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                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1472                         if (ret)
1473                                 return ret;
1474                 }
1475                 goto mapit;
1476         } else if (!skip_sum) {
1477                 /* csum items have already been cloned */
1478                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1479                         goto mapit;
1480                 /* we're doing a write, do the async checksumming */
1481                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1482                                    inode, rw, bio, mirror_num,
1483                                    bio_flags, bio_offset,
1484                                    __btrfs_submit_bio_start,
1485                                    __btrfs_submit_bio_done);
1486         }
1487
1488 mapit:
1489         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1490 }
1491
1492 /*
1493  * given a list of ordered sums record them in the inode.  This happens
1494  * at IO completion time based on sums calculated at bio submission time.
1495  */
1496 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1497                              struct inode *inode, u64 file_offset,
1498                              struct list_head *list)
1499 {
1500         struct btrfs_ordered_sum *sum;
1501
1502         btrfs_set_trans_block_group(trans, inode);
1503
1504         list_for_each_entry(sum, list, list) {
1505                 btrfs_csum_file_blocks(trans,
1506                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1507         }
1508         return 0;
1509 }
1510
1511 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1512                               struct extent_state **cached_state)
1513 {
1514         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1515                 WARN_ON(1);
1516         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1517                                    cached_state, GFP_NOFS);
1518 }
1519
1520 /* see btrfs_writepage_start_hook for details on why this is required */
1521 struct btrfs_writepage_fixup {
1522         struct page *page;
1523         struct btrfs_work work;
1524 };
1525
1526 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1527 {
1528         struct btrfs_writepage_fixup *fixup;
1529         struct btrfs_ordered_extent *ordered;
1530         struct extent_state *cached_state = NULL;
1531         struct page *page;
1532         struct inode *inode;
1533         u64 page_start;
1534         u64 page_end;
1535
1536         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1537         page = fixup->page;
1538 again:
1539         lock_page(page);
1540         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1541                 ClearPageChecked(page);
1542                 goto out_page;
1543         }
1544
1545         inode = page->mapping->host;
1546         page_start = page_offset(page);
1547         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1548
1549         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1550                          &cached_state, GFP_NOFS);
1551
1552         /* already ordered? We're done */
1553         if (PagePrivate2(page))
1554                 goto out;
1555
1556         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1557         if (ordered) {
1558                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1559                                      page_end, &cached_state, GFP_NOFS);
1560                 unlock_page(page);
1561                 btrfs_start_ordered_extent(inode, ordered, 1);
1562                 goto again;
1563         }
1564
1565         BUG();
1566         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1567         ClearPageChecked(page);
1568 out:
1569         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1570                              &cached_state, GFP_NOFS);
1571 out_page:
1572         unlock_page(page);
1573         page_cache_release(page);
1574         kfree(fixup);
1575 }
1576
1577 /*
1578  * There are a few paths in the higher layers of the kernel that directly
1579  * set the page dirty bit without asking the filesystem if it is a
1580  * good idea.  This causes problems because we want to make sure COW
1581  * properly happens and the data=ordered rules are followed.
1582  *
1583  * In our case any range that doesn't have the ORDERED bit set
1584  * hasn't been properly setup for IO.  We kick off an async process
1585  * to fix it up.  The async helper will wait for ordered extents, set
1586  * the delalloc bit and make it safe to write the page.
1587  */
1588 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1589 {
1590         struct inode *inode = page->mapping->host;
1591         struct btrfs_writepage_fixup *fixup;
1592         struct btrfs_root *root = BTRFS_I(inode)->root;
1593
1594         /* this page is properly in the ordered list */
1595         if (TestClearPagePrivate2(page))
1596                 return 0;
1597
1598         if (PageChecked(page))
1599                 return -EAGAIN;
1600
1601         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1602         if (!fixup)
1603                 return -EAGAIN;
1604
1605         SetPageChecked(page);
1606         page_cache_get(page);
1607         fixup->work.func = btrfs_writepage_fixup_worker;
1608         fixup->page = page;
1609         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1610         return -EAGAIN;
1611 }
1612
1613 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1614                                        struct inode *inode, u64 file_pos,
1615                                        u64 disk_bytenr, u64 disk_num_bytes,
1616                                        u64 num_bytes, u64 ram_bytes,
1617                                        u8 compression, u8 encryption,
1618                                        u16 other_encoding, int extent_type)
1619 {
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621         struct btrfs_file_extent_item *fi;
1622         struct btrfs_path *path;
1623         struct extent_buffer *leaf;
1624         struct btrfs_key ins;
1625         u64 hint;
1626         int ret;
1627
1628         path = btrfs_alloc_path();
1629         BUG_ON(!path);
1630
1631         path->leave_spinning = 1;
1632
1633         /*
1634          * we may be replacing one extent in the tree with another.
1635          * The new extent is pinned in the extent map, and we don't want
1636          * to drop it from the cache until it is completely in the btree.
1637          *
1638          * So, tell btrfs_drop_extents to leave this extent in the cache.
1639          * the caller is expected to unpin it and allow it to be merged
1640          * with the others.
1641          */
1642         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1643                                  &hint, 0);
1644         BUG_ON(ret);
1645
1646         ins.objectid = inode->i_ino;
1647         ins.offset = file_pos;
1648         ins.type = BTRFS_EXTENT_DATA_KEY;
1649         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1650         BUG_ON(ret);
1651         leaf = path->nodes[0];
1652         fi = btrfs_item_ptr(leaf, path->slots[0],
1653                             struct btrfs_file_extent_item);
1654         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1655         btrfs_set_file_extent_type(leaf, fi, extent_type);
1656         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1657         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1658         btrfs_set_file_extent_offset(leaf, fi, 0);
1659         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1660         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1661         btrfs_set_file_extent_compression(leaf, fi, compression);
1662         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1663         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1664
1665         btrfs_unlock_up_safe(path, 1);
1666         btrfs_set_lock_blocking(leaf);
1667
1668         btrfs_mark_buffer_dirty(leaf);
1669
1670         inode_add_bytes(inode, num_bytes);
1671
1672         ins.objectid = disk_bytenr;
1673         ins.offset = disk_num_bytes;
1674         ins.type = BTRFS_EXTENT_ITEM_KEY;
1675         ret = btrfs_alloc_reserved_file_extent(trans, root,
1676                                         root->root_key.objectid,
1677                                         inode->i_ino, file_pos, &ins);
1678         BUG_ON(ret);
1679         btrfs_free_path(path);
1680
1681         return 0;
1682 }
1683
1684 /*
1685  * helper function for btrfs_finish_ordered_io, this
1686  * just reads in some of the csum leaves to prime them into ram
1687  * before we start the transaction.  It limits the amount of btree
1688  * reads required while inside the transaction.
1689  */
1690 /* as ordered data IO finishes, this gets called so we can finish
1691  * an ordered extent if the range of bytes in the file it covers are
1692  * fully written.
1693  */
1694 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1695 {
1696         struct btrfs_root *root = BTRFS_I(inode)->root;
1697         struct btrfs_trans_handle *trans = NULL;
1698         struct btrfs_ordered_extent *ordered_extent = NULL;
1699         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1700         struct extent_state *cached_state = NULL;
1701         int compress_type = 0;
1702         int ret;
1703         bool nolock = false;
1704
1705         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1706                                              end - start + 1);
1707         if (!ret)
1708                 return 0;
1709         BUG_ON(!ordered_extent);
1710
1711         nolock = (root == root->fs_info->tree_root);
1712
1713         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1714                 BUG_ON(!list_empty(&ordered_extent->list));
1715                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1716                 if (!ret) {
1717                         if (nolock)
1718                                 trans = btrfs_join_transaction_nolock(root, 1);
1719                         else
1720                                 trans = btrfs_join_transaction(root, 1);
1721                         BUG_ON(IS_ERR(trans));
1722                         btrfs_set_trans_block_group(trans, inode);
1723                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1724                         ret = btrfs_update_inode(trans, root, inode);
1725                         BUG_ON(ret);
1726                 }
1727                 goto out;
1728         }
1729
1730         lock_extent_bits(io_tree, ordered_extent->file_offset,
1731                          ordered_extent->file_offset + ordered_extent->len - 1,
1732                          0, &cached_state, GFP_NOFS);
1733
1734         if (nolock)
1735                 trans = btrfs_join_transaction_nolock(root, 1);
1736         else
1737                 trans = btrfs_join_transaction(root, 1);
1738         BUG_ON(IS_ERR(trans));
1739         btrfs_set_trans_block_group(trans, inode);
1740         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1741
1742         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1743                 compress_type = ordered_extent->compress_type;
1744         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1745                 BUG_ON(compress_type);
1746                 ret = btrfs_mark_extent_written(trans, inode,
1747                                                 ordered_extent->file_offset,
1748                                                 ordered_extent->file_offset +
1749                                                 ordered_extent->len);
1750                 BUG_ON(ret);
1751         } else {
1752                 BUG_ON(root == root->fs_info->tree_root);
1753                 ret = insert_reserved_file_extent(trans, inode,
1754                                                 ordered_extent->file_offset,
1755                                                 ordered_extent->start,
1756                                                 ordered_extent->disk_len,
1757                                                 ordered_extent->len,
1758                                                 ordered_extent->len,
1759                                                 compress_type, 0, 0,
1760                                                 BTRFS_FILE_EXTENT_REG);
1761                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1762                                    ordered_extent->file_offset,
1763                                    ordered_extent->len);
1764                 BUG_ON(ret);
1765         }
1766         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1767                              ordered_extent->file_offset +
1768                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1769
1770         add_pending_csums(trans, inode, ordered_extent->file_offset,
1771                           &ordered_extent->list);
1772
1773         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1774         ret = btrfs_update_inode(trans, root, inode);
1775         BUG_ON(ret);
1776 out:
1777         if (nolock) {
1778                 if (trans)
1779                         btrfs_end_transaction_nolock(trans, root);
1780         } else {
1781                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1782                 if (trans)
1783                         btrfs_end_transaction(trans, root);
1784         }
1785
1786         /* once for us */
1787         btrfs_put_ordered_extent(ordered_extent);
1788         /* once for the tree */
1789         btrfs_put_ordered_extent(ordered_extent);
1790
1791         return 0;
1792 }
1793
1794 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1795                                 struct extent_state *state, int uptodate)
1796 {
1797         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1798
1799         ClearPagePrivate2(page);
1800         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1801 }
1802
1803 /*
1804  * When IO fails, either with EIO or csum verification fails, we
1805  * try other mirrors that might have a good copy of the data.  This
1806  * io_failure_record is used to record state as we go through all the
1807  * mirrors.  If another mirror has good data, the page is set up to date
1808  * and things continue.  If a good mirror can't be found, the original
1809  * bio end_io callback is called to indicate things have failed.
1810  */
1811 struct io_failure_record {
1812         struct page *page;
1813         u64 start;
1814         u64 len;
1815         u64 logical;
1816         unsigned long bio_flags;
1817         int last_mirror;
1818 };
1819
1820 static int btrfs_io_failed_hook(struct bio *failed_bio,
1821                          struct page *page, u64 start, u64 end,
1822                          struct extent_state *state)
1823 {
1824         struct io_failure_record *failrec = NULL;
1825         u64 private;
1826         struct extent_map *em;
1827         struct inode *inode = page->mapping->host;
1828         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1829         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1830         struct bio *bio;
1831         int num_copies;
1832         int ret;
1833         int rw;
1834         u64 logical;
1835
1836         ret = get_state_private(failure_tree, start, &private);
1837         if (ret) {
1838                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1839                 if (!failrec)
1840                         return -ENOMEM;
1841                 failrec->start = start;
1842                 failrec->len = end - start + 1;
1843                 failrec->last_mirror = 0;
1844                 failrec->bio_flags = 0;
1845
1846                 read_lock(&em_tree->lock);
1847                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1848                 if (em->start > start || em->start + em->len < start) {
1849                         free_extent_map(em);
1850                         em = NULL;
1851                 }
1852                 read_unlock(&em_tree->lock);
1853
1854                 if (!em || IS_ERR(em)) {
1855                         kfree(failrec);
1856                         return -EIO;
1857                 }
1858                 logical = start - em->start;
1859                 logical = em->block_start + logical;
1860                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1861                         logical = em->block_start;
1862                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1863                         extent_set_compress_type(&failrec->bio_flags,
1864                                                  em->compress_type);
1865                 }
1866                 failrec->logical = logical;
1867                 free_extent_map(em);
1868                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1869                                 EXTENT_DIRTY, GFP_NOFS);
1870                 set_state_private(failure_tree, start,
1871                                  (u64)(unsigned long)failrec);
1872         } else {
1873                 failrec = (struct io_failure_record *)(unsigned long)private;
1874         }
1875         num_copies = btrfs_num_copies(
1876                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1877                               failrec->logical, failrec->len);
1878         failrec->last_mirror++;
1879         if (!state) {
1880                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1881                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1882                                                     failrec->start,
1883                                                     EXTENT_LOCKED);
1884                 if (state && state->start != failrec->start)
1885                         state = NULL;
1886                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1887         }
1888         if (!state || failrec->last_mirror > num_copies) {
1889                 set_state_private(failure_tree, failrec->start, 0);
1890                 clear_extent_bits(failure_tree, failrec->start,
1891                                   failrec->start + failrec->len - 1,
1892                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1893                 kfree(failrec);
1894                 return -EIO;
1895         }
1896         bio = bio_alloc(GFP_NOFS, 1);
1897         bio->bi_private = state;
1898         bio->bi_end_io = failed_bio->bi_end_io;
1899         bio->bi_sector = failrec->logical >> 9;
1900         bio->bi_bdev = failed_bio->bi_bdev;
1901         bio->bi_size = 0;
1902
1903         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1904         if (failed_bio->bi_rw & REQ_WRITE)
1905                 rw = WRITE;
1906         else
1907                 rw = READ;
1908
1909         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1910                                                       failrec->last_mirror,
1911                                                       failrec->bio_flags, 0);
1912         return ret;
1913 }
1914
1915 /*
1916  * each time an IO finishes, we do a fast check in the IO failure tree
1917  * to see if we need to process or clean up an io_failure_record
1918  */
1919 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1920 {
1921         u64 private;
1922         u64 private_failure;
1923         struct io_failure_record *failure;
1924         int ret;
1925
1926         private = 0;
1927         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1928                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1929                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1930                                         start, &private_failure);
1931                 if (ret == 0) {
1932                         failure = (struct io_failure_record *)(unsigned long)
1933                                    private_failure;
1934                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1935                                           failure->start, 0);
1936                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1937                                           failure->start,
1938                                           failure->start + failure->len - 1,
1939                                           EXTENT_DIRTY | EXTENT_LOCKED,
1940                                           GFP_NOFS);
1941                         kfree(failure);
1942                 }
1943         }
1944         return 0;
1945 }
1946
1947 /*
1948  * when reads are done, we need to check csums to verify the data is correct
1949  * if there's a match, we allow the bio to finish.  If not, we go through
1950  * the io_failure_record routines to find good copies
1951  */
1952 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1953                                struct extent_state *state)
1954 {
1955         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1956         struct inode *inode = page->mapping->host;
1957         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1958         char *kaddr;
1959         u64 private = ~(u32)0;
1960         int ret;
1961         struct btrfs_root *root = BTRFS_I(inode)->root;
1962         u32 csum = ~(u32)0;
1963
1964         if (PageChecked(page)) {
1965                 ClearPageChecked(page);
1966                 goto good;
1967         }
1968
1969         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1970                 return 0;
1971
1972         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1973             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1974                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1975                                   GFP_NOFS);
1976                 return 0;
1977         }
1978
1979         if (state && state->start == start) {
1980                 private = state->private;
1981                 ret = 0;
1982         } else {
1983                 ret = get_state_private(io_tree, start, &private);
1984         }
1985         kaddr = kmap_atomic(page, KM_USER0);
1986         if (ret)
1987                 goto zeroit;
1988
1989         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1990         btrfs_csum_final(csum, (char *)&csum);
1991         if (csum != private)
1992                 goto zeroit;
1993
1994         kunmap_atomic(kaddr, KM_USER0);
1995 good:
1996         /* if the io failure tree for this inode is non-empty,
1997          * check to see if we've recovered from a failed IO
1998          */
1999         btrfs_clean_io_failures(inode, start);
2000         return 0;
2001
2002 zeroit:
2003         if (printk_ratelimit()) {
2004                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2005                        "private %llu\n", page->mapping->host->i_ino,
2006                        (unsigned long long)start, csum,
2007                        (unsigned long long)private);
2008         }
2009         memset(kaddr + offset, 1, end - start + 1);
2010         flush_dcache_page(page);
2011         kunmap_atomic(kaddr, KM_USER0);
2012         if (private == 0)
2013                 return 0;
2014         return -EIO;
2015 }
2016
2017 struct delayed_iput {
2018         struct list_head list;
2019         struct inode *inode;
2020 };
2021
2022 void btrfs_add_delayed_iput(struct inode *inode)
2023 {
2024         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2025         struct delayed_iput *delayed;
2026
2027         if (atomic_add_unless(&inode->i_count, -1, 1))
2028                 return;
2029
2030         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2031         delayed->inode = inode;
2032
2033         spin_lock(&fs_info->delayed_iput_lock);
2034         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2035         spin_unlock(&fs_info->delayed_iput_lock);
2036 }
2037
2038 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2039 {
2040         LIST_HEAD(list);
2041         struct btrfs_fs_info *fs_info = root->fs_info;
2042         struct delayed_iput *delayed;
2043         int empty;
2044
2045         spin_lock(&fs_info->delayed_iput_lock);
2046         empty = list_empty(&fs_info->delayed_iputs);
2047         spin_unlock(&fs_info->delayed_iput_lock);
2048         if (empty)
2049                 return;
2050
2051         down_read(&root->fs_info->cleanup_work_sem);
2052         spin_lock(&fs_info->delayed_iput_lock);
2053         list_splice_init(&fs_info->delayed_iputs, &list);
2054         spin_unlock(&fs_info->delayed_iput_lock);
2055
2056         while (!list_empty(&list)) {
2057                 delayed = list_entry(list.next, struct delayed_iput, list);
2058                 list_del(&delayed->list);
2059                 iput(delayed->inode);
2060                 kfree(delayed);
2061         }
2062         up_read(&root->fs_info->cleanup_work_sem);
2063 }
2064
2065 /*
2066  * calculate extra metadata reservation when snapshotting a subvolume
2067  * contains orphan files.
2068  */
2069 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2070                                 struct btrfs_pending_snapshot *pending,
2071                                 u64 *bytes_to_reserve)
2072 {
2073         struct btrfs_root *root;
2074         struct btrfs_block_rsv *block_rsv;
2075         u64 num_bytes;
2076         int index;
2077
2078         root = pending->root;
2079         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2080                 return;
2081
2082         block_rsv = root->orphan_block_rsv;
2083
2084         /* orphan block reservation for the snapshot */
2085         num_bytes = block_rsv->size;
2086
2087         /*
2088          * after the snapshot is created, COWing tree blocks may use more
2089          * space than it frees. So we should make sure there is enough
2090          * reserved space.
2091          */
2092         index = trans->transid & 0x1;
2093         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2094                 num_bytes += block_rsv->size -
2095                              (block_rsv->reserved + block_rsv->freed[index]);
2096         }
2097
2098         *bytes_to_reserve += num_bytes;
2099 }
2100
2101 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2102                                 struct btrfs_pending_snapshot *pending)
2103 {
2104         struct btrfs_root *root = pending->root;
2105         struct btrfs_root *snap = pending->snap;
2106         struct btrfs_block_rsv *block_rsv;
2107         u64 num_bytes;
2108         int index;
2109         int ret;
2110
2111         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2112                 return;
2113
2114         /* refill source subvolume's orphan block reservation */
2115         block_rsv = root->orphan_block_rsv;
2116         index = trans->transid & 0x1;
2117         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2118                 num_bytes = block_rsv->size -
2119                             (block_rsv->reserved + block_rsv->freed[index]);
2120                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2121                                               root->orphan_block_rsv,
2122                                               num_bytes);
2123                 BUG_ON(ret);
2124         }
2125
2126         /* setup orphan block reservation for the snapshot */
2127         block_rsv = btrfs_alloc_block_rsv(snap);
2128         BUG_ON(!block_rsv);
2129
2130         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2131         snap->orphan_block_rsv = block_rsv;
2132
2133         num_bytes = root->orphan_block_rsv->size;
2134         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2135                                       block_rsv, num_bytes);
2136         BUG_ON(ret);
2137
2138 #if 0
2139         /* insert orphan item for the snapshot */
2140         WARN_ON(!root->orphan_item_inserted);
2141         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2142                                        snap->root_key.objectid);
2143         BUG_ON(ret);
2144         snap->orphan_item_inserted = 1;
2145 #endif
2146 }
2147
2148 enum btrfs_orphan_cleanup_state {
2149         ORPHAN_CLEANUP_STARTED  = 1,
2150         ORPHAN_CLEANUP_DONE     = 2,
2151 };
2152
2153 /*
2154  * This is called in transaction commmit time. If there are no orphan
2155  * files in the subvolume, it removes orphan item and frees block_rsv
2156  * structure.
2157  */
2158 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2159                               struct btrfs_root *root)
2160 {
2161         int ret;
2162
2163         if (!list_empty(&root->orphan_list) ||
2164             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2165                 return;
2166
2167         if (root->orphan_item_inserted &&
2168             btrfs_root_refs(&root->root_item) > 0) {
2169                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2170                                             root->root_key.objectid);
2171                 BUG_ON(ret);
2172                 root->orphan_item_inserted = 0;
2173         }
2174
2175         if (root->orphan_block_rsv) {
2176                 WARN_ON(root->orphan_block_rsv->size > 0);
2177                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2178                 root->orphan_block_rsv = NULL;
2179         }
2180 }
2181
2182 /*
2183  * This creates an orphan entry for the given inode in case something goes
2184  * wrong in the middle of an unlink/truncate.
2185  *
2186  * NOTE: caller of this function should reserve 5 units of metadata for
2187  *       this function.
2188  */
2189 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2190 {
2191         struct btrfs_root *root = BTRFS_I(inode)->root;
2192         struct btrfs_block_rsv *block_rsv = NULL;
2193         int reserve = 0;
2194         int insert = 0;
2195         int ret;
2196
2197         if (!root->orphan_block_rsv) {
2198                 block_rsv = btrfs_alloc_block_rsv(root);
2199                 BUG_ON(!block_rsv);
2200         }
2201
2202         spin_lock(&root->orphan_lock);
2203         if (!root->orphan_block_rsv) {
2204                 root->orphan_block_rsv = block_rsv;
2205         } else if (block_rsv) {
2206                 btrfs_free_block_rsv(root, block_rsv);
2207                 block_rsv = NULL;
2208         }
2209
2210         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2211                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2212 #if 0
2213                 /*
2214                  * For proper ENOSPC handling, we should do orphan
2215                  * cleanup when mounting. But this introduces backward
2216                  * compatibility issue.
2217                  */
2218                 if (!xchg(&root->orphan_item_inserted, 1))
2219                         insert = 2;
2220                 else
2221                         insert = 1;
2222 #endif
2223                 insert = 1;
2224         } else {
2225                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2226         }
2227
2228         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2229                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2230                 reserve = 1;
2231         }
2232         spin_unlock(&root->orphan_lock);
2233
2234         if (block_rsv)
2235                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2236
2237         /* grab metadata reservation from transaction handle */
2238         if (reserve) {
2239                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2240                 BUG_ON(ret);
2241         }
2242
2243         /* insert an orphan item to track this unlinked/truncated file */
2244         if (insert >= 1) {
2245                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2246                 BUG_ON(ret);
2247         }
2248
2249         /* insert an orphan item to track subvolume contains orphan files */
2250         if (insert >= 2) {
2251                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2252                                                root->root_key.objectid);
2253                 BUG_ON(ret);
2254         }
2255         return 0;
2256 }
2257
2258 /*
2259  * We have done the truncate/delete so we can go ahead and remove the orphan
2260  * item for this particular inode.
2261  */
2262 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2263 {
2264         struct btrfs_root *root = BTRFS_I(inode)->root;
2265         int delete_item = 0;
2266         int release_rsv = 0;
2267         int ret = 0;
2268
2269         spin_lock(&root->orphan_lock);
2270         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2271                 list_del_init(&BTRFS_I(inode)->i_orphan);
2272                 delete_item = 1;
2273         }
2274
2275         if (BTRFS_I(inode)->orphan_meta_reserved) {
2276                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2277                 release_rsv = 1;
2278         }
2279         spin_unlock(&root->orphan_lock);
2280
2281         if (trans && delete_item) {
2282                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2283                 BUG_ON(ret);
2284         }
2285
2286         if (release_rsv)
2287                 btrfs_orphan_release_metadata(inode);
2288
2289         return 0;
2290 }
2291
2292 /*
2293  * this cleans up any orphans that may be left on the list from the last use
2294  * of this root.
2295  */
2296 int btrfs_orphan_cleanup(struct btrfs_root *root)
2297 {
2298         struct btrfs_path *path;
2299         struct extent_buffer *leaf;
2300         struct btrfs_key key, found_key;
2301         struct btrfs_trans_handle *trans;
2302         struct inode *inode;
2303         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2304
2305         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2306                 return 0;
2307
2308         path = btrfs_alloc_path();
2309         if (!path) {
2310                 ret = -ENOMEM;
2311                 goto out;
2312         }
2313         path->reada = -1;
2314
2315         key.objectid = BTRFS_ORPHAN_OBJECTID;
2316         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2317         key.offset = (u64)-1;
2318
2319         while (1) {
2320                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2321                 if (ret < 0)
2322                         goto out;
2323
2324                 /*
2325                  * if ret == 0 means we found what we were searching for, which
2326                  * is weird, but possible, so only screw with path if we didnt
2327                  * find the key and see if we have stuff that matches
2328                  */
2329                 if (ret > 0) {
2330                         ret = 0;
2331                         if (path->slots[0] == 0)
2332                                 break;
2333                         path->slots[0]--;
2334                 }
2335
2336                 /* pull out the item */
2337                 leaf = path->nodes[0];
2338                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2339
2340                 /* make sure the item matches what we want */
2341                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2342                         break;
2343                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2344                         break;
2345
2346                 /* release the path since we're done with it */
2347                 btrfs_release_path(root, path);
2348
2349                 /*
2350                  * this is where we are basically btrfs_lookup, without the
2351                  * crossing root thing.  we store the inode number in the
2352                  * offset of the orphan item.
2353                  */
2354                 found_key.objectid = found_key.offset;
2355                 found_key.type = BTRFS_INODE_ITEM_KEY;
2356                 found_key.offset = 0;
2357                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2358                 if (IS_ERR(inode)) {
2359                         ret = PTR_ERR(inode);
2360                         goto out;
2361                 }
2362
2363                 /*
2364                  * add this inode to the orphan list so btrfs_orphan_del does
2365                  * the proper thing when we hit it
2366                  */
2367                 spin_lock(&root->orphan_lock);
2368                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2369                 spin_unlock(&root->orphan_lock);
2370
2371                 /*
2372                  * if this is a bad inode, means we actually succeeded in
2373                  * removing the inode, but not the orphan record, which means
2374                  * we need to manually delete the orphan since iput will just
2375                  * do a destroy_inode
2376                  */
2377                 if (is_bad_inode(inode)) {
2378                         trans = btrfs_start_transaction(root, 0);
2379                         if (IS_ERR(trans)) {
2380                                 ret = PTR_ERR(trans);
2381                                 goto out;
2382                         }
2383                         btrfs_orphan_del(trans, inode);
2384                         btrfs_end_transaction(trans, root);
2385                         iput(inode);
2386                         continue;
2387                 }
2388
2389                 /* if we have links, this was a truncate, lets do that */
2390                 if (inode->i_nlink) {
2391                         if (!S_ISREG(inode->i_mode)) {
2392                                 WARN_ON(1);
2393                                 iput(inode);
2394                                 continue;
2395                         }
2396                         nr_truncate++;
2397                         ret = btrfs_truncate(inode);
2398                 } else {
2399                         nr_unlink++;
2400                 }
2401
2402                 /* this will do delete_inode and everything for us */
2403                 iput(inode);
2404                 if (ret)
2405                         goto out;
2406         }
2407         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2408
2409         if (root->orphan_block_rsv)
2410                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2411                                         (u64)-1);
2412
2413         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2414                 trans = btrfs_join_transaction(root, 1);
2415                 if (!IS_ERR(trans))
2416                         btrfs_end_transaction(trans, root);
2417         }
2418
2419         if (nr_unlink)
2420                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2421         if (nr_truncate)
2422                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2423
2424 out:
2425         if (ret)
2426                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2427         btrfs_free_path(path);
2428         return ret;
2429 }
2430
2431 /*
2432  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2433  * don't find any xattrs, we know there can't be any acls.
2434  *
2435  * slot is the slot the inode is in, objectid is the objectid of the inode
2436  */
2437 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2438                                           int slot, u64 objectid)
2439 {
2440         u32 nritems = btrfs_header_nritems(leaf);
2441         struct btrfs_key found_key;
2442         int scanned = 0;
2443
2444         slot++;
2445         while (slot < nritems) {
2446                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2447
2448                 /* we found a different objectid, there must not be acls */
2449                 if (found_key.objectid != objectid)
2450                         return 0;
2451
2452                 /* we found an xattr, assume we've got an acl */
2453                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2454                         return 1;
2455
2456                 /*
2457                  * we found a key greater than an xattr key, there can't
2458                  * be any acls later on
2459                  */
2460                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2461                         return 0;
2462
2463                 slot++;
2464                 scanned++;
2465
2466                 /*
2467                  * it goes inode, inode backrefs, xattrs, extents,
2468                  * so if there are a ton of hard links to an inode there can
2469                  * be a lot of backrefs.  Don't waste time searching too hard,
2470                  * this is just an optimization
2471                  */
2472                 if (scanned >= 8)
2473                         break;
2474         }
2475         /* we hit the end of the leaf before we found an xattr or
2476          * something larger than an xattr.  We have to assume the inode
2477          * has acls
2478          */
2479         return 1;
2480 }
2481
2482 /*
2483  * read an inode from the btree into the in-memory inode
2484  */
2485 static void btrfs_read_locked_inode(struct inode *inode)
2486 {
2487         struct btrfs_path *path;
2488         struct extent_buffer *leaf;
2489         struct btrfs_inode_item *inode_item;
2490         struct btrfs_timespec *tspec;
2491         struct btrfs_root *root = BTRFS_I(inode)->root;
2492         struct btrfs_key location;
2493         int maybe_acls;
2494         u64 alloc_group_block;
2495         u32 rdev;
2496         int ret;
2497
2498         path = btrfs_alloc_path();
2499         BUG_ON(!path);
2500         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2501
2502         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2503         if (ret)
2504                 goto make_bad;
2505
2506         leaf = path->nodes[0];
2507         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2508                                     struct btrfs_inode_item);
2509
2510         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2511         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2512         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2513         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2514         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2515
2516         tspec = btrfs_inode_atime(inode_item);
2517         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2518         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2519
2520         tspec = btrfs_inode_mtime(inode_item);
2521         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523
2524         tspec = btrfs_inode_ctime(inode_item);
2525         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2526         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2527
2528         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2529         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2530         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2531         inode->i_generation = BTRFS_I(inode)->generation;
2532         inode->i_rdev = 0;
2533         rdev = btrfs_inode_rdev(leaf, inode_item);
2534
2535         BTRFS_I(inode)->index_cnt = (u64)-1;
2536         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2537
2538         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2539         if (location.objectid == BTRFS_FREE_SPACE_OBJECTID)
2540                 inode->i_mapping->flags &= ~__GFP_FS;
2541
2542         /*
2543          * try to precache a NULL acl entry for files that don't have
2544          * any xattrs or acls
2545          */
2546         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2547         if (!maybe_acls)
2548                 cache_no_acl(inode);
2549
2550         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2551                                                 alloc_group_block, 0);
2552         btrfs_free_path(path);
2553         inode_item = NULL;
2554
2555         switch (inode->i_mode & S_IFMT) {
2556         case S_IFREG:
2557                 inode->i_mapping->a_ops = &btrfs_aops;
2558                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2559                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2560                 inode->i_fop = &btrfs_file_operations;
2561                 inode->i_op = &btrfs_file_inode_operations;
2562                 break;
2563         case S_IFDIR:
2564                 inode->i_fop = &btrfs_dir_file_operations;
2565                 if (root == root->fs_info->tree_root)
2566                         inode->i_op = &btrfs_dir_ro_inode_operations;
2567                 else
2568                         inode->i_op = &btrfs_dir_inode_operations;
2569                 break;
2570         case S_IFLNK:
2571                 inode->i_op = &btrfs_symlink_inode_operations;
2572                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2573                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2574                 break;
2575         default:
2576                 inode->i_op = &btrfs_special_inode_operations;
2577                 init_special_inode(inode, inode->i_mode, rdev);
2578                 break;
2579         }
2580
2581         btrfs_update_iflags(inode);
2582         return;
2583
2584 make_bad:
2585         btrfs_free_path(path);
2586         make_bad_inode(inode);
2587 }
2588
2589 /*
2590  * given a leaf and an inode, copy the inode fields into the leaf
2591  */
2592 static void fill_inode_item(struct btrfs_trans_handle *trans,
2593                             struct extent_buffer *leaf,
2594                             struct btrfs_inode_item *item,
2595                             struct inode *inode)
2596 {
2597         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2598         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2599         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2600         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2601         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2602
2603         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2604                                inode->i_atime.tv_sec);
2605         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2606                                 inode->i_atime.tv_nsec);
2607
2608         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2609                                inode->i_mtime.tv_sec);
2610         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2611                                 inode->i_mtime.tv_nsec);
2612
2613         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2614                                inode->i_ctime.tv_sec);
2615         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2616                                 inode->i_ctime.tv_nsec);
2617
2618         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2619         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2620         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2621         btrfs_set_inode_transid(leaf, item, trans->transid);
2622         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2623         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2624         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2625 }
2626
2627 /*
2628  * copy everything in the in-memory inode into the btree.
2629  */
2630 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2631                                 struct btrfs_root *root, struct inode *inode)
2632 {
2633         struct btrfs_inode_item *inode_item;
2634         struct btrfs_path *path;
2635         struct extent_buffer *leaf;
2636         int ret;
2637
2638         path = btrfs_alloc_path();
2639         BUG_ON(!path);
2640         path->leave_spinning = 1;
2641         ret = btrfs_lookup_inode(trans, root, path,
2642                                  &BTRFS_I(inode)->location, 1);
2643         if (ret) {
2644                 if (ret > 0)
2645                         ret = -ENOENT;
2646                 goto failed;
2647         }
2648
2649         btrfs_unlock_up_safe(path, 1);
2650         leaf = path->nodes[0];
2651         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2652                                   struct btrfs_inode_item);
2653
2654         fill_inode_item(trans, leaf, inode_item, inode);
2655         btrfs_mark_buffer_dirty(leaf);
2656         btrfs_set_inode_last_trans(trans, inode);
2657         ret = 0;
2658 failed:
2659         btrfs_free_path(path);
2660         return ret;
2661 }
2662
2663
2664 /*
2665  * unlink helper that gets used here in inode.c and in the tree logging
2666  * recovery code.  It remove a link in a directory with a given name, and
2667  * also drops the back refs in the inode to the directory
2668  */
2669 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2670                                 struct btrfs_root *root,
2671                                 struct inode *dir, struct inode *inode,
2672                                 const char *name, int name_len)
2673 {
2674         struct btrfs_path *path;
2675         int ret = 0;
2676         struct extent_buffer *leaf;
2677         struct btrfs_dir_item *di;
2678         struct btrfs_key key;
2679         u64 index;
2680
2681         path = btrfs_alloc_path();
2682         if (!path) {
2683                 ret = -ENOMEM;
2684                 goto out;
2685         }
2686
2687         path->leave_spinning = 1;
2688         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2689                                     name, name_len, -1);
2690         if (IS_ERR(di)) {
2691                 ret = PTR_ERR(di);
2692                 goto err;
2693         }
2694         if (!di) {
2695                 ret = -ENOENT;
2696                 goto err;
2697         }
2698         leaf = path->nodes[0];
2699         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2700         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2701         if (ret)
2702                 goto err;
2703         btrfs_release_path(root, path);
2704
2705         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2706                                   inode->i_ino,
2707                                   dir->i_ino, &index);
2708         if (ret) {
2709                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2710                        "inode %lu parent %lu\n", name_len, name,
2711                        inode->i_ino, dir->i_ino);
2712                 goto err;
2713         }
2714
2715         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2716                                          index, name, name_len, -1);
2717         if (IS_ERR(di)) {
2718                 ret = PTR_ERR(di);
2719                 goto err;
2720         }
2721         if (!di) {
2722                 ret = -ENOENT;
2723                 goto err;
2724         }
2725         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2726         btrfs_release_path(root, path);
2727
2728         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2729                                          inode, dir->i_ino);
2730         BUG_ON(ret != 0 && ret != -ENOENT);
2731
2732         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2733                                            dir, index);
2734         if (ret == -ENOENT)
2735                 ret = 0;
2736 err:
2737         btrfs_free_path(path);
2738         if (ret)
2739                 goto out;
2740
2741         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2742         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2743         btrfs_update_inode(trans, root, dir);
2744 out:
2745         return ret;
2746 }
2747
2748 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2749                        struct btrfs_root *root,
2750                        struct inode *dir, struct inode *inode,
2751                        const char *name, int name_len)
2752 {
2753         int ret;
2754         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2755         if (!ret) {
2756                 btrfs_drop_nlink(inode);
2757                 ret = btrfs_update_inode(trans, root, inode);
2758         }
2759         return ret;
2760 }
2761
2762
2763 /* helper to check if there is any shared block in the path */
2764 static int check_path_shared(struct btrfs_root *root,
2765                              struct btrfs_path *path)
2766 {
2767         struct extent_buffer *eb;
2768         int level;
2769         u64 refs = 1;
2770
2771         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2772                 int ret;
2773
2774                 if (!path->nodes[level])
2775                         break;
2776                 eb = path->nodes[level];
2777                 if (!btrfs_block_can_be_shared(root, eb))
2778                         continue;
2779                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2780                                                &refs, NULL);
2781                 if (refs > 1)
2782                         return 1;
2783         }
2784         return 0;
2785 }
2786
2787 /*
2788  * helper to start transaction for unlink and rmdir.
2789  *
2790  * unlink and rmdir are special in btrfs, they do not always free space.
2791  * so in enospc case, we should make sure they will free space before
2792  * allowing them to use the global metadata reservation.
2793  */
2794 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2795                                                        struct dentry *dentry)
2796 {
2797         struct btrfs_trans_handle *trans;
2798         struct btrfs_root *root = BTRFS_I(dir)->root;
2799         struct btrfs_path *path;
2800         struct btrfs_inode_ref *ref;
2801         struct btrfs_dir_item *di;
2802         struct inode *inode = dentry->d_inode;
2803         u64 index;
2804         int check_link = 1;
2805         int err = -ENOSPC;
2806         int ret;
2807
2808         trans = btrfs_start_transaction(root, 10);
2809         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2810                 return trans;
2811
2812         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2813                 return ERR_PTR(-ENOSPC);
2814
2815         /* check if there is someone else holds reference */
2816         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2817                 return ERR_PTR(-ENOSPC);
2818
2819         if (atomic_read(&inode->i_count) > 2)
2820                 return ERR_PTR(-ENOSPC);
2821
2822         if (xchg(&root->fs_info->enospc_unlink, 1))
2823                 return ERR_PTR(-ENOSPC);
2824
2825         path = btrfs_alloc_path();
2826         if (!path) {
2827                 root->fs_info->enospc_unlink = 0;
2828                 return ERR_PTR(-ENOMEM);
2829         }
2830
2831         trans = btrfs_start_transaction(root, 0);
2832         if (IS_ERR(trans)) {
2833                 btrfs_free_path(path);
2834                 root->fs_info->enospc_unlink = 0;
2835                 return trans;
2836         }
2837
2838         path->skip_locking = 1;
2839         path->search_commit_root = 1;
2840
2841         ret = btrfs_lookup_inode(trans, root, path,
2842                                 &BTRFS_I(dir)->location, 0);
2843         if (ret < 0) {
2844                 err = ret;
2845                 goto out;
2846         }
2847         if (ret == 0) {
2848                 if (check_path_shared(root, path))
2849                         goto out;
2850         } else {
2851                 check_link = 0;
2852         }
2853         btrfs_release_path(root, path);
2854
2855         ret = btrfs_lookup_inode(trans, root, path,
2856                                 &BTRFS_I(inode)->location, 0);
2857         if (ret < 0) {
2858                 err = ret;
2859                 goto out;
2860         }
2861         if (ret == 0) {
2862                 if (check_path_shared(root, path))
2863                         goto out;
2864         } else {
2865                 check_link = 0;
2866         }
2867         btrfs_release_path(root, path);
2868
2869         if (ret == 0 && S_ISREG(inode->i_mode)) {
2870                 ret = btrfs_lookup_file_extent(trans, root, path,
2871                                                inode->i_ino, (u64)-1, 0);
2872                 if (ret < 0) {
2873                         err = ret;
2874                         goto out;
2875                 }
2876                 BUG_ON(ret == 0);
2877                 if (check_path_shared(root, path))
2878                         goto out;
2879                 btrfs_release_path(root, path);
2880         }
2881
2882         if (!check_link) {
2883                 err = 0;
2884                 goto out;
2885         }
2886
2887         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2888                                 dentry->d_name.name, dentry->d_name.len, 0);
2889         if (IS_ERR(di)) {
2890                 err = PTR_ERR(di);
2891                 goto out;
2892         }
2893         if (di) {
2894                 if (check_path_shared(root, path))
2895                         goto out;
2896         } else {
2897                 err = 0;
2898                 goto out;
2899         }
2900         btrfs_release_path(root, path);
2901
2902         ref = btrfs_lookup_inode_ref(trans, root, path,
2903                                 dentry->d_name.name, dentry->d_name.len,
2904                                 inode->i_ino, dir->i_ino, 0);
2905         if (IS_ERR(ref)) {
2906                 err = PTR_ERR(ref);
2907                 goto out;
2908         }
2909         BUG_ON(!ref);
2910         if (check_path_shared(root, path))
2911                 goto out;
2912         index = btrfs_inode_ref_index(path->nodes[0], ref);
2913         btrfs_release_path(root, path);
2914
2915         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2916                                 dentry->d_name.name, dentry->d_name.len, 0);
2917         if (IS_ERR(di)) {
2918                 err = PTR_ERR(di);
2919                 goto out;
2920         }
2921         BUG_ON(ret == -ENOENT);
2922         if (check_path_shared(root, path))
2923                 goto out;
2924
2925         err = 0;
2926 out:
2927         btrfs_free_path(path);
2928         if (err) {
2929                 btrfs_end_transaction(trans, root);
2930                 root->fs_info->enospc_unlink = 0;
2931                 return ERR_PTR(err);
2932         }
2933
2934         trans->block_rsv = &root->fs_info->global_block_rsv;
2935         return trans;
2936 }
2937
2938 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2939                                struct btrfs_root *root)
2940 {
2941         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2942                 BUG_ON(!root->fs_info->enospc_unlink);
2943                 root->fs_info->enospc_unlink = 0;
2944         }
2945         btrfs_end_transaction_throttle(trans, root);
2946 }
2947
2948 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2949 {
2950         struct btrfs_root *root = BTRFS_I(dir)->root;
2951         struct btrfs_trans_handle *trans;
2952         struct inode *inode = dentry->d_inode;
2953         int ret;
2954         unsigned long nr = 0;
2955
2956         trans = __unlink_start_trans(dir, dentry);
2957         if (IS_ERR(trans))
2958                 return PTR_ERR(trans);
2959
2960         btrfs_set_trans_block_group(trans, dir);
2961
2962         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2963
2964         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2965                                  dentry->d_name.name, dentry->d_name.len);
2966         BUG_ON(ret);
2967
2968         if (inode->i_nlink == 0) {
2969                 ret = btrfs_orphan_add(trans, inode);
2970                 BUG_ON(ret);
2971         }
2972
2973         nr = trans->blocks_used;
2974         __unlink_end_trans(trans, root);
2975         btrfs_btree_balance_dirty(root, nr);
2976         return ret;
2977 }
2978
2979 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2980                         struct btrfs_root *root,
2981                         struct inode *dir, u64 objectid,
2982                         const char *name, int name_len)
2983 {
2984         struct btrfs_path *path;
2985         struct extent_buffer *leaf;
2986         struct btrfs_dir_item *di;
2987         struct btrfs_key key;
2988         u64 index;
2989         int ret;
2990
2991         path = btrfs_alloc_path();
2992         if (!path)
2993                 return -ENOMEM;
2994
2995         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2996                                    name, name_len, -1);
2997         BUG_ON(!di || IS_ERR(di));
2998
2999         leaf = path->nodes[0];
3000         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3001         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3002         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3003         BUG_ON(ret);
3004         btrfs_release_path(root, path);
3005
3006         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3007                                  objectid, root->root_key.objectid,
3008                                  dir->i_ino, &index, name, name_len);
3009         if (ret < 0) {
3010                 BUG_ON(ret != -ENOENT);
3011                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3012                                                  name, name_len);
3013                 BUG_ON(!di || IS_ERR(di));
3014
3015                 leaf = path->nodes[0];
3016                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3017                 btrfs_release_path(root, path);
3018                 index = key.offset;
3019         }
3020
3021         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3022                                          index, name, name_len, -1);
3023         BUG_ON(!di || IS_ERR(di));
3024
3025         leaf = path->nodes[0];
3026         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3027         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3028         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3029         BUG_ON(ret);
3030         btrfs_release_path(root, path);
3031
3032         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3033         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3034         ret = btrfs_update_inode(trans, root, dir);
3035         BUG_ON(ret);
3036
3037         btrfs_free_path(path);
3038         return 0;
3039 }
3040
3041 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3042 {
3043         struct inode *inode = dentry->d_inode;
3044         int err = 0;
3045         struct btrfs_root *root = BTRFS_I(dir)->root;
3046         struct btrfs_trans_handle *trans;
3047         unsigned long nr = 0;
3048
3049         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3050             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3051                 return -ENOTEMPTY;
3052
3053         trans = __unlink_start_trans(dir, dentry);
3054         if (IS_ERR(trans))
3055                 return PTR_ERR(trans);
3056
3057         btrfs_set_trans_block_group(trans, dir);
3058
3059         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3060                 err = btrfs_unlink_subvol(trans, root, dir,
3061                                           BTRFS_I(inode)->location.objectid,
3062                                           dentry->d_name.name,
3063                                           dentry->d_name.len);
3064                 goto out;
3065         }
3066
3067         err = btrfs_orphan_add(trans, inode);
3068         if (err)
3069                 goto out;
3070
3071         /* now the directory is empty */
3072         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3073                                  dentry->d_name.name, dentry->d_name.len);
3074         if (!err)
3075                 btrfs_i_size_write(inode, 0);
3076 out:
3077         nr = trans->blocks_used;
3078         __unlink_end_trans(trans, root);
3079         btrfs_btree_balance_dirty(root, nr);
3080
3081         return err;
3082 }
3083
3084 #if 0
3085 /*
3086  * when truncating bytes in a file, it is possible to avoid reading
3087  * the leaves that contain only checksum items.  This can be the
3088  * majority of the IO required to delete a large file, but it must
3089  * be done carefully.
3090  *
3091  * The keys in the level just above the leaves are checked to make sure
3092  * the lowest key in a given leaf is a csum key, and starts at an offset
3093  * after the new  size.
3094  *
3095  * Then the key for the next leaf is checked to make sure it also has
3096  * a checksum item for the same file.  If it does, we know our target leaf
3097  * contains only checksum items, and it can be safely freed without reading
3098  * it.
3099  *
3100  * This is just an optimization targeted at large files.  It may do
3101  * nothing.  It will return 0 unless things went badly.
3102  */
3103 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3104                                      struct btrfs_root *root,
3105                                      struct btrfs_path *path,
3106                                      struct inode *inode, u64 new_size)
3107 {
3108         struct btrfs_key key;
3109         int ret;
3110         int nritems;
3111         struct btrfs_key found_key;
3112         struct btrfs_key other_key;
3113         struct btrfs_leaf_ref *ref;
3114         u64 leaf_gen;
3115         u64 leaf_start;
3116
3117         path->lowest_level = 1;
3118         key.objectid = inode->i_ino;
3119         key.type = BTRFS_CSUM_ITEM_KEY;
3120         key.offset = new_size;
3121 again:
3122         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3123         if (ret < 0)
3124                 goto out;
3125
3126         if (path->nodes[1] == NULL) {
3127                 ret = 0;
3128                 goto out;
3129         }
3130         ret = 0;
3131         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3132         nritems = btrfs_header_nritems(path->nodes[1]);
3133
3134         if (!nritems)
3135                 goto out;
3136
3137         if (path->slots[1] >= nritems)
3138                 goto next_node;
3139
3140         /* did we find a key greater than anything we want to delete? */
3141         if (found_key.objectid > inode->i_ino ||
3142            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3143                 goto out;
3144
3145         /* we check the next key in the node to make sure the leave contains
3146          * only checksum items.  This comparison doesn't work if our
3147          * leaf is the last one in the node
3148          */
3149         if (path->slots[1] + 1 >= nritems) {
3150 next_node:
3151                 /* search forward from the last key in the node, this
3152                  * will bring us into the next node in the tree
3153                  */
3154                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3155
3156                 /* unlikely, but we inc below, so check to be safe */
3157                 if (found_key.offset == (u64)-1)
3158                         goto out;
3159
3160                 /* search_forward needs a path with locks held, do the
3161                  * search again for the original key.  It is possible
3162                  * this will race with a balance and return a path that
3163                  * we could modify, but this drop is just an optimization
3164                  * and is allowed to miss some leaves.
3165                  */
3166                 btrfs_release_path(root, path);
3167                 found_key.offset++;
3168
3169                 /* setup a max key for search_forward */
3170                 other_key.offset = (u64)-1;
3171                 other_key.type = key.type;
3172                 other_key.objectid = key.objectid;
3173
3174                 path->keep_locks = 1;
3175                 ret = btrfs_search_forward(root, &found_key, &other_key,
3176                                            path, 0, 0);
3177                 path->keep_locks = 0;
3178                 if (ret || found_key.objectid != key.objectid ||
3179                     found_key.type != key.type) {
3180                         ret = 0;
3181                         goto out;
3182                 }
3183
3184                 key.offset = found_key.offset;
3185                 btrfs_release_path(root, path);
3186                 cond_resched();
3187                 goto again;
3188         }
3189
3190         /* we know there's one more slot after us in the tree,
3191          * read that key so we can verify it is also a checksum item
3192          */
3193         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3194
3195         if (found_key.objectid < inode->i_ino)
3196                 goto next_key;
3197
3198         if (found_key.type != key.type || found_key.offset < new_size)
3199                 goto next_key;
3200
3201         /*
3202          * if the key for the next leaf isn't a csum key from this objectid,
3203          * we can't be sure there aren't good items inside this leaf.
3204          * Bail out
3205          */
3206         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3207                 goto out;
3208
3209         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3210         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3211         /*
3212          * it is safe to delete this leaf, it contains only
3213          * csum items from this inode at an offset >= new_size
3214          */
3215         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3216         BUG_ON(ret);
3217
3218         if (root->ref_cows && leaf_gen < trans->transid) {
3219                 ref = btrfs_alloc_leaf_ref(root, 0);
3220                 if (ref) {
3221                         ref->root_gen = root->root_key.offset;
3222                         ref->bytenr = leaf_start;
3223                         ref->owner = 0;
3224                         ref->generation = leaf_gen;
3225                         ref->nritems = 0;
3226
3227                         btrfs_sort_leaf_ref(ref);
3228
3229                         ret = btrfs_add_leaf_ref(root, ref, 0);
3230                         WARN_ON(ret);
3231                         btrfs_free_leaf_ref(root, ref);
3232                 } else {
3233                         WARN_ON(1);
3234                 }
3235         }
3236 next_key:
3237         btrfs_release_path(root, path);
3238
3239         if (other_key.objectid == inode->i_ino &&
3240             other_key.type == key.type && other_key.offset > key.offset) {
3241                 key.offset = other_key.offset;
3242                 cond_resched();
3243                 goto again;
3244         }
3245         ret = 0;
3246 out:
3247         /* fixup any changes we've made to the path */
3248         path->lowest_level = 0;
3249         path->keep_locks = 0;
3250         btrfs_release_path(root, path);
3251         return ret;
3252 }
3253
3254 #endif
3255
3256 /*
3257  * this can truncate away extent items, csum items and directory items.
3258  * It starts at a high offset and removes keys until it can't find
3259  * any higher than new_size
3260  *
3261  * csum items that cross the new i_size are truncated to the new size
3262  * as well.
3263  *
3264  * min_type is the minimum key type to truncate down to.  If set to 0, this
3265  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3266  */
3267 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3268                                struct btrfs_root *root,
3269                                struct inode *inode,
3270                                u64 new_size, u32 min_type)
3271 {
3272         struct btrfs_path *path;
3273         struct extent_buffer *leaf;
3274         struct btrfs_file_extent_item *fi;
3275         struct btrfs_key key;
3276         struct btrfs_key found_key;
3277         u64 extent_start = 0;
3278         u64 extent_num_bytes = 0;
3279         u64 extent_offset = 0;
3280         u64 item_end = 0;
3281         u64 mask = root->sectorsize - 1;
3282         u32 found_type = (u8)-1;
3283         int found_extent;
3284         int del_item;
3285         int pending_del_nr = 0;
3286         int pending_del_slot = 0;
3287         int extent_type = -1;
3288         int encoding;
3289         int ret;
3290         int err = 0;
3291
3292         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3293
3294         if (root->ref_cows || root == root->fs_info->tree_root)
3295                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3296
3297         path = btrfs_alloc_path();
3298         BUG_ON(!path);
3299         path->reada = -1;
3300
3301         key.objectid = inode->i_ino;
3302         key.offset = (u64)-1;
3303         key.type = (u8)-1;
3304
3305 search_again:
3306         path->leave_spinning = 1;
3307         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3308         if (ret < 0) {
3309                 err = ret;
3310                 goto out;
3311         }
3312
3313         if (ret > 0) {
3314                 /* there are no items in the tree for us to truncate, we're
3315                  * done
3316                  */
3317                 if (path->slots[0] == 0)
3318                         goto out;
3319                 path->slots[0]--;
3320         }
3321
3322         while (1) {
3323                 fi = NULL;
3324                 leaf = path->nodes[0];
3325                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3326                 found_type = btrfs_key_type(&found_key);
3327                 encoding = 0;
3328
3329                 if (found_key.objectid != inode->i_ino)
3330                         break;
3331
3332                 if (found_type < min_type)
3333                         break;
3334
3335                 item_end = found_key.offset;
3336                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3337                         fi = btrfs_item_ptr(leaf, path->slots[0],
3338                                             struct btrfs_file_extent_item);
3339                         extent_type = btrfs_file_extent_type(leaf, fi);
3340                         encoding = btrfs_file_extent_compression(leaf, fi);
3341                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3342                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3343
3344                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3345                                 item_end +=
3346                                     btrfs_file_extent_num_bytes(leaf, fi);
3347                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3348                                 item_end += btrfs_file_extent_inline_len(leaf,
3349                                                                          fi);
3350                         }
3351                         item_end--;
3352                 }
3353                 if (found_type > min_type) {
3354                         del_item = 1;
3355                 } else {
3356                         if (item_end < new_size)
3357                                 break;
3358                         if (found_key.offset >= new_size)
3359                                 del_item = 1;
3360                         else
3361                                 del_item = 0;
3362                 }
3363                 found_extent = 0;
3364                 /* FIXME, shrink the extent if the ref count is only 1 */
3365                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3366                         goto delete;
3367
3368                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3369                         u64 num_dec;
3370                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3371                         if (!del_item && !encoding) {
3372                                 u64 orig_num_bytes =
3373                                         btrfs_file_extent_num_bytes(leaf, fi);
3374                                 extent_num_bytes = new_size -
3375                                         found_key.offset + root->sectorsize - 1;
3376                                 extent_num_bytes = extent_num_bytes &
3377                                         ~((u64)root->sectorsize - 1);
3378                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3379                                                          extent_num_bytes);
3380                                 num_dec = (orig_num_bytes -
3381                                            extent_num_bytes);
3382                                 if (root->ref_cows && extent_start != 0)
3383                                         inode_sub_bytes(inode, num_dec);
3384                                 btrfs_mark_buffer_dirty(leaf);
3385                         } else {
3386                                 extent_num_bytes =
3387                                         btrfs_file_extent_disk_num_bytes(leaf,
3388                                                                          fi);
3389                                 extent_offset = found_key.offset -
3390                                         btrfs_file_extent_offset(leaf, fi);
3391
3392                                 /* FIXME blocksize != 4096 */
3393                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3394                                 if (extent_start != 0) {
3395                                         found_extent = 1;
3396                                         if (root->ref_cows)
3397                                                 inode_sub_bytes(inode, num_dec);
3398                                 }
3399                         }
3400                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3401                         /*
3402                          * we can't truncate inline items that have had
3403                          * special encodings
3404                          */
3405                         if (!del_item &&
3406                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3407                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3408                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3409                                 u32 size = new_size - found_key.offset;
3410
3411                                 if (root->ref_cows) {
3412                                         inode_sub_bytes(inode, item_end + 1 -
3413                                                         new_size);
3414                                 }
3415                                 size =
3416                                     btrfs_file_extent_calc_inline_size(size);
3417                                 ret = btrfs_truncate_item(trans, root, path,
3418                                                           size, 1);
3419                                 BUG_ON(ret);
3420                         } else if (root->ref_cows) {
3421                                 inode_sub_bytes(inode, item_end + 1 -
3422                                                 found_key.offset);
3423                         }
3424                 }
3425 delete:
3426                 if (del_item) {
3427                         if (!pending_del_nr) {
3428                                 /* no pending yet, add ourselves */
3429                                 pending_del_slot = path->slots[0];
3430                                 pending_del_nr = 1;
3431                         } else if (pending_del_nr &&
3432                                    path->slots[0] + 1 == pending_del_slot) {
3433                                 /* hop on the pending chunk */
3434                                 pending_del_nr++;
3435                                 pending_del_slot = path->slots[0];
3436                         } else {
3437                                 BUG();
3438                         }
3439                 } else {
3440                         break;
3441                 }
3442                 if (found_extent && (root->ref_cows ||
3443                                      root == root->fs_info->tree_root)) {
3444                         btrfs_set_path_blocking(path);
3445                         ret = btrfs_free_extent(trans, root, extent_start,
3446                                                 extent_num_bytes, 0,
3447                                                 btrfs_header_owner(leaf),
3448                                                 inode->i_ino, extent_offset);
3449                         BUG_ON(ret);
3450                 }
3451
3452                 if (found_type == BTRFS_INODE_ITEM_KEY)
3453                         break;
3454
3455                 if (path->slots[0] == 0 ||
3456                     path->slots[0] != pending_del_slot) {
3457                         if (root->ref_cows) {
3458                                 err = -EAGAIN;
3459                                 goto out;
3460                         }
3461                         if (pending_del_nr) {
3462                                 ret = btrfs_del_items(trans, root, path,
3463                                                 pending_del_slot,
3464                                                 pending_del_nr);
3465                                 BUG_ON(ret);
3466                                 pending_del_nr = 0;
3467                         }
3468                         btrfs_release_path(root, path);
3469                         goto search_again;
3470                 } else {
3471                         path->slots[0]--;
3472                 }
3473         }
3474 out:
3475         if (pending_del_nr) {
3476                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3477                                       pending_del_nr);
3478                 BUG_ON(ret);
3479         }
3480         btrfs_free_path(path);
3481         return err;
3482 }
3483
3484 /*
3485  * taken from block_truncate_page, but does cow as it zeros out
3486  * any bytes left in the last page in the file.
3487  */
3488 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3489 {
3490         struct inode *inode = mapping->host;
3491         struct btrfs_root *root = BTRFS_I(inode)->root;
3492         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3493         struct btrfs_ordered_extent *ordered;
3494         struct extent_state *cached_state = NULL;
3495         char *kaddr;
3496         u32 blocksize = root->sectorsize;
3497         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3498         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3499         struct page *page;
3500         int ret = 0;
3501         u64 page_start;
3502         u64 page_end;
3503
3504         if ((offset & (blocksize - 1)) == 0)
3505                 goto out;
3506         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3507         if (ret)
3508                 goto out;
3509
3510         ret = -ENOMEM;
3511 again:
3512         page = grab_cache_page(mapping, index);
3513         if (!page) {
3514                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3515                 goto out;
3516         }
3517
3518         page_start = page_offset(page);
3519         page_end = page_start + PAGE_CACHE_SIZE - 1;
3520
3521         if (!PageUptodate(page)) {
3522                 ret = btrfs_readpage(NULL, page);
3523                 lock_page(page);
3524                 if (page->mapping != mapping) {
3525                         unlock_page(page);
3526                         page_cache_release(page);
3527                         goto again;
3528                 }
3529                 if (!PageUptodate(page)) {
3530                         ret = -EIO;
3531                         goto out_unlock;
3532                 }
3533         }
3534         wait_on_page_writeback(page);
3535
3536         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3537                          GFP_NOFS);
3538         set_page_extent_mapped(page);
3539
3540         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3541         if (ordered) {
3542                 unlock_extent_cached(io_tree, page_start, page_end,
3543                                      &cached_state, GFP_NOFS);
3544                 unlock_page(page);
3545                 page_cache_release(page);
3546                 btrfs_start_ordered_extent(inode, ordered, 1);
3547                 btrfs_put_ordered_extent(ordered);
3548                 goto again;
3549         }
3550
3551         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3552                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3553                           0, 0, &cached_state, GFP_NOFS);
3554
3555         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3556                                         &cached_state);
3557         if (ret) {
3558                 unlock_extent_cached(io_tree, page_start, page_end,
3559                                      &cached_state, GFP_NOFS);
3560                 goto out_unlock;
3561         }
3562
3563         ret = 0;
3564         if (offset != PAGE_CACHE_SIZE) {
3565                 kaddr = kmap(page);
3566                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3567                 flush_dcache_page(page);
3568                 kunmap(page);
3569         }
3570         ClearPageChecked(page);
3571         set_page_dirty(page);
3572         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3573                              GFP_NOFS);
3574
3575 out_unlock:
3576         if (ret)
3577                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3578         unlock_page(page);
3579         page_cache_release(page);
3580 out:
3581         return ret;
3582 }
3583
3584 /*
3585  * This function puts in dummy file extents for the area we're creating a hole
3586  * for.  So if we are truncating this file to a larger size we need to insert
3587  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3588  * the range between oldsize and size
3589  */
3590 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3591 {
3592         struct btrfs_trans_handle *trans;
3593         struct btrfs_root *root = BTRFS_I(inode)->root;
3594         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3595         struct extent_map *em = NULL;
3596         struct extent_state *cached_state = NULL;
3597         u64 mask = root->sectorsize - 1;
3598         u64 hole_start = (oldsize + mask) & ~mask;
3599         u64 block_end = (size + mask) & ~mask;
3600         u64 last_byte;
3601         u64 cur_offset;
3602         u64 hole_size;
3603         int err = 0;
3604
3605         if (size <= hole_start)
3606                 return 0;
3607
3608         while (1) {
3609                 struct btrfs_ordered_extent *ordered;
3610                 btrfs_wait_ordered_range(inode, hole_start,
3611                                          block_end - hole_start);
3612                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3613                                  &cached_state, GFP_NOFS);
3614                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3615                 if (!ordered)
3616                         break;
3617                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3618                                      &cached_state, GFP_NOFS);
3619                 btrfs_put_ordered_extent(ordered);
3620         }
3621
3622         cur_offset = hole_start;
3623         while (1) {
3624                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3625                                 block_end - cur_offset, 0);
3626                 BUG_ON(IS_ERR(em) || !em);
3627                 last_byte = min(extent_map_end(em), block_end);
3628                 last_byte = (last_byte + mask) & ~mask;
3629                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3630                         u64 hint_byte = 0;
3631                         hole_size = last_byte - cur_offset;
3632
3633                         trans = btrfs_start_transaction(root, 2);
3634                         if (IS_ERR(trans)) {
3635                                 err = PTR_ERR(trans);
3636                                 break;
3637                         }
3638                         btrfs_set_trans_block_group(trans, inode);
3639
3640                         err = btrfs_drop_extents(trans, inode, cur_offset,
3641                                                  cur_offset + hole_size,
3642                                                  &hint_byte, 1);
3643                         if (err)
3644                                 break;
3645
3646                         err = btrfs_insert_file_extent(trans, root,
3647                                         inode->i_ino, cur_offset, 0,
3648                                         0, hole_size, 0, hole_size,
3649                                         0, 0, 0);
3650                         if (err)
3651                                 break;
3652
3653                         btrfs_drop_extent_cache(inode, hole_start,
3654                                         last_byte - 1, 0);
3655
3656                         btrfs_end_transaction(trans, root);
3657                 }
3658                 free_extent_map(em);
3659                 em = NULL;
3660                 cur_offset = last_byte;
3661                 if (cur_offset >= block_end)
3662                         break;
3663         }
3664
3665         free_extent_map(em);
3666         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3667                              GFP_NOFS);
3668         return err;
3669 }
3670
3671 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3672 {
3673         loff_t oldsize = i_size_read(inode);
3674         int ret;
3675
3676         if (newsize == oldsize)
3677                 return 0;
3678
3679         if (newsize > oldsize) {
3680                 i_size_write(inode, newsize);
3681                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3682                 truncate_pagecache(inode, oldsize, newsize);
3683                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3684                 if (ret) {
3685                         btrfs_setsize(inode, oldsize);
3686                         return ret;
3687                 }
3688
3689                 mark_inode_dirty(inode);
3690         } else {
3691
3692                 /*
3693                  * We're truncating a file that used to have good data down to
3694                  * zero. Make sure it gets into the ordered flush list so that
3695                  * any new writes get down to disk quickly.
3696                  */
3697                 if (newsize == 0)
3698                         BTRFS_I(inode)->ordered_data_close = 1;
3699
3700                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3701                 truncate_setsize(inode, newsize);
3702                 ret = btrfs_truncate(inode);
3703         }
3704
3705         return ret;
3706 }
3707
3708 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3709 {
3710         struct inode *inode = dentry->d_inode;
3711         struct btrfs_root *root = BTRFS_I(inode)->root;
3712         int err;
3713
3714         if (btrfs_root_readonly(root))
3715                 return -EROFS;
3716
3717         err = inode_change_ok(inode, attr);
3718         if (err)
3719                 return err;
3720
3721         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3722                 err = btrfs_setsize(inode, attr->ia_size);
3723                 if (err)
3724                         return err;
3725         }
3726
3727         if (attr->ia_valid) {
3728                 setattr_copy(inode, attr);
3729                 mark_inode_dirty(inode);
3730
3731                 if (attr->ia_valid & ATTR_MODE)
3732                         err = btrfs_acl_chmod(inode);
3733         }
3734
3735         return err;
3736 }
3737
3738 void btrfs_evict_inode(struct inode *inode)
3739 {
3740         struct btrfs_trans_handle *trans;
3741         struct btrfs_root *root = BTRFS_I(inode)->root;
3742         unsigned long nr;
3743         int ret;
3744
3745         trace_btrfs_inode_evict(inode);
3746
3747         truncate_inode_pages(&inode->i_data, 0);
3748         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3749                                root == root->fs_info->tree_root))
3750                 goto no_delete;
3751
3752         if (is_bad_inode(inode)) {
3753                 btrfs_orphan_del(NULL, inode);
3754                 goto no_delete;
3755         }
3756         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3757         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3758
3759         if (root->fs_info->log_root_recovering) {
3760                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3761                 goto no_delete;
3762         }
3763
3764         if (inode->i_nlink > 0) {
3765                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3766                 goto no_delete;
3767         }
3768
3769         btrfs_i_size_write(inode, 0);
3770
3771         while (1) {
3772                 trans = btrfs_start_transaction(root, 0);
3773                 BUG_ON(IS_ERR(trans));
3774                 btrfs_set_trans_block_group(trans, inode);
3775                 trans->block_rsv = root->orphan_block_rsv;
3776
3777                 ret = btrfs_block_rsv_check(trans, root,
3778                                             root->orphan_block_rsv, 0, 5);
3779                 if (ret) {
3780                         BUG_ON(ret != -EAGAIN);
3781                         ret = btrfs_commit_transaction(trans, root);
3782                         BUG_ON(ret);
3783                         continue;
3784                 }
3785
3786                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3787                 if (ret != -EAGAIN)
3788                         break;
3789
3790                 nr = trans->blocks_used;
3791                 btrfs_end_transaction(trans, root);
3792                 trans = NULL;
3793                 btrfs_btree_balance_dirty(root, nr);
3794
3795         }
3796
3797         if (ret == 0) {
3798                 ret = btrfs_orphan_del(trans, inode);
3799                 BUG_ON(ret);
3800         }
3801
3802         nr = trans->blocks_used;
3803         btrfs_end_transaction(trans, root);
3804         btrfs_btree_balance_dirty(root, nr);
3805 no_delete:
3806         end_writeback(inode);
3807         return;
3808 }
3809
3810 /*
3811  * this returns the key found in the dir entry in the location pointer.
3812  * If no dir entries were found, location->objectid is 0.
3813  */
3814 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3815                                struct btrfs_key *location)
3816 {
3817         const char *name = dentry->d_name.name;
3818         int namelen = dentry->d_name.len;
3819         struct btrfs_dir_item *di;
3820         struct btrfs_path *path;
3821         struct btrfs_root *root = BTRFS_I(dir)->root;
3822         int ret = 0;
3823
3824         path = btrfs_alloc_path();
3825         BUG_ON(!path);
3826
3827         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3828                                     namelen, 0);
3829         if (IS_ERR(di))
3830                 ret = PTR_ERR(di);
3831
3832         if (!di || IS_ERR(di))
3833                 goto out_err;
3834
3835         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3836 out:
3837         btrfs_free_path(path);
3838         return ret;
3839 out_err:
3840         location->objectid = 0;
3841         goto out;
3842 }
3843
3844 /*
3845  * when we hit a tree root in a directory, the btrfs part of the inode
3846  * needs to be changed to reflect the root directory of the tree root.  This
3847  * is kind of like crossing a mount point.
3848  */
3849 static int fixup_tree_root_location(struct btrfs_root *root,
3850                                     struct inode *dir,
3851                                     struct dentry *dentry,
3852                                     struct btrfs_key *location,
3853                                     struct btrfs_root **sub_root)
3854 {
3855         struct btrfs_path *path;
3856         struct btrfs_root *new_root;
3857         struct btrfs_root_ref *ref;
3858         struct extent_buffer *leaf;
3859         int ret;
3860         int err = 0;
3861
3862         path = btrfs_alloc_path();
3863         if (!path) {
3864                 err = -ENOMEM;
3865                 goto out;
3866         }
3867
3868         err = -ENOENT;
3869         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3870                                   BTRFS_I(dir)->root->root_key.objectid,
3871                                   location->objectid);
3872         if (ret) {
3873                 if (ret < 0)
3874                         err = ret;
3875                 goto out;
3876         }
3877
3878         leaf = path->nodes[0];
3879         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3880         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3881             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3882                 goto out;
3883
3884         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3885                                    (unsigned long)(ref + 1),
3886                                    dentry->d_name.len);
3887         if (ret)
3888                 goto out;
3889
3890         btrfs_release_path(root->fs_info->tree_root, path);
3891
3892         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3893         if (IS_ERR(new_root)) {
3894                 err = PTR_ERR(new_root);
3895                 goto out;
3896         }
3897
3898         if (btrfs_root_refs(&new_root->root_item) == 0) {
3899                 err = -ENOENT;
3900                 goto out;
3901         }
3902
3903         *sub_root = new_root;
3904         location->objectid = btrfs_root_dirid(&new_root->root_item);
3905         location->type = BTRFS_INODE_ITEM_KEY;
3906         location->offset = 0;
3907         err = 0;
3908 out:
3909         btrfs_free_path(path);
3910         return err;
3911 }
3912
3913 static void inode_tree_add(struct inode *inode)
3914 {
3915         struct btrfs_root *root = BTRFS_I(inode)->root;
3916         struct btrfs_inode *entry;
3917         struct rb_node **p;
3918         struct rb_node *parent;
3919 again:
3920         p = &root->inode_tree.rb_node;
3921         parent = NULL;
3922
3923         if (inode_unhashed(inode))
3924                 return;
3925
3926         spin_lock(&root->inode_lock);
3927         while (*p) {
3928                 parent = *p;
3929                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3930
3931                 if (inode->i_ino < entry->vfs_inode.i_ino)
3932                         p = &parent->rb_left;
3933                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3934                         p = &parent->rb_right;
3935                 else {
3936                         WARN_ON(!(entry->vfs_inode.i_state &
3937                                   (I_WILL_FREE | I_FREEING)));
3938                         rb_erase(parent, &root->inode_tree);
3939                         RB_CLEAR_NODE(parent);
3940                         spin_unlock(&root->inode_lock);
3941                         goto again;
3942                 }
3943         }
3944         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3945         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3946         spin_unlock(&root->inode_lock);
3947 }
3948
3949 static void inode_tree_del(struct inode *inode)
3950 {
3951         struct btrfs_root *root = BTRFS_I(inode)->root;
3952         int empty = 0;
3953
3954         spin_lock(&root->inode_lock);
3955         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3956                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3957                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3958                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3959         }
3960         spin_unlock(&root->inode_lock);
3961
3962         /*
3963          * Free space cache has inodes in the tree root, but the tree root has a
3964          * root_refs of 0, so this could end up dropping the tree root as a
3965          * snapshot, so we need the extra !root->fs_info->tree_root check to
3966          * make sure we don't drop it.
3967          */
3968         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3969             root != root->fs_info->tree_root) {
3970                 synchronize_srcu(&root->fs_info->subvol_srcu);
3971                 spin_lock(&root->inode_lock);
3972                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3973                 spin_unlock(&root->inode_lock);
3974                 if (empty)
3975                         btrfs_add_dead_root(root);
3976         }
3977 }
3978
3979 int btrfs_invalidate_inodes(struct btrfs_root *root)
3980 {
3981         struct rb_node *node;
3982         struct rb_node *prev;
3983         struct btrfs_inode *entry;
3984         struct inode *inode;
3985         u64 objectid = 0;
3986
3987         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3988
3989         spin_lock(&root->inode_lock);
3990 again:
3991         node = root->inode_tree.rb_node;
3992         prev = NULL;
3993         while (node) {
3994                 prev = node;
3995                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3996
3997                 if (objectid < entry->vfs_inode.i_ino)
3998                         node = node->rb_left;
3999                 else if (objectid > entry->vfs_inode.i_ino)
4000                         node = node->rb_right;
4001                 else
4002                         break;
4003         }
4004         if (!node) {
4005                 while (prev) {
4006                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4007                         if (objectid <= entry->vfs_inode.i_ino) {
4008                                 node = prev;
4009                                 break;
4010                         }
4011                         prev = rb_next(prev);
4012                 }
4013         }
4014         while (node) {
4015                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4016                 objectid = entry->vfs_inode.i_ino + 1;
4017                 inode = igrab(&entry->vfs_inode);
4018                 if (inode) {
4019                         spin_unlock(&root->inode_lock);
4020                         if (atomic_read(&inode->i_count) > 1)
4021                                 d_prune_aliases(inode);
4022                         /*
4023                          * btrfs_drop_inode will have it removed from
4024                          * the inode cache when its usage count
4025                          * hits zero.
4026                          */
4027                         iput(inode);
4028                         cond_resched();
4029                         spin_lock(&root->inode_lock);
4030                         goto again;
4031                 }
4032
4033                 if (cond_resched_lock(&root->inode_lock))
4034                         goto again;
4035
4036                 node = rb_next(node);
4037         }
4038         spin_unlock(&root->inode_lock);
4039         return 0;
4040 }
4041
4042 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4043 {
4044         struct btrfs_iget_args *args = p;
4045         inode->i_ino = args->ino;
4046         BTRFS_I(inode)->root = args->root;
4047         btrfs_set_inode_space_info(args->root, inode);
4048         return 0;
4049 }
4050
4051 static int btrfs_find_actor(struct inode *inode, void *opaque)
4052 {
4053         struct btrfs_iget_args *args = opaque;
4054         return args->ino == inode->i_ino &&
4055                 args->root == BTRFS_I(inode)->root;
4056 }
4057
4058 static struct inode *btrfs_iget_locked(struct super_block *s,
4059                                        u64 objectid,
4060                                        struct btrfs_root *root)
4061 {
4062         struct inode *inode;
4063         struct btrfs_iget_args args;
4064         args.ino = objectid;
4065         args.root = root;
4066
4067         inode = iget5_locked(s, objectid, btrfs_find_actor,
4068                              btrfs_init_locked_inode,
4069                              (void *)&args);
4070         return inode;
4071 }
4072
4073 /* Get an inode object given its location and corresponding root.
4074  * Returns in *is_new if the inode was read from disk
4075  */
4076 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4077                          struct btrfs_root *root, int *new)
4078 {
4079         struct inode *inode;
4080
4081         inode = btrfs_iget_locked(s, location->objectid, root);
4082         if (!inode)
4083                 return ERR_PTR(-ENOMEM);
4084
4085         if (inode->i_state & I_NEW) {
4086                 BTRFS_I(inode)->root = root;
4087                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4088                 btrfs_read_locked_inode(inode);
4089                 inode_tree_add(inode);
4090                 unlock_new_inode(inode);
4091                 if (new)
4092                         *new = 1;
4093         }
4094
4095         return inode;
4096 }
4097
4098 static struct inode *new_simple_dir(struct super_block *s,
4099                                     struct btrfs_key *key,
4100                                     struct btrfs_root *root)
4101 {
4102         struct inode *inode = new_inode(s);
4103
4104         if (!inode)
4105                 return ERR_PTR(-ENOMEM);
4106
4107         BTRFS_I(inode)->root = root;
4108         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4109         BTRFS_I(inode)->dummy_inode = 1;
4110
4111         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4112         inode->i_op = &simple_dir_inode_operations;
4113         inode->i_fop = &simple_dir_operations;
4114         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4115         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4116
4117         return inode;
4118 }
4119
4120 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4121 {
4122         struct inode *inode;
4123         struct btrfs_root *root = BTRFS_I(dir)->root;
4124         struct btrfs_root *sub_root = root;
4125         struct btrfs_key location;
4126         int index;
4127         int ret;
4128
4129         if (dentry->d_name.len > BTRFS_NAME_LEN)
4130                 return ERR_PTR(-ENAMETOOLONG);
4131
4132         ret = btrfs_inode_by_name(dir, dentry, &location);
4133
4134         if (ret < 0)
4135                 return ERR_PTR(ret);
4136
4137         if (location.objectid == 0)
4138                 return NULL;
4139
4140         if (location.type == BTRFS_INODE_ITEM_KEY) {
4141                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4142                 return inode;
4143         }
4144
4145         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4146
4147         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4148         ret = fixup_tree_root_location(root, dir, dentry,
4149                                        &location, &sub_root);
4150         if (ret < 0) {
4151                 if (ret != -ENOENT)
4152                         inode = ERR_PTR(ret);
4153                 else
4154                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4155         } else {
4156                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4157         }
4158         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4159
4160         if (!IS_ERR(inode) && root != sub_root) {
4161                 down_read(&root->fs_info->cleanup_work_sem);
4162                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4163                         ret = btrfs_orphan_cleanup(sub_root);
4164                 up_read(&root->fs_info->cleanup_work_sem);
4165                 if (ret)
4166                         inode = ERR_PTR(ret);
4167         }
4168
4169         return inode;
4170 }
4171
4172 static int btrfs_dentry_delete(const struct dentry *dentry)
4173 {
4174         struct btrfs_root *root;
4175
4176         if (!dentry->d_inode && !IS_ROOT(dentry))
4177                 dentry = dentry->d_parent;
4178
4179         if (dentry->d_inode) {
4180                 root = BTRFS_I(dentry->d_inode)->root;
4181                 if (btrfs_root_refs(&root->root_item) == 0)
4182                         return 1;
4183         }
4184         return 0;
4185 }
4186
4187 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4188                                    struct nameidata *nd)
4189 {
4190         struct inode *inode;
4191
4192         inode = btrfs_lookup_dentry(dir, dentry);
4193         if (IS_ERR(inode))
4194                 return ERR_CAST(inode);
4195
4196         return d_splice_alias(inode, dentry);
4197 }
4198
4199 static unsigned char btrfs_filetype_table[] = {
4200         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4201 };
4202
4203 static int btrfs_real_readdir(struct file *filp, void *dirent,
4204                               filldir_t filldir)
4205 {
4206         struct inode *inode = filp->f_dentry->d_inode;
4207         struct btrfs_root *root = BTRFS_I(inode)->root;
4208         struct btrfs_item *item;
4209         struct btrfs_dir_item *di;
4210         struct btrfs_key key;
4211         struct btrfs_key found_key;
4212         struct btrfs_path *path;
4213         int ret;
4214         u32 nritems;
4215         struct extent_buffer *leaf;
4216         int slot;
4217         int advance;
4218         unsigned char d_type;
4219         int over = 0;
4220         u32 di_cur;
4221         u32 di_total;
4222         u32 di_len;
4223         int key_type = BTRFS_DIR_INDEX_KEY;
4224         char tmp_name[32];
4225         char *name_ptr;
4226         int name_len;
4227
4228         /* FIXME, use a real flag for deciding about the key type */
4229         if (root->fs_info->tree_root == root)
4230                 key_type = BTRFS_DIR_ITEM_KEY;
4231
4232         /* special case for "." */
4233         if (filp->f_pos == 0) {
4234                 over = filldir(dirent, ".", 1,
4235                                1, inode->i_ino,
4236                                DT_DIR);
4237                 if (over)
4238                         return 0;
4239                 filp->f_pos = 1;
4240         }
4241         /* special case for .., just use the back ref */
4242         if (filp->f_pos == 1) {
4243                 u64 pino = parent_ino(filp->f_path.dentry);
4244                 over = filldir(dirent, "..", 2,
4245                                2, pino, DT_DIR);
4246                 if (over)
4247                         return 0;
4248                 filp->f_pos = 2;
4249         }
4250         path = btrfs_alloc_path();
4251         path->reada = 2;
4252
4253         btrfs_set_key_type(&key, key_type);
4254         key.offset = filp->f_pos;
4255         key.objectid = inode->i_ino;
4256
4257         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4258         if (ret < 0)
4259                 goto err;
4260         advance = 0;
4261
4262         while (1) {
4263                 leaf = path->nodes[0];
4264                 nritems = btrfs_header_nritems(leaf);
4265                 slot = path->slots[0];
4266                 if (advance || slot >= nritems) {
4267                         if (slot >= nritems - 1) {
4268                                 ret = btrfs_next_leaf(root, path);
4269                                 if (ret)
4270                                         break;
4271                                 leaf = path->nodes[0];
4272                                 nritems = btrfs_header_nritems(leaf);
4273                                 slot = path->slots[0];
4274                         } else {
4275                                 slot++;
4276                                 path->slots[0]++;
4277                         }
4278                 }
4279
4280                 advance = 1;
4281                 item = btrfs_item_nr(leaf, slot);
4282                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4283
4284                 if (found_key.objectid != key.objectid)
4285                         break;
4286                 if (btrfs_key_type(&found_key) != key_type)
4287                         break;
4288                 if (found_key.offset < filp->f_pos)
4289                         continue;
4290
4291                 filp->f_pos = found_key.offset;
4292
4293                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4294                 di_cur = 0;
4295                 di_total = btrfs_item_size(leaf, item);
4296
4297                 while (di_cur < di_total) {
4298                         struct btrfs_key location;
4299
4300                         if (verify_dir_item(root, leaf, di))
4301                                 break;
4302
4303                         name_len = btrfs_dir_name_len(leaf, di);
4304                         if (name_len <= sizeof(tmp_name)) {
4305                                 name_ptr = tmp_name;
4306                         } else {
4307                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4308                                 if (!name_ptr) {
4309                                         ret = -ENOMEM;
4310                                         goto err;
4311                                 }
4312                         }
4313                         read_extent_buffer(leaf, name_ptr,
4314                                            (unsigned long)(di + 1), name_len);
4315
4316                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4317                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4318
4319                         /* is this a reference to our own snapshot? If so
4320                          * skip it
4321                          */
4322                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4323                             location.objectid == root->root_key.objectid) {
4324                                 over = 0;
4325                                 goto skip;
4326                         }
4327                         over = filldir(dirent, name_ptr, name_len,
4328                                        found_key.offset, location.objectid,
4329                                        d_type);
4330
4331 skip:
4332                         if (name_ptr != tmp_name)
4333                                 kfree(name_ptr);
4334
4335                         if (over)
4336                                 goto nopos;
4337                         di_len = btrfs_dir_name_len(leaf, di) +
4338                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4339                         di_cur += di_len;
4340                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4341                 }
4342         }
4343
4344         /* Reached end of directory/root. Bump pos past the last item. */
4345         if (key_type == BTRFS_DIR_INDEX_KEY)
4346                 /*
4347                  * 32-bit glibc will use getdents64, but then strtol -
4348                  * so the last number we can serve is this.
4349                  */
4350                 filp->f_pos = 0x7fffffff;
4351         else
4352                 filp->f_pos++;
4353 nopos:
4354         ret = 0;
4355 err:
4356         btrfs_free_path(path);
4357         return ret;
4358 }
4359
4360 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4361 {
4362         struct btrfs_root *root = BTRFS_I(inode)->root;
4363         struct btrfs_trans_handle *trans;
4364         int ret = 0;
4365         bool nolock = false;
4366
4367         if (BTRFS_I(inode)->dummy_inode)
4368                 return 0;
4369
4370         smp_mb();
4371         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4372
4373         if (wbc->sync_mode == WB_SYNC_ALL) {
4374                 if (nolock)
4375                         trans = btrfs_join_transaction_nolock(root, 1);
4376                 else
4377                         trans = btrfs_join_transaction(root, 1);
4378                 if (IS_ERR(trans))
4379                         return PTR_ERR(trans);
4380                 btrfs_set_trans_block_group(trans, inode);
4381                 if (nolock)
4382                         ret = btrfs_end_transaction_nolock(trans, root);
4383                 else
4384                         ret = btrfs_commit_transaction(trans, root);
4385         }
4386         return ret;
4387 }
4388
4389 /*
4390  * This is somewhat expensive, updating the tree every time the
4391  * inode changes.  But, it is most likely to find the inode in cache.
4392  * FIXME, needs more benchmarking...there are no reasons other than performance
4393  * to keep or drop this code.
4394  */
4395 void btrfs_dirty_inode(struct inode *inode)
4396 {
4397         struct btrfs_root *root = BTRFS_I(inode)->root;
4398         struct btrfs_trans_handle *trans;
4399         int ret;
4400
4401         if (BTRFS_I(inode)->dummy_inode)
4402                 return;
4403
4404         trans = btrfs_join_transaction(root, 1);
4405         BUG_ON(IS_ERR(trans));
4406         btrfs_set_trans_block_group(trans, inode);
4407
4408         ret = btrfs_update_inode(trans, root, inode);
4409         if (ret && ret == -ENOSPC) {
4410                 /* whoops, lets try again with the full transaction */
4411                 btrfs_end_transaction(trans, root);
4412                 trans = btrfs_start_transaction(root, 1);
4413                 if (IS_ERR(trans)) {
4414                         if (printk_ratelimit()) {
4415                                 printk(KERN_ERR "btrfs: fail to "
4416                                        "dirty  inode %lu error %ld\n",
4417                                        inode->i_ino, PTR_ERR(trans));
4418                         }
4419                         return;
4420                 }
4421                 btrfs_set_trans_block_group(trans, inode);
4422
4423                 ret = btrfs_update_inode(trans, root, inode);
4424                 if (ret) {
4425                         if (printk_ratelimit()) {
4426                                 printk(KERN_ERR "btrfs: fail to "
4427                                        "dirty  inode %lu error %d\n",
4428                                        inode->i_ino, ret);
4429                         }
4430                 }
4431         }
4432         btrfs_end_transaction(trans, root);
4433 }
4434
4435 /*
4436  * find the highest existing sequence number in a directory
4437  * and then set the in-memory index_cnt variable to reflect
4438  * free sequence numbers
4439  */
4440 static int btrfs_set_inode_index_count(struct inode *inode)
4441 {
4442         struct btrfs_root *root = BTRFS_I(inode)->root;
4443         struct btrfs_key key, found_key;
4444         struct btrfs_path *path;
4445         struct extent_buffer *leaf;
4446         int ret;
4447
4448         key.objectid = inode->i_ino;
4449         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4450         key.offset = (u64)-1;
4451
4452         path = btrfs_alloc_path();
4453         if (!path)
4454                 return -ENOMEM;
4455
4456         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4457         if (ret < 0)
4458                 goto out;
4459         /* FIXME: we should be able to handle this */
4460         if (ret == 0)
4461                 goto out;
4462         ret = 0;
4463
4464         /*
4465          * MAGIC NUMBER EXPLANATION:
4466          * since we search a directory based on f_pos we have to start at 2
4467          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4468          * else has to start at 2
4469          */
4470         if (path->slots[0] == 0) {
4471                 BTRFS_I(inode)->index_cnt = 2;
4472                 goto out;
4473         }
4474
4475         path->slots[0]--;
4476
4477         leaf = path->nodes[0];
4478         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4479
4480         if (found_key.objectid != inode->i_ino ||
4481             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4482                 BTRFS_I(inode)->index_cnt = 2;
4483                 goto out;
4484         }
4485
4486         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4487 out:
4488         btrfs_free_path(path);
4489         return ret;
4490 }
4491
4492 /*
4493  * helper to find a free sequence number in a given directory.  This current
4494  * code is very simple, later versions will do smarter things in the btree
4495  */
4496 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4497 {
4498         int ret = 0;
4499
4500         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4501                 ret = btrfs_set_inode_index_count(dir);
4502                 if (ret)
4503                         return ret;
4504         }
4505
4506         *index = BTRFS_I(dir)->index_cnt;
4507         BTRFS_I(dir)->index_cnt++;
4508
4509         return ret;
4510 }
4511
4512 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4513                                      struct btrfs_root *root,
4514                                      struct inode *dir,
4515                                      const char *name, int name_len,
4516                                      u64 ref_objectid, u64 objectid,
4517                                      u64 alloc_hint, int mode, u64 *index)
4518 {
4519         struct inode *inode;
4520         struct btrfs_inode_item *inode_item;
4521         struct btrfs_key *location;
4522         struct btrfs_path *path;
4523         struct btrfs_inode_ref *ref;
4524         struct btrfs_key key[2];
4525         u32 sizes[2];
4526         unsigned long ptr;
4527         int ret;
4528         int owner;
4529
4530         path = btrfs_alloc_path();
4531         BUG_ON(!path);
4532
4533         inode = new_inode(root->fs_info->sb);
4534         if (!inode)
4535                 return ERR_PTR(-ENOMEM);
4536
4537         if (dir) {
4538                 trace_btrfs_inode_request(dir);
4539
4540                 ret = btrfs_set_inode_index(dir, index);
4541                 if (ret) {
4542                         iput(inode);
4543                         return ERR_PTR(ret);
4544                 }
4545         }
4546         /*
4547          * index_cnt is ignored for everything but a dir,
4548          * btrfs_get_inode_index_count has an explanation for the magic
4549          * number
4550          */
4551         BTRFS_I(inode)->index_cnt = 2;
4552         BTRFS_I(inode)->root = root;
4553         BTRFS_I(inode)->generation = trans->transid;
4554         inode->i_generation = BTRFS_I(inode)->generation;
4555         btrfs_set_inode_space_info(root, inode);
4556
4557         if (mode & S_IFDIR)
4558                 owner = 0;
4559         else
4560                 owner = 1;
4561         BTRFS_I(inode)->block_group =
4562                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4563
4564         key[0].objectid = objectid;
4565         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4566         key[0].offset = 0;
4567
4568         key[1].objectid = objectid;
4569         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4570         key[1].offset = ref_objectid;
4571
4572         sizes[0] = sizeof(struct btrfs_inode_item);
4573         sizes[1] = name_len + sizeof(*ref);
4574
4575         path->leave_spinning = 1;
4576         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4577         if (ret != 0)
4578                 goto fail;
4579
4580         inode_init_owner(inode, dir, mode);
4581         inode->i_ino = objectid;
4582         inode_set_bytes(inode, 0);
4583         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4584         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4585                                   struct btrfs_inode_item);
4586         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4587
4588         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4589                              struct btrfs_inode_ref);
4590         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4591         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4592         ptr = (unsigned long)(ref + 1);
4593         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4594
4595         btrfs_mark_buffer_dirty(path->nodes[0]);
4596         btrfs_free_path(path);
4597
4598         location = &BTRFS_I(inode)->location;
4599         location->objectid = objectid;
4600         location->offset = 0;
4601         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4602
4603         btrfs_inherit_iflags(inode, dir);
4604
4605         if ((mode & S_IFREG)) {
4606                 if (btrfs_test_opt(root, NODATASUM))
4607                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4608                 if (btrfs_test_opt(root, NODATACOW) ||
4609                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4610                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4611         }
4612
4613         insert_inode_hash(inode);
4614         inode_tree_add(inode);
4615
4616         trace_btrfs_inode_new(inode);
4617
4618         return inode;
4619 fail:
4620         if (dir)
4621                 BTRFS_I(dir)->index_cnt--;
4622         btrfs_free_path(path);
4623         iput(inode);
4624         return ERR_PTR(ret);
4625 }
4626
4627 static inline u8 btrfs_inode_type(struct inode *inode)
4628 {
4629         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4630 }
4631
4632 /*
4633  * utility function to add 'inode' into 'parent_inode' with
4634  * a give name and a given sequence number.
4635  * if 'add_backref' is true, also insert a backref from the
4636  * inode to the parent directory.
4637  */
4638 int btrfs_add_link(struct btrfs_trans_handle *trans,
4639                    struct inode *parent_inode, struct inode *inode,
4640                    const char *name, int name_len, int add_backref, u64 index)
4641 {
4642         int ret = 0;
4643         struct btrfs_key key;
4644         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4645
4646         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4647                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4648         } else {
4649                 key.objectid = inode->i_ino;
4650                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4651                 key.offset = 0;
4652         }
4653
4654         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4655                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4656                                          key.objectid, root->root_key.objectid,
4657                                          parent_inode->i_ino,
4658                                          index, name, name_len);
4659         } else if (add_backref) {
4660                 ret = btrfs_insert_inode_ref(trans, root,
4661                                              name, name_len, inode->i_ino,
4662                                              parent_inode->i_ino, index);
4663         }
4664
4665         if (ret == 0) {
4666                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4667                                             parent_inode->i_ino, &key,
4668                                             btrfs_inode_type(inode), index);
4669                 BUG_ON(ret);
4670
4671                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4672                                    name_len * 2);
4673                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4674                 ret = btrfs_update_inode(trans, root, parent_inode);
4675         }
4676         return ret;
4677 }
4678
4679 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4680                             struct inode *dir, struct dentry *dentry,
4681                             struct inode *inode, int backref, u64 index)
4682 {
4683         int err = btrfs_add_link(trans, dir, inode,
4684                                  dentry->d_name.name, dentry->d_name.len,
4685                                  backref, index);
4686         if (!err) {
4687                 d_instantiate(dentry, inode);
4688                 return 0;
4689         }
4690         if (err > 0)
4691                 err = -EEXIST;
4692         return err;
4693 }
4694
4695 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4696                         int mode, dev_t rdev)
4697 {
4698         struct btrfs_trans_handle *trans;
4699         struct btrfs_root *root = BTRFS_I(dir)->root;
4700         struct inode *inode = NULL;
4701         int err;
4702         int drop_inode = 0;
4703         u64 objectid;
4704         unsigned long nr = 0;
4705         u64 index = 0;
4706
4707         if (!new_valid_dev(rdev))
4708                 return -EINVAL;
4709
4710         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4711         if (err)
4712                 return err;
4713
4714         /*
4715          * 2 for inode item and ref
4716          * 2 for dir items
4717          * 1 for xattr if selinux is on
4718          */
4719         trans = btrfs_start_transaction(root, 5);
4720         if (IS_ERR(trans))
4721                 return PTR_ERR(trans);
4722
4723         btrfs_set_trans_block_group(trans, dir);
4724
4725         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4726                                 dentry->d_name.len, dir->i_ino, objectid,
4727                                 BTRFS_I(dir)->block_group, mode, &index);
4728         err = PTR_ERR(inode);
4729         if (IS_ERR(inode))
4730                 goto out_unlock;
4731
4732         err = btrfs_init_inode_security(trans, inode, dir);
4733         if (err) {
4734                 drop_inode = 1;
4735                 goto out_unlock;
4736         }
4737
4738         btrfs_set_trans_block_group(trans, inode);
4739         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4740         if (err)
4741                 drop_inode = 1;
4742         else {
4743                 inode->i_op = &btrfs_special_inode_operations;
4744                 init_special_inode(inode, inode->i_mode, rdev);
4745                 btrfs_update_inode(trans, root, inode);
4746         }
4747         btrfs_update_inode_block_group(trans, inode);
4748         btrfs_update_inode_block_group(trans, dir);
4749 out_unlock:
4750         nr = trans->blocks_used;
4751         btrfs_end_transaction_throttle(trans, root);
4752         btrfs_btree_balance_dirty(root, nr);
4753         if (drop_inode) {
4754                 inode_dec_link_count(inode);
4755                 iput(inode);
4756         }
4757         return err;
4758 }
4759
4760 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4761                         int mode, struct nameidata *nd)
4762 {
4763         struct btrfs_trans_handle *trans;
4764         struct btrfs_root *root = BTRFS_I(dir)->root;
4765         struct inode *inode = NULL;
4766         int drop_inode = 0;
4767         int err;
4768         unsigned long nr = 0;
4769         u64 objectid;
4770         u64 index = 0;
4771
4772         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4773         if (err)
4774                 return err;
4775         /*
4776          * 2 for inode item and ref
4777          * 2 for dir items
4778          * 1 for xattr if selinux is on
4779          */
4780         trans = btrfs_start_transaction(root, 5);
4781         if (IS_ERR(trans))
4782                 return PTR_ERR(trans);
4783
4784         btrfs_set_trans_block_group(trans, dir);
4785
4786         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4787                                 dentry->d_name.len, dir->i_ino, objectid,
4788                                 BTRFS_I(dir)->block_group, mode, &index);
4789         err = PTR_ERR(inode);
4790         if (IS_ERR(inode))
4791                 goto out_unlock;
4792
4793         err = btrfs_init_inode_security(trans, inode, dir);
4794         if (err) {
4795                 drop_inode = 1;
4796                 goto out_unlock;
4797         }
4798
4799         btrfs_set_trans_block_group(trans, inode);
4800         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4801         if (err)
4802                 drop_inode = 1;
4803         else {
4804                 inode->i_mapping->a_ops = &btrfs_aops;
4805                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4806                 inode->i_fop = &btrfs_file_operations;
4807                 inode->i_op = &btrfs_file_inode_operations;
4808                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4809         }
4810         btrfs_update_inode_block_group(trans, inode);
4811         btrfs_update_inode_block_group(trans, dir);
4812 out_unlock:
4813         nr = trans->blocks_used;
4814         btrfs_end_transaction_throttle(trans, root);
4815         if (drop_inode) {
4816                 inode_dec_link_count(inode);
4817                 iput(inode);
4818         }
4819         btrfs_btree_balance_dirty(root, nr);
4820         return err;
4821 }
4822
4823 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4824                       struct dentry *dentry)
4825 {
4826         struct btrfs_trans_handle *trans;
4827         struct btrfs_root *root = BTRFS_I(dir)->root;
4828         struct inode *inode = old_dentry->d_inode;
4829         u64 index;
4830         unsigned long nr = 0;
4831         int err;
4832         int drop_inode = 0;
4833
4834         if (inode->i_nlink == 0)
4835                 return -ENOENT;
4836
4837         /* do not allow sys_link's with other subvols of the same device */
4838         if (root->objectid != BTRFS_I(inode)->root->objectid)
4839                 return -EXDEV;
4840
4841         if (inode->i_nlink == ~0U)
4842                 return -EMLINK;
4843
4844         btrfs_inc_nlink(inode);
4845         inode->i_ctime = CURRENT_TIME;
4846
4847         err = btrfs_set_inode_index(dir, &index);
4848         if (err)
4849                 goto fail;
4850
4851         /*
4852          * 2 items for inode and inode ref
4853          * 2 items for dir items
4854          * 1 item for parent inode
4855          */
4856         trans = btrfs_start_transaction(root, 5);
4857         if (IS_ERR(trans)) {
4858                 err = PTR_ERR(trans);
4859                 goto fail;
4860         }
4861
4862         btrfs_set_trans_block_group(trans, dir);
4863         ihold(inode);
4864
4865         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4866
4867         if (err) {
4868                 drop_inode = 1;
4869         } else {
4870                 struct dentry *parent = dget_parent(dentry);
4871                 btrfs_update_inode_block_group(trans, dir);
4872                 err = btrfs_update_inode(trans, root, inode);
4873                 BUG_ON(err);
4874                 btrfs_log_new_name(trans, inode, NULL, parent);
4875                 dput(parent);
4876         }
4877
4878         nr = trans->blocks_used;
4879         btrfs_end_transaction_throttle(trans, root);
4880 fail:
4881         if (drop_inode) {
4882                 inode_dec_link_count(inode);
4883                 iput(inode);
4884         }
4885         btrfs_btree_balance_dirty(root, nr);
4886         return err;
4887 }
4888
4889 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4890 {
4891         struct inode *inode = NULL;
4892         struct btrfs_trans_handle *trans;
4893         struct btrfs_root *root = BTRFS_I(dir)->root;
4894         int err = 0;
4895         int drop_on_err = 0;
4896         u64 objectid = 0;
4897         u64 index = 0;
4898         unsigned long nr = 1;
4899
4900         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4901         if (err)
4902                 return err;
4903
4904         /*
4905          * 2 items for inode and ref
4906          * 2 items for dir items
4907          * 1 for xattr if selinux is on
4908          */
4909         trans = btrfs_start_transaction(root, 5);
4910         if (IS_ERR(trans))
4911                 return PTR_ERR(trans);
4912         btrfs_set_trans_block_group(trans, dir);
4913
4914         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4915                                 dentry->d_name.len, dir->i_ino, objectid,
4916                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4917                                 &index);
4918         if (IS_ERR(inode)) {
4919                 err = PTR_ERR(inode);
4920                 goto out_fail;
4921         }
4922
4923         drop_on_err = 1;
4924
4925         err = btrfs_init_inode_security(trans, inode, dir);
4926         if (err)
4927                 goto out_fail;
4928
4929         inode->i_op = &btrfs_dir_inode_operations;
4930         inode->i_fop = &btrfs_dir_file_operations;
4931         btrfs_set_trans_block_group(trans, inode);
4932
4933         btrfs_i_size_write(inode, 0);
4934         err = btrfs_update_inode(trans, root, inode);
4935         if (err)
4936                 goto out_fail;
4937
4938         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4939                              dentry->d_name.len, 0, index);
4940         if (err)
4941                 goto out_fail;
4942
4943         d_instantiate(dentry, inode);
4944         drop_on_err = 0;
4945         btrfs_update_inode_block_group(trans, inode);
4946         btrfs_update_inode_block_group(trans, dir);
4947
4948 out_fail:
4949         nr = trans->blocks_used;
4950         btrfs_end_transaction_throttle(trans, root);
4951         if (drop_on_err)
4952                 iput(inode);
4953         btrfs_btree_balance_dirty(root, nr);
4954         return err;
4955 }
4956
4957 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4958  * and an extent that you want to insert, deal with overlap and insert
4959  * the new extent into the tree.
4960  */
4961 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4962                                 struct extent_map *existing,
4963                                 struct extent_map *em,
4964                                 u64 map_start, u64 map_len)
4965 {
4966         u64 start_diff;
4967
4968         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4969         start_diff = map_start - em->start;
4970         em->start = map_start;
4971         em->len = map_len;
4972         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4973             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4974                 em->block_start += start_diff;
4975                 em->block_len -= start_diff;
4976         }
4977         return add_extent_mapping(em_tree, em);
4978 }
4979
4980 static noinline int uncompress_inline(struct btrfs_path *path,
4981                                       struct inode *inode, struct page *page,
4982                                       size_t pg_offset, u64 extent_offset,
4983                                       struct btrfs_file_extent_item *item)
4984 {
4985         int ret;
4986         struct extent_buffer *leaf = path->nodes[0];
4987         char *tmp;
4988         size_t max_size;
4989         unsigned long inline_size;
4990         unsigned long ptr;
4991         int compress_type;
4992
4993         WARN_ON(pg_offset != 0);
4994         compress_type = btrfs_file_extent_compression(leaf, item);
4995         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4996         inline_size = btrfs_file_extent_inline_item_len(leaf,
4997                                         btrfs_item_nr(leaf, path->slots[0]));
4998         tmp = kmalloc(inline_size, GFP_NOFS);
4999         ptr = btrfs_file_extent_inline_start(item);
5000
5001         read_extent_buffer(leaf, tmp, ptr, inline_size);
5002
5003         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5004         ret = btrfs_decompress(compress_type, tmp, page,
5005                                extent_offset, inline_size, max_size);
5006         if (ret) {
5007                 char *kaddr = kmap_atomic(page, KM_USER0);
5008                 unsigned long copy_size = min_t(u64,
5009                                   PAGE_CACHE_SIZE - pg_offset,
5010                                   max_size - extent_offset);
5011                 memset(kaddr + pg_offset, 0, copy_size);
5012                 kunmap_atomic(kaddr, KM_USER0);
5013         }
5014         kfree(tmp);
5015         return 0;
5016 }
5017
5018 /*
5019  * a bit scary, this does extent mapping from logical file offset to the disk.
5020  * the ugly parts come from merging extents from the disk with the in-ram
5021  * representation.  This gets more complex because of the data=ordered code,
5022  * where the in-ram extents might be locked pending data=ordered completion.
5023  *
5024  * This also copies inline extents directly into the page.
5025  */
5026
5027 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5028                                     size_t pg_offset, u64 start, u64 len,
5029                                     int create)
5030 {
5031         int ret;
5032         int err = 0;
5033         u64 bytenr;
5034         u64 extent_start = 0;
5035         u64 extent_end = 0;
5036         u64 objectid = inode->i_ino;
5037         u32 found_type;
5038         struct btrfs_path *path = NULL;
5039         struct btrfs_root *root = BTRFS_I(inode)->root;
5040         struct btrfs_file_extent_item *item;
5041         struct extent_buffer *leaf;
5042         struct btrfs_key found_key;
5043         struct extent_map *em = NULL;
5044         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5045         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5046         struct btrfs_trans_handle *trans = NULL;
5047         int compress_type;
5048
5049 again:
5050         read_lock(&em_tree->lock);
5051         em = lookup_extent_mapping(em_tree, start, len);
5052         if (em)
5053                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5054         read_unlock(&em_tree->lock);
5055
5056         if (em) {
5057                 if (em->start > start || em->start + em->len <= start)
5058                         free_extent_map(em);
5059                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5060                         free_extent_map(em);
5061                 else
5062                         goto out;
5063         }
5064         em = alloc_extent_map(GFP_NOFS);
5065         if (!em) {
5066                 err = -ENOMEM;
5067                 goto out;
5068         }
5069         em->bdev = root->fs_info->fs_devices->latest_bdev;
5070         em->start = EXTENT_MAP_HOLE;
5071         em->orig_start = EXTENT_MAP_HOLE;
5072         em->len = (u64)-1;
5073         em->block_len = (u64)-1;
5074
5075         if (!path) {
5076                 path = btrfs_alloc_path();
5077                 BUG_ON(!path);
5078         }
5079
5080         ret = btrfs_lookup_file_extent(trans, root, path,
5081                                        objectid, start, trans != NULL);
5082         if (ret < 0) {
5083                 err = ret;
5084                 goto out;
5085         }
5086
5087         if (ret != 0) {
5088                 if (path->slots[0] == 0)
5089                         goto not_found;
5090                 path->slots[0]--;
5091         }
5092
5093         leaf = path->nodes[0];
5094         item = btrfs_item_ptr(leaf, path->slots[0],
5095                               struct btrfs_file_extent_item);
5096         /* are we inside the extent that was found? */
5097         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5098         found_type = btrfs_key_type(&found_key);
5099         if (found_key.objectid != objectid ||
5100             found_type != BTRFS_EXTENT_DATA_KEY) {
5101                 goto not_found;
5102         }
5103
5104         found_type = btrfs_file_extent_type(leaf, item);
5105         extent_start = found_key.offset;
5106         compress_type = btrfs_file_extent_compression(leaf, item);
5107         if (found_type == BTRFS_FILE_EXTENT_REG ||
5108             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5109                 extent_end = extent_start +
5110                        btrfs_file_extent_num_bytes(leaf, item);
5111         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5112                 size_t size;
5113                 size = btrfs_file_extent_inline_len(leaf, item);
5114                 extent_end = (extent_start + size + root->sectorsize - 1) &
5115                         ~((u64)root->sectorsize - 1);
5116         }
5117
5118         if (start >= extent_end) {
5119                 path->slots[0]++;
5120                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5121                         ret = btrfs_next_leaf(root, path);
5122                         if (ret < 0) {
5123                                 err = ret;
5124                                 goto out;
5125                         }
5126                         if (ret > 0)
5127                                 goto not_found;
5128                         leaf = path->nodes[0];
5129                 }
5130                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5131                 if (found_key.objectid != objectid ||
5132                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5133                         goto not_found;
5134                 if (start + len <= found_key.offset)
5135                         goto not_found;
5136                 em->start = start;
5137                 em->len = found_key.offset - start;
5138                 goto not_found_em;
5139         }
5140
5141         if (found_type == BTRFS_FILE_EXTENT_REG ||
5142             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5143                 em->start = extent_start;
5144                 em->len = extent_end - extent_start;
5145                 em->orig_start = extent_start -
5146                                  btrfs_file_extent_offset(leaf, item);
5147                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5148                 if (bytenr == 0) {
5149                         em->block_start = EXTENT_MAP_HOLE;
5150                         goto insert;
5151                 }
5152                 if (compress_type != BTRFS_COMPRESS_NONE) {
5153                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5154                         em->compress_type = compress_type;
5155                         em->block_start = bytenr;
5156                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5157                                                                          item);
5158                 } else {
5159                         bytenr += btrfs_file_extent_offset(leaf, item);
5160                         em->block_start = bytenr;
5161                         em->block_len = em->len;
5162                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5163                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5164                 }
5165                 goto insert;
5166         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5167                 unsigned long ptr;
5168                 char *map;
5169                 size_t size;
5170                 size_t extent_offset;
5171                 size_t copy_size;
5172
5173                 em->block_start = EXTENT_MAP_INLINE;
5174                 if (!page || create) {
5175                         em->start = extent_start;
5176                         em->len = extent_end - extent_start;
5177                         goto out;
5178                 }
5179
5180                 size = btrfs_file_extent_inline_len(leaf, item);
5181                 extent_offset = page_offset(page) + pg_offset - extent_start;
5182                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5183                                 size - extent_offset);
5184                 em->start = extent_start + extent_offset;
5185                 em->len = (copy_size + root->sectorsize - 1) &
5186                         ~((u64)root->sectorsize - 1);
5187                 em->orig_start = EXTENT_MAP_INLINE;
5188                 if (compress_type) {
5189                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5190                         em->compress_type = compress_type;
5191                 }
5192                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5193                 if (create == 0 && !PageUptodate(page)) {
5194                         if (btrfs_file_extent_compression(leaf, item) !=
5195                             BTRFS_COMPRESS_NONE) {
5196                                 ret = uncompress_inline(path, inode, page,
5197                                                         pg_offset,
5198                                                         extent_offset, item);
5199                                 BUG_ON(ret);
5200                         } else {
5201                                 map = kmap(page);
5202                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5203                                                    copy_size);
5204                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5205                                         memset(map + pg_offset + copy_size, 0,
5206                                                PAGE_CACHE_SIZE - pg_offset -
5207                                                copy_size);
5208                                 }
5209                                 kunmap(page);
5210                         }
5211                         flush_dcache_page(page);
5212                 } else if (create && PageUptodate(page)) {
5213                         WARN_ON(1);
5214                         if (!trans) {
5215                                 kunmap(page);
5216                                 free_extent_map(em);
5217                                 em = NULL;
5218                                 btrfs_release_path(root, path);
5219                                 trans = btrfs_join_transaction(root, 1);
5220                                 if (IS_ERR(trans))
5221                                         return ERR_CAST(trans);
5222                                 goto again;
5223                         }
5224                         map = kmap(page);
5225                         write_extent_buffer(leaf, map + pg_offset, ptr,
5226                                             copy_size);
5227                         kunmap(page);
5228                         btrfs_mark_buffer_dirty(leaf);
5229                 }
5230                 set_extent_uptodate(io_tree, em->start,
5231                                     extent_map_end(em) - 1, GFP_NOFS);
5232                 goto insert;
5233         } else {
5234                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5235                 WARN_ON(1);
5236         }
5237 not_found:
5238         em->start = start;
5239         em->len = len;
5240 not_found_em:
5241         em->block_start = EXTENT_MAP_HOLE;
5242         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5243 insert:
5244         btrfs_release_path(root, path);
5245         if (em->start > start || extent_map_end(em) <= start) {
5246                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5247                        "[%llu %llu]\n", (unsigned long long)em->start,
5248                        (unsigned long long)em->len,
5249                        (unsigned long long)start,
5250                        (unsigned long long)len);
5251                 err = -EIO;
5252                 goto out;
5253         }
5254
5255         err = 0;
5256         write_lock(&em_tree->lock);
5257         ret = add_extent_mapping(em_tree, em);
5258         /* it is possible that someone inserted the extent into the tree
5259          * while we had the lock dropped.  It is also possible that
5260          * an overlapping map exists in the tree
5261          */
5262         if (ret == -EEXIST) {
5263                 struct extent_map *existing;
5264
5265                 ret = 0;
5266
5267                 existing = lookup_extent_mapping(em_tree, start, len);
5268                 if (existing && (existing->start > start ||
5269                     existing->start + existing->len <= start)) {
5270                         free_extent_map(existing);
5271                         existing = NULL;
5272                 }
5273                 if (!existing) {
5274                         existing = lookup_extent_mapping(em_tree, em->start,
5275                                                          em->len);
5276                         if (existing) {
5277                                 err = merge_extent_mapping(em_tree, existing,
5278                                                            em, start,
5279                                                            root->sectorsize);
5280                                 free_extent_map(existing);
5281                                 if (err) {
5282                                         free_extent_map(em);
5283                                         em = NULL;
5284                                 }
5285                         } else {
5286                                 err = -EIO;
5287                                 free_extent_map(em);
5288                                 em = NULL;
5289                         }
5290                 } else {
5291                         free_extent_map(em);
5292                         em = existing;
5293                         err = 0;
5294                 }
5295         }
5296         write_unlock(&em_tree->lock);
5297 out:
5298
5299         trace_btrfs_get_extent(root, em);
5300
5301         if (path)
5302                 btrfs_free_path(path);
5303         if (trans) {
5304                 ret = btrfs_end_transaction(trans, root);
5305                 if (!err)
5306                         err = ret;
5307         }
5308         if (err) {
5309                 free_extent_map(em);
5310                 return ERR_PTR(err);
5311         }
5312         return em;
5313 }
5314
5315 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5316                                            size_t pg_offset, u64 start, u64 len,
5317                                            int create)
5318 {
5319         struct extent_map *em;
5320         struct extent_map *hole_em = NULL;
5321         u64 range_start = start;
5322         u64 end;
5323         u64 found;
5324         u64 found_end;
5325         int err = 0;
5326
5327         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5328         if (IS_ERR(em))
5329                 return em;
5330         if (em) {
5331                 /*
5332                  * if our em maps to a hole, there might
5333                  * actually be delalloc bytes behind it
5334                  */
5335                 if (em->block_start != EXTENT_MAP_HOLE)
5336                         return em;
5337                 else
5338                         hole_em = em;
5339         }
5340
5341         /* check to see if we've wrapped (len == -1 or similar) */
5342         end = start + len;
5343         if (end < start)
5344                 end = (u64)-1;
5345         else
5346                 end -= 1;
5347
5348         em = NULL;
5349
5350         /* ok, we didn't find anything, lets look for delalloc */
5351         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5352                                  end, len, EXTENT_DELALLOC, 1);
5353         found_end = range_start + found;
5354         if (found_end < range_start)
5355                 found_end = (u64)-1;
5356
5357         /*
5358          * we didn't find anything useful, return
5359          * the original results from get_extent()
5360          */
5361         if (range_start > end || found_end <= start) {
5362                 em = hole_em;
5363                 hole_em = NULL;
5364                 goto out;
5365         }
5366
5367         /* adjust the range_start to make sure it doesn't
5368          * go backwards from the start they passed in
5369          */
5370         range_start = max(start,range_start);
5371         found = found_end - range_start;
5372
5373         if (found > 0) {
5374                 u64 hole_start = start;
5375                 u64 hole_len = len;
5376
5377                 em = alloc_extent_map(GFP_NOFS);
5378                 if (!em) {
5379                         err = -ENOMEM;
5380                         goto out;
5381                 }
5382                 /*
5383                  * when btrfs_get_extent can't find anything it
5384                  * returns one huge hole
5385                  *
5386                  * make sure what it found really fits our range, and
5387                  * adjust to make sure it is based on the start from
5388                  * the caller
5389                  */
5390                 if (hole_em) {
5391                         u64 calc_end = extent_map_end(hole_em);
5392
5393                         if (calc_end <= start || (hole_em->start > end)) {
5394                                 free_extent_map(hole_em);
5395                                 hole_em = NULL;
5396                         } else {
5397                                 hole_start = max(hole_em->start, start);
5398                                 hole_len = calc_end - hole_start;
5399                         }
5400                 }
5401                 em->bdev = NULL;
5402                 if (hole_em && range_start > hole_start) {
5403                         /* our hole starts before our delalloc, so we
5404                          * have to return just the parts of the hole
5405                          * that go until  the delalloc starts
5406                          */
5407                         em->len = min(hole_len,
5408                                       range_start - hole_start);
5409                         em->start = hole_start;
5410                         em->orig_start = hole_start;
5411                         /*
5412                          * don't adjust block start at all,
5413                          * it is fixed at EXTENT_MAP_HOLE
5414                          */
5415                         em->block_start = hole_em->block_start;
5416                         em->block_len = hole_len;
5417                 } else {
5418                         em->start = range_start;
5419                         em->len = found;
5420                         em->orig_start = range_start;
5421                         em->block_start = EXTENT_MAP_DELALLOC;
5422                         em->block_len = found;
5423                 }
5424         } else if (hole_em) {
5425                 return hole_em;
5426         }
5427 out:
5428
5429         free_extent_map(hole_em);
5430         if (err) {
5431                 free_extent_map(em);
5432                 return ERR_PTR(err);
5433         }
5434         return em;
5435 }
5436
5437 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5438                                                   u64 start, u64 len)
5439 {
5440         struct btrfs_root *root = BTRFS_I(inode)->root;
5441         struct btrfs_trans_handle *trans;
5442         struct extent_map *em;
5443         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5444         struct btrfs_key ins;
5445         u64 alloc_hint;
5446         int ret;
5447
5448         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5449
5450         trans = btrfs_join_transaction(root, 0);
5451         if (IS_ERR(trans))
5452                 return ERR_CAST(trans);
5453
5454         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5455
5456         alloc_hint = get_extent_allocation_hint(inode, start, len);
5457         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5458                                    alloc_hint, (u64)-1, &ins, 1);
5459         if (ret) {
5460                 em = ERR_PTR(ret);
5461                 goto out;
5462         }
5463
5464         em = alloc_extent_map(GFP_NOFS);
5465         if (!em) {
5466                 em = ERR_PTR(-ENOMEM);
5467                 goto out;
5468         }
5469
5470         em->start = start;
5471         em->orig_start = em->start;
5472         em->len = ins.offset;
5473
5474         em->block_start = ins.objectid;
5475         em->block_len = ins.offset;
5476         em->bdev = root->fs_info->fs_devices->latest_bdev;
5477         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5478
5479         while (1) {
5480                 write_lock(&em_tree->lock);
5481                 ret = add_extent_mapping(em_tree, em);
5482                 write_unlock(&em_tree->lock);
5483                 if (ret != -EEXIST)
5484                         break;
5485                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5486         }
5487
5488         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5489                                            ins.offset, ins.offset, 0);
5490         if (ret) {
5491                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5492                 em = ERR_PTR(ret);
5493         }
5494 out:
5495         btrfs_end_transaction(trans, root);
5496         return em;
5497 }
5498
5499 /*
5500  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5501  * block must be cow'd
5502  */
5503 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5504                                       struct inode *inode, u64 offset, u64 len)
5505 {
5506         struct btrfs_path *path;
5507         int ret;
5508         struct extent_buffer *leaf;
5509         struct btrfs_root *root = BTRFS_I(inode)->root;
5510         struct btrfs_file_extent_item *fi;
5511         struct btrfs_key key;
5512         u64 disk_bytenr;
5513         u64 backref_offset;
5514         u64 extent_end;
5515         u64 num_bytes;
5516         int slot;
5517         int found_type;
5518
5519         path = btrfs_alloc_path();
5520         if (!path)
5521                 return -ENOMEM;
5522
5523         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5524                                        offset, 0);
5525         if (ret < 0)
5526                 goto out;
5527
5528         slot = path->slots[0];
5529         if (ret == 1) {
5530                 if (slot == 0) {
5531                         /* can't find the item, must cow */
5532                         ret = 0;
5533                         goto out;
5534                 }
5535                 slot--;
5536         }
5537         ret = 0;
5538         leaf = path->nodes[0];
5539         btrfs_item_key_to_cpu(leaf, &key, slot);
5540         if (key.objectid != inode->i_ino ||
5541             key.type != BTRFS_EXTENT_DATA_KEY) {
5542                 /* not our file or wrong item type, must cow */
5543                 goto out;
5544         }
5545
5546         if (key.offset > offset) {
5547                 /* Wrong offset, must cow */
5548                 goto out;
5549         }
5550
5551         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5552         found_type = btrfs_file_extent_type(leaf, fi);
5553         if (found_type != BTRFS_FILE_EXTENT_REG &&
5554             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5555                 /* not a regular extent, must cow */
5556                 goto out;
5557         }
5558         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5559         backref_offset = btrfs_file_extent_offset(leaf, fi);
5560
5561         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5562         if (extent_end < offset + len) {
5563                 /* extent doesn't include our full range, must cow */
5564                 goto out;
5565         }
5566
5567         if (btrfs_extent_readonly(root, disk_bytenr))
5568                 goto out;
5569
5570         /*
5571          * look for other files referencing this extent, if we
5572          * find any we must cow
5573          */
5574         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5575                                   key.offset - backref_offset, disk_bytenr))
5576                 goto out;
5577
5578         /*
5579          * adjust disk_bytenr and num_bytes to cover just the bytes
5580          * in this extent we are about to write.  If there
5581          * are any csums in that range we have to cow in order
5582          * to keep the csums correct
5583          */
5584         disk_bytenr += backref_offset;
5585         disk_bytenr += offset - key.offset;
5586         num_bytes = min(offset + len, extent_end) - offset;
5587         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5588                                 goto out;
5589         /*
5590          * all of the above have passed, it is safe to overwrite this extent
5591          * without cow
5592          */
5593         ret = 1;
5594 out:
5595         btrfs_free_path(path);
5596         return ret;
5597 }
5598
5599 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5600                                    struct buffer_head *bh_result, int create)
5601 {
5602         struct extent_map *em;
5603         struct btrfs_root *root = BTRFS_I(inode)->root;
5604         u64 start = iblock << inode->i_blkbits;
5605         u64 len = bh_result->b_size;
5606         struct btrfs_trans_handle *trans;
5607
5608         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5609         if (IS_ERR(em))
5610                 return PTR_ERR(em);
5611
5612         /*
5613          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5614          * io.  INLINE is special, and we could probably kludge it in here, but
5615          * it's still buffered so for safety lets just fall back to the generic
5616          * buffered path.
5617          *
5618          * For COMPRESSED we _have_ to read the entire extent in so we can
5619          * decompress it, so there will be buffering required no matter what we
5620          * do, so go ahead and fallback to buffered.
5621          *
5622          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5623          * to buffered IO.  Don't blame me, this is the price we pay for using
5624          * the generic code.
5625          */
5626         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5627             em->block_start == EXTENT_MAP_INLINE) {
5628                 free_extent_map(em);
5629                 return -ENOTBLK;
5630         }
5631
5632         /* Just a good old fashioned hole, return */
5633         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5634                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5635                 free_extent_map(em);
5636                 /* DIO will do one hole at a time, so just unlock a sector */
5637                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5638                               start + root->sectorsize - 1, GFP_NOFS);
5639                 return 0;
5640         }
5641
5642         /*
5643          * We don't allocate a new extent in the following cases
5644          *
5645          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5646          * existing extent.
5647          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5648          * just use the extent.
5649          *
5650          */
5651         if (!create) {
5652                 len = em->len - (start - em->start);
5653                 goto map;
5654         }
5655
5656         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5657             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5658              em->block_start != EXTENT_MAP_HOLE)) {
5659                 int type;
5660                 int ret;
5661                 u64 block_start;
5662
5663                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5664                         type = BTRFS_ORDERED_PREALLOC;
5665                 else
5666                         type = BTRFS_ORDERED_NOCOW;
5667                 len = min(len, em->len - (start - em->start));
5668                 block_start = em->block_start + (start - em->start);
5669
5670                 /*
5671                  * we're not going to log anything, but we do need
5672                  * to make sure the current transaction stays open
5673                  * while we look for nocow cross refs
5674                  */
5675                 trans = btrfs_join_transaction(root, 0);
5676                 if (IS_ERR(trans))
5677                         goto must_cow;
5678
5679                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5680                         ret = btrfs_add_ordered_extent_dio(inode, start,
5681                                            block_start, len, len, type);
5682                         btrfs_end_transaction(trans, root);
5683                         if (ret) {
5684                                 free_extent_map(em);
5685                                 return ret;
5686                         }
5687                         goto unlock;
5688                 }
5689                 btrfs_end_transaction(trans, root);
5690         }
5691 must_cow:
5692         /*
5693          * this will cow the extent, reset the len in case we changed
5694          * it above
5695          */
5696         len = bh_result->b_size;
5697         free_extent_map(em);
5698         em = btrfs_new_extent_direct(inode, start, len);
5699         if (IS_ERR(em))
5700                 return PTR_ERR(em);
5701         len = min(len, em->len - (start - em->start));
5702 unlock:
5703         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5704                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5705                           0, NULL, GFP_NOFS);
5706 map:
5707         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5708                 inode->i_blkbits;
5709         bh_result->b_size = len;
5710         bh_result->b_bdev = em->bdev;
5711         set_buffer_mapped(bh_result);
5712         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5713                 set_buffer_new(bh_result);
5714
5715         free_extent_map(em);
5716
5717         return 0;
5718 }
5719
5720 struct btrfs_dio_private {
5721         struct inode *inode;
5722         u64 logical_offset;
5723         u64 disk_bytenr;
5724         u64 bytes;
5725         u32 *csums;
5726         void *private;
5727
5728         /* number of bios pending for this dio */
5729         atomic_t pending_bios;
5730
5731         /* IO errors */
5732         int errors;
5733
5734         struct bio *orig_bio;
5735 };
5736
5737 static void btrfs_endio_direct_read(struct bio *bio, int err)
5738 {
5739         struct btrfs_dio_private *dip = bio->bi_private;
5740         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5741         struct bio_vec *bvec = bio->bi_io_vec;
5742         struct inode *inode = dip->inode;
5743         struct btrfs_root *root = BTRFS_I(inode)->root;
5744         u64 start;
5745         u32 *private = dip->csums;
5746
5747         start = dip->logical_offset;
5748         do {
5749                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5750                         struct page *page = bvec->bv_page;
5751                         char *kaddr;
5752                         u32 csum = ~(u32)0;
5753                         unsigned long flags;
5754
5755                         local_irq_save(flags);
5756                         kaddr = kmap_atomic(page, KM_IRQ0);
5757                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5758                                                csum, bvec->bv_len);
5759                         btrfs_csum_final(csum, (char *)&csum);
5760                         kunmap_atomic(kaddr, KM_IRQ0);
5761                         local_irq_restore(flags);
5762
5763                         flush_dcache_page(bvec->bv_page);
5764                         if (csum != *private) {
5765                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5766                                       " %llu csum %u private %u\n",
5767                                       inode->i_ino, (unsigned long long)start,
5768                                       csum, *private);
5769                                 err = -EIO;
5770                         }
5771                 }
5772
5773                 start += bvec->bv_len;
5774                 private++;
5775                 bvec++;
5776         } while (bvec <= bvec_end);
5777
5778         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5779                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5780         bio->bi_private = dip->private;
5781
5782         kfree(dip->csums);
5783         kfree(dip);
5784
5785         /* If we had a csum failure make sure to clear the uptodate flag */
5786         if (err)
5787                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5788         dio_end_io(bio, err);
5789 }
5790
5791 static void btrfs_endio_direct_write(struct bio *bio, int err)
5792 {
5793         struct btrfs_dio_private *dip = bio->bi_private;
5794         struct inode *inode = dip->inode;
5795         struct btrfs_root *root = BTRFS_I(inode)->root;
5796         struct btrfs_trans_handle *trans;
5797         struct btrfs_ordered_extent *ordered = NULL;
5798         struct extent_state *cached_state = NULL;
5799         u64 ordered_offset = dip->logical_offset;
5800         u64 ordered_bytes = dip->bytes;
5801         int ret;
5802
5803         if (err)
5804                 goto out_done;
5805 again:
5806         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5807                                                    &ordered_offset,
5808                                                    ordered_bytes);
5809         if (!ret)
5810                 goto out_test;
5811
5812         BUG_ON(!ordered);
5813
5814         trans = btrfs_join_transaction(root, 1);
5815         if (IS_ERR(trans)) {
5816                 err = -ENOMEM;
5817                 goto out;
5818         }
5819         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5820
5821         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5822                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5823                 if (!ret)
5824                         ret = btrfs_update_inode(trans, root, inode);
5825                 err = ret;
5826                 goto out;
5827         }
5828
5829         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5830                          ordered->file_offset + ordered->len - 1, 0,
5831                          &cached_state, GFP_NOFS);
5832
5833         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5834                 ret = btrfs_mark_extent_written(trans, inode,
5835                                                 ordered->file_offset,
5836                                                 ordered->file_offset +
5837                                                 ordered->len);
5838                 if (ret) {
5839                         err = ret;
5840                         goto out_unlock;
5841                 }
5842         } else {
5843                 ret = insert_reserved_file_extent(trans, inode,
5844                                                   ordered->file_offset,
5845                                                   ordered->start,
5846                                                   ordered->disk_len,
5847                                                   ordered->len,
5848                                                   ordered->len,
5849                                                   0, 0, 0,
5850                                                   BTRFS_FILE_EXTENT_REG);
5851                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5852                                    ordered->file_offset, ordered->len);
5853                 if (ret) {
5854                         err = ret;
5855                         WARN_ON(1);
5856                         goto out_unlock;
5857                 }
5858         }
5859
5860         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5861         btrfs_ordered_update_i_size(inode, 0, ordered);
5862         btrfs_update_inode(trans, root, inode);
5863 out_unlock:
5864         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5865                              ordered->file_offset + ordered->len - 1,
5866                              &cached_state, GFP_NOFS);
5867 out:
5868         btrfs_delalloc_release_metadata(inode, ordered->len);
5869         btrfs_end_transaction(trans, root);
5870         ordered_offset = ordered->file_offset + ordered->len;
5871         btrfs_put_ordered_extent(ordered);
5872         btrfs_put_ordered_extent(ordered);
5873
5874 out_test:
5875         /*
5876          * our bio might span multiple ordered extents.  If we haven't
5877          * completed the accounting for the whole dio, go back and try again
5878          */
5879         if (ordered_offset < dip->logical_offset + dip->bytes) {
5880                 ordered_bytes = dip->logical_offset + dip->bytes -
5881                         ordered_offset;
5882                 goto again;
5883         }
5884 out_done:
5885         bio->bi_private = dip->private;
5886
5887         kfree(dip->csums);
5888         kfree(dip);
5889
5890         /* If we had an error make sure to clear the uptodate flag */
5891         if (err)
5892                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5893         dio_end_io(bio, err);
5894 }
5895
5896 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5897                                     struct bio *bio, int mirror_num,
5898                                     unsigned long bio_flags, u64 offset)
5899 {
5900         int ret;
5901         struct btrfs_root *root = BTRFS_I(inode)->root;
5902         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5903         BUG_ON(ret);
5904         return 0;
5905 }
5906
5907 static void btrfs_end_dio_bio(struct bio *bio, int err)
5908 {
5909         struct btrfs_dio_private *dip = bio->bi_private;
5910
5911         if (err) {
5912                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5913                       "sector %#Lx len %u err no %d\n",
5914                       dip->inode->i_ino, bio->bi_rw,
5915                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5916                 dip->errors = 1;
5917
5918                 /*
5919                  * before atomic variable goto zero, we must make sure
5920                  * dip->errors is perceived to be set.
5921                  */
5922                 smp_mb__before_atomic_dec();
5923         }
5924
5925         /* if there are more bios still pending for this dio, just exit */
5926         if (!atomic_dec_and_test(&dip->pending_bios))
5927                 goto out;
5928
5929         if (dip->errors)
5930                 bio_io_error(dip->orig_bio);
5931         else {
5932                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5933                 bio_endio(dip->orig_bio, 0);
5934         }
5935 out:
5936         bio_put(bio);
5937 }
5938
5939 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5940                                        u64 first_sector, gfp_t gfp_flags)
5941 {
5942         int nr_vecs = bio_get_nr_vecs(bdev);
5943         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5944 }
5945
5946 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5947                                          int rw, u64 file_offset, int skip_sum,
5948                                          u32 *csums)
5949 {
5950         int write = rw & REQ_WRITE;
5951         struct btrfs_root *root = BTRFS_I(inode)->root;
5952         int ret;
5953
5954         bio_get(bio);
5955         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5956         if (ret)
5957                 goto err;
5958
5959         if (write && !skip_sum) {
5960                 ret = btrfs_wq_submit_bio(root->fs_info,
5961                                    inode, rw, bio, 0, 0,
5962                                    file_offset,
5963                                    __btrfs_submit_bio_start_direct_io,
5964                                    __btrfs_submit_bio_done);
5965                 goto err;
5966         } else if (!skip_sum) {
5967                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5968                                           file_offset, csums);
5969                 if (ret)
5970                         goto err;
5971         }
5972
5973         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5974 err:
5975         bio_put(bio);
5976         return ret;
5977 }
5978
5979 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5980                                     int skip_sum)
5981 {
5982         struct inode *inode = dip->inode;
5983         struct btrfs_root *root = BTRFS_I(inode)->root;
5984         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5985         struct bio *bio;
5986         struct bio *orig_bio = dip->orig_bio;
5987         struct bio_vec *bvec = orig_bio->bi_io_vec;
5988         u64 start_sector = orig_bio->bi_sector;
5989         u64 file_offset = dip->logical_offset;
5990         u64 submit_len = 0;
5991         u64 map_length;
5992         int nr_pages = 0;
5993         u32 *csums = dip->csums;
5994         int ret = 0;
5995         int write = rw & REQ_WRITE;
5996
5997         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5998         if (!bio)
5999                 return -ENOMEM;
6000         bio->bi_private = dip;
6001         bio->bi_end_io = btrfs_end_dio_bio;
6002         atomic_inc(&dip->pending_bios);
6003
6004         map_length = orig_bio->bi_size;
6005         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6006                               &map_length, NULL, 0);
6007         if (ret) {
6008                 bio_put(bio);
6009                 return -EIO;
6010         }
6011
6012         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6013                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6014                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6015                                  bvec->bv_offset) < bvec->bv_len)) {
6016                         /*
6017                          * inc the count before we submit the bio so
6018                          * we know the end IO handler won't happen before
6019                          * we inc the count. Otherwise, the dip might get freed
6020                          * before we're done setting it up
6021                          */
6022                         atomic_inc(&dip->pending_bios);
6023                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6024                                                      file_offset, skip_sum,
6025                                                      csums);
6026                         if (ret) {
6027                                 bio_put(bio);
6028                                 atomic_dec(&dip->pending_bios);
6029                                 goto out_err;
6030                         }
6031
6032                         /* Write's use the ordered csums */
6033                         if (!write && !skip_sum)
6034                                 csums = csums + nr_pages;
6035                         start_sector += submit_len >> 9;
6036                         file_offset += submit_len;
6037
6038                         submit_len = 0;
6039                         nr_pages = 0;
6040
6041                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6042                                                   start_sector, GFP_NOFS);
6043                         if (!bio)
6044                                 goto out_err;
6045                         bio->bi_private = dip;
6046                         bio->bi_end_io = btrfs_end_dio_bio;
6047
6048                         map_length = orig_bio->bi_size;
6049                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6050                                               &map_length, NULL, 0);
6051                         if (ret) {
6052                                 bio_put(bio);
6053                                 goto out_err;
6054                         }
6055                 } else {
6056                         submit_len += bvec->bv_len;
6057                         nr_pages ++;
6058                         bvec++;
6059                 }
6060         }
6061
6062         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6063                                      csums);
6064         if (!ret)
6065                 return 0;
6066
6067         bio_put(bio);
6068 out_err:
6069         dip->errors = 1;
6070         /*
6071          * before atomic variable goto zero, we must
6072          * make sure dip->errors is perceived to be set.
6073          */
6074         smp_mb__before_atomic_dec();
6075         if (atomic_dec_and_test(&dip->pending_bios))
6076                 bio_io_error(dip->orig_bio);
6077
6078         /* bio_end_io() will handle error, so we needn't return it */
6079         return 0;
6080 }
6081
6082 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6083                                 loff_t file_offset)
6084 {
6085         struct btrfs_root *root = BTRFS_I(inode)->root;
6086         struct btrfs_dio_private *dip;
6087         struct bio_vec *bvec = bio->bi_io_vec;
6088         int skip_sum;
6089         int write = rw & REQ_WRITE;
6090         int ret = 0;
6091
6092         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6093
6094         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6095         if (!dip) {
6096                 ret = -ENOMEM;
6097                 goto free_ordered;
6098         }
6099         dip->csums = NULL;
6100
6101         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6102         if (!write && !skip_sum) {
6103                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6104                 if (!dip->csums) {
6105                         kfree(dip);
6106                         ret = -ENOMEM;
6107                         goto free_ordered;
6108                 }
6109         }
6110
6111         dip->private = bio->bi_private;
6112         dip->inode = inode;
6113         dip->logical_offset = file_offset;
6114
6115         dip->bytes = 0;
6116         do {
6117                 dip->bytes += bvec->bv_len;
6118                 bvec++;
6119         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6120
6121         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6122         bio->bi_private = dip;
6123         dip->errors = 0;
6124         dip->orig_bio = bio;
6125         atomic_set(&dip->pending_bios, 0);
6126
6127         if (write)
6128                 bio->bi_end_io = btrfs_endio_direct_write;
6129         else
6130                 bio->bi_end_io = btrfs_endio_direct_read;
6131
6132         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6133         if (!ret)
6134                 return;
6135 free_ordered:
6136         /*
6137          * If this is a write, we need to clean up the reserved space and kill
6138          * the ordered extent.
6139          */
6140         if (write) {
6141                 struct btrfs_ordered_extent *ordered;
6142                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6143                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6144                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6145                         btrfs_free_reserved_extent(root, ordered->start,
6146                                                    ordered->disk_len);
6147                 btrfs_put_ordered_extent(ordered);
6148                 btrfs_put_ordered_extent(ordered);
6149         }
6150         bio_endio(bio, ret);
6151 }
6152
6153 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6154                         const struct iovec *iov, loff_t offset,
6155                         unsigned long nr_segs)
6156 {
6157         int seg;
6158         size_t size;
6159         unsigned long addr;
6160         unsigned blocksize_mask = root->sectorsize - 1;
6161         ssize_t retval = -EINVAL;
6162         loff_t end = offset;
6163
6164         if (offset & blocksize_mask)
6165                 goto out;
6166
6167         /* Check the memory alignment.  Blocks cannot straddle pages */
6168         for (seg = 0; seg < nr_segs; seg++) {
6169                 addr = (unsigned long)iov[seg].iov_base;
6170                 size = iov[seg].iov_len;
6171                 end += size;
6172                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6173                         goto out;
6174         }
6175         retval = 0;
6176 out:
6177         return retval;
6178 }
6179 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6180                         const struct iovec *iov, loff_t offset,
6181                         unsigned long nr_segs)
6182 {
6183         struct file *file = iocb->ki_filp;
6184         struct inode *inode = file->f_mapping->host;
6185         struct btrfs_ordered_extent *ordered;
6186         struct extent_state *cached_state = NULL;
6187         u64 lockstart, lockend;
6188         ssize_t ret;
6189         int writing = rw & WRITE;
6190         int write_bits = 0;
6191         size_t count = iov_length(iov, nr_segs);
6192
6193         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6194                             offset, nr_segs)) {
6195                 return 0;
6196         }
6197
6198         lockstart = offset;
6199         lockend = offset + count - 1;
6200
6201         if (writing) {
6202                 ret = btrfs_delalloc_reserve_space(inode, count);
6203                 if (ret)
6204                         goto out;
6205         }
6206
6207         while (1) {
6208                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6209                                  0, &cached_state, GFP_NOFS);
6210                 /*
6211                  * We're concerned with the entire range that we're going to be
6212                  * doing DIO to, so we need to make sure theres no ordered
6213                  * extents in this range.
6214                  */
6215                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6216                                                      lockend - lockstart + 1);
6217                 if (!ordered)
6218                         break;
6219                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6220                                      &cached_state, GFP_NOFS);
6221                 btrfs_start_ordered_extent(inode, ordered, 1);
6222                 btrfs_put_ordered_extent(ordered);
6223                 cond_resched();
6224         }
6225
6226         /*
6227          * we don't use btrfs_set_extent_delalloc because we don't want
6228          * the dirty or uptodate bits
6229          */
6230         if (writing) {
6231                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6232                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6233                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6234                                      GFP_NOFS);
6235                 if (ret) {
6236                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6237                                          lockend, EXTENT_LOCKED | write_bits,
6238                                          1, 0, &cached_state, GFP_NOFS);
6239                         goto out;
6240                 }
6241         }
6242
6243         free_extent_state(cached_state);
6244         cached_state = NULL;
6245
6246         ret = __blockdev_direct_IO(rw, iocb, inode,
6247                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6248                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6249                    btrfs_submit_direct, 0);
6250
6251         if (ret < 0 && ret != -EIOCBQUEUED) {
6252                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6253                               offset + iov_length(iov, nr_segs) - 1,
6254                               EXTENT_LOCKED | write_bits, 1, 0,
6255                               &cached_state, GFP_NOFS);
6256         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6257                 /*
6258                  * We're falling back to buffered, unlock the section we didn't
6259                  * do IO on.
6260                  */
6261                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6262                               offset + iov_length(iov, nr_segs) - 1,
6263                               EXTENT_LOCKED | write_bits, 1, 0,
6264                               &cached_state, GFP_NOFS);
6265         }
6266 out:
6267         free_extent_state(cached_state);
6268         return ret;
6269 }
6270
6271 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6272                 __u64 start, __u64 len)
6273 {
6274         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6275 }
6276
6277 int btrfs_readpage(struct file *file, struct page *page)
6278 {
6279         struct extent_io_tree *tree;
6280         tree = &BTRFS_I(page->mapping->host)->io_tree;
6281         return extent_read_full_page(tree, page, btrfs_get_extent);
6282 }
6283
6284 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6285 {
6286         struct extent_io_tree *tree;
6287
6288
6289         if (current->flags & PF_MEMALLOC) {
6290                 redirty_page_for_writepage(wbc, page);
6291                 unlock_page(page);
6292                 return 0;
6293         }
6294         tree = &BTRFS_I(page->mapping->host)->io_tree;
6295         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6296 }
6297
6298 int btrfs_writepages(struct address_space *mapping,
6299                      struct writeback_control *wbc)
6300 {
6301         struct extent_io_tree *tree;
6302
6303         tree = &BTRFS_I(mapping->host)->io_tree;
6304         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6305 }
6306
6307 static int
6308 btrfs_readpages(struct file *file, struct address_space *mapping,
6309                 struct list_head *pages, unsigned nr_pages)
6310 {
6311         struct extent_io_tree *tree;
6312         tree = &BTRFS_I(mapping->host)->io_tree;
6313         return extent_readpages(tree, mapping, pages, nr_pages,
6314                                 btrfs_get_extent);
6315 }
6316 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6317 {
6318         struct extent_io_tree *tree;
6319         struct extent_map_tree *map;
6320         int ret;
6321
6322         tree = &BTRFS_I(page->mapping->host)->io_tree;
6323         map = &BTRFS_I(page->mapping->host)->extent_tree;
6324         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6325         if (ret == 1) {
6326                 ClearPagePrivate(page);
6327                 set_page_private(page, 0);
6328                 page_cache_release(page);
6329         }
6330         return ret;
6331 }
6332
6333 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6334 {
6335         if (PageWriteback(page) || PageDirty(page))
6336                 return 0;
6337         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6338 }
6339
6340 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6341 {
6342         struct extent_io_tree *tree;
6343         struct btrfs_ordered_extent *ordered;
6344         struct extent_state *cached_state = NULL;
6345         u64 page_start = page_offset(page);
6346         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6347
6348
6349         /*
6350          * we have the page locked, so new writeback can't start,
6351          * and the dirty bit won't be cleared while we are here.
6352          *
6353          * Wait for IO on this page so that we can safely clear
6354          * the PagePrivate2 bit and do ordered accounting
6355          */
6356         wait_on_page_writeback(page);
6357
6358         tree = &BTRFS_I(page->mapping->host)->io_tree;
6359         if (offset) {
6360                 btrfs_releasepage(page, GFP_NOFS);
6361                 return;
6362         }
6363         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6364                          GFP_NOFS);
6365         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6366                                            page_offset(page));
6367         if (ordered) {
6368                 /*
6369                  * IO on this page will never be started, so we need
6370                  * to account for any ordered extents now
6371                  */
6372                 clear_extent_bit(tree, page_start, page_end,
6373                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6374                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6375                                  &cached_state, GFP_NOFS);
6376                 /*
6377                  * whoever cleared the private bit is responsible
6378                  * for the finish_ordered_io
6379                  */
6380                 if (TestClearPagePrivate2(page)) {
6381                         btrfs_finish_ordered_io(page->mapping->host,
6382                                                 page_start, page_end);
6383                 }
6384                 btrfs_put_ordered_extent(ordered);
6385                 cached_state = NULL;
6386                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6387                                  GFP_NOFS);
6388         }
6389         clear_extent_bit(tree, page_start, page_end,
6390                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6391                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6392         __btrfs_releasepage(page, GFP_NOFS);
6393
6394         ClearPageChecked(page);
6395         if (PagePrivate(page)) {
6396                 ClearPagePrivate(page);
6397                 set_page_private(page, 0);
6398                 page_cache_release(page);
6399         }
6400 }
6401
6402 /*
6403  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6404  * called from a page fault handler when a page is first dirtied. Hence we must
6405  * be careful to check for EOF conditions here. We set the page up correctly
6406  * for a written page which means we get ENOSPC checking when writing into
6407  * holes and correct delalloc and unwritten extent mapping on filesystems that
6408  * support these features.
6409  *
6410  * We are not allowed to take the i_mutex here so we have to play games to
6411  * protect against truncate races as the page could now be beyond EOF.  Because
6412  * vmtruncate() writes the inode size before removing pages, once we have the
6413  * page lock we can determine safely if the page is beyond EOF. If it is not
6414  * beyond EOF, then the page is guaranteed safe against truncation until we
6415  * unlock the page.
6416  */
6417 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6418 {
6419         struct page *page = vmf->page;
6420         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6421         struct btrfs_root *root = BTRFS_I(inode)->root;
6422         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6423         struct btrfs_ordered_extent *ordered;
6424         struct extent_state *cached_state = NULL;
6425         char *kaddr;
6426         unsigned long zero_start;
6427         loff_t size;
6428         int ret;
6429         u64 page_start;
6430         u64 page_end;
6431
6432         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6433         if (ret) {
6434                 if (ret == -ENOMEM)
6435                         ret = VM_FAULT_OOM;
6436                 else /* -ENOSPC, -EIO, etc */
6437                         ret = VM_FAULT_SIGBUS;
6438                 goto out;
6439         }
6440
6441         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6442 again:
6443         lock_page(page);
6444         size = i_size_read(inode);
6445         page_start = page_offset(page);
6446         page_end = page_start + PAGE_CACHE_SIZE - 1;
6447
6448         if ((page->mapping != inode->i_mapping) ||
6449             (page_start >= size)) {
6450                 /* page got truncated out from underneath us */
6451                 goto out_unlock;
6452         }
6453         wait_on_page_writeback(page);
6454
6455         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6456                          GFP_NOFS);
6457         set_page_extent_mapped(page);
6458
6459         /*
6460          * we can't set the delalloc bits if there are pending ordered
6461          * extents.  Drop our locks and wait for them to finish
6462          */
6463         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6464         if (ordered) {
6465                 unlock_extent_cached(io_tree, page_start, page_end,
6466                                      &cached_state, GFP_NOFS);
6467                 unlock_page(page);
6468                 btrfs_start_ordered_extent(inode, ordered, 1);
6469                 btrfs_put_ordered_extent(ordered);
6470                 goto again;
6471         }
6472
6473         /*
6474          * XXX - page_mkwrite gets called every time the page is dirtied, even
6475          * if it was already dirty, so for space accounting reasons we need to
6476          * clear any delalloc bits for the range we are fixing to save.  There
6477          * is probably a better way to do this, but for now keep consistent with
6478          * prepare_pages in the normal write path.
6479          */
6480         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6481                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6482                           0, 0, &cached_state, GFP_NOFS);
6483
6484         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6485                                         &cached_state);
6486         if (ret) {
6487                 unlock_extent_cached(io_tree, page_start, page_end,
6488                                      &cached_state, GFP_NOFS);
6489                 ret = VM_FAULT_SIGBUS;
6490                 goto out_unlock;
6491         }
6492         ret = 0;
6493
6494         /* page is wholly or partially inside EOF */
6495         if (page_start + PAGE_CACHE_SIZE > size)
6496                 zero_start = size & ~PAGE_CACHE_MASK;
6497         else
6498                 zero_start = PAGE_CACHE_SIZE;
6499
6500         if (zero_start != PAGE_CACHE_SIZE) {
6501                 kaddr = kmap(page);
6502                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6503                 flush_dcache_page(page);
6504                 kunmap(page);
6505         }
6506         ClearPageChecked(page);
6507         set_page_dirty(page);
6508         SetPageUptodate(page);
6509
6510         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6511         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6512
6513         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6514
6515 out_unlock:
6516         if (!ret)
6517                 return VM_FAULT_LOCKED;
6518         unlock_page(page);
6519         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6520 out:
6521         return ret;
6522 }
6523
6524 static int btrfs_truncate(struct inode *inode)
6525 {
6526         struct btrfs_root *root = BTRFS_I(inode)->root;
6527         int ret;
6528         int err = 0;
6529         struct btrfs_trans_handle *trans;
6530         unsigned long nr;
6531         u64 mask = root->sectorsize - 1;
6532
6533         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6534         if (ret)
6535                 return ret;
6536
6537         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6538         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6539
6540         trans = btrfs_start_transaction(root, 5);
6541         if (IS_ERR(trans))
6542                 return PTR_ERR(trans);
6543
6544         btrfs_set_trans_block_group(trans, inode);
6545
6546         ret = btrfs_orphan_add(trans, inode);
6547         if (ret) {
6548                 btrfs_end_transaction(trans, root);
6549                 return ret;
6550         }
6551
6552         nr = trans->blocks_used;
6553         btrfs_end_transaction(trans, root);
6554         btrfs_btree_balance_dirty(root, nr);
6555
6556         /* Now start a transaction for the truncate */
6557         trans = btrfs_start_transaction(root, 0);
6558         if (IS_ERR(trans))
6559                 return PTR_ERR(trans);
6560         btrfs_set_trans_block_group(trans, inode);
6561         trans->block_rsv = root->orphan_block_rsv;
6562
6563         /*
6564          * setattr is responsible for setting the ordered_data_close flag,
6565          * but that is only tested during the last file release.  That
6566          * could happen well after the next commit, leaving a great big
6567          * window where new writes may get lost if someone chooses to write
6568          * to this file after truncating to zero
6569          *
6570          * The inode doesn't have any dirty data here, and so if we commit
6571          * this is a noop.  If someone immediately starts writing to the inode
6572          * it is very likely we'll catch some of their writes in this
6573          * transaction, and the commit will find this file on the ordered
6574          * data list with good things to send down.
6575          *
6576          * This is a best effort solution, there is still a window where
6577          * using truncate to replace the contents of the file will
6578          * end up with a zero length file after a crash.
6579          */
6580         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6581                 btrfs_add_ordered_operation(trans, root, inode);
6582
6583         while (1) {
6584                 if (!trans) {
6585                         trans = btrfs_start_transaction(root, 0);
6586                         if (IS_ERR(trans))
6587                                 return PTR_ERR(trans);
6588                         btrfs_set_trans_block_group(trans, inode);
6589                         trans->block_rsv = root->orphan_block_rsv;
6590                 }
6591
6592                 ret = btrfs_block_rsv_check(trans, root,
6593                                             root->orphan_block_rsv, 0, 5);
6594                 if (ret == -EAGAIN) {
6595                         ret = btrfs_commit_transaction(trans, root);
6596                         if (ret)
6597                                 return ret;
6598                         trans = NULL;
6599                         continue;
6600                 } else if (ret) {
6601                         err = ret;
6602                         break;
6603                 }
6604
6605                 ret = btrfs_truncate_inode_items(trans, root, inode,
6606                                                  inode->i_size,
6607                                                  BTRFS_EXTENT_DATA_KEY);
6608                 if (ret != -EAGAIN) {
6609                         err = ret;
6610                         break;
6611                 }
6612
6613                 ret = btrfs_update_inode(trans, root, inode);
6614                 if (ret) {
6615                         err = ret;
6616                         break;
6617                 }
6618
6619                 nr = trans->blocks_used;
6620                 btrfs_end_transaction(trans, root);
6621                 trans = NULL;
6622                 btrfs_btree_balance_dirty(root, nr);
6623         }
6624
6625         if (ret == 0 && inode->i_nlink > 0) {
6626                 ret = btrfs_orphan_del(trans, inode);
6627                 if (ret)
6628                         err = ret;
6629         } else if (ret && inode->i_nlink > 0) {
6630                 /*
6631                  * Failed to do the truncate, remove us from the in memory
6632                  * orphan list.
6633                  */
6634                 ret = btrfs_orphan_del(NULL, inode);
6635         }
6636
6637         ret = btrfs_update_inode(trans, root, inode);
6638         if (ret && !err)
6639                 err = ret;
6640
6641         nr = trans->blocks_used;
6642         ret = btrfs_end_transaction_throttle(trans, root);
6643         if (ret && !err)
6644                 err = ret;
6645         btrfs_btree_balance_dirty(root, nr);
6646
6647         return err;
6648 }
6649
6650 /*
6651  * create a new subvolume directory/inode (helper for the ioctl).
6652  */
6653 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6654                              struct btrfs_root *new_root,
6655                              u64 new_dirid, u64 alloc_hint)
6656 {
6657         struct inode *inode;
6658         int err;
6659         u64 index = 0;
6660
6661         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6662                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6663         if (IS_ERR(inode))
6664                 return PTR_ERR(inode);
6665         inode->i_op = &btrfs_dir_inode_operations;
6666         inode->i_fop = &btrfs_dir_file_operations;
6667
6668         inode->i_nlink = 1;
6669         btrfs_i_size_write(inode, 0);
6670
6671         err = btrfs_update_inode(trans, new_root, inode);
6672         BUG_ON(err);
6673
6674         iput(inode);
6675         return 0;
6676 }
6677
6678 /* helper function for file defrag and space balancing.  This
6679  * forces readahead on a given range of bytes in an inode
6680  */
6681 unsigned long btrfs_force_ra(struct address_space *mapping,
6682                               struct file_ra_state *ra, struct file *file,
6683                               pgoff_t offset, pgoff_t last_index)
6684 {
6685         pgoff_t req_size = last_index - offset + 1;
6686
6687         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6688         return offset + req_size;
6689 }
6690
6691 struct inode *btrfs_alloc_inode(struct super_block *sb)
6692 {
6693         struct btrfs_inode *ei;
6694         struct inode *inode;
6695
6696         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6697         if (!ei)
6698                 return NULL;
6699
6700         ei->root = NULL;
6701         ei->space_info = NULL;
6702         ei->generation = 0;
6703         ei->sequence = 0;
6704         ei->last_trans = 0;
6705         ei->last_sub_trans = 0;
6706         ei->logged_trans = 0;
6707         ei->delalloc_bytes = 0;
6708         ei->reserved_bytes = 0;
6709         ei->disk_i_size = 0;
6710         ei->flags = 0;
6711         ei->index_cnt = (u64)-1;
6712         ei->last_unlink_trans = 0;
6713
6714         atomic_set(&ei->outstanding_extents, 0);
6715         atomic_set(&ei->reserved_extents, 0);
6716
6717         ei->ordered_data_close = 0;
6718         ei->orphan_meta_reserved = 0;
6719         ei->dummy_inode = 0;
6720         ei->force_compress = BTRFS_COMPRESS_NONE;
6721
6722         inode = &ei->vfs_inode;
6723         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6724         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6725         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6726         mutex_init(&ei->log_mutex);
6727         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6728         INIT_LIST_HEAD(&ei->i_orphan);
6729         INIT_LIST_HEAD(&ei->delalloc_inodes);
6730         INIT_LIST_HEAD(&ei->ordered_operations);
6731         RB_CLEAR_NODE(&ei->rb_node);
6732
6733         return inode;
6734 }
6735
6736 static void btrfs_i_callback(struct rcu_head *head)
6737 {
6738         struct inode *inode = container_of(head, struct inode, i_rcu);
6739         INIT_LIST_HEAD(&inode->i_dentry);
6740         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6741 }
6742
6743 void btrfs_destroy_inode(struct inode *inode)
6744 {
6745         struct btrfs_ordered_extent *ordered;
6746         struct btrfs_root *root = BTRFS_I(inode)->root;
6747
6748         WARN_ON(!list_empty(&inode->i_dentry));
6749         WARN_ON(inode->i_data.nrpages);
6750         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6751         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6752
6753         /*
6754          * This can happen where we create an inode, but somebody else also
6755          * created the same inode and we need to destroy the one we already
6756          * created.
6757          */
6758         if (!root)
6759                 goto free;
6760
6761         /*
6762          * Make sure we're properly removed from the ordered operation
6763          * lists.
6764          */
6765         smp_mb();
6766         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6767                 spin_lock(&root->fs_info->ordered_extent_lock);
6768                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6769                 spin_unlock(&root->fs_info->ordered_extent_lock);
6770         }
6771
6772         if (root == root->fs_info->tree_root) {
6773                 struct btrfs_block_group_cache *block_group;
6774
6775                 block_group = btrfs_lookup_block_group(root->fs_info,
6776                                                 BTRFS_I(inode)->block_group);
6777                 if (block_group && block_group->inode == inode) {
6778                         spin_lock(&block_group->lock);
6779                         block_group->inode = NULL;
6780                         spin_unlock(&block_group->lock);
6781                         btrfs_put_block_group(block_group);
6782                 } else if (block_group) {
6783                         btrfs_put_block_group(block_group);
6784                 }
6785         }
6786
6787         spin_lock(&root->orphan_lock);
6788         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6789                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6790                        inode->i_ino);
6791                 list_del_init(&BTRFS_I(inode)->i_orphan);
6792         }
6793         spin_unlock(&root->orphan_lock);
6794
6795         while (1) {
6796                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6797                 if (!ordered)
6798                         break;
6799                 else {
6800                         printk(KERN_ERR "btrfs found ordered "
6801                                "extent %llu %llu on inode cleanup\n",
6802                                (unsigned long long)ordered->file_offset,
6803                                (unsigned long long)ordered->len);
6804                         btrfs_remove_ordered_extent(inode, ordered);
6805                         btrfs_put_ordered_extent(ordered);
6806                         btrfs_put_ordered_extent(ordered);
6807                 }
6808         }
6809         inode_tree_del(inode);
6810         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6811 free:
6812         call_rcu(&inode->i_rcu, btrfs_i_callback);
6813 }
6814
6815 int btrfs_drop_inode(struct inode *inode)
6816 {
6817         struct btrfs_root *root = BTRFS_I(inode)->root;
6818
6819         if (btrfs_root_refs(&root->root_item) == 0 &&
6820             root != root->fs_info->tree_root)
6821                 return 1;
6822         else
6823                 return generic_drop_inode(inode);
6824 }
6825
6826 static void init_once(void *foo)
6827 {
6828         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6829
6830         inode_init_once(&ei->vfs_inode);
6831 }
6832
6833 void btrfs_destroy_cachep(void)
6834 {
6835         if (btrfs_inode_cachep)
6836                 kmem_cache_destroy(btrfs_inode_cachep);
6837         if (btrfs_trans_handle_cachep)
6838                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6839         if (btrfs_transaction_cachep)
6840                 kmem_cache_destroy(btrfs_transaction_cachep);
6841         if (btrfs_path_cachep)
6842                 kmem_cache_destroy(btrfs_path_cachep);
6843         if (btrfs_free_space_cachep)
6844                 kmem_cache_destroy(btrfs_free_space_cachep);
6845 }
6846
6847 int btrfs_init_cachep(void)
6848 {
6849         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6850                         sizeof(struct btrfs_inode), 0,
6851                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6852         if (!btrfs_inode_cachep)
6853                 goto fail;
6854
6855         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6856                         sizeof(struct btrfs_trans_handle), 0,
6857                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6858         if (!btrfs_trans_handle_cachep)
6859                 goto fail;
6860
6861         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6862                         sizeof(struct btrfs_transaction), 0,
6863                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6864         if (!btrfs_transaction_cachep)
6865                 goto fail;
6866
6867         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6868                         sizeof(struct btrfs_path), 0,
6869                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6870         if (!btrfs_path_cachep)
6871                 goto fail;
6872
6873         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6874                         sizeof(struct btrfs_free_space), 0,
6875                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6876         if (!btrfs_free_space_cachep)
6877                 goto fail;
6878
6879         return 0;
6880 fail:
6881         btrfs_destroy_cachep();
6882         return -ENOMEM;
6883 }
6884
6885 static int btrfs_getattr(struct vfsmount *mnt,
6886                          struct dentry *dentry, struct kstat *stat)
6887 {
6888         struct inode *inode = dentry->d_inode;
6889         generic_fillattr(inode, stat);
6890         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6891         stat->blksize = PAGE_CACHE_SIZE;
6892         stat->blocks = (inode_get_bytes(inode) +
6893                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6894         return 0;
6895 }
6896
6897 /*
6898  * If a file is moved, it will inherit the cow and compression flags of the new
6899  * directory.
6900  */
6901 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6902 {
6903         struct btrfs_inode *b_dir = BTRFS_I(dir);
6904         struct btrfs_inode *b_inode = BTRFS_I(inode);
6905
6906         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6907                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6908         else
6909                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6910
6911         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6912                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6913         else
6914                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6915 }
6916
6917 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6918                            struct inode *new_dir, struct dentry *new_dentry)
6919 {
6920         struct btrfs_trans_handle *trans;
6921         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6922         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6923         struct inode *new_inode = new_dentry->d_inode;
6924         struct inode *old_inode = old_dentry->d_inode;
6925         struct timespec ctime = CURRENT_TIME;
6926         u64 index = 0;
6927         u64 root_objectid;
6928         int ret;
6929
6930         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6931                 return -EPERM;
6932
6933         /* we only allow rename subvolume link between subvolumes */
6934         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6935                 return -EXDEV;
6936
6937         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6938             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6939                 return -ENOTEMPTY;
6940
6941         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6942             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6943                 return -ENOTEMPTY;
6944         /*
6945          * we're using rename to replace one file with another.
6946          * and the replacement file is large.  Start IO on it now so
6947          * we don't add too much work to the end of the transaction
6948          */
6949         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6950             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6951                 filemap_flush(old_inode->i_mapping);
6952
6953         /* close the racy window with snapshot create/destroy ioctl */
6954         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6955                 down_read(&root->fs_info->subvol_sem);
6956         /*
6957          * We want to reserve the absolute worst case amount of items.  So if
6958          * both inodes are subvols and we need to unlink them then that would
6959          * require 4 item modifications, but if they are both normal inodes it
6960          * would require 5 item modifications, so we'll assume their normal
6961          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6962          * should cover the worst case number of items we'll modify.
6963          */
6964         trans = btrfs_start_transaction(root, 20);
6965         if (IS_ERR(trans))
6966                 return PTR_ERR(trans);
6967
6968         btrfs_set_trans_block_group(trans, new_dir);
6969
6970         if (dest != root)
6971                 btrfs_record_root_in_trans(trans, dest);
6972
6973         ret = btrfs_set_inode_index(new_dir, &index);
6974         if (ret)
6975                 goto out_fail;
6976
6977         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6978                 /* force full log commit if subvolume involved. */
6979                 root->fs_info->last_trans_log_full_commit = trans->transid;
6980         } else {
6981                 ret = btrfs_insert_inode_ref(trans, dest,
6982                                              new_dentry->d_name.name,
6983                                              new_dentry->d_name.len,
6984                                              old_inode->i_ino,
6985                                              new_dir->i_ino, index);
6986                 if (ret)
6987                         goto out_fail;
6988                 /*
6989                  * this is an ugly little race, but the rename is required
6990                  * to make sure that if we crash, the inode is either at the
6991                  * old name or the new one.  pinning the log transaction lets
6992                  * us make sure we don't allow a log commit to come in after
6993                  * we unlink the name but before we add the new name back in.
6994                  */
6995                 btrfs_pin_log_trans(root);
6996         }
6997         /*
6998          * make sure the inode gets flushed if it is replacing
6999          * something.
7000          */
7001         if (new_inode && new_inode->i_size &&
7002             old_inode && S_ISREG(old_inode->i_mode)) {
7003                 btrfs_add_ordered_operation(trans, root, old_inode);
7004         }
7005
7006         old_dir->i_ctime = old_dir->i_mtime = ctime;
7007         new_dir->i_ctime = new_dir->i_mtime = ctime;
7008         old_inode->i_ctime = ctime;
7009
7010         if (old_dentry->d_parent != new_dentry->d_parent)
7011                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7012
7013         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7014                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7015                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7016                                         old_dentry->d_name.name,
7017                                         old_dentry->d_name.len);
7018         } else {
7019                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7020                                         old_dentry->d_inode,
7021                                         old_dentry->d_name.name,
7022                                         old_dentry->d_name.len);
7023                 if (!ret)
7024                         ret = btrfs_update_inode(trans, root, old_inode);
7025         }
7026         BUG_ON(ret);
7027
7028         if (new_inode) {
7029                 new_inode->i_ctime = CURRENT_TIME;
7030                 if (unlikely(new_inode->i_ino ==
7031                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7032                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7033                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7034                                                 root_objectid,
7035                                                 new_dentry->d_name.name,
7036                                                 new_dentry->d_name.len);
7037                         BUG_ON(new_inode->i_nlink == 0);
7038                 } else {
7039                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7040                                                  new_dentry->d_inode,
7041                                                  new_dentry->d_name.name,
7042                                                  new_dentry->d_name.len);
7043                 }
7044                 BUG_ON(ret);
7045                 if (new_inode->i_nlink == 0) {
7046                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7047                         BUG_ON(ret);
7048                 }
7049         }
7050
7051         fixup_inode_flags(new_dir, old_inode);
7052
7053         ret = btrfs_add_link(trans, new_dir, old_inode,
7054                              new_dentry->d_name.name,
7055                              new_dentry->d_name.len, 0, index);
7056         BUG_ON(ret);
7057
7058         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7059                 struct dentry *parent = dget_parent(new_dentry);
7060                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7061                 dput(parent);
7062                 btrfs_end_log_trans(root);
7063         }
7064 out_fail:
7065         btrfs_end_transaction_throttle(trans, root);
7066
7067         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7068                 up_read(&root->fs_info->subvol_sem);
7069
7070         return ret;
7071 }
7072
7073 /*
7074  * some fairly slow code that needs optimization. This walks the list
7075  * of all the inodes with pending delalloc and forces them to disk.
7076  */
7077 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7078 {
7079         struct list_head *head = &root->fs_info->delalloc_inodes;
7080         struct btrfs_inode *binode;
7081         struct inode *inode;
7082
7083         if (root->fs_info->sb->s_flags & MS_RDONLY)
7084                 return -EROFS;
7085
7086         spin_lock(&root->fs_info->delalloc_lock);
7087         while (!list_empty(head)) {
7088                 binode = list_entry(head->next, struct btrfs_inode,
7089                                     delalloc_inodes);
7090                 inode = igrab(&binode->vfs_inode);
7091                 if (!inode)
7092                         list_del_init(&binode->delalloc_inodes);
7093                 spin_unlock(&root->fs_info->delalloc_lock);
7094                 if (inode) {
7095                         filemap_flush(inode->i_mapping);
7096                         if (delay_iput)
7097                                 btrfs_add_delayed_iput(inode);
7098                         else
7099                                 iput(inode);
7100                 }
7101                 cond_resched();
7102                 spin_lock(&root->fs_info->delalloc_lock);
7103         }
7104         spin_unlock(&root->fs_info->delalloc_lock);
7105
7106         /* the filemap_flush will queue IO into the worker threads, but
7107          * we have to make sure the IO is actually started and that
7108          * ordered extents get created before we return
7109          */
7110         atomic_inc(&root->fs_info->async_submit_draining);
7111         while (atomic_read(&root->fs_info->nr_async_submits) ||
7112               atomic_read(&root->fs_info->async_delalloc_pages)) {
7113                 wait_event(root->fs_info->async_submit_wait,
7114                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7115                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7116         }
7117         atomic_dec(&root->fs_info->async_submit_draining);
7118         return 0;
7119 }
7120
7121 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7122                                    int sync)
7123 {
7124         struct btrfs_inode *binode;
7125         struct inode *inode = NULL;
7126
7127         spin_lock(&root->fs_info->delalloc_lock);
7128         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7129                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7130                                     struct btrfs_inode, delalloc_inodes);
7131                 inode = igrab(&binode->vfs_inode);
7132                 if (inode) {
7133                         list_move_tail(&binode->delalloc_inodes,
7134                                        &root->fs_info->delalloc_inodes);
7135                         break;
7136                 }
7137
7138                 list_del_init(&binode->delalloc_inodes);
7139                 cond_resched_lock(&root->fs_info->delalloc_lock);
7140         }
7141         spin_unlock(&root->fs_info->delalloc_lock);
7142
7143         if (inode) {
7144                 if (sync) {
7145                         filemap_write_and_wait(inode->i_mapping);
7146                         /*
7147                          * We have to do this because compression doesn't
7148                          * actually set PG_writeback until it submits the pages
7149                          * for IO, which happens in an async thread, so we could
7150                          * race and not actually wait for any writeback pages
7151                          * because they've not been submitted yet.  Technically
7152                          * this could still be the case for the ordered stuff
7153                          * since the async thread may not have started to do its
7154                          * work yet.  If this becomes the case then we need to
7155                          * figure out a way to make sure that in writepage we
7156                          * wait for any async pages to be submitted before
7157                          * returning so that fdatawait does what its supposed to
7158                          * do.
7159                          */
7160                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7161                 } else {
7162                         filemap_flush(inode->i_mapping);
7163                 }
7164                 if (delay_iput)
7165                         btrfs_add_delayed_iput(inode);
7166                 else
7167                         iput(inode);
7168                 return 1;
7169         }
7170         return 0;
7171 }
7172
7173 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7174                          const char *symname)
7175 {
7176         struct btrfs_trans_handle *trans;
7177         struct btrfs_root *root = BTRFS_I(dir)->root;
7178         struct btrfs_path *path;
7179         struct btrfs_key key;
7180         struct inode *inode = NULL;
7181         int err;
7182         int drop_inode = 0;
7183         u64 objectid;
7184         u64 index = 0 ;
7185         int name_len;
7186         int datasize;
7187         unsigned long ptr;
7188         struct btrfs_file_extent_item *ei;
7189         struct extent_buffer *leaf;
7190         unsigned long nr = 0;
7191
7192         name_len = strlen(symname) + 1;
7193         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7194                 return -ENAMETOOLONG;
7195
7196         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7197         if (err)
7198                 return err;
7199         /*
7200          * 2 items for inode item and ref
7201          * 2 items for dir items
7202          * 1 item for xattr if selinux is on
7203          */
7204         trans = btrfs_start_transaction(root, 5);
7205         if (IS_ERR(trans))
7206                 return PTR_ERR(trans);
7207
7208         btrfs_set_trans_block_group(trans, dir);
7209
7210         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7211                                 dentry->d_name.len, dir->i_ino, objectid,
7212                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7213                                 &index);
7214         err = PTR_ERR(inode);
7215         if (IS_ERR(inode))
7216                 goto out_unlock;
7217
7218         err = btrfs_init_inode_security(trans, inode, dir);
7219         if (err) {
7220                 drop_inode = 1;
7221                 goto out_unlock;
7222         }
7223
7224         btrfs_set_trans_block_group(trans, inode);
7225         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7226         if (err)
7227                 drop_inode = 1;
7228         else {
7229                 inode->i_mapping->a_ops = &btrfs_aops;
7230                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7231                 inode->i_fop = &btrfs_file_operations;
7232                 inode->i_op = &btrfs_file_inode_operations;
7233                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7234         }
7235         btrfs_update_inode_block_group(trans, inode);
7236         btrfs_update_inode_block_group(trans, dir);
7237         if (drop_inode)
7238                 goto out_unlock;
7239
7240         path = btrfs_alloc_path();
7241         BUG_ON(!path);
7242         key.objectid = inode->i_ino;
7243         key.offset = 0;
7244         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7245         datasize = btrfs_file_extent_calc_inline_size(name_len);
7246         err = btrfs_insert_empty_item(trans, root, path, &key,
7247                                       datasize);
7248         if (err) {
7249                 drop_inode = 1;
7250                 goto out_unlock;
7251         }
7252         leaf = path->nodes[0];
7253         ei = btrfs_item_ptr(leaf, path->slots[0],
7254                             struct btrfs_file_extent_item);
7255         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7256         btrfs_set_file_extent_type(leaf, ei,
7257                                    BTRFS_FILE_EXTENT_INLINE);
7258         btrfs_set_file_extent_encryption(leaf, ei, 0);
7259         btrfs_set_file_extent_compression(leaf, ei, 0);
7260         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7261         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7262
7263         ptr = btrfs_file_extent_inline_start(ei);
7264         write_extent_buffer(leaf, symname, ptr, name_len);
7265         btrfs_mark_buffer_dirty(leaf);
7266         btrfs_free_path(path);
7267
7268         inode->i_op = &btrfs_symlink_inode_operations;
7269         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7270         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7271         inode_set_bytes(inode, name_len);
7272         btrfs_i_size_write(inode, name_len - 1);
7273         err = btrfs_update_inode(trans, root, inode);
7274         if (err)
7275                 drop_inode = 1;
7276
7277 out_unlock:
7278         nr = trans->blocks_used;
7279         btrfs_end_transaction_throttle(trans, root);
7280         if (drop_inode) {
7281                 inode_dec_link_count(inode);
7282                 iput(inode);
7283         }
7284         btrfs_btree_balance_dirty(root, nr);
7285         return err;
7286 }
7287
7288 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7289                                        u64 start, u64 num_bytes, u64 min_size,
7290                                        loff_t actual_len, u64 *alloc_hint,
7291                                        struct btrfs_trans_handle *trans)
7292 {
7293         struct btrfs_root *root = BTRFS_I(inode)->root;
7294         struct btrfs_key ins;
7295         u64 cur_offset = start;
7296         u64 i_size;
7297         int ret = 0;
7298         bool own_trans = true;
7299
7300         if (trans)
7301                 own_trans = false;
7302         while (num_bytes > 0) {
7303                 if (own_trans) {
7304                         trans = btrfs_start_transaction(root, 3);
7305                         if (IS_ERR(trans)) {
7306                                 ret = PTR_ERR(trans);
7307                                 break;
7308                         }
7309                 }
7310
7311                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7312                                            0, *alloc_hint, (u64)-1, &ins, 1);
7313                 if (ret) {
7314                         if (own_trans)
7315                                 btrfs_end_transaction(trans, root);
7316                         break;
7317                 }
7318
7319                 ret = insert_reserved_file_extent(trans, inode,
7320                                                   cur_offset, ins.objectid,
7321                                                   ins.offset, ins.offset,
7322                                                   ins.offset, 0, 0, 0,
7323                                                   BTRFS_FILE_EXTENT_PREALLOC);
7324                 BUG_ON(ret);
7325                 btrfs_drop_extent_cache(inode, cur_offset,
7326                                         cur_offset + ins.offset -1, 0);
7327
7328                 num_bytes -= ins.offset;
7329                 cur_offset += ins.offset;
7330                 *alloc_hint = ins.objectid + ins.offset;
7331
7332                 inode->i_ctime = CURRENT_TIME;
7333                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7334                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7335                     (actual_len > inode->i_size) &&
7336                     (cur_offset > inode->i_size)) {
7337                         if (cur_offset > actual_len)
7338                                 i_size = actual_len;
7339                         else
7340                                 i_size = cur_offset;
7341                         i_size_write(inode, i_size);
7342                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7343                 }
7344
7345                 ret = btrfs_update_inode(trans, root, inode);
7346                 BUG_ON(ret);
7347
7348                 if (own_trans)
7349                         btrfs_end_transaction(trans, root);
7350         }
7351         return ret;
7352 }
7353
7354 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7355                               u64 start, u64 num_bytes, u64 min_size,
7356                               loff_t actual_len, u64 *alloc_hint)
7357 {
7358         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7359                                            min_size, actual_len, alloc_hint,
7360                                            NULL);
7361 }
7362
7363 int btrfs_prealloc_file_range_trans(struct inode *inode,
7364                                     struct btrfs_trans_handle *trans, int mode,
7365                                     u64 start, u64 num_bytes, u64 min_size,
7366                                     loff_t actual_len, u64 *alloc_hint)
7367 {
7368         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7369                                            min_size, actual_len, alloc_hint, trans);
7370 }
7371
7372 static int btrfs_set_page_dirty(struct page *page)
7373 {
7374         return __set_page_dirty_nobuffers(page);
7375 }
7376
7377 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7378 {
7379         struct btrfs_root *root = BTRFS_I(inode)->root;
7380
7381         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7382                 return -EROFS;
7383         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7384                 return -EACCES;
7385         return generic_permission(inode, mask, flags, btrfs_check_acl);
7386 }
7387
7388 static const struct inode_operations btrfs_dir_inode_operations = {
7389         .getattr        = btrfs_getattr,
7390         .lookup         = btrfs_lookup,
7391         .create         = btrfs_create,
7392         .unlink         = btrfs_unlink,
7393         .link           = btrfs_link,
7394         .mkdir          = btrfs_mkdir,
7395         .rmdir          = btrfs_rmdir,
7396         .rename         = btrfs_rename,
7397         .symlink        = btrfs_symlink,
7398         .setattr        = btrfs_setattr,
7399         .mknod          = btrfs_mknod,
7400         .setxattr       = btrfs_setxattr,
7401         .getxattr       = btrfs_getxattr,
7402         .listxattr      = btrfs_listxattr,
7403         .removexattr    = btrfs_removexattr,
7404         .permission     = btrfs_permission,
7405 };
7406 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7407         .lookup         = btrfs_lookup,
7408         .permission     = btrfs_permission,
7409 };
7410
7411 static const struct file_operations btrfs_dir_file_operations = {
7412         .llseek         = generic_file_llseek,
7413         .read           = generic_read_dir,
7414         .readdir        = btrfs_real_readdir,
7415         .unlocked_ioctl = btrfs_ioctl,
7416 #ifdef CONFIG_COMPAT
7417         .compat_ioctl   = btrfs_ioctl,
7418 #endif
7419         .release        = btrfs_release_file,
7420         .fsync          = btrfs_sync_file,
7421 };
7422
7423 static struct extent_io_ops btrfs_extent_io_ops = {
7424         .fill_delalloc = run_delalloc_range,
7425         .submit_bio_hook = btrfs_submit_bio_hook,
7426         .merge_bio_hook = btrfs_merge_bio_hook,
7427         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7428         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7429         .writepage_start_hook = btrfs_writepage_start_hook,
7430         .readpage_io_failed_hook = btrfs_io_failed_hook,
7431         .set_bit_hook = btrfs_set_bit_hook,
7432         .clear_bit_hook = btrfs_clear_bit_hook,
7433         .merge_extent_hook = btrfs_merge_extent_hook,
7434         .split_extent_hook = btrfs_split_extent_hook,
7435 };
7436
7437 /*
7438  * btrfs doesn't support the bmap operation because swapfiles
7439  * use bmap to make a mapping of extents in the file.  They assume
7440  * these extents won't change over the life of the file and they
7441  * use the bmap result to do IO directly to the drive.
7442  *
7443  * the btrfs bmap call would return logical addresses that aren't
7444  * suitable for IO and they also will change frequently as COW
7445  * operations happen.  So, swapfile + btrfs == corruption.
7446  *
7447  * For now we're avoiding this by dropping bmap.
7448  */
7449 static const struct address_space_operations btrfs_aops = {
7450         .readpage       = btrfs_readpage,
7451         .writepage      = btrfs_writepage,
7452         .writepages     = btrfs_writepages,
7453         .readpages      = btrfs_readpages,
7454         .sync_page      = block_sync_page,
7455         .direct_IO      = btrfs_direct_IO,
7456         .invalidatepage = btrfs_invalidatepage,
7457         .releasepage    = btrfs_releasepage,
7458         .set_page_dirty = btrfs_set_page_dirty,
7459         .error_remove_page = generic_error_remove_page,
7460 };
7461
7462 static const struct address_space_operations btrfs_symlink_aops = {
7463         .readpage       = btrfs_readpage,
7464         .writepage      = btrfs_writepage,
7465         .invalidatepage = btrfs_invalidatepage,
7466         .releasepage    = btrfs_releasepage,
7467 };
7468
7469 static const struct inode_operations btrfs_file_inode_operations = {
7470         .getattr        = btrfs_getattr,
7471         .setattr        = btrfs_setattr,
7472         .setxattr       = btrfs_setxattr,
7473         .getxattr       = btrfs_getxattr,
7474         .listxattr      = btrfs_listxattr,
7475         .removexattr    = btrfs_removexattr,
7476         .permission     = btrfs_permission,
7477         .fiemap         = btrfs_fiemap,
7478 };
7479 static const struct inode_operations btrfs_special_inode_operations = {
7480         .getattr        = btrfs_getattr,
7481         .setattr        = btrfs_setattr,
7482         .permission     = btrfs_permission,
7483         .setxattr       = btrfs_setxattr,
7484         .getxattr       = btrfs_getxattr,
7485         .listxattr      = btrfs_listxattr,
7486         .removexattr    = btrfs_removexattr,
7487 };
7488 static const struct inode_operations btrfs_symlink_inode_operations = {
7489         .readlink       = generic_readlink,
7490         .follow_link    = page_follow_link_light,
7491         .put_link       = page_put_link,
7492         .getattr        = btrfs_getattr,
7493         .permission     = btrfs_permission,
7494         .setxattr       = btrfs_setxattr,
7495         .getxattr       = btrfs_getxattr,
7496         .listxattr      = btrfs_listxattr,
7497         .removexattr    = btrfs_removexattr,
7498 };
7499
7500 const struct dentry_operations btrfs_dentry_operations = {
7501         .d_delete       = btrfs_dentry_delete,
7502 };