extent_map(existing);
5251                         existing = NULL;
5252                 }
5253                 if (!existing) {
5254                         existing = lookup_extent_mapping(em_tree, em->start,
5255                                                          em->len);
5256                         if (existing) {
5257                                 err = merge_extent_mapping(em_tree, existing,
5258                                                            em, start,
5259                                                            root->sectorsize);
5260                                 free_extent_map(existing);
5261                                 if (err) {
5262                                         free_extent_map(em);
5263                                         em = NULL;
5264                                 }
5265                         } else {
5266                                 err = -EIO;
5267                                 free_extent_map(em);
5268                                 em = NULL;
5269                         }
5270                 } else {
5271                         free_extent_map(em);
5272                         em = existing;
5273                         err = 0;
5274                 }
5275         }
5276         write_unlock(&em_tree->lock);
5277 out:
5278
5279         trace_btrfs_get_extent(root, em);
5280
5281         if (path)
5282                 btrfs_free_path(path);
5283         if (trans) {
5284                 ret = btrfs_end_transaction(trans, root);
5285                 if (!err)
5286                         err = ret;
5287         }
5288         if (err) {
5289                 free_extent_map(em);
5290                 return ERR_PTR(err);
5291         }
5292         return em;
5293 }
5294
5295 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5296                                            size_t pg_offset, u64 start, u64 len,
5297                                            int create)
5298 {
5299         struct extent_map *em;
5300         struct extent_map *hole_em = NULL;
5301         u64 range_start = start;
5302         u64 end;
5303         u64 found;
5304         u64 found_end;
5305         int err = 0;
5306
5307         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5308         if (IS_ERR(em))
5309                 return em;
5310         if (em) {
5311                 /*
5312                  * if our em maps to a hole, there might
5313                  * actually be delalloc bytes behind it
5314                  */
5315                 if (em->block_start != EXTENT_MAP_HOLE)
5316                         return em;
5317                 else
5318                         hole_em = em;
5319         }
5320
5321         /* check to see if we've wrapped (len == -1 or similar) */
5322         end = start + len;
5323         if (end < start)
5324                 end = (u64)-1;
5325         else
5326                 end -= 1;
5327
5328         em = NULL;
5329
5330         /* ok, we didn't find anything, lets look for delalloc */
5331         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5332                                  end, len, EXTENT_DELALLOC, 1);
5333         found_end = range_start + found;
5334         if (found_end < range_start)
5335                 found_end = (u64)-1;
5336
5337         /*
5338          * we didn't find anything useful, return
5339          * the original results from get_extent()
5340          */
5341         if (range_start > end || found_end <= start) {
5342                 em = hole_em;
5343                 hole_em = NULL;
5344                 goto out;
5345         }
5346
5347         /* adjust the range_start to make sure it doesn't
5348          * go backwards from the start they passed in
5349          */
5350         range_start = max(start,range_start);
5351         found = found_end - range_start;
5352
5353         if (found > 0) {
5354                 u64 hole_start = start;
5355                 u64 hole_len = len;
5356
5357                 em = alloc_extent_map(GFP_NOFS);
5358                 if (!em) {
5359                         err = -ENOMEM;
5360                         goto out;
5361                 }
5362                 /*
5363                  * when btrfs_get_extent can't find anything it
5364                  * returns one huge hole
5365                  *
5366                  * make sure what it found really fits our range, and
5367                  * adjust to make sure it is based on the start from
5368                  * the caller
5369                  */
5370                 if (hole_em) {
5371                         u64 calc_end = extent_map_end(hole_em);
5372
5373                         if (calc_end <= start || (hole_em->start > end)) {
5374                                 free_extent_map(hole_em);
5375                                 hole_em = NULL;
5376                         } else {
5377                                 hole_start = max(hole_em->start, start);
5378                                 hole_len = calc_end - hole_start;
5379                         }
5380                 }
5381                 em->bdev = NULL;
5382                 if (hole_em && range_start > hole_start) {
5383                         /* our hole starts before our delalloc, so we
5384                          * have to return just the parts of the hole
5385                          * that go until  the delalloc starts
5386                          */
5387                         em->len = min(hole_len,
5388                                       range_start - hole_start);
5389                         em->start = hole_start;
5390                         em->orig_start = hole_start;
5391                         /*
5392                          * don't adjust block start at all,
5393                          * it is fixed at EXTENT_MAP_HOLE
5394                          */
5395                         em->block_start = hole_em->block_start;
5396                         em->block_len = hole_len;
5397                 } else {
5398                         em->start = range_start;
5399                         em->len = found;
5400                         em->orig_start = range_start;
5401                         em->block_start = EXTENT_MAP_DELALLOC;
5402                         em->block_len = found;
5403                 }
5404         } else if (hole_em) {
5405                 return hole_em;
5406         }
5407 out:
5408
5409         free_extent_map(hole_em);
5410         if (err) {
5411                 free_extent_map(em);
5412                 return ERR_PTR(err);
5413         }
5414         return em;
5415 }
5416
5417 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5418                                                   u64 start, u64 len)
5419 {
5420         struct btrfs_root *root = BTRFS_I(inode)->root;
5421         struct btrfs_trans_handle *trans;
5422         struct extent_map *em;
5423         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5424         struct btrfs_key ins;
5425         u64 alloc_hint;
5426         int ret;
5427
5428         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5429
5430         trans = btrfs_join_transaction(root, 0);
5431         if (IS_ERR(trans))
5432                 return ERR_CAST(trans);
5433
5434         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5435
5436         alloc_hint = get_extent_allocation_hint(inode, start, len);
5437         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5438                                    alloc_hint, (u64)-1, &ins, 1);
5439         if (ret) {
5440                 em = ERR_PTR(ret);
5441                 goto out;
5442         }
5443
5444         em = alloc_extent_map(GFP_NOFS);
5445         if (!em) {
5446                 em = ERR_PTR(-ENOMEM);
5447                 goto out;
5448         }
5449
5450         em->start = start;
5451         em->orig_start = em->start;
5452         em->len = ins.offset;
5453
5454         em->block_start = ins.objectid;
5455         em->block_len = ins.offset;
5456         em->bdev = root->fs_info->fs_devices->latest_bdev;
5457         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5458
5459         while (1) {
5460                 write_lock(&em_tree->lock);
5461                 ret = add_extent_mapping(em_tree, em);
5462                 write_unlock(&em_tree->lock);
5463                 if (ret != -EEXIST)
5464                         break;
5465                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5466         }
5467
5468         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5469                                            ins.offset, ins.offset, 0);
5470         if (ret) {
5471                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5472                 em = ERR_PTR(ret);
5473         }
5474 out:
5475         btrfs_end_transaction(trans, root);
5476         return em;
5477 }
5478
5479 /*
5480  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5481  * block must be cow'd
5482  */
5483 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5484                                       struct inode *inode, u64 offset, u64 len)
5485 {
5486         struct btrfs_path *path;
5487         int ret;
5488         struct extent_buffer *leaf;
5489         struct btrfs_root *root = BTRFS_I(inode)->root;
5490         struct btrfs_file_extent_item *fi;
5491         struct btrfs_key key;
5492         u64 disk_bytenr;
5493         u64 backref_offset;
5494         u64 extent_end;
5495         u64 num_bytes;
5496         int slot;
5497         int found_type;
5498
5499         path = btrfs_alloc_path();
5500         if (!path)
5501                 return -ENOMEM;
5502
5503         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5504                                        offset, 0);
5505         if (ret < 0)
5506                 goto out;
5507
5508         slot = path->slots[0];
5509         if (ret == 1) {
5510                 if (slot == 0) {
5511                         /* can't find the item, must cow */
5512                         ret = 0;
5513                         goto out;
5514                 }
5515                 slot--;
5516         }
5517         ret = 0;
5518         leaf = path->nodes[0];
5519         btrfs_item_key_to_cpu(leaf, &key, slot);
5520         if (key.objectid != inode->i_ino ||
5521             key.type != BTRFS_EXTENT_DATA_KEY) {
5522                 /* not our file or wrong item type, must cow */
5523                 goto out;
5524         }
5525
5526         if (key.offset > offset) {
5527                 /* Wrong offset, must cow */
5528                 goto out;
5529         }
5530
5531         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5532         found_type = btrfs_file_extent_type(leaf, fi);
5533         if (found_type != BTRFS_FILE_EXTENT_REG &&
5534             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5535                 /* not a regular extent, must cow */
5536                 goto out;
5537         }
5538         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5539         backref_offset = btrfs_file_extent_offset(leaf, fi);
5540
5541         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5542         if (extent_end < offset + len) {
5543                 /* extent doesn't include our full range, must cow */
5544                 goto out;
5545         }
5546
5547         if (btrfs_extent_readonly(root, disk_bytenr))
5548                 goto out;
5549
5550         /*
5551          * look for other files referencing this extent, if we
5552          * find any we must cow
5553          */
5554         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5555                                   key.offset - backref_offset, disk_bytenr))
5556                 goto out;
5557
5558         /*
5559          * adjust disk_bytenr and num_bytes to cover just the bytes
5560          * in this extent we are about to write.  If there
5561          * are any csums in that range we have to cow in order
5562          * to keep the csums correct
5563          */
5564         disk_bytenr += backref_offset;
5565         disk_bytenr += offset - key.offset;
5566         num_bytes = min(offset + len, extent_end) - offset;
5567         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5568                                 goto out;
5569         /*
5570          * all of the above have passed, it is safe to overwrite this extent
5571          * without cow
5572          */
5573         ret = 1;
5574 out:
5575         btrfs_free_path(path);
5576         return ret;
5577 }
5578
5579 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5580                                    struct buffer_head *bh_result, int create)
5581 {
5582         struct extent_map *em;
5583         struct btrfs_root *root = BTRFS_I(inode)->root;
5584         u64 start = iblock << inode->i_blkbits;
5585         u64 len = bh_result->b_size;
5586         struct btrfs_trans_handle *trans;
5587
5588         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5589         if (IS_ERR(em))
5590                 return PTR_ERR(em);
5591
5592         /*
5593          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5594          * io.  INLINE is special, and we could probably kludge it in here, but
5595          * it's still buffered so for safety lets just fall back to the generic
5596          * buffered path.
5597          *
5598          * For COMPRESSED we _have_ to read the entire extent in so we can
5599          * decompress it, so there will be buffering required no matter what we
5600          * do, so go ahead and fallback to buffered.
5601          *
5602          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5603          * to buffered IO.  Don't blame me, this is the price we pay for using
5604          * the generic code.
5605          */
5606         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5607             em->block_start == EXTENT_MAP_INLINE) {
5608                 free_extent_map(em);
5609                 return -ENOTBLK;
5610         }
5611
5612         /* Just a good old fashioned hole, return */
5613         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5614                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5615                 free_extent_map(em);
5616                 /* DIO will do one hole at a time, so just unlock a sector */
5617                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5618                               start + root->sectorsize - 1, GFP_NOFS);
5619                 return 0;
5620         }
5621
5622         /*
5623          * We don't allocate a new extent in the following cases
5624          *
5625          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5626          * existing extent.
5627          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5628          * just use the extent.
5629          *
5630          */
5631         if (!create) {
5632                 len = em->len - (start - em->start);
5633                 goto map;
5634         }
5635
5636         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5637             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5638              em->block_start != EXTENT_MAP_HOLE)) {
5639                 int type;
5640                 int ret;
5641                 u64 block_start;
5642
5643                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5644                         type = BTRFS_ORDERED_PREALLOC;
5645                 else
5646                         type = BTRFS_ORDERED_NOCOW;
5647                 len = min(len, em->len - (start - em->start));
5648                 block_start = em->block_start + (start - em->start);
5649
5650                 /*
5651                  * we're not going to log anything, but we do need
5652                  * to make sure the current transaction stays open
5653                  * while we look for nocow cross refs
5654                  */
5655                 trans = btrfs_join_transaction(root, 0);
5656                 if (IS_ERR(trans))
5657                         goto must_cow;
5658
5659                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5660                         ret = btrfs_add_ordered_extent_dio(inode, start,
5661                                            block_start, len, len, type);
5662                         btrfs_end_transaction(trans, root);
5663                         if (ret) {
5664                                 free_extent_map(em);
5665                                 return ret;
5666                         }
5667                         goto unlock;
5668                 }
5669                 btrfs_end_transaction(trans, root);
5670         }
5671 must_cow:
5672         /*
5673          * this will cow the extent, reset the len in case we changed
5674          * it above
5675          */
5676         len = bh_result->b_size;
5677         free_extent_map(em);
5678         em = btrfs_new_extent_direct(inode, start, len);
5679         if (IS_ERR(em))
5680                 return PTR_ERR(em);
5681         len = min(len, em->len - (start - em->start));
5682 unlock:
5683         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5684                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5685                           0, NULL, GFP_NOFS);
5686 map:
5687         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5688                 inode->i_blkbits;
5689         bh_result->b_size = len;
5690         bh_result->b_bdev = em->bdev;
5691         set_buffer_mapped(bh_result);
5692         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5693                 set_buffer_new(bh_result);
5694
5695         free_extent_map(em);
5696
5697         return 0;
5698 }
5699
5700 struct btrfs_dio_private {
5701         struct inode *inode;
5702         u64 logical_offset;
5703         u64 disk_bytenr;
5704         u64 bytes;
5705         u32 *csums;
5706         void *private;
5707
5708         /* number of bios pending for this dio */
5709         atomic_t pending_bios;
5710
5711         /* IO errors */
5712         int errors;
5713
5714         struct bio *orig_bio;
5715 };
5716
5717 static void btrfs_endio_direct_read(struct bio *bio, int err)
5718 {
5719         struct btrfs_dio_private *dip = bio->bi_private;
5720         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5721         struct bio_vec *bvec = bio->bi_io_vec;
5722         struct inode *inode = dip->inode;
5723         struct btrfs_root *root = BTRFS_I(inode)->root;
5724         u64 start;
5725         u32 *private = dip->csums;
5726
5727         start = dip->logical_offset;
5728         do {
5729                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5730                         struct page *page = bvec->bv_page;
5731                         char *kaddr;
5732                         u32 csum = ~(u32)0;
5733                         unsigned long flags;
5734
5735                         local_irq_save(flags);
5736                         kaddr = kmap_atomic(page, KM_IRQ0);
5737                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5738                                                csum, bvec->bv_len);
5739                         btrfs_csum_final(csum, (char *)&csum);
5740                         kunmap_atomic(kaddr, KM_IRQ0);
5741                         local_irq_restore(flags);
5742
5743                         flush_dcache_page(bvec->bv_page);
5744                         if (csum != *private) {
5745                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5746                                       " %llu csum %u private %u\n",
5747                                       inode->i_ino, (unsigned long long)start,
5748                                       csum, *private);
5749                                 err = -EIO;
5750                         }
5751                 }
5752
5753                 start += bvec->bv_len;
5754                 private++;
5755                 bvec++;
5756         } while (bvec <= bvec_end);
5757
5758         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5759                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5760         bio->bi_private = dip->private;
5761
5762         kfree(dip->csums);
5763         kfree(dip);
5764
5765         /* If we had a csum failure make sure to clear the uptodate flag */
5766         if (err)
5767                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5768         dio_end_io(bio, err);
5769 }
5770
5771 static void btrfs_endio_direct_write(struct bio *bio, int err)
5772 {
5773         struct btrfs_dio_private *dip = bio->bi_private;
5774         struct inode *inode = dip->inode;
5775         struct btrfs_root *root = BTRFS_I(inode)->root;
5776         struct btrfs_trans_handle *trans;
5777         struct btrfs_ordered_extent *ordered = NULL;
5778         struct extent_state *cached_state = NULL;
5779         u64 ordered_offset = dip->logical_offset;
5780         u64 ordered_bytes = dip->bytes;
5781         int ret;
5782
5783         if (err)
5784                 goto out_done;
5785 again:
5786         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5787                                                    &ordered_offset,
5788                                                    ordered_bytes);
5789         if (!ret)
5790                 goto out_test;
5791
5792         BUG_ON(!ordered);
5793
5794         trans = btrfs_join_transaction(root, 1);
5795         if (IS_ERR(trans)) {
5796                 err = -ENOMEM;
5797                 goto out;
5798         }
5799         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5800
5801         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5802                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5803                 if (!ret)
5804                         ret = btrfs_update_inode(trans, root, inode);
5805                 err = ret;
5806                 goto out;
5807         }
5808
5809         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5810                          ordered->file_offset + ordered->len - 1, 0,
5811                          &cached_state, GFP_NOFS);
5812
5813         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5814                 ret = btrfs_mark_extent_written(trans, inode,
5815                                                 ordered->file_offset,
5816                                                 ordered->file_offset +
5817                                                 ordered->len);
5818                 if (ret) {
5819                         err = ret;
5820                         goto out_unlock;
5821                 }
5822         } else {
5823                 ret = insert_reserved_file_extent(trans, inode,
5824                                                   ordered->file_offset,
5825                                                   ordered->start,
5826                                                   ordered->disk_len,
5827                                                   ordered->len,
5828                                                   ordered->len,
5829                                                   0, 0, 0,
5830                                                   BTRFS_FILE_EXTENT_REG);
5831                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5832                                    ordered->file_offset, ordered->len);
5833                 if (ret) {
5834                         err = ret;
5835                         WARN_ON(1);
5836                         goto out_unlock;
5837                 }
5838         }
5839
5840         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5841         btrfs_ordered_update_i_size(inode, 0, ordered);
5842         btrfs_update_inode(trans, root, inode);
5843 out_unlock:
5844         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5845                              ordered->file_offset + ordered->len - 1,
5846                              &cached_state, GFP_NOFS);
5847 out:
5848         btrfs_delalloc_release_metadata(inode, ordered->len);
5849         btrfs_end_transaction(trans, root);
5850         ordered_offset = ordered->file_offset + ordered->len;
5851         btrfs_put_ordered_extent(ordered);
5852         btrfs_put_ordered_extent(ordered);
5853
5854 out_test:
5855         /*
5856          * our bio might span multiple ordered extents.  If we haven't
5857          * completed the accounting for the whole dio, go back and try again
5858          */
5859         if (ordered_offset < dip->logical_offset + dip->bytes) {
5860                 ordered_bytes = dip->logical_offset + dip->bytes -
5861                         ordered_offset;
5862                 goto again;
5863         }
5864 out_done:
5865         bio->bi_private = dip->private;
5866
5867         kfree(dip->csums);
5868         kfree(dip);
5869
5870         /* If we had an error make sure to clear the uptodate flag */
5871         if (err)
5872                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5873         dio_end_io(bio, err);
5874 }
5875
5876 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5877                                     struct bio *bio, int mirror_num,
5878                                     unsigned long bio_flags, u64 offset)
5879 {
5880         int ret;
5881         struct btrfs_root *root = BTRFS_I(inode)->root;
5882         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5883         BUG_ON(ret);
5884         return 0;
5885 }
5886
5887 static void btrfs_end_dio_bio(struct bio *bio, int err)
5888 {
5889         struct btrfs_dio_private *dip = bio->bi_private;
5890
5891         if (err) {
5892                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5893                       "sector %#Lx len %u err no %d\n",
5894                       dip->inode->i_ino, bio->bi_rw,
5895                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5896                 dip->errors = 1;
5897
5898                 /*
5899                  * before atomic variable goto zero, we must make sure
5900                  * dip->errors is perceived to be set.
5901                  */
5902                 smp_mb__before_atomic_dec();
5903         }
5904
5905         /* if there are more bios still pending for this dio, just exit */
5906         if (!atomic_dec_and_test(&dip->pending_bios))
5907                 goto out;
5908
5909         if (dip->errors)
5910                 bio_io_error(dip->orig_bio);
5911         else {
5912                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5913                 bio_endio(dip->orig_bio, 0);
5914         }
5915 out:
5916         bio_put(bio);
5917 }
5918
5919 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5920                                        u64 first_sector, gfp_t gfp_flags)
5921 {
5922         int nr_vecs = bio_get_nr_vecs(bdev);
5923         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5924 }
5925
5926 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5927                                          int rw, u64 file_offset, int skip_sum,
5928                                          u32 *csums)
5929 {
5930         int write = rw & REQ_WRITE;
5931         struct btrfs_root *root = BTRFS_I(inode)->root;
5932         int ret;
5933
5934         bio_get(bio);
5935         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5936         if (ret)
5937                 goto err;
5938
5939         if (write && !skip_sum) {
5940                 ret = btrfs_wq_submit_bio(root->fs_info,
5941                                    inode, rw, bio, 0, 0,
5942                                    file_offset,
5943                                    __btrfs_submit_bio_start_direct_io,
5944                                    __btrfs_submit_bio_done);
5945                 goto err;
5946         } else if (!skip_sum)
5947                 btrfs_lookup_bio_sums_dio(root, inode, bio,
5948                                           file_offset, csums);
5949
5950         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5951 err:
5952         bio_put(bio);
5953         return ret;
5954 }
5955
5956 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5957                                     int skip_sum)
5958 {
5959         struct inode *inode = dip->inode;
5960         struct btrfs_root *root = BTRFS_I(inode)->root;
5961         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5962         struct bio *bio;
5963         struct bio *orig_bio = dip->orig_bio;
5964         struct bio_vec *bvec = orig_bio->bi_io_vec;
5965         u64 start_sector = orig_bio->bi_sector;
5966         u64 file_offset = dip->logical_offset;
5967         u64 submit_len = 0;
5968         u64 map_length;
5969         int nr_pages = 0;
5970         u32 *csums = dip->csums;
5971         int ret = 0;
5972         int write = rw & REQ_WRITE;
5973
5974         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5975         if (!bio)
5976                 return -ENOMEM;
5977         bio->bi_private = dip;
5978         bio->bi_end_io = btrfs_end_dio_bio;
5979         atomic_inc(&dip->pending_bios);
5980
5981         map_length = orig_bio->bi_size;
5982         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5983                               &map_length, NULL, 0);
5984         if (ret) {
5985                 bio_put(bio);
5986                 return -EIO;
5987         }
5988
5989         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5990                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5991                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5992                                  bvec->bv_offset) < bvec->bv_len)) {
5993                         /*
5994                          * inc the count before we submit the bio so
5995                          * we know the end IO handler won't happen before
5996                          * we inc the count. Otherwise, the dip might get freed
5997                          * before we're done setting it up
5998                          */
5999                         atomic_inc(&dip->pending_bios);
6000                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6001                                                      file_offset, skip_sum,
6002                                                      csums);
6003                         if (ret) {
6004                                 bio_put(bio);
6005                                 atomic_dec(&dip->pending_bios);
6006                                 goto out_err;
6007                         }
6008
6009                         /* Write's use the ordered csums */
6010                         if (!write && !skip_sum)
6011                                 csums = csums + nr_pages;
6012                         start_sector += submit_len >> 9;
6013                         file_offset += submit_len;
6014
6015                         submit_len = 0;
6016                         nr_pages = 0;
6017
6018                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6019                                                   start_sector, GFP_NOFS);
6020                         if (!bio)
6021                                 goto out_err;
6022                         bio->bi_private = dip;
6023                         bio->bi_end_io = btrfs_end_dio_bio;
6024
6025                         map_length = orig_bio->bi_size;
6026                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6027                                               &map_length, NULL, 0);
6028                         if (ret) {
6029                                 bio_put(bio);
6030                                 goto out_err;
6031                         }
6032                 } else {
6033                         submit_len += bvec->bv_len;
6034                         nr_pages ++;
6035                         bvec++;
6036                 }
6037         }
6038
6039         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6040                                      csums);
6041         if (!ret)
6042                 return 0;
6043
6044         bio_put(bio);
6045 out_err:
6046         dip->errors = 1;
6047         /*
6048          * before atomic variable goto zero, we must
6049          * make sure dip->errors is perceived to be set.
6050          */
6051         smp_mb__before_atomic_dec();
6052         if (atomic_dec_and_test(&dip->pending_bios))
6053                 bio_io_error(dip->orig_bio);
6054
6055         /* bio_end_io() will handle error, so we needn't return it */
6056         return 0;
6057 }
6058
6059 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6060                                 loff_t file_offset)
6061 {
6062         struct btrfs_root *root = BTRFS_I(inode)->root;
6063         struct btrfs_dio_private *dip;
6064         struct bio_vec *bvec = bio->bi_io_vec;
6065         int skip_sum;
6066         int write = rw & REQ_WRITE;
6067         int ret = 0;
6068
6069         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6070
6071         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6072         if (!dip) {
6073                 ret = -ENOMEM;
6074                 goto free_ordered;
6075         }
6076         dip->csums = NULL;
6077
6078         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6079         if (!write && !skip_sum) {
6080                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6081                 if (!dip->csums) {
6082                         kfree(dip);
6083                         ret = -ENOMEM;
6084                         goto free_ordered;
6085                 }
6086         }
6087
6088         dip->private = bio->bi_private;
6089         dip->inode = inode;
6090         dip->logical_offset = file_offset;
6091
6092         dip->bytes = 0;
6093         do {
6094                 dip->bytes += bvec->bv_len;
6095                 bvec++;
6096         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6097
6098         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6099         bio->bi_private = dip;
6100         dip->errors = 0;
6101         dip->orig_bio = bio;
6102         atomic_set(&dip->pending_bios, 0);
6103
6104         if (write)
6105                 bio->bi_end_io = btrfs_endio_direct_write;
6106         else
6107                 bio->bi_end_io = btrfs_endio_direct_read;
6108
6109         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6110         if (!ret)
6111                 return;
6112 free_ordered:
6113         /*
6114          * If this is a write, we need to clean up the reserved space and kill
6115          * the ordered extent.
6116          */
6117         if (write) {
6118                 struct btrfs_ordered_extent *ordered;
6119                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6120                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6121                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6122                         btrfs_free_reserved_extent(root, ordered->start,
6123                                                    ordered->disk_len);
6124                 btrfs_put_ordered_extent(ordered);
6125                 btrfs_put_ordered_extent(ordered);
6126         }
6127         bio_endio(bio, ret);
6128 }
6129
6130 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6131                         const struct iovec *iov, loff_t offset,
6132                         unsigned long nr_segs)
6133 {
6134         int seg;
6135         size_t size;
6136         unsigned long addr;
6137         unsigned blocksize_mask = root->sectorsize - 1;
6138         ssize_t retval = -EINVAL;
6139         loff_t end = offset;
6140
6141         if (offset & blocksize_mask)
6142                 goto out;
6143
6144         /* Check the memory alignment.  Blocks cannot straddle pages */
6145         for (seg = 0; seg < nr_segs; seg++) {
6146                 addr = (unsigned long)iov[seg].iov_base;
6147                 size = iov[seg].iov_len;
6148                 end += size;
6149                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6150                         goto out;
6151         }
6152         retval = 0;
6153 out:
6154         return retval;
6155 }
6156 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6157                         const struct iovec *iov, loff_t offset,
6158                         unsigned long nr_segs)
6159 {
6160         struct file *file = iocb->ki_filp;
6161         struct inode *inode = file->f_mapping->host;
6162         struct btrfs_ordered_extent *ordered;
6163         struct extent_state *cached_state = NULL;
6164         u64 lockstart, lockend;
6165         ssize_t ret;
6166         int writing = rw & WRITE;
6167         int write_bits = 0;
6168         size_t count = iov_length(iov, nr_segs);
6169
6170         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6171                             offset, nr_segs)) {
6172                 return 0;
6173         }
6174
6175         lockstart = offset;
6176         lockend = offset + count - 1;
6177
6178         if (writing) {
6179                 ret = btrfs_delalloc_reserve_space(inode, count);
6180                 if (ret)
6181                         goto out;
6182         }
6183
6184         while (1) {
6185                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186                                  0, &cached_state, GFP_NOFS);
6187                 /*
6188                  * We're concerned with the entire range that we're going to be
6189                  * doing DIO to, so we need to make sure theres no ordered
6190                  * extents in this range.
6191                  */
6192                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6193                                                      lockend - lockstart + 1);
6194                 if (!ordered)
6195                         break;
6196                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6197                                      &cached_state, GFP_NOFS);
6198                 btrfs_start_ordered_extent(inode, ordered, 1);
6199                 btrfs_put_ordered_extent(ordered);
6200                 cond_resched();
6201         }
6202
6203         /*
6204          * we don't use btrfs_set_extent_delalloc because we don't want
6205          * the dirty or uptodate bits
6206          */
6207         if (writing) {
6208                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6209                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6210                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6211                                      GFP_NOFS);
6212                 if (ret) {
6213                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6214                                          lockend, EXTENT_LOCKED | write_bits,
6215                                          1, 0, &cached_state, GFP_NOFS);
6216                         goto out;
6217                 }
6218         }
6219
6220         free_extent_state(cached_state);
6221         cached_state = NULL;
6222
6223         ret = __blockdev_direct_IO(rw, iocb, inode,
6224                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6225                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6226                    btrfs_submit_direct, 0);
6227
6228         if (ret < 0 && ret != -EIOCBQUEUED) {
6229                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6230                               offset + iov_length(iov, nr_segs) - 1,
6231                               EXTENT_LOCKED | write_bits, 1, 0,
6232                               &cached_state, GFP_NOFS);
6233         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6234                 /*
6235                  * We're falling back to buffered, unlock the section we didn't
6236                  * do IO on.
6237                  */
6238                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6239                               offset + iov_length(iov, nr_segs) - 1,
6240                               EXTENT_LOCKED | write_bits, 1, 0,
6241                               &cached_state, GFP_NOFS);
6242         }
6243 out:
6244         free_extent_state(cached_state);
6245         return ret;
6246 }
6247
6248 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6249                 __u64 start, __u64 len)
6250 {
6251         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6252 }
6253
6254 int btrfs_readpage(struct file *file, struct page *page)
6255 {
6256         struct extent_io_tree *tree;
6257         tree = &BTRFS_I(page->mapping->host)->io_tree;
6258         return extent_read_full_page(tree, page, btrfs_get_extent);
6259 }
6260
6261 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6262 {
6263         struct extent_io_tree *tree;
6264
6265
6266         if (current->flags & PF_MEMALLOC) {
6267                 redirty_page_for_writepage(wbc, page);
6268                 unlock_page(page);
6269                 return 0;
6270         }
6271         tree = &BTRFS_I(page->mapping->host)->io_tree;
6272         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6273 }
6274
6275 int btrfs_writepages(struct address_space *mapping,
6276                      struct writeback_control *wbc)
6277 {
6278         struct extent_io_tree *tree;
6279
6280         tree = &BTRFS_I(mapping->host)->io_tree;
6281         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6282 }
6283
6284 static int
6285 btrfs_readpages(struct file *file, struct address_space *mapping,
6286                 struct list_head *pages, unsigned nr_pages)
6287 {
6288         struct extent_io_tree *tree;
6289         tree = &BTRFS_I(mapping->host)->io_tree;
6290         return extent_readpages(tree, mapping, pages, nr_pages,
6291                                 btrfs_get_extent);
6292 }
6293 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6294 {
6295         struct extent_io_tree *tree;
6296         struct extent_map_tree *map;
6297         int ret;
6298
6299         tree = &BTRFS_I(page->mapping->host)->io_tree;
6300         map = &BTRFS_I(page->mapping->host)->extent_tree;
6301         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6302         if (ret == 1) {
6303                 ClearPagePrivate(page);
6304                 set_page_private(page, 0);
6305                 page_cache_release(page);
6306         }
6307         return ret;
6308 }
6309
6310 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6311 {
6312         if (PageWriteback(page) || PageDirty(page))
6313                 return 0;
6314         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6315 }
6316
6317 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6318 {
6319         struct extent_io_tree *tree;
6320         struct btrfs_ordered_extent *ordered;
6321         struct extent_state *cached_state = NULL;
6322         u64 page_start = page_offset(page);
6323         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6324
6325
6326         /*
6327          * we have the page locked, so new writeback can't start,
6328          * and the dirty bit won't be cleared while we are here.
6329          *
6330          * Wait for IO on this page so that we can safely clear
6331          * the PagePrivate2 bit and do ordered accounting
6332          */
6333         wait_on_page_writeback(page);
6334
6335         tree = &BTRFS_I(page->mapping->host)->io_tree;
6336         if (offset) {
6337                 btrfs_releasepage(page, GFP_NOFS);
6338                 return;
6339         }
6340         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6341                          GFP_NOFS);
6342         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6343                                            page_offset(page));
6344         if (ordered) {
6345                 /*
6346                  * IO on this page will never be started, so we need
6347                  * to account for any ordered extents now
6348                  */
6349                 clear_extent_bit(tree, page_start, page_end,
6350                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6351                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6352                                  &cached_state, GFP_NOFS);
6353                 /*
6354                  * whoever cleared the private bit is responsible
6355                  * for the finish_ordered_io
6356                  */
6357                 if (TestClearPagePrivate2(page)) {
6358                         btrfs_finish_ordered_io(page->mapping->host,
6359                                                 page_start, page_end);
6360                 }
6361                 btrfs_put_ordered_extent(ordered);
6362                 cached_state = NULL;
6363                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6364                                  GFP_NOFS);
6365         }
6366         clear_extent_bit(tree, page_start, page_end,
6367                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6368                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6369         __btrfs_releasepage(page, GFP_NOFS);
6370
6371         ClearPageChecked(page);
6372         if (PagePrivate(page)) {
6373                 ClearPagePrivate(page);
6374                 set_page_private(page, 0);
6375                 page_cache_release(page);
6376         }
6377 }
6378
6379 /*
6380  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6381  * called from a page fault handler when a page is first dirtied. Hence we must
6382  * be careful to check for EOF conditions here. We set the page up correctly
6383  * for a written page which means we get ENOSPC checking when writing into
6384  * holes and correct delalloc and unwritten extent mapping on filesystems that
6385  * support these features.
6386  *
6387  * We are not allowed to take the i_mutex here so we have to play games to
6388  * protect against truncate races as the page could now be beyond EOF.  Because
6389  * vmtruncate() writes the inode size before removing pages, once we have the
6390  * page lock we can determine safely if the page is beyond EOF. If it is not
6391  * beyond EOF, then the page is guaranteed safe against truncation until we
6392  * unlock the page.
6393  */
6394 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6395 {
6396         struct page *page = vmf->page;
6397         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6398         struct btrfs_root *root = BTRFS_I(inode)->root;
6399         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6400         struct btrfs_ordered_extent *ordered;
6401         struct extent_state *cached_state = NULL;
6402         char *kaddr;
6403         unsigned long zero_start;
6404         loff_t size;
6405         int ret;
6406         u64 page_start;
6407         u64 page_end;
6408
6409         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6410         if (ret) {
6411                 if (ret == -ENOMEM)
6412                         ret = VM_FAULT_OOM;
6413                 else /* -ENOSPC, -EIO, etc */
6414                         ret = VM_FAULT_SIGBUS;
6415                 goto out;
6416         }
6417
6418         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6419 again:
6420         lock_page(page);
6421         size = i_size_read(inode);
6422         page_start = page_offset(page);
6423         page_end = page_start + PAGE_CACHE_SIZE - 1;
6424
6425         if ((page->mapping != inode->i_mapping) ||
6426             (page_start >= size)) {
6427                 /* page got truncated out from underneath us */
6428                 goto out_unlock;
6429         }
6430         wait_on_page_writeback(page);
6431
6432         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6433                          GFP_NOFS);
6434         set_page_extent_mapped(page);
6435
6436         /*
6437          * we can't set the delalloc bits if there are pending ordered
6438          * extents.  Drop our locks and wait for them to finish
6439          */
6440         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6441         if (ordered) {
6442                 unlock_extent_cached(io_tree, page_start, page_end,
6443                                      &cached_state, GFP_NOFS);
6444                 unlock_page(page);
6445                 btrfs_start_ordered_extent(inode, ordered, 1);
6446                 btrfs_put_ordered_extent(ordered);
6447                 goto again;
6448         }
6449
6450         /*
6451          * XXX - page_mkwrite gets called every time the page is dirtied, even
6452          * if it was already dirty, so for space accounting reasons we need to
6453          * clear any delalloc bits for the range we are fixing to save.  There
6454          * is probably a better way to do this, but for now keep consistent with
6455          * prepare_pages in the normal write path.
6456          */
6457         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6458                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6459                           0, 0, &cached_state, GFP_NOFS);
6460
6461         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6462                                         &cached_state);
6463         if (ret) {
6464                 unlock_extent_cached(io_tree, page_start, page_end,
6465                                      &cached_state, GFP_NOFS);
6466                 ret = VM_FAULT_SIGBUS;
6467                 goto out_unlock;
6468         }
6469         ret = 0;
6470
6471         /* page is wholly or partially inside EOF */
6472         if (page_start + PAGE_CACHE_SIZE > size)
6473                 zero_start = size & ~PAGE_CACHE_MASK;
6474         else
6475                 zero_start = PAGE_CACHE_SIZE;
6476
6477         if (zero_start != PAGE_CACHE_SIZE) {
6478                 kaddr = kmap(page);
6479                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6480                 flush_dcache_page(page);
6481                 kunmap(page);
6482         }
6483         ClearPageChecked(page);
6484         set_page_dirty(page);
6485         SetPageUptodate(page);
6486
6487         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6488         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6489
6490         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6491
6492 out_unlock:
6493         if (!ret)
6494                 return VM_FAULT_LOCKED;
6495         unlock_page(page);
6496         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6497 out:
6498         return ret;
6499 }
6500
6501 static int btrfs_truncate(struct inode *inode)
6502 {
6503         struct btrfs_root *root = BTRFS_I(inode)->root;
6504         int ret;
6505         int err = 0;
6506         struct btrfs_trans_handle *trans;
6507         unsigned long nr;
6508         u64 mask = root->sectorsize - 1;
6509
6510         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6511         if (ret)
6512                 return ret;
6513
6514         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6515         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6516
6517         trans = btrfs_start_transaction(root, 5);
6518         if (IS_ERR(trans))
6519                 return PTR_ERR(trans);
6520
6521         btrfs_set_trans_block_group(trans, inode);
6522
6523         ret = btrfs_orphan_add(trans, inode);
6524         if (ret) {
6525                 btrfs_end_transaction(trans, root);
6526                 return ret;
6527         }
6528
6529         nr = trans->blocks_used;
6530         btrfs_end_transaction(trans, root);
6531         btrfs_btree_balance_dirty(root, nr);
6532
6533         /* Now start a transaction for the truncate */
6534         trans = btrfs_start_transaction(root, 0);
6535         if (IS_ERR(trans))
6536                 return PTR_ERR(trans);
6537         btrfs_set_trans_block_group(trans, inode);
6538         trans->block_rsv = root->orphan_block_rsv;
6539
6540         /*
6541          * setattr is responsible for setting the ordered_data_close flag,
6542          * but that is only tested during the last file release.  That
6543          * could happen well after the next commit, leaving a great big
6544          * window where new writes may get lost if someone chooses to write
6545          * to this file after truncating to zero
6546          *
6547          * The inode doesn't have any dirty data here, and so if we commit
6548          * this is a noop.  If someone immediately starts writing to the inode
6549          * it is very likely we'll catch some of their writes in this
6550          * transaction, and the commit will find this file on the ordered
6551          * data list with good things to send down.
6552          *
6553          * This is a best effort solution, there is still a window where
6554          * using truncate to replace the contents of the file will
6555          * end up with a zero length file after a crash.
6556          */
6557         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6558                 btrfs_add_ordered_operation(trans, root, inode);
6559
6560         while (1) {
6561                 if (!trans) {
6562                         trans = btrfs_start_transaction(root, 0);
6563                         if (IS_ERR(trans))
6564                                 return PTR_ERR(trans);
6565                         btrfs_set_trans_block_group(trans, inode);
6566                         trans->block_rsv = root->orphan_block_rsv;
6567                 }
6568
6569                 ret = btrfs_block_rsv_check(trans, root,
6570                                             root->orphan_block_rsv, 0, 5);
6571                 if (ret == -EAGAIN) {
6572                         ret = btrfs_commit_transaction(trans, root);
6573                         if (ret)
6574                                 return ret;
6575                         trans = NULL;
6576                         continue;
6577                 } else if (ret) {
6578                         err = ret;
6579                         break;
6580                 }
6581
6582                 ret = btrfs_truncate_inode_items(trans, root, inode,
6583                                                  inode->i_size,
6584                                                  BTRFS_EXTENT_DATA_KEY);
6585                 if (ret != -EAGAIN) {
6586                         err = ret;
6587                         break;
6588                 }
6589
6590                 ret = btrfs_update_inode(trans, root, inode);
6591                 if (ret) {
6592                         err = ret;
6593                         break;
6594                 }
6595
6596                 nr = trans->blocks_used;
6597                 btrfs_end_transaction(trans, root);
6598                 trans = NULL;
6599                 btrfs_btree_balance_dirty(root, nr);
6600         }
6601
6602         if (ret == 0 && inode->i_nlink > 0) {
6603                 ret = btrfs_orphan_del(trans, inode);
6604                 if (ret)
6605                         err = ret;
6606         } else if (ret && inode->i_nlink > 0) {
6607                 /*
6608                  * Failed to do the truncate, remove us from the in memory
6609                  * orphan list.
6610                  */
6611                 ret = btrfs_orphan_del(NULL, inode);
6612         }
6613
6614         ret = btrfs_update_inode(trans, root, inode);
6615         if (ret && !err)
6616                 err = ret;
6617
6618         nr = trans->blocks_used;
6619         ret = btrfs_end_transaction_throttle(trans, root);
6620         if (ret && !err)
6621                 err = ret;
6622         btrfs_btree_balance_dirty(root, nr);
6623
6624         return err;
6625 }
6626
6627 /*
6628  * create a new subvolume directory/inode (helper for the ioctl).
6629  */
6630 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6631                              struct btrfs_root *new_root,
6632                              u64 new_dirid, u64 alloc_hint)
6633 {
6634         struct inode *inode;
6635         int err;
6636         u64 index = 0;
6637
6638         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6639                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6640         if (IS_ERR(inode))
6641                 return PTR_ERR(inode);
6642         inode->i_op = &btrfs_dir_inode_operations;
6643         inode->i_fop = &btrfs_dir_file_operations;
6644
6645         inode->i_nlink = 1;
6646         btrfs_i_size_write(inode, 0);
6647
6648         err = btrfs_update_inode(trans, new_root, inode);
6649         BUG_ON(err);
6650
6651         iput(inode);
6652         return 0;
6653 }
6654
6655 /* helper function for file defrag and space balancing.  This
6656  * forces readahead on a given range of bytes in an inode
6657  */
6658 unsigned long btrfs_force_ra(struct address_space *mapping,
6659                               struct file_ra_state *ra, struct file *file,
6660                               pgoff_t offset, pgoff_t last_index)
6661 {
6662         pgoff_t req_size = last_index - offset + 1;
6663
6664         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6665         return offset + req_size;
6666 }
6667
6668 struct inode *btrfs_alloc_inode(struct super_block *sb)
6669 {
6670         struct btrfs_inode *ei;
6671         struct inode *inode;
6672
6673         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6674         if (!ei)
6675                 return NULL;
6676
6677         ei->root = NULL;
6678         ei->space_info = NULL;
6679         ei->generation = 0;
6680         ei->sequence = 0;
6681         ei->last_trans = 0;
6682         ei->last_sub_trans = 0;
6683         ei->logged_trans = 0;
6684         ei->delalloc_bytes = 0;
6685         ei->reserved_bytes = 0;
6686         ei->disk_i_size = 0;
6687         ei->flags = 0;
6688         ei->index_cnt = (u64)-1;
6689         ei->last_unlink_trans = 0;
6690
6691         atomic_set(&ei->outstanding_extents, 0);
6692         atomic_set(&ei->reserved_extents, 0);
6693
6694         ei->ordered_data_close = 0;
6695         ei->orphan_meta_reserved = 0;
6696         ei->dummy_inode = 0;
6697         ei->force_compress = BTRFS_COMPRESS_NONE;
6698
6699         inode = &ei->vfs_inode;
6700         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6701         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6702         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6703         mutex_init(&ei->log_mutex);
6704         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6705         INIT_LIST_HEAD(&ei->i_orphan);
6706         INIT_LIST_HEAD(&ei->delalloc_inodes);
6707         INIT_LIST_HEAD(&ei->ordered_operations);
6708         RB_CLEAR_NODE(&ei->rb_node);
6709
6710         return inode;
6711 }
6712
6713 static void btrfs_i_callback(struct rcu_head *head)
6714 {
6715         struct inode *inode = container_of(head, struct inode, i_rcu);
6716         INIT_LIST_HEAD(&inode->i_dentry);
6717         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6718 }
6719
6720 void btrfs_destroy_inode(struct inode *inode)
6721 {
6722         struct btrfs_ordered_extent *ordered;
6723         struct btrfs_root *root = BTRFS_I(inode)->root;
6724
6725         WARN_ON(!list_empty(&inode->i_dentry));
6726         WARN_ON(inode->i_data.nrpages);
6727         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6728         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6729
6730         /*
6731          * This can happen where we create an inode, but somebody else also
6732          * created the same inode and we need to destroy the one we already
6733          * created.
6734          */
6735         if (!root)
6736                 goto free;
6737
6738         /*
6739          * Make sure we're properly removed from the ordered operation
6740          * lists.
6741          */
6742         smp_mb();
6743         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6744                 spin_lock(&root->fs_info->ordered_extent_lock);
6745                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6746                 spin_unlock(&root->fs_info->ordered_extent_lock);
6747         }
6748
6749         if (root == root->fs_info->tree_root) {
6750                 struct btrfs_block_group_cache *block_group;
6751
6752                 block_group = btrfs_lookup_block_group(root->fs_info,
6753                                                 BTRFS_I(inode)->block_group);
6754                 if (block_group && block_group->inode == inode) {
6755                         spin_lock(&block_group->lock);
6756                         block_group->inode = NULL;
6757                         spin_unlock(&block_group->lock);
6758                         btrfs_put_block_group(block_group);
6759                 } else if (block_group) {
6760                         btrfs_put_block_group(block_group);
6761                 }
6762         }
6763
6764         spin_lock(&root->orphan_lock);
6765         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6766                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6767                        inode->i_ino);
6768                 list_del_init(&BTRFS_I(inode)->i_orphan);
6769         }
6770         spin_unlock(&root->orphan_lock);
6771
6772         while (1) {
6773                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6774                 if (!ordered)
6775                         break;
6776                 else {
6777                         printk(KERN_ERR "btrfs found ordered "
6778                                "extent %llu %llu on inode cleanup\n",
6779                                (unsigned long long)ordered->file_offset,
6780                                (unsigned long long)ordered->len);
6781                         btrfs_remove_ordered_extent(inode, ordered);
6782                         btrfs_put_ordered_extent(ordered);
6783                         btrfs_put_ordered_extent(ordered);
6784                 }
6785         }
6786         inode_tree_del(inode);
6787         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6788 free:
6789         call_rcu(&inode->i_rcu, btrfs_i_callback);
6790 }
6791
6792 int btrfs_drop_inode(struct inode *inode)
6793 {
6794         struct btrfs_root *root = BTRFS_I(inode)->root;
6795
6796         if (btrfs_root_refs(&root->root_item) == 0 &&
6797             root != root->fs_info->tree_root)
6798                 return 1;
6799         else
6800                 return generic_drop_inode(inode);
6801 }
6802
6803 static void init_once(void *foo)
6804 {
6805         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6806
6807         inode_init_once(&ei->vfs_inode);
6808 }
6809
6810 void btrfs_destroy_cachep(void)
6811 {
6812         if (btrfs_inode_cachep)
6813                 kmem_cache_destroy(btrfs_inode_cachep);
6814         if (btrfs_trans_handle_cachep)
6815                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6816         if (btrfs_transaction_cachep)
6817                 kmem_cache_destroy(btrfs_transaction_cachep);
6818         if (btrfs_path_cachep)
6819                 kmem_cache_destroy(btrfs_path_cachep);
6820         if (btrfs_free_space_cachep)
6821                 kmem_cache_destroy(btrfs_free_space_cachep);
6822 }
6823
6824 int btrfs_init_cachep(void)
6825 {
6826         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6827                         sizeof(struct btrfs_inode), 0,
6828                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6829         if (!btrfs_inode_cachep)
6830                 goto fail;
6831
6832         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6833                         sizeof(struct btrfs_trans_handle), 0,
6834                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6835         if (!btrfs_trans_handle_cachep)
6836                 goto fail;
6837
6838         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6839                         sizeof(struct btrfs_transaction), 0,
6840                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6841         if (!btrfs_transaction_cachep)
6842                 goto fail;
6843
6844         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6845                         sizeof(struct btrfs_path), 0,
6846                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6847         if (!btrfs_path_cachep)
6848                 goto fail;
6849
6850         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6851                         sizeof(struct btrfs_free_space), 0,
6852                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6853         if (!btrfs_free_space_cachep)
6854                 goto fail;
6855
6856         return 0;
6857 fail:
6858         btrfs_destroy_cachep();
6859         return -ENOMEM;
6860 }
6861
6862 static int btrfs_getattr(struct vfsmount *mnt,
6863                          struct dentry *dentry, struct kstat *stat)
6864 {
6865         struct inode *inode = dentry->d_inode;
6866         generic_fillattr(inode, stat);
6867         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6868         stat->blksize = PAGE_CACHE_SIZE;
6869         stat->blocks = (inode_get_bytes(inode) +
6870                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6871         return 0;
6872 }
6873
6874 /*
6875  * If a file is moved, it will inherit the cow and compression flags of the new
6876  * directory.
6877  */
6878 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6879 {
6880         struct btrfs_inode *b_dir = BTRFS_I(dir);
6881         struct btrfs_inode *b_inode = BTRFS_I(inode);
6882
6883         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6884                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6885         else
6886                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6887
6888         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6889                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6890         else
6891                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6892 }
6893
6894 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6895                            struct inode *new_dir, struct dentry *new_dentry)
6896 {
6897         struct btrfs_trans_handle *trans;
6898         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6899         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6900         struct inode *new_inode = new_dentry->d_inode;
6901         struct inode *old_inode = old_dentry->d_inode;
6902         struct timespec ctime = CURRENT_TIME;
6903         u64 index = 0;
6904         u64 root_objectid;
6905         int ret;
6906
6907         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6908                 return -EPERM;
6909
6910         /* we only allow rename subvolume link between subvolumes */
6911         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6912                 return -EXDEV;
6913
6914         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6915             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6916                 return -ENOTEMPTY;
6917
6918         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6919             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6920                 return -ENOTEMPTY;
6921         /*
6922          * we're using rename to replace one file with another.
6923          * and the replacement file is large.  Start IO on it now so
6924          * we don't add too much work to the end of the transaction
6925          */
6926         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6927             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6928                 filemap_flush(old_inode->i_mapping);
6929
6930         /* close the racy window with snapshot create/destroy ioctl */
6931         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6932                 down_read(&root->fs_info->subvol_sem);
6933         /*
6934          * We want to reserve the absolute worst case amount of items.  So if
6935          * both inodes are subvols and we need to unlink them then that would
6936          * require 4 item modifications, but if they are both normal inodes it
6937          * would require 5 item modifications, so we'll assume their normal
6938          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6939          * should cover the worst case number of items we'll modify.
6940          */
6941         trans = btrfs_start_transaction(root, 20);
6942         if (IS_ERR(trans))
6943                 return PTR_ERR(trans);
6944
6945         btrfs_set_trans_block_group(trans, new_dir);
6946
6947         if (dest != root)
6948                 btrfs_record_root_in_trans(trans, dest);
6949
6950         ret = btrfs_set_inode_index(new_dir, &index);
6951         if (ret)
6952                 goto out_fail;
6953
6954         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6955                 /* force full log commit if subvolume involved. */
6956                 root->fs_info->last_trans_log_full_commit = trans->transid;
6957         } else {
6958                 ret = btrfs_insert_inode_ref(trans, dest,
6959                                              new_dentry->d_name.name,
6960                                              new_dentry->d_name.len,
6961                                              old_inode->i_ino,
6962                                              new_dir->i_ino, index);
6963                 if (ret)
6964                         goto out_fail;
6965                 /*
6966                  * this is an ugly little race, but the rename is required
6967                  * to make sure that if we crash, the inode is either at the
6968                  * old name or the new one.  pinning the log transaction lets
6969                  * us make sure we don't allow a log commit to come in after
6970                  * we unlink the name but before we add the new name back in.
6971                  */
6972                 btrfs_pin_log_trans(root);
6973         }
6974         /*
6975          * make sure the inode gets flushed if it is replacing
6976          * something.
6977          */
6978         if (new_inode && new_inode->i_size &&
6979             old_inode && S_ISREG(old_inode->i_mode)) {
6980                 btrfs_add_ordered_operation(trans, root, old_inode);
6981         }
6982
6983         old_dir->i_ctime = old_dir->i_mtime = ctime;
6984         new_dir->i_ctime = new_dir->i_mtime = ctime;
6985         old_inode->i_ctime = ctime;
6986
6987         if (old_dentry->d_parent != new_dentry->d_parent)
6988                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6989
6990         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6991                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6992                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6993                                         old_dentry->d_name.name,
6994                                         old_dentry->d_name.len);
6995         } else {
6996                 btrfs_inc_nlink(old_dentry->d_inode);
6997                 ret = btrfs_unlink_inode(trans, root, old_dir,
6998                                          old_dentry->d_inode,
6999                                          old_dentry->d_name.name,
7000                                          old_dentry->d_name.len);
7001         }
7002         BUG_ON(ret);
7003
7004         if (new_inode) {
7005                 new_inode->i_ctime = CURRENT_TIME;
7006                 if (unlikely(new_inode->i_ino ==
7007                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7008                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7009                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7010                                                 root_objectid,
7011                                                 new_dentry->d_name.name,
7012                                                 new_dentry->d_name.len);
7013                         BUG_ON(new_inode->i_nlink == 0);
7014                 } else {
7015                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7016                                                  new_dentry->d_inode,
7017                                                  new_dentry->d_name.name,
7018                                                  new_dentry->d_name.len);
7019                 }
7020                 BUG_ON(ret);
7021                 if (new_inode->i_nlink == 0) {
7022                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7023                         BUG_ON(ret);
7024                 }
7025         }
7026
7027         fixup_inode_flags(new_dir, old_inode);
7028
7029         ret = btrfs_add_link(trans, new_dir, old_inode,
7030                              new_dentry->d_name.name,
7031                              new_dentry->d_name.len, 0, index);
7032         BUG_ON(ret);
7033
7034         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7035                 struct dentry *parent = dget_parent(new_dentry);
7036                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7037                 dput(parent);
7038                 btrfs_end_log_trans(root);
7039         }
7040 out_fail:
7041         btrfs_end_transaction_throttle(trans, root);
7042
7043         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7044                 up_read(&root->fs_info->subvol_sem);
7045
7046         return ret;
7047 }
7048
7049 /*
7050  * some fairly slow code that needs optimization. This walks the list
7051  * of all the inodes with pending delalloc and forces them to disk.
7052  */
7053 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7054 {
7055         struct list_head *head = &root->fs_info->delalloc_inodes;
7056         struct btrfs_inode *binode;
7057         struct inode *inode;
7058
7059         if (root->fs_info->sb->s_flags & MS_RDONLY)
7060                 return -EROFS;
7061
7062         spin_lock(&root->fs_info->delalloc_lock);
7063         while (!list_empty(head)) {
7064                 binode = list_entry(head->next, struct btrfs_inode,
7065                                     delalloc_inodes);
7066                 inode = igrab(&binode->vfs_inode);
7067                 if (!inode)
7068                         list_del_init(&binode->delalloc_inodes);
7069                 spin_unlock(&root->fs_info->delalloc_lock);
7070                 if (inode) {
7071                         filemap_flush(inode->i_mapping);
7072                         if (delay_iput)
7073                                 btrfs_add_delayed_iput(inode);
7074                         else
7075                                 iput(inode);
7076                 }
7077                 cond_resched();
7078                 spin_lock(&root->fs_info->delalloc_lock);
7079         }
7080         spin_unlock(&root->fs_info->delalloc_lock);
7081
7082         /* the filemap_flush will queue IO into the worker threads, but
7083          * we have to make sure the IO is actually started and that
7084          * ordered extents get created before we return
7085          */
7086         atomic_inc(&root->fs_info->async_submit_draining);
7087         while (atomic_read(&root->fs_info->nr_async_submits) ||
7088               atomic_read(&root->fs_info->async_delalloc_pages)) {
7089                 wait_event(root->fs_info->async_submit_wait,
7090                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7091                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7092         }
7093         atomic_dec(&root->fs_info->async_submit_draining);
7094         return 0;
7095 }
7096
7097 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7098                                    int sync)
7099 {
7100         struct btrfs_inode *binode;
7101         struct inode *inode = NULL;
7102
7103         spin_lock(&root->fs_info->delalloc_lock);
7104         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7105                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7106                                     struct btrfs_inode, delalloc_inodes);
7107                 inode = igrab(&binode->vfs_inode);
7108                 if (inode) {
7109                         list_move_tail(&binode->delalloc_inodes,
7110                                        &root->fs_info->delalloc_inodes);
7111                         break;
7112                 }
7113
7114                 list_del_init(&binode->delalloc_inodes);
7115                 cond_resched_lock(&root->fs_info->delalloc_lock);
7116         }
7117         spin_unlock(&root->fs_info->delalloc_lock);
7118
7119         if (inode) {
7120                 if (sync) {
7121                         filemap_write_and_wait(inode->i_mapping);
7122                         /*
7123                          * We have to do this because compression doesn't
7124                          * actually set PG_writeback until it submits the pages
7125                          * for IO, which happens in an async thread, so we could
7126                          * race and not actually wait for any writeback pages
7127                          * because they've not been submitted yet.  Technically
7128                          * this could still be the case for the ordered stuff
7129                          * since the async thread may not have started to do its
7130                          * work yet.  If this becomes the case then we need to
7131                          * figure out a way to make sure that in writepage we
7132                          * wait for any async pages to be submitted before
7133                          * returning so that fdatawait does what its supposed to
7134                          * do.
7135                          */
7136                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7137                 } else {
7138                         filemap_flush(inode->i_mapping);
7139                 }
7140                 if (delay_iput)
7141                         btrfs_add_delayed_iput(inode);
7142                 else
7143                         iput(inode);
7144                 return 1;
7145         }
7146         return 0;
7147 }
7148
7149 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7150                          const char *symname)
7151 {
7152         struct btrfs_trans_handle *trans;
7153         struct btrfs_root *root = BTRFS_I(dir)->root;
7154         struct btrfs_path *path;
7155         struct btrfs_key key;
7156         struct inode *inode = NULL;
7157         int err;
7158         int drop_inode = 0;
7159         u64 objectid;
7160         u64 index = 0 ;
7161         int name_len;
7162         int datasize;
7163         unsigned long ptr;
7164         struct btrfs_file_extent_item *ei;
7165         struct extent_buffer *leaf;
7166         unsigned long nr = 0;
7167
7168         name_len = strlen(symname) + 1;
7169         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7170                 return -ENAMETOOLONG;
7171
7172         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7173         if (err)
7174                 return err;
7175         /*
7176          * 2 items for inode item and ref
7177          * 2 items for dir items
7178          * 1 item for xattr if selinux is on
7179          */
7180         trans = btrfs_start_transaction(root, 5);
7181         if (IS_ERR(trans))
7182                 return PTR_ERR(trans);
7183
7184         btrfs_set_trans_block_group(trans, dir);
7185
7186         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7187                                 dentry->d_name.len, dir->i_ino, objectid,
7188                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7189                                 &index);
7190         err = PTR_ERR(inode);
7191         if (IS_ERR(inode))
7192                 goto out_unlock;
7193
7194         err = btrfs_init_inode_security(trans, inode, dir);
7195         if (err) {
7196                 drop_inode = 1;
7197                 goto out_unlock;
7198         }
7199
7200         btrfs_set_trans_block_group(trans, inode);
7201         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7202         if (err)
7203                 drop_inode = 1;
7204         else {
7205                 inode->i_mapping->a_ops = &btrfs_aops;
7206                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7207                 inode->i_fop = &btrfs_file_operations;
7208                 inode->i_op = &btrfs_file_inode_operations;
7209                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7210         }
7211         btrfs_update_inode_block_group(trans, inode);
7212         btrfs_update_inode_block_group(trans, dir);
7213         if (drop_inode)
7214                 goto out_unlock;
7215
7216         path = btrfs_alloc_path();
7217         BUG_ON(!path);
7218         key.objectid = inode->i_ino;
7219         key.offset = 0;
7220         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7221         datasize = btrfs_file_extent_calc_inline_size(name_len);
7222         err = btrfs_insert_empty_item(trans, root, path, &key,
7223                                       datasize);
7224         if (err) {
7225                 drop_inode = 1;
7226                 goto out_unlock;
7227         }
7228         leaf = path->nodes[0];
7229         ei = btrfs_item_ptr(leaf, path->slots[0],
7230                             struct btrfs_file_extent_item);
7231         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7232         btrfs_set_file_extent_type(leaf, ei,
7233                                    BTRFS_FILE_EXTENT_INLINE);
7234         btrfs_set_file_extent_encryption(leaf, ei, 0);
7235         btrfs_set_file_extent_compression(leaf, ei, 0);
7236         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7237         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7238
7239         ptr = btrfs_file_extent_inline_start(ei);
7240         write_extent_buffer(leaf, symname, ptr, name_len);
7241         btrfs_mark_buffer_dirty(leaf);
7242         btrfs_free_path(path);
7243
7244         inode->i_op = &btrfs_symlink_inode_operations;
7245         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7246         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7247         inode_set_bytes(inode, name_len);
7248         btrfs_i_size_write(inode, name_len - 1);
7249         err = btrfs_update_inode(trans, root, inode);
7250         if (err)
7251                 drop_inode = 1;
7252
7253 out_unlock:
7254         nr = trans->blocks_used;
7255         btrfs_end_transaction_throttle(trans, root);
7256         if (drop_inode) {
7257                 inode_dec_link_count(inode);
7258                 iput(inode);
7259         }
7260         btrfs_btree_balance_dirty(root, nr);
7261         return err;
7262 }
7263
7264 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7265                                        u64 start, u64 num_bytes, u64 min_size,
7266                                        loff_t actual_len, u64 *alloc_hint,
7267                                        struct btrfs_trans_handle *trans)
7268 {
7269         struct btrfs_root *root = BTRFS_I(inode)->root;
7270         struct btrfs_key ins;
7271         u64 cur_offset = start;
7272         u64 i_size;
7273         int ret = 0;
7274         bool own_trans = true;
7275
7276         if (trans)
7277                 own_trans = false;
7278         while (num_bytes > 0) {
7279                 if (own_trans) {
7280                         trans = btrfs_start_transaction(root, 3);
7281                         if (IS_ERR(trans)) {
7282                                 ret = PTR_ERR(trans);
7283                                 break;
7284                         }
7285                 }
7286
7287                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7288                                            0, *alloc_hint, (u64)-1, &ins, 1);
7289                 if (ret) {
7290                         if (own_trans)
7291                                 btrfs_end_transaction(trans, root);
7292                         break;
7293                 }
7294
7295                 ret = insert_reserved_file_extent(trans, inode,
7296                                                   cur_offset, ins.objectid,
7297                                                   ins.offset, ins.offset,
7298                                                   ins.offset, 0, 0, 0,
7299                                                   BTRFS_FILE_EXTENT_PREALLOC);
7300                 BUG_ON(ret);
7301                 btrfs_drop_extent_cache(inode, cur_offset,
7302                                         cur_offset + ins.offset -1, 0);
7303
7304                 num_bytes -= ins.offset;
7305                 cur_offset += ins.offset;
7306                 *alloc_hint = ins.objectid + ins.offset;
7307
7308                 inode->i_ctime = CURRENT_TIME;
7309                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7310                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7311                     (actual_len > inode->i_size) &&
7312                     (cur_offset > inode->i_size)) {
7313                         if (cur_offset > actual_len)
7314                                 i_size = actual_len;
7315                         else
7316                                 i_size = cur_offset;
7317                         i_size_write(inode, i_size);
7318                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7319                 }
7320
7321                 ret = btrfs_update_inode(trans, root, inode);
7322                 BUG_ON(ret);
7323
7324                 if (own_trans)
7325                         btrfs_end_transaction(trans, root);
7326         }
7327         return ret;
7328 }
7329
7330 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7331                               u64 start, u64 num_bytes, u64 min_size,
7332                               loff_t actual_len, u64 *alloc_hint)
7333 {
7334         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7335                                            min_size, actual_len, alloc_hint,
7336                                            NULL);
7337 }
7338
7339 int btrfs_prealloc_file_range_trans(struct inode *inode,
7340                                     struct btrfs_trans_handle *trans, int mode,
7341                                     u64 start, u64 num_bytes, u64 min_size,
7342                                     loff_t actual_len, u64 *alloc_hint)
7343 {
7344         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7345                                            min_size, actual_len, alloc_hint, trans);
7346 }
7347
7348 static int btrfs_set_page_dirty(struct page *page)
7349 {
7350         return __set_page_dirty_nobuffers(page);
7351 }
7352
7353 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7354 {
7355         struct btrfs_root *root = BTRFS_I(inode)->root;
7356
7357         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7358                 return -EROFS;
7359         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7360                 return -EACCES;
7361         return generic_permission(inode, mask, flags, btrfs_check_acl);
7362 }
7363
7364 static const struct inode_operations btrfs_dir_inode_operations = {
7365         .getattr        = btrfs_getattr,
7366         .lookup         = btrfs_lookup,
7367         .create         = btrfs_create,
7368         .unlink         = btrfs_unlink,
7369         .link           = btrfs_link,
7370         .mkdir          = btrfs_mkdir,
7371         .rmdir          = btrfs_rmdir,
7372         .rename         = btrfs_rename,
7373         .symlink        = btrfs_symlink,
7374         .setattr        = btrfs_setattr,
7375         .mknod          = btrfs_mknod,
7376         .setxattr       = btrfs_setxattr,
7377         .getxattr       = btrfs_getxattr,
7378         .listxattr      = btrfs_listxattr,
7379         .removexattr    = btrfs_removexattr,
7380         .permission     = btrfs_permission,
7381 };
7382 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7383         .lookup         = btrfs_lookup,
7384         .permission     = btrfs_permission,
7385 };
7386
7387 static const struct file_operations btrfs_dir_file_operations = {
7388         .llseek         = generic_file_llseek,
7389         .read           = generic_read_dir,
7390         .readdir        = btrfs_real_readdir,
7391         .unlocked_ioctl = btrfs_ioctl,
7392 #ifdef CONFIG_COMPAT
7393         .compat_ioctl   = btrfs_ioctl,
7394 #endif
7395         .release        = btrfs_release_file,
7396         .fsync          = btrfs_sync_file,
7397 };
7398
7399 static struct extent_io_ops btrfs_extent_io_ops = {
7400         .fill_delalloc = run_delalloc_range,
7401         .submit_bio_hook = btrfs_submit_bio_hook,
7402         .merge_bio_hook = btrfs_merge_bio_hook,
7403         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7404         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7405         .writepage_start_hook = btrfs_writepage_start_hook,
7406         .readpage_io_failed_hook = btrfs_io_failed_hook,
7407         .set_bit_hook = btrfs_set_bit_hook,
7408         .clear_bit_hook = btrfs_clear_bit_hook,
7409         .merge_extent_hook = btrfs_merge_extent_hook,
7410         .split_extent_hook = btrfs_split_extent_hook,
7411 };
7412
7413 /*
7414  * btrfs doesn't support the bmap operation because swapfiles
7415  * use bmap to make a mapping of extents in the file.  They assume
7416  * these extents won't change over the life of the file and they
7417  * use the bmap result to do IO directly to the drive.
7418  *
7419  * the btrfs bmap call would return logical addresses that aren't
7420  * suitable for IO and they also will change frequently as COW
7421  * operations happen.  So, swapfile + btrfs == corruption.
7422  *
7423  * For now we're avoiding this by dropping bmap.
7424  */
7425 static const struct address_space_operations btrfs_aops = {
7426         .readpage       = btrfs_readpage,
7427         .writepage      = btrfs_writepage,
7428         .writepages     = btrfs_writepages,
7429         .readpages      = btrfs_readpages,
7430         .sync_page      = block_sync_page,
7431         .direct_IO      = btrfs_direct_IO,
7432         .invalidatepage = btrfs_invalidatepage,
7433         .releasepage    = btrfs_releasepage,
7434         .set_page_dirty = btrfs_set_page_dirty,
7435         .error_remove_page = generic_error_remove_page,
7436 };
7437
7438 static const struct address_space_operations btrfs_symlink_aops = {
7439         .readpage       = btrfs_readpage,
7440         .writepage      = btrfs_writepage,
7441         .invalidatepage = btrfs_invalidatepage,
7442         .releasepage    = btrfs_releasepage,
7443 };
7444
7445 static const struct inode_operations btrfs_file_inode_operations = {
7446         .getattr        = btrfs_getattr,
7447         .setattr        = btrfs_setattr,
7448         .setxattr       = btrfs_setxattr,
7449         .getxattr       = btrfs_getxattr,
7450         .listxattr      = btrfs_listxattr,
7451         .removexattr    = btrfs_removexattr,
7452         .permission     = btrfs_permission,
7453         .fiemap         = btrfs_fiemap,
7454 };
7455 static const struct inode_operations btrfs_special_inode_operations = {
7456         .getattr        = btrfs_getattr,
7457         .setattr        = btrfs_setattr,
7458         .permission     = btrfs_permission,
7459         .setxattr       = btrfs_setxattr,
7460         .getxattr       = btrfs_getxattr,
7461         .listxattr      = btrfs_listxattr,
7462         .removexattr    = btrfs_removexattr,
7463 };
7464 static const struct inode_operations btrfs_symlink_inode_operations = {
7465         .readlink       = generic_readlink,
7466         .follow_link    = page_follow_link_light,
7467         .put_link       = page_put_link,
7468         .getattr        = btrfs_getattr,
7469         .permission     = btrfs_permission,
7470         .setxattr       = btrfs_setxattr,
7471         .getxattr       = btrfs_getxattr,
7472         .listxattr      = btrfs_listxattr,
7473         .removexattr    = btrfs_removexattr,
7474 };
7475
7476 const struct dentry_operations btrfs_dentry_operations = {
7477         .d_delete       = btrfs_dentry_delete,
7478 };