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/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "compat.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "ordered-data.h"
  53 #include "xattr.h"
  54 #include "tree-log.h"
  55 #include "volumes.h"
  56 #include "compression.h"
  57 #include "locking.h"
  58 #include "free-space-cache.h"
  59 #include "inode-map.h"
  60 #include "backref.h"
  61 #include "hash.h"
  62
  63 struct btrfs_iget_args {
  64         u64 ino;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_root *root, struct inode *inode,
 130                                 u64 start, size_t size, size_t compressed_size,
 131                                 int compress_type,
 132                                 struct page **compressed_pages)
 133 {
 134         struct btrfs_key key;
 135         struct btrfs_path *path;
 136         struct extent_buffer *leaf;
 137         struct page *page = NULL;
 138         char *kaddr;
 139         unsigned long ptr;
 140         struct btrfs_file_extent_item *ei;
 141         int err = 0;
 142         int ret;
 143         size_t cur_size = size;
 144         size_t datasize;
 145         unsigned long offset;
 146
 147         if (compressed_size && compressed_pages)
 148                 cur_size = compressed_size;
 149
 150         path = btrfs_alloc_path();
 151         if (!path)
 152                 return -ENOMEM;
 153
 154         path->leave_spinning = 1;
 155
 156         key.objectid = btrfs_ino(inode);
 157         key.offset = start;
 158         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 159         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 160
 161         inode_add_bytes(inode, size);
 162         ret = btrfs_insert_empty_item(trans, root, path, &key,
 163                                       datasize);
 164         if (ret) {
 165                 err = ret;
 166                 goto fail;
 167         }
 168         leaf = path->nodes[0];
 169         ei = btrfs_item_ptr(leaf, path->slots[0],
 170                             struct btrfs_file_extent_item);
 171         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 172         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 173         btrfs_set_file_extent_encryption(leaf, ei, 0);
 174         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 175         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 176         ptr = btrfs_file_extent_inline_start(ei);
 177
 178         if (compress_type != BTRFS_COMPRESS_NONE) {
 179                 struct page *cpage;
 180                 int i = 0;
 181                 while (compressed_size > 0) {
 182                         cpage = compressed_pages[i];
 183                         cur_size = min_t(unsigned long, compressed_size,
 184                                        PAGE_CACHE_SIZE);
 185
 186                         kaddr = kmap_atomic(cpage);
 187                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 188                         kunmap_atomic(kaddr);
 189
 190                         i++;
 191                         ptr += cur_size;
 192                         compressed_size -= cur_size;
 193                 }
 194                 btrfs_set_file_extent_compression(leaf, ei,
 195                                                   compress_type);
 196         } else {
 197                 page = find_get_page(inode->i_mapping,
 198                                      start >> PAGE_CACHE_SHIFT);
 199                 btrfs_set_file_extent_compression(leaf, ei, 0);
 200                 kaddr = kmap_atomic(page);
 201                 offset = start & (PAGE_CACHE_SIZE - 1);
 202                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 203                 kunmap_atomic(kaddr);
 204                 page_cache_release(page);
 205         }
 206         btrfs_mark_buffer_dirty(leaf);
 207         btrfs_free_path(path);
 208
 209         /*
 210          * we're an inline extent, so nobody can
 211          * extend the file past i_size without locking
 212          * a page we already have locked.
 213          *
 214          * We must do any isize and inode updates
 215          * before we unlock the pages.  Otherwise we
 216          * could end up racing with unlink.
 217          */
 218         BTRFS_I(inode)->disk_i_size = inode->i_size;
 219         ret = btrfs_update_inode(trans, root, inode);
 220
 221         return ret;
 222 fail:
 223         btrfs_free_path(path);
 224         return err;
 225 }
 226
 227
 228 /*
 229  * conditionally insert an inline extent into the file.  This
 230  * does the checks required to make sure the data is small enough
 231  * to fit as an inline extent.
 232  */
 233 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 234                                  struct btrfs_root *root,
 235                                  struct inode *inode, u64 start, u64 end,
 236                                  size_t compressed_size, int compress_type,
 237                                  struct page **compressed_pages)
 238 {
 239         u64 isize = i_size_read(inode);
 240         u64 actual_end = min(end + 1, isize);
 241         u64 inline_len = actual_end - start;
 242         u64 aligned_end = ALIGN(end, root->sectorsize);
 243         u64 data_len = inline_len;
 244         int ret;
 245
 246         if (compressed_size)
 247                 data_len = compressed_size;
 248
 249         if (start > 0 ||
 250             actual_end >= PAGE_CACHE_SIZE ||
 251             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 252             (!compressed_size &&
 253             (actual_end & (root->sectorsize - 1)) == 0) ||
 254             end + 1 < isize ||
 255             data_len > root->fs_info->max_inline) {
 256                 return 1;
 257         }
 258
 259         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 260         if (ret)
 261                 return ret;
 262
 263         if (isize > actual_end)
 264                 inline_len = min_t(u64, isize, actual_end);
 265         ret = insert_inline_extent(trans, root, inode, start,
 266                                    inline_len, compressed_size,
 267                                    compress_type, compressed_pages);
 268         if (ret && ret != -ENOSPC) {
 269                 btrfs_abort_transaction(trans, root, ret);
 270                 return ret;
 271         } else if (ret == -ENOSPC) {
 272                 return 1;
 273         }
 274
 275         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 276         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 277         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 278         return 0;
 279 }
 280
 281 struct async_extent {
 282         u64 start;
 283         u64 ram_size;
 284         u64 compressed_size;
 285         struct page **pages;
 286         unsigned long nr_pages;
 287         int compress_type;
 288         struct list_head list;
 289 };
 290
 291 struct async_cow {
 292         struct inode *inode;
 293         struct btrfs_root *root;
 294         struct page *locked_page;
 295         u64 start;
 296         u64 end;
 297         struct list_head extents;
 298         struct btrfs_work work;
 299 };
 300
 301 static noinline int add_async_extent(struct async_cow *cow,
 302                                      u64 start, u64 ram_size,
 303                                      u64 compressed_size,
 304                                      struct page **pages,
 305                                      unsigned long nr_pages,
 306                                      int compress_type)
 307 {
 308         struct async_extent *async_extent;
 309
 310         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 311         BUG_ON(!async_extent); /* -ENOMEM */
 312         async_extent->start = start;
 313         async_extent->ram_size = ram_size;
 314         async_extent->compressed_size = compressed_size;
 315         async_extent->pages = pages;
 316         async_extent->nr_pages = nr_pages;
 317         async_extent->compress_type = compress_type;
 318         list_add_tail(&async_extent->list, &cow->extents);
 319         return 0;
 320 }
 321
 322 /*
 323  * we create compressed extents in two phases.  The first
 324  * phase compresses a range of pages that have already been
 325  * locked (both pages and state bits are locked).
 326  *
 327  * This is done inside an ordered work queue, and the compression
 328  * is spread across many cpus.  The actual IO submission is step
 329  * two, and the ordered work queue takes care of making sure that
 330  * happens in the same order things were put onto the queue by
 331  * writepages and friends.
 332  *
 333  * If this code finds it can't get good compression, it puts an
 334  * entry onto the work queue to write the uncompressed bytes.  This
 335  * makes sure that both compressed inodes and uncompressed inodes
 336  * are written in the same order that the flusher thread sent them
 337  * down.
 338  */
 339 static noinline int compress_file_range(struct inode *inode,
 340                                         struct page *locked_page,
 341                                         u64 start, u64 end,
 342                                         struct async_cow *async_cow,
 343                                         int *num_added)
 344 {
 345         struct btrfs_root *root = BTRFS_I(inode)->root;
 346         struct btrfs_trans_handle *trans;
 347         u64 num_bytes;
 348         u64 blocksize = root->sectorsize;
 349         u64 actual_end;
 350         u64 isize = i_size_read(inode);
 351         int ret = 0;
 352         struct page **pages = NULL;
 353         unsigned long nr_pages;
 354         unsigned long nr_pages_ret = 0;
 355         unsigned long total_compressed = 0;
 356         unsigned long total_in = 0;
 357         unsigned long max_compressed = 128 * 1024;
 358         unsigned long max_uncompressed = 128 * 1024;
 359         int i;
 360         int will_compress;
 361         int compress_type = root->fs_info->compress_type;
 362         int redirty = 0;
 363
 364         /* if this is a small write inside eof, kick off a defrag */
 365         if ((end - start + 1) < 16 * 1024 &&
 366             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 367                 btrfs_add_inode_defrag(NULL, inode);
 368
 369         actual_end = min_t(u64, isize, end + 1);
 370 again:
 371         will_compress = 0;
 372         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 373         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 374
 375         /*
 376          * we don't want to send crud past the end of i_size through
 377          * compression, that's just a waste of CPU time.  So, if the
 378          * end of the file is before the start of our current
 379          * requested range of bytes, we bail out to the uncompressed
 380          * cleanup code that can deal with all of this.
 381          *
 382          * It isn't really the fastest way to fix things, but this is a
 383          * very uncommon corner.
 384          */
 385         if (actual_end <= start)
 386                 goto cleanup_and_bail_uncompressed;
 387
 388         total_compressed = actual_end - start;
 389
 390         /* we want to make sure that amount of ram required to uncompress
 391          * an extent is reasonable, so we limit the total size in ram
 392          * of a compressed extent to 128k.  This is a crucial number
 393          * because it also controls how easily we can spread reads across
 394          * cpus for decompression.
 395          *
 396          * We also want to make sure the amount of IO required to do
 397          * a random read is reasonably small, so we limit the size of
 398          * a compressed extent to 128k.
 399          */
 400         total_compressed = min(total_compressed, max_uncompressed);
 401         num_bytes = ALIGN(end - start + 1, blocksize);
 402         num_bytes = max(blocksize,  num_bytes);
 403         total_in = 0;
 404         ret = 0;
 405
 406         /*
 407          * we do compression for mount -o compress and when the
 408          * inode has not been flagged as nocompress.  This flag can
 409          * change at any time if we discover bad compression ratios.
 410          */
 411         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 412             (btrfs_test_opt(root, COMPRESS) ||
 413              (BTRFS_I(inode)->force_compress) ||
 414              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 415                 WARN_ON(pages);
 416                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 417                 if (!pages) {
 418                         /* just bail out to the uncompressed code */
 419                         goto cont;
 420                 }
 421
 422                 if (BTRFS_I(inode)->force_compress)
 423                         compress_type = BTRFS_I(inode)->force_compress;
 424
 425                 /*
 426                  * we need to call clear_page_dirty_for_io on each
 427                  * page in the range.  Otherwise applications with the file
 428                  * mmap'd can wander in and change the page contents while
 429                  * we are compressing them.
 430                  *
 431                  * If the compression fails for any reason, we set the pages
 432                  * dirty again later on.
 433                  */
 434                 extent_range_clear_dirty_for_io(inode, start, end);
 435                 redirty = 1;
 436                 ret = btrfs_compress_pages(compress_type,
 437                                            inode->i_mapping, start,
 438                                            total_compressed, pages,
 439                                            nr_pages, &nr_pages_ret,
 440                                            &total_in,
 441                                            &total_compressed,
 442                                            max_compressed);
 443
 444                 if (!ret) {
 445                         unsigned long offset = total_compressed &
 446                                 (PAGE_CACHE_SIZE - 1);
 447                         struct page *page = pages[nr_pages_ret - 1];
 448                         char *kaddr;
 449
 450                         /* zero the tail end of the last page, we might be
 451                          * sending it down to disk
 452                          */
 453                         if (offset) {
 454                                 kaddr = kmap_atomic(page);
 455                                 memset(kaddr + offset, 0,
 456                                        PAGE_CACHE_SIZE - offset);
 457                                 kunmap_atomic(kaddr);
 458                         }
 459                         will_compress = 1;
 460                 }
 461         }
 462 cont:
 463         if (start == 0) {
 464                 trans = btrfs_join_transaction(root);
 465                 if (IS_ERR(trans)) {
 466                         ret = PTR_ERR(trans);
 467                         trans = NULL;
 468                         goto cleanup_and_out;
 469                 }
 470                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 471
 472                 /* lets try to make an inline extent */
 473                 if (ret || total_in < (actual_end - start)) {
 474                         /* we didn't compress the entire range, try
 475                          * to make an uncompressed inline extent.
 476                          */
 477                         ret = cow_file_range_inline(trans, root, inode,
 478                                                     start, end, 0, 0, NULL);
 479                 } else {
 480                         /* try making a compressed inline extent */
 481                         ret = cow_file_range_inline(trans, root, inode,
 482                                                     start, end,
 483                                                     total_compressed,
 484                                                     compress_type, pages);
 485                 }
 486                 if (ret <= 0) {
 487                         /*
 488                          * inline extent creation worked or returned error,
 489                          * we don't need to create any more async work items.
 490                          * Unlock and free up our temp pages.
 491                          */
 492                         extent_clear_unlock_delalloc(inode,
 493                              &BTRFS_I(inode)->io_tree,
 494                              start, end, NULL,
 495                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 496                              EXTENT_CLEAR_DELALLOC |
 497                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 498
 499                         btrfs_end_transaction(trans, root);
 500                         goto free_pages_out;
 501                 }
 502                 btrfs_end_transaction(trans, root);
 503         }
 504
 505         if (will_compress) {
 506                 /*
 507                  * we aren't doing an inline extent round the compressed size
 508                  * up to a block size boundary so the allocator does sane
 509                  * things
 510                  */
 511                 total_compressed = ALIGN(total_compressed, blocksize);
 512
 513                 /*
 514                  * one last check to make sure the compression is really a
 515                  * win, compare the page count read with the blocks on disk
 516                  */
 517                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 518                 if (total_compressed >= total_in) {
 519                         will_compress = 0;
 520                 } else {
 521                         num_bytes = total_in;
 522                 }
 523         }
 524         if (!will_compress && pages) {
 525                 /*
 526                  * the compression code ran but failed to make things smaller,
 527                  * free any pages it allocated and our page pointer array
 528                  */
 529                 for (i = 0; i < nr_pages_ret; i++) {
 530                         WARN_ON(pages[i]->mapping);
 531                         page_cache_release(pages[i]);
 532                 }
 533                 kfree(pages);
 534                 pages = NULL;
 535                 total_compressed = 0;
 536                 nr_pages_ret = 0;
 537
 538                 /* flag the file so we don't compress in the future */
 539                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 540                     !(BTRFS_I(inode)->force_compress)) {
 541                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 542                 }
 543         }
 544         if (will_compress) {
 545                 *num_added += 1;
 546
 547                 /* the async work queues will take care of doing actual
 548                  * allocation on disk for these compressed pages,
 549                  * and will submit them to the elevator.
 550                  */
 551                 add_async_extent(async_cow, start, num_bytes,
 552                                  total_compressed, pages, nr_pages_ret,
 553                                  compress_type);
 554
 555                 if (start + num_bytes < end) {
 556                         start += num_bytes;
 557                         pages = NULL;
 558                         cond_resched();
 559                         goto again;
 560                 }
 561         } else {
 562 cleanup_and_bail_uncompressed:
 563                 /*
 564                  * No compression, but we still need to write the pages in
 565                  * the file we've been given so far.  redirty the locked
 566                  * page if it corresponds to our extent and set things up
 567                  * for the async work queue to run cow_file_range to do
 568                  * the normal delalloc dance
 569                  */
 570                 if (page_offset(locked_page) >= start &&
 571                     page_offset(locked_page) <= end) {
 572                         __set_page_dirty_nobuffers(locked_page);
 573                         /* unlocked later on in the async handlers */
 574                 }
 575                 if (redirty)
 576                         extent_range_redirty_for_io(inode, start, end);
 577                 add_async_extent(async_cow, start, end - start + 1,
 578                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 579                 *num_added += 1;
 580         }
 581
 582 out:
 583         return ret;
 584
 585 free_pages_out:
 586         for (i = 0; i < nr_pages_ret; i++) {
 587                 WARN_ON(pages[i]->mapping);
 588                 page_cache_release(pages[i]);
 589         }
 590         kfree(pages);
 591
 592         goto out;
 593
 594 cleanup_and_out:
 595         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 596                                      start, end, NULL,
 597                                      EXTENT_CLEAR_UNLOCK_PAGE |
 598                                      EXTENT_CLEAR_DIRTY |
 599                                      EXTENT_CLEAR_DELALLOC |
 600                                      EXTENT_SET_WRITEBACK |
 601                                      EXTENT_END_WRITEBACK);
 602         if (!trans || IS_ERR(trans))
 603                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 604         else
 605                 btrfs_abort_transaction(trans, root, ret);
 606         goto free_pages_out;
 607 }
 608
 609 /*
 610  * phase two of compressed writeback.  This is the ordered portion
 611  * of the code, which only gets called in the order the work was
 612  * queued.  We walk all the async extents created by compress_file_range
 613  * and send them down to the disk.
 614  */
 615 static noinline int submit_compressed_extents(struct inode *inode,
 616                                               struct async_cow *async_cow)
 617 {
 618         struct async_extent *async_extent;
 619         u64 alloc_hint = 0;
 620         struct btrfs_trans_handle *trans;
 621         struct btrfs_key ins;
 622         struct extent_map *em;
 623         struct btrfs_root *root = BTRFS_I(inode)->root;
 624         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 625         struct extent_io_tree *io_tree;
 626         int ret = 0;
 627
 628         if (list_empty(&async_cow->extents))
 629                 return 0;
 630
 631 again:
 632         while (!list_empty(&async_cow->extents)) {
 633                 async_extent = list_entry(async_cow->extents.next,
 634                                           struct async_extent, list);
 635                 list_del(&async_extent->list);
 636
 637                 io_tree = &BTRFS_I(inode)->io_tree;
 638
 639 retry:
 640                 /* did the compression code fall back to uncompressed IO? */
 641                 if (!async_extent->pages) {
 642                         int page_started = 0;
 643                         unsigned long nr_written = 0;
 644
 645                         lock_extent(io_tree, async_extent->start,
 646                                          async_extent->start +
 647                                          async_extent->ram_size - 1);
 648
 649                         /* allocate blocks */
 650                         ret = cow_file_range(inode, async_cow->locked_page,
 651                                              async_extent->start,
 652                                              async_extent->start +
 653                                              async_extent->ram_size - 1,
 654                                              &page_started, &nr_written, 0);
 655
 656                         /* JDM XXX */
 657
 658                         /*
 659                          * if page_started, cow_file_range inserted an
 660                          * inline extent and took care of all the unlocking
 661                          * and IO for us.  Otherwise, we need to submit
 662                          * all those pages down to the drive.
 663                          */
 664                         if (!page_started && !ret)
 665                                 extent_write_locked_range(io_tree,
 666                                                   inode, async_extent->start,
 667                                                   async_extent->start +
 668                                                   async_extent->ram_size - 1,
 669                                                   btrfs_get_extent,
 670                                                   WB_SYNC_ALL);
 671                         else if (ret)
 672                                 unlock_page(async_cow->locked_page);
 673                         kfree(async_extent);
 674                         cond_resched();
 675                         continue;
 676                 }
 677
 678                 lock_extent(io_tree, async_extent->start,
 679                             async_extent->start + async_extent->ram_size - 1);
 680
 681                 trans = btrfs_join_transaction(root);
 682                 if (IS_ERR(trans)) {
 683                         ret = PTR_ERR(trans);
 684                 } else {
 685                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 686                         ret = btrfs_reserve_extent(trans, root,
 687                                            async_extent->compressed_size,
 688                                            async_extent->compressed_size,
 689                                            0, alloc_hint, &ins, 1);
 690                         if (ret && ret != -ENOSPC)
 691                                 btrfs_abort_transaction(trans, root, ret);
 692                         btrfs_end_transaction(trans, root);
 693                 }
 694
 695                 if (ret) {
 696                         int i;
 697
 698                         for (i = 0; i < async_extent->nr_pages; i++) {
 699                                 WARN_ON(async_extent->pages[i]->mapping);
 700                                 page_cache_release(async_extent->pages[i]);
 701                         }
 702                         kfree(async_extent->pages);
 703                         async_extent->nr_pages = 0;
 704                         async_extent->pages = NULL;
 705
 706                         if (ret == -ENOSPC) {
 707                                 unlock_extent(io_tree, async_extent->start,
 708                                               async_extent->start +
 709                                               async_extent->ram_size - 1);
 710                                 goto retry;
 711                         }
 712                         goto out_free;
 713                 }
 714
 715                 /*
 716                  * here we're doing allocation and writeback of the
 717                  * compressed pages
 718                  */
 719                 btrfs_drop_extent_cache(inode, async_extent->start,
 720                                         async_extent->start +
 721                                         async_extent->ram_size - 1, 0);
 722
 723                 em = alloc_extent_map();
 724                 if (!em) {
 725                         ret = -ENOMEM;
 726                         goto out_free_reserve;
 727                 }
 728                 em->start = async_extent->start;
 729                 em->len = async_extent->ram_size;
 730                 em->orig_start = em->start;
 731                 em->mod_start = em->start;
 732                 em->mod_len = em->len;
 733
 734                 em->block_start = ins.objectid;
 735                 em->block_len = ins.offset;
 736                 em->orig_block_len = ins.offset;
 737                 em->ram_bytes = async_extent->ram_size;
 738                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 739                 em->compress_type = async_extent->compress_type;
 740                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 741                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 742                 em->generation = -1;
 743
 744                 while (1) {
 745                         write_lock(&em_tree->lock);
 746                         ret = add_extent_mapping(em_tree, em, 1);
 747                         write_unlock(&em_tree->lock);
 748                         if (ret != -EEXIST) {
 749                                 free_extent_map(em);
 750                                 break;
 751                         }
 752                         btrfs_drop_extent_cache(inode, async_extent->start,
 753                                                 async_extent->start +
 754                                                 async_extent->ram_size - 1, 0);
 755                 }
 756
 757                 if (ret)
 758                         goto out_free_reserve;
 759
 760                 ret = btrfs_add_ordered_extent_compress(inode,
 761                                                 async_extent->start,
 762                                                 ins.objectid,
 763                                                 async_extent->ram_size,
 764                                                 ins.offset,
 765                                                 BTRFS_ORDERED_COMPRESSED,
 766                                                 async_extent->compress_type);
 767                 if (ret)
 768                         goto out_free_reserve;
 769
 770                 /*
 771                  * clear dirty, set writeback and unlock the pages.
 772                  */
 773                 extent_clear_unlock_delalloc(inode,
 774                                 &BTRFS_I(inode)->io_tree,
 775                                 async_extent->start,
 776                                 async_extent->start +
 777                                 async_extent->ram_size - 1,
 778                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 779                                 EXTENT_CLEAR_UNLOCK |
 780                                 EXTENT_CLEAR_DELALLOC |
 781                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 782
 783                 ret = btrfs_submit_compressed_write(inode,
 784                                     async_extent->start,
 785                                     async_extent->ram_size,
 786                                     ins.objectid,
 787                                     ins.offset, async_extent->pages,
 788                                     async_extent->nr_pages);
 789                 alloc_hint = ins.objectid + ins.offset;
 790                 kfree(async_extent);
 791                 if (ret)
 792                         goto out;
 793                 cond_resched();
 794         }
 795         ret = 0;
 796 out:
 797         return ret;
 798 out_free_reserve:
 799         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 800 out_free:
 801         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 802                                      async_extent->start,
 803                                      async_extent->start +
 804                                      async_extent->ram_size - 1,
 805                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 806                                      EXTENT_CLEAR_UNLOCK |
 807                                      EXTENT_CLEAR_DELALLOC |
 808                                      EXTENT_CLEAR_DIRTY |
 809                                      EXTENT_SET_WRITEBACK |
 810                                      EXTENT_END_WRITEBACK);
 811         kfree(async_extent);
 812         goto again;
 813 }
 814
 815 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 816                                       u64 num_bytes)
 817 {
 818         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 819         struct extent_map *em;
 820         u64 alloc_hint = 0;
 821
 822         read_lock(&em_tree->lock);
 823         em = search_extent_mapping(em_tree, start, num_bytes);
 824         if (em) {
 825                 /*
 826                  * if block start isn't an actual block number then find the
 827                  * first block in this inode and use that as a hint.  If that
 828                  * block is also bogus then just don't worry about it.
 829                  */
 830                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 831                         free_extent_map(em);
 832                         em = search_extent_mapping(em_tree, 0, 0);
 833                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 834                                 alloc_hint = em->block_start;
 835                         if (em)
 836                                 free_extent_map(em);
 837                 } else {
 838                         alloc_hint = em->block_start;
 839                         free_extent_map(em);
 840                 }
 841         }
 842         read_unlock(&em_tree->lock);
 843
 844         return alloc_hint;
 845 }
 846
 847 /*
 848  * when extent_io.c finds a delayed allocation range in the file,
 849  * the call backs end up in this code.  The basic idea is to
 850  * allocate extents on disk for the range, and create ordered data structs
 851  * in ram to track those extents.
 852  *
 853  * locked_page is the page that writepage had locked already.  We use
 854  * it to make sure we don't do extra locks or unlocks.
 855  *
 856  * *page_started is set to one if we unlock locked_page and do everything
 857  * required to start IO on it.  It may be clean and already done with
 858  * IO when we return.
 859  */
 860 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 861                                      struct inode *inode,
 862                                      struct btrfs_root *root,
 863                                      struct page *locked_page,
 864                                      u64 start, u64 end, int *page_started,
 865                                      unsigned long *nr_written,
 866                                      int unlock)
 867 {
 868         u64 alloc_hint = 0;
 869         u64 num_bytes;
 870         unsigned long ram_size;
 871         u64 disk_num_bytes;
 872         u64 cur_alloc_size;
 873         u64 blocksize = root->sectorsize;
 874         struct btrfs_key ins;
 875         struct extent_map *em;
 876         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 877         int ret = 0;
 878
 879         BUG_ON(btrfs_is_free_space_inode(inode));
 880
 881         num_bytes = ALIGN(end - start + 1, blocksize);
 882         num_bytes = max(blocksize,  num_bytes);
 883         disk_num_bytes = num_bytes;
 884
 885         /* if this is a small write inside eof, kick off defrag */
 886         if (num_bytes < 64 * 1024 &&
 887             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 888                 btrfs_add_inode_defrag(trans, inode);
 889
 890         if (start == 0) {
 891                 /* lets try to make an inline extent */
 892                 ret = cow_file_range_inline(trans, root, inode,
 893                                             start, end, 0, 0, NULL);
 894                 if (ret == 0) {
 895                         extent_clear_unlock_delalloc(inode,
 896                                      &BTRFS_I(inode)->io_tree,
 897                                      start, end, NULL,
 898                                      EXTENT_CLEAR_UNLOCK_PAGE |
 899                                      EXTENT_CLEAR_UNLOCK |
 900                                      EXTENT_CLEAR_DELALLOC |
 901                                      EXTENT_CLEAR_DIRTY |
 902                                      EXTENT_SET_WRITEBACK |
 903                                      EXTENT_END_WRITEBACK);
 904
 905                         *nr_written = *nr_written +
 906                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 907                         *page_started = 1;
 908                         goto out;
 909                 } else if (ret < 0) {
 910                         btrfs_abort_transaction(trans, root, ret);
 911                         goto out_unlock;
 912                 }
 913         }
 914
 915         BUG_ON(disk_num_bytes >
 916                btrfs_super_total_bytes(root->fs_info->super_copy));
 917
 918         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 919         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 920
 921         while (disk_num_bytes > 0) {
 922                 unsigned long op;
 923
 924                 cur_alloc_size = disk_num_bytes;
 925                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 926                                            root->sectorsize, 0, alloc_hint,
 927                                            &ins, 1);
 928                 if (ret < 0) {
 929                         btrfs_abort_transaction(trans, root, ret);
 930                         goto out_unlock;
 931                 }
 932
 933                 em = alloc_extent_map();
 934                 if (!em) {
 935                         ret = -ENOMEM;
 936                         goto out_reserve;
 937                 }
 938                 em->start = start;
 939                 em->orig_start = em->start;
 940                 ram_size = ins.offset;
 941                 em->len = ins.offset;
 942                 em->mod_start = em->start;
 943                 em->mod_len = em->len;
 944
 945                 em->block_start = ins.objectid;
 946                 em->block_len = ins.offset;
 947                 em->orig_block_len = ins.offset;
 948                 em->ram_bytes = ram_size;
 949                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 950                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 951                 em->generation = -1;
 952
 953                 while (1) {
 954                         write_lock(&em_tree->lock);
 955                         ret = add_extent_mapping(em_tree, em, 1);
 956                         write_unlock(&em_tree->lock);
 957                         if (ret != -EEXIST) {
 958                                 free_extent_map(em);
 959                                 break;
 960                         }
 961                         btrfs_drop_extent_cache(inode, start,
 962                                                 start + ram_size - 1, 0);
 963                 }
 964                 if (ret)
 965                         goto out_reserve;
 966
 967                 cur_alloc_size = ins.offset;
 968                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 969                                                ram_size, cur_alloc_size, 0);
 970                 if (ret)
 971                         goto out_reserve;
 972
 973                 if (root->root_key.objectid ==
 974                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 975                         ret = btrfs_reloc_clone_csums(inode, start,
 976                                                       cur_alloc_size);
 977                         if (ret) {
 978                                 btrfs_abort_transaction(trans, root, ret);
 979                                 goto out_reserve;
 980                         }
 981                 }
 982
 983                 if (disk_num_bytes < cur_alloc_size)
 984                         break;
 985
 986                 /* we're not doing compressed IO, don't unlock the first
 987                  * page (which the caller expects to stay locked), don't
 988                  * clear any dirty bits and don't set any writeback bits
 989                  *
 990                  * Do set the Private2 bit so we know this page was properly
 991                  * setup for writepage
 992                  */
 993                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 994                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 995                         EXTENT_SET_PRIVATE2;
 996
 997                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 998                                              start, start + ram_size - 1,
 999                                              locked_page, op);
1000                 disk_num_bytes -= cur_alloc_size;
1001                 num_bytes -= cur_alloc_size;
1002                 alloc_hint = ins.objectid + ins.offset;
1003                 start += cur_alloc_size;
1004         }
1005 out:
1006         return ret;
1007
1008 out_reserve:
1009         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1010 out_unlock:
1011         extent_clear_unlock_delalloc(inode,
1012                      &BTRFS_I(inode)->io_tree,
1013                      start, end, locked_page,
1014                      EXTENT_CLEAR_UNLOCK_PAGE |
1015                      EXTENT_CLEAR_UNLOCK |
1016                      EXTENT_CLEAR_DELALLOC |
1017                      EXTENT_CLEAR_DIRTY |
1018                      EXTENT_SET_WRITEBACK |
1019                      EXTENT_END_WRITEBACK);
1020
1021         goto out;
1022 }
1023
1024 static noinline int cow_file_range(struct inode *inode,
1025                                    struct page *locked_page,
1026                                    u64 start, u64 end, int *page_started,
1027                                    unsigned long *nr_written,
1028                                    int unlock)
1029 {
1030         struct btrfs_trans_handle *trans;
1031         struct btrfs_root *root = BTRFS_I(inode)->root;
1032         int ret;
1033
1034         trans = btrfs_join_transaction(root);
1035         if (IS_ERR(trans)) {
1036                 extent_clear_unlock_delalloc(inode,
1037                              &BTRFS_I(inode)->io_tree,
1038                              start, end, locked_page,
1039                              EXTENT_CLEAR_UNLOCK_PAGE |
1040                              EXTENT_CLEAR_UNLOCK |
1041                              EXTENT_CLEAR_DELALLOC |
1042                              EXTENT_CLEAR_DIRTY |
1043                              EXTENT_SET_WRITEBACK |
1044                              EXTENT_END_WRITEBACK);
1045                 return PTR_ERR(trans);
1046         }
1047         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1048
1049         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1050                                page_started, nr_written, unlock);
1051
1052         btrfs_end_transaction(trans, root);
1053
1054         return ret;
1055 }
1056
1057 /*
1058  * work queue call back to started compression on a file and pages
1059  */
1060 static noinline void async_cow_start(struct btrfs_work *work)
1061 {
1062         struct async_cow *async_cow;
1063         int num_added = 0;
1064         async_cow = container_of(work, struct async_cow, work);
1065
1066         compress_file_range(async_cow->inode, async_cow->locked_page,
1067                             async_cow->start, async_cow->end, async_cow,
1068                             &num_added);
1069         if (num_added == 0) {
1070                 btrfs_add_delayed_iput(async_cow->inode);
1071                 async_cow->inode = NULL;
1072         }
1073 }
1074
1075 /*
1076  * work queue call back to submit previously compressed pages
1077  */
1078 static noinline void async_cow_submit(struct btrfs_work *work)
1079 {
1080         struct async_cow *async_cow;
1081         struct btrfs_root *root;
1082         unsigned long nr_pages;
1083
1084         async_cow = container_of(work, struct async_cow, work);
1085
1086         root = async_cow->root;
1087         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1088                 PAGE_CACHE_SHIFT;
1089
1090         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1091             5 * 1024 * 1024 &&
1092             waitqueue_active(&root->fs_info->async_submit_wait))
1093                 wake_up(&root->fs_info->async_submit_wait);
1094
1095         if (async_cow->inode)
1096                 submit_compressed_extents(async_cow->inode, async_cow);
1097 }
1098
1099 static noinline void async_cow_free(struct btrfs_work *work)
1100 {
1101         struct async_cow *async_cow;
1102         async_cow = container_of(work, struct async_cow, work);
1103         if (async_cow->inode)
1104                 btrfs_add_delayed_iput(async_cow->inode);
1105         kfree(async_cow);
1106 }
1107
1108 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1109                                 u64 start, u64 end, int *page_started,
1110                                 unsigned long *nr_written)
1111 {
1112         struct async_cow *async_cow;
1113         struct btrfs_root *root = BTRFS_I(inode)->root;
1114         unsigned long nr_pages;
1115         u64 cur_end;
1116         int limit = 10 * 1024 * 1024;
1117
1118         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1119                          1, 0, NULL, GFP_NOFS);
1120         while (start < end) {
1121                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1122                 BUG_ON(!async_cow); /* -ENOMEM */
1123                 async_cow->inode = igrab(inode);
1124                 async_cow->root = root;
1125                 async_cow->locked_page = locked_page;
1126                 async_cow->start = start;
1127
1128                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1129                         cur_end = end;
1130                 else
1131                         cur_end = min(end, start + 512 * 1024 - 1);
1132
1133                 async_cow->end = cur_end;
1134                 INIT_LIST_HEAD(&async_cow->extents);
1135
1136                 async_cow->work.func = async_cow_start;
1137                 async_cow->work.ordered_func = async_cow_submit;
1138                 async_cow->work.ordered_free = async_cow_free;
1139                 async_cow->work.flags = 0;
1140
1141                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1142                         PAGE_CACHE_SHIFT;
1143                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1144
1145                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1146                                    &async_cow->work);
1147
1148                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1149                         wait_event(root->fs_info->async_submit_wait,
1150                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1151                             limit));
1152                 }
1153
1154                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1155                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1156                         wait_event(root->fs_info->async_submit_wait,
1157                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1158                            0));
1159                 }
1160
1161                 *nr_written += nr_pages;
1162                 start = cur_end + 1;
1163         }
1164         *page_started = 1;
1165         return 0;
1166 }
1167
1168 static noinline int csum_exist_in_range(struct btrfs_root *root,
1169                                         u64 bytenr, u64 num_bytes)
1170 {
1171         int ret;
1172         struct btrfs_ordered_sum *sums;
1173         LIST_HEAD(list);
1174
1175         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1176                                        bytenr + num_bytes - 1, &list, 0);
1177         if (ret == 0 && list_empty(&list))
1178                 return 0;
1179
1180         while (!list_empty(&list)) {
1181                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1182                 list_del(&sums->list);
1183                 kfree(sums);
1184         }
1185         return 1;
1186 }
1187
1188 /*
1189  * when nowcow writeback call back.  This checks for snapshots or COW copies
1190  * of the extents that exist in the file, and COWs the file as required.
1191  *
1192  * If no cow copies or snapshots exist, we write directly to the existing
1193  * blocks on disk
1194  */
1195 static noinline int run_delalloc_nocow(struct inode *inode,
1196                                        struct page *locked_page,
1197                               u64 start, u64 end, int *page_started, int force,
1198                               unsigned long *nr_written)
1199 {
1200         struct btrfs_root *root = BTRFS_I(inode)->root;
1201         struct btrfs_trans_handle *trans;
1202         struct extent_buffer *leaf;
1203         struct btrfs_path *path;
1204         struct btrfs_file_extent_item *fi;
1205         struct btrfs_key found_key;
1206         u64 cow_start;
1207         u64 cur_offset;
1208         u64 extent_end;
1209         u64 extent_offset;
1210         u64 disk_bytenr;
1211         u64 num_bytes;
1212         u64 disk_num_bytes;
1213         u64 ram_bytes;
1214         int extent_type;
1215         int ret, err;
1216         int type;
1217         int nocow;
1218         int check_prev = 1;
1219         bool nolock;
1220         u64 ino = btrfs_ino(inode);
1221
1222         path = btrfs_alloc_path();
1223         if (!path) {
1224                 extent_clear_unlock_delalloc(inode,
1225                              &BTRFS_I(inode)->io_tree,
1226                              start, end, locked_page,
1227                              EXTENT_CLEAR_UNLOCK_PAGE |
1228                              EXTENT_CLEAR_UNLOCK |
1229                              EXTENT_CLEAR_DELALLOC |
1230                              EXTENT_CLEAR_DIRTY |
1231                              EXTENT_SET_WRITEBACK |
1232                              EXTENT_END_WRITEBACK);
1233                 return -ENOMEM;
1234         }
1235
1236         nolock = btrfs_is_free_space_inode(inode);
1237
1238         if (nolock)
1239                 trans = btrfs_join_transaction_nolock(root);
1240         else
1241                 trans = btrfs_join_transaction(root);
1242
1243         if (IS_ERR(trans)) {
1244                 extent_clear_unlock_delalloc(inode,
1245                              &BTRFS_I(inode)->io_tree,
1246                              start, end, locked_page,
1247                              EXTENT_CLEAR_UNLOCK_PAGE |
1248                              EXTENT_CLEAR_UNLOCK |
1249                              EXTENT_CLEAR_DELALLOC |
1250                              EXTENT_CLEAR_DIRTY |
1251                              EXTENT_SET_WRITEBACK |
1252                              EXTENT_END_WRITEBACK);
1253                 btrfs_free_path(path);
1254                 return PTR_ERR(trans);
1255         }
1256
1257         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1258
1259         cow_start = (u64)-1;
1260         cur_offset = start;
1261         while (1) {
1262                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1263                                                cur_offset, 0);
1264                 if (ret < 0) {
1265                         btrfs_abort_transaction(trans, root, ret);
1266                         goto error;
1267                 }
1268                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1269                         leaf = path->nodes[0];
1270                         btrfs_item_key_to_cpu(leaf, &found_key,
1271                                               path->slots[0] - 1);
1272                         if (found_key.objectid == ino &&
1273                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1274                                 path->slots[0]--;
1275                 }
1276                 check_prev = 0;
1277 next_slot:
1278                 leaf = path->nodes[0];
1279                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1280                         ret = btrfs_next_leaf(root, path);
1281                         if (ret < 0) {
1282                                 btrfs_abort_transaction(trans, root, ret);
1283                                 goto error;
1284                         }
1285                         if (ret > 0)
1286                                 break;
1287                         leaf = path->nodes[0];
1288                 }
1289
1290                 nocow = 0;
1291                 disk_bytenr = 0;
1292                 num_bytes = 0;
1293                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1294
1295                 if (found_key.objectid > ino ||
1296                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1297                     found_key.offset > end)
1298                         break;
1299
1300                 if (found_key.offset > cur_offset) {
1301                         extent_end = found_key.offset;
1302                         extent_type = 0;
1303                         goto out_check;
1304                 }
1305
1306                 fi = btrfs_item_ptr(leaf, path->slots[0],
1307                                     struct btrfs_file_extent_item);
1308                 extent_type = btrfs_file_extent_type(leaf, fi);
1309
1310                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1311                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1312                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1313                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1314                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1315                         extent_end = found_key.offset +
1316                                 btrfs_file_extent_num_bytes(leaf, fi);
1317                         disk_num_bytes =
1318                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1319                         if (extent_end <= start) {
1320                                 path->slots[0]++;
1321                                 goto next_slot;
1322                         }
1323                         if (disk_bytenr == 0)
1324                                 goto out_check;
1325                         if (btrfs_file_extent_compression(leaf, fi) ||
1326                             btrfs_file_extent_encryption(leaf, fi) ||
1327                             btrfs_file_extent_other_encoding(leaf, fi))
1328                                 goto out_check;
1329                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1330                                 goto out_check;
1331                         if (btrfs_extent_readonly(root, disk_bytenr))
1332                                 goto out_check;
1333                         if (btrfs_cross_ref_exist(trans, root, ino,
1334                                                   found_key.offset -
1335                                                   extent_offset, disk_bytenr))
1336                                 goto out_check;
1337                         disk_bytenr += extent_offset;
1338                         disk_bytenr += cur_offset - found_key.offset;
1339                         num_bytes = min(end + 1, extent_end) - cur_offset;
1340                         /*
1341                          * force cow if csum exists in the range.
1342                          * this ensure that csum for a given extent are
1343                          * either valid or do not exist.
1344                          */
1345                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1346                                 goto out_check;
1347                         nocow = 1;
1348                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1349                         extent_end = found_key.offset +
1350                                 btrfs_file_extent_inline_len(leaf, fi);
1351                         extent_end = ALIGN(extent_end, root->sectorsize);
1352                 } else {
1353                         BUG_ON(1);
1354                 }
1355 out_check:
1356                 if (extent_end <= start) {
1357                         path->slots[0]++;
1358                         goto next_slot;
1359                 }
1360                 if (!nocow) {
1361                         if (cow_start == (u64)-1)
1362                                 cow_start = cur_offset;
1363                         cur_offset = extent_end;
1364                         if (cur_offset > end)
1365                                 break;
1366                         path->slots[0]++;
1367                         goto next_slot;
1368                 }
1369
1370                 btrfs_release_path(path);
1371                 if (cow_start != (u64)-1) {
1372                         ret = __cow_file_range(trans, inode, root, locked_page,
1373                                                cow_start, found_key.offset - 1,
1374                                                page_started, nr_written, 1);
1375                         if (ret) {
1376                                 btrfs_abort_transaction(trans, root, ret);
1377                                 goto error;
1378                         }
1379                         cow_start = (u64)-1;
1380                 }
1381
1382                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1383                         struct extent_map *em;
1384                         struct extent_map_tree *em_tree;
1385                         em_tree = &BTRFS_I(inode)->extent_tree;
1386                         em = alloc_extent_map();
1387                         BUG_ON(!em); /* -ENOMEM */
1388                         em->start = cur_offset;
1389                         em->orig_start = found_key.offset - extent_offset;
1390                         em->len = num_bytes;
1391                         em->block_len = num_bytes;
1392                         em->block_start = disk_bytenr;
1393                         em->orig_block_len = disk_num_bytes;
1394                         em->ram_bytes = ram_bytes;
1395                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1396                         em->mod_start = em->start;
1397                         em->mod_len = em->len;
1398                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1399                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1400                         em->generation = -1;
1401                         while (1) {
1402                                 write_lock(&em_tree->lock);
1403                                 ret = add_extent_mapping(em_tree, em, 1);
1404                                 write_unlock(&em_tree->lock);
1405                                 if (ret != -EEXIST) {
1406                                         free_extent_map(em);
1407                                         break;
1408                                 }
1409                                 btrfs_drop_extent_cache(inode, em->start,
1410                                                 em->start + em->len - 1, 0);
1411                         }
1412                         type = BTRFS_ORDERED_PREALLOC;
1413                 } else {
1414                         type = BTRFS_ORDERED_NOCOW;
1415                 }
1416
1417                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1418                                                num_bytes, num_bytes, type);
1419                 BUG_ON(ret); /* -ENOMEM */
1420
1421                 if (root->root_key.objectid ==
1422                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1423                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1424                                                       num_bytes);
1425                         if (ret) {
1426                                 btrfs_abort_transaction(trans, root, ret);
1427                                 goto error;
1428                         }
1429                 }
1430
1431                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1432                                 cur_offset, cur_offset + num_bytes - 1,
1433                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1434                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1435                                 EXTENT_SET_PRIVATE2);
1436                 cur_offset = extent_end;
1437                 if (cur_offset > end)
1438                         break;
1439         }
1440         btrfs_release_path(path);
1441
1442         if (cur_offset <= end && cow_start == (u64)-1) {
1443                 cow_start = cur_offset;
1444                 cur_offset = end;
1445         }
1446
1447         if (cow_start != (u64)-1) {
1448                 ret = __cow_file_range(trans, inode, root, locked_page,
1449                                        cow_start, end,
1450                                        page_started, nr_written, 1);
1451                 if (ret) {
1452                         btrfs_abort_transaction(trans, root, ret);
1453                         goto error;
1454                 }
1455         }
1456
1457 error:
1458         err = btrfs_end_transaction(trans, root);
1459         if (!ret)
1460                 ret = err;
1461
1462         if (ret && cur_offset < end)
1463                 extent_clear_unlock_delalloc(inode,
1464                              &BTRFS_I(inode)->io_tree,
1465                              cur_offset, end, locked_page,
1466                              EXTENT_CLEAR_UNLOCK_PAGE |
1467                              EXTENT_CLEAR_UNLOCK |
1468                              EXTENT_CLEAR_DELALLOC |
1469                              EXTENT_CLEAR_DIRTY |
1470                              EXTENT_SET_WRITEBACK |
1471                              EXTENT_END_WRITEBACK);
1472
1473         btrfs_free_path(path);
1474         return ret;
1475 }
1476
1477 /*
1478  * extent_io.c call back to do delayed allocation processing
1479  */
1480 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1481                               u64 start, u64 end, int *page_started,
1482                               unsigned long *nr_written)
1483 {
1484         int ret;
1485         struct btrfs_root *root = BTRFS_I(inode)->root;
1486
1487         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1488                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1489                                          page_started, 1, nr_written);
1490         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1491                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1492                                          page_started, 0, nr_written);
1493         } else if (!btrfs_test_opt(root, COMPRESS) &&
1494                    !(BTRFS_I(inode)->force_compress) &&
1495                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1496                 ret = cow_file_range(inode, locked_page, start, end,
1497                                       page_started, nr_written, 1);
1498         } else {
1499                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1500                         &BTRFS_I(inode)->runtime_flags);
1501                 ret = cow_file_range_async(inode, locked_page, start, end,
1502                                            page_started, nr_written);
1503         }
1504         return ret;
1505 }
1506
1507 static void btrfs_split_extent_hook(struct inode *inode,
1508                                     struct extent_state *orig, u64 split)
1509 {
1510         /* not delalloc, ignore it */
1511         if (!(orig->state & EXTENT_DELALLOC))
1512                 return;
1513
1514         spin_lock(&BTRFS_I(inode)->lock);
1515         BTRFS_I(inode)->outstanding_extents++;
1516         spin_unlock(&BTRFS_I(inode)->lock);
1517 }
1518
1519 /*
1520  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1521  * extents so we can keep track of new extents that are just merged onto old
1522  * extents, such as when we are doing sequential writes, so we can properly
1523  * account for the metadata space we'll need.
1524  */
1525 static void btrfs_merge_extent_hook(struct inode *inode,
1526                                     struct extent_state *new,
1527                                     struct extent_state *other)
1528 {
1529         /* not delalloc, ignore it */
1530         if (!(other->state & EXTENT_DELALLOC))
1531                 return;
1532
1533         spin_lock(&BTRFS_I(inode)->lock);
1534         BTRFS_I(inode)->outstanding_extents--;
1535         spin_unlock(&BTRFS_I(inode)->lock);
1536 }
1537
1538 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1539                                       struct inode *inode)
1540 {
1541         spin_lock(&root->delalloc_lock);
1542         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1543                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1544                               &root->delalloc_inodes);
1545                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1546                         &BTRFS_I(inode)->runtime_flags);
1547                 root->nr_delalloc_inodes++;
1548                 if (root->nr_delalloc_inodes == 1) {
1549                         spin_lock(&root->fs_info->delalloc_root_lock);
1550                         BUG_ON(!list_empty(&root->delalloc_root));
1551                         list_add_tail(&root->delalloc_root,
1552                                       &root->fs_info->delalloc_roots);
1553                         spin_unlock(&root->fs_info->delalloc_root_lock);
1554                 }
1555         }
1556         spin_unlock(&root->delalloc_lock);
1557 }
1558
1559 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1560                                      struct inode *inode)
1561 {
1562         spin_lock(&root->delalloc_lock);
1563         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1564                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1565                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1566                           &BTRFS_I(inode)->runtime_flags);
1567                 root->nr_delalloc_inodes--;
1568                 if (!root->nr_delalloc_inodes) {
1569                         spin_lock(&root->fs_info->delalloc_root_lock);
1570                         BUG_ON(list_empty(&root->delalloc_root));
1571                         list_del_init(&root->delalloc_root);
1572                         spin_unlock(&root->fs_info->delalloc_root_lock);
1573                 }
1574         }
1575         spin_unlock(&root->delalloc_lock);
1576 }
1577
1578 /*
1579  * extent_io.c set_bit_hook, used to track delayed allocation
1580  * bytes in this file, and to maintain the list of inodes that
1581  * have pending delalloc work to be done.
1582  */
1583 static void btrfs_set_bit_hook(struct inode *inode,
1584                                struct extent_state *state, unsigned long *bits)
1585 {
1586
1587         /*
1588          * set_bit and clear bit hooks normally require _irqsave/restore
1589          * but in this case, we are only testing for the DELALLOC
1590          * bit, which is only set or cleared with irqs on
1591          */
1592         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1593                 struct btrfs_root *root = BTRFS_I(inode)->root;
1594                 u64 len = state->end + 1 - state->start;
1595                 bool do_list = !btrfs_is_free_space_inode(inode);
1596
1597                 if (*bits & EXTENT_FIRST_DELALLOC) {
1598                         *bits &= ~EXTENT_FIRST_DELALLOC;
1599                 } else {
1600                         spin_lock(&BTRFS_I(inode)->lock);
1601                         BTRFS_I(inode)->outstanding_extents++;
1602                         spin_unlock(&BTRFS_I(inode)->lock);
1603                 }
1604
1605                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1606                                      root->fs_info->delalloc_batch);
1607                 spin_lock(&BTRFS_I(inode)->lock);
1608                 BTRFS_I(inode)->delalloc_bytes += len;
1609                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1610                                          &BTRFS_I(inode)->runtime_flags))
1611                         btrfs_add_delalloc_inodes(root, inode);
1612                 spin_unlock(&BTRFS_I(inode)->lock);
1613         }
1614 }
1615
1616 /*
1617  * extent_io.c clear_bit_hook, see set_bit_hook for why
1618  */
1619 static void btrfs_clear_bit_hook(struct inode *inode,
1620                                  struct extent_state *state,
1621                                  unsigned long *bits)
1622 {
1623         /*
1624          * set_bit and clear bit hooks normally require _irqsave/restore
1625          * but in this case, we are only testing for the DELALLOC
1626          * bit, which is only set or cleared with irqs on
1627          */
1628         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1629                 struct btrfs_root *root = BTRFS_I(inode)->root;
1630                 u64 len = state->end + 1 - state->start;
1631                 bool do_list = !btrfs_is_free_space_inode(inode);
1632
1633                 if (*bits & EXTENT_FIRST_DELALLOC) {
1634                         *bits &= ~EXTENT_FIRST_DELALLOC;
1635                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1636                         spin_lock(&BTRFS_I(inode)->lock);
1637                         BTRFS_I(inode)->outstanding_extents--;
1638                         spin_unlock(&BTRFS_I(inode)->lock);
1639                 }
1640
1641                 if (*bits & EXTENT_DO_ACCOUNTING)
1642                         btrfs_delalloc_release_metadata(inode, len);
1643
1644                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1645                     && do_list && !(state->state & EXTENT_NORESERVE))
1646                         btrfs_free_reserved_data_space(inode, len);
1647
1648                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1649                                      root->fs_info->delalloc_batch);
1650                 spin_lock(&BTRFS_I(inode)->lock);
1651                 BTRFS_I(inode)->delalloc_bytes -= len;
1652                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1653                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1654                              &BTRFS_I(inode)->runtime_flags))
1655                         btrfs_del_delalloc_inode(root, inode);
1656                 spin_unlock(&BTRFS_I(inode)->lock);
1657         }
1658 }
1659
1660 /*
1661  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1662  * we don't create bios that span stripes or chunks
1663  */
1664 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1665                          size_t size, struct bio *bio,
1666                          unsigned long bio_flags)
1667 {
1668         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1669         u64 logical = (u64)bio->bi_sector << 9;
1670         u64 length = 0;
1671         u64 map_length;
1672         int ret;
1673
1674         if (bio_flags & EXTENT_BIO_COMPRESSED)
1675                 return 0;
1676
1677         length = bio->bi_size;
1678         map_length = length;
1679         ret = btrfs_map_block(root->fs_info, rw, logical,
1680                               &map_length, NULL, 0);
1681         /* Will always return 0 with map_multi == NULL */
1682         BUG_ON(ret < 0);
1683         if (map_length < length + size)
1684                 return 1;
1685         return 0;
1686 }
1687
1688 /*
1689  * in order to insert checksums into the metadata in large chunks,
1690  * we wait until bio submission time.   All the pages in the bio are
1691  * checksummed and sums are attached onto the ordered extent record.
1692  *
1693  * At IO completion time the cums attached on the ordered extent record
1694  * are inserted into the btree
1695  */
1696 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1697                                     struct bio *bio, int mirror_num,
1698                                     unsigned long bio_flags,
1699                                     u64 bio_offset)
1700 {
1701         struct btrfs_root *root = BTRFS_I(inode)->root;
1702         int ret = 0;
1703
1704         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1705         BUG_ON(ret); /* -ENOMEM */
1706         return 0;
1707 }
1708
1709 /*
1710  * in order to insert checksums into the metadata in large chunks,
1711  * we wait until bio submission time.   All the pages in the bio are
1712  * checksummed and sums are attached onto the ordered extent record.
1713  *
1714  * At IO completion time the cums attached on the ordered extent record
1715  * are inserted into the btree
1716  */
1717 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1718                           int mirror_num, unsigned long bio_flags,
1719                           u64 bio_offset)
1720 {
1721         struct btrfs_root *root = BTRFS_I(inode)->root;
1722         int ret;
1723
1724         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1725         if (ret)
1726                 bio_endio(bio, ret);
1727         return ret;
1728 }
1729
1730 /*
1731  * extent_io.c submission hook. This does the right thing for csum calculation
1732  * on write, or reading the csums from the tree before a read
1733  */
1734 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1735                           int mirror_num, unsigned long bio_flags,
1736                           u64 bio_offset)
1737 {
1738         struct btrfs_root *root = BTRFS_I(inode)->root;
1739         int ret = 0;
1740         int skip_sum;
1741         int metadata = 0;
1742         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1743
1744         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1745
1746         if (btrfs_is_free_space_inode(inode))
1747                 metadata = 2;
1748
1749         if (!(rw & REQ_WRITE)) {
1750                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1751                 if (ret)
1752                         goto out;
1753
1754                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1755                         ret = btrfs_submit_compressed_read(inode, bio,
1756                                                            mirror_num,
1757                                                            bio_flags);
1758                         goto out;
1759                 } else if (!skip_sum) {
1760                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1761                         if (ret)
1762                                 goto out;
1763                 }
1764                 goto mapit;
1765         } else if (async && !skip_sum) {
1766                 /* csum items have already been cloned */
1767                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1768                         goto mapit;
1769                 /* we're doing a write, do the async checksumming */
1770                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1771                                    inode, rw, bio, mirror_num,
1772                                    bio_flags, bio_offset,
1773                                    __btrfs_submit_bio_start,
1774                                    __btrfs_submit_bio_done);
1775                 goto out;
1776         } else if (!skip_sum) {
1777                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1778                 if (ret)
1779                         goto out;
1780         }
1781
1782 mapit:
1783         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1784
1785 out:
1786         if (ret < 0)
1787                 bio_endio(bio, ret);
1788         return ret;
1789 }
1790
1791 /*
1792  * given a list of ordered sums record them in the inode.  This happens
1793  * at IO completion time based on sums calculated at bio submission time.
1794  */
1795 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1796                              struct inode *inode, u64 file_offset,
1797                              struct list_head *list)
1798 {
1799         struct btrfs_ordered_sum *sum;
1800
1801         list_for_each_entry(sum, list, list) {
1802                 trans->adding_csums = 1;
1803                 btrfs_csum_file_blocks(trans,
1804                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1805                 trans->adding_csums = 0;
1806         }
1807         return 0;
1808 }
1809
1810 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1811                               struct extent_state **cached_state)
1812 {
1813         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1814         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1815                                    cached_state, GFP_NOFS);
1816 }
1817
1818 /* see btrfs_writepage_start_hook for details on why this is required */
1819 struct btrfs_writepage_fixup {
1820         struct page *page;
1821         struct btrfs_work work;
1822 };
1823
1824 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1825 {
1826         struct btrfs_writepage_fixup *fixup;
1827         struct btrfs_ordered_extent *ordered;
1828         struct extent_state *cached_state = NULL;
1829         struct page *page;
1830         struct inode *inode;
1831         u64 page_start;
1832         u64 page_end;
1833         int ret;
1834
1835         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1836         page = fixup->page;
1837 again:
1838         lock_page(page);
1839         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1840                 ClearPageChecked(page);
1841                 goto out_page;
1842         }
1843
1844         inode = page->mapping->host;
1845         page_start = page_offset(page);
1846         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1847
1848         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1849                          &cached_state);
1850
1851         /* already ordered? We're done */
1852         if (PagePrivate2(page))
1853                 goto out;
1854
1855         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1856         if (ordered) {
1857                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1858                                      page_end, &cached_state, GFP_NOFS);
1859                 unlock_page(page);
1860                 btrfs_start_ordered_extent(inode, ordered, 1);
1861                 btrfs_put_ordered_extent(ordered);
1862                 goto again;
1863         }
1864
1865         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1866         if (ret) {
1867                 mapping_set_error(page->mapping, ret);
1868                 end_extent_writepage(page, ret, page_start, page_end);
1869                 ClearPageChecked(page);
1870                 goto out;
1871          }
1872
1873         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1874         ClearPageChecked(page);
1875         set_page_dirty(page);
1876 out:
1877         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1878                              &cached_state, GFP_NOFS);
1879 out_page:
1880         unlock_page(page);
1881         page_cache_release(page);
1882         kfree(fixup);
1883 }
1884
1885 /*
1886  * There are a few paths in the higher layers of the kernel that directly
1887  * set the page dirty bit without asking the filesystem if it is a
1888  * good idea.  This causes problems because we want to make sure COW
1889  * properly happens and the data=ordered rules are followed.
1890  *
1891  * In our case any range that doesn't have the ORDERED bit set
1892  * hasn't been properly setup for IO.  We kick off an async process
1893  * to fix it up.  The async helper will wait for ordered extents, set
1894  * the delalloc bit and make it safe to write the page.
1895  */
1896 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1897 {
1898         struct inode *inode = page->mapping->host;
1899         struct btrfs_writepage_fixup *fixup;
1900         struct btrfs_root *root = BTRFS_I(inode)->root;
1901
1902         /* this page is properly in the ordered list */
1903         if (TestClearPagePrivate2(page))
1904                 return 0;
1905
1906         if (PageChecked(page))
1907                 return -EAGAIN;
1908
1909         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1910         if (!fixup)
1911                 return -EAGAIN;
1912
1913         SetPageChecked(page);
1914         page_cache_get(page);
1915         fixup->work.func = btrfs_writepage_fixup_worker;
1916         fixup->page = page;
1917         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1918         return -EBUSY;
1919 }
1920
1921 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1922                                        struct inode *inode, u64 file_pos,
1923                                        u64 disk_bytenr, u64 disk_num_bytes,
1924                                        u64 num_bytes, u64 ram_bytes,
1925                                        u8 compression, u8 encryption,
1926                                        u16 other_encoding, int extent_type)
1927 {
1928         struct btrfs_root *root = BTRFS_I(inode)->root;
1929         struct btrfs_file_extent_item *fi;
1930         struct btrfs_path *path;
1931         struct extent_buffer *leaf;
1932         struct btrfs_key ins;
1933         int ret;
1934
1935         path = btrfs_alloc_path();
1936         if (!path)
1937                 return -ENOMEM;
1938
1939         path->leave_spinning = 1;
1940
1941         /*
1942          * we may be replacing one extent in the tree with another.
1943          * The new extent is pinned in the extent map, and we don't want
1944          * to drop it from the cache until it is completely in the btree.
1945          *
1946          * So, tell btrfs_drop_extents to leave this extent in the cache.
1947          * the caller is expected to unpin it and allow it to be merged
1948          * with the others.
1949          */
1950         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1951                                  file_pos + num_bytes, 0);
1952         if (ret)
1953                 goto out;
1954
1955         ins.objectid = btrfs_ino(inode);
1956         ins.offset = file_pos;
1957         ins.type = BTRFS_EXTENT_DATA_KEY;
1958         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1959         if (ret)
1960                 goto out;
1961         leaf = path->nodes[0];
1962         fi = btrfs_item_ptr(leaf, path->slots[0],
1963                             struct btrfs_file_extent_item);
1964         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1965         btrfs_set_file_extent_type(leaf, fi, extent_type);
1966         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1967         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1968         btrfs_set_file_extent_offset(leaf, fi, 0);
1969         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1970         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1971         btrfs_set_file_extent_compression(leaf, fi, compression);
1972         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1973         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1974
1975         btrfs_mark_buffer_dirty(leaf);
1976         btrfs_release_path(path);
1977
1978         inode_add_bytes(inode, num_bytes);
1979
1980         ins.objectid = disk_bytenr;
1981         ins.offset = disk_num_bytes;
1982         ins.type = BTRFS_EXTENT_ITEM_KEY;
1983         ret = btrfs_alloc_reserved_file_extent(trans, root,
1984                                         root->root_key.objectid,
1985                                         btrfs_ino(inode), file_pos, &ins);
1986 out:
1987         btrfs_free_path(path);
1988
1989         return ret;
1990 }
1991
1992 /* snapshot-aware defrag */
1993 struct sa_defrag_extent_backref {
1994         struct rb_node node;
1995         struct old_sa_defrag_extent *old;
1996         u64 root_id;
1997         u64 inum;
1998         u64 file_pos;
1999         u64 extent_offset;
2000         u64 num_bytes;
2001         u64 generation;
2002 };
2003
2004 struct old_sa_defrag_extent {
2005         struct list_head list;
2006         struct new_sa_defrag_extent *new;
2007
2008         u64 extent_offset;
2009         u64 bytenr;
2010         u64 offset;
2011         u64 len;
2012         int count;
2013 };
2014
2015 struct new_sa_defrag_extent {
2016         struct rb_root root;
2017         struct list_head head;
2018         struct btrfs_path *path;
2019         struct inode *inode;
2020         u64 file_pos;
2021         u64 len;
2022         u64 bytenr;
2023         u64 disk_len;
2024         u8 compress_type;
2025 };
2026
2027 static int backref_comp(struct sa_defrag_extent_backref *b1,
2028                         struct sa_defrag_extent_backref *b2)
2029 {
2030         if (b1->root_id < b2->root_id)
2031                 return -1;
2032         else if (b1->root_id > b2->root_id)
2033                 return 1;
2034
2035         if (b1->inum < b2->inum)
2036                 return -1;
2037         else if (b1->inum > b2->inum)
2038                 return 1;
2039
2040         if (b1->file_pos < b2->file_pos)
2041                 return -1;
2042         else if (b1->file_pos > b2->file_pos)
2043                 return 1;
2044
2045         /*
2046          * [------------------------------] ===> (a range of space)
2047          *     |<--->|   |<---->| =============> (fs/file tree A)
2048          * |<---------------------------->| ===> (fs/file tree B)
2049          *
2050          * A range of space can refer to two file extents in one tree while
2051          * refer to only one file extent in another tree.
2052          *
2053          * So we may process a disk offset more than one time(two extents in A)
2054          * and locate at the same extent(one extent in B), then insert two same
2055          * backrefs(both refer to the extent in B).
2056          */
2057         return 0;
2058 }
2059
2060 static void backref_insert(struct rb_root *root,
2061                            struct sa_defrag_extent_backref *backref)
2062 {
2063         struct rb_node **p = &root->rb_node;
2064         struct rb_node *parent = NULL;
2065         struct sa_defrag_extent_backref *entry;
2066         int ret;
2067
2068         while (*p) {
2069                 parent = *p;
2070                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2071
2072                 ret = backref_comp(backref, entry);
2073                 if (ret < 0)
2074                         p = &(*p)->rb_left;
2075                 else
2076                         p = &(*p)->rb_right;
2077         }
2078
2079         rb_link_node(&backref->node, parent, p);
2080         rb_insert_color(&backref->node, root);
2081 }
2082
2083 /*
2084  * Note the backref might has changed, and in this case we just return 0.
2085  */
2086 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2087                                        void *ctx)
2088 {
2089         struct btrfs_file_extent_item *extent;
2090         struct btrfs_fs_info *fs_info;
2091         struct old_sa_defrag_extent *old = ctx;
2092         struct new_sa_defrag_extent *new = old->new;
2093         struct btrfs_path *path = new->path;
2094         struct btrfs_key key;
2095         struct btrfs_root *root;
2096         struct sa_defrag_extent_backref *backref;
2097         struct extent_buffer *leaf;
2098         struct inode *inode = new->inode;
2099         int slot;
2100         int ret;
2101         u64 extent_offset;
2102         u64 num_bytes;
2103
2104         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2105             inum == btrfs_ino(inode))
2106                 return 0;
2107
2108         key.objectid = root_id;
2109         key.type = BTRFS_ROOT_ITEM_KEY;
2110         key.offset = (u64)-1;
2111
2112         fs_info = BTRFS_I(inode)->root->fs_info;
2113         root = btrfs_read_fs_root_no_name(fs_info, &key);
2114         if (IS_ERR(root)) {
2115                 if (PTR_ERR(root) == -ENOENT)
2116                         return 0;
2117                 WARN_ON(1);
2118                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2119                          inum, offset, root_id);
2120                 return PTR_ERR(root);
2121         }
2122
2123         key.objectid = inum;
2124         key.type = BTRFS_EXTENT_DATA_KEY;
2125         if (offset > (u64)-1 << 32)
2126                 key.offset = 0;
2127         else
2128                 key.offset = offset;
2129
2130         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2131         if (ret < 0) {
2132                 WARN_ON(1);
2133                 return ret;
2134         }
2135
2136         while (1) {
2137                 cond_resched();
2138
2139                 leaf = path->nodes[0];
2140                 slot = path->slots[0];
2141
2142                 if (slot >= btrfs_header_nritems(leaf)) {
2143                         ret = btrfs_next_leaf(root, path);
2144                         if (ret < 0) {
2145                                 goto out;
2146                         } else if (ret > 0) {
2147                                 ret = 0;
2148                                 goto out;
2149                         }
2150                         continue;
2151                 }
2152
2153                 path->slots[0]++;
2154
2155                 btrfs_item_key_to_cpu(leaf, &key, slot);
2156
2157                 if (key.objectid > inum)
2158                         goto out;
2159
2160                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2161                         continue;
2162
2163                 extent = btrfs_item_ptr(leaf, slot,
2164                                         struct btrfs_file_extent_item);
2165
2166                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2167                         continue;
2168
2169                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2170                 if (key.offset - extent_offset != offset)
2171                         continue;
2172
2173                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2174                 if (extent_offset >= old->extent_offset + old->offset +
2175                     old->len || extent_offset + num_bytes <=
2176                     old->extent_offset + old->offset)
2177                         continue;
2178
2179                 break;
2180         }
2181
2182         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2183         if (!backref) {
2184                 ret = -ENOENT;
2185                 goto out;
2186         }
2187
2188         backref->root_id = root_id;
2189         backref->inum = inum;
2190         backref->file_pos = offset + extent_offset;
2191         backref->num_bytes = num_bytes;
2192         backref->extent_offset = extent_offset;
2193         backref->generation = btrfs_file_extent_generation(leaf, extent);
2194         backref->old = old;
2195         backref_insert(&new->root, backref);
2196         old->count++;
2197 out:
2198         btrfs_release_path(path);
2199         WARN_ON(ret);
2200         return ret;
2201 }
2202
2203 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2204                                    struct new_sa_defrag_extent *new)
2205 {
2206         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2207         struct old_sa_defrag_extent *old, *tmp;
2208         int ret;
2209
2210         new->path = path;
2211
2212         list_for_each_entry_safe(old, tmp, &new->head, list) {
2213                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2214                                                   path, record_one_backref,
2215                                                   old);
2216                 BUG_ON(ret < 0 && ret != -ENOENT);
2217
2218                 /* no backref to be processed for this extent */
2219                 if (!old->count) {
2220                         list_del(&old->list);
2221                         kfree(old);
2222                 }
2223         }
2224
2225         if (list_empty(&new->head))
2226                 return false;
2227
2228         return true;
2229 }
2230
2231 static int relink_is_mergable(struct extent_buffer *leaf,
2232                               struct btrfs_file_extent_item *fi,
2233                               u64 disk_bytenr)
2234 {
2235         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2236                 return 0;
2237
2238         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2239                 return 0;
2240
2241         if (btrfs_file_extent_compression(leaf, fi) ||
2242             btrfs_file_extent_encryption(leaf, fi) ||
2243             btrfs_file_extent_other_encoding(leaf, fi))
2244                 return 0;
2245
2246         return 1;
2247 }
2248
2249 /*
2250  * Note the backref might has changed, and in this case we just return 0.
2251  */
2252 static noinline int relink_extent_backref(struct btrfs_path *path,
2253                                  struct sa_defrag_extent_backref *prev,
2254                                  struct sa_defrag_extent_backref *backref)
2255 {
2256         struct btrfs_file_extent_item *extent;
2257         struct btrfs_file_extent_item *item;
2258         struct btrfs_ordered_extent *ordered;
2259         struct btrfs_trans_handle *trans;
2260         struct btrfs_fs_info *fs_info;
2261         struct btrfs_root *root;
2262         struct btrfs_key key;
2263         struct extent_buffer *leaf;
2264         struct old_sa_defrag_extent *old = backref->old;
2265         struct new_sa_defrag_extent *new = old->new;
2266         struct inode *src_inode = new->inode;
2267         struct inode *inode;
2268         struct extent_state *cached = NULL;
2269         int ret = 0;
2270         u64 start;
2271         u64 len;
2272         u64 lock_start;
2273         u64 lock_end;
2274         bool merge = false;
2275         int index;
2276
2277         if (prev && prev->root_id == backref->root_id &&
2278             prev->inum == backref->inum &&
2279             prev->file_pos + prev->num_bytes == backref->file_pos)
2280                 merge = true;
2281
2282         /* step 1: get root */
2283         key.objectid = backref->root_id;
2284         key.type = BTRFS_ROOT_ITEM_KEY;
2285         key.offset = (u64)-1;
2286
2287         fs_info = BTRFS_I(src_inode)->root->fs_info;
2288         index = srcu_read_lock(&fs_info->subvol_srcu);
2289
2290         root = btrfs_read_fs_root_no_name(fs_info, &key);
2291         if (IS_ERR(root)) {
2292                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2293                 if (PTR_ERR(root) == -ENOENT)
2294                         return 0;
2295                 return PTR_ERR(root);
2296         }
2297
2298         /* step 2: get inode */
2299         key.objectid = backref->inum;
2300         key.type = BTRFS_INODE_ITEM_KEY;
2301         key.offset = 0;
2302
2303         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2304         if (IS_ERR(inode)) {
2305                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2306                 return 0;
2307         }
2308
2309         srcu_read_unlock(&fs_info->subvol_srcu, index);
2310
2311         /* step 3: relink backref */
2312         lock_start = backref->file_pos;
2313         lock_end = backref->file_pos + backref->num_bytes - 1;
2314         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2315                          0, &cached);
2316
2317         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2318         if (ordered) {
2319                 btrfs_put_ordered_extent(ordered);
2320                 goto out_unlock;
2321         }
2322
2323         trans = btrfs_join_transaction(root);
2324         if (IS_ERR(trans)) {
2325                 ret = PTR_ERR(trans);
2326                 goto out_unlock;
2327         }
2328
2329         key.objectid = backref->inum;
2330         key.type = BTRFS_EXTENT_DATA_KEY;
2331         key.offset = backref->file_pos;
2332
2333         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2334         if (ret < 0) {
2335                 goto out_free_path;
2336         } else if (ret > 0) {
2337                 ret = 0;
2338                 goto out_free_path;
2339         }
2340
2341         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2342                                 struct btrfs_file_extent_item);
2343
2344         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2345             backref->generation)
2346                 goto out_free_path;
2347
2348         btrfs_release_path(path);
2349
2350         start = backref->file_pos;
2351         if (backref->extent_offset < old->extent_offset + old->offset)
2352                 start += old->extent_offset + old->offset -
2353                          backref->extent_offset;
2354
2355         len = min(backref->extent_offset + backref->num_bytes,
2356                   old->extent_offset + old->offset + old->len);
2357         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2358
2359         ret = btrfs_drop_extents(trans, root, inode, start,
2360                                  start + len, 1);
2361         if (ret)
2362                 goto out_free_path;
2363 again:
2364         key.objectid = btrfs_ino(inode);
2365         key.type = BTRFS_EXTENT_DATA_KEY;
2366         key.offset = start;
2367
2368         path->leave_spinning = 1;
2369         if (merge) {
2370                 struct btrfs_file_extent_item *fi;
2371                 u64 extent_len;
2372                 struct btrfs_key found_key;
2373
2374                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2375                 if (ret < 0)
2376                         goto out_free_path;
2377
2378                 path->slots[0]--;
2379                 leaf = path->nodes[0];
2380                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2381
2382                 fi = btrfs_item_ptr(leaf, path->slots[0],
2383                                     struct btrfs_file_extent_item);
2384                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2385
2386                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2387                     extent_len + found_key.offset == start) {
2388                         btrfs_set_file_extent_num_bytes(leaf, fi,
2389                                                         extent_len + len);
2390                         btrfs_mark_buffer_dirty(leaf);
2391                         inode_add_bytes(inode, len);
2392
2393                         ret = 1;
2394                         goto out_free_path;
2395                 } else {
2396                         merge = false;
2397                         btrfs_release_path(path);
2398                         goto again;
2399                 }
2400         }
2401
2402         ret = btrfs_insert_empty_item(trans, root, path, &key,
2403                                         sizeof(*extent));
2404         if (ret) {
2405                 btrfs_abort_transaction(trans, root, ret);
2406                 goto out_free_path;
2407         }
2408
2409         leaf = path->nodes[0];
2410         item = btrfs_item_ptr(leaf, path->slots[0],
2411                                 struct btrfs_file_extent_item);
2412         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2413         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2414         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2415         btrfs_set_file_extent_num_bytes(leaf, item, len);
2416         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2417         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2418         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2419         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2420         btrfs_set_file_extent_encryption(leaf, item, 0);
2421         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2422
2423         btrfs_mark_buffer_dirty(leaf);
2424         inode_add_bytes(inode, len);
2425         btrfs_release_path(path);
2426
2427         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2428                         new->disk_len, 0,
2429                         backref->root_id, backref->inum,
2430                         new->file_pos, 0);      /* start - extent_offset */
2431         if (ret) {
2432                 btrfs_abort_transaction(trans, root, ret);
2433                 goto out_free_path;
2434         }
2435
2436         ret = 1;
2437 out_free_path:
2438         btrfs_release_path(path);
2439         path->leave_spinning = 0;
2440         btrfs_end_transaction(trans, root);
2441 out_unlock:
2442         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2443                              &cached, GFP_NOFS);
2444         iput(inode);
2445         return ret;
2446 }
2447
2448 static void relink_file_extents(struct new_sa_defrag_extent *new)
2449 {
2450         struct btrfs_path *path;
2451         struct old_sa_defrag_extent *old, *tmp;
2452         struct sa_defrag_extent_backref *backref;
2453         struct sa_defrag_extent_backref *prev = NULL;
2454         struct inode *inode;
2455         struct btrfs_root *root;
2456         struct rb_node *node;
2457         int ret;
2458
2459         inode = new->inode;
2460         root = BTRFS_I(inode)->root;
2461
2462         path = btrfs_alloc_path();
2463         if (!path)
2464                 return;
2465
2466         if (!record_extent_backrefs(path, new)) {
2467                 btrfs_free_path(path);
2468                 goto out;
2469         }
2470         btrfs_release_path(path);
2471
2472         while (1) {
2473                 node = rb_first(&new->root);
2474                 if (!node)
2475                         break;
2476                 rb_erase(node, &new->root);
2477
2478                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2479
2480                 ret = relink_extent_backref(path, prev, backref);
2481                 WARN_ON(ret < 0);
2482
2483                 kfree(prev);
2484
2485                 if (ret == 1)
2486                         prev = backref;
2487                 else
2488                         prev = NULL;
2489                 cond_resched();
2490         }
2491         kfree(prev);
2492
2493         btrfs_free_path(path);
2494
2495         list_for_each_entry_safe(old, tmp, &new->head, list) {
2496                 list_del(&old->list);
2497                 kfree(old);
2498         }
2499 out:
2500         atomic_dec(&root->fs_info->defrag_running);
2501         wake_up(&root->fs_info->transaction_wait);
2502
2503         kfree(new);
2504 }
2505
2506 static struct new_sa_defrag_extent *
2507 record_old_file_extents(struct inode *inode,
2508                         struct btrfs_ordered_extent *ordered)
2509 {
2510         struct btrfs_root *root = BTRFS_I(inode)->root;
2511         struct btrfs_path *path;
2512         struct btrfs_key key;
2513         struct old_sa_defrag_extent *old, *tmp;
2514         struct new_sa_defrag_extent *new;
2515         int ret;
2516
2517         new = kmalloc(sizeof(*new), GFP_NOFS);
2518         if (!new)
2519                 return NULL;
2520
2521         new->inode = inode;
2522         new->file_pos = ordered->file_offset;
2523         new->len = ordered->len;
2524         new->bytenr = ordered->start;
2525         new->disk_len = ordered->disk_len;
2526         new->compress_type = ordered->compress_type;
2527         new->root = RB_ROOT;
2528         INIT_LIST_HEAD(&new->head);
2529
2530         path = btrfs_alloc_path();
2531         if (!path)
2532                 goto out_kfree;
2533
2534         key.objectid = btrfs_ino(inode);
2535         key.type = BTRFS_EXTENT_DATA_KEY;
2536         key.offset = new->file_pos;
2537
2538         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2539         if (ret < 0)
2540                 goto out_free_path;
2541         if (ret > 0 && path->slots[0] > 0)
2542                 path->slots[0]--;
2543
2544         /* find out all the old extents for the file range */
2545         while (1) {
2546                 struct btrfs_file_extent_item *extent;
2547                 struct extent_buffer *l;
2548                 int slot;
2549                 u64 num_bytes;
2550                 u64 offset;
2551                 u64 end;
2552                 u64 disk_bytenr;
2553                 u64 extent_offset;
2554
2555                 l = path->nodes[0];
2556                 slot = path->slots[0];
2557
2558                 if (slot >= btrfs_header_nritems(l)) {
2559                         ret = btrfs_next_leaf(root, path);
2560                         if (ret < 0)
2561                                 goto out_free_list;
2562                         else if (ret > 0)
2563                                 break;
2564                         continue;
2565                 }
2566
2567                 btrfs_item_key_to_cpu(l, &key, slot);
2568
2569                 if (key.objectid != btrfs_ino(inode))
2570                         break;
2571                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2572                         break;
2573                 if (key.offset >= new->file_pos + new->len)
2574                         break;
2575
2576                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2577
2578                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2579                 if (key.offset + num_bytes < new->file_pos)
2580                         goto next;
2581
2582                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2583                 if (!disk_bytenr)
2584                         goto next;
2585
2586                 extent_offset = btrfs_file_extent_offset(l, extent);
2587
2588                 old = kmalloc(sizeof(*old), GFP_NOFS);
2589                 if (!old)
2590                         goto out_free_list;
2591
2592                 offset = max(new->file_pos, key.offset);
2593                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2594
2595                 old->bytenr = disk_bytenr;
2596                 old->extent_offset = extent_offset;
2597                 old->offset = offset - key.offset;
2598                 old->len = end - offset;
2599                 old->new = new;
2600                 old->count = 0;
2601                 list_add_tail(&old->list, &new->head);
2602 next:
2603                 path->slots[0]++;
2604                 cond_resched();
2605         }
2606
2607         btrfs_free_path(path);
2608         atomic_inc(&root->fs_info->defrag_running);
2609
2610         return new;
2611
2612 out_free_list:
2613         list_for_each_entry_safe(old, tmp, &new->head, list) {
2614                 list_del(&old->list);
2615                 kfree(old);
2616         }
2617 out_free_path:
2618         btrfs_free_path(path);
2619 out_kfree:
2620         kfree(new);
2621         return NULL;
2622 }
2623
2624 /*
2625  * helper function for btrfs_finish_ordered_io, this
2626  * just reads in some of the csum leaves to prime them into ram
2627  * before we start the transaction.  It limits the amount of btree
2628  * reads required while inside the transaction.
2629  */
2630 /* as ordered data IO finishes, this gets called so we can finish
2631  * an ordered extent if the range of bytes in the file it covers are
2632  * fully written.
2633  */
2634 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2635 {
2636         struct inode *inode = ordered_extent->inode;
2637         struct btrfs_root *root = BTRFS_I(inode)->root;
2638         struct btrfs_trans_handle *trans = NULL;
2639         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2640         struct extent_state *cached_state = NULL;
2641         struct new_sa_defrag_extent *new = NULL;
2642         int compress_type = 0;
2643         int ret;
2644         bool nolock;
2645
2646         nolock = btrfs_is_free_space_inode(inode);
2647
2648         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2649                 ret = -EIO;
2650                 goto out;
2651         }
2652
2653         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2654                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2655                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2656                 if (nolock)
2657                         trans = btrfs_join_transaction_nolock(root);
2658                 else
2659                         trans = btrfs_join_transaction(root);
2660                 if (IS_ERR(trans)) {
2661                         ret = PTR_ERR(trans);
2662                         trans = NULL;
2663                         goto out;
2664                 }
2665                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2666                 ret = btrfs_update_inode_fallback(trans, root, inode);
2667                 if (ret) /* -ENOMEM or corruption */
2668                         btrfs_abort_transaction(trans, root, ret);
2669                 goto out;
2670         }
2671
2672         lock_extent_bits(io_tree, ordered_extent->file_offset,
2673                          ordered_extent->file_offset + ordered_extent->len - 1,
2674                          0, &cached_state);
2675
2676         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2677                         ordered_extent->file_offset + ordered_extent->len - 1,
2678                         EXTENT_DEFRAG, 1, cached_state);
2679         if (ret) {
2680                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2681                 if (last_snapshot >= BTRFS_I(inode)->generation)
2682                         /* the inode is shared */
2683                         new = record_old_file_extents(inode, ordered_extent);
2684
2685                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2686                         ordered_extent->file_offset + ordered_extent->len - 1,
2687                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2688         }
2689
2690         if (nolock)
2691                 trans = btrfs_join_transaction_nolock(root);
2692         else
2693                 trans = btrfs_join_transaction(root);
2694         if (IS_ERR(trans)) {
2695                 ret = PTR_ERR(trans);
2696                 trans = NULL;
2697                 goto out_unlock;
2698         }
2699         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2700
2701         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2702                 compress_type = ordered_extent->compress_type;
2703         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2704                 BUG_ON(compress_type);
2705                 ret = btrfs_mark_extent_written(trans, inode,
2706                                                 ordered_extent->file_offset,
2707                                                 ordered_extent->file_offset +
2708                                                 ordered_extent->len);
2709         } else {
2710                 BUG_ON(root == root->fs_info->tree_root);
2711                 ret = insert_reserved_file_extent(trans, inode,
2712                                                 ordered_extent->file_offset,
2713                                                 ordered_extent->start,
2714                                                 ordered_extent->disk_len,
2715                                                 ordered_extent->len,
2716                                                 ordered_extent->len,
2717                                                 compress_type, 0, 0,
2718                                                 BTRFS_FILE_EXTENT_REG);
2719         }
2720         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2721                            ordered_extent->file_offset, ordered_extent->len,
2722                            trans->transid);
2723         if (ret < 0) {
2724                 btrfs_abort_transaction(trans, root, ret);
2725                 goto out_unlock;
2726         }
2727
2728         add_pending_csums(trans, inode, ordered_extent->file_offset,
2729                           &ordered_extent->list);
2730
2731         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2732         ret = btrfs_update_inode_fallback(trans, root, inode);
2733         if (ret) { /* -ENOMEM or corruption */
2734                 btrfs_abort_transaction(trans, root, ret);
2735                 goto out_unlock;
2736         }
2737         ret = 0;
2738 out_unlock:
2739         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2740                              ordered_extent->file_offset +
2741                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2742 out:
2743         if (root != root->fs_info->tree_root)
2744                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2745         if (trans)
2746                 btrfs_end_transaction(trans, root);
2747
2748         if (ret) {
2749                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2750                                       ordered_extent->file_offset +
2751                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2752
2753                 /*
2754                  * If the ordered extent had an IOERR or something else went
2755                  * wrong we need to return the space for this ordered extent
2756                  * back to the allocator.
2757                  */
2758                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2759                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2760                         btrfs_free_reserved_extent(root, ordered_extent->start,
2761                                                    ordered_extent->disk_len);
2762         }
2763
2764
2765         /*
2766          * This needs to be done to make sure anybody waiting knows we are done
2767          * updating everything for this ordered extent.
2768          */
2769         btrfs_remove_ordered_extent(inode, ordered_extent);
2770
2771         /* for snapshot-aware defrag */
2772         if (new)
2773                 relink_file_extents(new);
2774
2775         /* once for us */
2776         btrfs_put_ordered_extent(ordered_extent);
2777         /* once for the tree */
2778         btrfs_put_ordered_extent(ordered_extent);
2779
2780         return ret;
2781 }
2782
2783 static void finish_ordered_fn(struct btrfs_work *work)
2784 {
2785         struct btrfs_ordered_extent *ordered_extent;
2786         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2787         btrfs_finish_ordered_io(ordered_extent);
2788 }
2789
2790 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2791                                 struct extent_state *state, int uptodate)
2792 {
2793         struct inode *inode = page->mapping->host;
2794         struct btrfs_root *root = BTRFS_I(inode)->root;
2795         struct btrfs_ordered_extent *ordered_extent = NULL;
2796         struct btrfs_workers *workers;
2797
2798         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2799
2800         ClearPagePrivate2(page);
2801         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2802                                             end - start + 1, uptodate))
2803                 return 0;
2804
2805         ordered_extent->work.func = finish_ordered_fn;
2806         ordered_extent->work.flags = 0;
2807
2808         if (btrfs_is_free_space_inode(inode))
2809                 workers = &root->fs_info->endio_freespace_worker;
2810         else
2811                 workers = &root->fs_info->endio_write_workers;
2812         btrfs_queue_worker(workers, &ordered_extent->work);
2813
2814         return 0;
2815 }
2816
2817 /*
2818  * when reads are done, we need to check csums to verify the data is correct
2819  * if there's a match, we allow the bio to finish.  If not, the code in
2820  * extent_io.c will try to find good copies for us.
2821  */
2822 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2823                                struct extent_state *state, int mirror)
2824 {
2825         size_t offset = start - page_offset(page);
2826         struct inode *inode = page->mapping->host;
2827         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2828         char *kaddr;
2829         u64 private = ~(u32)0;
2830         int ret;
2831         struct btrfs_root *root = BTRFS_I(inode)->root;
2832         u32 csum = ~(u32)0;
2833         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2834                                       DEFAULT_RATELIMIT_BURST);
2835
2836         if (PageChecked(page)) {
2837                 ClearPageChecked(page);
2838                 goto good;
2839         }
2840
2841         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2842                 goto good;
2843
2844         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2845             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2846                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2847                                   GFP_NOFS);
2848                 return 0;
2849         }
2850
2851         if (state && state->start == start) {
2852                 private = state->private;
2853                 ret = 0;
2854         } else {
2855                 ret = get_state_private(io_tree, start, &private);
2856         }
2857         kaddr = kmap_atomic(page);
2858         if (ret)
2859                 goto zeroit;
2860
2861         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2862         btrfs_csum_final(csum, (char *)&csum);
2863         if (csum != private)
2864                 goto zeroit;
2865
2866         kunmap_atomic(kaddr);
2867 good:
2868         return 0;
2869
2870 zeroit:
2871         if (__ratelimit(&_rs))
2872                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2873                         (unsigned long long)btrfs_ino(page->mapping->host),
2874                         (unsigned long long)start, csum,
2875                         (unsigned long long)private);
2876         memset(kaddr + offset, 1, end - start + 1);
2877         flush_dcache_page(page);
2878         kunmap_atomic(kaddr);
2879         if (private == 0)
2880                 return 0;
2881         return -EIO;
2882 }
2883
2884 struct delayed_iput {
2885         struct list_head list;
2886         struct inode *inode;
2887 };
2888
2889 /* JDM: If this is fs-wide, why can't we add a pointer to
2890  * btrfs_inode instead and avoid the allocation? */
2891 void btrfs_add_delayed_iput(struct inode *inode)
2892 {
2893         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2894         struct delayed_iput *delayed;
2895
2896         if (atomic_add_unless(&inode->i_count, -1, 1))
2897                 return;
2898
2899         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2900         delayed->inode = inode;
2901
2902         spin_lock(&fs_info->delayed_iput_lock);
2903         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2904         spin_unlock(&fs_info->delayed_iput_lock);
2905 }
2906
2907 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2908 {
2909         LIST_HEAD(list);
2910         struct btrfs_fs_info *fs_info = root->fs_info;
2911         struct delayed_iput *delayed;
2912         int empty;
2913
2914         spin_lock(&fs_info->delayed_iput_lock);
2915         empty = list_empty(&fs_info->delayed_iputs);
2916         spin_unlock(&fs_info->delayed_iput_lock);
2917         if (empty)
2918                 return;
2919
2920         spin_lock(&fs_info->delayed_iput_lock);
2921         list_splice_init(&fs_info->delayed_iputs, &list);
2922         spin_unlock(&fs_info->delayed_iput_lock);
2923
2924         while (!list_empty(&list)) {
2925                 delayed = list_entry(list.next, struct delayed_iput, list);
2926                 list_del(&delayed->list);
2927                 iput(delayed->inode);
2928                 kfree(delayed);
2929         }
2930 }
2931
2932 /*
2933  * This is called in transaction commit time. If there are no orphan
2934  * files in the subvolume, it removes orphan item and frees block_rsv
2935  * structure.
2936  */
2937 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2938                               struct btrfs_root *root)
2939 {
2940         struct btrfs_block_rsv *block_rsv;
2941         int ret;
2942
2943         if (atomic_read(&root->orphan_inodes) ||
2944             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2945                 return;
2946
2947         spin_lock(&root->orphan_lock);
2948         if (atomic_read(&root->orphan_inodes)) {
2949                 spin_unlock(&root->orphan_lock);
2950                 return;
2951         }
2952
2953         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2954                 spin_unlock(&root->orphan_lock);
2955                 return;
2956         }
2957
2958         block_rsv = root->orphan_block_rsv;
2959         root->orphan_block_rsv = NULL;
2960         spin_unlock(&root->orphan_lock);
2961
2962         if (root->orphan_item_inserted &&
2963             btrfs_root_refs(&root->root_item) > 0) {
2964                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2965                                             root->root_key.objectid);
2966                 BUG_ON(ret);
2967                 root->orphan_item_inserted = 0;
2968         }
2969
2970         if (block_rsv) {
2971                 WARN_ON(block_rsv->size > 0);
2972                 btrfs_free_block_rsv(root, block_rsv);
2973         }
2974 }
2975
2976 /*
2977  * This creates an orphan entry for the given inode in case something goes
2978  * wrong in the middle of an unlink/truncate.
2979  *
2980  * NOTE: caller of this function should reserve 5 units of metadata for
2981  *       this function.
2982  */
2983 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2984 {
2985         struct btrfs_root *root = BTRFS_I(inode)->root;
2986         struct btrfs_block_rsv *block_rsv = NULL;
2987         int reserve = 0;
2988         int insert = 0;
2989         int ret;
2990
2991         if (!root->orphan_block_rsv) {
2992                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2993                 if (!block_rsv)
2994                         return -ENOMEM;
2995         }
2996
2997         spin_lock(&root->orphan_lock);
2998         if (!root->orphan_block_rsv) {
2999                 root->orphan_block_rsv = block_rsv;
3000         } else if (block_rsv) {
3001                 btrfs_free_block_rsv(root, block_rsv);
3002                 block_rsv = NULL;
3003         }
3004
3005         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3006                               &BTRFS_I(inode)->runtime_flags)) {
3007 #if 0
3008                 /*
3009                  * For proper ENOSPC handling, we should do orphan
3010                  * cleanup when mounting. But this introduces backward
3011                  * compatibility issue.
3012                  */
3013                 if (!xchg(&root->orphan_item_inserted, 1))
3014                         insert = 2;
3015                 else
3016                         insert = 1;
3017 #endif
3018                 insert = 1;
3019                 atomic_inc(&root->orphan_inodes);
3020         }
3021
3022         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3023                               &BTRFS_I(inode)->runtime_flags))
3024                 reserve = 1;
3025         spin_unlock(&root->orphan_lock);
3026
3027         /* grab metadata reservation from transaction handle */
3028         if (reserve) {
3029                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3030                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3031         }
3032
3033         /* insert an orphan item to track this unlinked/truncated file */
3034         if (insert >= 1) {
3035                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3036                 if (ret && ret != -EEXIST) {
3037                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3038                                   &BTRFS_I(inode)->runtime_flags);
3039                         btrfs_abort_transaction(trans, root, ret);
3040                         return ret;
3041                 }
3042                 ret = 0;
3043         }
3044
3045         /* insert an orphan item to track subvolume contains orphan files */
3046         if (insert >= 2) {
3047                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3048                                                root->root_key.objectid);
3049                 if (ret && ret != -EEXIST) {
3050                         btrfs_abort_transaction(trans, root, ret);
3051                         return ret;
3052                 }
3053         }
3054         return 0;
3055 }
3056
3057 /*
3058  * We have done the truncate/delete so we can go ahead and remove the orphan
3059  * item for this particular inode.
3060  */
3061 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3062                             struct inode *inode)
3063 {
3064         struct btrfs_root *root = BTRFS_I(inode)->root;
3065         int delete_item = 0;
3066         int release_rsv = 0;
3067         int ret = 0;
3068
3069         spin_lock(&root->orphan_lock);
3070         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3071                                &BTRFS_I(inode)->runtime_flags))
3072                 delete_item = 1;
3073
3074         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3075                                &BTRFS_I(inode)->runtime_flags))
3076                 release_rsv = 1;
3077         spin_unlock(&root->orphan_lock);
3078
3079         if (trans && delete_item) {
3080                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3081                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3082         }
3083
3084         if (release_rsv) {
3085                 btrfs_orphan_release_metadata(inode);
3086                 atomic_dec(&root->orphan_inodes);
3087         }
3088
3089         return 0;
3090 }
3091
3092 /*
3093  * this cleans up any orphans that may be left on the list from the last use
3094  * of this root.
3095  */
3096 int btrfs_orphan_cleanup(struct btrfs_root *root)
3097 {
3098         struct btrfs_path *path;
3099         struct extent_buffer *leaf;
3100         struct btrfs_key key, found_key;
3101         struct btrfs_trans_handle *trans;
3102         struct inode *inode;
3103         u64 last_objectid = 0;
3104         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3105
3106         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3107                 return 0;
3108
3109         path = btrfs_alloc_path();
3110         if (!path) {
3111                 ret = -ENOMEM;
3112                 goto out;
3113         }
3114         path->reada = -1;
3115
3116         key.objectid = BTRFS_ORPHAN_OBJECTID;
3117         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3118         key.offset = (u64)-1;
3119
3120         while (1) {
3121                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3122                 if (ret < 0)
3123                         goto out;
3124
3125                 /*
3126                  * if ret == 0 means we found what we were searching for, which
3127                  * is weird, but possible, so only screw with path if we didn't
3128                  * find the key and see if we have stuff that matches
3129                  */
3130                 if (ret > 0) {
3131                         ret = 0;
3132                         if (path->slots[0] == 0)
3133                                 break;
3134                         path->slots[0]--;
3135                 }
3136
3137                 /* pull out the item */
3138                 leaf = path->nodes[0];
3139                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3140
3141                 /* make sure the item matches what we want */
3142                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3143                         break;
3144                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3145                         break;
3146
3147                 /* release the path since we're done with it */
3148                 btrfs_release_path(path);
3149
3150                 /*
3151                  * this is where we are basically btrfs_lookup, without the
3152                  * crossing root thing.  we store the inode number in the
3153                  * offset of the orphan item.
3154                  */
3155
3156                 if (found_key.offset == last_objectid) {
3157                         btrfs_err(root->fs_info,
3158                                 "Error removing orphan entry, stopping orphan cleanup");
3159                         ret = -EINVAL;
3160                         goto out;
3161                 }
3162
3163                 last_objectid = found_key.offset;
3164
3165                 found_key.objectid = found_key.offset;
3166                 found_key.type = BTRFS_INODE_ITEM_KEY;
3167                 found_key.offset = 0;
3168                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3169                 ret = PTR_RET(inode);
3170                 if (ret && ret != -ESTALE)
3171                         goto out;
3172
3173                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3174                         struct btrfs_root *dead_root;
3175                         struct btrfs_fs_info *fs_info = root->fs_info;
3176                         int is_dead_root = 0;
3177
3178                         /*
3179                          * this is an orphan in the tree root. Currently these
3180                          * could come from 2 sources:
3181                          *  a) a snapshot deletion in progress
3182                          *  b) a free space cache inode
3183                          * We need to distinguish those two, as the snapshot
3184                          * orphan must not get deleted.
3185                          * find_dead_roots already ran before us, so if this
3186                          * is a snapshot deletion, we should find the root
3187                          * in the dead_roots list
3188                          */
3189                         spin_lock(&fs_info->trans_lock);
3190                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3191                                             root_list) {
3192                                 if (dead_root->root_key.objectid ==
3193                                     found_key.objectid) {
3194                                         is_dead_root = 1;
3195                                         break;
3196                                 }
3197                         }
3198                         spin_unlock(&fs_info->trans_lock);
3199                         if (is_dead_root) {
3200                                 /* prevent this orphan from being found again */
3201                                 key.offset = found_key.objectid - 1;
3202                                 continue;
3203                         }
3204                 }
3205                 /*
3206                  * Inode is already gone but the orphan item is still there,
3207                  * kill the orphan item.
3208                  */
3209                 if (ret == -ESTALE) {
3210                         trans = btrfs_start_transaction(root, 1);
3211                         if (IS_ERR(trans)) {
3212                                 ret = PTR_ERR(trans);
3213                                 goto out;
3214                         }
3215                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3216                                 found_key.objectid);
3217                         ret = btrfs_del_orphan_item(trans, root,
3218                                                     found_key.objectid);
3219                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3220                         btrfs_end_transaction(trans, root);
3221                         continue;
3222                 }
3223
3224                 /*
3225                  * add this inode to the orphan list so btrfs_orphan_del does
3226                  * the proper thing when we hit it
3227                  */
3228                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3229                         &BTRFS_I(inode)->runtime_flags);
3230                 atomic_inc(&root->orphan_inodes);
3231
3232                 /* if we have links, this was a truncate, lets do that */
3233                 if (inode->i_nlink) {
3234                         if (!S_ISREG(inode->i_mode)) {
3235                                 WARN_ON(1);
3236                                 iput(inode);
3237                                 continue;
3238                         }
3239                         nr_truncate++;
3240
3241                         /* 1 for the orphan item deletion. */
3242                         trans = btrfs_start_transaction(root, 1);
3243                         if (IS_ERR(trans)) {
3244                                 iput(inode);
3245                                 ret = PTR_ERR(trans);
3246                                 goto out;
3247                         }
3248                         ret = btrfs_orphan_add(trans, inode);
3249                         btrfs_end_transaction(trans, root);
3250                         if (ret) {
3251                                 iput(inode);
3252                                 goto out;
3253                         }
3254
3255                         ret = btrfs_truncate(inode);
3256                         if (ret)
3257                                 btrfs_orphan_del(NULL, inode);
3258                 } else {
3259                         nr_unlink++;
3260                 }
3261
3262                 /* this will do delete_inode and everything for us */
3263                 iput(inode);
3264                 if (ret)
3265                         goto out;
3266         }
3267         /* release the path since we're done with it */
3268         btrfs_release_path(path);
3269
3270         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3271
3272         if (root->orphan_block_rsv)
3273                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3274                                         (u64)-1);
3275
3276         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3277                 trans = btrfs_join_transaction(root);
3278                 if (!IS_ERR(trans))
3279                         btrfs_end_transaction(trans, root);
3280         }
3281
3282         if (nr_unlink)
3283                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3284         if (nr_truncate)
3285                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3286
3287 out:
3288         if (ret)
3289                 btrfs_crit(root->fs_info,
3290                         "could not do orphan cleanup %d", ret);
3291         btrfs_free_path(path);
3292         return ret;
3293 }
3294
3295 /*
3296  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3297  * don't find any xattrs, we know there can't be any acls.
3298  *
3299  * slot is the slot the inode is in, objectid is the objectid of the inode
3300  */
3301 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3302                                           int slot, u64 objectid)
3303 {
3304         u32 nritems = btrfs_header_nritems(leaf);
3305         struct btrfs_key found_key;
3306         static u64 xattr_access = 0;
3307         static u64 xattr_default = 0;
3308         int scanned = 0;
3309
3310         if (!xattr_access) {
3311                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3312                                         strlen(POSIX_ACL_XATTR_ACCESS));
3313                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3314                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3315         }
3316
3317         slot++;
3318         while (slot < nritems) {
3319                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3320
3321                 /* we found a different objectid, there must not be acls */
3322                 if (found_key.objectid != objectid)
3323                         return 0;
3324
3325                 /* we found an xattr, assume we've got an acl */
3326                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3327                         if (found_key.offset == xattr_access ||
3328                             found_key.offset == xattr_default)
3329                                 return 1;
3330                 }
3331
3332                 /*
3333                  * we found a key greater than an xattr key, there can't
3334                  * be any acls later on
3335                  */
3336                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3337                         return 0;
3338
3339                 slot++;
3340                 scanned++;
3341
3342                 /*
3343                  * it goes inode, inode backrefs, xattrs, extents,
3344                  * so if there are a ton of hard links to an inode there can
3345                  * be a lot of backrefs.  Don't waste time searching too hard,
3346                  * this is just an optimization
3347                  */
3348                 if (scanned >= 8)
3349                         break;
3350         }
3351         /* we hit the end of the leaf before we found an xattr or
3352          * something larger than an xattr.  We have to assume the inode
3353          * has acls
3354          */
3355         return 1;
3356 }
3357
3358 /*
3359  * read an inode from the btree into the in-memory inode
3360  */
3361 static void btrfs_read_locked_inode(struct inode *inode)
3362 {
3363         struct btrfs_path *path;
3364         struct extent_buffer *leaf;
3365         struct btrfs_inode_item *inode_item;
3366         struct btrfs_timespec *tspec;
3367         struct btrfs_root *root = BTRFS_I(inode)->root;
3368         struct btrfs_key location;
3369         int maybe_acls;
3370         u32 rdev;
3371         int ret;
3372         bool filled = false;
3373
3374         ret = btrfs_fill_inode(inode, &rdev);
3375         if (!ret)
3376                 filled = true;
3377
3378         path = btrfs_alloc_path();
3379         if (!path)
3380                 goto make_bad;
3381
3382         path->leave_spinning = 1;
3383         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3384
3385         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3386         if (ret)
3387                 goto make_bad;
3388
3389         leaf = path->nodes[0];
3390
3391         if (filled)
3392                 goto cache_acl;
3393
3394         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3395                                     struct btrfs_inode_item);
3396         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3397         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3398         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3399         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3400         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3401
3402         tspec = btrfs_inode_atime(inode_item);
3403         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3404         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3405
3406         tspec = btrfs_inode_mtime(inode_item);
3407         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3408         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3409
3410         tspec = btrfs_inode_ctime(inode_item);
3411         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3412         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3413
3414         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3415         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3416         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3417
3418         /*
3419          * If we were modified in the current generation and evicted from memory
3420          * and then re-read we need to do a full sync since we don't have any
3421          * idea about which extents were modified before we were evicted from
3422          * cache.
3423          */
3424         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3425                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3426                         &BTRFS_I(inode)->runtime_flags);
3427
3428         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3429         inode->i_generation = BTRFS_I(inode)->generation;
3430         inode->i_rdev = 0;
3431         rdev = btrfs_inode_rdev(leaf, inode_item);
3432
3433         BTRFS_I(inode)->index_cnt = (u64)-1;
3434         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3435 cache_acl:
3436         /*
3437          * try to precache a NULL acl entry for files that don't have
3438          * any xattrs or acls
3439          */
3440         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3441                                            btrfs_ino(inode));
3442         if (!maybe_acls)
3443                 cache_no_acl(inode);
3444
3445         btrfs_free_path(path);
3446
3447         switch (inode->i_mode & S_IFMT) {
3448         case S_IFREG:
3449                 inode->i_mapping->a_ops = &btrfs_aops;
3450                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3451                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3452                 inode->i_fop = &btrfs_file_operations;
3453                 inode->i_op = &btrfs_file_inode_operations;
3454                 break;
3455         case S_IFDIR:
3456                 inode->i_fop = &btrfs_dir_file_operations;
3457                 if (root == root->fs_info->tree_root)
3458                         inode->i_op = &btrfs_dir_ro_inode_operations;
3459                 else
3460                         inode->i_op = &btrfs_dir_inode_operations;
3461                 break;
3462         case S_IFLNK:
3463                 inode->i_op = &btrfs_symlink_inode_operations;
3464                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3465                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3466                 break;
3467         default:
3468                 inode->i_op = &btrfs_special_inode_operations;
3469                 init_special_inode(inode, inode->i_mode, rdev);
3470                 break;
3471         }
3472
3473         btrfs_update_iflags(inode);
3474         return;
3475
3476 make_bad:
3477         btrfs_free_path(path);
3478         make_bad_inode(inode);
3479 }
3480
3481 /*
3482  * given a leaf and an inode, copy the inode fields into the leaf
3483  */
3484 static void fill_inode_item(struct btrfs_trans_handle *trans,
3485                             struct extent_buffer *leaf,
3486                             struct btrfs_inode_item *item,
3487                             struct inode *inode)
3488 {
3489         struct btrfs_map_token token;
3490
3491         btrfs_init_map_token(&token);
3492
3493         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3494         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3495         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3496                                    &token);
3497         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3498         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3499
3500         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3501                                      inode->i_atime.tv_sec, &token);
3502         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3503                                       inode->i_atime.tv_nsec, &token);
3504
3505         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3506                                      inode->i_mtime.tv_sec, &token);
3507         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3508                                       inode->i_mtime.tv_nsec, &token);
3509
3510         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3511                                      inode->i_ctime.tv_sec, &token);
3512         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3513                                       inode->i_ctime.tv_nsec, &token);
3514
3515         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3516                                      &token);
3517         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3518                                          &token);
3519         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3520         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3521         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3522         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3523         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3524 }
3525
3526 /*
3527  * copy everything in the in-memory inode into the btree.
3528  */
3529 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3530                                 struct btrfs_root *root, struct inode *inode)
3531 {
3532         struct btrfs_inode_item *inode_item;
3533         struct btrfs_path *path;
3534         struct extent_buffer *leaf;
3535         int ret;
3536
3537         path = btrfs_alloc_path();
3538         if (!path)
3539                 return -ENOMEM;
3540
3541         path->leave_spinning = 1;
3542         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3543                                  1);
3544         if (ret) {
3545                 if (ret > 0)
3546                         ret = -ENOENT;
3547                 goto failed;
3548         }
3549
3550         btrfs_unlock_up_safe(path, 1);
3551         leaf = path->nodes[0];
3552         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3553                                     struct btrfs_inode_item);
3554
3555         fill_inode_item(trans, leaf, inode_item, inode);
3556         btrfs_mark_buffer_dirty(leaf);
3557         btrfs_set_inode_last_trans(trans, inode);
3558         ret = 0;
3559 failed:
3560         btrfs_free_path(path);
3561         return ret;
3562 }
3563
3564 /*
3565  * copy everything in the in-memory inode into the btree.
3566  */
3567 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3568                                 struct btrfs_root *root, struct inode *inode)
3569 {
3570         int ret;
3571
3572         /*
3573          * If the inode is a free space inode, we can deadlock during commit
3574          * if we put it into the delayed code.
3575          *
3576          * The data relocation inode should also be directly updated
3577          * without delay
3578          */
3579         if (!btrfs_is_free_space_inode(inode)
3580             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3581                 btrfs_update_root_times(trans, root);
3582
3583                 ret = btrfs_delayed_update_inode(trans, root, inode);
3584                 if (!ret)
3585                         btrfs_set_inode_last_trans(trans, inode);
3586                 return ret;
3587         }
3588
3589         return btrfs_update_inode_item(trans, root, inode);
3590 }
3591
3592 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3593                                          struct btrfs_root *root,
3594                                          struct inode *inode)
3595 {
3596         int ret;
3597
3598         ret = btrfs_update_inode(trans, root, inode);
3599         if (ret == -ENOSPC)
3600                 return btrfs_update_inode_item(trans, root, inode);
3601         return ret;
3602 }
3603
3604 /*
3605  * unlink helper that gets used here in inode.c and in the tree logging
3606  * recovery code.  It remove a link in a directory with a given name, and
3607  * also drops the back refs in the inode to the directory
3608  */
3609 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3610                                 struct btrfs_root *root,
3611                                 struct inode *dir, struct inode *inode,
3612                                 const char *name, int name_len)
3613 {
3614         struct btrfs_path *path;
3615         int ret = 0;
3616         struct extent_buffer *leaf;
3617         struct btrfs_dir_item *di;
3618         struct btrfs_key key;
3619         u64 index;
3620         u64 ino = btrfs_ino(inode);
3621         u64 dir_ino = btrfs_ino(dir);
3622
3623         path = btrfs_alloc_path();
3624         if (!path) {
3625                 ret = -ENOMEM;
3626                 goto out;
3627         }
3628
3629         path->leave_spinning = 1;
3630         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3631                                     name, name_len, -1);
3632         if (IS_ERR(di)) {
3633                 ret = PTR_ERR(di);
3634                 goto err;
3635         }
3636         if (!di) {
3637                 ret = -ENOENT;
3638                 goto err;
3639         }
3640         leaf = path->nodes[0];
3641         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3642         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3643         if (ret)
3644                 goto err;
3645         btrfs_release_path(path);
3646
3647         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3648                                   dir_ino, &index);
3649         if (ret) {
3650                 btrfs_info(root->fs_info,
3651                         "failed to delete reference to %.*s, inode %llu parent %llu",
3652                         name_len, name,
3653                         (unsigned long long)ino, (unsigned long long)dir_ino);
3654                 btrfs_abort_transaction(trans, root, ret);
3655                 goto err;
3656         }
3657
3658         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3659         if (ret) {
3660                 btrfs_abort_transaction(trans, root, ret);
3661                 goto err;
3662         }
3663
3664         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3665                                          inode, dir_ino);
3666         if (ret != 0 && ret != -ENOENT) {
3667                 btrfs_abort_transaction(trans, root, ret);
3668                 goto err;
3669         }
3670
3671         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3672                                            dir, index);
3673         if (ret == -ENOENT)
3674                 ret = 0;
3675         else if (ret)
3676                 btrfs_abort_transaction(trans, root, ret);
3677 err:
3678         btrfs_free_path(path);
3679         if (ret)
3680                 goto out;
3681
3682         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3683         inode_inc_iversion(inode);
3684         inode_inc_iversion(dir);
3685         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3686         ret = btrfs_update_inode(trans, root, dir);
3687 out:
3688         return ret;
3689 }
3690
3691 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3692                        struct btrfs_root *root,
3693                        struct inode *dir, struct inode *inode,
3694                        const char *name, int name_len)
3695 {
3696         int ret;
3697         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3698         if (!ret) {
3699                 btrfs_drop_nlink(inode);
3700                 ret = btrfs_update_inode(trans, root, inode);
3701         }
3702         return ret;
3703 }
3704
3705 /*
3706  * helper to start transaction for unlink and rmdir.
3707  *
3708  * unlink and rmdir are special in btrfs, they do not always free space, so
3709  * if we cannot make our reservations the normal way try and see if there is
3710  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3711  * allow the unlink to occur.
3712  */
3713 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3714 {
3715         struct btrfs_trans_handle *trans;
3716         struct btrfs_root *root = BTRFS_I(dir)->root;
3717         int ret;
3718
3719         /*
3720          * 1 for the possible orphan item
3721          * 1 for the dir item
3722          * 1 for the dir index
3723          * 1 for the inode ref
3724          * 1 for the inode
3725          */
3726         trans = btrfs_start_transaction(root, 5);
3727         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3728                 return trans;
3729
3730         if (PTR_ERR(trans) == -ENOSPC) {
3731                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3732
3733                 trans = btrfs_start_transaction(root, 0);
3734                 if (IS_ERR(trans))
3735                         return trans;
3736                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3737                                                &root->fs_info->trans_block_rsv,
3738                                                num_bytes, 5);
3739                 if (ret) {
3740                         btrfs_end_transaction(trans, root);
3741                         return ERR_PTR(ret);
3742                 }
3743                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3744                 trans->bytes_reserved = num_bytes;
3745         }
3746         return trans;
3747 }
3748
3749 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3750 {
3751         struct btrfs_root *root = BTRFS_I(dir)->root;
3752         struct btrfs_trans_handle *trans;
3753         struct inode *inode = dentry->d_inode;
3754         int ret;
3755
3756         trans = __unlink_start_trans(dir);
3757         if (IS_ERR(trans))
3758                 return PTR_ERR(trans);
3759
3760         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3761
3762         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3763                                  dentry->d_name.name, dentry->d_name.len);
3764         if (ret)
3765                 goto out;
3766
3767         if (inode->i_nlink == 0) {
3768                 ret = btrfs_orphan_add(trans, inode);
3769                 if (ret)
3770                         goto out;
3771         }
3772
3773 out:
3774         btrfs_end_transaction(trans, root);
3775         btrfs_btree_balance_dirty(root);
3776         return ret;
3777 }
3778
3779 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3780                         struct btrfs_root *root,
3781                         struct inode *dir, u64 objectid,
3782                         const char *name, int name_len)
3783 {
3784         struct btrfs_path *path;
3785         struct extent_buffer *leaf;
3786         struct btrfs_dir_item *di;
3787         struct btrfs_key key;
3788         u64 index;
3789         int ret;
3790         u64 dir_ino = btrfs_ino(dir);
3791
3792         path = btrfs_alloc_path();
3793         if (!path)
3794                 return -ENOMEM;
3795
3796         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3797                                    name, name_len, -1);
3798         if (IS_ERR_OR_NULL(di)) {
3799                 if (!di)
3800                         ret = -ENOENT;
3801                 else
3802                         ret = PTR_ERR(di);
3803                 goto out;
3804         }
3805
3806         leaf = path->nodes[0];
3807         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3808         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3809         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3810         if (ret) {
3811                 btrfs_abort_transaction(trans, root, ret);
3812                 goto out;
3813         }
3814         btrfs_release_path(path);
3815
3816         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3817                                  objectid, root->root_key.objectid,
3818                                  dir_ino, &index, name, name_len);
3819         if (ret < 0) {
3820                 if (ret != -ENOENT) {
3821                         btrfs_abort_transaction(trans, root, ret);
3822                         goto out;
3823                 }
3824                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3825                                                  name, name_len);
3826                 if (IS_ERR_OR_NULL(di)) {
3827                         if (!di)
3828                                 ret = -ENOENT;
3829                         else
3830                                 ret = PTR_ERR(di);
3831                         btrfs_abort_transaction(trans, root, ret);
3832                         goto out;
3833                 }
3834
3835                 leaf = path->nodes[0];
3836                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3837                 btrfs_release_path(path);
3838                 index = key.offset;
3839         }
3840         btrfs_release_path(path);
3841
3842         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3843         if (ret) {
3844                 btrfs_abort_transaction(trans, root, ret);
3845                 goto out;
3846         }
3847
3848         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3849         inode_inc_iversion(dir);
3850         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3851         ret = btrfs_update_inode_fallback(trans, root, dir);
3852         if (ret)
3853                 btrfs_abort_transaction(trans, root, ret);
3854 out:
3855         btrfs_free_path(path);
3856         return ret;
3857 }
3858
3859 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3860 {
3861         struct inode *inode = dentry->d_inode;
3862         int err = 0;
3863         struct btrfs_root *root = BTRFS_I(dir)->root;
3864         struct btrfs_trans_handle *trans;
3865
3866         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3867                 return -ENOTEMPTY;
3868         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3869                 return -EPERM;
3870
3871         trans = __unlink_start_trans(dir);
3872         if (IS_ERR(trans))
3873                 return PTR_ERR(trans);
3874
3875         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3876                 err = btrfs_unlink_subvol(trans, root, dir,
3877                                           BTRFS_I(inode)->location.objectid,
3878                                           dentry->d_name.name,
3879                                           dentry->d_name.len);
3880                 goto out;
3881         }
3882
3883         err = btrfs_orphan_add(trans, inode);
3884         if (err)
3885                 goto out;
3886
3887         /* now the directory is empty */
3888         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3889                                  dentry->d_name.name, dentry->d_name.len);
3890         if (!err)
3891                 btrfs_i_size_write(inode, 0);
3892 out:
3893         btrfs_end_transaction(trans, root);
3894         btrfs_btree_balance_dirty(root);
3895
3896         return err;
3897 }
3898
3899 /*
3900  * this can truncate away extent items, csum items and directory items.
3901  * It starts at a high offset and removes keys until it can't find
3902  * any higher than new_size
3903  *
3904  * csum items that cross the new i_size are truncated to the new size
3905  * as well.
3906  *
3907  * min_type is the minimum key type to truncate down to.  If set to 0, this
3908  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3909  */
3910 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3911                                struct btrfs_root *root,
3912                                struct inode *inode,
3913                                u64 new_size, u32 min_type)
3914 {
3915         struct btrfs_path *path;
3916         struct extent_buffer *leaf;
3917         struct btrfs_file_extent_item *fi;
3918         struct btrfs_key key;
3919         struct btrfs_key found_key;
3920         u64 extent_start = 0;
3921         u64 extent_num_bytes = 0;
3922         u64 extent_offset = 0;
3923         u64 item_end = 0;
3924         u32 found_type = (u8)-1;
3925         int found_extent;
3926         int del_item;
3927         int pending_del_nr = 0;
3928         int pending_del_slot = 0;
3929         int extent_type = -1;
3930         int ret;
3931         int err = 0;
3932         u64 ino = btrfs_ino(inode);
3933
3934         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3935
3936         path = btrfs_alloc_path();
3937         if (!path)
3938                 return -ENOMEM;
3939         path->reada = -1;
3940
3941         /*
3942          * We want to drop from the next block forward in case this new size is
3943          * not block aligned since we will be keeping the last block of the
3944          * extent just the way it is.
3945          */
3946         if (root->ref_cows || root == root->fs_info->tree_root)
3947                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3948                                         root->sectorsize), (u64)-1, 0);
3949
3950         /*
3951          * This function is also used to drop the items in the log tree before
3952          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3953          * it is used to drop the loged items. So we shouldn't kill the delayed
3954          * items.
3955          */
3956         if (min_type == 0 && root == BTRFS_I(inode)->root)
3957                 btrfs_kill_delayed_inode_items(inode);
3958
3959         key.objectid = ino;
3960         key.offset = (u64)-1;
3961         key.type = (u8)-1;
3962
3963 search_again:
3964         path->leave_spinning = 1;
3965         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3966         if (ret < 0) {
3967                 err = ret;
3968                 goto out;
3969         }
3970
3971         if (ret > 0) {
3972                 /* there are no items in the tree for us to truncate, we're
3973                  * done
3974                  */
3975                 if (path->slots[0] == 0)
3976                         goto out;
3977                 path->slots[0]--;
3978         }
3979
3980         while (1) {
3981                 fi = NULL;
3982                 leaf = path->nodes[0];
3983                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3984                 found_type = btrfs_key_type(&found_key);
3985
3986                 if (found_key.objectid != ino)
3987                         break;
3988
3989                 if (found_type < min_type)
3990                         break;
3991
3992                 item_end = found_key.offset;
3993                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3994                         fi = btrfs_item_ptr(leaf, path->slots[0],
3995                                             struct btrfs_file_extent_item);
3996                         extent_type = btrfs_file_extent_type(leaf, fi);
3997                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3998                                 item_end +=
3999                                     btrfs_file_extent_num_bytes(leaf, fi);
4000                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4001                                 item_end += btrfs_file_extent_inline_len(leaf,
4002                                                                          fi);
4003                         }
4004                         item_end--;
4005                 }
4006                 if (found_type > min_type) {
4007                         del_item = 1;
4008                 } else {
4009                         if (item_end < new_size)
4010                                 break;
4011                         if (found_key.offset >= new_size)
4012                                 del_item = 1;
4013                         else
4014                                 del_item = 0;
4015                 }
4016                 found_extent = 0;
4017                 /* FIXME, shrink the extent if the ref count is only 1 */
4018                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4019                         goto delete;
4020
4021                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4022                         u64 num_dec;
4023                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4024                         if (!del_item) {
4025                                 u64 orig_num_bytes =
4026                                         btrfs_file_extent_num_bytes(leaf, fi);
4027                                 extent_num_bytes = ALIGN(new_size -
4028                                                 found_key.offset,
4029                                                 root->sectorsize);
4030                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4031                                                          extent_num_bytes);
4032                                 num_dec = (orig_num_bytes -
4033                                            extent_num_bytes);
4034                                 if (root->ref_cows && extent_start != 0)
4035                                         inode_sub_bytes(inode, num_dec);
4036                                 btrfs_mark_buffer_dirty(leaf);
4037                         } else {
4038                                 extent_num_bytes =
4039                                         btrfs_file_extent_disk_num_bytes(leaf,
4040                                                                          fi);
4041                                 extent_offset = found_key.offset -
4042                                         btrfs_file_extent_offset(leaf, fi);
4043
4044                                 /* FIXME blocksize != 4096 */
4045                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4046                                 if (extent_start != 0) {
4047                                         found_extent = 1;
4048                                         if (root->ref_cows)
4049                                                 inode_sub_bytes(inode, num_dec);
4050                                 }
4051                         }
4052                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4053                         /*
4054                          * we can't truncate inline items that have had
4055                          * special encodings
4056                          */
4057                         if (!del_item &&
4058                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4059                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4060                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4061                                 u32 size = new_size - found_key.offset;
4062
4063                                 if (root->ref_cows) {
4064                                         inode_sub_bytes(inode, item_end + 1 -
4065                                                         new_size);
4066                                 }
4067                                 size =
4068                                     btrfs_file_extent_calc_inline_size(size);
4069                                 btrfs_truncate_item(root, path, size, 1);
4070                         } else if (root->ref_cows) {
4071                                 inode_sub_bytes(inode, item_end + 1 -
4072                                                 found_key.offset);
4073                         }
4074                 }
4075 delete:
4076                 if (del_item) {
4077                         if (!pending_del_nr) {
4078                                 /* no pending yet, add ourselves */
4079                                 pending_del_slot = path->slots[0];
4080                                 pending_del_nr = 1;
4081                         } else if (pending_del_nr &&
4082                                    path->slots[0] + 1 == pending_del_slot) {
4083                                 /* hop on the pending chunk */
4084                                 pending_del_nr++;
4085                                 pending_del_slot = path->slots[0];
4086                         } else {
4087                                 BUG();
4088                         }
4089                 } else {
4090                         break;
4091                 }
4092                 if (found_extent && (root->ref_cows ||
4093                                      root == root->fs_info->tree_root)) {
4094                         btrfs_set_path_blocking(path);
4095                         ret = btrfs_free_extent(trans, root, extent_start,
4096                                                 extent_num_bytes, 0,
4097                                                 btrfs_header_owner(leaf),
4098                                                 ino, extent_offset, 0);
4099                         BUG_ON(ret);
4100                 }
4101
4102                 if (found_type == BTRFS_INODE_ITEM_KEY)
4103                         break;
4104
4105                 if (path->slots[0] == 0 ||
4106                     path->slots[0] != pending_del_slot) {
4107                         if (pending_del_nr) {
4108                                 ret = btrfs_del_items(trans, root, path,
4109                                                 pending_del_slot,
4110                                                 pending_del_nr);
4111                                 if (ret) {
4112                                         btrfs_abort_transaction(trans,
4113                                                                 root, ret);
4114                                         goto error;
4115                                 }
4116                                 pending_del_nr = 0;
4117                         }
4118                         btrfs_release_path(path);
4119                         goto search_again;
4120                 } else {
4121                         path->slots[0]--;
4122                 }
4123         }
4124 out:
4125         if (pending_del_nr) {
4126                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4127                                       pending_del_nr);
4128                 if (ret)
4129                         btrfs_abort_transaction(trans, root, ret);
4130         }
4131 error:
4132         btrfs_free_path(path);
4133         return err;
4134 }
4135
4136 /*
4137  * btrfs_truncate_page - read, zero a chunk and write a page
4138  * @inode - inode that we're zeroing
4139  * @from - the offset to start zeroing
4140  * @len - the length to zero, 0 to zero the entire range respective to the
4141  *      offset
4142  * @front - zero up to the offset instead of from the offset on
4143  *
4144  * This will find the page for the "from" offset and cow the page and zero the
4145  * part we want to zero.  This is used with truncate and hole punching.
4146  */
4147 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4148                         int front)
4149 {
4150         struct address_space *mapping = inode->i_mapping;
4151         struct btrfs_root *root = BTRFS_I(inode)->root;
4152         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4153         struct btrfs_ordered_extent *ordered;
4154         struct extent_state *cached_state = NULL;
4155         char *kaddr;
4156         u32 blocksize = root->sectorsize;
4157         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4158         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4159         struct page *page;
4160         gfp_t mask = btrfs_alloc_write_mask(mapping);
4161         int ret = 0;
4162         u64 page_start;
4163         u64 page_end;
4164
4165         if ((offset & (blocksize - 1)) == 0 &&
4166             (!len || ((len & (blocksize - 1)) == 0)))
4167                 goto out;
4168         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4169         if (ret)
4170                 goto out;
4171
4172 again:
4173         page = find_or_create_page(mapping, index, mask);
4174         if (!page) {
4175                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4176                 ret = -ENOMEM;
4177                 goto out;
4178         }
4179
4180         page_start = page_offset(page);
4181         page_end = page_start + PAGE_CACHE_SIZE - 1;
4182
4183         if (!PageUptodate(page)) {
4184                 ret = btrfs_readpage(NULL, page);
4185                 lock_page(page);
4186                 if (page->mapping != mapping) {
4187                         unlock_page(page);
4188                         page_cache_release(page);
4189                         goto again;
4190                 }
4191                 if (!PageUptodate(page)) {
4192                         ret = -EIO;
4193                         goto out_unlock;
4194                 }
4195         }
4196         wait_on_page_writeback(page);
4197
4198         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4199         set_page_extent_mapped(page);
4200
4201         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4202         if (ordered) {
4203                 unlock_extent_cached(io_tree, page_start, page_end,
4204                                      &cached_state, GFP_NOFS);
4205                 unlock_page(page);
4206                 page_cache_release(page);
4207                 btrfs_start_ordered_extent(inode, ordered, 1);
4208                 btrfs_put_ordered_extent(ordered);
4209                 goto again;
4210         }
4211
4212         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4213                           EXTENT_DIRTY | EXTENT_DELALLOC |
4214                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4215                           0, 0, &cached_state, GFP_NOFS);
4216
4217         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4218                                         &cached_state);
4219         if (ret) {
4220                 unlock_extent_cached(io_tree, page_start, page_end,
4221                                      &cached_state, GFP_NOFS);
4222                 goto out_unlock;
4223         }
4224
4225         if (offset != PAGE_CACHE_SIZE) {
4226                 if (!len)
4227                         len = PAGE_CACHE_SIZE - offset;
4228                 kaddr = kmap(page);
4229                 if (front)
4230                         memset(kaddr, 0, offset);
4231                 else
4232                         memset(kaddr + offset, 0, len);
4233                 flush_dcache_page(page);
4234                 kunmap(page);
4235         }
4236         ClearPageChecked(page);
4237         set_page_dirty(page);
4238         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4239                              GFP_NOFS);
4240
4241 out_unlock:
4242         if (ret)
4243                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4244         unlock_page(page);
4245         page_cache_release(page);
4246 out:
4247         return ret;
4248 }
4249
4250 /*
4251  * This function puts in dummy file extents for the area we're creating a hole
4252  * for.  So if we are truncating this file to a larger size we need to insert
4253  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4254  * the range between oldsize and size
4255  */
4256 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4257 {
4258         struct btrfs_trans_handle *trans;
4259         struct btrfs_root *root = BTRFS_I(inode)->root;
4260         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4261         struct extent_map *em = NULL;
4262         struct extent_state *cached_state = NULL;
4263         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4264         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4265         u64 block_end = ALIGN(size, root->sectorsize);
4266         u64 last_byte;
4267         u64 cur_offset;
4268         u64 hole_size;
4269         int err = 0;
4270
4271         /*
4272          * If our size started in the middle of a page we need to zero out the
4273          * rest of the page before we expand the i_size, otherwise we could
4274          * expose stale data.
4275          */
4276         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4277         if (err)
4278                 return err;
4279
4280         if (size <= hole_start)
4281                 return 0;
4282
4283         while (1) {
4284                 struct btrfs_ordered_extent *ordered;
4285                 btrfs_wait_ordered_range(inode, hole_start,
4286                                          block_end - hole_start);
4287                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4288                                  &cached_state);
4289                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4290                 if (!ordered)
4291                         break;
4292                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4293                                      &cached_state, GFP_NOFS);
4294                 btrfs_put_ordered_extent(ordered);
4295         }
4296
4297         cur_offset = hole_start;
4298         while (1) {
4299                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4300                                 block_end - cur_offset, 0);
4301                 if (IS_ERR(em)) {
4302                         err = PTR_ERR(em);
4303                         em = NULL;
4304                         break;
4305                 }
4306                 last_byte = min(extent_map_end(em), block_end);
4307                 last_byte = ALIGN(last_byte , root->sectorsize);
4308                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4309                         struct extent_map *hole_em;
4310                         hole_size = last_byte - cur_offset;
4311
4312                         trans = btrfs_start_transaction(root, 3);
4313                         if (IS_ERR(trans)) {
4314                                 err = PTR_ERR(trans);
4315                                 break;
4316                         }
4317
4318                         err = btrfs_drop_extents(trans, root, inode,
4319                                                  cur_offset,
4320                                                  cur_offset + hole_size, 1);
4321                         if (err) {
4322                                 btrfs_abort_transaction(trans, root, err);
4323                                 btrfs_end_transaction(trans, root);
4324                                 break;
4325                         }
4326
4327                         err = btrfs_insert_file_extent(trans, root,
4328                                         btrfs_ino(inode), cur_offset, 0,
4329                                         0, hole_size, 0, hole_size,
4330                                         0, 0, 0);
4331                         if (err) {
4332                                 btrfs_abort_transaction(trans, root, err);
4333                                 btrfs_end_transaction(trans, root);
4334                                 break;
4335                         }
4336
4337                         btrfs_drop_extent_cache(inode, cur_offset,
4338                                                 cur_offset + hole_size - 1, 0);
4339                         hole_em = alloc_extent_map();
4340                         if (!hole_em) {
4341                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4342                                         &BTRFS_I(inode)->runtime_flags);
4343                                 goto next;
4344                         }
4345                         hole_em->start = cur_offset;
4346                         hole_em->len = hole_size;
4347                         hole_em->orig_start = cur_offset;
4348
4349                         hole_em->block_start = EXTENT_MAP_HOLE;
4350                         hole_em->block_len = 0;
4351                         hole_em->orig_block_len = 0;
4352                         hole_em->ram_bytes = hole_size;
4353                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4354                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4355                         hole_em->generation = trans->transid;
4356
4357                         while (1) {
4358                                 write_lock(&em_tree->lock);
4359                                 err = add_extent_mapping(em_tree, hole_em, 1);
4360                                 write_unlock(&em_tree->lock);
4361                                 if (err != -EEXIST)
4362                                         break;
4363                                 btrfs_drop_extent_cache(inode, cur_offset,
4364                                                         cur_offset +
4365                                                         hole_size - 1, 0);
4366                         }
4367                         free_extent_map(hole_em);
4368 next:
4369                         btrfs_update_inode(trans, root, inode);
4370                         btrfs_end_transaction(trans, root);
4371                 }
4372                 free_extent_map(em);
4373                 em = NULL;
4374                 cur_offset = last_byte;
4375                 if (cur_offset >= block_end)
4376                         break;
4377         }
4378
4379         free_extent_map(em);
4380         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4381                              GFP_NOFS);
4382         return err;
4383 }
4384
4385 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4386 {
4387         struct btrfs_root *root = BTRFS_I(inode)->root;
4388         struct btrfs_trans_handle *trans;
4389         loff_t oldsize = i_size_read(inode);
4390         loff_t newsize = attr->ia_size;
4391         int mask = attr->ia_valid;
4392         int ret;
4393
4394         if (newsize == oldsize)
4395                 return 0;
4396
4397         /*
4398          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4399          * special case where we need to update the times despite not having
4400          * these flags set.  For all other operations the VFS set these flags
4401          * explicitly if it wants a timestamp update.
4402          */
4403         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4404                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4405
4406         if (newsize > oldsize) {
4407                 truncate_pagecache(inode, oldsize, newsize);
4408                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4409                 if (ret)
4410                         return ret;
4411
4412                 trans = btrfs_start_transaction(root, 1);
4413                 if (IS_ERR(trans))
4414                         return PTR_ERR(trans);
4415
4416                 i_size_write(inode, newsize);
4417                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4418                 ret = btrfs_update_inode(trans, root, inode);
4419                 btrfs_end_transaction(trans, root);
4420         } else {
4421
4422                 /*
4423                  * We're truncating a file that used to have good data down to
4424                  * zero. Make sure it gets into the ordered flush list so that
4425                  * any new writes get down to disk quickly.
4426                  */
4427                 if (newsize == 0)
4428                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4429                                 &BTRFS_I(inode)->runtime_flags);
4430
4431                 /*
4432                  * 1 for the orphan item we're going to add
4433                  * 1 for the orphan item deletion.
4434                  */
4435                 trans = btrfs_start_transaction(root, 2);
4436                 if (IS_ERR(trans))
4437                         return PTR_ERR(trans);
4438
4439                 /*
4440                  * We need to do this in case we fail at _any_ point during the
4441                  * actual truncate.  Once we do the truncate_setsize we could
4442                  * invalidate pages which forces any outstanding ordered io to
4443                  * be instantly completed which will give us extents that need
4444                  * to be truncated.  If we fail to get an orphan inode down we
4445                  * could have left over extents that were never meant to live,
4446                  * so we need to garuntee from this point on that everything
4447                  * will be consistent.
4448                  */
4449                 ret = btrfs_orphan_add(trans, inode);
4450                 btrfs_end_transaction(trans, root);
4451                 if (ret)
4452                         return ret;
4453
4454                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4455                 truncate_setsize(inode, newsize);
4456
4457                 /* Disable nonlocked read DIO to avoid the end less truncate */
4458                 btrfs_inode_block_unlocked_dio(inode);
4459                 inode_dio_wait(inode);
4460                 btrfs_inode_resume_unlocked_dio(inode);
4461
4462                 ret = btrfs_truncate(inode);
4463                 if (ret && inode->i_nlink)
4464                         btrfs_orphan_del(NULL, inode);
4465         }
4466
4467         return ret;
4468 }
4469
4470 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4471 {
4472         struct inode *inode = dentry->d_inode;
4473         struct btrfs_root *root = BTRFS_I(inode)->root;
4474         int err;
4475
4476         if (btrfs_root_readonly(root))
4477                 return -EROFS;
4478
4479         err = inode_change_ok(inode, attr);
4480         if (err)
4481                 return err;
4482
4483         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4484                 err = btrfs_setsize(inode, attr);
4485                 if (err)
4486                         return err;
4487         }
4488
4489         if (attr->ia_valid) {
4490                 setattr_copy(inode, attr);
4491                 inode_inc_iversion(inode);
4492                 err = btrfs_dirty_inode(inode);
4493
4494                 if (!err && attr->ia_valid & ATTR_MODE)
4495                         err = btrfs_acl_chmod(inode);
4496         }
4497
4498         return err;
4499 }
4500
4501 void btrfs_evict_inode(struct inode *inode)
4502 {
4503         struct btrfs_trans_handle *trans;
4504         struct btrfs_root *root = BTRFS_I(inode)->root;
4505         struct btrfs_block_rsv *rsv, *global_rsv;
4506         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4507         int ret;
4508
4509         trace_btrfs_inode_evict(inode);
4510
4511         truncate_inode_pages(&inode->i_data, 0);
4512         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4513                                btrfs_is_free_space_inode(inode)))
4514                 goto no_delete;
4515
4516         if (is_bad_inode(inode)) {
4517                 btrfs_orphan_del(NULL, inode);
4518                 goto no_delete;
4519         }
4520         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4521         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4522
4523         if (root->fs_info->log_root_recovering) {
4524                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4525                                  &BTRFS_I(inode)->runtime_flags));
4526                 goto no_delete;
4527         }
4528
4529         if (inode->i_nlink > 0) {
4530                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4531                 goto no_delete;
4532         }
4533
4534         ret = btrfs_commit_inode_delayed_inode(inode);
4535         if (ret) {
4536                 btrfs_orphan_del(NULL, inode);
4537                 goto no_delete;
4538         }
4539
4540         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4541         if (!rsv) {
4542                 btrfs_orphan_del(NULL, inode);
4543                 goto no_delete;
4544         }
4545         rsv->size = min_size;
4546         rsv->failfast = 1;
4547         global_rsv = &root->fs_info->global_block_rsv;
4548
4549         btrfs_i_size_write(inode, 0);
4550
4551         /*
4552          * This is a bit simpler than btrfs_truncate since we've already
4553          * reserved our space for our orphan item in the unlink, so we just
4554          * need to reserve some slack space in case we add bytes and update
4555          * inode item when doing the truncate.
4556          */
4557         while (1) {
4558                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4559                                              BTRFS_RESERVE_FLUSH_LIMIT);
4560
4561                 /*
4562                  * Try and steal from the global reserve since we will
4563                  * likely not use this space anyway, we want to try as
4564                  * hard as possible to get this to work.
4565                  */
4566                 if (ret)
4567                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4568
4569                 if (ret) {
4570                         btrfs_warn(root->fs_info,
4571                                 "Could not get space for a delete, will truncate on mount %d",
4572                                 ret);
4573                         btrfs_orphan_del(NULL, inode);
4574                         btrfs_free_block_rsv(root, rsv);
4575                         goto no_delete;
4576                 }
4577
4578                 trans = btrfs_join_transaction(root);
4579                 if (IS_ERR(trans)) {
4580                         btrfs_orphan_del(NULL, inode);
4581                         btrfs_free_block_rsv(root, rsv);
4582                         goto no_delete;
4583                 }
4584
4585                 trans->block_rsv = rsv;
4586
4587                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4588                 if (ret != -ENOSPC)
4589                         break;
4590
4591                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4592                 btrfs_end_transaction(trans, root);
4593                 trans = NULL;
4594                 btrfs_btree_balance_dirty(root);
4595         }
4596
4597         btrfs_free_block_rsv(root, rsv);
4598
4599         if (ret == 0) {
4600                 trans->block_rsv = root->orphan_block_rsv;
4601                 ret = btrfs_orphan_del(trans, inode);
4602                 BUG_ON(ret);
4603         }
4604
4605         trans->block_rsv = &root->fs_info->trans_block_rsv;
4606         if (!(root == root->fs_info->tree_root ||
4607               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4608                 btrfs_return_ino(root, btrfs_ino(inode));
4609
4610         btrfs_end_transaction(trans, root);
4611         btrfs_btree_balance_dirty(root);
4612 no_delete:
4613         btrfs_remove_delayed_node(inode);
4614         clear_inode(inode);
4615         return;
4616 }
4617
4618 /*
4619  * this returns the key found in the dir entry in the location pointer.
4620  * If no dir entries were found, location->objectid is 0.
4621  */
4622 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4623                                struct btrfs_key *location)
4624 {
4625         const char *name = dentry->d_name.name;
4626         int namelen = dentry->d_name.len;
4627         struct btrfs_dir_item *di;
4628         struct btrfs_path *path;
4629         struct btrfs_root *root = BTRFS_I(dir)->root;
4630         int ret = 0;
4631
4632         path = btrfs_alloc_path();
4633         if (!path)
4634                 return -ENOMEM;
4635
4636         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4637                                     namelen, 0);
4638         if (IS_ERR(di))
4639                 ret = PTR_ERR(di);
4640
4641         if (IS_ERR_OR_NULL(di))
4642                 goto out_err;
4643
4644         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4645 out:
4646         btrfs_free_path(path);
4647         return ret;
4648 out_err:
4649         location->objectid = 0;
4650         goto out;
4651 }
4652
4653 /*
4654  * when we hit a tree root in a directory, the btrfs part of the inode
4655  * needs to be changed to reflect the root directory of the tree root.  This
4656  * is kind of like crossing a mount point.
4657  */
4658 static int fixup_tree_root_location(struct btrfs_root *root,
4659                                     struct inode *dir,
4660                                     struct dentry *dentry,
4661                                     struct btrfs_key *location,
4662                                     struct btrfs_root **sub_root)
4663 {
4664         struct btrfs_path *path;
4665         struct btrfs_root *new_root;
4666         struct btrfs_root_ref *ref;
4667         struct extent_buffer *leaf;
4668         int ret;
4669         int err = 0;
4670
4671         path = btrfs_alloc_path();
4672         if (!path) {
4673                 err = -ENOMEM;
4674                 goto out;
4675         }
4676
4677         err = -ENOENT;
4678         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4679                                   BTRFS_I(dir)->root->root_key.objectid,
4680                                   location->objectid);
4681         if (ret) {
4682                 if (ret < 0)
4683                         err = ret;
4684                 goto out;
4685         }
4686
4687         leaf = path->nodes[0];
4688         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4689         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4690             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4691                 goto out;
4692
4693         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4694                                    (unsigned long)(ref + 1),
4695                                    dentry->d_name.len);
4696         if (ret)
4697                 goto out;
4698
4699         btrfs_release_path(path);
4700
4701         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4702         if (IS_ERR(new_root)) {
4703                 err = PTR_ERR(new_root);
4704                 goto out;
4705         }
4706
4707         *sub_root = new_root;
4708         location->objectid = btrfs_root_dirid(&new_root->root_item);
4709         location->type = BTRFS_INODE_ITEM_KEY;
4710         location->offset = 0;
4711         err = 0;
4712 out:
4713         btrfs_free_path(path);
4714         return err;
4715 }
4716
4717 static void inode_tree_add(struct inode *inode)
4718 {
4719         struct btrfs_root *root = BTRFS_I(inode)->root;
4720         struct btrfs_inode *entry;
4721         struct rb_node **p;
4722         struct rb_node *parent;
4723         u64 ino = btrfs_ino(inode);
4724
4725         if (inode_unhashed(inode))
4726                 return;
4727 again:
4728         parent = NULL;
4729         spin_lock(&root->inode_lock);
4730         p = &root->inode_tree.rb_node;
4731         while (*p) {
4732                 parent = *p;
4733                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4734
4735                 if (ino < btrfs_ino(&entry->vfs_inode))
4736                         p = &parent->rb_left;
4737                 else if (ino > btrfs_ino(&entry->vfs_inode))
4738                         p = &parent->rb_right;
4739                 else {
4740                         WARN_ON(!(entry->vfs_inode.i_state &
4741                                   (I_WILL_FREE | I_FREEING)));
4742                         rb_erase(parent, &root->inode_tree);
4743                         RB_CLEAR_NODE(parent);
4744                         spin_unlock(&root->inode_lock);
4745                         goto again;
4746                 }
4747         }
4748         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4749         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4750         spin_unlock(&root->inode_lock);
4751 }
4752
4753 static void inode_tree_del(struct inode *inode)
4754 {
4755         struct btrfs_root *root = BTRFS_I(inode)->root;
4756         int empty = 0;
4757
4758         spin_lock(&root->inode_lock);
4759         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4760                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4761                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4762                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4763         }
4764         spin_unlock(&root->inode_lock);
4765
4766         /*
4767          * Free space cache has inodes in the tree root, but the tree root has a
4768          * root_refs of 0, so this could end up dropping the tree root as a
4769          * snapshot, so we need the extra !root->fs_info->tree_root check to
4770          * make sure we don't drop it.
4771          */
4772         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4773             root != root->fs_info->tree_root) {
4774                 synchronize_srcu(&root->fs_info->subvol_srcu);
4775                 spin_lock(&root->inode_lock);
4776                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4777                 spin_unlock(&root->inode_lock);
4778                 if (empty)
4779                         btrfs_add_dead_root(root);
4780         }
4781 }
4782
4783 void btrfs_invalidate_inodes(struct btrfs_root *root)
4784 {
4785         struct rb_node *node;
4786         struct rb_node *prev;
4787         struct btrfs_inode *entry;
4788         struct inode *inode;
4789         u64 objectid = 0;
4790
4791         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4792
4793         spin_lock(&root->inode_lock);
4794 again:
4795         node = root->inode_tree.rb_node;
4796         prev = NULL;
4797         while (node) {
4798                 prev = node;
4799                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4800
4801                 if (objectid < btrfs_ino(&entry->vfs_inode))
4802                         node = node->rb_left;
4803                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4804                         node = node->rb_right;
4805                 else
4806                         break;
4807         }
4808         if (!node) {
4809                 while (prev) {
4810                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4811                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4812                                 node = prev;
4813                                 break;
4814                         }
4815                         prev = rb_next(prev);
4816                 }
4817         }
4818         while (node) {
4819                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4820                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4821                 inode = igrab(&entry->vfs_inode);
4822                 if (inode) {
4823                         spin_unlock(&root->inode_lock);
4824                         if (atomic_read(&inode->i_count) > 1)
4825                                 d_prune_aliases(inode);
4826                         /*
4827                          * btrfs_drop_inode will have it removed from
4828                          * the inode cache when its usage count
4829                          * hits zero.
4830                          */
4831                         iput(inode);
4832                         cond_resched();
4833                         spin_lock(&root->inode_lock);
4834                         goto again;
4835                 }
4836
4837                 if (cond_resched_lock(&root->inode_lock))
4838                         goto again;
4839
4840                 node = rb_next(node);
4841         }
4842         spin_unlock(&root->inode_lock);
4843 }
4844
4845 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4846 {
4847         struct btrfs_iget_args *args = p;
4848         inode->i_ino = args->ino;
4849         BTRFS_I(inode)->root = args->root;
4850         return 0;
4851 }
4852
4853 static int btrfs_find_actor(struct inode *inode, void *opaque)
4854 {
4855         struct btrfs_iget_args *args = opaque;
4856         return args->ino == btrfs_ino(inode) &&
4857                 args->root == BTRFS_I(inode)->root;
4858 }
4859
4860 static struct inode *btrfs_iget_locked(struct super_block *s,
4861                                        u64 objectid,
4862                                        struct btrfs_root *root)
4863 {
4864         struct inode *inode;
4865         struct btrfs_iget_args args;
4866         args.ino = objectid;
4867         args.root = root;
4868
4869         inode = iget5_locked(s, objectid, btrfs_find_actor,
4870                              btrfs_init_locked_inode,
4871                              (void *)&args);
4872         return inode;
4873 }
4874
4875 /* Get an inode object given its location and corresponding root.
4876  * Returns in *is_new if the inode was read from disk
4877  */
4878 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4879                          struct btrfs_root *root, int *new)
4880 {
4881         struct inode *inode;
4882
4883         inode = btrfs_iget_locked(s, location->objectid, root);
4884         if (!inode)
4885                 return ERR_PTR(-ENOMEM);
4886
4887         if (inode->i_state & I_NEW) {
4888                 BTRFS_I(inode)->root = root;
4889                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4890                 btrfs_read_locked_inode(inode);
4891                 if (!is_bad_inode(inode)) {
4892                         inode_tree_add(inode);
4893                         unlock_new_inode(inode);
4894                         if (new)
4895                                 *new = 1;
4896                 } else {
4897                         unlock_new_inode(inode);
4898                         iput(inode);
4899                         inode = ERR_PTR(-ESTALE);
4900                 }
4901         }
4902
4903         return inode;
4904 }
4905
4906 static struct inode *new_simple_dir(struct super_block *s,
4907                                     struct btrfs_key *key,
4908                                     struct btrfs_root *root)
4909 {
4910         struct inode *inode = new_inode(s);
4911
4912         if (!inode)
4913                 return ERR_PTR(-ENOMEM);
4914
4915         BTRFS_I(inode)->root = root;
4916         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4917         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4918
4919         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4920         inode->i_op = &btrfs_dir_ro_inode_operations;
4921         inode->i_fop = &simple_dir_operations;
4922         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4923         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4924
4925         return inode;
4926 }
4927
4928 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4929 {
4930         struct inode *inode;
4931         struct btrfs_root *root = BTRFS_I(dir)->root;
4932         struct btrfs_root *sub_root = root;
4933         struct btrfs_key location;
4934         int index;
4935         int ret = 0;
4936
4937         if (dentry->d_name.len > BTRFS_NAME_LEN)
4938                 return ERR_PTR(-ENAMETOOLONG);
4939
4940         ret = btrfs_inode_by_name(dir, dentry, &location);
4941         if (ret < 0)
4942                 return ERR_PTR(ret);
4943
4944         if (location.objectid == 0)
4945                 return NULL;
4946
4947         if (location.type == BTRFS_INODE_ITEM_KEY) {
4948                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4949                 return inode;
4950         }
4951
4952         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4953
4954         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4955         ret = fixup_tree_root_location(root, dir, dentry,
4956                                        &location, &sub_root);
4957         if (ret < 0) {
4958                 if (ret != -ENOENT)
4959                         inode = ERR_PTR(ret);
4960                 else
4961                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4962         } else {
4963                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4964         }
4965         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4966
4967         if (!IS_ERR(inode) && root != sub_root) {
4968                 down_read(&root->fs_info->cleanup_work_sem);
4969                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4970                         ret = btrfs_orphan_cleanup(sub_root);
4971                 up_read(&root->fs_info->cleanup_work_sem);
4972                 if (ret) {
4973                         iput(inode);
4974                         inode = ERR_PTR(ret);
4975                 }
4976         }
4977
4978         return inode;
4979 }
4980
4981 static int btrfs_dentry_delete(const struct dentry *dentry)
4982 {
4983         struct btrfs_root *root;
4984         struct inode *inode = dentry->d_inode;
4985
4986         if (!inode && !IS_ROOT(dentry))
4987                 inode = dentry->d_parent->d_inode;
4988
4989         if (inode) {
4990                 root = BTRFS_I(inode)->root;
4991                 if (btrfs_root_refs(&root->root_item) == 0)
4992                         return 1;
4993
4994                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4995                         return 1;
4996         }
4997         return 0;
4998 }
4999
5000 static void btrfs_dentry_release(struct dentry *dentry)
5001 {
5002         if (dentry->d_fsdata)
5003                 kfree(dentry->d_fsdata);
5004 }
5005
5006 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5007                                    unsigned int flags)
5008 {
5009         struct dentry *ret;
5010
5011         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5012         return ret;
5013 }
5014
5015 unsigned char btrfs_filetype_table[] = {
5016         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5017 };
5018
5019 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5020 {
5021         struct inode *inode = file_inode(file);
5022         struct btrfs_root *root = BTRFS_I(inode)->root;
5023         struct btrfs_item *item;
5024         struct btrfs_dir_item *di;
5025         struct btrfs_key key;
5026         struct btrfs_key found_key;
5027         struct btrfs_path *path;
5028         struct list_head ins_list;
5029         struct list_head del_list;
5030         int ret;
5031         struct extent_buffer *leaf;
5032         int slot;
5033         unsigned char d_type;
5034         int over = 0;
5035         u32 di_cur;
5036         u32 di_total;
5037         u32 di_len;
5038         int key_type = BTRFS_DIR_INDEX_KEY;
5039         char tmp_name[32];
5040         char *name_ptr;
5041         int name_len;
5042         int is_curr = 0;        /* ctx->pos points to the current index? */
5043
5044         /* FIXME, use a real flag for deciding about the key type */
5045         if (root->fs_info->tree_root == root)
5046                 key_type = BTRFS_DIR_ITEM_KEY;
5047
5048         if (!dir_emit_dots(file, ctx))
5049                 return 0;
5050
5051         path = btrfs_alloc_path();
5052         if (!path)
5053                 return -ENOMEM;
5054
5055         path->reada = 1;
5056
5057         if (key_type == BTRFS_DIR_INDEX_KEY) {
5058                 INIT_LIST_HEAD(&ins_list);
5059                 INIT_LIST_HEAD(&del_list);
5060                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5061         }
5062
5063         btrfs_set_key_type(&key, key_type);
5064         key.offset = ctx->pos;
5065         key.objectid = btrfs_ino(inode);
5066
5067         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5068         if (ret < 0)
5069                 goto err;
5070
5071         while (1) {
5072                 leaf = path->nodes[0];
5073                 slot = path->slots[0];
5074                 if (slot >= btrfs_header_nritems(leaf)) {
5075                         ret = btrfs_next_leaf(root, path);
5076                         if (ret < 0)
5077                                 goto err;
5078                         else if (ret > 0)
5079                                 break;
5080                         continue;
5081                 }
5082
5083                 item = btrfs_item_nr(leaf, slot);
5084                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5085
5086                 if (found_key.objectid != key.objectid)
5087                         break;
5088                 if (btrfs_key_type(&found_key) != key_type)
5089                         break;
5090                 if (found_key.offset < ctx->pos)
5091                         goto next;
5092                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5093                     btrfs_should_delete_dir_index(&del_list,
5094                                                   found_key.offset))
5095                         goto next;
5096
5097                 ctx->pos = found_key.offset;
5098                 is_curr = 1;
5099
5100                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5101                 di_cur = 0;
5102                 di_total = btrfs_item_size(leaf, item);
5103
5104                 while (di_cur < di_total) {
5105                         struct btrfs_key location;
5106
5107                         if (verify_dir_item(root, leaf, di))
5108                                 break;
5109
5110                         name_len = btrfs_dir_name_len(leaf, di);
5111                         if (name_len <= sizeof(tmp_name)) {
5112                                 name_ptr = tmp_name;
5113                         } else {
5114                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5115                                 if (!name_ptr) {
5116                                         ret = -ENOMEM;
5117                                         goto err;
5118                                 }
5119                         }
5120                         read_extent_buffer(leaf, name_ptr,
5121                                            (unsigned long)(di + 1), name_len);
5122
5123                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5124                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5125
5126
5127                         /* is this a reference to our own snapshot? If so
5128                          * skip it.
5129                          *
5130                          * In contrast to old kernels, we insert the snapshot's
5131                          * dir item and dir index after it has been created, so
5132                          * we won't find a reference to our own snapshot. We
5133                          * still keep the following code for backward
5134                          * compatibility.
5135                          */
5136                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5137                             location.objectid == root->root_key.objectid) {
5138                                 over = 0;
5139                                 goto skip;
5140                         }
5141                         over = !dir_emit(ctx, name_ptr, name_len,
5142                                        location.objectid, d_type);
5143
5144 skip:
5145                         if (name_ptr != tmp_name)
5146                                 kfree(name_ptr);
5147
5148                         if (over)
5149                                 goto nopos;
5150                         di_len = btrfs_dir_name_len(leaf, di) +
5151                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5152                         di_cur += di_len;
5153                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5154                 }
5155 next:
5156                 path->slots[0]++;
5157         }
5158
5159         if (key_type == BTRFS_DIR_INDEX_KEY) {
5160                 if (is_curr)
5161                         ctx->pos++;
5162                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5163                 if (ret)
5164                         goto nopos;
5165         }
5166
5167         /* Reached end of directory/root. Bump pos past the last item. */
5168         if (key_type == BTRFS_DIR_INDEX_KEY)
5169                 /*
5170                  * 32-bit glibc will use getdents64, but then strtol -
5171                  * so the last number we can serve is this.
5172                  */
5173                 ctx->pos = 0x7fffffff;
5174         else
5175                 ctx->pos++;
5176 nopos:
5177         ret = 0;
5178 err:
5179         if (key_type == BTRFS_DIR_INDEX_KEY)
5180                 btrfs_put_delayed_items(&ins_list, &del_list);
5181         btrfs_free_path(path);
5182         return ret;
5183 }
5184
5185 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5186 {
5187         struct btrfs_root *root = BTRFS_I(inode)->root;
5188         struct btrfs_trans_handle *trans;
5189         int ret = 0;
5190         bool nolock = false;
5191
5192         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5193                 return 0;
5194
5195         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5196                 nolock = true;
5197
5198         if (wbc->sync_mode == WB_SYNC_ALL) {
5199                 if (nolock)
5200                         trans = btrfs_join_transaction_nolock(root);
5201                 else
5202                         trans = btrfs_join_transaction(root);
5203                 if (IS_ERR(trans))
5204                         return PTR_ERR(trans);
5205                 ret = btrfs_commit_transaction(trans, root);
5206         }
5207         return ret;
5208 }
5209
5210 /*
5211  * This is somewhat expensive, updating the tree every time the
5212  * inode changes.  But, it is most likely to find the inode in cache.
5213  * FIXME, needs more benchmarking...there are no reasons other than performance
5214  * to keep or drop this code.
5215  */
5216 static int btrfs_dirty_inode(struct inode *inode)
5217 {
5218         struct btrfs_root *root = BTRFS_I(inode)->root;
5219         struct btrfs_trans_handle *trans;
5220         int ret;
5221
5222         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5223                 return 0;
5224
5225         trans = btrfs_join_transaction(root);
5226         if (IS_ERR(trans))
5227                 return PTR_ERR(trans);
5228
5229         ret = btrfs_update_inode(trans, root, inode);
5230         if (ret && ret == -ENOSPC) {
5231                 /* whoops, lets try again with the full transaction */
5232                 btrfs_end_transaction(trans, root);
5233                 trans = btrfs_start_transaction(root, 1);
5234                 if (IS_ERR(trans))
5235                         return PTR_ERR(trans);
5236
5237                 ret = btrfs_update_inode(trans, root, inode);
5238         }
5239         btrfs_end_transaction(trans, root);
5240         if (BTRFS_I(inode)->delayed_node)
5241                 btrfs_balance_delayed_items(root);
5242
5243         return ret;
5244 }
5245
5246 /*
5247  * This is a copy of file_update_time.  We need this so we can return error on
5248  * ENOSPC for updating the inode in the case of file write and mmap writes.
5249  */
5250 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5251                              int flags)
5252 {
5253         struct btrfs_root *root = BTRFS_I(inode)->root;
5254
5255         if (btrfs_root_readonly(root))
5256                 return -EROFS;
5257
5258         if (flags & S_VERSION)
5259                 inode_inc_iversion(inode);
5260         if (flags & S_CTIME)
5261                 inode->i_ctime = *now;
5262         if (flags & S_MTIME)
5263                 inode->i_mtime = *now;
5264         if (flags & S_ATIME)
5265                 inode->i_atime = *now;
5266         return btrfs_dirty_inode(inode);
5267 }
5268
5269 /*
5270  * find the highest existing sequence number in a directory
5271  * and then set the in-memory index_cnt variable to reflect
5272  * free sequence numbers
5273  */
5274 static int btrfs_set_inode_index_count(struct inode *inode)
5275 {
5276         struct btrfs_root *root = BTRFS_I(inode)->root;
5277         struct btrfs_key key, found_key;
5278         struct btrfs_path *path;
5279         struct extent_buffer *leaf;
5280         int ret;
5281
5282         key.objectid = btrfs_ino(inode);
5283         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5284         key.offset = (u64)-1;
5285
5286         path = btrfs_alloc_path();
5287         if (!path)
5288                 return -ENOMEM;
5289
5290         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5291         if (ret < 0)
5292                 goto out;
5293         /* FIXME: we should be able to handle this */
5294         if (ret == 0)
5295                 goto out;
5296         ret = 0;
5297
5298         /*
5299          * MAGIC NUMBER EXPLANATION:
5300          * since we search a directory based on f_pos we have to start at 2
5301          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5302          * else has to start at 2
5303          */
5304         if (path->slots[0] == 0) {
5305                 BTRFS_I(inode)->index_cnt = 2;
5306                 goto out;
5307         }
5308
5309         path->slots[0]--;
5310
5311         leaf = path->nodes[0];
5312         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5313
5314         if (found_key.objectid != btrfs_ino(inode) ||
5315             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5316                 BTRFS_I(inode)->index_cnt = 2;
5317                 goto out;
5318         }
5319
5320         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5321 out:
5322         btrfs_free_path(path);
5323         return ret;
5324 }
5325
5326 /*
5327  * helper to find a free sequence number in a given directory.  This current
5328  * code is very simple, later versions will do smarter things in the btree
5329  */
5330 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5331 {
5332         int ret = 0;
5333
5334         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5335                 ret = btrfs_inode_delayed_dir_index_count(dir);
5336                 if (ret) {
5337                         ret = btrfs_set_inode_index_count(dir);
5338                         if (ret)
5339                                 return ret;
5340                 }
5341         }
5342
5343         *index = BTRFS_I(dir)->index_cnt;
5344         BTRFS_I(dir)->index_cnt++;
5345
5346         return ret;
5347 }
5348
5349 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5350                                      struct btrfs_root *root,
5351                                      struct inode *dir,
5352                                      const char *name, int name_len,
5353                                      u64 ref_objectid, u64 objectid,
5354                                      umode_t mode, u64 *index)
5355 {
5356         struct inode *inode;
5357         struct btrfs_inode_item *inode_item;
5358         struct btrfs_key *location;
5359         struct btrfs_path *path;
5360         struct btrfs_inode_ref *ref;
5361         struct btrfs_key key[2];
5362         u32 sizes[2];
5363         unsigned long ptr;
5364         int ret;
5365         int owner;
5366
5367         path = btrfs_alloc_path();
5368         if (!path)
5369                 return ERR_PTR(-ENOMEM);
5370
5371         inode = new_inode(root->fs_info->sb);
5372         if (!inode) {
5373                 btrfs_free_path(path);
5374                 return ERR_PTR(-ENOMEM);
5375         }
5376
5377         /*
5378          * we have to initialize this early, so we can reclaim the inode
5379          * number if we fail afterwards in this function.
5380          */
5381         inode->i_ino = objectid;
5382
5383         if (dir) {
5384                 trace_btrfs_inode_request(dir);
5385
5386                 ret = btrfs_set_inode_index(dir, index);
5387                 if (ret) {
5388                         btrfs_free_path(path);
5389                         iput(inode);
5390                         return ERR_PTR(ret);
5391                 }
5392         }
5393         /*
5394          * index_cnt is ignored for everything but a dir,
5395          * btrfs_get_inode_index_count has an explanation for the magic
5396          * number
5397          */
5398         BTRFS_I(inode)->index_cnt = 2;
5399         BTRFS_I(inode)->root = root;
5400         BTRFS_I(inode)->generation = trans->transid;
5401         inode->i_generation = BTRFS_I(inode)->generation;
5402
5403         /*
5404          * We could have gotten an inode number from somebody who was fsynced
5405          * and then removed in this same transaction, so let's just set full
5406          * sync since it will be a full sync anyway and this will blow away the
5407          * old info in the log.
5408          */
5409         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5410
5411         if (S_ISDIR(mode))
5412                 owner = 0;
5413         else
5414                 owner = 1;
5415
5416         key[0].objectid = objectid;
5417         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5418         key[0].offset = 0;
5419
5420         /*
5421          * Start new inodes with an inode_ref. This is slightly more
5422          * efficient for small numbers of hard links since they will
5423          * be packed into one item. Extended refs will kick in if we
5424          * add more hard links than can fit in the ref item.
5425          */
5426         key[1].objectid = objectid;
5427         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5428         key[1].offset = ref_objectid;
5429
5430         sizes[0] = sizeof(struct btrfs_inode_item);
5431         sizes[1] = name_len + sizeof(*ref);
5432
5433         path->leave_spinning = 1;
5434         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5435         if (ret != 0)
5436                 goto fail;
5437
5438         inode_init_owner(inode, dir, mode);
5439         inode_set_bytes(inode, 0);
5440         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5441         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5442                                   struct btrfs_inode_item);
5443         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5444                              sizeof(*inode_item));
5445         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5446
5447         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5448                              struct btrfs_inode_ref);
5449         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5450         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5451         ptr = (unsigned long)(ref + 1);
5452         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5453
5454         btrfs_mark_buffer_dirty(path->nodes[0]);
5455         btrfs_free_path(path);
5456
5457         location = &BTRFS_I(inode)->location;
5458         location->objectid = objectid;
5459         location->offset = 0;
5460         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5461
5462         btrfs_inherit_iflags(inode, dir);
5463
5464         if (S_ISREG(mode)) {
5465                 if (btrfs_test_opt(root, NODATASUM))
5466                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5467                 if (btrfs_test_opt(root, NODATACOW))
5468                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5469                                 BTRFS_INODE_NODATASUM;
5470         }
5471
5472         insert_inode_hash(inode);
5473         inode_tree_add(inode);
5474
5475         trace_btrfs_inode_new(inode);
5476         btrfs_set_inode_last_trans(trans, inode);
5477
5478         btrfs_update_root_times(trans, root);
5479
5480         return inode;
5481 fail:
5482         if (dir)
5483                 BTRFS_I(dir)->index_cnt--;
5484         btrfs_free_path(path);
5485         iput(inode);
5486         return ERR_PTR(ret);
5487 }
5488
5489 static inline u8 btrfs_inode_type(struct inode *inode)
5490 {
5491         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5492 }
5493
5494 /*
5495  * utility function to add 'inode' into 'parent_inode' with
5496  * a give name and a given sequence number.
5497  * if 'add_backref' is true, also insert a backref from the
5498  * inode to the parent directory.
5499  */
5500 int btrfs_add_link(struct btrfs_trans_handle *trans,
5501                    struct inode *parent_inode, struct inode *inode,
5502                    const char *name, int name_len, int add_backref, u64 index)
5503 {
5504         int ret = 0;
5505         struct btrfs_key key;
5506         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5507         u64 ino = btrfs_ino(inode);
5508         u64 parent_ino = btrfs_ino(parent_inode);
5509
5510         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5511                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5512         } else {
5513                 key.objectid = ino;
5514                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5515                 key.offset = 0;
5516         }
5517
5518         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5519                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5520                                          key.objectid, root->root_key.objectid,
5521                                          parent_ino, index, name, name_len);
5522         } else if (add_backref) {
5523                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5524                                              parent_ino, index);
5525         }
5526
5527         /* Nothing to clean up yet */
5528         if (ret)
5529                 return ret;
5530
5531         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5532                                     parent_inode, &key,
5533                                     btrfs_inode_type(inode), index);
5534         if (ret == -EEXIST || ret == -EOVERFLOW)
5535                 goto fail_dir_item;
5536         else if (ret) {
5537                 btrfs_abort_transaction(trans, root, ret);
5538                 return ret;
5539         }
5540
5541         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5542                            name_len * 2);
5543         inode_inc_iversion(parent_inode);
5544         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5545         ret = btrfs_update_inode(trans, root, parent_inode);
5546         if (ret)
5547                 btrfs_abort_transaction(trans, root, ret);
5548         return ret;
5549
5550 fail_dir_item:
5551         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5552                 u64 local_index;
5553                 int err;
5554                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5555                                  key.objectid, root->root_key.objectid,
5556                                  parent_ino, &local_index, name, name_len);
5557
5558         } else if (add_backref) {
5559                 u64 local_index;
5560                 int err;
5561
5562                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5563                                           ino, parent_ino, &local_index);
5564         }
5565         return ret;
5566 }
5567
5568 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5569                             struct inode *dir, struct dentry *dentry,
5570                             struct inode *inode, int backref, u64 index)
5571 {
5572         int err = btrfs_add_link(trans, dir, inode,
5573                                  dentry->d_name.name, dentry->d_name.len,
5574                                  backref, index);
5575         if (err > 0)
5576                 err = -EEXIST;
5577         return err;
5578 }
5579
5580 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5581                         umode_t mode, dev_t rdev)
5582 {
5583         struct btrfs_trans_handle *trans;
5584         struct btrfs_root *root = BTRFS_I(dir)->root;
5585         struct inode *inode = NULL;
5586         int err;
5587         int drop_inode = 0;
5588         u64 objectid;
5589         u64 index = 0;
5590
5591         if (!new_valid_dev(rdev))
5592                 return -EINVAL;
5593
5594         /*
5595          * 2 for inode item and ref
5596          * 2 for dir items
5597          * 1 for xattr if selinux is on
5598          */
5599         trans = btrfs_start_transaction(root, 5);
5600         if (IS_ERR(trans))
5601                 return PTR_ERR(trans);
5602
5603         err = btrfs_find_free_ino(root, &objectid);
5604         if (err)
5605                 goto out_unlock;
5606
5607         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5608                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5609                                 mode, &index);
5610         if (IS_ERR(inode)) {
5611                 err = PTR_ERR(inode);
5612                 goto out_unlock;
5613         }
5614
5615         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5616         if (err) {
5617                 drop_inode = 1;
5618                 goto out_unlock;
5619         }
5620
5621         /*
5622         * If the active LSM wants to access the inode during
5623         * d_instantiate it needs these. Smack checks to see
5624         * if the filesystem supports xattrs by looking at the
5625         * ops vector.
5626         */
5627
5628         inode->i_op = &btrfs_special_inode_operations;
5629         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5630         if (err)
5631                 drop_inode = 1;
5632         else {
5633                 init_special_inode(inode, inode->i_mode, rdev);
5634                 btrfs_update_inode(trans, root, inode);
5635                 d_instantiate(dentry, inode);
5636         }
5637 out_unlock:
5638         btrfs_end_transaction(trans, root);
5639         btrfs_btree_balance_dirty(root);
5640         if (drop_inode) {
5641                 inode_dec_link_count(inode);
5642                 iput(inode);
5643         }
5644         return err;
5645 }
5646
5647 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5648                         umode_t mode, bool excl)
5649 {
5650         struct btrfs_trans_handle *trans;
5651         struct btrfs_root *root = BTRFS_I(dir)->root;
5652         struct inode *inode = NULL;
5653         int drop_inode_on_err = 0;
5654         int err;
5655         u64 objectid;
5656         u64 index = 0;
5657
5658         /*
5659          * 2 for inode item and ref
5660          * 2 for dir items
5661          * 1 for xattr if selinux is on
5662          */
5663         trans = btrfs_start_transaction(root, 5);
5664         if (IS_ERR(trans))
5665                 return PTR_ERR(trans);
5666
5667         err = btrfs_find_free_ino(root, &objectid);
5668         if (err)
5669                 goto out_unlock;
5670
5671         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5672                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5673                                 mode, &index);
5674         if (IS_ERR(inode)) {
5675                 err = PTR_ERR(inode);
5676                 goto out_unlock;
5677         }
5678         drop_inode_on_err = 1;
5679
5680         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5681         if (err)
5682                 goto out_unlock;
5683
5684         err = btrfs_update_inode(trans, root, inode);
5685         if (err)
5686                 goto out_unlock;
5687
5688         /*
5689         * If the active LSM wants to access the inode during
5690         * d_instantiate it needs these. Smack checks to see
5691         * if the filesystem supports xattrs by looking at the
5692         * ops vector.
5693         */
5694         inode->i_fop = &btrfs_file_operations;
5695         inode->i_op = &btrfs_file_inode_operations;
5696
5697         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5698         if (err)
5699                 goto out_unlock;
5700
5701         inode->i_mapping->a_ops = &btrfs_aops;
5702         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5703         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5704         d_instantiate(dentry, inode);
5705
5706 out_unlock:
5707         btrfs_end_transaction(trans, root);
5708         if (err && drop_inode_on_err) {
5709                 inode_dec_link_count(inode);
5710                 iput(inode);
5711         }
5712         btrfs_btree_balance_dirty(root);
5713         return err;
5714 }
5715
5716 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5717                       struct dentry *dentry)
5718 {
5719         struct btrfs_trans_handle *trans;
5720         struct btrfs_root *root = BTRFS_I(dir)->root;
5721         struct inode *inode = old_dentry->d_inode;
5722         u64 index;
5723         int err;
5724         int drop_inode = 0;
5725
5726         /* do not allow sys_link's with other subvols of the same device */
5727         if (root->objectid != BTRFS_I(inode)->root->objectid)
5728                 return -EXDEV;
5729
5730         if (inode->i_nlink >= BTRFS_LINK_MAX)
5731                 return -EMLINK;
5732
5733         err = btrfs_set_inode_index(dir, &index);
5734         if (err)
5735                 goto fail;
5736
5737         /*
5738          * 2 items for inode and inode ref
5739          * 2 items for dir items
5740          * 1 item for parent inode
5741          */
5742         trans = btrfs_start_transaction(root, 5);
5743         if (IS_ERR(trans)) {
5744                 err = PTR_ERR(trans);
5745                 goto fail;
5746         }
5747
5748         btrfs_inc_nlink(inode);
5749         inode_inc_iversion(inode);
5750         inode->i_ctime = CURRENT_TIME;
5751         ihold(inode);
5752         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5753
5754         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5755
5756         if (err) {
5757                 drop_inode = 1;
5758         } else {
5759                 struct dentry *parent = dentry->d_parent;
5760                 err = btrfs_update_inode(trans, root, inode);
5761                 if (err)
5762                         goto fail;
5763                 d_instantiate(dentry, inode);
5764                 btrfs_log_new_name(trans, inode, NULL, parent);
5765         }
5766
5767         btrfs_end_transaction(trans, root);
5768 fail:
5769         if (drop_inode) {
5770                 inode_dec_link_count(inode);
5771                 iput(inode);
5772         }
5773         btrfs_btree_balance_dirty(root);
5774         return err;
5775 }
5776
5777 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5778 {
5779         struct inode *inode = NULL;
5780         struct btrfs_trans_handle *trans;
5781         struct btrfs_root *root = BTRFS_I(dir)->root;
5782         int err = 0;
5783         int drop_on_err = 0;
5784         u64 objectid = 0;
5785         u64 index = 0;
5786
5787         /*
5788          * 2 items for inode and ref
5789          * 2 items for dir items
5790          * 1 for xattr if selinux is on
5791          */
5792         trans = btrfs_start_transaction(root, 5);
5793         if (IS_ERR(trans))
5794                 return PTR_ERR(trans);
5795
5796         err = btrfs_find_free_ino(root, &objectid);
5797         if (err)
5798                 goto out_fail;
5799
5800         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5801                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5802                                 S_IFDIR | mode, &index);
5803         if (IS_ERR(inode)) {
5804                 err = PTR_ERR(inode);
5805                 goto out_fail;
5806         }
5807
5808         drop_on_err = 1;
5809
5810         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5811         if (err)
5812                 goto out_fail;
5813
5814         inode->i_op = &btrfs_dir_inode_operations;
5815         inode->i_fop = &btrfs_dir_file_operations;
5816
5817         btrfs_i_size_write(inode, 0);
5818         err = btrfs_update_inode(trans, root, inode);
5819         if (err)
5820                 goto out_fail;
5821
5822         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5823                              dentry->d_name.len, 0, index);
5824         if (err)
5825                 goto out_fail;
5826
5827         d_instantiate(dentry, inode);
5828         drop_on_err = 0;
5829
5830 out_fail:
5831         btrfs_end_transaction(trans, root);
5832         if (drop_on_err)
5833                 iput(inode);
5834         btrfs_btree_balance_dirty(root);
5835         return err;
5836 }
5837
5838 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5839  * and an extent that you want to insert, deal with overlap and insert
5840  * the new extent into the tree.
5841  */
5842 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5843                                 struct extent_map *existing,
5844                                 struct extent_map *em,
5845                                 u64 map_start, u64 map_len)
5846 {
5847         u64 start_diff;
5848
5849         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5850         start_diff = map_start - em->start;
5851         em->start = map_start;
5852         em->len = map_len;
5853         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5854             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5855                 em->block_start += start_diff;
5856                 em->block_len -= start_diff;
5857         }
5858         return add_extent_mapping(em_tree, em, 0);
5859 }
5860
5861 static noinline int uncompress_inline(struct btrfs_path *path,
5862                                       struct inode *inode, struct page *page,
5863                                       size_t pg_offset, u64 extent_offset,
5864                                       struct btrfs_file_extent_item *item)
5865 {
5866         int ret;
5867         struct extent_buffer *leaf = path->nodes[0];
5868         char *tmp;
5869         size_t max_size;
5870         unsigned long inline_size;
5871         unsigned long ptr;
5872         int compress_type;
5873
5874         WARN_ON(pg_offset != 0);
5875         compress_type = btrfs_file_extent_compression(leaf, item);
5876         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5877         inline_size = btrfs_file_extent_inline_item_len(leaf,
5878                                         btrfs_item_nr(leaf, path->slots[0]));
5879         tmp = kmalloc(inline_size, GFP_NOFS);
5880         if (!tmp)
5881                 return -ENOMEM;
5882         ptr = btrfs_file_extent_inline_start(item);
5883
5884         read_extent_buffer(leaf, tmp, ptr, inline_size);
5885
5886         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5887         ret = btrfs_decompress(compress_type, tmp, page,
5888                                extent_offset, inline_size, max_size);
5889         if (ret) {
5890                 char *kaddr = kmap_atomic(page);
5891                 unsigned long copy_size = min_t(u64,
5892                                   PAGE_CACHE_SIZE - pg_offset,
5893                                   max_size - extent_offset);
5894                 memset(kaddr + pg_offset, 0, copy_size);
5895                 kunmap_atomic(kaddr);
5896         }
5897         kfree(tmp);
5898         return 0;
5899 }
5900
5901 /*
5902  * a bit scary, this does extent mapping from logical file offset to the disk.
5903  * the ugly parts come from merging extents from the disk with the in-ram
5904  * representation.  This gets more complex because of the data=ordered code,
5905  * where the in-ram extents might be locked pending data=ordered completion.
5906  *
5907  * This also copies inline extents directly into the page.
5908  */
5909
5910 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5911                                     size_t pg_offset, u64 start, u64 len,
5912                                     int create)
5913 {
5914         int ret;
5915         int err = 0;
5916         u64 bytenr;
5917         u64 extent_start = 0;
5918         u64 extent_end = 0;
5919         u64 objectid = btrfs_ino(inode);
5920         u32 found_type;
5921         struct btrfs_path *path = NULL;
5922         struct btrfs_root *root = BTRFS_I(inode)->root;
5923         struct btrfs_file_extent_item *item;
5924         struct extent_buffer *leaf;
5925         struct btrfs_key found_key;
5926         struct extent_map *em = NULL;
5927         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5928         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5929         struct btrfs_trans_handle *trans = NULL;
5930         int compress_type;
5931
5932 again:
5933         read_lock(&em_tree->lock);
5934         em = lookup_extent_mapping(em_tree, start, len);
5935         if (em)
5936                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5937         read_unlock(&em_tree->lock);
5938
5939         if (em) {
5940                 if (em->start > start || em->start + em->len <= start)
5941                         free_extent_map(em);
5942                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5943                         free_extent_map(em);
5944                 else
5945                         goto out;
5946         }
5947         em = alloc_extent_map();
5948         if (!em) {
5949                 err = -ENOMEM;
5950                 goto out;
5951         }
5952         em->bdev = root->fs_info->fs_devices->latest_bdev;
5953         em->start = EXTENT_MAP_HOLE;
5954         em->orig_start = EXTENT_MAP_HOLE;
5955         em->len = (u64)-1;
5956         em->block_len = (u64)-1;
5957
5958         if (!path) {
5959                 path = btrfs_alloc_path();
5960                 if (!path) {
5961                         err = -ENOMEM;
5962                         goto out;
5963                 }
5964                 /*
5965                  * Chances are we'll be called again, so go ahead and do
5966                  * readahead
5967                  */
5968                 path->reada = 1;
5969         }
5970
5971         ret = btrfs_lookup_file_extent(trans, root, path,
5972                                        objectid, start, trans != NULL);
5973         if (ret < 0) {
5974                 err = ret;
5975                 goto out;
5976         }
5977
5978         if (ret != 0) {
5979                 if (path->slots[0] == 0)
5980                         goto not_found;
5981                 path->slots[0]--;
5982         }
5983
5984         leaf = path->nodes[0];
5985         item = btrfs_item_ptr(leaf, path->slots[0],
5986                               struct btrfs_file_extent_item);
5987         /* are we inside the extent that was found? */
5988         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5989         found_type = btrfs_key_type(&found_key);
5990         if (found_key.objectid != objectid ||
5991             found_type != BTRFS_EXTENT_DATA_KEY) {
5992                 goto not_found;
5993         }
5994
5995         found_type = btrfs_file_extent_type(leaf, item);
5996         extent_start = found_key.offset;
5997         compress_type = btrfs_file_extent_compression(leaf, item);
5998         if (found_type == BTRFS_FILE_EXTENT_REG ||
5999             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6000                 extent_end = extent_start +
6001                        btrfs_file_extent_num_bytes(leaf, item);
6002         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6003                 size_t size;
6004                 size = btrfs_file_extent_inline_len(leaf, item);
6005                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6006         }
6007
6008         if (start >= extent_end) {
6009                 path->slots[0]++;
6010                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6011                         ret = btrfs_next_leaf(root, path);
6012                         if (ret < 0) {
6013                                 err = ret;
6014                                 goto out;
6015                         }
6016                         if (ret > 0)
6017                                 goto not_found;
6018                         leaf = path->nodes[0];
6019                 }
6020                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6021                 if (found_key.objectid != objectid ||
6022                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6023                         goto not_found;
6024                 if (start + len <= found_key.offset)
6025                         goto not_found;
6026                 em->start = start;
6027                 em->orig_start = start;
6028                 em->len = found_key.offset - start;
6029                 goto not_found_em;
6030         }
6031
6032         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6033         if (found_type == BTRFS_FILE_EXTENT_REG ||
6034             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6035                 em->start = extent_start;
6036                 em->len = extent_end - extent_start;
6037                 em->orig_start = extent_start -
6038                                  btrfs_file_extent_offset(leaf, item);
6039                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6040                                                                       item);
6041                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6042                 if (bytenr == 0) {
6043                         em->block_start = EXTENT_MAP_HOLE;
6044                         goto insert;
6045                 }
6046                 if (compress_type != BTRFS_COMPRESS_NONE) {
6047                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6048                         em->compress_type = compress_type;
6049                         em->block_start = bytenr;
6050                         em->block_len = em->orig_block_len;
6051                 } else {
6052                         bytenr += btrfs_file_extent_offset(leaf, item);
6053                         em->block_start = bytenr;
6054                         em->block_len = em->len;
6055                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6056                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6057                 }
6058                 goto insert;
6059         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6060                 unsigned long ptr;
6061                 char *map;
6062                 size_t size;
6063                 size_t extent_offset;
6064                 size_t copy_size;
6065
6066                 em->block_start = EXTENT_MAP_INLINE;
6067                 if (!page || create) {
6068                         em->start = extent_start;
6069                         em->len = extent_end - extent_start;
6070                         goto out;
6071                 }
6072
6073                 size = btrfs_file_extent_inline_len(leaf, item);
6074                 extent_offset = page_offset(page) + pg_offset - extent_start;
6075                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6076                                 size - extent_offset);
6077                 em->start = extent_start + extent_offset;
6078                 em->len = ALIGN(copy_size, root->sectorsize);
6079                 em->orig_block_len = em->len;
6080                 em->orig_start = em->start;
6081                 if (compress_type) {
6082                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6083                         em->compress_type = compress_type;
6084                 }
6085                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6086                 if (create == 0 && !PageUptodate(page)) {
6087                         if (btrfs_file_extent_compression(leaf, item) !=
6088                             BTRFS_COMPRESS_NONE) {
6089                                 ret = uncompress_inline(path, inode, page,
6090                                                         pg_offset,
6091                                                         extent_offset, item);
6092                                 BUG_ON(ret); /* -ENOMEM */
6093                         } else {
6094                                 map = kmap(page);
6095                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6096                                                    copy_size);
6097                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6098                                         memset(map + pg_offset + copy_size, 0,
6099                                                PAGE_CACHE_SIZE - pg_offset -
6100                                                copy_size);
6101                                 }
6102                                 kunmap(page);
6103                         }
6104                         flush_dcache_page(page);
6105                 } else if (create && PageUptodate(page)) {
6106                         BUG();
6107                         if (!trans) {
6108                                 kunmap(page);
6109                                 free_extent_map(em);
6110                                 em = NULL;
6111
6112                                 btrfs_release_path(path);
6113                                 trans = btrfs_join_transaction(root);
6114
6115                                 if (IS_ERR(trans))
6116                                         return ERR_CAST(trans);
6117                                 goto again;
6118                         }
6119                         map = kmap(page);
6120                         write_extent_buffer(leaf, map + pg_offset, ptr,
6121                                             copy_size);
6122                         kunmap(page);
6123                         btrfs_mark_buffer_dirty(leaf);
6124                 }
6125                 set_extent_uptodate(io_tree, em->start,
6126                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6127                 goto insert;
6128         } else {
6129                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6130         }
6131 not_found:
6132         em->start = start;
6133         em->orig_start = start;
6134         em->len = len;
6135 not_found_em:
6136         em->block_start = EXTENT_MAP_HOLE;
6137         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6138 insert:
6139         btrfs_release_path(path);
6140         if (em->start > start || extent_map_end(em) <= start) {
6141                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6142                         (unsigned long long)em->start,
6143                         (unsigned long long)em->len,
6144                         (unsigned long long)start,
6145                         (unsigned long long)len);
6146                 err = -EIO;
6147                 goto out;
6148         }
6149
6150         err = 0;
6151         write_lock(&em_tree->lock);
6152         ret = add_extent_mapping(em_tree, em, 0);
6153         /* it is possible that someone inserted the extent into the tree
6154          * while we had the lock dropped.  It is also possible that
6155          * an overlapping map exists in the tree
6156          */
6157         if (ret == -EEXIST) {
6158                 struct extent_map *existing;
6159
6160                 ret = 0;
6161
6162                 existing = lookup_extent_mapping(em_tree, start, len);
6163                 if (existing && (existing->start > start ||
6164                     existing->start + existing->len <= start)) {
6165                         free_extent_map(existing);
6166                         existing = NULL;
6167                 }
6168                 if (!existing) {
6169                         existing = lookup_extent_mapping(em_tree, em->start,
6170                                                          em->len);
6171                         if (existing) {
6172                                 err = merge_extent_mapping(em_tree, existing,
6173                                                            em, start,
6174                                                            root->sectorsize);
6175                                 free_extent_map(existing);
6176                                 if (err) {
6177                                         free_extent_map(em);
6178                                         em = NULL;
6179                                 }
6180                         } else {
6181                                 err = -EIO;
6182                                 free_extent_map(em);
6183                                 em = NULL;
6184                         }
6185                 } else {
6186                         free_extent_map(em);
6187                         em = existing;
6188                         err = 0;
6189                 }
6190         }
6191         write_unlock(&em_tree->lock);
6192 out:
6193
6194         if (em)
6195                 trace_btrfs_get_extent(root, em);
6196
6197         if (path)
6198                 btrfs_free_path(path);
6199         if (trans) {
6200                 ret = btrfs_end_transaction(trans, root);
6201                 if (!err)
6202                         err = ret;
6203         }
6204         if (err) {
6205                 free_extent_map(em);
6206                 return ERR_PTR(err);
6207         }
6208         BUG_ON(!em); /* Error is always set */
6209         return em;
6210 }
6211
6212 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6213                                            size_t pg_offset, u64 start, u64 len,
6214                                            int create)
6215 {
6216         struct extent_map *em;
6217         struct extent_map *hole_em = NULL;
6218         u64 range_start = start;
6219         u64 end;
6220         u64 found;
6221         u64 found_end;
6222         int err = 0;
6223
6224         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6225         if (IS_ERR(em))
6226                 return em;
6227         if (em) {
6228                 /*
6229                  * if our em maps to
6230                  * -  a hole or
6231                  * -  a pre-alloc extent,
6232                  * there might actually be delalloc bytes behind it.
6233                  */
6234                 if (em->block_start != EXTENT_MAP_HOLE &&
6235                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6236                         return em;
6237                 else
6238                         hole_em = em;
6239         }
6240
6241         /* check to see if we've wrapped (len == -1 or similar) */
6242         end = start + len;
6243         if (end < start)
6244                 end = (u64)-1;
6245         else
6246                 end -= 1;
6247
6248         em = NULL;
6249
6250         /* ok, we didn't find anything, lets look for delalloc */
6251         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6252                                  end, len, EXTENT_DELALLOC, 1);
6253         found_end = range_start + found;
6254         if (found_end < range_start)
6255                 found_end = (u64)-1;
6256
6257         /*
6258          * we didn't find anything useful, return
6259          * the original results from get_extent()
6260          */
6261         if (range_start > end || found_end <= start) {
6262                 em = hole_em;
6263                 hole_em = NULL;
6264                 goto out;
6265         }
6266
6267         /* adjust the range_start to make sure it doesn't
6268          * go backwards from the start they passed in
6269          */
6270         range_start = max(start,range_start);
6271         found = found_end - range_start;
6272
6273         if (found > 0) {
6274                 u64 hole_start = start;
6275                 u64 hole_len = len;
6276
6277                 em = alloc_extent_map();
6278                 if (!em) {
6279                         err = -ENOMEM;
6280                         goto out;
6281                 }
6282                 /*
6283                  * when btrfs_get_extent can't find anything it
6284                  * returns one huge hole
6285                  *
6286                  * make sure what it found really fits our range, and
6287                  * adjust to make sure it is based on the start from
6288                  * the caller
6289                  */
6290                 if (hole_em) {
6291                         u64 calc_end = extent_map_end(hole_em);
6292
6293                         if (calc_end <= start || (hole_em->start > end)) {
6294                                 free_extent_map(hole_em);
6295                                 hole_em = NULL;
6296                         } else {
6297                                 hole_start = max(hole_em->start, start);
6298                                 hole_len = calc_end - hole_start;
6299                         }
6300                 }
6301                 em->bdev = NULL;
6302                 if (hole_em && range_start > hole_start) {
6303                         /* our hole starts before our delalloc, so we
6304                          * have to return just the parts of the hole
6305                          * that go until  the delalloc starts
6306                          */
6307                         em->len = min(hole_len,
6308                                       range_start - hole_start);
6309                         em->start = hole_start;
6310                         em->orig_start = hole_start;
6311                         /*
6312                          * don't adjust block start at all,
6313                          * it is fixed at EXTENT_MAP_HOLE
6314                          */
6315                         em->block_start = hole_em->block_start;
6316                         em->block_len = hole_len;
6317                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6318                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6319                 } else {
6320                         em->start = range_start;
6321                         em->len = found;
6322                         em->orig_start = range_start;
6323                         em->block_start = EXTENT_MAP_DELALLOC;
6324                         em->block_len = found;
6325                 }
6326         } else if (hole_em) {
6327                 return hole_em;
6328         }
6329 out:
6330
6331         free_extent_map(hole_em);
6332         if (err) {
6333                 free_extent_map(em);
6334                 return ERR_PTR(err);
6335         }
6336         return em;
6337 }
6338
6339 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6340                                                   u64 start, u64 len)
6341 {
6342         struct btrfs_root *root = BTRFS_I(inode)->root;
6343         struct btrfs_trans_handle *trans;
6344         struct extent_map *em;
6345         struct btrfs_key ins;
6346         u64 alloc_hint;
6347         int ret;
6348
6349         trans = btrfs_join_transaction(root);
6350         if (IS_ERR(trans))
6351                 return ERR_CAST(trans);
6352
6353         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6354
6355         alloc_hint = get_extent_allocation_hint(inode, start, len);
6356         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6357                                    alloc_hint, &ins, 1);
6358         if (ret) {
6359                 em = ERR_PTR(ret);
6360                 goto out;
6361         }
6362
6363         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6364                               ins.offset, ins.offset, ins.offset, 0);
6365         if (IS_ERR(em))
6366                 goto out;
6367
6368         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6369                                            ins.offset, ins.offset, 0);
6370         if (ret) {
6371                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6372                 em = ERR_PTR(ret);
6373         }
6374 out:
6375         btrfs_end_transaction(trans, root);
6376         return em;
6377 }
6378
6379 /*
6380  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6381  * block must be cow'd
6382  */
6383 noinline int can_nocow_extent(struct btrfs_trans_handle *trans,
6384                               struct inode *inode, u64 offset, u64 *len,
6385                               u64 *orig_start, u64 *orig_block_len,
6386                               u64 *ram_bytes)
6387 {
6388         struct btrfs_path *path;
6389         int ret;
6390         struct extent_buffer *leaf;
6391         struct btrfs_root *root = BTRFS_I(inode)->root;
6392         struct btrfs_file_extent_item *fi;
6393         struct btrfs_key key;
6394         u64 disk_bytenr;
6395         u64 backref_offset;
6396         u64 extent_end;
6397         u64 num_bytes;
6398         int slot;
6399         int found_type;
6400         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6401         path = btrfs_alloc_path();
6402         if (!path)
6403                 return -ENOMEM;
6404
6405         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6406                                        offset, 0);
6407         if (ret < 0)
6408                 goto out;
6409
6410         slot = path->slots[0];
6411         if (ret == 1) {
6412                 if (slot == 0) {
6413                         /* can't find the item, must cow */
6414                         ret = 0;
6415                         goto out;
6416                 }
6417                 slot--;
6418         }
6419         ret = 0;
6420         leaf = path->nodes[0];
6421         btrfs_item_key_to_cpu(leaf, &key, slot);
6422         if (key.objectid != btrfs_ino(inode) ||
6423             key.type != BTRFS_EXTENT_DATA_KEY) {
6424                 /* not our file or wrong item type, must cow */
6425                 goto out;
6426         }
6427
6428         if (key.offset > offset) {
6429                 /* Wrong offset, must cow */
6430                 goto out;
6431         }
6432
6433         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6434         found_type = btrfs_file_extent_type(leaf, fi);
6435         if (found_type != BTRFS_FILE_EXTENT_REG &&
6436             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6437                 /* not a regular extent, must cow */
6438                 goto out;
6439         }
6440
6441         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6442                 goto out;
6443
6444         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6445         if (disk_bytenr == 0)
6446                 goto out;
6447
6448         if (btrfs_file_extent_compression(leaf, fi) ||
6449             btrfs_file_extent_encryption(leaf, fi) ||
6450             btrfs_file_extent_other_encoding(leaf, fi))
6451                 goto out;
6452
6453         backref_offset = btrfs_file_extent_offset(leaf, fi);
6454
6455         if (orig_start) {
6456                 *orig_start = key.offset - backref_offset;
6457                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6458                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6459         }
6460
6461         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6462
6463         if (btrfs_extent_readonly(root, disk_bytenr))
6464                 goto out;
6465
6466         /*
6467          * look for other files referencing this extent, if we
6468          * find any we must cow
6469          */
6470         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6471                                   key.offset - backref_offset, disk_bytenr))
6472                 goto out;
6473
6474         /*
6475          * adjust disk_bytenr and num_bytes to cover just the bytes
6476          * in this extent we are about to write.  If there
6477          * are any csums in that range we have to cow in order
6478          * to keep the csums correct
6479          */
6480         disk_bytenr += backref_offset;
6481         disk_bytenr += offset - key.offset;
6482         num_bytes = min(offset + *len, extent_end) - offset;
6483         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6484                                 goto out;
6485         /*
6486          * all of the above have passed, it is safe to overwrite this extent
6487          * without cow
6488          */
6489         *len = num_bytes;
6490         ret = 1;
6491 out:
6492         btrfs_free_path(path);
6493         return ret;
6494 }
6495
6496 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6497                               struct extent_state **cached_state, int writing)
6498 {
6499         struct btrfs_ordered_extent *ordered;
6500         int ret = 0;
6501
6502         while (1) {
6503                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6504                                  0, cached_state);
6505                 /*
6506                  * We're concerned with the entire range that we're going to be
6507                  * doing DIO to, so we need to make sure theres no ordered
6508                  * extents in this range.
6509                  */
6510                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6511                                                      lockend - lockstart + 1);
6512
6513                 /*
6514                  * We need to make sure there are no buffered pages in this
6515                  * range either, we could have raced between the invalidate in
6516                  * generic_file_direct_write and locking the extent.  The
6517                  * invalidate needs to happen so that reads after a write do not
6518                  * get stale data.
6519                  */
6520                 if (!ordered && (!writing ||
6521                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6522                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6523                                     *cached_state)))
6524                         break;
6525
6526                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6527                                      cached_state, GFP_NOFS);
6528
6529                 if (ordered) {
6530                         btrfs_start_ordered_extent(inode, ordered, 1);
6531                         btrfs_put_ordered_extent(ordered);
6532                 } else {
6533                         /* Screw you mmap */
6534                         ret = filemap_write_and_wait_range(inode->i_mapping,
6535                                                            lockstart,
6536                                                            lockend);
6537                         if (ret)
6538                                 break;
6539
6540                         /*
6541                          * If we found a page that couldn't be invalidated just
6542                          * fall back to buffered.
6543                          */
6544                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6545                                         lockstart >> PAGE_CACHE_SHIFT,
6546                                         lockend >> PAGE_CACHE_SHIFT);
6547                         if (ret)
6548                                 break;
6549                 }
6550
6551                 cond_resched();
6552         }
6553
6554         return ret;
6555 }
6556
6557 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6558                                            u64 len, u64 orig_start,
6559                                            u64 block_start, u64 block_len,
6560                                            u64 orig_block_len, u64 ram_bytes,
6561                                            int type)
6562 {
6563         struct extent_map_tree *em_tree;
6564         struct extent_map *em;
6565         struct btrfs_root *root = BTRFS_I(inode)->root;
6566         int ret;
6567
6568         em_tree = &BTRFS_I(inode)->extent_tree;
6569         em = alloc_extent_map();
6570         if (!em)
6571                 return ERR_PTR(-ENOMEM);
6572
6573         em->start = start;
6574         em->orig_start = orig_start;
6575         em->mod_start = start;
6576         em->mod_len = len;
6577         em->len = len;
6578         em->block_len = block_len;
6579         em->block_start = block_start;
6580         em->bdev = root->fs_info->fs_devices->latest_bdev;
6581         em->orig_block_len = orig_block_len;
6582         em->ram_bytes = ram_bytes;
6583         em->generation = -1;
6584         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6585         if (type == BTRFS_ORDERED_PREALLOC)
6586                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6587
6588         do {
6589                 btrfs_drop_extent_cache(inode, em->start,
6590                                 em->start + em->len - 1, 0);
6591                 write_lock(&em_tree->lock);
6592                 ret = add_extent_mapping(em_tree, em, 1);
6593                 write_unlock(&em_tree->lock);
6594         } while (ret == -EEXIST);
6595
6596         if (ret) {
6597                 free_extent_map(em);
6598                 return ERR_PTR(ret);
6599         }
6600
6601         return em;
6602 }
6603
6604
6605 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6606                                    struct buffer_head *bh_result, int create)
6607 {
6608         struct extent_map *em;
6609         struct btrfs_root *root = BTRFS_I(inode)->root;
6610         struct extent_state *cached_state = NULL;
6611         u64 start = iblock << inode->i_blkbits;
6612         u64 lockstart, lockend;
6613         u64 len = bh_result->b_size;
6614         struct btrfs_trans_handle *trans;
6615         int unlock_bits = EXTENT_LOCKED;
6616         int ret = 0;
6617
6618         if (create)
6619                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6620         else
6621                 len = min_t(u64, len, root->sectorsize);
6622
6623         lockstart = start;
6624         lockend = start + len - 1;
6625
6626         /*
6627          * If this errors out it's because we couldn't invalidate pagecache for
6628          * this range and we need to fallback to buffered.
6629          */
6630         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6631                 return -ENOTBLK;
6632
6633         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6634         if (IS_ERR(em)) {
6635                 ret = PTR_ERR(em);
6636                 goto unlock_err;
6637         }
6638
6639         /*
6640          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6641          * io.  INLINE is special, and we could probably kludge it in here, but
6642          * it's still buffered so for safety lets just fall back to the generic
6643          * buffered path.
6644          *
6645          * For COMPRESSED we _have_ to read the entire extent in so we can
6646          * decompress it, so there will be buffering required no matter what we
6647          * do, so go ahead and fallback to buffered.
6648          *
6649          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6650          * to buffered IO.  Don't blame me, this is the price we pay for using
6651          * the generic code.
6652          */
6653         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6654             em->block_start == EXTENT_MAP_INLINE) {
6655                 free_extent_map(em);
6656                 ret = -ENOTBLK;
6657                 goto unlock_err;
6658         }
6659
6660         /* Just a good old fashioned hole, return */
6661         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6662                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6663                 free_extent_map(em);
6664                 goto unlock_err;
6665         }
6666
6667         /*
6668          * We don't allocate a new extent in the following cases
6669          *
6670          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6671          * existing extent.
6672          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6673          * just use the extent.
6674          *
6675          */
6676         if (!create) {
6677                 len = min(len, em->len - (start - em->start));
6678                 lockstart = start + len;
6679                 goto unlock;
6680         }
6681
6682         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6683             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6684              em->block_start != EXTENT_MAP_HOLE)) {
6685                 int type;
6686                 int ret;
6687                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6688
6689                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6690                         type = BTRFS_ORDERED_PREALLOC;
6691                 else
6692                         type = BTRFS_ORDERED_NOCOW;
6693                 len = min(len, em->len - (start - em->start));
6694                 block_start = em->block_start + (start - em->start);
6695
6696                 /*
6697                  * we're not going to log anything, but we do need
6698                  * to make sure the current transaction stays open
6699                  * while we look for nocow cross refs
6700                  */
6701                 trans = btrfs_join_transaction(root);
6702                 if (IS_ERR(trans))
6703                         goto must_cow;
6704
6705                 if (can_nocow_extent(trans, inode, start, &len, &orig_start,
6706                                      &orig_block_len, &ram_bytes) == 1) {
6707                         if (type == BTRFS_ORDERED_PREALLOC) {
6708                                 free_extent_map(em);
6709                                 em = create_pinned_em(inode, start, len,
6710                                                        orig_start,
6711                                                        block_start, len,
6712                                                        orig_block_len,
6713                                                        ram_bytes, type);
6714                                 if (IS_ERR(em)) {
6715                                         btrfs_end_transaction(trans, root);
6716                                         goto unlock_err;
6717                                 }
6718                         }
6719
6720                         ret = btrfs_add_ordered_extent_dio(inode, start,
6721                                            block_start, len, len, type);
6722                         btrfs_end_transaction(trans, root);
6723                         if (ret) {
6724                                 free_extent_map(em);
6725                                 goto unlock_err;
6726                         }
6727                         goto unlock;
6728                 }
6729                 btrfs_end_transaction(trans, root);
6730         }
6731 must_cow:
6732         /*
6733          * this will cow the extent, reset the len in case we changed
6734          * it above
6735          */
6736         len = bh_result->b_size;
6737         free_extent_map(em);
6738         em = btrfs_new_extent_direct(inode, start, len);
6739         if (IS_ERR(em)) {
6740                 ret = PTR_ERR(em);
6741                 goto unlock_err;
6742         }
6743         len = min(len, em->len - (start - em->start));
6744 unlock:
6745         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6746                 inode->i_blkbits;
6747         bh_result->b_size = len;
6748         bh_result->b_bdev = em->bdev;
6749         set_buffer_mapped(bh_result);
6750         if (create) {
6751                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6752                         set_buffer_new(bh_result);
6753
6754                 /*
6755                  * Need to update the i_size under the extent lock so buffered
6756                  * readers will get the updated i_size when we unlock.
6757                  */
6758                 if (start + len > i_size_read(inode))
6759                         i_size_write(inode, start + len);
6760
6761                 spin_lock(&BTRFS_I(inode)->lock);
6762                 BTRFS_I(inode)->outstanding_extents++;
6763                 spin_unlock(&BTRFS_I(inode)->lock);
6764
6765                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6766                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6767                                      &cached_state, GFP_NOFS);
6768                 BUG_ON(ret);
6769         }
6770
6771         /*
6772          * In the case of write we need to clear and unlock the entire range,
6773          * in the case of read we need to unlock only the end area that we
6774          * aren't using if there is any left over space.
6775          */
6776         if (lockstart < lockend) {
6777                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6778                                  lockend, unlock_bits, 1, 0,
6779                                  &cached_state, GFP_NOFS);
6780         } else {
6781                 free_extent_state(cached_state);
6782         }
6783
6784         free_extent_map(em);
6785
6786         return 0;
6787
6788 unlock_err:
6789         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6790                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6791         return ret;
6792 }
6793
6794 struct btrfs_dio_private {
6795         struct inode *inode;
6796         u64 logical_offset;
6797         u64 disk_bytenr;
6798         u64 bytes;
6799         void *private;
6800
6801         /* number of bios pending for this dio */
6802         atomic_t pending_bios;
6803
6804         /* IO errors */
6805         int errors;
6806
6807         /* orig_bio is our btrfs_io_bio */
6808         struct bio *orig_bio;
6809
6810         /* dio_bio came from fs/direct-io.c */
6811         struct bio *dio_bio;
6812 };
6813
6814 static void btrfs_endio_direct_read(struct bio *bio, int err)
6815 {
6816         struct btrfs_dio_private *dip = bio->bi_private;
6817         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6818         struct bio_vec *bvec = bio->bi_io_vec;
6819         struct inode *inode = dip->inode;
6820         struct btrfs_root *root = BTRFS_I(inode)->root;
6821         struct bio *dio_bio;
6822         u64 start;
6823
6824         start = dip->logical_offset;
6825         do {
6826                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6827                         struct page *page = bvec->bv_page;
6828                         char *kaddr;
6829                         u32 csum = ~(u32)0;
6830                         u64 private = ~(u32)0;
6831                         unsigned long flags;
6832
6833                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6834                                               start, &private))
6835                                 goto failed;
6836                         local_irq_save(flags);
6837                         kaddr = kmap_atomic(page);
6838                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6839                                                csum, bvec->bv_len);
6840                         btrfs_csum_final(csum, (char *)&csum);
6841                         kunmap_atomic(kaddr);
6842                         local_irq_restore(flags);
6843
6844                         flush_dcache_page(bvec->bv_page);
6845                         if (csum != private) {
6846 failed:
6847                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6848                                         (unsigned long long)btrfs_ino(inode),
6849                                         (unsigned long long)start,
6850                                         csum, (unsigned)private);
6851                                 err = -EIO;
6852                         }
6853                 }
6854
6855                 start += bvec->bv_len;
6856                 bvec++;
6857         } while (bvec <= bvec_end);
6858
6859         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6860                       dip->logical_offset + dip->bytes - 1);
6861         dio_bio = dip->dio_bio;
6862
6863         kfree(dip);
6864
6865         /* If we had a csum failure make sure to clear the uptodate flag */
6866         if (err)
6867                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6868         dio_end_io(dio_bio, err);
6869         bio_put(bio);
6870 }
6871
6872 static void btrfs_endio_direct_write(struct bio *bio, int err)
6873 {
6874         struct btrfs_dio_private *dip = bio->bi_private;
6875         struct inode *inode = dip->inode;
6876         struct btrfs_root *root = BTRFS_I(inode)->root;
6877         struct btrfs_ordered_extent *ordered = NULL;
6878         u64 ordered_offset = dip->logical_offset;
6879         u64 ordered_bytes = dip->bytes;
6880         struct bio *dio_bio;
6881         int ret;
6882
6883         if (err)
6884                 goto out_done;
6885 again:
6886         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6887                                                    &ordered_offset,
6888                                                    ordered_bytes, !err);
6889         if (!ret)
6890                 goto out_test;
6891
6892         ordered->work.func = finish_ordered_fn;
6893         ordered->work.flags = 0;
6894         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6895                            &ordered->work);
6896 out_test:
6897         /*
6898          * our bio might span multiple ordered extents.  If we haven't
6899          * completed the accounting for the whole dio, go back and try again
6900          */
6901         if (ordered_offset < dip->logical_offset + dip->bytes) {
6902                 ordered_bytes = dip->logical_offset + dip->bytes -
6903                         ordered_offset;
6904                 ordered = NULL;
6905                 goto again;
6906         }
6907 out_done:
6908         dio_bio = dip->dio_bio;
6909
6910         kfree(dip);
6911
6912         /* If we had an error make sure to clear the uptodate flag */
6913         if (err)
6914                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6915         dio_end_io(dio_bio, err);
6916         bio_put(bio);
6917 }
6918
6919 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6920                                     struct bio *bio, int mirror_num,
6921                                     unsigned long bio_flags, u64 offset)
6922 {
6923         int ret;
6924         struct btrfs_root *root = BTRFS_I(inode)->root;
6925         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6926         BUG_ON(ret); /* -ENOMEM */
6927         return 0;
6928 }
6929
6930 static void btrfs_end_dio_bio(struct bio *bio, int err)
6931 {
6932         struct btrfs_dio_private *dip = bio->bi_private;
6933
6934         if (err) {
6935                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6936                       "sector %#Lx len %u err no %d\n",
6937                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6938                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6939                 dip->errors = 1;
6940
6941                 /*
6942                  * before atomic variable goto zero, we must make sure
6943                  * dip->errors is perceived to be set.
6944                  */
6945                 smp_mb__before_atomic_dec();
6946         }
6947
6948         /* if there are more bios still pending for this dio, just exit */
6949         if (!atomic_dec_and_test(&dip->pending_bios))
6950                 goto out;
6951
6952         if (dip->errors) {
6953                 bio_io_error(dip->orig_bio);
6954         } else {
6955                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6956                 bio_endio(dip->orig_bio, 0);
6957         }
6958 out:
6959         bio_put(bio);
6960 }
6961
6962 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6963                                        u64 first_sector, gfp_t gfp_flags)
6964 {
6965         int nr_vecs = bio_get_nr_vecs(bdev);
6966         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6967 }
6968
6969 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6970                                          int rw, u64 file_offset, int skip_sum,
6971                                          int async_submit)
6972 {
6973         int write = rw & REQ_WRITE;
6974         struct btrfs_root *root = BTRFS_I(inode)->root;
6975         int ret;
6976
6977         if (async_submit)
6978                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6979
6980         bio_get(bio);
6981
6982         if (!write) {
6983                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6984                 if (ret)
6985                         goto err;
6986         }
6987
6988         if (skip_sum)
6989                 goto map;
6990
6991         if (write && async_submit) {
6992                 ret = btrfs_wq_submit_bio(root->fs_info,
6993                                    inode, rw, bio, 0, 0,
6994                                    file_offset,
6995                                    __btrfs_submit_bio_start_direct_io,
6996                                    __btrfs_submit_bio_done);
6997                 goto err;
6998         } else if (write) {
6999                 /*
7000                  * If we aren't doing async submit, calculate the csum of the
7001                  * bio now.
7002                  */
7003                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7004                 if (ret)
7005                         goto err;
7006         } else if (!skip_sum) {
7007                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7008                 if (ret)
7009                         goto err;
7010         }
7011
7012 map:
7013         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7014 err:
7015         bio_put(bio);
7016         return ret;
7017 }
7018
7019 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7020                                     int skip_sum)
7021 {
7022         struct inode *inode = dip->inode;
7023         struct btrfs_root *root = BTRFS_I(inode)->root;
7024         struct bio *bio;
7025         struct bio *orig_bio = dip->orig_bio;
7026         struct bio_vec *bvec = orig_bio->bi_io_vec;
7027         u64 start_sector = orig_bio->bi_sector;
7028         u64 file_offset = dip->logical_offset;
7029         u64 submit_len = 0;
7030         u64 map_length;
7031         int nr_pages = 0;
7032         int ret = 0;
7033         int async_submit = 0;
7034
7035         map_length = orig_bio->bi_size;
7036         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7037                               &map_length, NULL, 0);
7038         if (ret) {
7039                 bio_put(orig_bio);
7040                 return -EIO;
7041         }
7042         if (map_length >= orig_bio->bi_size) {
7043                 bio = orig_bio;
7044                 goto submit;
7045         }
7046
7047         /* async crcs make it difficult to collect full stripe writes. */
7048         if (btrfs_get_alloc_profile(root, 1) &
7049             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7050                 async_submit = 0;
7051         else
7052                 async_submit = 1;
7053
7054         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7055         if (!bio)
7056                 return -ENOMEM;
7057         bio->bi_private = dip;
7058         bio->bi_end_io = btrfs_end_dio_bio;
7059         atomic_inc(&dip->pending_bios);
7060
7061         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7062                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7063                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7064                                  bvec->bv_offset) < bvec->bv_len)) {
7065                         /*
7066                          * inc the count before we submit the bio so
7067                          * we know the end IO handler won't happen before
7068                          * we inc the count. Otherwise, the dip might get freed
7069                          * before we're done setting it up
7070                          */
7071                         atomic_inc(&dip->pending_bios);
7072                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7073                                                      file_offset, skip_sum,
7074                                                      async_submit);
7075                         if (ret) {
7076                                 bio_put(bio);
7077                                 atomic_dec(&dip->pending_bios);
7078                                 goto out_err;
7079                         }
7080
7081                         start_sector += submit_len >> 9;
7082                         file_offset += submit_len;
7083
7084                         submit_len = 0;
7085                         nr_pages = 0;
7086
7087                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7088                                                   start_sector, GFP_NOFS);
7089                         if (!bio)
7090                                 goto out_err;
7091                         bio->bi_private = dip;
7092                         bio->bi_end_io = btrfs_end_dio_bio;
7093
7094                         map_length = orig_bio->bi_size;
7095                         ret = btrfs_map_block(root->fs_info, rw,
7096                                               start_sector << 9,
7097                                               &map_length, NULL, 0);
7098                         if (ret) {
7099                                 bio_put(bio);
7100                                 goto out_err;
7101                         }
7102                 } else {
7103                         submit_len += bvec->bv_len;
7104                         nr_pages ++;
7105                         bvec++;
7106                 }
7107         }
7108
7109 submit:
7110         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7111                                      async_submit);
7112         if (!ret)
7113                 return 0;
7114
7115         bio_put(bio);
7116 out_err:
7117         dip->errors = 1;
7118         /*
7119          * before atomic variable goto zero, we must
7120          * make sure dip->errors is perceived to be set.
7121          */
7122         smp_mb__before_atomic_dec();
7123         if (atomic_dec_and_test(&dip->pending_bios))
7124                 bio_io_error(dip->orig_bio);
7125
7126         /* bio_end_io() will handle error, so we needn't return it */
7127         return 0;
7128 }
7129
7130 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7131                                 struct inode *inode, loff_t file_offset)
7132 {
7133         struct btrfs_root *root = BTRFS_I(inode)->root;
7134         struct btrfs_dio_private *dip;
7135         struct bio *io_bio;
7136         int skip_sum;
7137         int write = rw & REQ_WRITE;
7138         int ret = 0;
7139
7140         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7141
7142         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7143
7144         if (!io_bio) {
7145                 ret = -ENOMEM;
7146                 goto free_ordered;
7147         }
7148
7149         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7150         if (!dip) {
7151                 ret = -ENOMEM;
7152                 goto free_io_bio;
7153         }
7154
7155         dip->private = dio_bio->bi_private;
7156         dip->inode = inode;
7157         dip->logical_offset = file_offset;
7158         dip->bytes = dio_bio->bi_size;
7159         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7160         io_bio->bi_private = dip;
7161         dip->errors = 0;
7162         dip->orig_bio = io_bio;
7163         dip->dio_bio = dio_bio;
7164         atomic_set(&dip->pending_bios, 0);
7165
7166         if (write)
7167                 io_bio->bi_end_io = btrfs_endio_direct_write;
7168         else
7169                 io_bio->bi_end_io = btrfs_endio_direct_read;
7170
7171         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7172         if (!ret)
7173                 return;
7174
7175 free_io_bio:
7176         bio_put(io_bio);
7177
7178 free_ordered:
7179         /*
7180          * If this is a write, we need to clean up the reserved space and kill
7181          * the ordered extent.
7182          */
7183         if (write) {
7184                 struct btrfs_ordered_extent *ordered;
7185                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7186                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7187                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7188                         btrfs_free_reserved_extent(root, ordered->start,
7189                                                    ordered->disk_len);
7190                 btrfs_put_ordered_extent(ordered);
7191                 btrfs_put_ordered_extent(ordered);
7192         }
7193         bio_endio(dio_bio, ret);
7194 }
7195
7196 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7197                         const struct iovec *iov, loff_t offset,
7198                         unsigned long nr_segs)
7199 {
7200         int seg;
7201         int i;
7202         size_t size;
7203         unsigned long addr;
7204         unsigned blocksize_mask = root->sectorsize - 1;
7205         ssize_t retval = -EINVAL;
7206         loff_t end = offset;
7207
7208         if (offset & blocksize_mask)
7209                 goto out;
7210
7211         /* Check the memory alignment.  Blocks cannot straddle pages */
7212         for (seg = 0; seg < nr_segs; seg++) {
7213                 addr = (unsigned long)iov[seg].iov_base;
7214                 size = iov[seg].iov_len;
7215                 end += size;
7216                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7217                         goto out;
7218
7219                 /* If this is a write we don't need to check anymore */
7220                 if (rw & WRITE)
7221                         continue;
7222
7223                 /*
7224                  * Check to make sure we don't have duplicate iov_base's in this
7225                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7226                  * when reading back.
7227                  */
7228                 for (i = seg + 1; i < nr_segs; i++) {
7229                         if (iov[seg].iov_base == iov[i].iov_base)
7230                                 goto out;
7231                 }
7232         }
7233         retval = 0;
7234 out:
7235         return retval;
7236 }
7237
7238 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7239                         const struct iovec *iov, loff_t offset,
7240                         unsigned long nr_segs)
7241 {
7242         struct file *file = iocb->ki_filp;
7243         struct inode *inode = file->f_mapping->host;
7244         size_t count = 0;
7245         int flags = 0;
7246         bool wakeup = true;
7247         bool relock = false;
7248         ssize_t ret;
7249
7250         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7251                             offset, nr_segs))
7252                 return 0;
7253
7254         atomic_inc(&inode->i_dio_count);
7255         smp_mb__after_atomic_inc();
7256
7257         /*
7258          * The generic stuff only does filemap_write_and_wait_range, which isn't
7259          * enough if we've written compressed pages to this area, so we need to
7260          * call btrfs_wait_ordered_range to make absolutely sure that any
7261          * outstanding dirty pages are on disk.
7262          */
7263         count = iov_length(iov, nr_segs);
7264         btrfs_wait_ordered_range(inode, offset, count);
7265
7266         if (rw & WRITE) {
7267                 /*
7268                  * If the write DIO is beyond the EOF, we need update
7269                  * the isize, but it is protected by i_mutex. So we can
7270                  * not unlock the i_mutex at this case.
7271                  */
7272                 if (offset + count <= inode->i_size) {
7273                         mutex_unlock(&inode->i_mutex);
7274                         relock = true;
7275                 }
7276                 ret = btrfs_delalloc_reserve_space(inode, count);
7277                 if (ret)
7278                         goto out;
7279         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7280                                      &BTRFS_I(inode)->runtime_flags))) {
7281                 inode_dio_done(inode);
7282                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7283                 wakeup = false;
7284         }
7285
7286         ret = __blockdev_direct_IO(rw, iocb, inode,
7287                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7288                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7289                         btrfs_submit_direct, flags);
7290         if (rw & WRITE) {
7291                 if (ret < 0 && ret != -EIOCBQUEUED)
7292                         btrfs_delalloc_release_space(inode, count);
7293                 else if (ret >= 0 && (size_t)ret < count)
7294                         btrfs_delalloc_release_space(inode,
7295                                                      count - (size_t)ret);
7296                 else
7297                         btrfs_delalloc_release_metadata(inode, 0);
7298         }
7299 out:
7300         if (wakeup)
7301                 inode_dio_done(inode);
7302         if (relock)
7303                 mutex_lock(&inode->i_mutex);
7304
7305         return ret;
7306 }
7307
7308 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7309
7310 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7311                 __u64 start, __u64 len)
7312 {
7313         int     ret;
7314
7315         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7316         if (ret)
7317                 return ret;
7318
7319         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7320 }
7321
7322 int btrfs_readpage(struct file *file, struct page *page)
7323 {
7324         struct extent_io_tree *tree;
7325         tree = &BTRFS_I(page->mapping->host)->io_tree;
7326         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7327 }
7328
7329 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7330 {
7331         struct extent_io_tree *tree;
7332
7333
7334         if (current->flags & PF_MEMALLOC) {
7335                 redirty_page_for_writepage(wbc, page);
7336                 unlock_page(page);
7337                 return 0;
7338         }
7339         tree = &BTRFS_I(page->mapping->host)->io_tree;
7340         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7341 }
7342
7343 static int btrfs_writepages(struct address_space *mapping,
7344                             struct writeback_control *wbc)
7345 {
7346         struct extent_io_tree *tree;
7347
7348         tree = &BTRFS_I(mapping->host)->io_tree;
7349         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7350 }
7351
7352 static int
7353 btrfs_readpages(struct file *file, struct address_space *mapping,
7354                 struct list_head *pages, unsigned nr_pages)
7355 {
7356         struct extent_io_tree *tree;
7357         tree = &BTRFS_I(mapping->host)->io_tree;
7358         return extent_readpages(tree, mapping, pages, nr_pages,
7359                                 btrfs_get_extent);
7360 }
7361 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7362 {
7363         struct extent_io_tree *tree;
7364         struct extent_map_tree *map;
7365         int ret;
7366
7367         tree = &BTRFS_I(page->mapping->host)->io_tree;
7368         map = &BTRFS_I(page->mapping->host)->extent_tree;
7369         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7370         if (ret == 1) {
7371                 ClearPagePrivate(page);
7372                 set_page_private(page, 0);
7373                 page_cache_release(page);
7374         }
7375         return ret;
7376 }
7377
7378 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7379 {
7380         if (PageWriteback(page) || PageDirty(page))
7381                 return 0;
7382         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7383 }
7384
7385 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7386                                  unsigned int length)
7387 {
7388         struct inode *inode = page->mapping->host;
7389         struct extent_io_tree *tree;
7390         struct btrfs_ordered_extent *ordered;
7391         struct extent_state *cached_state = NULL;
7392         u64 page_start = page_offset(page);
7393         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7394
7395         /*
7396          * we have the page locked, so new writeback can't start,
7397          * and the dirty bit won't be cleared while we are here.
7398          *
7399          * Wait for IO on this page so that we can safely clear
7400          * the PagePrivate2 bit and do ordered accounting
7401          */
7402         wait_on_page_writeback(page);
7403
7404         tree = &BTRFS_I(inode)->io_tree;
7405         if (offset) {
7406                 btrfs_releasepage(page, GFP_NOFS);
7407                 return;
7408         }
7409         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7410         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7411         if (ordered) {
7412                 /*
7413                  * IO on this page will never be started, so we need
7414                  * to account for any ordered extents now
7415                  */
7416                 clear_extent_bit(tree, page_start, page_end,
7417                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7418                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7419                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7420                 /*
7421                  * whoever cleared the private bit is responsible
7422                  * for the finish_ordered_io
7423                  */
7424                 if (TestClearPagePrivate2(page) &&
7425                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7426                                                    PAGE_CACHE_SIZE, 1)) {
7427                         btrfs_finish_ordered_io(ordered);
7428                 }
7429                 btrfs_put_ordered_extent(ordered);
7430                 cached_state = NULL;
7431                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7432         }
7433         clear_extent_bit(tree, page_start, page_end,
7434                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7435                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7436                  &cached_state, GFP_NOFS);
7437         __btrfs_releasepage(page, GFP_NOFS);
7438
7439         ClearPageChecked(page);
7440         if (PagePrivate(page)) {
7441                 ClearPagePrivate(page);
7442                 set_page_private(page, 0);
7443                 page_cache_release(page);
7444         }
7445 }
7446
7447 /*
7448  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7449  * called from a page fault handler when a page is first dirtied. Hence we must
7450  * be careful to check for EOF conditions here. We set the page up correctly
7451  * for a written page which means we get ENOSPC checking when writing into
7452  * holes and correct delalloc and unwritten extent mapping on filesystems that
7453  * support these features.
7454  *
7455  * We are not allowed to take the i_mutex here so we have to play games to
7456  * protect against truncate races as the page could now be beyond EOF.  Because
7457  * vmtruncate() writes the inode size before removing pages, once we have the
7458  * page lock we can determine safely if the page is beyond EOF. If it is not
7459  * beyond EOF, then the page is guaranteed safe against truncation until we
7460  * unlock the page.
7461  */
7462 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7463 {
7464         struct page *page = vmf->page;
7465         struct inode *inode = file_inode(vma->vm_file);
7466         struct btrfs_root *root = BTRFS_I(inode)->root;
7467         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7468         struct btrfs_ordered_extent *ordered;
7469         struct extent_state *cached_state = NULL;
7470         char *kaddr;
7471         unsigned long zero_start;
7472         loff_t size;
7473         int ret;
7474         int reserved = 0;
7475         u64 page_start;
7476         u64 page_end;
7477
7478         sb_start_pagefault(inode->i_sb);
7479         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7480         if (!ret) {
7481                 ret = file_update_time(vma->vm_file);
7482                 reserved = 1;
7483         }
7484         if (ret) {
7485                 if (ret == -ENOMEM)
7486                         ret = VM_FAULT_OOM;
7487                 else /* -ENOSPC, -EIO, etc */
7488                         ret = VM_FAULT_SIGBUS;
7489                 if (reserved)
7490                         goto out;
7491                 goto out_noreserve;
7492         }
7493
7494         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7495 again:
7496         lock_page(page);
7497         size = i_size_read(inode);
7498         page_start = page_offset(page);
7499         page_end = page_start + PAGE_CACHE_SIZE - 1;
7500
7501         if ((page->mapping != inode->i_mapping) ||
7502             (page_start >= size)) {
7503                 /* page got truncated out from underneath us */
7504                 goto out_unlock;
7505         }
7506         wait_on_page_writeback(page);
7507
7508         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7509         set_page_extent_mapped(page);
7510
7511         /*
7512          * we can't set the delalloc bits if there are pending ordered
7513          * extents.  Drop our locks and wait for them to finish
7514          */
7515         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7516         if (ordered) {
7517                 unlock_extent_cached(io_tree, page_start, page_end,
7518                                      &cached_state, GFP_NOFS);
7519                 unlock_page(page);
7520                 btrfs_start_ordered_extent(inode, ordered, 1);
7521                 btrfs_put_ordered_extent(ordered);
7522                 goto again;
7523         }
7524
7525         /*
7526          * XXX - page_mkwrite gets called every time the page is dirtied, even
7527          * if it was already dirty, so for space accounting reasons we need to
7528          * clear any delalloc bits for the range we are fixing to save.  There
7529          * is probably a better way to do this, but for now keep consistent with
7530          * prepare_pages in the normal write path.
7531          */
7532         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7533                           EXTENT_DIRTY | EXTENT_DELALLOC |
7534                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7535                           0, 0, &cached_state, GFP_NOFS);
7536
7537         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7538                                         &cached_state);
7539         if (ret) {
7540                 unlock_extent_cached(io_tree, page_start, page_end,
7541                                      &cached_state, GFP_NOFS);
7542                 ret = VM_FAULT_SIGBUS;
7543                 goto out_unlock;
7544         }
7545         ret = 0;
7546
7547         /* page is wholly or partially inside EOF */
7548         if (page_start + PAGE_CACHE_SIZE > size)
7549                 zero_start = size & ~PAGE_CACHE_MASK;
7550         else
7551                 zero_start = PAGE_CACHE_SIZE;
7552
7553         if (zero_start != PAGE_CACHE_SIZE) {
7554                 kaddr = kmap(page);
7555                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7556                 flush_dcache_page(page);
7557                 kunmap(page);
7558         }
7559         ClearPageChecked(page);
7560         set_page_dirty(page);
7561         SetPageUptodate(page);
7562
7563         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7564         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7565         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7566
7567         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7568
7569 out_unlock:
7570         if (!ret) {
7571                 sb_end_pagefault(inode->i_sb);
7572                 return VM_FAULT_LOCKED;
7573         }
7574         unlock_page(page);
7575 out:
7576         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7577 out_noreserve:
7578         sb_end_pagefault(inode->i_sb);
7579         return ret;
7580 }
7581
7582 static int btrfs_truncate(struct inode *inode)
7583 {
7584         struct btrfs_root *root = BTRFS_I(inode)->root;
7585         struct btrfs_block_rsv *rsv;
7586         int ret = 0;
7587         int err = 0;
7588         struct btrfs_trans_handle *trans;
7589         u64 mask = root->sectorsize - 1;
7590         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7591
7592         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7593         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7594
7595         /*
7596          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7597          * 3 things going on here
7598          *
7599          * 1) We need to reserve space for our orphan item and the space to
7600          * delete our orphan item.  Lord knows we don't want to have a dangling
7601          * orphan item because we didn't reserve space to remove it.
7602          *
7603          * 2) We need to reserve space to update our inode.
7604          *
7605          * 3) We need to have something to cache all the space that is going to
7606          * be free'd up by the truncate operation, but also have some slack
7607          * space reserved in case it uses space during the truncate (thank you
7608          * very much snapshotting).
7609          *
7610          * And we need these to all be seperate.  The fact is we can use alot of
7611          * space doing the truncate, and we have no earthly idea how much space
7612          * we will use, so we need the truncate reservation to be seperate so it
7613          * doesn't end up using space reserved for updating the inode or
7614          * removing the orphan item.  We also need to be able to stop the
7615          * transaction and start a new one, which means we need to be able to
7616          * update the inode several times, and we have no idea of knowing how
7617          * many times that will be, so we can't just reserve 1 item for the
7618          * entirety of the opration, so that has to be done seperately as well.
7619          * Then there is the orphan item, which does indeed need to be held on
7620          * to for the whole operation, and we need nobody to touch this reserved
7621          * space except the orphan code.
7622          *
7623          * So that leaves us with
7624          *
7625          * 1) root->orphan_block_rsv - for the orphan deletion.
7626          * 2) rsv - for the truncate reservation, which we will steal from the
7627          * transaction reservation.
7628          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7629          * updating the inode.
7630          */
7631         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7632         if (!rsv)
7633                 return -ENOMEM;
7634         rsv->size = min_size;
7635         rsv->failfast = 1;
7636
7637         /*
7638          * 1 for the truncate slack space
7639          * 1 for updating the inode.
7640          */
7641         trans = btrfs_start_transaction(root, 2);
7642         if (IS_ERR(trans)) {
7643                 err = PTR_ERR(trans);
7644                 goto out;
7645         }
7646
7647         /* Migrate the slack space for the truncate to our reserve */
7648         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7649                                       min_size);
7650         BUG_ON(ret);
7651
7652         /*
7653          * setattr is responsible for setting the ordered_data_close flag,
7654          * but that is only tested during the last file release.  That
7655          * could happen well after the next commit, leaving a great big
7656          * window where new writes may get lost if someone chooses to write
7657          * to this file after truncating to zero
7658          *
7659          * The inode doesn't have any dirty data here, and so if we commit
7660          * this is a noop.  If someone immediately starts writing to the inode
7661          * it is very likely we'll catch some of their writes in this
7662          * transaction, and the commit will find this file on the ordered
7663          * data list with good things to send down.
7664          *
7665          * This is a best effort solution, there is still a window where
7666          * using truncate to replace the contents of the file will
7667          * end up with a zero length file after a crash.
7668          */
7669         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7670                                            &BTRFS_I(inode)->runtime_flags))
7671                 btrfs_add_ordered_operation(trans, root, inode);
7672
7673         /*
7674          * So if we truncate and then write and fsync we normally would just
7675          * write the extents that changed, which is a problem if we need to
7676          * first truncate that entire inode.  So set this flag so we write out
7677          * all of the extents in the inode to the sync log so we're completely
7678          * safe.
7679          */
7680         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7681         trans->block_rsv = rsv;
7682
7683         while (1) {
7684                 ret = btrfs_truncate_inode_items(trans, root, inode,
7685                                                  inode->i_size,
7686                                                  BTRFS_EXTENT_DATA_KEY);
7687                 if (ret != -ENOSPC) {
7688                         err = ret;
7689                         break;
7690                 }
7691
7692                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7693                 ret = btrfs_update_inode(trans, root, inode);
7694                 if (ret) {
7695                         err = ret;
7696                         break;
7697                 }
7698
7699                 btrfs_end_transaction(trans, root);
7700                 btrfs_btree_balance_dirty(root);
7701
7702                 trans = btrfs_start_transaction(root, 2);
7703                 if (IS_ERR(trans)) {
7704                         ret = err = PTR_ERR(trans);
7705                         trans = NULL;
7706                         break;
7707                 }
7708
7709                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7710                                               rsv, min_size);
7711                 BUG_ON(ret);    /* shouldn't happen */
7712                 trans->block_rsv = rsv;
7713         }
7714
7715         if (ret == 0 && inode->i_nlink > 0) {
7716                 trans->block_rsv = root->orphan_block_rsv;
7717                 ret = btrfs_orphan_del(trans, inode);
7718                 if (ret)
7719                         err = ret;
7720         }
7721
7722         if (trans) {
7723                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7724                 ret = btrfs_update_inode(trans, root, inode);
7725                 if (ret && !err)
7726                         err = ret;
7727
7728                 ret = btrfs_end_transaction(trans, root);
7729                 btrfs_btree_balance_dirty(root);
7730         }
7731
7732 out:
7733         btrfs_free_block_rsv(root, rsv);
7734
7735         if (ret && !err)
7736                 err = ret;
7737
7738         return err;
7739 }
7740
7741 /*
7742  * create a new subvolume directory/inode (helper for the ioctl).
7743  */
7744 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7745                              struct btrfs_root *new_root, u64 new_dirid)
7746 {
7747         struct inode *inode;
7748         int err;
7749         u64 index = 0;
7750
7751         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7752                                 new_dirid, new_dirid,
7753                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7754                                 &index);
7755         if (IS_ERR(inode))
7756                 return PTR_ERR(inode);
7757         inode->i_op = &btrfs_dir_inode_operations;
7758         inode->i_fop = &btrfs_dir_file_operations;
7759
7760         set_nlink(inode, 1);
7761         btrfs_i_size_write(inode, 0);
7762
7763         err = btrfs_update_inode(trans, new_root, inode);
7764
7765         iput(inode);
7766         return err;
7767 }
7768
7769 struct inode *btrfs_alloc_inode(struct super_block *sb)
7770 {
7771         struct btrfs_inode *ei;
7772         struct inode *inode;
7773
7774         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7775         if (!ei)
7776                 return NULL;
7777
7778         ei->root = NULL;
7779         ei->generation = 0;
7780         ei->last_trans = 0;
7781         ei->last_sub_trans = 0;
7782         ei->logged_trans = 0;
7783         ei->delalloc_bytes = 0;
7784         ei->disk_i_size = 0;
7785         ei->flags = 0;
7786         ei->csum_bytes = 0;
7787         ei->index_cnt = (u64)-1;
7788         ei->last_unlink_trans = 0;
7789         ei->last_log_commit = 0;
7790
7791         spin_lock_init(&ei->lock);
7792         ei->outstanding_extents = 0;
7793         ei->reserved_extents = 0;
7794
7795         ei->runtime_flags = 0;
7796         ei->force_compress = BTRFS_COMPRESS_NONE;
7797
7798         ei->delayed_node = NULL;
7799
7800         inode = &ei->vfs_inode;
7801         extent_map_tree_init(&ei->extent_tree);
7802         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7803         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7804         ei->io_tree.track_uptodate = 1;
7805         ei->io_failure_tree.track_uptodate = 1;
7806         atomic_set(&ei->sync_writers, 0);
7807         mutex_init(&ei->log_mutex);
7808         mutex_init(&ei->delalloc_mutex);
7809         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7810         INIT_LIST_HEAD(&ei->delalloc_inodes);
7811         INIT_LIST_HEAD(&ei->ordered_operations);
7812         RB_CLEAR_NODE(&ei->rb_node);
7813
7814         return inode;
7815 }
7816
7817 static void btrfs_i_callback(struct rcu_head *head)
7818 {
7819         struct inode *inode = container_of(head, struct inode, i_rcu);
7820         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7821 }
7822
7823 void btrfs_destroy_inode(struct inode *inode)
7824 {
7825         struct btrfs_ordered_extent *ordered;
7826         struct btrfs_root *root = BTRFS_I(inode)->root;
7827
7828         WARN_ON(!hlist_empty(&inode->i_dentry));
7829         WARN_ON(inode->i_data.nrpages);
7830         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7831         WARN_ON(BTRFS_I(inode)->reserved_extents);
7832         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7833         WARN_ON(BTRFS_I(inode)->csum_bytes);
7834
7835         /*
7836          * This can happen where we create an inode, but somebody else also
7837          * created the same inode and we need to destroy the one we already
7838          * created.
7839          */
7840         if (!root)
7841                 goto free;
7842
7843         /*
7844          * Make sure we're properly removed from the ordered operation
7845          * lists.
7846          */
7847         smp_mb();
7848         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7849                 spin_lock(&root->fs_info->ordered_root_lock);
7850                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7851                 spin_unlock(&root->fs_info->ordered_root_lock);
7852         }
7853
7854         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7855                      &BTRFS_I(inode)->runtime_flags)) {
7856                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7857                         (unsigned long long)btrfs_ino(inode));
7858                 atomic_dec(&root->orphan_inodes);
7859         }
7860
7861         while (1) {
7862                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7863                 if (!ordered)
7864                         break;
7865                 else {
7866                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7867                                 (unsigned long long)ordered->file_offset,
7868                                 (unsigned long long)ordered->len);
7869                         btrfs_remove_ordered_extent(inode, ordered);
7870                         btrfs_put_ordered_extent(ordered);
7871                         btrfs_put_ordered_extent(ordered);
7872                 }
7873         }
7874         inode_tree_del(inode);
7875         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7876 free:
7877         call_rcu(&inode->i_rcu, btrfs_i_callback);
7878 }
7879
7880 int btrfs_drop_inode(struct inode *inode)
7881 {
7882         struct btrfs_root *root = BTRFS_I(inode)->root;
7883
7884         if (root == NULL)
7885                 return 1;
7886
7887         /* the snap/subvol tree is on deleting */
7888         if (btrfs_root_refs(&root->root_item) == 0 &&
7889             root != root->fs_info->tree_root)
7890                 return 1;
7891         else
7892                 return generic_drop_inode(inode);
7893 }
7894
7895 static void init_once(void *foo)
7896 {
7897         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7898
7899         inode_init_once(&ei->vfs_inode);
7900 }
7901
7902 void btrfs_destroy_cachep(void)
7903 {
7904         /*
7905          * Make sure all delayed rcu free inodes are flushed before we
7906          * destroy cache.
7907          */
7908         rcu_barrier();
7909         if (btrfs_inode_cachep)
7910                 kmem_cache_destroy(btrfs_inode_cachep);
7911         if (btrfs_trans_handle_cachep)
7912                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7913         if (btrfs_transaction_cachep)
7914                 kmem_cache_destroy(btrfs_transaction_cachep);
7915         if (btrfs_path_cachep)
7916                 kmem_cache_destroy(btrfs_path_cachep);
7917         if (btrfs_free_space_cachep)
7918                 kmem_cache_destroy(btrfs_free_space_cachep);
7919         if (btrfs_delalloc_work_cachep)
7920                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7921 }
7922
7923 int btrfs_init_cachep(void)
7924 {
7925         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7926                         sizeof(struct btrfs_inode), 0,
7927                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7928         if (!btrfs_inode_cachep)
7929                 goto fail;
7930
7931         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7932                         sizeof(struct btrfs_trans_handle), 0,
7933                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7934         if (!btrfs_trans_handle_cachep)
7935                 goto fail;
7936
7937         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7938                         sizeof(struct btrfs_transaction), 0,
7939                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7940         if (!btrfs_transaction_cachep)
7941                 goto fail;
7942
7943         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7944                         sizeof(struct btrfs_path), 0,
7945                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7946         if (!btrfs_path_cachep)
7947                 goto fail;
7948
7949         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7950                         sizeof(struct btrfs_free_space), 0,
7951                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7952         if (!btrfs_free_space_cachep)
7953                 goto fail;
7954
7955         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7956                         sizeof(struct btrfs_delalloc_work), 0,
7957                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7958                         NULL);
7959         if (!btrfs_delalloc_work_cachep)
7960                 goto fail;
7961
7962         return 0;
7963 fail:
7964         btrfs_destroy_cachep();
7965         return -ENOMEM;
7966 }
7967
7968 static int btrfs_getattr(struct vfsmount *mnt,
7969                          struct dentry *dentry, struct kstat *stat)
7970 {
7971         u64 delalloc_bytes;
7972         struct inode *inode = dentry->d_inode;
7973         u32 blocksize = inode->i_sb->s_blocksize;
7974
7975         generic_fillattr(inode, stat);
7976         stat->dev = BTRFS_I(inode)->root->anon_dev;
7977         stat->blksize = PAGE_CACHE_SIZE;
7978
7979         spin_lock(&BTRFS_I(inode)->lock);
7980         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7981         spin_unlock(&BTRFS_I(inode)->lock);
7982         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7983                         ALIGN(delalloc_bytes, blocksize)) >> 9;
7984         return 0;
7985 }
7986
7987 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7988                            struct inode *new_dir, struct dentry *new_dentry)
7989 {
7990         struct btrfs_trans_handle *trans;
7991         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7992         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7993         struct inode *new_inode = new_dentry->d_inode;
7994         struct inode *old_inode = old_dentry->d_inode;
7995         struct timespec ctime = CURRENT_TIME;
7996         u64 index = 0;
7997         u64 root_objectid;
7998         int ret;
7999         u64 old_ino = btrfs_ino(old_inode);
8000
8001         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8002                 return -EPERM;
8003
8004         /* we only allow rename subvolume link between subvolumes */
8005         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8006                 return -EXDEV;
8007
8008         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8009             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8010                 return -ENOTEMPTY;
8011
8012         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8013             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8014                 return -ENOTEMPTY;
8015
8016
8017         /* check for collisions, even if the  name isn't there */
8018         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8019                              new_dentry->d_name.name,
8020                              new_dentry->d_name.len);
8021
8022         if (ret) {
8023                 if (ret == -EEXIST) {
8024                         /* we shouldn't get
8025                          * eexist without a new_inode */
8026                         if (!new_inode) {
8027                                 WARN_ON(1);
8028                                 return ret;
8029                         }
8030                 } else {
8031                         /* maybe -EOVERFLOW */
8032                         return ret;
8033                 }
8034         }
8035         ret = 0;
8036
8037         /*
8038          * we're using rename to replace one file with another.
8039          * and the replacement file is large.  Start IO on it now so
8040          * we don't add too much work to the end of the transaction
8041          */
8042         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8043             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8044                 filemap_flush(old_inode->i_mapping);
8045
8046         /* close the racy window with snapshot create/destroy ioctl */
8047         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8048                 down_read(&root->fs_info->subvol_sem);
8049         /*
8050          * We want to reserve the absolute worst case amount of items.  So if
8051          * both inodes are subvols and we need to unlink them then that would
8052          * require 4 item modifications, but if they are both normal inodes it
8053          * would require 5 item modifications, so we'll assume their normal
8054          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8055          * should cover the worst case number of items we'll modify.
8056          */
8057         trans = btrfs_start_transaction(root, 11);
8058         if (IS_ERR(trans)) {
8059                 ret = PTR_ERR(trans);
8060                 goto out_notrans;
8061         }
8062
8063         if (dest != root)
8064                 btrfs_record_root_in_trans(trans, dest);
8065
8066         ret = btrfs_set_inode_index(new_dir, &index);
8067         if (ret)
8068                 goto out_fail;
8069
8070         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8071                 /* force full log commit if subvolume involved. */
8072                 root->fs_info->last_trans_log_full_commit = trans->transid;
8073         } else {
8074                 ret = btrfs_insert_inode_ref(trans, dest,
8075                                              new_dentry->d_name.name,
8076                                              new_dentry->d_name.len,
8077                                              old_ino,
8078                                              btrfs_ino(new_dir), index);
8079                 if (ret)
8080                         goto out_fail;
8081                 /*
8082                  * this is an ugly little race, but the rename is required
8083                  * to make sure that if we crash, the inode is either at the
8084                  * old name or the new one.  pinning the log transaction lets
8085                  * us make sure we don't allow a log commit to come in after
8086                  * we unlink the name but before we add the new name back in.
8087                  */
8088                 btrfs_pin_log_trans(root);
8089         }
8090         /*
8091          * make sure the inode gets flushed if it is replacing
8092          * something.
8093          */
8094         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8095                 btrfs_add_ordered_operation(trans, root, old_inode);
8096
8097         inode_inc_iversion(old_dir);
8098         inode_inc_iversion(new_dir);
8099         inode_inc_iversion(old_inode);
8100         old_dir->i_ctime = old_dir->i_mtime = ctime;
8101         new_dir->i_ctime = new_dir->i_mtime = ctime;
8102         old_inode->i_ctime = ctime;
8103
8104         if (old_dentry->d_parent != new_dentry->d_parent)
8105                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8106
8107         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8108                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8109                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8110                                         old_dentry->d_name.name,
8111                                         old_dentry->d_name.len);
8112         } else {
8113                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8114                                         old_dentry->d_inode,
8115                                         old_dentry->d_name.name,
8116                                         old_dentry->d_name.len);
8117                 if (!ret)
8118                         ret = btrfs_update_inode(trans, root, old_inode);
8119         }
8120         if (ret) {
8121                 btrfs_abort_transaction(trans, root, ret);
8122                 goto out_fail;
8123         }
8124
8125         if (new_inode) {
8126                 inode_inc_iversion(new_inode);
8127                 new_inode->i_ctime = CURRENT_TIME;
8128                 if (unlikely(btrfs_ino(new_inode) ==
8129                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8130                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8131                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8132                                                 root_objectid,
8133                                                 new_dentry->d_name.name,
8134                                                 new_dentry->d_name.len);
8135                         BUG_ON(new_inode->i_nlink == 0);
8136                 } else {
8137                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8138                                                  new_dentry->d_inode,
8139                                                  new_dentry->d_name.name,
8140                                                  new_dentry->d_name.len);
8141                 }
8142                 if (!ret && new_inode->i_nlink == 0) {
8143                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8144                         BUG_ON(ret);
8145                 }
8146                 if (ret) {
8147                         btrfs_abort_transaction(trans, root, ret);
8148                         goto out_fail;
8149                 }
8150         }
8151
8152         ret = btrfs_add_link(trans, new_dir, old_inode,
8153                              new_dentry->d_name.name,
8154                              new_dentry->d_name.len, 0, index);
8155         if (ret) {
8156                 btrfs_abort_transaction(trans, root, ret);
8157                 goto out_fail;
8158         }
8159
8160         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8161                 struct dentry *parent = new_dentry->d_parent;
8162                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8163                 btrfs_end_log_trans(root);
8164         }
8165 out_fail:
8166         btrfs_end_transaction(trans, root);
8167 out_notrans:
8168         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8169                 up_read(&root->fs_info->subvol_sem);
8170
8171         return ret;
8172 }
8173
8174 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8175 {
8176         struct btrfs_delalloc_work *delalloc_work;
8177
8178         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8179                                      work);
8180         if (delalloc_work->wait)
8181                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8182         else
8183                 filemap_flush(delalloc_work->inode->i_mapping);
8184
8185         if (delalloc_work->delay_iput)
8186                 btrfs_add_delayed_iput(delalloc_work->inode);
8187         else
8188                 iput(delalloc_work->inode);
8189         complete(&delalloc_work->completion);
8190 }
8191
8192 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8193                                                     int wait, int delay_iput)
8194 {
8195         struct btrfs_delalloc_work *work;
8196
8197         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8198         if (!work)
8199                 return NULL;
8200
8201         init_completion(&work->completion);
8202         INIT_LIST_HEAD(&work->list);
8203         work->inode = inode;
8204         work->wait = wait;
8205         work->delay_iput = delay_iput;
8206         work->work.func = btrfs_run_delalloc_work;
8207
8208         return work;
8209 }
8210
8211 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8212 {
8213         wait_for_completion(&work->completion);
8214         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8215 }
8216
8217 /*
8218  * some fairly slow code that needs optimization. This walks the list
8219  * of all the inodes with pending delalloc and forces them to disk.
8220  */
8221 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8222 {
8223         struct btrfs_inode *binode;
8224         struct inode *inode;
8225         struct btrfs_delalloc_work *work, *next;
8226         struct list_head works;
8227         struct list_head splice;
8228         int ret = 0;
8229
8230         INIT_LIST_HEAD(&works);
8231         INIT_LIST_HEAD(&splice);
8232
8233         spin_lock(&root->delalloc_lock);
8234         list_splice_init(&root->delalloc_inodes, &splice);
8235         while (!list_empty(&splice)) {
8236                 binode = list_entry(splice.next, struct btrfs_inode,
8237                                     delalloc_inodes);
8238
8239                 list_move_tail(&binode->delalloc_inodes,
8240                                &root->delalloc_inodes);
8241                 inode = igrab(&binode->vfs_inode);
8242                 if (!inode) {
8243                         cond_resched_lock(&root->delalloc_lock);
8244                         continue;
8245                 }
8246                 spin_unlock(&root->delalloc_lock);
8247
8248                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8249                 if (unlikely(!work)) {
8250                         ret = -ENOMEM;
8251                         goto out;
8252                 }
8253                 list_add_tail(&work->list, &works);
8254                 btrfs_queue_worker(&root->fs_info->flush_workers,
8255                                    &work->work);
8256
8257                 cond_resched();
8258                 spin_lock(&root->delalloc_lock);
8259         }
8260         spin_unlock(&root->delalloc_lock);
8261
8262         list_for_each_entry_safe(work, next, &works, list) {
8263                 list_del_init(&work->list);
8264                 btrfs_wait_and_free_delalloc_work(work);
8265         }
8266         return 0;
8267 out:
8268         list_for_each_entry_safe(work, next, &works, list) {
8269                 list_del_init(&work->list);
8270                 btrfs_wait_and_free_delalloc_work(work);
8271         }
8272
8273         if (!list_empty_careful(&splice)) {
8274                 spin_lock(&root->delalloc_lock);
8275                 list_splice_tail(&splice, &root->delalloc_inodes);
8276                 spin_unlock(&root->delalloc_lock);
8277         }
8278         return ret;
8279 }
8280
8281 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8282 {
8283         int ret;
8284
8285         if (root->fs_info->sb->s_flags & MS_RDONLY)
8286                 return -EROFS;
8287
8288         ret = __start_delalloc_inodes(root, delay_iput);
8289         /*
8290          * the filemap_flush will queue IO into the worker threads, but
8291          * we have to make sure the IO is actually started and that
8292          * ordered extents get created before we return
8293          */
8294         atomic_inc(&root->fs_info->async_submit_draining);
8295         while (atomic_read(&root->fs_info->nr_async_submits) ||
8296               atomic_read(&root->fs_info->async_delalloc_pages)) {
8297                 wait_event(root->fs_info->async_submit_wait,
8298                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8299                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8300         }
8301         atomic_dec(&root->fs_info->async_submit_draining);
8302         return ret;
8303 }
8304
8305 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8306                                     int delay_iput)
8307 {
8308         struct btrfs_root *root;
8309         struct list_head splice;
8310         int ret;
8311
8312         if (fs_info->sb->s_flags & MS_RDONLY)
8313                 return -EROFS;
8314
8315         INIT_LIST_HEAD(&splice);
8316
8317         spin_lock(&fs_info->delalloc_root_lock);
8318         list_splice_init(&fs_info->delalloc_roots, &splice);
8319         while (!list_empty(&splice)) {
8320                 root = list_first_entry(&splice, struct btrfs_root,
8321                                         delalloc_root);
8322                 root = btrfs_grab_fs_root(root);
8323                 BUG_ON(!root);
8324                 list_move_tail(&root->delalloc_root,
8325                                &fs_info->delalloc_roots);
8326                 spin_unlock(&fs_info->delalloc_root_lock);
8327
8328                 ret = __start_delalloc_inodes(root, delay_iput);
8329                 btrfs_put_fs_root(root);
8330                 if (ret)
8331                         goto out;
8332
8333                 spin_lock(&fs_info->delalloc_root_lock);
8334         }
8335         spin_unlock(&fs_info->delalloc_root_lock);
8336
8337         atomic_inc(&fs_info->async_submit_draining);
8338         while (atomic_read(&fs_info->nr_async_submits) ||
8339               atomic_read(&fs_info->async_delalloc_pages)) {
8340                 wait_event(fs_info->async_submit_wait,
8341                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8342                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8343         }
8344         atomic_dec(&fs_info->async_submit_draining);
8345         return 0;
8346 out:
8347         if (!list_empty_careful(&splice)) {
8348                 spin_lock(&fs_info->delalloc_root_lock);
8349                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8350                 spin_unlock(&fs_info->delalloc_root_lock);
8351         }
8352         return ret;
8353 }
8354
8355 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8356                          const char *symname)
8357 {
8358         struct btrfs_trans_handle *trans;
8359         struct btrfs_root *root = BTRFS_I(dir)->root;
8360         struct btrfs_path *path;
8361         struct btrfs_key key;
8362         struct inode *inode = NULL;
8363         int err;
8364         int drop_inode = 0;
8365         u64 objectid;
8366         u64 index = 0 ;
8367         int name_len;
8368         int datasize;
8369         unsigned long ptr;
8370         struct btrfs_file_extent_item *ei;
8371         struct extent_buffer *leaf;
8372
8373         name_len = strlen(symname) + 1;
8374         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8375                 return -ENAMETOOLONG;
8376
8377         /*
8378          * 2 items for inode item and ref
8379          * 2 items for dir items
8380          * 1 item for xattr if selinux is on
8381          */
8382         trans = btrfs_start_transaction(root, 5);
8383         if (IS_ERR(trans))
8384                 return PTR_ERR(trans);
8385
8386         err = btrfs_find_free_ino(root, &objectid);
8387         if (err)
8388                 goto out_unlock;
8389
8390         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8391                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8392                                 S_IFLNK|S_IRWXUGO, &index);
8393         if (IS_ERR(inode)) {
8394                 err = PTR_ERR(inode);
8395                 goto out_unlock;
8396         }
8397
8398         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8399         if (err) {
8400                 drop_inode = 1;
8401                 goto out_unlock;
8402         }
8403
8404         /*
8405         * If the active LSM wants to access the inode during
8406         * d_instantiate it needs these. Smack checks to see
8407         * if the filesystem supports xattrs by looking at the
8408         * ops vector.
8409         */
8410         inode->i_fop = &btrfs_file_operations;
8411         inode->i_op = &btrfs_file_inode_operations;
8412
8413         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8414         if (err)
8415                 drop_inode = 1;
8416         else {
8417                 inode->i_mapping->a_ops = &btrfs_aops;
8418                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8419                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8420         }
8421         if (drop_inode)
8422                 goto out_unlock;
8423
8424         path = btrfs_alloc_path();
8425         if (!path) {
8426                 err = -ENOMEM;
8427                 drop_inode = 1;
8428                 goto out_unlock;
8429         }
8430         key.objectid = btrfs_ino(inode);
8431         key.offset = 0;
8432         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8433         datasize = btrfs_file_extent_calc_inline_size(name_len);
8434         err = btrfs_insert_empty_item(trans, root, path, &key,
8435                                       datasize);
8436         if (err) {
8437                 drop_inode = 1;
8438                 btrfs_free_path(path);
8439                 goto out_unlock;
8440         }
8441         leaf = path->nodes[0];
8442         ei = btrfs_item_ptr(leaf, path->slots[0],
8443                             struct btrfs_file_extent_item);
8444         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8445         btrfs_set_file_extent_type(leaf, ei,
8446                                    BTRFS_FILE_EXTENT_INLINE);
8447         btrfs_set_file_extent_encryption(leaf, ei, 0);
8448         btrfs_set_file_extent_compression(leaf, ei, 0);
8449         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8450         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8451
8452         ptr = btrfs_file_extent_inline_start(ei);
8453         write_extent_buffer(leaf, symname, ptr, name_len);
8454         btrfs_mark_buffer_dirty(leaf);
8455         btrfs_free_path(path);
8456
8457         inode->i_op = &btrfs_symlink_inode_operations;
8458         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8459         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8460         inode_set_bytes(inode, name_len);
8461         btrfs_i_size_write(inode, name_len - 1);
8462         err = btrfs_update_inode(trans, root, inode);
8463         if (err)
8464                 drop_inode = 1;
8465
8466 out_unlock:
8467         if (!err)
8468                 d_instantiate(dentry, inode);
8469         btrfs_end_transaction(trans, root);
8470         if (drop_inode) {
8471                 inode_dec_link_count(inode);
8472                 iput(inode);
8473         }
8474         btrfs_btree_balance_dirty(root);
8475         return err;
8476 }
8477
8478 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8479                                        u64 start, u64 num_bytes, u64 min_size,
8480                                        loff_t actual_len, u64 *alloc_hint,
8481                                        struct btrfs_trans_handle *trans)
8482 {
8483         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8484         struct extent_map *em;
8485         struct btrfs_root *root = BTRFS_I(inode)->root;
8486         struct btrfs_key ins;
8487         u64 cur_offset = start;
8488         u64 i_size;
8489         u64 cur_bytes;
8490         int ret = 0;
8491         bool own_trans = true;
8492
8493         if (trans)
8494                 own_trans = false;
8495         while (num_bytes > 0) {
8496                 if (own_trans) {
8497                         trans = btrfs_start_transaction(root, 3);
8498                         if (IS_ERR(trans)) {
8499                                 ret = PTR_ERR(trans);
8500                                 break;
8501                         }
8502                 }
8503
8504                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8505                 cur_bytes = max(cur_bytes, min_size);
8506                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8507                                            min_size, 0, *alloc_hint, &ins, 1);
8508                 if (ret) {
8509                         if (own_trans)
8510                                 btrfs_end_transaction(trans, root);
8511                         break;
8512                 }
8513
8514                 ret = insert_reserved_file_extent(trans, inode,
8515                                                   cur_offset, ins.objectid,
8516                                                   ins.offset, ins.offset,
8517                                                   ins.offset, 0, 0, 0,
8518                                                   BTRFS_FILE_EXTENT_PREALLOC);
8519                 if (ret) {
8520                         btrfs_abort_transaction(trans, root, ret);
8521                         if (own_trans)
8522                                 btrfs_end_transaction(trans, root);
8523                         break;
8524                 }
8525                 btrfs_drop_extent_cache(inode, cur_offset,
8526                                         cur_offset + ins.offset -1, 0);
8527
8528                 em = alloc_extent_map();
8529                 if (!em) {
8530                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8531                                 &BTRFS_I(inode)->runtime_flags);
8532                         goto next;
8533                 }
8534
8535                 em->start = cur_offset;
8536                 em->orig_start = cur_offset;
8537                 em->len = ins.offset;
8538                 em->block_start = ins.objectid;
8539                 em->block_len = ins.offset;
8540                 em->orig_block_len = ins.offset;
8541                 em->ram_bytes = ins.offset;
8542                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8543                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8544                 em->generation = trans->transid;
8545
8546                 while (1) {
8547                         write_lock(&em_tree->lock);
8548                         ret = add_extent_mapping(em_tree, em, 1);
8549                         write_unlock(&em_tree->lock);
8550                         if (ret != -EEXIST)
8551                                 break;
8552                         btrfs_drop_extent_cache(inode, cur_offset,
8553                                                 cur_offset + ins.offset - 1,
8554                                                 0);
8555                 }
8556                 free_extent_map(em);
8557 next:
8558                 num_bytes -= ins.offset;
8559                 cur_offset += ins.offset;
8560                 *alloc_hint = ins.objectid + ins.offset;
8561
8562                 inode_inc_iversion(inode);
8563                 inode->i_ctime = CURRENT_TIME;
8564                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8565                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8566                     (actual_len > inode->i_size) &&
8567                     (cur_offset > inode->i_size)) {
8568                         if (cur_offset > actual_len)
8569                                 i_size = actual_len;
8570                         else
8571                                 i_size = cur_offset;
8572                         i_size_write(inode, i_size);
8573                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8574                 }
8575
8576                 ret = btrfs_update_inode(trans, root, inode);
8577
8578                 if (ret) {
8579                         btrfs_abort_transaction(trans, root, ret);
8580                         if (own_trans)
8581                                 btrfs_end_transaction(trans, root);
8582                         break;
8583                 }
8584
8585                 if (own_trans)
8586                         btrfs_end_transaction(trans, root);
8587         }
8588         return ret;
8589 }
8590
8591 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8592                               u64 start, u64 num_bytes, u64 min_size,
8593                               loff_t actual_len, u64 *alloc_hint)
8594 {
8595         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8596                                            min_size, actual_len, alloc_hint,
8597                                            NULL);
8598 }
8599
8600 int btrfs_prealloc_file_range_trans(struct inode *inode,
8601                                     struct btrfs_trans_handle *trans, int mode,
8602                                     u64 start, u64 num_bytes, u64 min_size,
8603                                     loff_t actual_len, u64 *alloc_hint)
8604 {
8605         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8606                                            min_size, actual_len, alloc_hint, trans);
8607 }
8608
8609 static int btrfs_set_page_dirty(struct page *page)
8610 {
8611         return __set_page_dirty_nobuffers(page);
8612 }
8613
8614 static int btrfs_permission(struct inode *inode, int mask)
8615 {
8616         struct btrfs_root *root = BTRFS_I(inode)->root;
8617         umode_t mode = inode->i_mode;
8618
8619         if (mask & MAY_WRITE &&
8620             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8621                 if (btrfs_root_readonly(root))
8622                         return -EROFS;
8623                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8624                         return -EACCES;
8625         }
8626         return generic_permission(inode, mask);
8627 }
8628
8629 static const struct inode_operations btrfs_dir_inode_operations = {
8630         .getattr        = btrfs_getattr,
8631         .lookup         = btrfs_lookup,
8632         .create         = btrfs_create,
8633         .unlink         = btrfs_unlink,
8634         .link           = btrfs_link,
8635         .mkdir          = btrfs_mkdir,
8636         .rmdir          = btrfs_rmdir,
8637         .rename         = btrfs_rename,
8638         .symlink        = btrfs_symlink,
8639         .setattr        = btrfs_setattr,
8640         .mknod          = btrfs_mknod,
8641         .setxattr       = btrfs_setxattr,
8642         .getxattr       = btrfs_getxattr,
8643         .listxattr      = btrfs_listxattr,
8644         .removexattr    = btrfs_removexattr,
8645         .permission     = btrfs_permission,
8646         .get_acl        = btrfs_get_acl,
8647 };
8648 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8649         .lookup         = btrfs_lookup,
8650         .permission     = btrfs_permission,
8651         .get_acl        = btrfs_get_acl,
8652 };
8653
8654 static const struct file_operations btrfs_dir_file_operations = {
8655         .llseek         = generic_file_llseek,
8656         .read           = generic_read_dir,
8657         .iterate        = btrfs_real_readdir,
8658         .unlocked_ioctl = btrfs_ioctl,
8659 #ifdef CONFIG_COMPAT
8660         .compat_ioctl   = btrfs_ioctl,
8661 #endif
8662         .release        = btrfs_release_file,
8663         .fsync          = btrfs_sync_file,
8664 };
8665
8666 static struct extent_io_ops btrfs_extent_io_ops = {
8667         .fill_delalloc = run_delalloc_range,
8668         .submit_bio_hook = btrfs_submit_bio_hook,
8669         .merge_bio_hook = btrfs_merge_bio_hook,
8670         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8671         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8672         .writepage_start_hook = btrfs_writepage_start_hook,
8673         .set_bit_hook = btrfs_set_bit_hook,
8674         .clear_bit_hook = btrfs_clear_bit_hook,
8675         .merge_extent_hook = btrfs_merge_extent_hook,
8676         .split_extent_hook = btrfs_split_extent_hook,
8677 };
8678
8679 /*
8680  * btrfs doesn't support the bmap operation because swapfiles
8681  * use bmap to make a mapping of extents in the file.  They assume
8682  * these extents won't change over the life of the file and they
8683  * use the bmap result to do IO directly to the drive.
8684  *
8685  * the btrfs bmap call would return logical addresses that aren't
8686  * suitable for IO and they also will change frequently as COW
8687  * operations happen.  So, swapfile + btrfs == corruption.
8688  *
8689  * For now we're avoiding this by dropping bmap.
8690  */
8691 static const struct address_space_operations btrfs_aops = {
8692         .readpage       = btrfs_readpage,
8693         .writepage      = btrfs_writepage,
8694         .writepages     = btrfs_writepages,
8695         .readpages      = btrfs_readpages,
8696         .direct_IO      = btrfs_direct_IO,
8697         .invalidatepage = btrfs_invalidatepage,
8698         .releasepage    = btrfs_releasepage,
8699         .set_page_dirty = btrfs_set_page_dirty,
8700         .error_remove_page = generic_error_remove_page,
8701 };
8702
8703 static const struct address_space_operations btrfs_symlink_aops = {
8704         .readpage       = btrfs_readpage,
8705         .writepage      = btrfs_writepage,
8706         .invalidatepage = btrfs_invalidatepage,
8707         .releasepage    = btrfs_releasepage,
8708 };
8709
8710 static const struct inode_operations btrfs_file_inode_operations = {
8711         .getattr        = btrfs_getattr,
8712         .setattr        = btrfs_setattr,
8713         .setxattr       = btrfs_setxattr,
8714         .getxattr       = btrfs_getxattr,
8715         .listxattr      = btrfs_listxattr,
8716         .removexattr    = btrfs_removexattr,
8717         .permission     = btrfs_permission,
8718         .fiemap         = btrfs_fiemap,
8719         .get_acl        = btrfs_get_acl,
8720         .update_time    = btrfs_update_time,
8721 };
8722 static const struct inode_operations btrfs_special_inode_operations = {
8723         .getattr        = btrfs_getattr,
8724         .setattr        = btrfs_setattr,
8725         .permission     = btrfs_permission,
8726         .setxattr       = btrfs_setxattr,
8727         .getxattr       = btrfs_getxattr,
8728         .listxattr      = btrfs_listxattr,
8729         .removexattr    = btrfs_removexattr,
8730         .get_acl        = btrfs_get_acl,
8731         .update_time    = btrfs_update_time,
8732 };
8733 static const struct inode_operations btrfs_symlink_inode_operations = {
8734         .readlink       = generic_readlink,
8735         .follow_link    = page_follow_link_light,
8736         .put_link       = page_put_link,
8737         .getattr        = btrfs_getattr,
8738         .setattr        = btrfs_setattr,
8739         .permission     = btrfs_permission,
8740         .setxattr       = btrfs_setxattr,
8741         .getxattr       = btrfs_getxattr,
8742         .listxattr      = btrfs_listxattr,
8743         .removexattr    = btrfs_removexattr,
8744         .get_acl        = btrfs_get_acl,
8745         .update_time    = btrfs_update_time,
8746 };
8747
8748 const struct dentry_operations btrfs_dentry_operations = {
8749         .d_delete       = btrfs_dentry_delete,
8750         .d_release      = btrfs_dentry_release,
8751 };