fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include <linux/blkdev.h>
  44 #include "compat.h"
  45 #include "ctree.h"
  46 #include "disk-io.h"
  47 #include "transaction.h"
  48 #include "btrfs_inode.h"
  49 #include "print-tree.h"
  50 #include "ordered-data.h"
  51 #include "xattr.h"
  52 #include "tree-log.h"
  53 #include "volumes.h"
  54 #include "compression.h"
  55 #include "locking.h"
  56 #include "free-space-cache.h"
  57 #include "inode-map.h"
  58 #include "backref.h"
  59
  60 struct btrfs_iget_args {
  61         u64 ino;
  62         struct btrfs_root *root;
  63 };
  64
  65 static const struct inode_operations btrfs_dir_inode_operations;
  66 static const struct inode_operations btrfs_symlink_inode_operations;
  67 static const struct inode_operations btrfs_dir_ro_inode_operations;
  68 static const struct inode_operations btrfs_special_inode_operations;
  69 static const struct inode_operations btrfs_file_inode_operations;
  70 static const struct address_space_operations btrfs_aops;
  71 static const struct address_space_operations btrfs_symlink_aops;
  72 static const struct file_operations btrfs_dir_file_operations;
  73 static struct extent_io_ops btrfs_extent_io_ops;
  74
  75 static struct kmem_cache *btrfs_inode_cachep;
  76 static struct kmem_cache *btrfs_delalloc_work_cachep;
  77 struct kmem_cache *btrfs_trans_handle_cachep;
  78 struct kmem_cache *btrfs_transaction_cachep;
  79 struct kmem_cache *btrfs_path_cachep;
  80 struct kmem_cache *btrfs_free_space_cachep;
  81
  82 #define S_SHIFT 12
  83 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  84         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  85         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  86         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  87         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  88         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  89         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  90         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  91 };
  92
  93 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  94 static int btrfs_truncate(struct inode *inode);
  95 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  96 static noinline int cow_file_range(struct inode *inode,
  97                                    struct page *locked_page,
  98                                    u64 start, u64 end, int *page_started,
  99                                    unsigned long *nr_written, int unlock);
 100 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 101                                            u64 len, u64 orig_start,
 102                                            u64 block_start, u64 block_len,
 103                                            u64 orig_block_len, u64 ram_bytes,
 104                                            int type);
 105
 106 static int btrfs_dirty_inode(struct inode *inode);
 107
 108 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 109                                      struct inode *inode,  struct inode *dir,
 110                                      const struct qstr *qstr)
 111 {
 112         int err;
 113
 114         err = btrfs_init_acl(trans, inode, dir);
 115         if (!err)
 116                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 117         return err;
 118 }
 119
 120 /*
 121  * this does all the hard work for inserting an inline extent into
 122  * the btree.  The caller should have done a btrfs_drop_extents so that
 123  * no overlapping inline items exist in the btree
 124  */
 125 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 126                                 struct btrfs_root *root, struct inode *inode,
 127                                 u64 start, size_t size, size_t compressed_size,
 128                                 int compress_type,
 129                                 struct page **compressed_pages)
 130 {
 131         struct btrfs_key key;
 132         struct btrfs_path *path;
 133         struct extent_buffer *leaf;
 134         struct page *page = NULL;
 135         char *kaddr;
 136         unsigned long ptr;
 137         struct btrfs_file_extent_item *ei;
 138         int err = 0;
 139         int ret;
 140         size_t cur_size = size;
 141         size_t datasize;
 142         unsigned long offset;
 143
 144         if (compressed_size && compressed_pages)
 145                 cur_size = compressed_size;
 146
 147         path = btrfs_alloc_path();
 148         if (!path)
 149                 return -ENOMEM;
 150
 151         path->leave_spinning = 1;
 152
 153         key.objectid = btrfs_ino(inode);
 154         key.offset = start;
 155         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 156         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 157
 158         inode_add_bytes(inode, size);
 159         ret = btrfs_insert_empty_item(trans, root, path, &key,
 160                                       datasize);
 161         if (ret) {
 162                 err = ret;
 163                 goto fail;
 164         }
 165         leaf = path->nodes[0];
 166         ei = btrfs_item_ptr(leaf, path->slots[0],
 167                             struct btrfs_file_extent_item);
 168         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 169         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 170         btrfs_set_file_extent_encryption(leaf, ei, 0);
 171         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 172         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 173         ptr = btrfs_file_extent_inline_start(ei);
 174
 175         if (compress_type != BTRFS_COMPRESS_NONE) {
 176                 struct page *cpage;
 177                 int i = 0;
 178                 while (compressed_size > 0) {
 179                         cpage = compressed_pages[i];
 180                         cur_size = min_t(unsigned long, compressed_size,
 181                                        PAGE_CACHE_SIZE);
 182
 183                         kaddr = kmap_atomic(cpage);
 184                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 185                         kunmap_atomic(kaddr);
 186
 187                         i++;
 188                         ptr += cur_size;
 189                         compressed_size -= cur_size;
 190                 }
 191                 btrfs_set_file_extent_compression(leaf, ei,
 192                                                   compress_type);
 193         } else {
 194                 page = find_get_page(inode->i_mapping,
 195                                      start >> PAGE_CACHE_SHIFT);
 196                 btrfs_set_file_extent_compression(leaf, ei, 0);
 197                 kaddr = kmap_atomic(page);
 198                 offset = start & (PAGE_CACHE_SIZE - 1);
 199                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 200                 kunmap_atomic(kaddr);
 201                 page_cache_release(page);
 202         }
 203         btrfs_mark_buffer_dirty(leaf);
 204         btrfs_free_path(path);
 205
 206         /*
 207          * we're an inline extent, so nobody can
 208          * extend the file past i_size without locking
 209          * a page we already have locked.
 210          *
 211          * We must do any isize and inode updates
 212          * before we unlock the pages.  Otherwise we
 213          * could end up racing with unlink.
 214          */
 215         BTRFS_I(inode)->disk_i_size = inode->i_size;
 216         ret = btrfs_update_inode(trans, root, inode);
 217
 218         return ret;
 219 fail:
 220         btrfs_free_path(path);
 221         return err;
 222 }
 223
 224
 225 /*
 226  * conditionally insert an inline extent into the file.  This
 227  * does the checks required to make sure the data is small enough
 228  * to fit as an inline extent.
 229  */
 230 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 231                                  struct btrfs_root *root,
 232                                  struct inode *inode, u64 start, u64 end,
 233                                  size_t compressed_size, int compress_type,
 234                                  struct page **compressed_pages)
 235 {
 236         u64 isize = i_size_read(inode);
 237         u64 actual_end = min(end + 1, isize);
 238         u64 inline_len = actual_end - start;
 239         u64 aligned_end = ALIGN(end, root->sectorsize);
 240         u64 data_len = inline_len;
 241         int ret;
 242
 243         if (compressed_size)
 244                 data_len = compressed_size;
 245
 246         if (start > 0 ||
 247             actual_end >= PAGE_CACHE_SIZE ||
 248             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 249             (!compressed_size &&
 250             (actual_end & (root->sectorsize - 1)) == 0) ||
 251             end + 1 < isize ||
 252             data_len > root->fs_info->max_inline) {
 253                 return 1;
 254         }
 255
 256         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 257         if (ret)
 258                 return ret;
 259
 260         if (isize > actual_end)
 261                 inline_len = min_t(u64, isize, actual_end);
 262         ret = insert_inline_extent(trans, root, inode, start,
 263                                    inline_len, compressed_size,
 264                                    compress_type, compressed_pages);
 265         if (ret && ret != -ENOSPC) {
 266                 btrfs_abort_transaction(trans, root, ret);
 267                 return ret;
 268         } else if (ret == -ENOSPC) {
 269                 return 1;
 270         }
 271
 272         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 273         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 274         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 275         return 0;
 276 }
 277
 278 struct async_extent {
 279         u64 start;
 280         u64 ram_size;
 281         u64 compressed_size;
 282         struct page **pages;
 283         unsigned long nr_pages;
 284         int compress_type;
 285         struct list_head list;
 286 };
 287
 288 struct async_cow {
 289         struct inode *inode;
 290         struct btrfs_root *root;
 291         struct page *locked_page;
 292         u64 start;
 293         u64 end;
 294         struct list_head extents;
 295         struct btrfs_work work;
 296 };
 297
 298 static noinline int add_async_extent(struct async_cow *cow,
 299                                      u64 start, u64 ram_size,
 300                                      u64 compressed_size,
 301                                      struct page **pages,
 302                                      unsigned long nr_pages,
 303                                      int compress_type)
 304 {
 305         struct async_extent *async_extent;
 306
 307         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 308         BUG_ON(!async_extent); /* -ENOMEM */
 309         async_extent->start = start;
 310         async_extent->ram_size = ram_size;
 311         async_extent->compressed_size = compressed_size;
 312         async_extent->pages = pages;
 313         async_extent->nr_pages = nr_pages;
 314         async_extent->compress_type = compress_type;
 315         list_add_tail(&async_extent->list, &cow->extents);
 316         return 0;
 317 }
 318
 319 /*
 320  * we create compressed extents in two phases.  The first
 321  * phase compresses a range of pages that have already been
 322  * locked (both pages and state bits are locked).
 323  *
 324  * This is done inside an ordered work queue, and the compression
 325  * is spread across many cpus.  The actual IO submission is step
 326  * two, and the ordered work queue takes care of making sure that
 327  * happens in the same order things were put onto the queue by
 328  * writepages and friends.
 329  *
 330  * If this code finds it can't get good compression, it puts an
 331  * entry onto the work queue to write the uncompressed bytes.  This
 332  * makes sure that both compressed inodes and uncompressed inodes
 333  * are written in the same order that the flusher thread sent them
 334  * down.
 335  */
 336 static noinline int compress_file_range(struct inode *inode,
 337                                         struct page *locked_page,
 338                                         u64 start, u64 end,
 339                                         struct async_cow *async_cow,
 340                                         int *num_added)
 341 {
 342         struct btrfs_root *root = BTRFS_I(inode)->root;
 343         struct btrfs_trans_handle *trans;
 344         u64 num_bytes;
 345         u64 blocksize = root->sectorsize;
 346         u64 actual_end;
 347         u64 isize = i_size_read(inode);
 348         int ret = 0;
 349         struct page **pages = NULL;
 350         unsigned long nr_pages;
 351         unsigned long nr_pages_ret = 0;
 352         unsigned long total_compressed = 0;
 353         unsigned long total_in = 0;
 354         unsigned long max_compressed = 128 * 1024;
 355         unsigned long max_uncompressed = 128 * 1024;
 356         int i;
 357         int will_compress;
 358         int compress_type = root->fs_info->compress_type;
 359         int redirty = 0;
 360
 361         /* if this is a small write inside eof, kick off a defrag */
 362         if ((end - start + 1) < 16 * 1024 &&
 363             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 364                 btrfs_add_inode_defrag(NULL, inode);
 365
 366         actual_end = min_t(u64, isize, end + 1);
 367 again:
 368         will_compress = 0;
 369         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 370         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 371
 372         /*
 373          * we don't want to send crud past the end of i_size through
 374          * compression, that's just a waste of CPU time.  So, if the
 375          * end of the file is before the start of our current
 376          * requested range of bytes, we bail out to the uncompressed
 377          * cleanup code that can deal with all of this.
 378          *
 379          * It isn't really the fastest way to fix things, but this is a
 380          * very uncommon corner.
 381          */
 382         if (actual_end <= start)
 383                 goto cleanup_and_bail_uncompressed;
 384
 385         total_compressed = actual_end - start;
 386
 387         /* we want to make sure that amount of ram required to uncompress
 388          * an extent is reasonable, so we limit the total size in ram
 389          * of a compressed extent to 128k.  This is a crucial number
 390          * because it also controls how easily we can spread reads across
 391          * cpus for decompression.
 392          *
 393          * We also want to make sure the amount of IO required to do
 394          * a random read is reasonably small, so we limit the size of
 395          * a compressed extent to 128k.
 396          */
 397         total_compressed = min(total_compressed, max_uncompressed);
 398         num_bytes = ALIGN(end - start + 1, blocksize);
 399         num_bytes = max(blocksize,  num_bytes);
 400         total_in = 0;
 401         ret = 0;
 402
 403         /*
 404          * we do compression for mount -o compress and when the
 405          * inode has not been flagged as nocompress.  This flag can
 406          * change at any time if we discover bad compression ratios.
 407          */
 408         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 409             (btrfs_test_opt(root, COMPRESS) ||
 410              (BTRFS_I(inode)->force_compress) ||
 411              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 412                 WARN_ON(pages);
 413                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 414                 if (!pages) {
 415                         /* just bail out to the uncompressed code */
 416                         goto cont;
 417                 }
 418
 419                 if (BTRFS_I(inode)->force_compress)
 420                         compress_type = BTRFS_I(inode)->force_compress;
 421
 422                 /*
 423                  * we need to call clear_page_dirty_for_io on each
 424                  * page in the range.  Otherwise applications with the file
 425                  * mmap'd can wander in and change the page contents while
 426                  * we are compressing them.
 427                  *
 428                  * If the compression fails for any reason, we set the pages
 429                  * dirty again later on.
 430                  */
 431                 extent_range_clear_dirty_for_io(inode, start, end);
 432                 redirty = 1;
 433                 ret = btrfs_compress_pages(compress_type,
 434                                            inode->i_mapping, start,
 435                                            total_compressed, pages,
 436                                            nr_pages, &nr_pages_ret,
 437                                            &total_in,
 438                                            &total_compressed,
 439                                            max_compressed);
 440
 441                 if (!ret) {
 442                         unsigned long offset = total_compressed &
 443                                 (PAGE_CACHE_SIZE - 1);
 444                         struct page *page = pages[nr_pages_ret - 1];
 445                         char *kaddr;
 446
 447                         /* zero the tail end of the last page, we might be
 448                          * sending it down to disk
 449                          */
 450                         if (offset) {
 451                                 kaddr = kmap_atomic(page);
 452                                 memset(kaddr + offset, 0,
 453                                        PAGE_CACHE_SIZE - offset);
 454                                 kunmap_atomic(kaddr);
 455                         }
 456                         will_compress = 1;
 457                 }
 458         }
 459 cont:
 460         if (start == 0) {
 461                 trans = btrfs_join_transaction(root);
 462                 if (IS_ERR(trans)) {
 463                         ret = PTR_ERR(trans);
 464                         trans = NULL;
 465                         goto cleanup_and_out;
 466                 }
 467                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 468
 469                 /* lets try to make an inline extent */
 470                 if (ret || total_in < (actual_end - start)) {
 471                         /* we didn't compress the entire range, try
 472                          * to make an uncompressed inline extent.
 473                          */
 474                         ret = cow_file_range_inline(trans, root, inode,
 475                                                     start, end, 0, 0, NULL);
 476                 } else {
 477                         /* try making a compressed inline extent */
 478                         ret = cow_file_range_inline(trans, root, inode,
 479                                                     start, end,
 480                                                     total_compressed,
 481                                                     compress_type, pages);
 482                 }
 483                 if (ret <= 0) {
 484                         /*
 485                          * inline extent creation worked or returned error,
 486                          * we don't need to create any more async work items.
 487                          * Unlock and free up our temp pages.
 488                          */
 489                         extent_clear_unlock_delalloc(inode,
 490                              &BTRFS_I(inode)->io_tree,
 491                              start, end, NULL,
 492                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 493                              EXTENT_CLEAR_DELALLOC |
 494                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 495
 496                         btrfs_end_transaction(trans, root);
 497                         goto free_pages_out;
 498                 }
 499                 btrfs_end_transaction(trans, root);
 500         }
 501
 502         if (will_compress) {
 503                 /*
 504                  * we aren't doing an inline extent round the compressed size
 505                  * up to a block size boundary so the allocator does sane
 506                  * things
 507                  */
 508                 total_compressed = ALIGN(total_compressed, blocksize);
 509
 510                 /*
 511                  * one last check to make sure the compression is really a
 512                  * win, compare the page count read with the blocks on disk
 513                  */
 514                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 515                 if (total_compressed >= total_in) {
 516                         will_compress = 0;
 517                 } else {
 518                         num_bytes = total_in;
 519                 }
 520         }
 521         if (!will_compress && pages) {
 522                 /*
 523                  * the compression code ran but failed to make things smaller,
 524                  * free any pages it allocated and our page pointer array
 525                  */
 526                 for (i = 0; i < nr_pages_ret; i++) {
 527                         WARN_ON(pages[i]->mapping);
 528                         page_cache_release(pages[i]);
 529                 }
 530                 kfree(pages);
 531                 pages = NULL;
 532                 total_compressed = 0;
 533                 nr_pages_ret = 0;
 534
 535                 /* flag the file so we don't compress in the future */
 536                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 537                     !(BTRFS_I(inode)->force_compress)) {
 538                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 539                 }
 540         }
 541         if (will_compress) {
 542                 *num_added += 1;
 543
 544                 /* the async work queues will take care of doing actual
 545                  * allocation on disk for these compressed pages,
 546                  * and will submit them to the elevator.
 547                  */
 548                 add_async_extent(async_cow, start, num_bytes,
 549                                  total_compressed, pages, nr_pages_ret,
 550                                  compress_type);
 551
 552                 if (start + num_bytes < end) {
 553                         start += num_bytes;
 554                         pages = NULL;
 555                         cond_resched();
 556                         goto again;
 557                 }
 558         } else {
 559 cleanup_and_bail_uncompressed:
 560                 /*
 561                  * No compression, but we still need to write the pages in
 562                  * the file we've been given so far.  redirty the locked
 563                  * page if it corresponds to our extent and set things up
 564                  * for the async work queue to run cow_file_range to do
 565                  * the normal delalloc dance
 566                  */
 567                 if (page_offset(locked_page) >= start &&
 568                     page_offset(locked_page) <= end) {
 569                         __set_page_dirty_nobuffers(locked_page);
 570                         /* unlocked later on in the async handlers */
 571                 }
 572                 if (redirty)
 573                         extent_range_redirty_for_io(inode, start, end);
 574                 add_async_extent(async_cow, start, end - start + 1,
 575                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 576                 *num_added += 1;
 577         }
 578
 579 out:
 580         return ret;
 581
 582 free_pages_out:
 583         for (i = 0; i < nr_pages_ret; i++) {
 584                 WARN_ON(pages[i]->mapping);
 585                 page_cache_release(pages[i]);
 586         }
 587         kfree(pages);
 588
 589         goto out;
 590
 591 cleanup_and_out:
 592         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 593                                      start, end, NULL,
 594                                      EXTENT_CLEAR_UNLOCK_PAGE |
 595                                      EXTENT_CLEAR_DIRTY |
 596                                      EXTENT_CLEAR_DELALLOC |
 597                                      EXTENT_SET_WRITEBACK |
 598                                      EXTENT_END_WRITEBACK);
 599         if (!trans || IS_ERR(trans))
 600                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 601         else
 602                 btrfs_abort_transaction(trans, root, ret);
 603         goto free_pages_out;
 604 }
 605
 606 /*
 607  * phase two of compressed writeback.  This is the ordered portion
 608  * of the code, which only gets called in the order the work was
 609  * queued.  We walk all the async extents created by compress_file_range
 610  * and send them down to the disk.
 611  */
 612 static noinline int submit_compressed_extents(struct inode *inode,
 613                                               struct async_cow *async_cow)
 614 {
 615         struct async_extent *async_extent;
 616         u64 alloc_hint = 0;
 617         struct btrfs_trans_handle *trans;
 618         struct btrfs_key ins;
 619         struct extent_map *em;
 620         struct btrfs_root *root = BTRFS_I(inode)->root;
 621         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 622         struct extent_io_tree *io_tree;
 623         int ret = 0;
 624
 625         if (list_empty(&async_cow->extents))
 626                 return 0;
 627
 628 again:
 629         while (!list_empty(&async_cow->extents)) {
 630                 async_extent = list_entry(async_cow->extents.next,
 631                                           struct async_extent, list);
 632                 list_del(&async_extent->list);
 633
 634                 io_tree = &BTRFS_I(inode)->io_tree;
 635
 636 retry:
 637                 /* did the compression code fall back to uncompressed IO? */
 638                 if (!async_extent->pages) {
 639                         int page_started = 0;
 640                         unsigned long nr_written = 0;
 641
 642                         lock_extent(io_tree, async_extent->start,
 643                                          async_extent->start +
 644                                          async_extent->ram_size - 1);
 645
 646                         /* allocate blocks */
 647                         ret = cow_file_range(inode, async_cow->locked_page,
 648                                              async_extent->start,
 649                                              async_extent->start +
 650                                              async_extent->ram_size - 1,
 651                                              &page_started, &nr_written, 0);
 652
 653                         /* JDM XXX */
 654
 655                         /*
 656                          * if page_started, cow_file_range inserted an
 657                          * inline extent and took care of all the unlocking
 658                          * and IO for us.  Otherwise, we need to submit
 659                          * all those pages down to the drive.
 660                          */
 661                         if (!page_started && !ret)
 662                                 extent_write_locked_range(io_tree,
 663                                                   inode, async_extent->start,
 664                                                   async_extent->start +
 665                                                   async_extent->ram_size - 1,
 666                                                   btrfs_get_extent,
 667                                                   WB_SYNC_ALL);
 668                         else if (ret)
 669                                 unlock_page(async_cow->locked_page);
 670                         kfree(async_extent);
 671                         cond_resched();
 672                         continue;
 673                 }
 674
 675                 lock_extent(io_tree, async_extent->start,
 676                             async_extent->start + async_extent->ram_size - 1);
 677
 678                 trans = btrfs_join_transaction(root);
 679                 if (IS_ERR(trans)) {
 680                         ret = PTR_ERR(trans);
 681                 } else {
 682                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 683                         ret = btrfs_reserve_extent(trans, root,
 684                                            async_extent->compressed_size,
 685                                            async_extent->compressed_size,
 686                                            0, alloc_hint, &ins, 1);
 687                         if (ret && ret != -ENOSPC)
 688                                 btrfs_abort_transaction(trans, root, ret);
 689                         btrfs_end_transaction(trans, root);
 690                 }
 691
 692                 if (ret) {
 693                         int i;
 694
 695                         for (i = 0; i < async_extent->nr_pages; i++) {
 696                                 WARN_ON(async_extent->pages[i]->mapping);
 697                                 page_cache_release(async_extent->pages[i]);
 698                         }
 699                         kfree(async_extent->pages);
 700                         async_extent->nr_pages = 0;
 701                         async_extent->pages = NULL;
 702
 703                         if (ret == -ENOSPC)
 704                                 goto retry;
 705                         goto out_free;
 706                 }
 707
 708                 /*
 709                  * here we're doing allocation and writeback of the
 710                  * compressed pages
 711                  */
 712                 btrfs_drop_extent_cache(inode, async_extent->start,
 713                                         async_extent->start +
 714                                         async_extent->ram_size - 1, 0);
 715
 716                 em = alloc_extent_map();
 717                 if (!em) {
 718                         ret = -ENOMEM;
 719                         goto out_free_reserve;
 720                 }
 721                 em->start = async_extent->start;
 722                 em->len = async_extent->ram_size;
 723                 em->orig_start = em->start;
 724                 em->mod_start = em->start;
 725                 em->mod_len = em->len;
 726
 727                 em->block_start = ins.objectid;
 728                 em->block_len = ins.offset;
 729                 em->orig_block_len = ins.offset;
 730                 em->ram_bytes = async_extent->ram_size;
 731                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 732                 em->compress_type = async_extent->compress_type;
 733                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 734                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 735                 em->generation = -1;
 736
 737                 while (1) {
 738                         write_lock(&em_tree->lock);
 739                         ret = add_extent_mapping(em_tree, em, 1);
 740                         write_unlock(&em_tree->lock);
 741                         if (ret != -EEXIST) {
 742                                 free_extent_map(em);
 743                                 break;
 744                         }
 745                         btrfs_drop_extent_cache(inode, async_extent->start,
 746                                                 async_extent->start +
 747                                                 async_extent->ram_size - 1, 0);
 748                 }
 749
 750                 if (ret)
 751                         goto out_free_reserve;
 752
 753                 ret = btrfs_add_ordered_extent_compress(inode,
 754                                                 async_extent->start,
 755                                                 ins.objectid,
 756                                                 async_extent->ram_size,
 757                                                 ins.offset,
 758                                                 BTRFS_ORDERED_COMPRESSED,
 759                                                 async_extent->compress_type);
 760                 if (ret)
 761                         goto out_free_reserve;
 762
 763                 /*
 764                  * clear dirty, set writeback and unlock the pages.
 765                  */
 766                 extent_clear_unlock_delalloc(inode,
 767                                 &BTRFS_I(inode)->io_tree,
 768                                 async_extent->start,
 769                                 async_extent->start +
 770                                 async_extent->ram_size - 1,
 771                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 772                                 EXTENT_CLEAR_UNLOCK |
 773                                 EXTENT_CLEAR_DELALLOC |
 774                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 775
 776                 ret = btrfs_submit_compressed_write(inode,
 777                                     async_extent->start,
 778                                     async_extent->ram_size,
 779                                     ins.objectid,
 780                                     ins.offset, async_extent->pages,
 781                                     async_extent->nr_pages);
 782                 alloc_hint = ins.objectid + ins.offset;
 783                 kfree(async_extent);
 784                 if (ret)
 785                         goto out;
 786                 cond_resched();
 787         }
 788         ret = 0;
 789 out:
 790         return ret;
 791 out_free_reserve:
 792         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 793 out_free:
 794         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 795                                      async_extent->start,
 796                                      async_extent->start +
 797                                      async_extent->ram_size - 1,
 798                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 799                                      EXTENT_CLEAR_UNLOCK |
 800                                      EXTENT_CLEAR_DELALLOC |
 801                                      EXTENT_CLEAR_DIRTY |
 802                                      EXTENT_SET_WRITEBACK |
 803                                      EXTENT_END_WRITEBACK);
 804         kfree(async_extent);
 805         goto again;
 806 }
 807
 808 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 809                                       u64 num_bytes)
 810 {
 811         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 812         struct extent_map *em;
 813         u64 alloc_hint = 0;
 814
 815         read_lock(&em_tree->lock);
 816         em = search_extent_mapping(em_tree, start, num_bytes);
 817         if (em) {
 818                 /*
 819                  * if block start isn't an actual block number then find the
 820                  * first block in this inode and use that as a hint.  If that
 821                  * block is also bogus then just don't worry about it.
 822                  */
 823                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 824                         free_extent_map(em);
 825                         em = search_extent_mapping(em_tree, 0, 0);
 826                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 827                                 alloc_hint = em->block_start;
 828                         if (em)
 829                                 free_extent_map(em);
 830                 } else {
 831                         alloc_hint = em->block_start;
 832                         free_extent_map(em);
 833                 }
 834         }
 835         read_unlock(&em_tree->lock);
 836
 837         return alloc_hint;
 838 }
 839
 840 /*
 841  * when extent_io.c finds a delayed allocation range in the file,
 842  * the call backs end up in this code.  The basic idea is to
 843  * allocate extents on disk for the range, and create ordered data structs
 844  * in ram to track those extents.
 845  *
 846  * locked_page is the page that writepage had locked already.  We use
 847  * it to make sure we don't do extra locks or unlocks.
 848  *
 849  * *page_started is set to one if we unlock locked_page and do everything
 850  * required to start IO on it.  It may be clean and already done with
 851  * IO when we return.
 852  */
 853 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 854                                      struct inode *inode,
 855                                      struct btrfs_root *root,
 856                                      struct page *locked_page,
 857                                      u64 start, u64 end, int *page_started,
 858                                      unsigned long *nr_written,
 859                                      int unlock)
 860 {
 861         u64 alloc_hint = 0;
 862         u64 num_bytes;
 863         unsigned long ram_size;
 864         u64 disk_num_bytes;
 865         u64 cur_alloc_size;
 866         u64 blocksize = root->sectorsize;
 867         struct btrfs_key ins;
 868         struct extent_map *em;
 869         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 870         int ret = 0;
 871
 872         BUG_ON(btrfs_is_free_space_inode(inode));
 873
 874         num_bytes = ALIGN(end - start + 1, blocksize);
 875         num_bytes = max(blocksize,  num_bytes);
 876         disk_num_bytes = num_bytes;
 877
 878         /* if this is a small write inside eof, kick off defrag */
 879         if (num_bytes < 64 * 1024 &&
 880             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 881                 btrfs_add_inode_defrag(trans, inode);
 882
 883         if (start == 0) {
 884                 /* lets try to make an inline extent */
 885                 ret = cow_file_range_inline(trans, root, inode,
 886                                             start, end, 0, 0, NULL);
 887                 if (ret == 0) {
 888                         extent_clear_unlock_delalloc(inode,
 889                                      &BTRFS_I(inode)->io_tree,
 890                                      start, end, NULL,
 891                                      EXTENT_CLEAR_UNLOCK_PAGE |
 892                                      EXTENT_CLEAR_UNLOCK |
 893                                      EXTENT_CLEAR_DELALLOC |
 894                                      EXTENT_CLEAR_DIRTY |
 895                                      EXTENT_SET_WRITEBACK |
 896                                      EXTENT_END_WRITEBACK);
 897
 898                         *nr_written = *nr_written +
 899                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 900                         *page_started = 1;
 901                         goto out;
 902                 } else if (ret < 0) {
 903                         btrfs_abort_transaction(trans, root, ret);
 904                         goto out_unlock;
 905                 }
 906         }
 907
 908         BUG_ON(disk_num_bytes >
 909                btrfs_super_total_bytes(root->fs_info->super_copy));
 910
 911         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 912         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 913
 914         while (disk_num_bytes > 0) {
 915                 unsigned long op;
 916
 917                 cur_alloc_size = disk_num_bytes;
 918                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 919                                            root->sectorsize, 0, alloc_hint,
 920                                            &ins, 1);
 921                 if (ret < 0) {
 922                         btrfs_abort_transaction(trans, root, ret);
 923                         goto out_unlock;
 924                 }
 925
 926                 em = alloc_extent_map();
 927                 if (!em) {
 928                         ret = -ENOMEM;
 929                         goto out_reserve;
 930                 }
 931                 em->start = start;
 932                 em->orig_start = em->start;
 933                 ram_size = ins.offset;
 934                 em->len = ins.offset;
 935                 em->mod_start = em->start;
 936                 em->mod_len = em->len;
 937
 938                 em->block_start = ins.objectid;
 939                 em->block_len = ins.offset;
 940                 em->orig_block_len = ins.offset;
 941                 em->ram_bytes = ram_size;
 942                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 943                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 944                 em->generation = -1;
 945
 946                 while (1) {
 947                         write_lock(&em_tree->lock);
 948                         ret = add_extent_mapping(em_tree, em, 1);
 949                         write_unlock(&em_tree->lock);
 950                         if (ret != -EEXIST) {
 951                                 free_extent_map(em);
 952                                 break;
 953                         }
 954                         btrfs_drop_extent_cache(inode, start,
 955                                                 start + ram_size - 1, 0);
 956                 }
 957                 if (ret)
 958                         goto out_reserve;
 959
 960                 cur_alloc_size = ins.offset;
 961                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 962                                                ram_size, cur_alloc_size, 0);
 963                 if (ret)
 964                         goto out_reserve;
 965
 966                 if (root->root_key.objectid ==
 967                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 968                         ret = btrfs_reloc_clone_csums(inode, start,
 969                                                       cur_alloc_size);
 970                         if (ret) {
 971                                 btrfs_abort_transaction(trans, root, ret);
 972                                 goto out_reserve;
 973                         }
 974                 }
 975
 976                 if (disk_num_bytes < cur_alloc_size)
 977                         break;
 978
 979                 /* we're not doing compressed IO, don't unlock the first
 980                  * page (which the caller expects to stay locked), don't
 981                  * clear any dirty bits and don't set any writeback bits
 982                  *
 983                  * Do set the Private2 bit so we know this page was properly
 984                  * setup for writepage
 985                  */
 986                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 987                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 988                         EXTENT_SET_PRIVATE2;
 989
 990                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 991                                              start, start + ram_size - 1,
 992                                              locked_page, op);
 993                 disk_num_bytes -= cur_alloc_size;
 994                 num_bytes -= cur_alloc_size;
 995                 alloc_hint = ins.objectid + ins.offset;
 996                 start += cur_alloc_size;
 997         }
 998 out:
 999         return ret;
1000
1001 out_reserve:
1002         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
1003 out_unlock:
1004         extent_clear_unlock_delalloc(inode,
1005                      &BTRFS_I(inode)->io_tree,
1006                      start, end, locked_page,
1007                      EXTENT_CLEAR_UNLOCK_PAGE |
1008                      EXTENT_CLEAR_UNLOCK |
1009                      EXTENT_CLEAR_DELALLOC |
1010                      EXTENT_CLEAR_DIRTY |
1011                      EXTENT_SET_WRITEBACK |
1012                      EXTENT_END_WRITEBACK);
1013
1014         goto out;
1015 }
1016
1017 static noinline int cow_file_range(struct inode *inode,
1018                                    struct page *locked_page,
1019                                    u64 start, u64 end, int *page_started,
1020                                    unsigned long *nr_written,
1021                                    int unlock)
1022 {
1023         struct btrfs_trans_handle *trans;
1024         struct btrfs_root *root = BTRFS_I(inode)->root;
1025         int ret;
1026
1027         trans = btrfs_join_transaction(root);
1028         if (IS_ERR(trans)) {
1029                 extent_clear_unlock_delalloc(inode,
1030                              &BTRFS_I(inode)->io_tree,
1031                              start, end, locked_page,
1032                              EXTENT_CLEAR_UNLOCK_PAGE |
1033                              EXTENT_CLEAR_UNLOCK |
1034                              EXTENT_CLEAR_DELALLOC |
1035                              EXTENT_CLEAR_DIRTY |
1036                              EXTENT_SET_WRITEBACK |
1037                              EXTENT_END_WRITEBACK);
1038                 return PTR_ERR(trans);
1039         }
1040         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1041
1042         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1043                                page_started, nr_written, unlock);
1044
1045         btrfs_end_transaction(trans, root);
1046
1047         return ret;
1048 }
1049
1050 /*
1051  * work queue call back to started compression on a file and pages
1052  */
1053 static noinline void async_cow_start(struct btrfs_work *work)
1054 {
1055         struct async_cow *async_cow;
1056         int num_added = 0;
1057         async_cow = container_of(work, struct async_cow, work);
1058
1059         compress_file_range(async_cow->inode, async_cow->locked_page,
1060                             async_cow->start, async_cow->end, async_cow,
1061                             &num_added);
1062         if (num_added == 0) {
1063                 btrfs_add_delayed_iput(async_cow->inode);
1064                 async_cow->inode = NULL;
1065         }
1066 }
1067
1068 /*
1069  * work queue call back to submit previously compressed pages
1070  */
1071 static noinline void async_cow_submit(struct btrfs_work *work)
1072 {
1073         struct async_cow *async_cow;
1074         struct btrfs_root *root;
1075         unsigned long nr_pages;
1076
1077         async_cow = container_of(work, struct async_cow, work);
1078
1079         root = async_cow->root;
1080         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1081                 PAGE_CACHE_SHIFT;
1082
1083         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1084             5 * 1024 * 1024 &&
1085             waitqueue_active(&root->fs_info->async_submit_wait))
1086                 wake_up(&root->fs_info->async_submit_wait);
1087
1088         if (async_cow->inode)
1089                 submit_compressed_extents(async_cow->inode, async_cow);
1090 }
1091
1092 static noinline void async_cow_free(struct btrfs_work *work)
1093 {
1094         struct async_cow *async_cow;
1095         async_cow = container_of(work, struct async_cow, work);
1096         if (async_cow->inode)
1097                 btrfs_add_delayed_iput(async_cow->inode);
1098         kfree(async_cow);
1099 }
1100
1101 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1102                                 u64 start, u64 end, int *page_started,
1103                                 unsigned long *nr_written)
1104 {
1105         struct async_cow *async_cow;
1106         struct btrfs_root *root = BTRFS_I(inode)->root;
1107         unsigned long nr_pages;
1108         u64 cur_end;
1109         int limit = 10 * 1024 * 1024;
1110
1111         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1112                          1, 0, NULL, GFP_NOFS);
1113         while (start < end) {
1114                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1115                 BUG_ON(!async_cow); /* -ENOMEM */
1116                 async_cow->inode = igrab(inode);
1117                 async_cow->root = root;
1118                 async_cow->locked_page = locked_page;
1119                 async_cow->start = start;
1120
1121                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1122                         cur_end = end;
1123                 else
1124                         cur_end = min(end, start + 512 * 1024 - 1);
1125
1126                 async_cow->end = cur_end;
1127                 INIT_LIST_HEAD(&async_cow->extents);
1128
1129                 async_cow->work.func = async_cow_start;
1130                 async_cow->work.ordered_func = async_cow_submit;
1131                 async_cow->work.ordered_free = async_cow_free;
1132                 async_cow->work.flags = 0;
1133
1134                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1135                         PAGE_CACHE_SHIFT;
1136                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1137
1138                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1139                                    &async_cow->work);
1140
1141                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1142                         wait_event(root->fs_info->async_submit_wait,
1143                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1144                             limit));
1145                 }
1146
1147                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1148                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1149                         wait_event(root->fs_info->async_submit_wait,
1150                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1151                            0));
1152                 }
1153
1154                 *nr_written += nr_pages;
1155                 start = cur_end + 1;
1156         }
1157         *page_started = 1;
1158         return 0;
1159 }
1160
1161 static noinline int csum_exist_in_range(struct btrfs_root *root,
1162                                         u64 bytenr, u64 num_bytes)
1163 {
1164         int ret;
1165         struct btrfs_ordered_sum *sums;
1166         LIST_HEAD(list);
1167
1168         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1169                                        bytenr + num_bytes - 1, &list, 0);
1170         if (ret == 0 && list_empty(&list))
1171                 return 0;
1172
1173         while (!list_empty(&list)) {
1174                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1175                 list_del(&sums->list);
1176                 kfree(sums);
1177         }
1178         return 1;
1179 }
1180
1181 /*
1182  * when nowcow writeback call back.  This checks for snapshots or COW copies
1183  * of the extents that exist in the file, and COWs the file as required.
1184  *
1185  * If no cow copies or snapshots exist, we write directly to the existing
1186  * blocks on disk
1187  */
1188 static noinline int run_delalloc_nocow(struct inode *inode,
1189                                        struct page *locked_page,
1190                               u64 start, u64 end, int *page_started, int force,
1191                               unsigned long *nr_written)
1192 {
1193         struct btrfs_root *root = BTRFS_I(inode)->root;
1194         struct btrfs_trans_handle *trans;
1195         struct extent_buffer *leaf;
1196         struct btrfs_path *path;
1197         struct btrfs_file_extent_item *fi;
1198         struct btrfs_key found_key;
1199         u64 cow_start;
1200         u64 cur_offset;
1201         u64 extent_end;
1202         u64 extent_offset;
1203         u64 disk_bytenr;
1204         u64 num_bytes;
1205         u64 disk_num_bytes;
1206         u64 ram_bytes;
1207         int extent_type;
1208         int ret, err;
1209         int type;
1210         int nocow;
1211         int check_prev = 1;
1212         bool nolock;
1213         u64 ino = btrfs_ino(inode);
1214
1215         path = btrfs_alloc_path();
1216         if (!path) {
1217                 extent_clear_unlock_delalloc(inode,
1218                              &BTRFS_I(inode)->io_tree,
1219                              start, end, locked_page,
1220                              EXTENT_CLEAR_UNLOCK_PAGE |
1221                              EXTENT_CLEAR_UNLOCK |
1222                              EXTENT_CLEAR_DELALLOC |
1223                              EXTENT_CLEAR_DIRTY |
1224                              EXTENT_SET_WRITEBACK |
1225                              EXTENT_END_WRITEBACK);
1226                 return -ENOMEM;
1227         }
1228
1229         nolock = btrfs_is_free_space_inode(inode);
1230
1231         if (nolock)
1232                 trans = btrfs_join_transaction_nolock(root);
1233         else
1234                 trans = btrfs_join_transaction(root);
1235
1236         if (IS_ERR(trans)) {
1237                 extent_clear_unlock_delalloc(inode,
1238                              &BTRFS_I(inode)->io_tree,
1239                              start, end, locked_page,
1240                              EXTENT_CLEAR_UNLOCK_PAGE |
1241                              EXTENT_CLEAR_UNLOCK |
1242                              EXTENT_CLEAR_DELALLOC |
1243                              EXTENT_CLEAR_DIRTY |
1244                              EXTENT_SET_WRITEBACK |
1245                              EXTENT_END_WRITEBACK);
1246                 btrfs_free_path(path);
1247                 return PTR_ERR(trans);
1248         }
1249
1250         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1251
1252         cow_start = (u64)-1;
1253         cur_offset = start;
1254         while (1) {
1255                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1256                                                cur_offset, 0);
1257                 if (ret < 0) {
1258                         btrfs_abort_transaction(trans, root, ret);
1259                         goto error;
1260                 }
1261                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1262                         leaf = path->nodes[0];
1263                         btrfs_item_key_to_cpu(leaf, &found_key,
1264                                               path->slots[0] - 1);
1265                         if (found_key.objectid == ino &&
1266                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1267                                 path->slots[0]--;
1268                 }
1269                 check_prev = 0;
1270 next_slot:
1271                 leaf = path->nodes[0];
1272                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1273                         ret = btrfs_next_leaf(root, path);
1274                         if (ret < 0) {
1275                                 btrfs_abort_transaction(trans, root, ret);
1276                                 goto error;
1277                         }
1278                         if (ret > 0)
1279                                 break;
1280                         leaf = path->nodes[0];
1281                 }
1282
1283                 nocow = 0;
1284                 disk_bytenr = 0;
1285                 num_bytes = 0;
1286                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1287
1288                 if (found_key.objectid > ino ||
1289                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1290                     found_key.offset > end)
1291                         break;
1292
1293                 if (found_key.offset > cur_offset) {
1294                         extent_end = found_key.offset;
1295                         extent_type = 0;
1296                         goto out_check;
1297                 }
1298
1299                 fi = btrfs_item_ptr(leaf, path->slots[0],
1300                                     struct btrfs_file_extent_item);
1301                 extent_type = btrfs_file_extent_type(leaf, fi);
1302
1303                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1304                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1305                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1306                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1307                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1308                         extent_end = found_key.offset +
1309                                 btrfs_file_extent_num_bytes(leaf, fi);
1310                         disk_num_bytes =
1311                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1312                         if (extent_end <= start) {
1313                                 path->slots[0]++;
1314                                 goto next_slot;
1315                         }
1316                         if (disk_bytenr == 0)
1317                                 goto out_check;
1318                         if (btrfs_file_extent_compression(leaf, fi) ||
1319                             btrfs_file_extent_encryption(leaf, fi) ||
1320                             btrfs_file_extent_other_encoding(leaf, fi))
1321                                 goto out_check;
1322                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1323                                 goto out_check;
1324                         if (btrfs_extent_readonly(root, disk_bytenr))
1325                                 goto out_check;
1326                         if (btrfs_cross_ref_exist(trans, root, ino,
1327                                                   found_key.offset -
1328                                                   extent_offset, disk_bytenr))
1329                                 goto out_check;
1330                         disk_bytenr += extent_offset;
1331                         disk_bytenr += cur_offset - found_key.offset;
1332                         num_bytes = min(end + 1, extent_end) - cur_offset;
1333                         /*
1334                          * force cow if csum exists in the range.
1335                          * this ensure that csum for a given extent are
1336                          * either valid or do not exist.
1337                          */
1338                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1339                                 goto out_check;
1340                         nocow = 1;
1341                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1342                         extent_end = found_key.offset +
1343                                 btrfs_file_extent_inline_len(leaf, fi);
1344                         extent_end = ALIGN(extent_end, root->sectorsize);
1345                 } else {
1346                         BUG_ON(1);
1347                 }
1348 out_check:
1349                 if (extent_end <= start) {
1350                         path->slots[0]++;
1351                         goto next_slot;
1352                 }
1353                 if (!nocow) {
1354                         if (cow_start == (u64)-1)
1355                                 cow_start = cur_offset;
1356                         cur_offset = extent_end;
1357                         if (cur_offset > end)
1358                                 break;
1359                         path->slots[0]++;
1360                         goto next_slot;
1361                 }
1362
1363                 btrfs_release_path(path);
1364                 if (cow_start != (u64)-1) {
1365                         ret = __cow_file_range(trans, inode, root, locked_page,
1366                                                cow_start, found_key.offset - 1,
1367                                                page_started, nr_written, 1);
1368                         if (ret) {
1369                                 btrfs_abort_transaction(trans, root, ret);
1370                                 goto error;
1371                         }
1372                         cow_start = (u64)-1;
1373                 }
1374
1375                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1376                         struct extent_map *em;
1377                         struct extent_map_tree *em_tree;
1378                         em_tree = &BTRFS_I(inode)->extent_tree;
1379                         em = alloc_extent_map();
1380                         BUG_ON(!em); /* -ENOMEM */
1381                         em->start = cur_offset;
1382                         em->orig_start = found_key.offset - extent_offset;
1383                         em->len = num_bytes;
1384                         em->block_len = num_bytes;
1385                         em->block_start = disk_bytenr;
1386                         em->orig_block_len = disk_num_bytes;
1387                         em->ram_bytes = ram_bytes;
1388                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1389                         em->mod_start = em->start;
1390                         em->mod_len = em->len;
1391                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1392                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1393                         em->generation = -1;
1394                         while (1) {
1395                                 write_lock(&em_tree->lock);
1396                                 ret = add_extent_mapping(em_tree, em, 1);
1397                                 write_unlock(&em_tree->lock);
1398                                 if (ret != -EEXIST) {
1399                                         free_extent_map(em);
1400                                         break;
1401                                 }
1402                                 btrfs_drop_extent_cache(inode, em->start,
1403                                                 em->start + em->len - 1, 0);
1404                         }
1405                         type = BTRFS_ORDERED_PREALLOC;
1406                 } else {
1407                         type = BTRFS_ORDERED_NOCOW;
1408                 }
1409
1410                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1411                                                num_bytes, num_bytes, type);
1412                 BUG_ON(ret); /* -ENOMEM */
1413
1414                 if (root->root_key.objectid ==
1415                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1416                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1417                                                       num_bytes);
1418                         if (ret) {
1419                                 btrfs_abort_transaction(trans, root, ret);
1420                                 goto error;
1421                         }
1422                 }
1423
1424                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1425                                 cur_offset, cur_offset + num_bytes - 1,
1426                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1427                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1428                                 EXTENT_SET_PRIVATE2);
1429                 cur_offset = extent_end;
1430                 if (cur_offset > end)
1431                         break;
1432         }
1433         btrfs_release_path(path);
1434
1435         if (cur_offset <= end && cow_start == (u64)-1) {
1436                 cow_start = cur_offset;
1437                 cur_offset = end;
1438         }
1439
1440         if (cow_start != (u64)-1) {
1441                 ret = __cow_file_range(trans, inode, root, locked_page,
1442                                        cow_start, end,
1443                                        page_started, nr_written, 1);
1444                 if (ret) {
1445                         btrfs_abort_transaction(trans, root, ret);
1446                         goto error;
1447                 }
1448         }
1449
1450 error:
1451         err = btrfs_end_transaction(trans, root);
1452         if (!ret)
1453                 ret = err;
1454
1455         if (ret && cur_offset < end)
1456                 extent_clear_unlock_delalloc(inode,
1457                              &BTRFS_I(inode)->io_tree,
1458                              cur_offset, end, locked_page,
1459                              EXTENT_CLEAR_UNLOCK_PAGE |
1460                              EXTENT_CLEAR_UNLOCK |
1461                              EXTENT_CLEAR_DELALLOC |
1462                              EXTENT_CLEAR_DIRTY |
1463                              EXTENT_SET_WRITEBACK |
1464                              EXTENT_END_WRITEBACK);
1465
1466         btrfs_free_path(path);
1467         return ret;
1468 }
1469
1470 /*
1471  * extent_io.c call back to do delayed allocation processing
1472  */
1473 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1474                               u64 start, u64 end, int *page_started,
1475                               unsigned long *nr_written)
1476 {
1477         int ret;
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479
1480         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1481                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1482                                          page_started, 1, nr_written);
1483         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1484                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1485                                          page_started, 0, nr_written);
1486         } else if (!btrfs_test_opt(root, COMPRESS) &&
1487                    !(BTRFS_I(inode)->force_compress) &&
1488                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1489                 ret = cow_file_range(inode, locked_page, start, end,
1490                                       page_started, nr_written, 1);
1491         } else {
1492                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1493                         &BTRFS_I(inode)->runtime_flags);
1494                 ret = cow_file_range_async(inode, locked_page, start, end,
1495                                            page_started, nr_written);
1496         }
1497         return ret;
1498 }
1499
1500 static void btrfs_split_extent_hook(struct inode *inode,
1501                                     struct extent_state *orig, u64 split)
1502 {
1503         /* not delalloc, ignore it */
1504         if (!(orig->state & EXTENT_DELALLOC))
1505                 return;
1506
1507         spin_lock(&BTRFS_I(inode)->lock);
1508         BTRFS_I(inode)->outstanding_extents++;
1509         spin_unlock(&BTRFS_I(inode)->lock);
1510 }
1511
1512 /*
1513  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1514  * extents so we can keep track of new extents that are just merged onto old
1515  * extents, such as when we are doing sequential writes, so we can properly
1516  * account for the metadata space we'll need.
1517  */
1518 static void btrfs_merge_extent_hook(struct inode *inode,
1519                                     struct extent_state *new,
1520                                     struct extent_state *other)
1521 {
1522         /* not delalloc, ignore it */
1523         if (!(other->state & EXTENT_DELALLOC))
1524                 return;
1525
1526         spin_lock(&BTRFS_I(inode)->lock);
1527         BTRFS_I(inode)->outstanding_extents--;
1528         spin_unlock(&BTRFS_I(inode)->lock);
1529 }
1530
1531 /*
1532  * extent_io.c set_bit_hook, used to track delayed allocation
1533  * bytes in this file, and to maintain the list of inodes that
1534  * have pending delalloc work to be done.
1535  */
1536 static void btrfs_set_bit_hook(struct inode *inode,
1537                                struct extent_state *state, unsigned long *bits)
1538 {
1539
1540         /*
1541          * set_bit and clear bit hooks normally require _irqsave/restore
1542          * but in this case, we are only testing for the DELALLOC
1543          * bit, which is only set or cleared with irqs on
1544          */
1545         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1546                 struct btrfs_root *root = BTRFS_I(inode)->root;
1547                 u64 len = state->end + 1 - state->start;
1548                 bool do_list = !btrfs_is_free_space_inode(inode);
1549
1550                 if (*bits & EXTENT_FIRST_DELALLOC) {
1551                         *bits &= ~EXTENT_FIRST_DELALLOC;
1552                 } else {
1553                         spin_lock(&BTRFS_I(inode)->lock);
1554                         BTRFS_I(inode)->outstanding_extents++;
1555                         spin_unlock(&BTRFS_I(inode)->lock);
1556                 }
1557
1558                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1559                                      root->fs_info->delalloc_batch);
1560                 spin_lock(&BTRFS_I(inode)->lock);
1561                 BTRFS_I(inode)->delalloc_bytes += len;
1562                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1563                                          &BTRFS_I(inode)->runtime_flags)) {
1564                         spin_lock(&root->fs_info->delalloc_lock);
1565                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1566                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1567                                               &root->fs_info->delalloc_inodes);
1568                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1569                                         &BTRFS_I(inode)->runtime_flags);
1570                         }
1571                         spin_unlock(&root->fs_info->delalloc_lock);
1572                 }
1573                 spin_unlock(&BTRFS_I(inode)->lock);
1574         }
1575 }
1576
1577 /*
1578  * extent_io.c clear_bit_hook, see set_bit_hook for why
1579  */
1580 static void btrfs_clear_bit_hook(struct inode *inode,
1581                                  struct extent_state *state,
1582                                  unsigned long *bits)
1583 {
1584         /*
1585          * set_bit and clear bit hooks normally require _irqsave/restore
1586          * but in this case, we are only testing for the DELALLOC
1587          * bit, which is only set or cleared with irqs on
1588          */
1589         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1590                 struct btrfs_root *root = BTRFS_I(inode)->root;
1591                 u64 len = state->end + 1 - state->start;
1592                 bool do_list = !btrfs_is_free_space_inode(inode);
1593
1594                 if (*bits & EXTENT_FIRST_DELALLOC) {
1595                         *bits &= ~EXTENT_FIRST_DELALLOC;
1596                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1597                         spin_lock(&BTRFS_I(inode)->lock);
1598                         BTRFS_I(inode)->outstanding_extents--;
1599                         spin_unlock(&BTRFS_I(inode)->lock);
1600                 }
1601
1602                 if (*bits & EXTENT_DO_ACCOUNTING)
1603                         btrfs_delalloc_release_metadata(inode, len);
1604
1605                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1606                     && do_list)
1607                         btrfs_free_reserved_data_space(inode, len);
1608
1609                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1610                                      root->fs_info->delalloc_batch);
1611                 spin_lock(&BTRFS_I(inode)->lock);
1612                 BTRFS_I(inode)->delalloc_bytes -= len;
1613                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1614                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1615                              &BTRFS_I(inode)->runtime_flags)) {
1616                         spin_lock(&root->fs_info->delalloc_lock);
1617                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1618                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1619                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1620                                           &BTRFS_I(inode)->runtime_flags);
1621                         }
1622                         spin_unlock(&root->fs_info->delalloc_lock);
1623                 }
1624                 spin_unlock(&BTRFS_I(inode)->lock);
1625         }
1626 }
1627
1628 /*
1629  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1630  * we don't create bios that span stripes or chunks
1631  */
1632 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1633                          size_t size, struct bio *bio,
1634                          unsigned long bio_flags)
1635 {
1636         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1637         u64 logical = (u64)bio->bi_sector << 9;
1638         u64 length = 0;
1639         u64 map_length;
1640         int ret;
1641
1642         if (bio_flags & EXTENT_BIO_COMPRESSED)
1643                 return 0;
1644
1645         length = bio->bi_size;
1646         map_length = length;
1647         ret = btrfs_map_block(root->fs_info, rw, logical,
1648                               &map_length, NULL, 0);
1649         /* Will always return 0 with map_multi == NULL */
1650         BUG_ON(ret < 0);
1651         if (map_length < length + size)
1652                 return 1;
1653         return 0;
1654 }
1655
1656 /*
1657  * in order to insert checksums into the metadata in large chunks,
1658  * we wait until bio submission time.   All the pages in the bio are
1659  * checksummed and sums are attached onto the ordered extent record.
1660  *
1661  * At IO completion time the cums attached on the ordered extent record
1662  * are inserted into the btree
1663  */
1664 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1665                                     struct bio *bio, int mirror_num,
1666                                     unsigned long bio_flags,
1667                                     u64 bio_offset)
1668 {
1669         struct btrfs_root *root = BTRFS_I(inode)->root;
1670         int ret = 0;
1671
1672         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1673         BUG_ON(ret); /* -ENOMEM */
1674         return 0;
1675 }
1676
1677 /*
1678  * in order to insert checksums into the metadata in large chunks,
1679  * we wait until bio submission time.   All the pages in the bio are
1680  * checksummed and sums are attached onto the ordered extent record.
1681  *
1682  * At IO completion time the cums attached on the ordered extent record
1683  * are inserted into the btree
1684  */
1685 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1686                           int mirror_num, unsigned long bio_flags,
1687                           u64 bio_offset)
1688 {
1689         struct btrfs_root *root = BTRFS_I(inode)->root;
1690         int ret;
1691
1692         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1693         if (ret)
1694                 bio_endio(bio, ret);
1695         return ret;
1696 }
1697
1698 /*
1699  * extent_io.c submission hook. This does the right thing for csum calculation
1700  * on write, or reading the csums from the tree before a read
1701  */
1702 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1703                           int mirror_num, unsigned long bio_flags,
1704                           u64 bio_offset)
1705 {
1706         struct btrfs_root *root = BTRFS_I(inode)->root;
1707         int ret = 0;
1708         int skip_sum;
1709         int metadata = 0;
1710         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1711
1712         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1713
1714         if (btrfs_is_free_space_inode(inode))
1715                 metadata = 2;
1716
1717         if (!(rw & REQ_WRITE)) {
1718                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1719                 if (ret)
1720                         goto out;
1721
1722                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1723                         ret = btrfs_submit_compressed_read(inode, bio,
1724                                                            mirror_num,
1725                                                            bio_flags);
1726                         goto out;
1727                 } else if (!skip_sum) {
1728                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1729                         if (ret)
1730                                 goto out;
1731                 }
1732                 goto mapit;
1733         } else if (async && !skip_sum) {
1734                 /* csum items have already been cloned */
1735                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1736                         goto mapit;
1737                 /* we're doing a write, do the async checksumming */
1738                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1739                                    inode, rw, bio, mirror_num,
1740                                    bio_flags, bio_offset,
1741                                    __btrfs_submit_bio_start,
1742                                    __btrfs_submit_bio_done);
1743                 goto out;
1744         } else if (!skip_sum) {
1745                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1746                 if (ret)
1747                         goto out;
1748         }
1749
1750 mapit:
1751         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1752
1753 out:
1754         if (ret < 0)
1755                 bio_endio(bio, ret);
1756         return ret;
1757 }
1758
1759 /*
1760  * given a list of ordered sums record them in the inode.  This happens
1761  * at IO completion time based on sums calculated at bio submission time.
1762  */
1763 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1764                              struct inode *inode, u64 file_offset,
1765                              struct list_head *list)
1766 {
1767         struct btrfs_ordered_sum *sum;
1768
1769         list_for_each_entry(sum, list, list) {
1770                 trans->adding_csums = 1;
1771                 btrfs_csum_file_blocks(trans,
1772                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1773                 trans->adding_csums = 0;
1774         }
1775         return 0;
1776 }
1777
1778 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1779                               struct extent_state **cached_state)
1780 {
1781         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1782         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1783                                    cached_state, GFP_NOFS);
1784 }
1785
1786 /* see btrfs_writepage_start_hook for details on why this is required */
1787 struct btrfs_writepage_fixup {
1788         struct page *page;
1789         struct btrfs_work work;
1790 };
1791
1792 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1793 {
1794         struct btrfs_writepage_fixup *fixup;
1795         struct btrfs_ordered_extent *ordered;
1796         struct extent_state *cached_state = NULL;
1797         struct page *page;
1798         struct inode *inode;
1799         u64 page_start;
1800         u64 page_end;
1801         int ret;
1802
1803         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1804         page = fixup->page;
1805 again:
1806         lock_page(page);
1807         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1808                 ClearPageChecked(page);
1809                 goto out_page;
1810         }
1811
1812         inode = page->mapping->host;
1813         page_start = page_offset(page);
1814         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1815
1816         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1817                          &cached_state);
1818
1819         /* already ordered? We're done */
1820         if (PagePrivate2(page))
1821                 goto out;
1822
1823         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1824         if (ordered) {
1825                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1826                                      page_end, &cached_state, GFP_NOFS);
1827                 unlock_page(page);
1828                 btrfs_start_ordered_extent(inode, ordered, 1);
1829                 btrfs_put_ordered_extent(ordered);
1830                 goto again;
1831         }
1832
1833         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1834         if (ret) {
1835                 mapping_set_error(page->mapping, ret);
1836                 end_extent_writepage(page, ret, page_start, page_end);
1837                 ClearPageChecked(page);
1838                 goto out;
1839          }
1840
1841         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1842         ClearPageChecked(page);
1843         set_page_dirty(page);
1844 out:
1845         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1846                              &cached_state, GFP_NOFS);
1847 out_page:
1848         unlock_page(page);
1849         page_cache_release(page);
1850         kfree(fixup);
1851 }
1852
1853 /*
1854  * There are a few paths in the higher layers of the kernel that directly
1855  * set the page dirty bit without asking the filesystem if it is a
1856  * good idea.  This causes problems because we want to make sure COW
1857  * properly happens and the data=ordered rules are followed.
1858  *
1859  * In our case any range that doesn't have the ORDERED bit set
1860  * hasn't been properly setup for IO.  We kick off an async process
1861  * to fix it up.  The async helper will wait for ordered extents, set
1862  * the delalloc bit and make it safe to write the page.
1863  */
1864 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1865 {
1866         struct inode *inode = page->mapping->host;
1867         struct btrfs_writepage_fixup *fixup;
1868         struct btrfs_root *root = BTRFS_I(inode)->root;
1869
1870         /* this page is properly in the ordered list */
1871         if (TestClearPagePrivate2(page))
1872                 return 0;
1873
1874         if (PageChecked(page))
1875                 return -EAGAIN;
1876
1877         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1878         if (!fixup)
1879                 return -EAGAIN;
1880
1881         SetPageChecked(page);
1882         page_cache_get(page);
1883         fixup->work.func = btrfs_writepage_fixup_worker;
1884         fixup->page = page;
1885         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1886         return -EBUSY;
1887 }
1888
1889 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1890                                        struct inode *inode, u64 file_pos,
1891                                        u64 disk_bytenr, u64 disk_num_bytes,
1892                                        u64 num_bytes, u64 ram_bytes,
1893                                        u8 compression, u8 encryption,
1894                                        u16 other_encoding, int extent_type)
1895 {
1896         struct btrfs_root *root = BTRFS_I(inode)->root;
1897         struct btrfs_file_extent_item *fi;
1898         struct btrfs_path *path;
1899         struct extent_buffer *leaf;
1900         struct btrfs_key ins;
1901         int ret;
1902
1903         path = btrfs_alloc_path();
1904         if (!path)
1905                 return -ENOMEM;
1906
1907         path->leave_spinning = 1;
1908
1909         /*
1910          * we may be replacing one extent in the tree with another.
1911          * The new extent is pinned in the extent map, and we don't want
1912          * to drop it from the cache until it is completely in the btree.
1913          *
1914          * So, tell btrfs_drop_extents to leave this extent in the cache.
1915          * the caller is expected to unpin it and allow it to be merged
1916          * with the others.
1917          */
1918         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1919                                  file_pos + num_bytes, 0);
1920         if (ret)
1921                 goto out;
1922
1923         ins.objectid = btrfs_ino(inode);
1924         ins.offset = file_pos;
1925         ins.type = BTRFS_EXTENT_DATA_KEY;
1926         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1927         if (ret)
1928                 goto out;
1929         leaf = path->nodes[0];
1930         fi = btrfs_item_ptr(leaf, path->slots[0],
1931                             struct btrfs_file_extent_item);
1932         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1933         btrfs_set_file_extent_type(leaf, fi, extent_type);
1934         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1935         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1936         btrfs_set_file_extent_offset(leaf, fi, 0);
1937         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1938         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1939         btrfs_set_file_extent_compression(leaf, fi, compression);
1940         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1941         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1942
1943         btrfs_mark_buffer_dirty(leaf);
1944         btrfs_release_path(path);
1945
1946         inode_add_bytes(inode, num_bytes);
1947
1948         ins.objectid = disk_bytenr;
1949         ins.offset = disk_num_bytes;
1950         ins.type = BTRFS_EXTENT_ITEM_KEY;
1951         ret = btrfs_alloc_reserved_file_extent(trans, root,
1952                                         root->root_key.objectid,
1953                                         btrfs_ino(inode), file_pos, &ins);
1954 out:
1955         btrfs_free_path(path);
1956
1957         return ret;
1958 }
1959
1960 /* snapshot-aware defrag */
1961 struct sa_defrag_extent_backref {
1962         struct rb_node node;
1963         struct old_sa_defrag_extent *old;
1964         u64 root_id;
1965         u64 inum;
1966         u64 file_pos;
1967         u64 extent_offset;
1968         u64 num_bytes;
1969         u64 generation;
1970 };
1971
1972 struct old_sa_defrag_extent {
1973         struct list_head list;
1974         struct new_sa_defrag_extent *new;
1975
1976         u64 extent_offset;
1977         u64 bytenr;
1978         u64 offset;
1979         u64 len;
1980         int count;
1981 };
1982
1983 struct new_sa_defrag_extent {
1984         struct rb_root root;
1985         struct list_head head;
1986         struct btrfs_path *path;
1987         struct inode *inode;
1988         u64 file_pos;
1989         u64 len;
1990         u64 bytenr;
1991         u64 disk_len;
1992         u8 compress_type;
1993 };
1994
1995 static int backref_comp(struct sa_defrag_extent_backref *b1,
1996                         struct sa_defrag_extent_backref *b2)
1997 {
1998         if (b1->root_id < b2->root_id)
1999                 return -1;
2000         else if (b1->root_id > b2->root_id)
2001                 return 1;
2002
2003         if (b1->inum < b2->inum)
2004                 return -1;
2005         else if (b1->inum > b2->inum)
2006                 return 1;
2007
2008         if (b1->file_pos < b2->file_pos)
2009                 return -1;
2010         else if (b1->file_pos > b2->file_pos)
2011                 return 1;
2012
2013         /*
2014          * [------------------------------] ===> (a range of space)
2015          *     |<--->|   |<---->| =============> (fs/file tree A)
2016          * |<---------------------------->| ===> (fs/file tree B)
2017          *
2018          * A range of space can refer to two file extents in one tree while
2019          * refer to only one file extent in another tree.
2020          *
2021          * So we may process a disk offset more than one time(two extents in A)
2022          * and locate at the same extent(one extent in B), then insert two same
2023          * backrefs(both refer to the extent in B).
2024          */
2025         return 0;
2026 }
2027
2028 static void backref_insert(struct rb_root *root,
2029                            struct sa_defrag_extent_backref *backref)
2030 {
2031         struct rb_node **p = &root->rb_node;
2032         struct rb_node *parent = NULL;
2033         struct sa_defrag_extent_backref *entry;
2034         int ret;
2035
2036         while (*p) {
2037                 parent = *p;
2038                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
2039
2040                 ret = backref_comp(backref, entry);
2041                 if (ret < 0)
2042                         p = &(*p)->rb_left;
2043                 else
2044                         p = &(*p)->rb_right;
2045         }
2046
2047         rb_link_node(&backref->node, parent, p);
2048         rb_insert_color(&backref->node, root);
2049 }
2050
2051 /*
2052  * Note the backref might has changed, and in this case we just return 0.
2053  */
2054 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2055                                        void *ctx)
2056 {
2057         struct btrfs_file_extent_item *extent;
2058         struct btrfs_fs_info *fs_info;
2059         struct old_sa_defrag_extent *old = ctx;
2060         struct new_sa_defrag_extent *new = old->new;
2061         struct btrfs_path *path = new->path;
2062         struct btrfs_key key;
2063         struct btrfs_root *root;
2064         struct sa_defrag_extent_backref *backref;
2065         struct extent_buffer *leaf;
2066         struct inode *inode = new->inode;
2067         int slot;
2068         int ret;
2069         u64 extent_offset;
2070         u64 num_bytes;
2071
2072         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2073             inum == btrfs_ino(inode))
2074                 return 0;
2075
2076         key.objectid = root_id;
2077         key.type = BTRFS_ROOT_ITEM_KEY;
2078         key.offset = (u64)-1;
2079
2080         fs_info = BTRFS_I(inode)->root->fs_info;
2081         root = btrfs_read_fs_root_no_name(fs_info, &key);
2082         if (IS_ERR(root)) {
2083                 if (PTR_ERR(root) == -ENOENT)
2084                         return 0;
2085                 WARN_ON(1);
2086                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2087                          inum, offset, root_id);
2088                 return PTR_ERR(root);
2089         }
2090
2091         key.objectid = inum;
2092         key.type = BTRFS_EXTENT_DATA_KEY;
2093         if (offset > (u64)-1 << 32)
2094                 key.offset = 0;
2095         else
2096                 key.offset = offset;
2097
2098         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2099         if (ret < 0) {
2100                 WARN_ON(1);
2101                 return ret;
2102         }
2103
2104         while (1) {
2105                 cond_resched();
2106
2107                 leaf = path->nodes[0];
2108                 slot = path->slots[0];
2109
2110                 if (slot >= btrfs_header_nritems(leaf)) {
2111                         ret = btrfs_next_leaf(root, path);
2112                         if (ret < 0) {
2113                                 goto out;
2114                         } else if (ret > 0) {
2115                                 ret = 0;
2116                                 goto out;
2117                         }
2118                         continue;
2119                 }
2120
2121                 path->slots[0]++;
2122
2123                 btrfs_item_key_to_cpu(leaf, &key, slot);
2124
2125                 if (key.objectid > inum)
2126                         goto out;
2127
2128                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2129                         continue;
2130
2131                 extent = btrfs_item_ptr(leaf, slot,
2132                                         struct btrfs_file_extent_item);
2133
2134                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2135                         continue;
2136
2137                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2138                 if (key.offset - extent_offset != offset)
2139                         continue;
2140
2141                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2142                 if (extent_offset >= old->extent_offset + old->offset +
2143                     old->len || extent_offset + num_bytes <=
2144                     old->extent_offset + old->offset)
2145                         continue;
2146
2147                 break;
2148         }
2149
2150         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2151         if (!backref) {
2152                 ret = -ENOENT;
2153                 goto out;
2154         }
2155
2156         backref->root_id = root_id;
2157         backref->inum = inum;
2158         backref->file_pos = offset + extent_offset;
2159         backref->num_bytes = num_bytes;
2160         backref->extent_offset = extent_offset;
2161         backref->generation = btrfs_file_extent_generation(leaf, extent);
2162         backref->old = old;
2163         backref_insert(&new->root, backref);
2164         old->count++;
2165 out:
2166         btrfs_release_path(path);
2167         WARN_ON(ret);
2168         return ret;
2169 }
2170
2171 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2172                                    struct new_sa_defrag_extent *new)
2173 {
2174         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2175         struct old_sa_defrag_extent *old, *tmp;
2176         int ret;
2177
2178         new->path = path;
2179
2180         list_for_each_entry_safe(old, tmp, &new->head, list) {
2181                 ret = iterate_inodes_from_logical(old->bytenr, fs_info,
2182                                                   path, record_one_backref,
2183                                                   old);
2184                 BUG_ON(ret < 0 && ret != -ENOENT);
2185
2186                 /* no backref to be processed for this extent */
2187                 if (!old->count) {
2188                         list_del(&old->list);
2189                         kfree(old);
2190                 }
2191         }
2192
2193         if (list_empty(&new->head))
2194                 return false;
2195
2196         return true;
2197 }
2198
2199 static int relink_is_mergable(struct extent_buffer *leaf,
2200                               struct btrfs_file_extent_item *fi,
2201                               u64 disk_bytenr)
2202 {
2203         if (btrfs_file_extent_disk_bytenr(leaf, fi) != disk_bytenr)
2204                 return 0;
2205
2206         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2207                 return 0;
2208
2209         if (btrfs_file_extent_compression(leaf, fi) ||
2210             btrfs_file_extent_encryption(leaf, fi) ||
2211             btrfs_file_extent_other_encoding(leaf, fi))
2212                 return 0;
2213
2214         return 1;
2215 }
2216
2217 /*
2218  * Note the backref might has changed, and in this case we just return 0.
2219  */
2220 static noinline int relink_extent_backref(struct btrfs_path *path,
2221                                  struct sa_defrag_extent_backref *prev,
2222                                  struct sa_defrag_extent_backref *backref)
2223 {
2224         struct btrfs_file_extent_item *extent;
2225         struct btrfs_file_extent_item *item;
2226         struct btrfs_ordered_extent *ordered;
2227         struct btrfs_trans_handle *trans;
2228         struct btrfs_fs_info *fs_info;
2229         struct btrfs_root *root;
2230         struct btrfs_key key;
2231         struct extent_buffer *leaf;
2232         struct old_sa_defrag_extent *old = backref->old;
2233         struct new_sa_defrag_extent *new = old->new;
2234         struct inode *src_inode = new->inode;
2235         struct inode *inode;
2236         struct extent_state *cached = NULL;
2237         int ret = 0;
2238         u64 start;
2239         u64 len;
2240         u64 lock_start;
2241         u64 lock_end;
2242         bool merge = false;
2243         int index;
2244
2245         if (prev && prev->root_id == backref->root_id &&
2246             prev->inum == backref->inum &&
2247             prev->file_pos + prev->num_bytes == backref->file_pos)
2248                 merge = true;
2249
2250         /* step 1: get root */
2251         key.objectid = backref->root_id;
2252         key.type = BTRFS_ROOT_ITEM_KEY;
2253         key.offset = (u64)-1;
2254
2255         fs_info = BTRFS_I(src_inode)->root->fs_info;
2256         index = srcu_read_lock(&fs_info->subvol_srcu);
2257
2258         root = btrfs_read_fs_root_no_name(fs_info, &key);
2259         if (IS_ERR(root)) {
2260                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2261                 if (PTR_ERR(root) == -ENOENT)
2262                         return 0;
2263                 return PTR_ERR(root);
2264         }
2265         if (btrfs_root_refs(&root->root_item) == 0) {
2266                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2267                 /* parse ENOENT to 0 */
2268                 return 0;
2269         }
2270
2271         /* step 2: get inode */
2272         key.objectid = backref->inum;
2273         key.type = BTRFS_INODE_ITEM_KEY;
2274         key.offset = 0;
2275
2276         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2277         if (IS_ERR(inode)) {
2278                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2279                 return 0;
2280         }
2281
2282         srcu_read_unlock(&fs_info->subvol_srcu, index);
2283
2284         /* step 3: relink backref */
2285         lock_start = backref->file_pos;
2286         lock_end = backref->file_pos + backref->num_bytes - 1;
2287         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2288                          0, &cached);
2289
2290         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2291         if (ordered) {
2292                 btrfs_put_ordered_extent(ordered);
2293                 goto out_unlock;
2294         }
2295
2296         trans = btrfs_join_transaction(root);
2297         if (IS_ERR(trans)) {
2298                 ret = PTR_ERR(trans);
2299                 goto out_unlock;
2300         }
2301
2302         key.objectid = backref->inum;
2303         key.type = BTRFS_EXTENT_DATA_KEY;
2304         key.offset = backref->file_pos;
2305
2306         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2307         if (ret < 0) {
2308                 goto out_free_path;
2309         } else if (ret > 0) {
2310                 ret = 0;
2311                 goto out_free_path;
2312         }
2313
2314         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2315                                 struct btrfs_file_extent_item);
2316
2317         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2318             backref->generation)
2319                 goto out_free_path;
2320
2321         btrfs_release_path(path);
2322
2323         start = backref->file_pos;
2324         if (backref->extent_offset < old->extent_offset + old->offset)
2325                 start += old->extent_offset + old->offset -
2326                          backref->extent_offset;
2327
2328         len = min(backref->extent_offset + backref->num_bytes,
2329                   old->extent_offset + old->offset + old->len);
2330         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2331
2332         ret = btrfs_drop_extents(trans, root, inode, start,
2333                                  start + len, 1);
2334         if (ret)
2335                 goto out_free_path;
2336 again:
2337         key.objectid = btrfs_ino(inode);
2338         key.type = BTRFS_EXTENT_DATA_KEY;
2339         key.offset = start;
2340
2341         path->leave_spinning = 1;
2342         if (merge) {
2343                 struct btrfs_file_extent_item *fi;
2344                 u64 extent_len;
2345                 struct btrfs_key found_key;
2346
2347                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2348                 if (ret < 0)
2349                         goto out_free_path;
2350
2351                 path->slots[0]--;
2352                 leaf = path->nodes[0];
2353                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2354
2355                 fi = btrfs_item_ptr(leaf, path->slots[0],
2356                                     struct btrfs_file_extent_item);
2357                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2358
2359                 if (relink_is_mergable(leaf, fi, new->bytenr) &&
2360                     extent_len + found_key.offset == start) {
2361                         btrfs_set_file_extent_num_bytes(leaf, fi,
2362                                                         extent_len + len);
2363                         btrfs_mark_buffer_dirty(leaf);
2364                         inode_add_bytes(inode, len);
2365
2366                         ret = 1;
2367                         goto out_free_path;
2368                 } else {
2369                         merge = false;
2370                         btrfs_release_path(path);
2371                         goto again;
2372                 }
2373         }
2374
2375         ret = btrfs_insert_empty_item(trans, root, path, &key,
2376                                         sizeof(*extent));
2377         if (ret) {
2378                 btrfs_abort_transaction(trans, root, ret);
2379                 goto out_free_path;
2380         }
2381
2382         leaf = path->nodes[0];
2383         item = btrfs_item_ptr(leaf, path->slots[0],
2384                                 struct btrfs_file_extent_item);
2385         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2386         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2387         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2388         btrfs_set_file_extent_num_bytes(leaf, item, len);
2389         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2390         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2391         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2392         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2393         btrfs_set_file_extent_encryption(leaf, item, 0);
2394         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2395
2396         btrfs_mark_buffer_dirty(leaf);
2397         inode_add_bytes(inode, len);
2398         btrfs_release_path(path);
2399
2400         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2401                         new->disk_len, 0,
2402                         backref->root_id, backref->inum,
2403                         new->file_pos, 0);      /* start - extent_offset */
2404         if (ret) {
2405                 btrfs_abort_transaction(trans, root, ret);
2406                 goto out_free_path;
2407         }
2408
2409         ret = 1;
2410 out_free_path:
2411         btrfs_release_path(path);
2412         path->leave_spinning = 0;
2413         btrfs_end_transaction(trans, root);
2414 out_unlock:
2415         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2416                              &cached, GFP_NOFS);
2417         iput(inode);
2418         return ret;
2419 }
2420
2421 static void relink_file_extents(struct new_sa_defrag_extent *new)
2422 {
2423         struct btrfs_path *path;
2424         struct old_sa_defrag_extent *old, *tmp;
2425         struct sa_defrag_extent_backref *backref;
2426         struct sa_defrag_extent_backref *prev = NULL;
2427         struct inode *inode;
2428         struct btrfs_root *root;
2429         struct rb_node *node;
2430         int ret;
2431
2432         inode = new->inode;
2433         root = BTRFS_I(inode)->root;
2434
2435         path = btrfs_alloc_path();
2436         if (!path)
2437                 return;
2438
2439         if (!record_extent_backrefs(path, new)) {
2440                 btrfs_free_path(path);
2441                 goto out;
2442         }
2443         btrfs_release_path(path);
2444
2445         while (1) {
2446                 node = rb_first(&new->root);
2447                 if (!node)
2448                         break;
2449                 rb_erase(node, &new->root);
2450
2451                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2452
2453                 ret = relink_extent_backref(path, prev, backref);
2454                 WARN_ON(ret < 0);
2455
2456                 kfree(prev);
2457
2458                 if (ret == 1)
2459                         prev = backref;
2460                 else
2461                         prev = NULL;
2462                 cond_resched();
2463         }
2464         kfree(prev);
2465
2466         btrfs_free_path(path);
2467
2468         list_for_each_entry_safe(old, tmp, &new->head, list) {
2469                 list_del(&old->list);
2470                 kfree(old);
2471         }
2472 out:
2473         atomic_dec(&root->fs_info->defrag_running);
2474         wake_up(&root->fs_info->transaction_wait);
2475
2476         kfree(new);
2477 }
2478
2479 static struct new_sa_defrag_extent *
2480 record_old_file_extents(struct inode *inode,
2481                         struct btrfs_ordered_extent *ordered)
2482 {
2483         struct btrfs_root *root = BTRFS_I(inode)->root;
2484         struct btrfs_path *path;
2485         struct btrfs_key key;
2486         struct old_sa_defrag_extent *old, *tmp;
2487         struct new_sa_defrag_extent *new;
2488         int ret;
2489
2490         new = kmalloc(sizeof(*new), GFP_NOFS);
2491         if (!new)
2492                 return NULL;
2493
2494         new->inode = inode;
2495         new->file_pos = ordered->file_offset;
2496         new->len = ordered->len;
2497         new->bytenr = ordered->start;
2498         new->disk_len = ordered->disk_len;
2499         new->compress_type = ordered->compress_type;
2500         new->root = RB_ROOT;
2501         INIT_LIST_HEAD(&new->head);
2502
2503         path = btrfs_alloc_path();
2504         if (!path)
2505                 goto out_kfree;
2506
2507         key.objectid = btrfs_ino(inode);
2508         key.type = BTRFS_EXTENT_DATA_KEY;
2509         key.offset = new->file_pos;
2510
2511         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2512         if (ret < 0)
2513                 goto out_free_path;
2514         if (ret > 0 && path->slots[0] > 0)
2515                 path->slots[0]--;
2516
2517         /* find out all the old extents for the file range */
2518         while (1) {
2519                 struct btrfs_file_extent_item *extent;
2520                 struct extent_buffer *l;
2521                 int slot;
2522                 u64 num_bytes;
2523                 u64 offset;
2524                 u64 end;
2525                 u64 disk_bytenr;
2526                 u64 extent_offset;
2527
2528                 l = path->nodes[0];
2529                 slot = path->slots[0];
2530
2531                 if (slot >= btrfs_header_nritems(l)) {
2532                         ret = btrfs_next_leaf(root, path);
2533                         if (ret < 0)
2534                                 goto out_free_list;
2535                         else if (ret > 0)
2536                                 break;
2537                         continue;
2538                 }
2539
2540                 btrfs_item_key_to_cpu(l, &key, slot);
2541
2542                 if (key.objectid != btrfs_ino(inode))
2543                         break;
2544                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2545                         break;
2546                 if (key.offset >= new->file_pos + new->len)
2547                         break;
2548
2549                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2550
2551                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2552                 if (key.offset + num_bytes < new->file_pos)
2553                         goto next;
2554
2555                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2556                 if (!disk_bytenr)
2557                         goto next;
2558
2559                 extent_offset = btrfs_file_extent_offset(l, extent);
2560
2561                 old = kmalloc(sizeof(*old), GFP_NOFS);
2562                 if (!old)
2563                         goto out_free_list;
2564
2565                 offset = max(new->file_pos, key.offset);
2566                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2567
2568                 old->bytenr = disk_bytenr;
2569                 old->extent_offset = extent_offset;
2570                 old->offset = offset - key.offset;
2571                 old->len = end - offset;
2572                 old->new = new;
2573                 old->count = 0;
2574                 list_add_tail(&old->list, &new->head);
2575 next:
2576                 path->slots[0]++;
2577                 cond_resched();
2578         }
2579
2580         btrfs_free_path(path);
2581         atomic_inc(&root->fs_info->defrag_running);
2582
2583         return new;
2584
2585 out_free_list:
2586         list_for_each_entry_safe(old, tmp, &new->head, list) {
2587                 list_del(&old->list);
2588                 kfree(old);
2589         }
2590 out_free_path:
2591         btrfs_free_path(path);
2592 out_kfree:
2593         kfree(new);
2594         return NULL;
2595 }
2596
2597 /*
2598  * helper function for btrfs_finish_ordered_io, this
2599  * just reads in some of the csum leaves to prime them into ram
2600  * before we start the transaction.  It limits the amount of btree
2601  * reads required while inside the transaction.
2602  */
2603 /* as ordered data IO finishes, this gets called so we can finish
2604  * an ordered extent if the range of bytes in the file it covers are
2605  * fully written.
2606  */
2607 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2608 {
2609         struct inode *inode = ordered_extent->inode;
2610         struct btrfs_root *root = BTRFS_I(inode)->root;
2611         struct btrfs_trans_handle *trans = NULL;
2612         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2613         struct extent_state *cached_state = NULL;
2614         struct new_sa_defrag_extent *new = NULL;
2615         int compress_type = 0;
2616         int ret;
2617         bool nolock;
2618
2619         nolock = btrfs_is_free_space_inode(inode);
2620
2621         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2622                 ret = -EIO;
2623                 goto out;
2624         }
2625
2626         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2627                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2628                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2629                 if (nolock)
2630                         trans = btrfs_join_transaction_nolock(root);
2631                 else
2632                         trans = btrfs_join_transaction(root);
2633                 if (IS_ERR(trans)) {
2634                         ret = PTR_ERR(trans);
2635                         trans = NULL;
2636                         goto out;
2637                 }
2638                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2639                 ret = btrfs_update_inode_fallback(trans, root, inode);
2640                 if (ret) /* -ENOMEM or corruption */
2641                         btrfs_abort_transaction(trans, root, ret);
2642                 goto out;
2643         }
2644
2645         lock_extent_bits(io_tree, ordered_extent->file_offset,
2646                          ordered_extent->file_offset + ordered_extent->len - 1,
2647                          0, &cached_state);
2648
2649         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2650                         ordered_extent->file_offset + ordered_extent->len - 1,
2651                         EXTENT_DEFRAG, 1, cached_state);
2652         if (ret) {
2653                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2654                 if (last_snapshot >= BTRFS_I(inode)->generation)
2655                         /* the inode is shared */
2656                         new = record_old_file_extents(inode, ordered_extent);
2657
2658                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2659                         ordered_extent->file_offset + ordered_extent->len - 1,
2660                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2661         }
2662
2663         if (nolock)
2664                 trans = btrfs_join_transaction_nolock(root);
2665         else
2666                 trans = btrfs_join_transaction(root);
2667         if (IS_ERR(trans)) {
2668                 ret = PTR_ERR(trans);
2669                 trans = NULL;
2670                 goto out_unlock;
2671         }
2672         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2673
2674         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2675                 compress_type = ordered_extent->compress_type;
2676         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2677                 BUG_ON(compress_type);
2678                 ret = btrfs_mark_extent_written(trans, inode,
2679                                                 ordered_extent->file_offset,
2680                                                 ordered_extent->file_offset +
2681                                                 ordered_extent->len);
2682         } else {
2683                 BUG_ON(root == root->fs_info->tree_root);
2684                 ret = insert_reserved_file_extent(trans, inode,
2685                                                 ordered_extent->file_offset,
2686                                                 ordered_extent->start,
2687                                                 ordered_extent->disk_len,
2688                                                 ordered_extent->len,
2689                                                 ordered_extent->len,
2690                                                 compress_type, 0, 0,
2691                                                 BTRFS_FILE_EXTENT_REG);
2692         }
2693         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2694                            ordered_extent->file_offset, ordered_extent->len,
2695                            trans->transid);
2696         if (ret < 0) {
2697                 btrfs_abort_transaction(trans, root, ret);
2698                 goto out_unlock;
2699         }
2700
2701         add_pending_csums(trans, inode, ordered_extent->file_offset,
2702                           &ordered_extent->list);
2703
2704         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2705         ret = btrfs_update_inode_fallback(trans, root, inode);
2706         if (ret) { /* -ENOMEM or corruption */
2707                 btrfs_abort_transaction(trans, root, ret);
2708                 goto out_unlock;
2709         }
2710         ret = 0;
2711 out_unlock:
2712         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2713                              ordered_extent->file_offset +
2714                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2715 out:
2716         if (root != root->fs_info->tree_root)
2717                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2718         if (trans)
2719                 btrfs_end_transaction(trans, root);
2720
2721         if (ret) {
2722                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2723                                       ordered_extent->file_offset +
2724                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2725
2726                 /*
2727                  * If the ordered extent had an IOERR or something else went
2728                  * wrong we need to return the space for this ordered extent
2729                  * back to the allocator.
2730                  */
2731                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2732                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2733                         btrfs_free_reserved_extent(root, ordered_extent->start,
2734                                                    ordered_extent->disk_len);
2735         }
2736
2737
2738         /*
2739          * This needs to be done to make sure anybody waiting knows we are done
2740          * updating everything for this ordered extent.
2741          */
2742         btrfs_remove_ordered_extent(inode, ordered_extent);
2743
2744         /* for snapshot-aware defrag */
2745         if (new)
2746                 relink_file_extents(new);
2747
2748         /* once for us */
2749         btrfs_put_ordered_extent(ordered_extent);
2750         /* once for the tree */
2751         btrfs_put_ordered_extent(ordered_extent);
2752
2753         return ret;
2754 }
2755
2756 static void finish_ordered_fn(struct btrfs_work *work)
2757 {
2758         struct btrfs_ordered_extent *ordered_extent;
2759         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2760         btrfs_finish_ordered_io(ordered_extent);
2761 }
2762
2763 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2764                                 struct extent_state *state, int uptodate)
2765 {
2766         struct inode *inode = page->mapping->host;
2767         struct btrfs_root *root = BTRFS_I(inode)->root;
2768         struct btrfs_ordered_extent *ordered_extent = NULL;
2769         struct btrfs_workers *workers;
2770
2771         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2772
2773         ClearPagePrivate2(page);
2774         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2775                                             end - start + 1, uptodate))
2776                 return 0;
2777
2778         ordered_extent->work.func = finish_ordered_fn;
2779         ordered_extent->work.flags = 0;
2780
2781         if (btrfs_is_free_space_inode(inode))
2782                 workers = &root->fs_info->endio_freespace_worker;
2783         else
2784                 workers = &root->fs_info->endio_write_workers;
2785         btrfs_queue_worker(workers, &ordered_extent->work);
2786
2787         return 0;
2788 }
2789
2790 /*
2791  * when reads are done, we need to check csums to verify the data is correct
2792  * if there's a match, we allow the bio to finish.  If not, the code in
2793  * extent_io.c will try to find good copies for us.
2794  */
2795 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2796                                struct extent_state *state, int mirror)
2797 {
2798         size_t offset = start - page_offset(page);
2799         struct inode *inode = page->mapping->host;
2800         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2801         char *kaddr;
2802         u64 private = ~(u32)0;
2803         int ret;
2804         struct btrfs_root *root = BTRFS_I(inode)->root;
2805         u32 csum = ~(u32)0;
2806         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2807                                       DEFAULT_RATELIMIT_BURST);
2808
2809         if (PageChecked(page)) {
2810                 ClearPageChecked(page);
2811                 goto good;
2812         }
2813
2814         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2815                 goto good;
2816
2817         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2818             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2819                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2820                                   GFP_NOFS);
2821                 return 0;
2822         }
2823
2824         if (state && state->start == start) {
2825                 private = state->private;
2826                 ret = 0;
2827         } else {
2828                 ret = get_state_private(io_tree, start, &private);
2829         }
2830         kaddr = kmap_atomic(page);
2831         if (ret)
2832                 goto zeroit;
2833
2834         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2835         btrfs_csum_final(csum, (char *)&csum);
2836         if (csum != private)
2837                 goto zeroit;
2838
2839         kunmap_atomic(kaddr);
2840 good:
2841         return 0;
2842
2843 zeroit:
2844         if (__ratelimit(&_rs))
2845                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u private %llu",
2846                         (unsigned long long)btrfs_ino(page->mapping->host),
2847                         (unsigned long long)start, csum,
2848                         (unsigned long long)private);
2849         memset(kaddr + offset, 1, end - start + 1);
2850         flush_dcache_page(page);
2851         kunmap_atomic(kaddr);
2852         if (private == 0)
2853                 return 0;
2854         return -EIO;
2855 }
2856
2857 struct delayed_iput {
2858         struct list_head list;
2859         struct inode *inode;
2860 };
2861
2862 /* JDM: If this is fs-wide, why can't we add a pointer to
2863  * btrfs_inode instead and avoid the allocation? */
2864 void btrfs_add_delayed_iput(struct inode *inode)
2865 {
2866         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2867         struct delayed_iput *delayed;
2868
2869         if (atomic_add_unless(&inode->i_count, -1, 1))
2870                 return;
2871
2872         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2873         delayed->inode = inode;
2874
2875         spin_lock(&fs_info->delayed_iput_lock);
2876         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2877         spin_unlock(&fs_info->delayed_iput_lock);
2878 }
2879
2880 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2881 {
2882         LIST_HEAD(list);
2883         struct btrfs_fs_info *fs_info = root->fs_info;
2884         struct delayed_iput *delayed;
2885         int empty;
2886
2887         spin_lock(&fs_info->delayed_iput_lock);
2888         empty = list_empty(&fs_info->delayed_iputs);
2889         spin_unlock(&fs_info->delayed_iput_lock);
2890         if (empty)
2891                 return;
2892
2893         spin_lock(&fs_info->delayed_iput_lock);
2894         list_splice_init(&fs_info->delayed_iputs, &list);
2895         spin_unlock(&fs_info->delayed_iput_lock);
2896
2897         while (!list_empty(&list)) {
2898                 delayed = list_entry(list.next, struct delayed_iput, list);
2899                 list_del(&delayed->list);
2900                 iput(delayed->inode);
2901                 kfree(delayed);
2902         }
2903 }
2904
2905 /*
2906  * This is called in transaction commit time. If there are no orphan
2907  * files in the subvolume, it removes orphan item and frees block_rsv
2908  * structure.
2909  */
2910 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2911                               struct btrfs_root *root)
2912 {
2913         struct btrfs_block_rsv *block_rsv;
2914         int ret;
2915
2916         if (atomic_read(&root->orphan_inodes) ||
2917             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2918                 return;
2919
2920         spin_lock(&root->orphan_lock);
2921         if (atomic_read(&root->orphan_inodes)) {
2922                 spin_unlock(&root->orphan_lock);
2923                 return;
2924         }
2925
2926         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2927                 spin_unlock(&root->orphan_lock);
2928                 return;
2929         }
2930
2931         block_rsv = root->orphan_block_rsv;
2932         root->orphan_block_rsv = NULL;
2933         spin_unlock(&root->orphan_lock);
2934
2935         if (root->orphan_item_inserted &&
2936             btrfs_root_refs(&root->root_item) > 0) {
2937                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2938                                             root->root_key.objectid);
2939                 BUG_ON(ret);
2940                 root->orphan_item_inserted = 0;
2941         }
2942
2943         if (block_rsv) {
2944                 WARN_ON(block_rsv->size > 0);
2945                 btrfs_free_block_rsv(root, block_rsv);
2946         }
2947 }
2948
2949 /*
2950  * This creates an orphan entry for the given inode in case something goes
2951  * wrong in the middle of an unlink/truncate.
2952  *
2953  * NOTE: caller of this function should reserve 5 units of metadata for
2954  *       this function.
2955  */
2956 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2957 {
2958         struct btrfs_root *root = BTRFS_I(inode)->root;
2959         struct btrfs_block_rsv *block_rsv = NULL;
2960         int reserve = 0;
2961         int insert = 0;
2962         int ret;
2963
2964         if (!root->orphan_block_rsv) {
2965                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2966                 if (!block_rsv)
2967                         return -ENOMEM;
2968         }
2969
2970         spin_lock(&root->orphan_lock);
2971         if (!root->orphan_block_rsv) {
2972                 root->orphan_block_rsv = block_rsv;
2973         } else if (block_rsv) {
2974                 btrfs_free_block_rsv(root, block_rsv);
2975                 block_rsv = NULL;
2976         }
2977
2978         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2979                               &BTRFS_I(inode)->runtime_flags)) {
2980 #if 0
2981                 /*
2982                  * For proper ENOSPC handling, we should do orphan
2983                  * cleanup when mounting. But this introduces backward
2984                  * compatibility issue.
2985                  */
2986                 if (!xchg(&root->orphan_item_inserted, 1))
2987                         insert = 2;
2988                 else
2989                         insert = 1;
2990 #endif
2991                 insert = 1;
2992                 atomic_inc(&root->orphan_inodes);
2993         }
2994
2995         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2996                               &BTRFS_I(inode)->runtime_flags))
2997                 reserve = 1;
2998         spin_unlock(&root->orphan_lock);
2999
3000         /* grab metadata reservation from transaction handle */
3001         if (reserve) {
3002                 ret = btrfs_orphan_reserve_metadata(trans, inode);
3003                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
3004         }
3005
3006         /* insert an orphan item to track this unlinked/truncated file */
3007         if (insert >= 1) {
3008                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
3009                 if (ret && ret != -EEXIST) {
3010                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3011                                   &BTRFS_I(inode)->runtime_flags);
3012                         btrfs_abort_transaction(trans, root, ret);
3013                         return ret;
3014                 }
3015                 ret = 0;
3016         }
3017
3018         /* insert an orphan item to track subvolume contains orphan files */
3019         if (insert >= 2) {
3020                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
3021                                                root->root_key.objectid);
3022                 if (ret && ret != -EEXIST) {
3023                         btrfs_abort_transaction(trans, root, ret);
3024                         return ret;
3025                 }
3026         }
3027         return 0;
3028 }
3029
3030 /*
3031  * We have done the truncate/delete so we can go ahead and remove the orphan
3032  * item for this particular inode.
3033  */
3034 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3035                             struct inode *inode)
3036 {
3037         struct btrfs_root *root = BTRFS_I(inode)->root;
3038         int delete_item = 0;
3039         int release_rsv = 0;
3040         int ret = 0;
3041
3042         spin_lock(&root->orphan_lock);
3043         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3044                                &BTRFS_I(inode)->runtime_flags))
3045                 delete_item = 1;
3046
3047         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3048                                &BTRFS_I(inode)->runtime_flags))
3049                 release_rsv = 1;
3050         spin_unlock(&root->orphan_lock);
3051
3052         if (trans && delete_item) {
3053                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
3054                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3055         }
3056
3057         if (release_rsv) {
3058                 btrfs_orphan_release_metadata(inode);
3059                 atomic_dec(&root->orphan_inodes);
3060         }
3061
3062         return 0;
3063 }
3064
3065 /*
3066  * this cleans up any orphans that may be left on the list from the last use
3067  * of this root.
3068  */
3069 int btrfs_orphan_cleanup(struct btrfs_root *root)
3070 {
3071         struct btrfs_path *path;
3072         struct extent_buffer *leaf;
3073         struct btrfs_key key, found_key;
3074         struct btrfs_trans_handle *trans;
3075         struct inode *inode;
3076         u64 last_objectid = 0;
3077         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3078
3079         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3080                 return 0;
3081
3082         path = btrfs_alloc_path();
3083         if (!path) {
3084                 ret = -ENOMEM;
3085                 goto out;
3086         }
3087         path->reada = -1;
3088
3089         key.objectid = BTRFS_ORPHAN_OBJECTID;
3090         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3091         key.offset = (u64)-1;
3092
3093         while (1) {
3094                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3095                 if (ret < 0)
3096                         goto out;
3097
3098                 /*
3099                  * if ret == 0 means we found what we were searching for, which
3100                  * is weird, but possible, so only screw with path if we didn't
3101                  * find the key and see if we have stuff that matches
3102                  */
3103                 if (ret > 0) {
3104                         ret = 0;
3105                         if (path->slots[0] == 0)
3106                                 break;
3107                         path->slots[0]--;
3108                 }
3109
3110                 /* pull out the item */
3111                 leaf = path->nodes[0];
3112                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3113
3114                 /* make sure the item matches what we want */
3115                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3116                         break;
3117                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3118                         break;
3119
3120                 /* release the path since we're done with it */
3121                 btrfs_release_path(path);
3122
3123                 /*
3124                  * this is where we are basically btrfs_lookup, without the
3125                  * crossing root thing.  we store the inode number in the
3126                  * offset of the orphan item.
3127                  */
3128
3129                 if (found_key.offset == last_objectid) {
3130                         btrfs_err(root->fs_info,
3131                                 "Error removing orphan entry, stopping orphan cleanup");
3132                         ret = -EINVAL;
3133                         goto out;
3134                 }
3135
3136                 last_objectid = found_key.offset;
3137
3138                 found_key.objectid = found_key.offset;
3139                 found_key.type = BTRFS_INODE_ITEM_KEY;
3140                 found_key.offset = 0;
3141                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3142                 ret = PTR_RET(inode);
3143                 if (ret && ret != -ESTALE)
3144                         goto out;
3145
3146                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3147                         struct btrfs_root *dead_root;
3148                         struct btrfs_fs_info *fs_info = root->fs_info;
3149                         int is_dead_root = 0;
3150
3151                         /*
3152                          * this is an orphan in the tree root. Currently these
3153                          * could come from 2 sources:
3154                          *  a) a snapshot deletion in progress
3155                          *  b) a free space cache inode
3156                          * We need to distinguish those two, as the snapshot
3157                          * orphan must not get deleted.
3158                          * find_dead_roots already ran before us, so if this
3159                          * is a snapshot deletion, we should find the root
3160                          * in the dead_roots list
3161                          */
3162                         spin_lock(&fs_info->trans_lock);
3163                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3164                                             root_list) {
3165                                 if (dead_root->root_key.objectid ==
3166                                     found_key.objectid) {
3167                                         is_dead_root = 1;
3168                                         break;
3169                                 }
3170                         }
3171                         spin_unlock(&fs_info->trans_lock);
3172                         if (is_dead_root) {
3173                                 /* prevent this orphan from being found again */
3174                                 key.offset = found_key.objectid - 1;
3175                                 continue;
3176                         }
3177                 }
3178                 /*
3179                  * Inode is already gone but the orphan item is still there,
3180                  * kill the orphan item.
3181                  */
3182                 if (ret == -ESTALE) {
3183                         trans = btrfs_start_transaction(root, 1);
3184                         if (IS_ERR(trans)) {
3185                                 ret = PTR_ERR(trans);
3186                                 goto out;
3187                         }
3188                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3189                                 found_key.objectid);
3190                         ret = btrfs_del_orphan_item(trans, root,
3191                                                     found_key.objectid);
3192                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
3193                         btrfs_end_transaction(trans, root);
3194                         continue;
3195                 }
3196
3197                 /*
3198                  * add this inode to the orphan list so btrfs_orphan_del does
3199                  * the proper thing when we hit it
3200                  */
3201                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3202                         &BTRFS_I(inode)->runtime_flags);
3203                 atomic_inc(&root->orphan_inodes);
3204
3205                 /* if we have links, this was a truncate, lets do that */
3206                 if (inode->i_nlink) {
3207                         if (!S_ISREG(inode->i_mode)) {
3208                                 WARN_ON(1);
3209                                 iput(inode);
3210                                 continue;
3211                         }
3212                         nr_truncate++;
3213
3214                         /* 1 for the orphan item deletion. */
3215                         trans = btrfs_start_transaction(root, 1);
3216                         if (IS_ERR(trans)) {
3217                                 ret = PTR_ERR(trans);
3218                                 goto out;
3219                         }
3220                         ret = btrfs_orphan_add(trans, inode);
3221                         btrfs_end_transaction(trans, root);
3222                         if (ret)
3223                                 goto out;
3224
3225                         ret = btrfs_truncate(inode);
3226                         if (ret)
3227                                 btrfs_orphan_del(NULL, inode);
3228                 } else {
3229                         nr_unlink++;
3230                 }
3231
3232                 /* this will do delete_inode and everything for us */
3233                 iput(inode);
3234                 if (ret)
3235                         goto out;
3236         }
3237         /* release the path since we're done with it */
3238         btrfs_release_path(path);
3239
3240         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3241
3242         if (root->orphan_block_rsv)
3243                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3244                                         (u64)-1);
3245
3246         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3247                 trans = btrfs_join_transaction(root);
3248                 if (!IS_ERR(trans))
3249                         btrfs_end_transaction(trans, root);
3250         }
3251
3252         if (nr_unlink)
3253                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3254         if (nr_truncate)
3255                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3256
3257 out:
3258         if (ret)
3259                 btrfs_crit(root->fs_info,
3260                         "could not do orphan cleanup %d", ret);
3261         btrfs_free_path(path);
3262         return ret;
3263 }
3264
3265 /*
3266  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3267  * don't find any xattrs, we know there can't be any acls.
3268  *
3269  * slot is the slot the inode is in, objectid is the objectid of the inode
3270  */
3271 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3272                                           int slot, u64 objectid)
3273 {
3274         u32 nritems = btrfs_header_nritems(leaf);
3275         struct btrfs_key found_key;
3276         int scanned = 0;
3277
3278         slot++;
3279         while (slot < nritems) {
3280                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3281
3282                 /* we found a different objectid, there must not be acls */
3283                 if (found_key.objectid != objectid)
3284                         return 0;
3285
3286                 /* we found an xattr, assume we've got an acl */
3287                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
3288                         return 1;
3289
3290                 /*
3291                  * we found a key greater than an xattr key, there can't
3292                  * be any acls later on
3293                  */
3294                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3295                         return 0;
3296
3297                 slot++;
3298                 scanned++;
3299
3300                 /*
3301                  * it goes inode, inode backrefs, xattrs, extents,
3302                  * so if there are a ton of hard links to an inode there can
3303                  * be a lot of backrefs.  Don't waste time searching too hard,
3304                  * this is just an optimization
3305                  */
3306                 if (scanned >= 8)
3307                         break;
3308         }
3309         /* we hit the end of the leaf before we found an xattr or
3310          * something larger than an xattr.  We have to assume the inode
3311          * has acls
3312          */
3313         return 1;
3314 }
3315
3316 /*
3317  * read an inode from the btree into the in-memory inode
3318  */
3319 static void btrfs_read_locked_inode(struct inode *inode)
3320 {
3321         struct btrfs_path *path;
3322         struct extent_buffer *leaf;
3323         struct btrfs_inode_item *inode_item;
3324         struct btrfs_timespec *tspec;
3325         struct btrfs_root *root = BTRFS_I(inode)->root;
3326         struct btrfs_key location;
3327         int maybe_acls;
3328         u32 rdev;
3329         int ret;
3330         bool filled = false;
3331
3332         ret = btrfs_fill_inode(inode, &rdev);
3333         if (!ret)
3334                 filled = true;
3335
3336         path = btrfs_alloc_path();
3337         if (!path)
3338                 goto make_bad;
3339
3340         path->leave_spinning = 1;
3341         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3342
3343         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3344         if (ret)
3345                 goto make_bad;
3346
3347         leaf = path->nodes[0];
3348
3349         if (filled)
3350                 goto cache_acl;
3351
3352         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3353                                     struct btrfs_inode_item);
3354         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3355         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3356         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3357         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3358         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3359
3360         tspec = btrfs_inode_atime(inode_item);
3361         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3362         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3363
3364         tspec = btrfs_inode_mtime(inode_item);
3365         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3366         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3367
3368         tspec = btrfs_inode_ctime(inode_item);
3369         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3370         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3371
3372         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3373         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3374         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3375
3376         /*
3377          * If we were modified in the current generation and evicted from memory
3378          * and then re-read we need to do a full sync since we don't have any
3379          * idea about which extents were modified before we were evicted from
3380          * cache.
3381          */
3382         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3383                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3384                         &BTRFS_I(inode)->runtime_flags);
3385
3386         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3387         inode->i_generation = BTRFS_I(inode)->generation;
3388         inode->i_rdev = 0;
3389         rdev = btrfs_inode_rdev(leaf, inode_item);
3390
3391         BTRFS_I(inode)->index_cnt = (u64)-1;
3392         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3393 cache_acl:
3394         /*
3395          * try to precache a NULL acl entry for files that don't have
3396          * any xattrs or acls
3397          */
3398         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3399                                            btrfs_ino(inode));
3400         if (!maybe_acls)
3401                 cache_no_acl(inode);
3402
3403         btrfs_free_path(path);
3404
3405         switch (inode->i_mode & S_IFMT) {
3406         case S_IFREG:
3407                 inode->i_mapping->a_ops = &btrfs_aops;
3408                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3409                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3410                 inode->i_fop = &btrfs_file_operations;
3411                 inode->i_op = &btrfs_file_inode_operations;
3412                 break;
3413         case S_IFDIR:
3414                 inode->i_fop = &btrfs_dir_file_operations;
3415                 if (root == root->fs_info->tree_root)
3416                         inode->i_op = &btrfs_dir_ro_inode_operations;
3417                 else
3418                         inode->i_op = &btrfs_dir_inode_operations;
3419                 break;
3420         case S_IFLNK:
3421                 inode->i_op = &btrfs_symlink_inode_operations;
3422                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3423                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3424                 break;
3425         default:
3426                 inode->i_op = &btrfs_special_inode_operations;
3427                 init_special_inode(inode, inode->i_mode, rdev);
3428                 break;
3429         }
3430
3431         btrfs_update_iflags(inode);
3432         return;
3433
3434 make_bad:
3435         btrfs_free_path(path);
3436         make_bad_inode(inode);
3437 }
3438
3439 /*
3440  * given a leaf and an inode, copy the inode fields into the leaf
3441  */
3442 static void fill_inode_item(struct btrfs_trans_handle *trans,
3443                             struct extent_buffer *leaf,
3444                             struct btrfs_inode_item *item,
3445                             struct inode *inode)
3446 {
3447         struct btrfs_map_token token;
3448
3449         btrfs_init_map_token(&token);
3450
3451         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3452         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3453         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3454                                    &token);
3455         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3456         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3457
3458         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3459                                      inode->i_atime.tv_sec, &token);
3460         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3461                                       inode->i_atime.tv_nsec, &token);
3462
3463         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3464                                      inode->i_mtime.tv_sec, &token);
3465         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3466                                       inode->i_mtime.tv_nsec, &token);
3467
3468         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3469                                      inode->i_ctime.tv_sec, &token);
3470         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3471                                       inode->i_ctime.tv_nsec, &token);
3472
3473         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3474                                      &token);
3475         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3476                                          &token);
3477         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3478         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3479         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3480         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3481         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3482 }
3483
3484 /*
3485  * copy everything in the in-memory inode into the btree.
3486  */
3487 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3488                                 struct btrfs_root *root, struct inode *inode)
3489 {
3490         struct btrfs_inode_item *inode_item;
3491         struct btrfs_path *path;
3492         struct extent_buffer *leaf;
3493         int ret;
3494
3495         path = btrfs_alloc_path();
3496         if (!path)
3497                 return -ENOMEM;
3498
3499         path->leave_spinning = 1;
3500         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3501                                  1);
3502         if (ret) {
3503                 if (ret > 0)
3504                         ret = -ENOENT;
3505                 goto failed;
3506         }
3507
3508         btrfs_unlock_up_safe(path, 1);
3509         leaf = path->nodes[0];
3510         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3511                                     struct btrfs_inode_item);
3512
3513         fill_inode_item(trans, leaf, inode_item, inode);
3514         btrfs_mark_buffer_dirty(leaf);
3515         btrfs_set_inode_last_trans(trans, inode);
3516         ret = 0;
3517 failed:
3518         btrfs_free_path(path);
3519         return ret;
3520 }
3521
3522 /*
3523  * copy everything in the in-memory inode into the btree.
3524  */
3525 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3526                                 struct btrfs_root *root, struct inode *inode)
3527 {
3528         int ret;
3529
3530         /*
3531          * If the inode is a free space inode, we can deadlock during commit
3532          * if we put it into the delayed code.
3533          *
3534          * The data relocation inode should also be directly updated
3535          * without delay
3536          */
3537         if (!btrfs_is_free_space_inode(inode)
3538             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3539                 btrfs_update_root_times(trans, root);
3540
3541                 ret = btrfs_delayed_update_inode(trans, root, inode);
3542                 if (!ret)
3543                         btrfs_set_inode_last_trans(trans, inode);
3544                 return ret;
3545         }
3546
3547         return btrfs_update_inode_item(trans, root, inode);
3548 }
3549
3550 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3551                                          struct btrfs_root *root,
3552                                          struct inode *inode)
3553 {
3554         int ret;
3555
3556         ret = btrfs_update_inode(trans, root, inode);
3557         if (ret == -ENOSPC)
3558                 return btrfs_update_inode_item(trans, root, inode);
3559         return ret;
3560 }
3561
3562 /*
3563  * unlink helper that gets used here in inode.c and in the tree logging
3564  * recovery code.  It remove a link in a directory with a given name, and
3565  * also drops the back refs in the inode to the directory
3566  */
3567 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3568                                 struct btrfs_root *root,
3569                                 struct inode *dir, struct inode *inode,
3570                                 const char *name, int name_len)
3571 {
3572         struct btrfs_path *path;
3573         int ret = 0;
3574         struct extent_buffer *leaf;
3575         struct btrfs_dir_item *di;
3576         struct btrfs_key key;
3577         u64 index;
3578         u64 ino = btrfs_ino(inode);
3579         u64 dir_ino = btrfs_ino(dir);
3580
3581         path = btrfs_alloc_path();
3582         if (!path) {
3583                 ret = -ENOMEM;
3584                 goto out;
3585         }
3586
3587         path->leave_spinning = 1;
3588         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3589                                     name, name_len, -1);
3590         if (IS_ERR(di)) {
3591                 ret = PTR_ERR(di);
3592                 goto err;
3593         }
3594         if (!di) {
3595                 ret = -ENOENT;
3596                 goto err;
3597         }
3598         leaf = path->nodes[0];
3599         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3600         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3601         if (ret)
3602                 goto err;
3603         btrfs_release_path(path);
3604
3605         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3606                                   dir_ino, &index);
3607         if (ret) {
3608                 btrfs_info(root->fs_info,
3609                         "failed to delete reference to %.*s, inode %llu parent %llu",
3610                         name_len, name,
3611                         (unsigned long long)ino, (unsigned long long)dir_ino);
3612                 btrfs_abort_transaction(trans, root, ret);
3613                 goto err;
3614         }
3615
3616         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3617         if (ret) {
3618                 btrfs_abort_transaction(trans, root, ret);
3619                 goto err;
3620         }
3621
3622         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3623                                          inode, dir_ino);
3624         if (ret != 0 && ret != -ENOENT) {
3625                 btrfs_abort_transaction(trans, root, ret);
3626                 goto err;
3627         }
3628
3629         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3630                                            dir, index);
3631         if (ret == -ENOENT)
3632                 ret = 0;
3633         else if (ret)
3634                 btrfs_abort_transaction(trans, root, ret);
3635 err:
3636         btrfs_free_path(path);
3637         if (ret)
3638                 goto out;
3639
3640         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3641         inode_inc_iversion(inode);
3642         inode_inc_iversion(dir);
3643         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3644         ret = btrfs_update_inode(trans, root, dir);
3645 out:
3646         return ret;
3647 }
3648
3649 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3650                        struct btrfs_root *root,
3651                        struct inode *dir, struct inode *inode,
3652                        const char *name, int name_len)
3653 {
3654         int ret;
3655         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3656         if (!ret) {
3657                 btrfs_drop_nlink(inode);
3658                 ret = btrfs_update_inode(trans, root, inode);
3659         }
3660         return ret;
3661 }
3662
3663
3664 /* helper to check if there is any shared block in the path */
3665 static int check_path_shared(struct btrfs_root *root,
3666                              struct btrfs_path *path)
3667 {
3668         struct extent_buffer *eb;
3669         int level;
3670         u64 refs = 1;
3671
3672         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
3673                 int ret;
3674
3675                 if (!path->nodes[level])
3676                         break;
3677                 eb = path->nodes[level];
3678                 if (!btrfs_block_can_be_shared(root, eb))
3679                         continue;
3680                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, level, 1,
3681                                                &refs, NULL);
3682                 if (refs > 1)
3683                         return 1;
3684         }
3685         return 0;
3686 }
3687
3688 /*
3689  * helper to start transaction for unlink and rmdir.
3690  *
3691  * unlink and rmdir are special in btrfs, they do not always free space.
3692  * so in enospc case, we should make sure they will free space before
3693  * allowing them to use the global metadata reservation.
3694  */
3695 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3696                                                        struct dentry *dentry)
3697 {
3698         struct btrfs_trans_handle *trans;
3699         struct btrfs_root *root = BTRFS_I(dir)->root;
3700         struct btrfs_path *path;
3701         struct btrfs_dir_item *di;
3702         struct inode *inode = dentry->d_inode;
3703         u64 index;
3704         int check_link = 1;
3705         int err = -ENOSPC;
3706         int ret;
3707         u64 ino = btrfs_ino(inode);
3708         u64 dir_ino = btrfs_ino(dir);
3709
3710         /*
3711          * 1 for the possible orphan item
3712          * 1 for the dir item
3713          * 1 for the dir index
3714          * 1 for the inode ref
3715          * 1 for the inode
3716          */
3717         trans = btrfs_start_transaction(root, 5);
3718         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3719                 return trans;
3720
3721         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3722                 return ERR_PTR(-ENOSPC);
3723
3724         /* check if there is someone else holds reference */
3725         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3726                 return ERR_PTR(-ENOSPC);
3727
3728         if (atomic_read(&inode->i_count) > 2)
3729                 return ERR_PTR(-ENOSPC);
3730
3731         if (xchg(&root->fs_info->enospc_unlink, 1))
3732                 return ERR_PTR(-ENOSPC);
3733
3734         path = btrfs_alloc_path();
3735         if (!path) {
3736                 root->fs_info->enospc_unlink = 0;
3737                 return ERR_PTR(-ENOMEM);
3738         }
3739
3740         /* 1 for the orphan item */
3741         trans = btrfs_start_transaction(root, 1);
3742         if (IS_ERR(trans)) {
3743                 btrfs_free_path(path);
3744                 root->fs_info->enospc_unlink = 0;
3745                 return trans;
3746         }
3747
3748         path->skip_locking = 1;
3749         path->search_commit_root = 1;
3750
3751         ret = btrfs_lookup_inode(trans, root, path,
3752                                 &BTRFS_I(dir)->location, 0);
3753         if (ret < 0) {
3754                 err = ret;
3755                 goto out;
3756         }
3757         if (ret == 0) {
3758                 if (check_path_shared(root, path))
3759                         goto out;
3760         } else {
3761                 check_link = 0;
3762         }
3763         btrfs_release_path(path);
3764
3765         ret = btrfs_lookup_inode(trans, root, path,
3766                                 &BTRFS_I(inode)->location, 0);
3767         if (ret < 0) {
3768                 err = ret;
3769                 goto out;
3770         }
3771         if (ret == 0) {
3772                 if (check_path_shared(root, path))
3773                         goto out;
3774         } else {
3775                 check_link = 0;
3776         }
3777         btrfs_release_path(path);
3778
3779         if (ret == 0 && S_ISREG(inode->i_mode)) {
3780                 ret = btrfs_lookup_file_extent(trans, root, path,
3781                                                ino, (u64)-1, 0);
3782                 if (ret < 0) {
3783                         err = ret;
3784                         goto out;
3785                 }
3786                 BUG_ON(ret == 0); /* Corruption */
3787                 if (check_path_shared(root, path))
3788                         goto out;
3789                 btrfs_release_path(path);
3790         }
3791
3792         if (!check_link) {
3793                 err = 0;
3794                 goto out;
3795         }
3796
3797         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3798                                 dentry->d_name.name, dentry->d_name.len, 0);
3799         if (IS_ERR(di)) {
3800                 err = PTR_ERR(di);
3801                 goto out;
3802         }
3803         if (di) {
3804                 if (check_path_shared(root, path))
3805                         goto out;
3806         } else {
3807                 err = 0;
3808                 goto out;
3809         }
3810         btrfs_release_path(path);
3811
3812         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3813                                         dentry->d_name.len, ino, dir_ino, 0,
3814                                         &index);
3815         if (ret) {
3816                 err = ret;
3817                 goto out;
3818         }
3819
3820         if (check_path_shared(root, path))
3821                 goto out;
3822
3823         btrfs_release_path(path);
3824
3825         /*
3826          * This is a commit root search, if we can lookup inode item and other
3827          * relative items in the commit root, it means the transaction of
3828          * dir/file creation has been committed, and the dir index item that we
3829          * delay to insert has also been inserted into the commit root. So
3830          * we needn't worry about the delayed insertion of the dir index item
3831          * here.
3832          */
3833         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3834                                 dentry->d_name.name, dentry->d_name.len, 0);
3835         if (IS_ERR(di)) {
3836                 err = PTR_ERR(di);
3837                 goto out;
3838         }
3839         BUG_ON(ret == -ENOENT);
3840         if (check_path_shared(root, path))
3841                 goto out;
3842
3843         err = 0;
3844 out:
3845         btrfs_free_path(path);
3846         /* Migrate the orphan reservation over */
3847         if (!err)
3848                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3849                                 &root->fs_info->global_block_rsv,
3850                                 trans->bytes_reserved);
3851
3852         if (err) {
3853                 btrfs_end_transaction(trans, root);
3854                 root->fs_info->enospc_unlink = 0;
3855                 return ERR_PTR(err);
3856         }
3857
3858         trans->block_rsv = &root->fs_info->global_block_rsv;
3859         return trans;
3860 }
3861
3862 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3863                                struct btrfs_root *root)
3864 {
3865         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3866                 btrfs_block_rsv_release(root, trans->block_rsv,
3867                                         trans->bytes_reserved);
3868                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3869                 BUG_ON(!root->fs_info->enospc_unlink);
3870                 root->fs_info->enospc_unlink = 0;
3871         }
3872         btrfs_end_transaction(trans, root);
3873 }
3874
3875 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3876 {
3877         struct btrfs_root *root = BTRFS_I(dir)->root;
3878         struct btrfs_trans_handle *trans;
3879         struct inode *inode = dentry->d_inode;
3880         int ret;
3881
3882         trans = __unlink_start_trans(dir, dentry);
3883         if (IS_ERR(trans))
3884                 return PTR_ERR(trans);
3885
3886         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3887
3888         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3889                                  dentry->d_name.name, dentry->d_name.len);
3890         if (ret)
3891                 goto out;
3892
3893         if (inode->i_nlink == 0) {
3894                 ret = btrfs_orphan_add(trans, inode);
3895                 if (ret)
3896                         goto out;
3897         }
3898
3899 out:
3900         __unlink_end_trans(trans, root);
3901         btrfs_btree_balance_dirty(root);
3902         return ret;
3903 }
3904
3905 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3906                         struct btrfs_root *root,
3907                         struct inode *dir, u64 objectid,
3908                         const char *name, int name_len)
3909 {
3910         struct btrfs_path *path;
3911         struct extent_buffer *leaf;
3912         struct btrfs_dir_item *di;
3913         struct btrfs_key key;
3914         u64 index;
3915         int ret;
3916         u64 dir_ino = btrfs_ino(dir);
3917
3918         path = btrfs_alloc_path();
3919         if (!path)
3920                 return -ENOMEM;
3921
3922         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3923                                    name, name_len, -1);
3924         if (IS_ERR_OR_NULL(di)) {
3925                 if (!di)
3926                         ret = -ENOENT;
3927                 else
3928                         ret = PTR_ERR(di);
3929                 goto out;
3930         }
3931
3932         leaf = path->nodes[0];
3933         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3934         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3935         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3936         if (ret) {
3937                 btrfs_abort_transaction(trans, root, ret);
3938                 goto out;
3939         }
3940         btrfs_release_path(path);
3941
3942         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3943                                  objectid, root->root_key.objectid,
3944                                  dir_ino, &index, name, name_len);
3945         if (ret < 0) {
3946                 if (ret != -ENOENT) {
3947                         btrfs_abort_transaction(trans, root, ret);
3948                         goto out;
3949                 }
3950                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3951                                                  name, name_len);
3952                 if (IS_ERR_OR_NULL(di)) {
3953                         if (!di)
3954                                 ret = -ENOENT;
3955                         else
3956                                 ret = PTR_ERR(di);
3957                         btrfs_abort_transaction(trans, root, ret);
3958                         goto out;
3959                 }
3960
3961                 leaf = path->nodes[0];
3962                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3963                 btrfs_release_path(path);
3964                 index = key.offset;
3965         }
3966         btrfs_release_path(path);
3967
3968         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3969         if (ret) {
3970                 btrfs_abort_transaction(trans, root, ret);
3971                 goto out;
3972         }
3973
3974         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3975         inode_inc_iversion(dir);
3976         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3977         ret = btrfs_update_inode_fallback(trans, root, dir);
3978         if (ret)
3979                 btrfs_abort_transaction(trans, root, ret);
3980 out:
3981         btrfs_free_path(path);
3982         return ret;
3983 }
3984
3985 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3986 {
3987         struct inode *inode = dentry->d_inode;
3988         int err = 0;
3989         struct btrfs_root *root = BTRFS_I(dir)->root;
3990         struct btrfs_trans_handle *trans;
3991
3992         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3993                 return -ENOTEMPTY;
3994         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3995                 return -EPERM;
3996
3997         trans = __unlink_start_trans(dir, dentry);
3998         if (IS_ERR(trans))
3999                 return PTR_ERR(trans);
4000
4001         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
4002                 err = btrfs_unlink_subvol(trans, root, dir,
4003                                           BTRFS_I(inode)->location.objectid,
4004                                           dentry->d_name.name,
4005                                           dentry->d_name.len);
4006                 goto out;
4007         }
4008
4009         err = btrfs_orphan_add(trans, inode);
4010         if (err)
4011                 goto out;
4012
4013         /* now the directory is empty */
4014         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
4015                                  dentry->d_name.name, dentry->d_name.len);
4016         if (!err)
4017                 btrfs_i_size_write(inode, 0);
4018 out:
4019         __unlink_end_trans(trans, root);
4020         btrfs_btree_balance_dirty(root);
4021
4022         return err;
4023 }
4024
4025 /*
4026  * this can truncate away extent items, csum items and directory items.
4027  * It starts at a high offset and removes keys until it can't find
4028  * any higher than new_size
4029  *
4030  * csum items that cross the new i_size are truncated to the new size
4031  * as well.
4032  *
4033  * min_type is the minimum key type to truncate down to.  If set to 0, this
4034  * will kill all the items on this inode, including the INODE_ITEM_KEY.
4035  */
4036 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
4037                                struct btrfs_root *root,
4038                                struct inode *inode,
4039                                u64 new_size, u32 min_type)
4040 {
4041         struct btrfs_path *path;
4042         struct extent_buffer *leaf;
4043         struct btrfs_file_extent_item *fi;
4044         struct btrfs_key key;
4045         struct btrfs_key found_key;
4046         u64 extent_start = 0;
4047         u64 extent_num_bytes = 0;
4048         u64 extent_offset = 0;
4049         u64 item_end = 0;
4050         u32 found_type = (u8)-1;
4051         int found_extent;
4052         int del_item;
4053         int pending_del_nr = 0;
4054         int pending_del_slot = 0;
4055         int extent_type = -1;
4056         int ret;
4057         int err = 0;
4058         u64 ino = btrfs_ino(inode);
4059
4060         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
4061
4062         path = btrfs_alloc_path();
4063         if (!path)
4064                 return -ENOMEM;
4065         path->reada = -1;
4066
4067         /*
4068          * We want to drop from the next block forward in case this new size is
4069          * not block aligned since we will be keeping the last block of the
4070          * extent just the way it is.
4071          */
4072         if (root->ref_cows || root == root->fs_info->tree_root)
4073                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
4074                                         root->sectorsize), (u64)-1, 0);
4075
4076         /*
4077          * This function is also used to drop the items in the log tree before
4078          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
4079          * it is used to drop the loged items. So we shouldn't kill the delayed
4080          * items.
4081          */
4082         if (min_type == 0 && root == BTRFS_I(inode)->root)
4083                 btrfs_kill_delayed_inode_items(inode);
4084
4085         key.objectid = ino;
4086         key.offset = (u64)-1;
4087         key.type = (u8)-1;
4088
4089 search_again:
4090         path->leave_spinning = 1;
4091         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
4092         if (ret < 0) {
4093                 err = ret;
4094                 goto out;
4095         }
4096
4097         if (ret > 0) {
4098                 /* there are no items in the tree for us to truncate, we're
4099                  * done
4100                  */
4101                 if (path->slots[0] == 0)
4102                         goto out;
4103                 path->slots[0]--;
4104         }
4105
4106         while (1) {
4107                 fi = NULL;
4108                 leaf = path->nodes[0];
4109                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4110                 found_type = btrfs_key_type(&found_key);
4111
4112                 if (found_key.objectid != ino)
4113                         break;
4114
4115                 if (found_type < min_type)
4116                         break;
4117
4118                 item_end = found_key.offset;
4119                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
4120                         fi = btrfs_item_ptr(leaf, path->slots[0],
4121                                             struct btrfs_file_extent_item);
4122                         extent_type = btrfs_file_extent_type(leaf, fi);
4123                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4124                                 item_end +=
4125                                     btrfs_file_extent_num_bytes(leaf, fi);
4126                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4127                                 item_end += btrfs_file_extent_inline_len(leaf,
4128                                                                          fi);
4129                         }
4130                         item_end--;
4131                 }
4132                 if (found_type > min_type) {
4133                         del_item = 1;
4134                 } else {
4135                         if (item_end < new_size)
4136                                 break;
4137                         if (found_key.offset >= new_size)
4138                                 del_item = 1;
4139                         else
4140                                 del_item = 0;
4141                 }
4142                 found_extent = 0;
4143                 /* FIXME, shrink the extent if the ref count is only 1 */
4144                 if (found_type != BTRFS_EXTENT_DATA_KEY)
4145                         goto delete;
4146
4147                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
4148                         u64 num_dec;
4149                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
4150                         if (!del_item) {
4151                                 u64 orig_num_bytes =
4152                                         btrfs_file_extent_num_bytes(leaf, fi);
4153                                 extent_num_bytes = ALIGN(new_size -
4154                                                 found_key.offset,
4155                                                 root->sectorsize);
4156                                 btrfs_set_file_extent_num_bytes(leaf, fi,
4157                                                          extent_num_bytes);
4158                                 num_dec = (orig_num_bytes -
4159                                            extent_num_bytes);
4160                                 if (root->ref_cows && extent_start != 0)
4161                                         inode_sub_bytes(inode, num_dec);
4162                                 btrfs_mark_buffer_dirty(leaf);
4163                         } else {
4164                                 extent_num_bytes =
4165                                         btrfs_file_extent_disk_num_bytes(leaf,
4166                                                                          fi);
4167                                 extent_offset = found_key.offset -
4168                                         btrfs_file_extent_offset(leaf, fi);
4169
4170                                 /* FIXME blocksize != 4096 */
4171                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4172                                 if (extent_start != 0) {
4173                                         found_extent = 1;
4174                                         if (root->ref_cows)
4175                                                 inode_sub_bytes(inode, num_dec);
4176                                 }
4177                         }
4178                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4179                         /*
4180                          * we can't truncate inline items that have had
4181                          * special encodings
4182                          */
4183                         if (!del_item &&
4184                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4185                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4186                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4187                                 u32 size = new_size - found_key.offset;
4188
4189                                 if (root->ref_cows) {
4190                                         inode_sub_bytes(inode, item_end + 1 -
4191                                                         new_size);
4192                                 }
4193                                 size =
4194                                     btrfs_file_extent_calc_inline_size(size);
4195                                 btrfs_truncate_item(root, path, size, 1);
4196                         } else if (root->ref_cows) {
4197                                 inode_sub_bytes(inode, item_end + 1 -
4198                                                 found_key.offset);
4199                         }
4200                 }
4201 delete:
4202                 if (del_item) {
4203                         if (!pending_del_nr) {
4204                                 /* no pending yet, add ourselves */
4205                                 pending_del_slot = path->slots[0];
4206                                 pending_del_nr = 1;
4207                         } else if (pending_del_nr &&
4208                                    path->slots[0] + 1 == pending_del_slot) {
4209                                 /* hop on the pending chunk */
4210                                 pending_del_nr++;
4211                                 pending_del_slot = path->slots[0];
4212                         } else {
4213                                 BUG();
4214                         }
4215                 } else {
4216                         break;
4217                 }
4218                 if (found_extent && (root->ref_cows ||
4219                                      root == root->fs_info->tree_root)) {
4220                         btrfs_set_path_blocking(path);
4221                         ret = btrfs_free_extent(trans, root, extent_start,
4222                                                 extent_num_bytes, 0,
4223                                                 btrfs_header_owner(leaf),
4224                                                 ino, extent_offset, 0);
4225                         BUG_ON(ret);
4226                 }
4227
4228                 if (found_type == BTRFS_INODE_ITEM_KEY)
4229                         break;
4230
4231                 if (path->slots[0] == 0 ||
4232                     path->slots[0] != pending_del_slot) {
4233                         if (pending_del_nr) {
4234                                 ret = btrfs_del_items(trans, root, path,
4235                                                 pending_del_slot,
4236                                                 pending_del_nr);
4237                                 if (ret) {
4238                                         btrfs_abort_transaction(trans,
4239                                                                 root, ret);
4240                                         goto error;
4241                                 }
4242                                 pending_del_nr = 0;
4243                         }
4244                         btrfs_release_path(path);
4245                         goto search_again;
4246                 } else {
4247                         path->slots[0]--;
4248                 }
4249         }
4250 out:
4251         if (pending_del_nr) {
4252                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4253                                       pending_del_nr);
4254                 if (ret)
4255                         btrfs_abort_transaction(trans, root, ret);
4256         }
4257 error:
4258         btrfs_free_path(path);
4259         return err;
4260 }
4261
4262 /*
4263  * btrfs_truncate_page - read, zero a chunk and write a page
4264  * @inode - inode that we're zeroing
4265  * @from - the offset to start zeroing
4266  * @len - the length to zero, 0 to zero the entire range respective to the
4267  *      offset
4268  * @front - zero up to the offset instead of from the offset on
4269  *
4270  * This will find the page for the "from" offset and cow the page and zero the
4271  * part we want to zero.  This is used with truncate and hole punching.
4272  */
4273 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4274                         int front)
4275 {
4276         struct address_space *mapping = inode->i_mapping;
4277         struct btrfs_root *root = BTRFS_I(inode)->root;
4278         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4279         struct btrfs_ordered_extent *ordered;
4280         struct extent_state *cached_state = NULL;
4281         char *kaddr;
4282         u32 blocksize = root->sectorsize;
4283         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4284         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4285         struct page *page;
4286         gfp_t mask = btrfs_alloc_write_mask(mapping);
4287         int ret = 0;
4288         u64 page_start;
4289         u64 page_end;
4290
4291         if ((offset & (blocksize - 1)) == 0 &&
4292             (!len || ((len & (blocksize - 1)) == 0)))
4293                 goto out;
4294         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4295         if (ret)
4296                 goto out;
4297
4298 again:
4299         page = find_or_create_page(mapping, index, mask);
4300         if (!page) {
4301                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4302                 ret = -ENOMEM;
4303                 goto out;
4304         }
4305
4306         page_start = page_offset(page);
4307         page_end = page_start + PAGE_CACHE_SIZE - 1;
4308
4309         if (!PageUptodate(page)) {
4310                 ret = btrfs_readpage(NULL, page);
4311                 lock_page(page);
4312                 if (page->mapping != mapping) {
4313                         unlock_page(page);
4314                         page_cache_release(page);
4315                         goto again;
4316                 }
4317                 if (!PageUptodate(page)) {
4318                         ret = -EIO;
4319                         goto out_unlock;
4320                 }
4321         }
4322         wait_on_page_writeback(page);
4323
4324         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4325         set_page_extent_mapped(page);
4326
4327         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4328         if (ordered) {
4329                 unlock_extent_cached(io_tree, page_start, page_end,
4330                                      &cached_state, GFP_NOFS);
4331                 unlock_page(page);
4332                 page_cache_release(page);
4333                 btrfs_start_ordered_extent(inode, ordered, 1);
4334                 btrfs_put_ordered_extent(ordered);
4335                 goto again;
4336         }
4337
4338         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4339                           EXTENT_DIRTY | EXTENT_DELALLOC |
4340                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4341                           0, 0, &cached_state, GFP_NOFS);
4342
4343         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4344                                         &cached_state);
4345         if (ret) {
4346                 unlock_extent_cached(io_tree, page_start, page_end,
4347                                      &cached_state, GFP_NOFS);
4348                 goto out_unlock;
4349         }
4350
4351         if (offset != PAGE_CACHE_SIZE) {
4352                 if (!len)
4353                         len = PAGE_CACHE_SIZE - offset;
4354                 kaddr = kmap(page);
4355                 if (front)
4356                         memset(kaddr, 0, offset);
4357                 else
4358                         memset(kaddr + offset, 0, len);
4359                 flush_dcache_page(page);
4360                 kunmap(page);
4361         }
4362         ClearPageChecked(page);
4363         set_page_dirty(page);
4364         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4365                              GFP_NOFS);
4366
4367 out_unlock:
4368         if (ret)
4369                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4370         unlock_page(page);
4371         page_cache_release(page);
4372 out:
4373         return ret;
4374 }
4375
4376 /*
4377  * This function puts in dummy file extents for the area we're creating a hole
4378  * for.  So if we are truncating this file to a larger size we need to insert
4379  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4380  * the range between oldsize and size
4381  */
4382 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4383 {
4384         struct btrfs_trans_handle *trans;
4385         struct btrfs_root *root = BTRFS_I(inode)->root;
4386         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4387         struct extent_map *em = NULL;
4388         struct extent_state *cached_state = NULL;
4389         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4390         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4391         u64 block_end = ALIGN(size, root->sectorsize);
4392         u64 last_byte;
4393         u64 cur_offset;
4394         u64 hole_size;
4395         int err = 0;
4396
4397         if (size <= hole_start)
4398                 return 0;
4399
4400         while (1) {
4401                 struct btrfs_ordered_extent *ordered;
4402                 btrfs_wait_ordered_range(inode, hole_start,
4403                                          block_end - hole_start);
4404                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4405                                  &cached_state);
4406                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
4407                 if (!ordered)
4408                         break;
4409                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4410                                      &cached_state, GFP_NOFS);
4411                 btrfs_put_ordered_extent(ordered);
4412         }
4413
4414         cur_offset = hole_start;
4415         while (1) {
4416                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4417                                 block_end - cur_offset, 0);
4418                 if (IS_ERR(em)) {
4419                         err = PTR_ERR(em);
4420                         em = NULL;
4421                         break;
4422                 }
4423                 last_byte = min(extent_map_end(em), block_end);
4424                 last_byte = ALIGN(last_byte , root->sectorsize);
4425                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4426                         struct extent_map *hole_em;
4427                         hole_size = last_byte - cur_offset;
4428
4429                         trans = btrfs_start_transaction(root, 3);
4430                         if (IS_ERR(trans)) {
4431                                 err = PTR_ERR(trans);
4432                                 break;
4433                         }
4434
4435                         err = btrfs_drop_extents(trans, root, inode,
4436                                                  cur_offset,
4437                                                  cur_offset + hole_size, 1);
4438                         if (err) {
4439                                 btrfs_abort_transaction(trans, root, err);
4440                                 btrfs_end_transaction(trans, root);
4441                                 break;
4442                         }
4443
4444                         err = btrfs_insert_file_extent(trans, root,
4445                                         btrfs_ino(inode), cur_offset, 0,
4446                                         0, hole_size, 0, hole_size,
4447                                         0, 0, 0);
4448                         if (err) {
4449                                 btrfs_abort_transaction(trans, root, err);
4450                                 btrfs_end_transaction(trans, root);
4451                                 break;
4452                         }
4453
4454                         btrfs_drop_extent_cache(inode, cur_offset,
4455                                                 cur_offset + hole_size - 1, 0);
4456                         hole_em = alloc_extent_map();
4457                         if (!hole_em) {
4458                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4459                                         &BTRFS_I(inode)->runtime_flags);
4460                                 goto next;
4461                         }
4462                         hole_em->start = cur_offset;
4463                         hole_em->len = hole_size;
4464                         hole_em->orig_start = cur_offset;
4465
4466                         hole_em->block_start = EXTENT_MAP_HOLE;
4467                         hole_em->block_len = 0;
4468                         hole_em->orig_block_len = 0;
4469                         hole_em->ram_bytes = hole_size;
4470                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4471                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4472                         hole_em->generation = trans->transid;
4473
4474                         while (1) {
4475                                 write_lock(&em_tree->lock);
4476                                 err = add_extent_mapping(em_tree, hole_em, 1);
4477                                 write_unlock(&em_tree->lock);
4478                                 if (err != -EEXIST)
4479                                         break;
4480                                 btrfs_drop_extent_cache(inode, cur_offset,
4481                                                         cur_offset +
4482                                                         hole_size - 1, 0);
4483                         }
4484                         free_extent_map(hole_em);
4485 next:
4486                         btrfs_update_inode(trans, root, inode);
4487                         btrfs_end_transaction(trans, root);
4488                 }
4489                 free_extent_map(em);
4490                 em = NULL;
4491                 cur_offset = last_byte;
4492                 if (cur_offset >= block_end)
4493                         break;
4494         }
4495
4496         free_extent_map(em);
4497         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4498                              GFP_NOFS);
4499         return err;
4500 }
4501
4502 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4503 {
4504         struct btrfs_root *root = BTRFS_I(inode)->root;
4505         struct btrfs_trans_handle *trans;
4506         loff_t oldsize = i_size_read(inode);
4507         loff_t newsize = attr->ia_size;
4508         int mask = attr->ia_valid;
4509         int ret;
4510
4511         if (newsize == oldsize)
4512                 return 0;
4513
4514         /*
4515          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4516          * special case where we need to update the times despite not having
4517          * these flags set.  For all other operations the VFS set these flags
4518          * explicitly if it wants a timestamp update.
4519          */
4520         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4521                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4522
4523         if (newsize > oldsize) {
4524                 truncate_pagecache(inode, oldsize, newsize);
4525                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4526                 if (ret)
4527                         return ret;
4528
4529                 trans = btrfs_start_transaction(root, 1);
4530                 if (IS_ERR(trans))
4531                         return PTR_ERR(trans);
4532
4533                 i_size_write(inode, newsize);
4534                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4535                 ret = btrfs_update_inode(trans, root, inode);
4536                 btrfs_end_transaction(trans, root);
4537         } else {
4538
4539                 /*
4540                  * We're truncating a file that used to have good data down to
4541                  * zero. Make sure it gets into the ordered flush list so that
4542                  * any new writes get down to disk quickly.
4543                  */
4544                 if (newsize == 0)
4545                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4546                                 &BTRFS_I(inode)->runtime_flags);
4547
4548                 /*
4549                  * 1 for the orphan item we're going to add
4550                  * 1 for the orphan item deletion.
4551                  */
4552                 trans = btrfs_start_transaction(root, 2);
4553                 if (IS_ERR(trans))
4554                         return PTR_ERR(trans);
4555
4556                 /*
4557                  * We need to do this in case we fail at _any_ point during the
4558                  * actual truncate.  Once we do the truncate_setsize we could
4559                  * invalidate pages which forces any outstanding ordered io to
4560                  * be instantly completed which will give us extents that need
4561                  * to be truncated.  If we fail to get an orphan inode down we
4562                  * could have left over extents that were never meant to live,
4563                  * so we need to garuntee from this point on that everything
4564                  * will be consistent.
4565                  */
4566                 ret = btrfs_orphan_add(trans, inode);
4567                 btrfs_end_transaction(trans, root);
4568                 if (ret)
4569                         return ret;
4570
4571                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4572                 truncate_setsize(inode, newsize);
4573
4574                 /* Disable nonlocked read DIO to avoid the end less truncate */
4575                 btrfs_inode_block_unlocked_dio(inode);
4576                 inode_dio_wait(inode);
4577                 btrfs_inode_resume_unlocked_dio(inode);
4578
4579                 ret = btrfs_truncate(inode);
4580                 if (ret && inode->i_nlink)
4581                         btrfs_orphan_del(NULL, inode);
4582         }
4583
4584         return ret;
4585 }
4586
4587 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4588 {
4589         struct inode *inode = dentry->d_inode;
4590         struct btrfs_root *root = BTRFS_I(inode)->root;
4591         int err;
4592
4593         if (btrfs_root_readonly(root))
4594                 return -EROFS;
4595
4596         err = inode_change_ok(inode, attr);
4597         if (err)
4598                 return err;
4599
4600         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4601                 err = btrfs_setsize(inode, attr);
4602                 if (err)
4603                         return err;
4604         }
4605
4606         if (attr->ia_valid) {
4607                 setattr_copy(inode, attr);
4608                 inode_inc_iversion(inode);
4609                 err = btrfs_dirty_inode(inode);
4610
4611                 if (!err && attr->ia_valid & ATTR_MODE)
4612                         err = btrfs_acl_chmod(inode);
4613         }
4614
4615         return err;
4616 }
4617
4618 void btrfs_evict_inode(struct inode *inode)
4619 {
4620         struct btrfs_trans_handle *trans;
4621         struct btrfs_root *root = BTRFS_I(inode)->root;
4622         struct btrfs_block_rsv *rsv, *global_rsv;
4623         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4624         int ret;
4625
4626         trace_btrfs_inode_evict(inode);
4627
4628         truncate_inode_pages(&inode->i_data, 0);
4629         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
4630                                btrfs_is_free_space_inode(inode)))
4631                 goto no_delete;
4632
4633         if (is_bad_inode(inode)) {
4634                 btrfs_orphan_del(NULL, inode);
4635                 goto no_delete;
4636         }
4637         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4638         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4639
4640         if (root->fs_info->log_root_recovering) {
4641                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4642                                  &BTRFS_I(inode)->runtime_flags));
4643                 goto no_delete;
4644         }
4645
4646         if (inode->i_nlink > 0) {
4647                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
4648                 goto no_delete;
4649         }
4650
4651         ret = btrfs_commit_inode_delayed_inode(inode);
4652         if (ret) {
4653                 btrfs_orphan_del(NULL, inode);
4654                 goto no_delete;
4655         }
4656
4657         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4658         if (!rsv) {
4659                 btrfs_orphan_del(NULL, inode);
4660                 goto no_delete;
4661         }
4662         rsv->size = min_size;
4663         rsv->failfast = 1;
4664         global_rsv = &root->fs_info->global_block_rsv;
4665
4666         btrfs_i_size_write(inode, 0);
4667
4668         /*
4669          * This is a bit simpler than btrfs_truncate since we've already
4670          * reserved our space for our orphan item in the unlink, so we just
4671          * need to reserve some slack space in case we add bytes and update
4672          * inode item when doing the truncate.
4673          */
4674         while (1) {
4675                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4676                                              BTRFS_RESERVE_FLUSH_LIMIT);
4677
4678                 /*
4679                  * Try and steal from the global reserve since we will
4680                  * likely not use this space anyway, we want to try as
4681                  * hard as possible to get this to work.
4682                  */
4683                 if (ret)
4684                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4685
4686                 if (ret) {
4687                         btrfs_warn(root->fs_info,
4688                                 "Could not get space for a delete, will truncate on mount %d",
4689                                 ret);
4690                         btrfs_orphan_del(NULL, inode);
4691                         btrfs_free_block_rsv(root, rsv);
4692                         goto no_delete;
4693                 }
4694
4695                 trans = btrfs_join_transaction(root);
4696                 if (IS_ERR(trans)) {
4697                         btrfs_orphan_del(NULL, inode);
4698                         btrfs_free_block_rsv(root, rsv);
4699                         goto no_delete;
4700                 }
4701
4702                 trans->block_rsv = rsv;
4703
4704                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4705                 if (ret != -ENOSPC)
4706                         break;
4707
4708                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4709                 btrfs_end_transaction(trans, root);
4710                 trans = NULL;
4711                 btrfs_btree_balance_dirty(root);
4712         }
4713
4714         btrfs_free_block_rsv(root, rsv);
4715
4716         if (ret == 0) {
4717                 trans->block_rsv = root->orphan_block_rsv;
4718                 ret = btrfs_orphan_del(trans, inode);
4719                 BUG_ON(ret);
4720         }
4721
4722         trans->block_rsv = &root->fs_info->trans_block_rsv;
4723         if (!(root == root->fs_info->tree_root ||
4724               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4725                 btrfs_return_ino(root, btrfs_ino(inode));
4726
4727         btrfs_end_transaction(trans, root);
4728         btrfs_btree_balance_dirty(root);
4729 no_delete:
4730         btrfs_remove_delayed_node(inode);
4731         clear_inode(inode);
4732         return;
4733 }
4734
4735 /*
4736  * this returns the key found in the dir entry in the location pointer.
4737  * If no dir entries were found, location->objectid is 0.
4738  */
4739 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4740                                struct btrfs_key *location)
4741 {
4742         const char *name = dentry->d_name.name;
4743         int namelen = dentry->d_name.len;
4744         struct btrfs_dir_item *di;
4745         struct btrfs_path *path;
4746         struct btrfs_root *root = BTRFS_I(dir)->root;
4747         int ret = 0;
4748
4749         path = btrfs_alloc_path();
4750         if (!path)
4751                 return -ENOMEM;
4752
4753         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4754                                     namelen, 0);
4755         if (IS_ERR(di))
4756                 ret = PTR_ERR(di);
4757
4758         if (IS_ERR_OR_NULL(di))
4759                 goto out_err;
4760
4761         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4762 out:
4763         btrfs_free_path(path);
4764         return ret;
4765 out_err:
4766         location->objectid = 0;
4767         goto out;
4768 }
4769
4770 /*
4771  * when we hit a tree root in a directory, the btrfs part of the inode
4772  * needs to be changed to reflect the root directory of the tree root.  This
4773  * is kind of like crossing a mount point.
4774  */
4775 static int fixup_tree_root_location(struct btrfs_root *root,
4776                                     struct inode *dir,
4777                                     struct dentry *dentry,
4778                                     struct btrfs_key *location,
4779                                     struct btrfs_root **sub_root)
4780 {
4781         struct btrfs_path *path;
4782         struct btrfs_root *new_root;
4783         struct btrfs_root_ref *ref;
4784         struct extent_buffer *leaf;
4785         int ret;
4786         int err = 0;
4787
4788         path = btrfs_alloc_path();
4789         if (!path) {
4790                 err = -ENOMEM;
4791                 goto out;
4792         }
4793
4794         err = -ENOENT;
4795         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4796                                   BTRFS_I(dir)->root->root_key.objectid,
4797                                   location->objectid);
4798         if (ret) {
4799                 if (ret < 0)
4800                         err = ret;
4801                 goto out;
4802         }
4803
4804         leaf = path->nodes[0];
4805         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4806         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4807             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4808                 goto out;
4809
4810         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4811                                    (unsigned long)(ref + 1),
4812                                    dentry->d_name.len);
4813         if (ret)
4814                 goto out;
4815
4816         btrfs_release_path(path);
4817
4818         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4819         if (IS_ERR(new_root)) {
4820                 err = PTR_ERR(new_root);
4821                 goto out;
4822         }
4823
4824         if (btrfs_root_refs(&new_root->root_item) == 0) {
4825                 err = -ENOENT;
4826                 goto out;
4827         }
4828
4829         *sub_root = new_root;
4830         location->objectid = btrfs_root_dirid(&new_root->root_item);
4831         location->type = BTRFS_INODE_ITEM_KEY;
4832         location->offset = 0;
4833         err = 0;
4834 out:
4835         btrfs_free_path(path);
4836         return err;
4837 }
4838
4839 static void inode_tree_add(struct inode *inode)
4840 {
4841         struct btrfs_root *root = BTRFS_I(inode)->root;
4842         struct btrfs_inode *entry;
4843         struct rb_node **p;
4844         struct rb_node *parent;
4845         u64 ino = btrfs_ino(inode);
4846
4847         if (inode_unhashed(inode))
4848                 return;
4849 again:
4850         parent = NULL;
4851         spin_lock(&root->inode_lock);
4852         p = &root->inode_tree.rb_node;
4853         while (*p) {
4854                 parent = *p;
4855                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4856
4857                 if (ino < btrfs_ino(&entry->vfs_inode))
4858                         p = &parent->rb_left;
4859                 else if (ino > btrfs_ino(&entry->vfs_inode))
4860                         p = &parent->rb_right;
4861                 else {
4862                         WARN_ON(!(entry->vfs_inode.i_state &
4863                                   (I_WILL_FREE | I_FREEING)));
4864                         rb_erase(parent, &root->inode_tree);
4865                         RB_CLEAR_NODE(parent);
4866                         spin_unlock(&root->inode_lock);
4867                         goto again;
4868                 }
4869         }
4870         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4871         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4872         spin_unlock(&root->inode_lock);
4873 }
4874
4875 static void inode_tree_del(struct inode *inode)
4876 {
4877         struct btrfs_root *root = BTRFS_I(inode)->root;
4878         int empty = 0;
4879
4880         spin_lock(&root->inode_lock);
4881         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4882                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4883                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4884                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4885         }
4886         spin_unlock(&root->inode_lock);
4887
4888         /*
4889          * Free space cache has inodes in the tree root, but the tree root has a
4890          * root_refs of 0, so this could end up dropping the tree root as a
4891          * snapshot, so we need the extra !root->fs_info->tree_root check to
4892          * make sure we don't drop it.
4893          */
4894         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4895             root != root->fs_info->tree_root) {
4896                 synchronize_srcu(&root->fs_info->subvol_srcu);
4897                 spin_lock(&root->inode_lock);
4898                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4899                 spin_unlock(&root->inode_lock);
4900                 if (empty)
4901                         btrfs_add_dead_root(root);
4902         }
4903 }
4904
4905 void btrfs_invalidate_inodes(struct btrfs_root *root)
4906 {
4907         struct rb_node *node;
4908         struct rb_node *prev;
4909         struct btrfs_inode *entry;
4910         struct inode *inode;
4911         u64 objectid = 0;
4912
4913         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4914
4915         spin_lock(&root->inode_lock);
4916 again:
4917         node = root->inode_tree.rb_node;
4918         prev = NULL;
4919         while (node) {
4920                 prev = node;
4921                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4922
4923                 if (objectid < btrfs_ino(&entry->vfs_inode))
4924                         node = node->rb_left;
4925                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4926                         node = node->rb_right;
4927                 else
4928                         break;
4929         }
4930         if (!node) {
4931                 while (prev) {
4932                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4933                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4934                                 node = prev;
4935                                 break;
4936                         }
4937                         prev = rb_next(prev);
4938                 }
4939         }
4940         while (node) {
4941                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4942                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4943                 inode = igrab(&entry->vfs_inode);
4944                 if (inode) {
4945                         spin_unlock(&root->inode_lock);
4946                         if (atomic_read(&inode->i_count) > 1)
4947                                 d_prune_aliases(inode);
4948                         /*
4949                          * btrfs_drop_inode will have it removed from
4950                          * the inode cache when its usage count
4951                          * hits zero.
4952                          */
4953                         iput(inode);
4954                         cond_resched();
4955                         spin_lock(&root->inode_lock);
4956                         goto again;
4957                 }
4958
4959                 if (cond_resched_lock(&root->inode_lock))
4960                         goto again;
4961
4962                 node = rb_next(node);
4963         }
4964         spin_unlock(&root->inode_lock);
4965 }
4966
4967 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4968 {
4969         struct btrfs_iget_args *args = p;
4970         inode->i_ino = args->ino;
4971         BTRFS_I(inode)->root = args->root;
4972         return 0;
4973 }
4974
4975 static int btrfs_find_actor(struct inode *inode, void *opaque)
4976 {
4977         struct btrfs_iget_args *args = opaque;
4978         return args->ino == btrfs_ino(inode) &&
4979                 args->root == BTRFS_I(inode)->root;
4980 }
4981
4982 static struct inode *btrfs_iget_locked(struct super_block *s,
4983                                        u64 objectid,
4984                                        struct btrfs_root *root)
4985 {
4986         struct inode *inode;
4987         struct btrfs_iget_args args;
4988         args.ino = objectid;
4989         args.root = root;
4990
4991         inode = iget5_locked(s, objectid, btrfs_find_actor,
4992                              btrfs_init_locked_inode,
4993                              (void *)&args);
4994         return inode;
4995 }
4996
4997 /* Get an inode object given its location and corresponding root.
4998  * Returns in *is_new if the inode was read from disk
4999  */
5000 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
5001                          struct btrfs_root *root, int *new)
5002 {
5003         struct inode *inode;
5004
5005         inode = btrfs_iget_locked(s, location->objectid, root);
5006         if (!inode)
5007                 return ERR_PTR(-ENOMEM);
5008
5009         if (inode->i_state & I_NEW) {
5010                 BTRFS_I(inode)->root = root;
5011                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
5012                 btrfs_read_locked_inode(inode);
5013                 if (!is_bad_inode(inode)) {
5014                         inode_tree_add(inode);
5015                         unlock_new_inode(inode);
5016                         if (new)
5017                                 *new = 1;
5018                 } else {
5019                         unlock_new_inode(inode);
5020                         iput(inode);
5021                         inode = ERR_PTR(-ESTALE);
5022                 }
5023         }
5024
5025         return inode;
5026 }
5027
5028 static struct inode *new_simple_dir(struct super_block *s,
5029                                     struct btrfs_key *key,
5030                                     struct btrfs_root *root)
5031 {
5032         struct inode *inode = new_inode(s);
5033
5034         if (!inode)
5035                 return ERR_PTR(-ENOMEM);
5036
5037         BTRFS_I(inode)->root = root;
5038         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
5039         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
5040
5041         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
5042         inode->i_op = &btrfs_dir_ro_inode_operations;
5043         inode->i_fop = &simple_dir_operations;
5044         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
5045         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5046
5047         return inode;
5048 }
5049
5050 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
5051 {
5052         struct inode *inode;
5053         struct btrfs_root *root = BTRFS_I(dir)->root;
5054         struct btrfs_root *sub_root = root;
5055         struct btrfs_key location;
5056         int index;
5057         int ret = 0;
5058
5059         if (dentry->d_name.len > BTRFS_NAME_LEN)
5060                 return ERR_PTR(-ENAMETOOLONG);
5061
5062         ret = btrfs_inode_by_name(dir, dentry, &location);
5063         if (ret < 0)
5064                 return ERR_PTR(ret);
5065
5066         if (location.objectid == 0)
5067                 return NULL;
5068
5069         if (location.type == BTRFS_INODE_ITEM_KEY) {
5070                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
5071                 return inode;
5072         }
5073
5074         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
5075
5076         index = srcu_read_lock(&root->fs_info->subvol_srcu);
5077         ret = fixup_tree_root_location(root, dir, dentry,
5078                                        &location, &sub_root);
5079         if (ret < 0) {
5080                 if (ret != -ENOENT)
5081                         inode = ERR_PTR(ret);
5082                 else
5083                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5084         } else {
5085                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5086         }
5087         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5088
5089         if (!IS_ERR(inode) && root != sub_root) {
5090                 down_read(&root->fs_info->cleanup_work_sem);
5091                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5092                         ret = btrfs_orphan_cleanup(sub_root);
5093                 up_read(&root->fs_info->cleanup_work_sem);
5094                 if (ret)
5095                         inode = ERR_PTR(ret);
5096         }
5097
5098         return inode;
5099 }
5100
5101 static int btrfs_dentry_delete(const struct dentry *dentry)
5102 {
5103         struct btrfs_root *root;
5104         struct inode *inode = dentry->d_inode;
5105
5106         if (!inode && !IS_ROOT(dentry))
5107                 inode = dentry->d_parent->d_inode;
5108
5109         if (inode) {
5110                 root = BTRFS_I(inode)->root;
5111                 if (btrfs_root_refs(&root->root_item) == 0)
5112                         return 1;
5113
5114                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5115                         return 1;
5116         }
5117         return 0;
5118 }
5119
5120 static void btrfs_dentry_release(struct dentry *dentry)
5121 {
5122         if (dentry->d_fsdata)
5123                 kfree(dentry->d_fsdata);
5124 }
5125
5126 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5127                                    unsigned int flags)
5128 {
5129         struct dentry *ret;
5130
5131         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
5132         return ret;
5133 }
5134
5135 unsigned char btrfs_filetype_table[] = {
5136         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5137 };
5138
5139 static int btrfs_real_readdir(struct file *filp, void *dirent,
5140                               filldir_t filldir)
5141 {
5142         struct inode *inode = file_inode(filp);
5143         struct btrfs_root *root = BTRFS_I(inode)->root;
5144         struct btrfs_item *item;
5145         struct btrfs_dir_item *di;
5146         struct btrfs_key key;
5147         struct btrfs_key found_key;
5148         struct btrfs_path *path;
5149         struct list_head ins_list;
5150         struct list_head del_list;
5151         int ret;
5152         struct extent_buffer *leaf;
5153         int slot;
5154         unsigned char d_type;
5155         int over = 0;
5156         u32 di_cur;
5157         u32 di_total;
5158         u32 di_len;
5159         int key_type = BTRFS_DIR_INDEX_KEY;
5160         char tmp_name[32];
5161         char *name_ptr;
5162         int name_len;
5163         int is_curr = 0;        /* filp->f_pos points to the current index? */
5164
5165         /* FIXME, use a real flag for deciding about the key type */
5166         if (root->fs_info->tree_root == root)
5167                 key_type = BTRFS_DIR_ITEM_KEY;
5168
5169         /* special case for "." */
5170         if (filp->f_pos == 0) {
5171                 over = filldir(dirent, ".", 1,
5172                                filp->f_pos, btrfs_ino(inode), DT_DIR);
5173                 if (over)
5174                         return 0;
5175                 filp->f_pos = 1;
5176         }
5177         /* special case for .., just use the back ref */
5178         if (filp->f_pos == 1) {
5179                 u64 pino = parent_ino(filp->f_path.dentry);
5180                 over = filldir(dirent, "..", 2,
5181                                filp->f_pos, pino, DT_DIR);
5182                 if (over)
5183                         return 0;
5184                 filp->f_pos = 2;
5185         }
5186         path = btrfs_alloc_path();
5187         if (!path)
5188                 return -ENOMEM;
5189
5190         path->reada = 1;
5191
5192         if (key_type == BTRFS_DIR_INDEX_KEY) {
5193                 INIT_LIST_HEAD(&ins_list);
5194                 INIT_LIST_HEAD(&del_list);
5195                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5196         }
5197
5198         btrfs_set_key_type(&key, key_type);
5199         key.offset = filp->f_pos;
5200         key.objectid = btrfs_ino(inode);
5201
5202         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5203         if (ret < 0)
5204                 goto err;
5205
5206         while (1) {
5207                 leaf = path->nodes[0];
5208                 slot = path->slots[0];
5209                 if (slot >= btrfs_header_nritems(leaf)) {
5210                         ret = btrfs_next_leaf(root, path);
5211                         if (ret < 0)
5212                                 goto err;
5213                         else if (ret > 0)
5214                                 break;
5215                         continue;
5216                 }
5217
5218                 item = btrfs_item_nr(leaf, slot);
5219                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5220
5221                 if (found_key.objectid != key.objectid)
5222                         break;
5223                 if (btrfs_key_type(&found_key) != key_type)
5224                         break;
5225                 if (found_key.offset < filp->f_pos)
5226                         goto next;
5227                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5228                     btrfs_should_delete_dir_index(&del_list,
5229                                                   found_key.offset))
5230                         goto next;
5231
5232                 filp->f_pos = found_key.offset;
5233                 is_curr = 1;
5234
5235                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5236                 di_cur = 0;
5237                 di_total = btrfs_item_size(leaf, item);
5238
5239                 while (di_cur < di_total) {
5240                         struct btrfs_key location;
5241
5242                         if (verify_dir_item(root, leaf, di))
5243                                 break;
5244
5245                         name_len = btrfs_dir_name_len(leaf, di);
5246                         if (name_len <= sizeof(tmp_name)) {
5247                                 name_ptr = tmp_name;
5248                         } else {
5249                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5250                                 if (!name_ptr) {
5251                                         ret = -ENOMEM;
5252                                         goto err;
5253                                 }
5254                         }
5255                         read_extent_buffer(leaf, name_ptr,
5256                                            (unsigned long)(di + 1), name_len);
5257
5258                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5259                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5260
5261
5262                         /* is this a reference to our own snapshot? If so
5263                          * skip it.
5264                          *
5265                          * In contrast to old kernels, we insert the snapshot's
5266                          * dir item and dir index after it has been created, so
5267                          * we won't find a reference to our own snapshot. We
5268                          * still keep the following code for backward
5269                          * compatibility.
5270                          */
5271                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5272                             location.objectid == root->root_key.objectid) {
5273                                 over = 0;
5274                                 goto skip;
5275                         }
5276                         over = filldir(dirent, name_ptr, name_len,
5277                                        found_key.offset, location.objectid,
5278                                        d_type);
5279
5280 skip:
5281                         if (name_ptr != tmp_name)
5282                                 kfree(name_ptr);
5283
5284                         if (over)
5285                                 goto nopos;
5286                         di_len = btrfs_dir_name_len(leaf, di) +
5287                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5288                         di_cur += di_len;
5289                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5290                 }
5291 next:
5292                 path->slots[0]++;
5293         }
5294
5295         if (key_type == BTRFS_DIR_INDEX_KEY) {
5296                 if (is_curr)
5297                         filp->f_pos++;
5298                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
5299                                                       &ins_list);
5300                 if (ret)
5301                         goto nopos;
5302         }
5303
5304         /* Reached end of directory/root. Bump pos past the last item. */
5305         if (key_type == BTRFS_DIR_INDEX_KEY)
5306                 /*
5307                  * 32-bit glibc will use getdents64, but then strtol -
5308                  * so the last number we can serve is this.
5309                  */
5310                 filp->f_pos = 0x7fffffff;
5311         else
5312                 filp->f_pos++;
5313 nopos:
5314         ret = 0;
5315 err:
5316         if (key_type == BTRFS_DIR_INDEX_KEY)
5317                 btrfs_put_delayed_items(&ins_list, &del_list);
5318         btrfs_free_path(path);
5319         return ret;
5320 }
5321
5322 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5323 {
5324         struct btrfs_root *root = BTRFS_I(inode)->root;
5325         struct btrfs_trans_handle *trans;
5326         int ret = 0;
5327         bool nolock = false;
5328
5329         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5330                 return 0;
5331
5332         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5333                 nolock = true;
5334
5335         if (wbc->sync_mode == WB_SYNC_ALL) {
5336                 if (nolock)
5337                         trans = btrfs_join_transaction_nolock(root);
5338                 else
5339                         trans = btrfs_join_transaction(root);
5340                 if (IS_ERR(trans))
5341                         return PTR_ERR(trans);
5342                 ret = btrfs_commit_transaction(trans, root);
5343         }
5344         return ret;
5345 }
5346
5347 /*
5348  * This is somewhat expensive, updating the tree every time the
5349  * inode changes.  But, it is most likely to find the inode in cache.
5350  * FIXME, needs more benchmarking...there are no reasons other than performance
5351  * to keep or drop this code.
5352  */
5353 static int btrfs_dirty_inode(struct inode *inode)
5354 {
5355         struct btrfs_root *root = BTRFS_I(inode)->root;
5356         struct btrfs_trans_handle *trans;
5357         int ret;
5358
5359         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5360                 return 0;
5361
5362         trans = btrfs_join_transaction(root);
5363         if (IS_ERR(trans))
5364                 return PTR_ERR(trans);
5365
5366         ret = btrfs_update_inode(trans, root, inode);
5367         if (ret && ret == -ENOSPC) {
5368                 /* whoops, lets try again with the full transaction */
5369                 btrfs_end_transaction(trans, root);
5370                 trans = btrfs_start_transaction(root, 1);
5371                 if (IS_ERR(trans))
5372                         return PTR_ERR(trans);
5373
5374                 ret = btrfs_update_inode(trans, root, inode);
5375         }
5376         btrfs_end_transaction(trans, root);
5377         if (BTRFS_I(inode)->delayed_node)
5378                 btrfs_balance_delayed_items(root);
5379
5380         return ret;
5381 }
5382
5383 /*
5384  * This is a copy of file_update_time.  We need this so we can return error on
5385  * ENOSPC for updating the inode in the case of file write and mmap writes.
5386  */
5387 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5388                              int flags)
5389 {
5390         struct btrfs_root *root = BTRFS_I(inode)->root;
5391
5392         if (btrfs_root_readonly(root))
5393                 return -EROFS;
5394
5395         if (flags & S_VERSION)
5396                 inode_inc_iversion(inode);
5397         if (flags & S_CTIME)
5398                 inode->i_ctime = *now;
5399         if (flags & S_MTIME)
5400                 inode->i_mtime = *now;
5401         if (flags & S_ATIME)
5402                 inode->i_atime = *now;
5403         return btrfs_dirty_inode(inode);
5404 }
5405
5406 /*
5407  * find the highest existing sequence number in a directory
5408  * and then set the in-memory index_cnt variable to reflect
5409  * free sequence numbers
5410  */
5411 static int btrfs_set_inode_index_count(struct inode *inode)
5412 {
5413         struct btrfs_root *root = BTRFS_I(inode)->root;
5414         struct btrfs_key key, found_key;
5415         struct btrfs_path *path;
5416         struct extent_buffer *leaf;
5417         int ret;
5418
5419         key.objectid = btrfs_ino(inode);
5420         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5421         key.offset = (u64)-1;
5422
5423         path = btrfs_alloc_path();
5424         if (!path)
5425                 return -ENOMEM;
5426
5427         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5428         if (ret < 0)
5429                 goto out;
5430         /* FIXME: we should be able to handle this */
5431         if (ret == 0)
5432                 goto out;
5433         ret = 0;
5434
5435         /*
5436          * MAGIC NUMBER EXPLANATION:
5437          * since we search a directory based on f_pos we have to start at 2
5438          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5439          * else has to start at 2
5440          */
5441         if (path->slots[0] == 0) {
5442                 BTRFS_I(inode)->index_cnt = 2;
5443                 goto out;
5444         }
5445
5446         path->slots[0]--;
5447
5448         leaf = path->nodes[0];
5449         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5450
5451         if (found_key.objectid != btrfs_ino(inode) ||
5452             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5453                 BTRFS_I(inode)->index_cnt = 2;
5454                 goto out;
5455         }
5456
5457         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5458 out:
5459         btrfs_free_path(path);
5460         return ret;
5461 }
5462
5463 /*
5464  * helper to find a free sequence number in a given directory.  This current
5465  * code is very simple, later versions will do smarter things in the btree
5466  */
5467 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5468 {
5469         int ret = 0;
5470
5471         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5472                 ret = btrfs_inode_delayed_dir_index_count(dir);
5473                 if (ret) {
5474                         ret = btrfs_set_inode_index_count(dir);
5475                         if (ret)
5476                                 return ret;
5477                 }
5478         }
5479
5480         *index = BTRFS_I(dir)->index_cnt;
5481         BTRFS_I(dir)->index_cnt++;
5482
5483         return ret;
5484 }
5485
5486 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5487                                      struct btrfs_root *root,
5488                                      struct inode *dir,
5489                                      const char *name, int name_len,
5490                                      u64 ref_objectid, u64 objectid,
5491                                      umode_t mode, u64 *index)
5492 {
5493         struct inode *inode;
5494         struct btrfs_inode_item *inode_item;
5495         struct btrfs_key *location;
5496         struct btrfs_path *path;
5497         struct btrfs_inode_ref *ref;
5498         struct btrfs_key key[2];
5499         u32 sizes[2];
5500         unsigned long ptr;
5501         int ret;
5502         int owner;
5503
5504         path = btrfs_alloc_path();
5505         if (!path)
5506                 return ERR_PTR(-ENOMEM);
5507
5508         inode = new_inode(root->fs_info->sb);
5509         if (!inode) {
5510                 btrfs_free_path(path);
5511                 return ERR_PTR(-ENOMEM);
5512         }
5513
5514         /*
5515          * we have to initialize this early, so we can reclaim the inode
5516          * number if we fail afterwards in this function.
5517          */
5518         inode->i_ino = objectid;
5519
5520         if (dir) {
5521                 trace_btrfs_inode_request(dir);
5522
5523                 ret = btrfs_set_inode_index(dir, index);
5524                 if (ret) {
5525                         btrfs_free_path(path);
5526                         iput(inode);
5527                         return ERR_PTR(ret);
5528                 }
5529         }
5530         /*
5531          * index_cnt is ignored for everything but a dir,
5532          * btrfs_get_inode_index_count has an explanation for the magic
5533          * number
5534          */
5535         BTRFS_I(inode)->index_cnt = 2;
5536         BTRFS_I(inode)->root = root;
5537         BTRFS_I(inode)->generation = trans->transid;
5538         inode->i_generation = BTRFS_I(inode)->generation;
5539
5540         /*
5541          * We could have gotten an inode number from somebody who was fsynced
5542          * and then removed in this same transaction, so let's just set full
5543          * sync since it will be a full sync anyway and this will blow away the
5544          * old info in the log.
5545          */
5546         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5547
5548         if (S_ISDIR(mode))
5549                 owner = 0;
5550         else
5551                 owner = 1;
5552
5553         key[0].objectid = objectid;
5554         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5555         key[0].offset = 0;
5556
5557         /*
5558          * Start new inodes with an inode_ref. This is slightly more
5559          * efficient for small numbers of hard links since they will
5560          * be packed into one item. Extended refs will kick in if we
5561          * add more hard links than can fit in the ref item.
5562          */
5563         key[1].objectid = objectid;
5564         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5565         key[1].offset = ref_objectid;
5566
5567         sizes[0] = sizeof(struct btrfs_inode_item);
5568         sizes[1] = name_len + sizeof(*ref);
5569
5570         path->leave_spinning = 1;
5571         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5572         if (ret != 0)
5573                 goto fail;
5574
5575         inode_init_owner(inode, dir, mode);
5576         inode_set_bytes(inode, 0);
5577         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5578         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5579                                   struct btrfs_inode_item);
5580         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5581                              sizeof(*inode_item));
5582         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5583
5584         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5585                              struct btrfs_inode_ref);
5586         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5587         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5588         ptr = (unsigned long)(ref + 1);
5589         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5590
5591         btrfs_mark_buffer_dirty(path->nodes[0]);
5592         btrfs_free_path(path);
5593
5594         location = &BTRFS_I(inode)->location;
5595         location->objectid = objectid;
5596         location->offset = 0;
5597         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5598
5599         btrfs_inherit_iflags(inode, dir);
5600
5601         if (S_ISREG(mode)) {
5602                 if (btrfs_test_opt(root, NODATASUM))
5603                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5604                 if (btrfs_test_opt(root, NODATACOW))
5605                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5606                                 BTRFS_INODE_NODATASUM;
5607         }
5608
5609         insert_inode_hash(inode);
5610         inode_tree_add(inode);
5611
5612         trace_btrfs_inode_new(inode);
5613         btrfs_set_inode_last_trans(trans, inode);
5614
5615         btrfs_update_root_times(trans, root);
5616
5617         return inode;
5618 fail:
5619         if (dir)
5620                 BTRFS_I(dir)->index_cnt--;
5621         btrfs_free_path(path);
5622         iput(inode);
5623         return ERR_PTR(ret);
5624 }
5625
5626 static inline u8 btrfs_inode_type(struct inode *inode)
5627 {
5628         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5629 }
5630
5631 /*
5632  * utility function to add 'inode' into 'parent_inode' with
5633  * a give name and a given sequence number.
5634  * if 'add_backref' is true, also insert a backref from the
5635  * inode to the parent directory.
5636  */
5637 int btrfs_add_link(struct btrfs_trans_handle *trans,
5638                    struct inode *parent_inode, struct inode *inode,
5639                    const char *name, int name_len, int add_backref, u64 index)
5640 {
5641         int ret = 0;
5642         struct btrfs_key key;
5643         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5644         u64 ino = btrfs_ino(inode);
5645         u64 parent_ino = btrfs_ino(parent_inode);
5646
5647         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5648                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5649         } else {
5650                 key.objectid = ino;
5651                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5652                 key.offset = 0;
5653         }
5654
5655         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5656                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5657                                          key.objectid, root->root_key.objectid,
5658                                          parent_ino, index, name, name_len);
5659         } else if (add_backref) {
5660                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5661                                              parent_ino, index);
5662         }
5663
5664         /* Nothing to clean up yet */
5665         if (ret)
5666                 return ret;
5667
5668         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5669                                     parent_inode, &key,
5670                                     btrfs_inode_type(inode), index);
5671         if (ret == -EEXIST || ret == -EOVERFLOW)
5672                 goto fail_dir_item;
5673         else if (ret) {
5674                 btrfs_abort_transaction(trans, root, ret);
5675                 return ret;
5676         }
5677
5678         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5679                            name_len * 2);
5680         inode_inc_iversion(parent_inode);
5681         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5682         ret = btrfs_update_inode(trans, root, parent_inode);
5683         if (ret)
5684                 btrfs_abort_transaction(trans, root, ret);
5685         return ret;
5686
5687 fail_dir_item:
5688         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5689                 u64 local_index;
5690                 int err;
5691                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5692                                  key.objectid, root->root_key.objectid,
5693                                  parent_ino, &local_index, name, name_len);
5694
5695         } else if (add_backref) {
5696                 u64 local_index;
5697                 int err;
5698
5699                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5700                                           ino, parent_ino, &local_index);
5701         }
5702         return ret;
5703 }
5704
5705 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5706                             struct inode *dir, struct dentry *dentry,
5707                             struct inode *inode, int backref, u64 index)
5708 {
5709         int err = btrfs_add_link(trans, dir, inode,
5710                                  dentry->d_name.name, dentry->d_name.len,
5711                                  backref, index);
5712         if (err > 0)
5713                 err = -EEXIST;
5714         return err;
5715 }
5716
5717 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5718                         umode_t mode, dev_t rdev)
5719 {
5720         struct btrfs_trans_handle *trans;
5721         struct btrfs_root *root = BTRFS_I(dir)->root;
5722         struct inode *inode = NULL;
5723         int err;
5724         int drop_inode = 0;
5725         u64 objectid;
5726         u64 index = 0;
5727
5728         if (!new_valid_dev(rdev))
5729                 return -EINVAL;
5730
5731         /*
5732          * 2 for inode item and ref
5733          * 2 for dir items
5734          * 1 for xattr if selinux is on
5735          */
5736         trans = btrfs_start_transaction(root, 5);
5737         if (IS_ERR(trans))
5738                 return PTR_ERR(trans);
5739
5740         err = btrfs_find_free_ino(root, &objectid);
5741         if (err)
5742                 goto out_unlock;
5743
5744         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5745                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5746                                 mode, &index);
5747         if (IS_ERR(inode)) {
5748                 err = PTR_ERR(inode);
5749                 goto out_unlock;
5750         }
5751
5752         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5753         if (err) {
5754                 drop_inode = 1;
5755                 goto out_unlock;
5756         }
5757
5758         /*
5759         * If the active LSM wants to access the inode during
5760         * d_instantiate it needs these. Smack checks to see
5761         * if the filesystem supports xattrs by looking at the
5762         * ops vector.
5763         */
5764
5765         inode->i_op = &btrfs_special_inode_operations;
5766         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5767         if (err)
5768                 drop_inode = 1;
5769         else {
5770                 init_special_inode(inode, inode->i_mode, rdev);
5771                 btrfs_update_inode(trans, root, inode);
5772                 d_instantiate(dentry, inode);
5773         }
5774 out_unlock:
5775         btrfs_end_transaction(trans, root);
5776         btrfs_btree_balance_dirty(root);
5777         if (drop_inode) {
5778                 inode_dec_link_count(inode);
5779                 iput(inode);
5780         }
5781         return err;
5782 }
5783
5784 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5785                         umode_t mode, bool excl)
5786 {
5787         struct btrfs_trans_handle *trans;
5788         struct btrfs_root *root = BTRFS_I(dir)->root;
5789         struct inode *inode = NULL;
5790         int drop_inode_on_err = 0;
5791         int err;
5792         u64 objectid;
5793         u64 index = 0;
5794
5795         /*
5796          * 2 for inode item and ref
5797          * 2 for dir items
5798          * 1 for xattr if selinux is on
5799          */
5800         trans = btrfs_start_transaction(root, 5);
5801         if (IS_ERR(trans))
5802                 return PTR_ERR(trans);
5803
5804         err = btrfs_find_free_ino(root, &objectid);
5805         if (err)
5806                 goto out_unlock;
5807
5808         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5809                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5810                                 mode, &index);
5811         if (IS_ERR(inode)) {
5812                 err = PTR_ERR(inode);
5813                 goto out_unlock;
5814         }
5815         drop_inode_on_err = 1;
5816
5817         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5818         if (err)
5819                 goto out_unlock;
5820
5821         err = btrfs_update_inode(trans, root, inode);
5822         if (err)
5823                 goto out_unlock;
5824
5825         /*
5826         * If the active LSM wants to access the inode during
5827         * d_instantiate it needs these. Smack checks to see
5828         * if the filesystem supports xattrs by looking at the
5829         * ops vector.
5830         */
5831         inode->i_fop = &btrfs_file_operations;
5832         inode->i_op = &btrfs_file_inode_operations;
5833
5834         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5835         if (err)
5836                 goto out_unlock;
5837
5838         inode->i_mapping->a_ops = &btrfs_aops;
5839         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5840         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5841         d_instantiate(dentry, inode);
5842
5843 out_unlock:
5844         btrfs_end_transaction(trans, root);
5845         if (err && drop_inode_on_err) {
5846                 inode_dec_link_count(inode);
5847                 iput(inode);
5848         }
5849         btrfs_btree_balance_dirty(root);
5850         return err;
5851 }
5852
5853 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5854                       struct dentry *dentry)
5855 {
5856         struct btrfs_trans_handle *trans;
5857         struct btrfs_root *root = BTRFS_I(dir)->root;
5858         struct inode *inode = old_dentry->d_inode;
5859         u64 index;
5860         int err;
5861         int drop_inode = 0;
5862
5863         /* do not allow sys_link's with other subvols of the same device */
5864         if (root->objectid != BTRFS_I(inode)->root->objectid)
5865                 return -EXDEV;
5866
5867         if (inode->i_nlink >= BTRFS_LINK_MAX)
5868                 return -EMLINK;
5869
5870         err = btrfs_set_inode_index(dir, &index);
5871         if (err)
5872                 goto fail;
5873
5874         /*
5875          * 2 items for inode and inode ref
5876          * 2 items for dir items
5877          * 1 item for parent inode
5878          */
5879         trans = btrfs_start_transaction(root, 5);
5880         if (IS_ERR(trans)) {
5881                 err = PTR_ERR(trans);
5882                 goto fail;
5883         }
5884
5885         btrfs_inc_nlink(inode);
5886         inode_inc_iversion(inode);
5887         inode->i_ctime = CURRENT_TIME;
5888         ihold(inode);
5889         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5890
5891         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5892
5893         if (err) {
5894                 drop_inode = 1;
5895         } else {
5896                 struct dentry *parent = dentry->d_parent;
5897                 err = btrfs_update_inode(trans, root, inode);
5898                 if (err)
5899                         goto fail;
5900                 d_instantiate(dentry, inode);
5901                 btrfs_log_new_name(trans, inode, NULL, parent);
5902         }
5903
5904         btrfs_end_transaction(trans, root);
5905 fail:
5906         if (drop_inode) {
5907                 inode_dec_link_count(inode);
5908                 iput(inode);
5909         }
5910         btrfs_btree_balance_dirty(root);
5911         return err;
5912 }
5913
5914 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5915 {
5916         struct inode *inode = NULL;
5917         struct btrfs_trans_handle *trans;
5918         struct btrfs_root *root = BTRFS_I(dir)->root;
5919         int err = 0;
5920         int drop_on_err = 0;
5921         u64 objectid = 0;
5922         u64 index = 0;
5923
5924         /*
5925          * 2 items for inode and ref
5926          * 2 items for dir items
5927          * 1 for xattr if selinux is on
5928          */
5929         trans = btrfs_start_transaction(root, 5);
5930         if (IS_ERR(trans))
5931                 return PTR_ERR(trans);
5932
5933         err = btrfs_find_free_ino(root, &objectid);
5934         if (err)
5935                 goto out_fail;
5936
5937         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5938                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5939                                 S_IFDIR | mode, &index);
5940         if (IS_ERR(inode)) {
5941                 err = PTR_ERR(inode);
5942                 goto out_fail;
5943         }
5944
5945         drop_on_err = 1;
5946
5947         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5948         if (err)
5949                 goto out_fail;
5950
5951         inode->i_op = &btrfs_dir_inode_operations;
5952         inode->i_fop = &btrfs_dir_file_operations;
5953
5954         btrfs_i_size_write(inode, 0);
5955         err = btrfs_update_inode(trans, root, inode);
5956         if (err)
5957                 goto out_fail;
5958
5959         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5960                              dentry->d_name.len, 0, index);
5961         if (err)
5962                 goto out_fail;
5963
5964         d_instantiate(dentry, inode);
5965         drop_on_err = 0;
5966
5967 out_fail:
5968         btrfs_end_transaction(trans, root);
5969         if (drop_on_err)
5970                 iput(inode);
5971         btrfs_btree_balance_dirty(root);
5972         return err;
5973 }
5974
5975 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5976  * and an extent that you want to insert, deal with overlap and insert
5977  * the new extent into the tree.
5978  */
5979 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5980                                 struct extent_map *existing,
5981                                 struct extent_map *em,
5982                                 u64 map_start, u64 map_len)
5983 {
5984         u64 start_diff;
5985
5986         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5987         start_diff = map_start - em->start;
5988         em->start = map_start;
5989         em->len = map_len;
5990         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5991             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5992                 em->block_start += start_diff;
5993                 em->block_len -= start_diff;
5994         }
5995         return add_extent_mapping(em_tree, em, 0);
5996 }
5997
5998 static noinline int uncompress_inline(struct btrfs_path *path,
5999                                       struct inode *inode, struct page *page,
6000                                       size_t pg_offset, u64 extent_offset,
6001                                       struct btrfs_file_extent_item *item)
6002 {
6003         int ret;
6004         struct extent_buffer *leaf = path->nodes[0];
6005         char *tmp;
6006         size_t max_size;
6007         unsigned long inline_size;
6008         unsigned long ptr;
6009         int compress_type;
6010
6011         WARN_ON(pg_offset != 0);
6012         compress_type = btrfs_file_extent_compression(leaf, item);
6013         max_size = btrfs_file_extent_ram_bytes(leaf, item);
6014         inline_size = btrfs_file_extent_inline_item_len(leaf,
6015                                         btrfs_item_nr(leaf, path->slots[0]));
6016         tmp = kmalloc(inline_size, GFP_NOFS);
6017         if (!tmp)
6018                 return -ENOMEM;
6019         ptr = btrfs_file_extent_inline_start(item);
6020
6021         read_extent_buffer(leaf, tmp, ptr, inline_size);
6022
6023         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
6024         ret = btrfs_decompress(compress_type, tmp, page,
6025                                extent_offset, inline_size, max_size);
6026         if (ret) {
6027                 char *kaddr = kmap_atomic(page);
6028                 unsigned long copy_size = min_t(u64,
6029                                   PAGE_CACHE_SIZE - pg_offset,
6030                                   max_size - extent_offset);
6031                 memset(kaddr + pg_offset, 0, copy_size);
6032                 kunmap_atomic(kaddr);
6033         }
6034         kfree(tmp);
6035         return 0;
6036 }
6037
6038 /*
6039  * a bit scary, this does extent mapping from logical file offset to the disk.
6040  * the ugly parts come from merging extents from the disk with the in-ram
6041  * representation.  This gets more complex because of the data=ordered code,
6042  * where the in-ram extents might be locked pending data=ordered completion.
6043  *
6044  * This also copies inline extents directly into the page.
6045  */
6046
6047 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
6048                                     size_t pg_offset, u64 start, u64 len,
6049                                     int create)
6050 {
6051         int ret;
6052         int err = 0;
6053         u64 bytenr;
6054         u64 extent_start = 0;
6055         u64 extent_end = 0;
6056         u64 objectid = btrfs_ino(inode);
6057         u32 found_type;
6058         struct btrfs_path *path = NULL;
6059         struct btrfs_root *root = BTRFS_I(inode)->root;
6060         struct btrfs_file_extent_item *item;
6061         struct extent_buffer *leaf;
6062         struct btrfs_key found_key;
6063         struct extent_map *em = NULL;
6064         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
6065         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6066         struct btrfs_trans_handle *trans = NULL;
6067         int compress_type;
6068
6069 again:
6070         read_lock(&em_tree->lock);
6071         em = lookup_extent_mapping(em_tree, start, len);
6072         if (em)
6073                 em->bdev = root->fs_info->fs_devices->latest_bdev;
6074         read_unlock(&em_tree->lock);
6075
6076         if (em) {
6077                 if (em->start > start || em->start + em->len <= start)
6078                         free_extent_map(em);
6079                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6080                         free_extent_map(em);
6081                 else
6082                         goto out;
6083         }
6084         em = alloc_extent_map();
6085         if (!em) {
6086                 err = -ENOMEM;
6087                 goto out;
6088         }
6089         em->bdev = root->fs_info->fs_devices->latest_bdev;
6090         em->start = EXTENT_MAP_HOLE;
6091         em->orig_start = EXTENT_MAP_HOLE;
6092         em->len = (u64)-1;
6093         em->block_len = (u64)-1;
6094
6095         if (!path) {
6096                 path = btrfs_alloc_path();
6097                 if (!path) {
6098                         err = -ENOMEM;
6099                         goto out;
6100                 }
6101                 /*
6102                  * Chances are we'll be called again, so go ahead and do
6103                  * readahead
6104                  */
6105                 path->reada = 1;
6106         }
6107
6108         ret = btrfs_lookup_file_extent(trans, root, path,
6109                                        objectid, start, trans != NULL);
6110         if (ret < 0) {
6111                 err = ret;
6112                 goto out;
6113         }
6114
6115         if (ret != 0) {
6116                 if (path->slots[0] == 0)
6117                         goto not_found;
6118                 path->slots[0]--;
6119         }
6120
6121         leaf = path->nodes[0];
6122         item = btrfs_item_ptr(leaf, path->slots[0],
6123                               struct btrfs_file_extent_item);
6124         /* are we inside the extent that was found? */
6125         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6126         found_type = btrfs_key_type(&found_key);
6127         if (found_key.objectid != objectid ||
6128             found_type != BTRFS_EXTENT_DATA_KEY) {
6129                 goto not_found;
6130         }
6131
6132         found_type = btrfs_file_extent_type(leaf, item);
6133         extent_start = found_key.offset;
6134         compress_type = btrfs_file_extent_compression(leaf, item);
6135         if (found_type == BTRFS_FILE_EXTENT_REG ||
6136             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6137                 extent_end = extent_start +
6138                        btrfs_file_extent_num_bytes(leaf, item);
6139         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6140                 size_t size;
6141                 size = btrfs_file_extent_inline_len(leaf, item);
6142                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6143         }
6144
6145         if (start >= extent_end) {
6146                 path->slots[0]++;
6147                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6148                         ret = btrfs_next_leaf(root, path);
6149                         if (ret < 0) {
6150                                 err = ret;
6151                                 goto out;
6152                         }
6153                         if (ret > 0)
6154                                 goto not_found;
6155                         leaf = path->nodes[0];
6156                 }
6157                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6158                 if (found_key.objectid != objectid ||
6159                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6160                         goto not_found;
6161                 if (start + len <= found_key.offset)
6162                         goto not_found;
6163                 em->start = start;
6164                 em->orig_start = start;
6165                 em->len = found_key.offset - start;
6166                 goto not_found_em;
6167         }
6168
6169         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6170         if (found_type == BTRFS_FILE_EXTENT_REG ||
6171             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6172                 em->start = extent_start;
6173                 em->len = extent_end - extent_start;
6174                 em->orig_start = extent_start -
6175                                  btrfs_file_extent_offset(leaf, item);
6176                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6177                                                                       item);
6178                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6179                 if (bytenr == 0) {
6180                         em->block_start = EXTENT_MAP_HOLE;
6181                         goto insert;
6182                 }
6183                 if (compress_type != BTRFS_COMPRESS_NONE) {
6184                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6185                         em->compress_type = compress_type;
6186                         em->block_start = bytenr;
6187                         em->block_len = em->orig_block_len;
6188                 } else {
6189                         bytenr += btrfs_file_extent_offset(leaf, item);
6190                         em->block_start = bytenr;
6191                         em->block_len = em->len;
6192                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6193                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6194                 }
6195                 goto insert;
6196         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6197                 unsigned long ptr;
6198                 char *map;
6199                 size_t size;
6200                 size_t extent_offset;
6201                 size_t copy_size;
6202
6203                 em->block_start = EXTENT_MAP_INLINE;
6204                 if (!page || create) {
6205                         em->start = extent_start;
6206                         em->len = extent_end - extent_start;
6207                         goto out;
6208                 }
6209
6210                 size = btrfs_file_extent_inline_len(leaf, item);
6211                 extent_offset = page_offset(page) + pg_offset - extent_start;
6212                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6213                                 size - extent_offset);
6214                 em->start = extent_start + extent_offset;
6215                 em->len = ALIGN(copy_size, root->sectorsize);
6216                 em->orig_block_len = em->len;
6217                 em->orig_start = em->start;
6218                 if (compress_type) {
6219                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6220                         em->compress_type = compress_type;
6221                 }
6222                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6223                 if (create == 0 && !PageUptodate(page)) {
6224                         if (btrfs_file_extent_compression(leaf, item) !=
6225                             BTRFS_COMPRESS_NONE) {
6226                                 ret = uncompress_inline(path, inode, page,
6227                                                         pg_offset,
6228                                                         extent_offset, item);
6229                                 BUG_ON(ret); /* -ENOMEM */
6230                         } else {
6231                                 map = kmap(page);
6232                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6233                                                    copy_size);
6234                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6235                                         memset(map + pg_offset + copy_size, 0,
6236                                                PAGE_CACHE_SIZE - pg_offset -
6237                                                copy_size);
6238                                 }
6239                                 kunmap(page);
6240                         }
6241                         flush_dcache_page(page);
6242                 } else if (create && PageUptodate(page)) {
6243                         BUG();
6244                         if (!trans) {
6245                                 kunmap(page);
6246                                 free_extent_map(em);
6247                                 em = NULL;
6248
6249                                 btrfs_release_path(path);
6250                                 trans = btrfs_join_transaction(root);
6251
6252                                 if (IS_ERR(trans))
6253                                         return ERR_CAST(trans);
6254                                 goto again;
6255                         }
6256                         map = kmap(page);
6257                         write_extent_buffer(leaf, map + pg_offset, ptr,
6258                                             copy_size);
6259                         kunmap(page);
6260                         btrfs_mark_buffer_dirty(leaf);
6261                 }
6262                 set_extent_uptodate(io_tree, em->start,
6263                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6264                 goto insert;
6265         } else {
6266                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6267         }
6268 not_found:
6269         em->start = start;
6270         em->orig_start = start;
6271         em->len = len;
6272 not_found_em:
6273         em->block_start = EXTENT_MAP_HOLE;
6274         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6275 insert:
6276         btrfs_release_path(path);
6277         if (em->start > start || extent_map_end(em) <= start) {
6278                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6279                         (unsigned long long)em->start,
6280                         (unsigned long long)em->len,
6281                         (unsigned long long)start,
6282                         (unsigned long long)len);
6283                 err = -EIO;
6284                 goto out;
6285         }
6286
6287         err = 0;
6288         write_lock(&em_tree->lock);
6289         ret = add_extent_mapping(em_tree, em, 0);
6290         /* it is possible that someone inserted the extent into the tree
6291          * while we had the lock dropped.  It is also possible that
6292          * an overlapping map exists in the tree
6293          */
6294         if (ret == -EEXIST) {
6295                 struct extent_map *existing;
6296
6297                 ret = 0;
6298
6299                 existing = lookup_extent_mapping(em_tree, start, len);
6300                 if (existing && (existing->start > start ||
6301                     existing->start + existing->len <= start)) {
6302                         free_extent_map(existing);
6303                         existing = NULL;
6304                 }
6305                 if (!existing) {
6306                         existing = lookup_extent_mapping(em_tree, em->start,
6307                                                          em->len);
6308                         if (existing) {
6309                                 err = merge_extent_mapping(em_tree, existing,
6310                                                            em, start,
6311                                                            root->sectorsize);
6312                                 free_extent_map(existing);
6313                                 if (err) {
6314                                         free_extent_map(em);
6315                                         em = NULL;
6316                                 }
6317                         } else {
6318                                 err = -EIO;
6319                                 free_extent_map(em);
6320                                 em = NULL;
6321                         }
6322                 } else {
6323                         free_extent_map(em);
6324                         em = existing;
6325                         err = 0;
6326                 }
6327         }
6328         write_unlock(&em_tree->lock);
6329 out:
6330
6331         if (em)
6332                 trace_btrfs_get_extent(root, em);
6333
6334         if (path)
6335                 btrfs_free_path(path);
6336         if (trans) {
6337                 ret = btrfs_end_transaction(trans, root);
6338                 if (!err)
6339                         err = ret;
6340         }
6341         if (err) {
6342                 free_extent_map(em);
6343                 return ERR_PTR(err);
6344         }
6345         BUG_ON(!em); /* Error is always set */
6346         return em;
6347 }
6348
6349 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6350                                            size_t pg_offset, u64 start, u64 len,
6351                                            int create)
6352 {
6353         struct extent_map *em;
6354         struct extent_map *hole_em = NULL;
6355         u64 range_start = start;
6356         u64 end;
6357         u64 found;
6358         u64 found_end;
6359         int err = 0;
6360
6361         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6362         if (IS_ERR(em))
6363                 return em;
6364         if (em) {
6365                 /*
6366                  * if our em maps to
6367                  * -  a hole or
6368                  * -  a pre-alloc extent,
6369                  * there might actually be delalloc bytes behind it.
6370                  */
6371                 if (em->block_start != EXTENT_MAP_HOLE &&
6372                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6373                         return em;
6374                 else
6375                         hole_em = em;
6376         }
6377
6378         /* check to see if we've wrapped (len == -1 or similar) */
6379         end = start + len;
6380         if (end < start)
6381                 end = (u64)-1;
6382         else
6383                 end -= 1;
6384
6385         em = NULL;
6386
6387         /* ok, we didn't find anything, lets look for delalloc */
6388         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6389                                  end, len, EXTENT_DELALLOC, 1);
6390         found_end = range_start + found;
6391         if (found_end < range_start)
6392                 found_end = (u64)-1;
6393
6394         /*
6395          * we didn't find anything useful, return
6396          * the original results from get_extent()
6397          */
6398         if (range_start > end || found_end <= start) {
6399                 em = hole_em;
6400                 hole_em = NULL;
6401                 goto out;
6402         }
6403
6404         /* adjust the range_start to make sure it doesn't
6405          * go backwards from the start they passed in
6406          */
6407         range_start = max(start,range_start);
6408         found = found_end - range_start;
6409
6410         if (found > 0) {
6411                 u64 hole_start = start;
6412                 u64 hole_len = len;
6413
6414                 em = alloc_extent_map();
6415                 if (!em) {
6416                         err = -ENOMEM;
6417                         goto out;
6418                 }
6419                 /*
6420                  * when btrfs_get_extent can't find anything it
6421                  * returns one huge hole
6422                  *
6423                  * make sure what it found really fits our range, and
6424                  * adjust to make sure it is based on the start from
6425                  * the caller
6426                  */
6427                 if (hole_em) {
6428                         u64 calc_end = extent_map_end(hole_em);
6429
6430                         if (calc_end <= start || (hole_em->start > end)) {
6431                                 free_extent_map(hole_em);
6432                                 hole_em = NULL;
6433                         } else {
6434                                 hole_start = max(hole_em->start, start);
6435                                 hole_len = calc_end - hole_start;
6436                         }
6437                 }
6438                 em->bdev = NULL;
6439                 if (hole_em && range_start > hole_start) {
6440                         /* our hole starts before our delalloc, so we
6441                          * have to return just the parts of the hole
6442                          * that go until  the delalloc starts
6443                          */
6444                         em->len = min(hole_len,
6445                                       range_start - hole_start);
6446                         em->start = hole_start;
6447                         em->orig_start = hole_start;
6448                         /*
6449                          * don't adjust block start at all,
6450                          * it is fixed at EXTENT_MAP_HOLE
6451                          */
6452                         em->block_start = hole_em->block_start;
6453                         em->block_len = hole_len;
6454                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6455                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6456                 } else {
6457                         em->start = range_start;
6458                         em->len = found;
6459                         em->orig_start = range_start;
6460                         em->block_start = EXTENT_MAP_DELALLOC;
6461                         em->block_len = found;
6462                 }
6463         } else if (hole_em) {
6464                 return hole_em;
6465         }
6466 out:
6467
6468         free_extent_map(hole_em);
6469         if (err) {
6470                 free_extent_map(em);
6471                 return ERR_PTR(err);
6472         }
6473         return em;
6474 }
6475
6476 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6477                                                   u64 start, u64 len)
6478 {
6479         struct btrfs_root *root = BTRFS_I(inode)->root;
6480         struct btrfs_trans_handle *trans;
6481         struct extent_map *em;
6482         struct btrfs_key ins;
6483         u64 alloc_hint;
6484         int ret;
6485
6486         trans = btrfs_join_transaction(root);
6487         if (IS_ERR(trans))
6488                 return ERR_CAST(trans);
6489
6490         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
6491
6492         alloc_hint = get_extent_allocation_hint(inode, start, len);
6493         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
6494                                    alloc_hint, &ins, 1);
6495         if (ret) {
6496                 em = ERR_PTR(ret);
6497                 goto out;
6498         }
6499
6500         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6501                               ins.offset, ins.offset, ins.offset, 0);
6502         if (IS_ERR(em))
6503                 goto out;
6504
6505         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6506                                            ins.offset, ins.offset, 0);
6507         if (ret) {
6508                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6509                 em = ERR_PTR(ret);
6510         }
6511 out:
6512         btrfs_end_transaction(trans, root);
6513         return em;
6514 }
6515
6516 /*
6517  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6518  * block must be cow'd
6519  */
6520 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
6521                                       struct inode *inode, u64 offset, u64 *len,
6522                                       u64 *orig_start, u64 *orig_block_len,
6523                                       u64 *ram_bytes)
6524 {
6525         struct btrfs_path *path;
6526         int ret;
6527         struct extent_buffer *leaf;
6528         struct btrfs_root *root = BTRFS_I(inode)->root;
6529         struct btrfs_file_extent_item *fi;
6530         struct btrfs_key key;
6531         u64 disk_bytenr;
6532         u64 backref_offset;
6533         u64 extent_end;
6534         u64 num_bytes;
6535         int slot;
6536         int found_type;
6537
6538         path = btrfs_alloc_path();
6539         if (!path)
6540                 return -ENOMEM;
6541
6542         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
6543                                        offset, 0);
6544         if (ret < 0)
6545                 goto out;
6546
6547         slot = path->slots[0];
6548         if (ret == 1) {
6549                 if (slot == 0) {
6550                         /* can't find the item, must cow */
6551                         ret = 0;
6552                         goto out;
6553                 }
6554                 slot--;
6555         }
6556         ret = 0;
6557         leaf = path->nodes[0];
6558         btrfs_item_key_to_cpu(leaf, &key, slot);
6559         if (key.objectid != btrfs_ino(inode) ||
6560             key.type != BTRFS_EXTENT_DATA_KEY) {
6561                 /* not our file or wrong item type, must cow */
6562                 goto out;
6563         }
6564
6565         if (key.offset > offset) {
6566                 /* Wrong offset, must cow */
6567                 goto out;
6568         }
6569
6570         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6571         found_type = btrfs_file_extent_type(leaf, fi);
6572         if (found_type != BTRFS_FILE_EXTENT_REG &&
6573             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6574                 /* not a regular extent, must cow */
6575                 goto out;
6576         }
6577         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6578         backref_offset = btrfs_file_extent_offset(leaf, fi);
6579
6580         *orig_start = key.offset - backref_offset;
6581         *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6582         *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6583
6584         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6585         if (extent_end < offset + *len) {
6586                 /* extent doesn't include our full range, must cow */
6587                 goto out;
6588         }
6589
6590         if (btrfs_extent_readonly(root, disk_bytenr))
6591                 goto out;
6592
6593         /*
6594          * look for other files referencing this extent, if we
6595          * find any we must cow
6596          */
6597         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6598                                   key.offset - backref_offset, disk_bytenr))
6599                 goto out;
6600
6601         /*
6602          * adjust disk_bytenr and num_bytes to cover just the bytes
6603          * in this extent we are about to write.  If there
6604          * are any csums in that range we have to cow in order
6605          * to keep the csums correct
6606          */
6607         disk_bytenr += backref_offset;
6608         disk_bytenr += offset - key.offset;
6609         num_bytes = min(offset + *len, extent_end) - offset;
6610         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6611                                 goto out;
6612         /*
6613          * all of the above have passed, it is safe to overwrite this extent
6614          * without cow
6615          */
6616         *len = num_bytes;
6617         ret = 1;
6618 out:
6619         btrfs_free_path(path);
6620         return ret;
6621 }
6622
6623 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6624                               struct extent_state **cached_state, int writing)
6625 {
6626         struct btrfs_ordered_extent *ordered;
6627         int ret = 0;
6628
6629         while (1) {
6630                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6631                                  0, cached_state);
6632                 /*
6633                  * We're concerned with the entire range that we're going to be
6634                  * doing DIO to, so we need to make sure theres no ordered
6635                  * extents in this range.
6636                  */
6637                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6638                                                      lockend - lockstart + 1);
6639
6640                 /*
6641                  * We need to make sure there are no buffered pages in this
6642                  * range either, we could have raced between the invalidate in
6643                  * generic_file_direct_write and locking the extent.  The
6644                  * invalidate needs to happen so that reads after a write do not
6645                  * get stale data.
6646                  */
6647                 if (!ordered && (!writing ||
6648                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6649                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6650                                     *cached_state)))
6651                         break;
6652
6653                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6654                                      cached_state, GFP_NOFS);
6655
6656                 if (ordered) {
6657                         btrfs_start_ordered_extent(inode, ordered, 1);
6658                         btrfs_put_ordered_extent(ordered);
6659                 } else {
6660                         /* Screw you mmap */
6661                         ret = filemap_write_and_wait_range(inode->i_mapping,
6662                                                            lockstart,
6663                                                            lockend);
6664                         if (ret)
6665                                 break;
6666
6667                         /*
6668                          * If we found a page that couldn't be invalidated just
6669                          * fall back to buffered.
6670                          */
6671                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6672                                         lockstart >> PAGE_CACHE_SHIFT,
6673                                         lockend >> PAGE_CACHE_SHIFT);
6674                         if (ret)
6675                                 break;
6676                 }
6677
6678                 cond_resched();
6679         }
6680
6681         return ret;
6682 }
6683
6684 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6685                                            u64 len, u64 orig_start,
6686                                            u64 block_start, u64 block_len,
6687                                            u64 orig_block_len, u64 ram_bytes,
6688                                            int type)
6689 {
6690         struct extent_map_tree *em_tree;
6691         struct extent_map *em;
6692         struct btrfs_root *root = BTRFS_I(inode)->root;
6693         int ret;
6694
6695         em_tree = &BTRFS_I(inode)->extent_tree;
6696         em = alloc_extent_map();
6697         if (!em)
6698                 return ERR_PTR(-ENOMEM);
6699
6700         em->start = start;
6701         em->orig_start = orig_start;
6702         em->mod_start = start;
6703         em->mod_len = len;
6704         em->len = len;
6705         em->block_len = block_len;
6706         em->block_start = block_start;
6707         em->bdev = root->fs_info->fs_devices->latest_bdev;
6708         em->orig_block_len = orig_block_len;
6709         em->ram_bytes = ram_bytes;
6710         em->generation = -1;
6711         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6712         if (type == BTRFS_ORDERED_PREALLOC)
6713                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6714
6715         do {
6716                 btrfs_drop_extent_cache(inode, em->start,
6717                                 em->start + em->len - 1, 0);
6718                 write_lock(&em_tree->lock);
6719                 ret = add_extent_mapping(em_tree, em, 1);
6720                 write_unlock(&em_tree->lock);
6721         } while (ret == -EEXIST);
6722
6723         if (ret) {
6724                 free_extent_map(em);
6725                 return ERR_PTR(ret);
6726         }
6727
6728         return em;
6729 }
6730
6731
6732 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6733                                    struct buffer_head *bh_result, int create)
6734 {
6735         struct extent_map *em;
6736         struct btrfs_root *root = BTRFS_I(inode)->root;
6737         struct extent_state *cached_state = NULL;
6738         u64 start = iblock << inode->i_blkbits;
6739         u64 lockstart, lockend;
6740         u64 len = bh_result->b_size;
6741         struct btrfs_trans_handle *trans;
6742         int unlock_bits = EXTENT_LOCKED;
6743         int ret = 0;
6744
6745         if (create)
6746                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6747         else
6748                 len = min_t(u64, len, root->sectorsize);
6749
6750         lockstart = start;
6751         lockend = start + len - 1;
6752
6753         /*
6754          * If this errors out it's because we couldn't invalidate pagecache for
6755          * this range and we need to fallback to buffered.
6756          */
6757         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6758                 return -ENOTBLK;
6759
6760         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6761         if (IS_ERR(em)) {
6762                 ret = PTR_ERR(em);
6763                 goto unlock_err;
6764         }
6765
6766         /*
6767          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6768          * io.  INLINE is special, and we could probably kludge it in here, but
6769          * it's still buffered so for safety lets just fall back to the generic
6770          * buffered path.
6771          *
6772          * For COMPRESSED we _have_ to read the entire extent in so we can
6773          * decompress it, so there will be buffering required no matter what we
6774          * do, so go ahead and fallback to buffered.
6775          *
6776          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6777          * to buffered IO.  Don't blame me, this is the price we pay for using
6778          * the generic code.
6779          */
6780         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6781             em->block_start == EXTENT_MAP_INLINE) {
6782                 free_extent_map(em);
6783                 ret = -ENOTBLK;
6784                 goto unlock_err;
6785         }
6786
6787         /* Just a good old fashioned hole, return */
6788         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6789                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6790                 free_extent_map(em);
6791                 goto unlock_err;
6792         }
6793
6794         /*
6795          * We don't allocate a new extent in the following cases
6796          *
6797          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6798          * existing extent.
6799          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6800          * just use the extent.
6801          *
6802          */
6803         if (!create) {
6804                 len = min(len, em->len - (start - em->start));
6805                 lockstart = start + len;
6806                 goto unlock;
6807         }
6808
6809         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6810             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6811              em->block_start != EXTENT_MAP_HOLE)) {
6812                 int type;
6813                 int ret;
6814                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6815
6816                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6817                         type = BTRFS_ORDERED_PREALLOC;
6818                 else
6819                         type = BTRFS_ORDERED_NOCOW;
6820                 len = min(len, em->len - (start - em->start));
6821                 block_start = em->block_start + (start - em->start);
6822
6823                 /*
6824                  * we're not going to log anything, but we do need
6825                  * to make sure the current transaction stays open
6826                  * while we look for nocow cross refs
6827                  */
6828                 trans = btrfs_join_transaction(root);
6829                 if (IS_ERR(trans))
6830                         goto must_cow;
6831
6832                 if (can_nocow_odirect(trans, inode, start, &len, &orig_start,
6833                                       &orig_block_len, &ram_bytes) == 1) {
6834                         if (type == BTRFS_ORDERED_PREALLOC) {
6835                                 free_extent_map(em);
6836                                 em = create_pinned_em(inode, start, len,
6837                                                        orig_start,
6838                                                        block_start, len,
6839                                                        orig_block_len,
6840                                                        ram_bytes, type);
6841                                 if (IS_ERR(em)) {
6842                                         btrfs_end_transaction(trans, root);
6843                                         goto unlock_err;
6844                                 }
6845                         }
6846
6847                         ret = btrfs_add_ordered_extent_dio(inode, start,
6848                                            block_start, len, len, type);
6849                         btrfs_end_transaction(trans, root);
6850                         if (ret) {
6851                                 free_extent_map(em);
6852                                 goto unlock_err;
6853                         }
6854                         goto unlock;
6855                 }
6856                 btrfs_end_transaction(trans, root);
6857         }
6858 must_cow:
6859         /*
6860          * this will cow the extent, reset the len in case we changed
6861          * it above
6862          */
6863         len = bh_result->b_size;
6864         free_extent_map(em);
6865         em = btrfs_new_extent_direct(inode, start, len);
6866         if (IS_ERR(em)) {
6867                 ret = PTR_ERR(em);
6868                 goto unlock_err;
6869         }
6870         len = min(len, em->len - (start - em->start));
6871 unlock:
6872         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6873                 inode->i_blkbits;
6874         bh_result->b_size = len;
6875         bh_result->b_bdev = em->bdev;
6876         set_buffer_mapped(bh_result);
6877         if (create) {
6878                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6879                         set_buffer_new(bh_result);
6880
6881                 /*
6882                  * Need to update the i_size under the extent lock so buffered
6883                  * readers will get the updated i_size when we unlock.
6884                  */
6885                 if (start + len > i_size_read(inode))
6886                         i_size_write(inode, start + len);
6887
6888                 spin_lock(&BTRFS_I(inode)->lock);
6889                 BTRFS_I(inode)->outstanding_extents++;
6890                 spin_unlock(&BTRFS_I(inode)->lock);
6891
6892                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6893                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6894                                      &cached_state, GFP_NOFS);
6895                 BUG_ON(ret);
6896         }
6897
6898         /*
6899          * In the case of write we need to clear and unlock the entire range,
6900          * in the case of read we need to unlock only the end area that we
6901          * aren't using if there is any left over space.
6902          */
6903         if (lockstart < lockend) {
6904                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6905                                  lockend, unlock_bits, 1, 0,
6906                                  &cached_state, GFP_NOFS);
6907         } else {
6908                 free_extent_state(cached_state);
6909         }
6910
6911         free_extent_map(em);
6912
6913         return 0;
6914
6915 unlock_err:
6916         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6917                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6918         return ret;
6919 }
6920
6921 struct btrfs_dio_private {
6922         struct inode *inode;
6923         u64 logical_offset;
6924         u64 disk_bytenr;
6925         u64 bytes;
6926         void *private;
6927
6928         /* number of bios pending for this dio */
6929         atomic_t pending_bios;
6930
6931         /* IO errors */
6932         int errors;
6933
6934         /* orig_bio is our btrfs_io_bio */
6935         struct bio *orig_bio;
6936
6937         /* dio_bio came from fs/direct-io.c */
6938         struct bio *dio_bio;
6939 };
6940
6941 static void btrfs_endio_direct_read(struct bio *bio, int err)
6942 {
6943         struct btrfs_dio_private *dip = bio->bi_private;
6944         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6945         struct bio_vec *bvec = bio->bi_io_vec;
6946         struct inode *inode = dip->inode;
6947         struct btrfs_root *root = BTRFS_I(inode)->root;
6948         struct bio *dio_bio;
6949         u64 start;
6950
6951         start = dip->logical_offset;
6952         do {
6953                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6954                         struct page *page = bvec->bv_page;
6955                         char *kaddr;
6956                         u32 csum = ~(u32)0;
6957                         u64 private = ~(u32)0;
6958                         unsigned long flags;
6959
6960                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6961                                               start, &private))
6962                                 goto failed;
6963                         local_irq_save(flags);
6964                         kaddr = kmap_atomic(page);
6965                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6966                                                csum, bvec->bv_len);
6967                         btrfs_csum_final(csum, (char *)&csum);
6968                         kunmap_atomic(kaddr);
6969                         local_irq_restore(flags);
6970
6971                         flush_dcache_page(bvec->bv_page);
6972                         if (csum != private) {
6973 failed:
6974                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u private %u",
6975                                         (unsigned long long)btrfs_ino(inode),
6976                                         (unsigned long long)start,
6977                                         csum, (unsigned)private);
6978                                 err = -EIO;
6979                         }
6980                 }
6981
6982                 start += bvec->bv_len;
6983                 bvec++;
6984         } while (bvec <= bvec_end);
6985
6986         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6987                       dip->logical_offset + dip->bytes - 1);
6988         dio_bio = dip->dio_bio;
6989
6990         kfree(dip);
6991
6992         /* If we had a csum failure make sure to clear the uptodate flag */
6993         if (err)
6994                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6995         dio_end_io(dio_bio, err);
6996         bio_put(bio);
6997 }
6998
6999 static void btrfs_endio_direct_write(struct bio *bio, int err)
7000 {
7001         struct btrfs_dio_private *dip = bio->bi_private;
7002         struct inode *inode = dip->inode;
7003         struct btrfs_root *root = BTRFS_I(inode)->root;
7004         struct btrfs_ordered_extent *ordered = NULL;
7005         u64 ordered_offset = dip->logical_offset;
7006         u64 ordered_bytes = dip->bytes;
7007         struct bio *dio_bio;
7008         int ret;
7009
7010         if (err)
7011                 goto out_done;
7012 again:
7013         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
7014                                                    &ordered_offset,
7015                                                    ordered_bytes, !err);
7016         if (!ret)
7017                 goto out_test;
7018
7019         ordered->work.func = finish_ordered_fn;
7020         ordered->work.flags = 0;
7021         btrfs_queue_worker(&root->fs_info->endio_write_workers,
7022                            &ordered->work);
7023 out_test:
7024         /*
7025          * our bio might span multiple ordered extents.  If we haven't
7026          * completed the accounting for the whole dio, go back and try again
7027          */
7028         if (ordered_offset < dip->logical_offset + dip->bytes) {
7029                 ordered_bytes = dip->logical_offset + dip->bytes -
7030                         ordered_offset;
7031                 ordered = NULL;
7032                 goto again;
7033         }
7034 out_done:
7035         dio_bio = dip->dio_bio;
7036
7037         kfree(dip);
7038
7039         /* If we had an error make sure to clear the uptodate flag */
7040         if (err)
7041                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
7042         dio_end_io(dio_bio, err);
7043         bio_put(bio);
7044 }
7045
7046 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
7047                                     struct bio *bio, int mirror_num,
7048                                     unsigned long bio_flags, u64 offset)
7049 {
7050         int ret;
7051         struct btrfs_root *root = BTRFS_I(inode)->root;
7052         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
7053         BUG_ON(ret); /* -ENOMEM */
7054         return 0;
7055 }
7056
7057 static void btrfs_end_dio_bio(struct bio *bio, int err)
7058 {
7059         struct btrfs_dio_private *dip = bio->bi_private;
7060
7061         if (err) {
7062                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
7063                       "sector %#Lx len %u err no %d\n",
7064                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
7065                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
7066                 dip->errors = 1;
7067
7068                 /*
7069                  * before atomic variable goto zero, we must make sure
7070                  * dip->errors is perceived to be set.
7071                  */
7072                 smp_mb__before_atomic_dec();
7073         }
7074
7075         /* if there are more bios still pending for this dio, just exit */
7076         if (!atomic_dec_and_test(&dip->pending_bios))
7077                 goto out;
7078
7079         if (dip->errors) {
7080                 bio_io_error(dip->orig_bio);
7081         } else {
7082                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
7083                 bio_endio(dip->orig_bio, 0);
7084         }
7085 out:
7086         bio_put(bio);
7087 }
7088
7089 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
7090                                        u64 first_sector, gfp_t gfp_flags)
7091 {
7092         int nr_vecs = bio_get_nr_vecs(bdev);
7093         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
7094 }
7095
7096 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7097                                          int rw, u64 file_offset, int skip_sum,
7098                                          int async_submit)
7099 {
7100         int write = rw & REQ_WRITE;
7101         struct btrfs_root *root = BTRFS_I(inode)->root;
7102         int ret;
7103
7104         if (async_submit)
7105                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7106
7107         bio_get(bio);
7108
7109         if (!write) {
7110                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7111                 if (ret)
7112                         goto err;
7113         }
7114
7115         if (skip_sum)
7116                 goto map;
7117
7118         if (write && async_submit) {
7119                 ret = btrfs_wq_submit_bio(root->fs_info,
7120                                    inode, rw, bio, 0, 0,
7121                                    file_offset,
7122                                    __btrfs_submit_bio_start_direct_io,
7123                                    __btrfs_submit_bio_done);
7124                 goto err;
7125         } else if (write) {
7126                 /*
7127                  * If we aren't doing async submit, calculate the csum of the
7128                  * bio now.
7129                  */
7130                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7131                 if (ret)
7132                         goto err;
7133         } else if (!skip_sum) {
7134                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
7135                 if (ret)
7136                         goto err;
7137         }
7138
7139 map:
7140         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7141 err:
7142         bio_put(bio);
7143         return ret;
7144 }
7145
7146 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7147                                     int skip_sum)
7148 {
7149         struct inode *inode = dip->inode;
7150         struct btrfs_root *root = BTRFS_I(inode)->root;
7151         struct bio *bio;
7152         struct bio *orig_bio = dip->orig_bio;
7153         struct bio_vec *bvec = orig_bio->bi_io_vec;
7154         u64 start_sector = orig_bio->bi_sector;
7155         u64 file_offset = dip->logical_offset;
7156         u64 submit_len = 0;
7157         u64 map_length;
7158         int nr_pages = 0;
7159         int ret = 0;
7160         int async_submit = 0;
7161
7162         map_length = orig_bio->bi_size;
7163         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7164                               &map_length, NULL, 0);
7165         if (ret) {
7166                 bio_put(orig_bio);
7167                 return -EIO;
7168         }
7169         if (map_length >= orig_bio->bi_size) {
7170                 bio = orig_bio;
7171                 goto submit;
7172         }
7173
7174         /* async crcs make it difficult to collect full stripe writes. */
7175         if (btrfs_get_alloc_profile(root, 1) &
7176             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7177                 async_submit = 0;
7178         else
7179                 async_submit = 1;
7180
7181         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7182         if (!bio)
7183                 return -ENOMEM;
7184         bio->bi_private = dip;
7185         bio->bi_end_io = btrfs_end_dio_bio;
7186         atomic_inc(&dip->pending_bios);
7187
7188         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7189                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7190                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7191                                  bvec->bv_offset) < bvec->bv_len)) {
7192                         /*
7193                          * inc the count before we submit the bio so
7194                          * we know the end IO handler won't happen before
7195                          * we inc the count. Otherwise, the dip might get freed
7196                          * before we're done setting it up
7197                          */
7198                         atomic_inc(&dip->pending_bios);
7199                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7200                                                      file_offset, skip_sum,
7201                                                      async_submit);
7202                         if (ret) {
7203                                 bio_put(bio);
7204                                 atomic_dec(&dip->pending_bios);
7205                                 goto out_err;
7206                         }
7207
7208                         start_sector += submit_len >> 9;
7209                         file_offset += submit_len;
7210
7211                         submit_len = 0;
7212                         nr_pages = 0;
7213
7214                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7215                                                   start_sector, GFP_NOFS);
7216                         if (!bio)
7217                                 goto out_err;
7218                         bio->bi_private = dip;
7219                         bio->bi_end_io = btrfs_end_dio_bio;
7220
7221                         map_length = orig_bio->bi_size;
7222                         ret = btrfs_map_block(root->fs_info, rw,
7223                                               start_sector << 9,
7224                                               &map_length, NULL, 0);
7225                         if (ret) {
7226                                 bio_put(bio);
7227                                 goto out_err;
7228                         }
7229                 } else {
7230                         submit_len += bvec->bv_len;
7231                         nr_pages ++;
7232                         bvec++;
7233                 }
7234         }
7235
7236 submit:
7237         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7238                                      async_submit);
7239         if (!ret)
7240                 return 0;
7241
7242         bio_put(bio);
7243 out_err:
7244         dip->errors = 1;
7245         /*
7246          * before atomic variable goto zero, we must
7247          * make sure dip->errors is perceived to be set.
7248          */
7249         smp_mb__before_atomic_dec();
7250         if (atomic_dec_and_test(&dip->pending_bios))
7251                 bio_io_error(dip->orig_bio);
7252
7253         /* bio_end_io() will handle error, so we needn't return it */
7254         return 0;
7255 }
7256
7257 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7258                                 struct inode *inode, loff_t file_offset)
7259 {
7260         struct btrfs_root *root = BTRFS_I(inode)->root;
7261         struct btrfs_dio_private *dip;
7262         struct bio_vec *bvec = dio_bio->bi_io_vec;
7263         struct bio *io_bio;
7264         int skip_sum;
7265         int write = rw & REQ_WRITE;
7266         int ret = 0;
7267
7268         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7269
7270         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7271
7272         if (!io_bio) {
7273                 ret = -ENOMEM;
7274                 goto free_ordered;
7275         }
7276
7277         dip = kmalloc(sizeof(*dip), GFP_NOFS);
7278         if (!dip) {
7279                 ret = -ENOMEM;
7280                 goto free_io_bio;
7281         }
7282
7283         dip->private = dio_bio->bi_private;
7284         io_bio->bi_private = dio_bio->bi_private;
7285         dip->inode = inode;
7286         dip->logical_offset = file_offset;
7287
7288         dip->bytes = 0;
7289         do {
7290                 dip->bytes += bvec->bv_len;
7291                 bvec++;
7292         } while (bvec <= (dio_bio->bi_io_vec + dio_bio->bi_vcnt - 1));
7293
7294         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7295         io_bio->bi_private = dip;
7296         dip->errors = 0;
7297         dip->orig_bio = io_bio;
7298         dip->dio_bio = dio_bio;
7299         atomic_set(&dip->pending_bios, 0);
7300
7301         if (write)
7302                 io_bio->bi_end_io = btrfs_endio_direct_write;
7303         else
7304                 io_bio->bi_end_io = btrfs_endio_direct_read;
7305
7306         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7307         if (!ret)
7308                 return;
7309
7310 free_io_bio:
7311         bio_put(io_bio);
7312
7313 free_ordered:
7314         /*
7315          * If this is a write, we need to clean up the reserved space and kill
7316          * the ordered extent.
7317          */
7318         if (write) {
7319                 struct btrfs_ordered_extent *ordered;
7320                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7321                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7322                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7323                         btrfs_free_reserved_extent(root, ordered->start,
7324                                                    ordered->disk_len);
7325                 btrfs_put_ordered_extent(ordered);
7326                 btrfs_put_ordered_extent(ordered);
7327         }
7328         bio_endio(dio_bio, ret);
7329 }
7330
7331 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7332                         const struct iovec *iov, loff_t offset,
7333                         unsigned long nr_segs)
7334 {
7335         int seg;
7336         int i;
7337         size_t size;
7338         unsigned long addr;
7339         unsigned blocksize_mask = root->sectorsize - 1;
7340         ssize_t retval = -EINVAL;
7341         loff_t end = offset;
7342
7343         if (offset & blocksize_mask)
7344                 goto out;
7345
7346         /* Check the memory alignment.  Blocks cannot straddle pages */
7347         for (seg = 0; seg < nr_segs; seg++) {
7348                 addr = (unsigned long)iov[seg].iov_base;
7349                 size = iov[seg].iov_len;
7350                 end += size;
7351                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7352                         goto out;
7353
7354                 /* If this is a write we don't need to check anymore */
7355                 if (rw & WRITE)
7356                         continue;
7357
7358                 /*
7359                  * Check to make sure we don't have duplicate iov_base's in this
7360                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7361                  * when reading back.
7362                  */
7363                 for (i = seg + 1; i < nr_segs; i++) {
7364                         if (iov[seg].iov_base == iov[i].iov_base)
7365                                 goto out;
7366                 }
7367         }
7368         retval = 0;
7369 out:
7370         return retval;
7371 }
7372
7373 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7374                         const struct iovec *iov, loff_t offset,
7375                         unsigned long nr_segs)
7376 {
7377         struct file *file = iocb->ki_filp;
7378         struct inode *inode = file->f_mapping->host;
7379         size_t count = 0;
7380         int flags = 0;
7381         bool wakeup = true;
7382         bool relock = false;
7383         ssize_t ret;
7384
7385         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7386                             offset, nr_segs))
7387                 return 0;
7388
7389         atomic_inc(&inode->i_dio_count);
7390         smp_mb__after_atomic_inc();
7391
7392         if (rw & WRITE) {
7393                 count = iov_length(iov, nr_segs);
7394                 /*
7395                  * If the write DIO is beyond the EOF, we need update
7396                  * the isize, but it is protected by i_mutex. So we can
7397                  * not unlock the i_mutex at this case.
7398                  */
7399                 if (offset + count <= inode->i_size) {
7400                         mutex_unlock(&inode->i_mutex);
7401                         relock = true;
7402                 }
7403                 ret = btrfs_delalloc_reserve_space(inode, count);
7404                 if (ret)
7405                         goto out;
7406         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7407                                      &BTRFS_I(inode)->runtime_flags))) {
7408                 inode_dio_done(inode);
7409                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7410                 wakeup = false;
7411         }
7412
7413         ret = __blockdev_direct_IO(rw, iocb, inode,
7414                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7415                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7416                         btrfs_submit_direct, flags);
7417         if (rw & WRITE) {
7418                 if (ret < 0 && ret != -EIOCBQUEUED)
7419                         btrfs_delalloc_release_space(inode, count);
7420                 else if (ret >= 0 && (size_t)ret < count)
7421                         btrfs_delalloc_release_space(inode,
7422                                                      count - (size_t)ret);
7423                 else
7424                         btrfs_delalloc_release_metadata(inode, 0);
7425         }
7426 out:
7427         if (wakeup)
7428                 inode_dio_done(inode);
7429         if (relock)
7430                 mutex_lock(&inode->i_mutex);
7431
7432         return ret;
7433 }
7434
7435 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7436
7437 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7438                 __u64 start, __u64 len)
7439 {
7440         int     ret;
7441
7442         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7443         if (ret)
7444                 return ret;
7445
7446         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7447 }
7448
7449 int btrfs_readpage(struct file *file, struct page *page)
7450 {
7451         struct extent_io_tree *tree;
7452         tree = &BTRFS_I(page->mapping->host)->io_tree;
7453         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7454 }
7455
7456 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7457 {
7458         struct extent_io_tree *tree;
7459
7460
7461         if (current->flags & PF_MEMALLOC) {
7462                 redirty_page_for_writepage(wbc, page);
7463                 unlock_page(page);
7464                 return 0;
7465         }
7466         tree = &BTRFS_I(page->mapping->host)->io_tree;
7467         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7468 }
7469
7470 static int btrfs_writepages(struct address_space *mapping,
7471                             struct writeback_control *wbc)
7472 {
7473         struct extent_io_tree *tree;
7474
7475         tree = &BTRFS_I(mapping->host)->io_tree;
7476         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7477 }
7478
7479 static int
7480 btrfs_readpages(struct file *file, struct address_space *mapping,
7481                 struct list_head *pages, unsigned nr_pages)
7482 {
7483         struct extent_io_tree *tree;
7484         tree = &BTRFS_I(mapping->host)->io_tree;
7485         return extent_readpages(tree, mapping, pages, nr_pages,
7486                                 btrfs_get_extent);
7487 }
7488 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7489 {
7490         struct extent_io_tree *tree;
7491         struct extent_map_tree *map;
7492         int ret;
7493
7494         tree = &BTRFS_I(page->mapping->host)->io_tree;
7495         map = &BTRFS_I(page->mapping->host)->extent_tree;
7496         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7497         if (ret == 1) {
7498                 ClearPagePrivate(page);
7499                 set_page_private(page, 0);
7500                 page_cache_release(page);
7501         }
7502         return ret;
7503 }
7504
7505 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7506 {
7507         if (PageWriteback(page) || PageDirty(page))
7508                 return 0;
7509         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7510 }
7511
7512 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
7513 {
7514         struct inode *inode = page->mapping->host;
7515         struct extent_io_tree *tree;
7516         struct btrfs_ordered_extent *ordered;
7517         struct extent_state *cached_state = NULL;
7518         u64 page_start = page_offset(page);
7519         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7520
7521         /*
7522          * we have the page locked, so new writeback can't start,
7523          * and the dirty bit won't be cleared while we are here.
7524          *
7525          * Wait for IO on this page so that we can safely clear
7526          * the PagePrivate2 bit and do ordered accounting
7527          */
7528         wait_on_page_writeback(page);
7529
7530         tree = &BTRFS_I(inode)->io_tree;
7531         if (offset) {
7532                 btrfs_releasepage(page, GFP_NOFS);
7533                 return;
7534         }
7535         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7536         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7537         if (ordered) {
7538                 /*
7539                  * IO on this page will never be started, so we need
7540                  * to account for any ordered extents now
7541                  */
7542                 clear_extent_bit(tree, page_start, page_end,
7543                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7544                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7545                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7546                 /*
7547                  * whoever cleared the private bit is responsible
7548                  * for the finish_ordered_io
7549                  */
7550                 if (TestClearPagePrivate2(page) &&
7551                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
7552                                                    PAGE_CACHE_SIZE, 1)) {
7553                         btrfs_finish_ordered_io(ordered);
7554                 }
7555                 btrfs_put_ordered_extent(ordered);
7556                 cached_state = NULL;
7557                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7558         }
7559         clear_extent_bit(tree, page_start, page_end,
7560                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7561                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7562                  &cached_state, GFP_NOFS);
7563         __btrfs_releasepage(page, GFP_NOFS);
7564
7565         ClearPageChecked(page);
7566         if (PagePrivate(page)) {
7567                 ClearPagePrivate(page);
7568                 set_page_private(page, 0);
7569                 page_cache_release(page);
7570         }
7571 }
7572
7573 /*
7574  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7575  * called from a page fault handler when a page is first dirtied. Hence we must
7576  * be careful to check for EOF conditions here. We set the page up correctly
7577  * for a written page which means we get ENOSPC checking when writing into
7578  * holes and correct delalloc and unwritten extent mapping on filesystems that
7579  * support these features.
7580  *
7581  * We are not allowed to take the i_mutex here so we have to play games to
7582  * protect against truncate races as the page could now be beyond EOF.  Because
7583  * vmtruncate() writes the inode size before removing pages, once we have the
7584  * page lock we can determine safely if the page is beyond EOF. If it is not
7585  * beyond EOF, then the page is guaranteed safe against truncation until we
7586  * unlock the page.
7587  */
7588 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7589 {
7590         struct page *page = vmf->page;
7591         struct inode *inode = file_inode(vma->vm_file);
7592         struct btrfs_root *root = BTRFS_I(inode)->root;
7593         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7594         struct btrfs_ordered_extent *ordered;
7595         struct extent_state *cached_state = NULL;
7596         char *kaddr;
7597         unsigned long zero_start;
7598         loff_t size;
7599         int ret;
7600         int reserved = 0;
7601         u64 page_start;
7602         u64 page_end;
7603
7604         sb_start_pagefault(inode->i_sb);
7605         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7606         if (!ret) {
7607                 ret = file_update_time(vma->vm_file);
7608                 reserved = 1;
7609         }
7610         if (ret) {
7611                 if (ret == -ENOMEM)
7612                         ret = VM_FAULT_OOM;
7613                 else /* -ENOSPC, -EIO, etc */
7614                         ret = VM_FAULT_SIGBUS;
7615                 if (reserved)
7616                         goto out;
7617                 goto out_noreserve;
7618         }
7619
7620         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7621 again:
7622         lock_page(page);
7623         size = i_size_read(inode);
7624         page_start = page_offset(page);
7625         page_end = page_start + PAGE_CACHE_SIZE - 1;
7626
7627         if ((page->mapping != inode->i_mapping) ||
7628             (page_start >= size)) {
7629                 /* page got truncated out from underneath us */
7630                 goto out_unlock;
7631         }
7632         wait_on_page_writeback(page);
7633
7634         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7635         set_page_extent_mapped(page);
7636
7637         /*
7638          * we can't set the delalloc bits if there are pending ordered
7639          * extents.  Drop our locks and wait for them to finish
7640          */
7641         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7642         if (ordered) {
7643                 unlock_extent_cached(io_tree, page_start, page_end,
7644                                      &cached_state, GFP_NOFS);
7645                 unlock_page(page);
7646                 btrfs_start_ordered_extent(inode, ordered, 1);
7647                 btrfs_put_ordered_extent(ordered);
7648                 goto again;
7649         }
7650
7651         /*
7652          * XXX - page_mkwrite gets called every time the page is dirtied, even
7653          * if it was already dirty, so for space accounting reasons we need to
7654          * clear any delalloc bits for the range we are fixing to save.  There
7655          * is probably a better way to do this, but for now keep consistent with
7656          * prepare_pages in the normal write path.
7657          */
7658         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7659                           EXTENT_DIRTY | EXTENT_DELALLOC |
7660                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7661                           0, 0, &cached_state, GFP_NOFS);
7662
7663         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7664                                         &cached_state);
7665         if (ret) {
7666                 unlock_extent_cached(io_tree, page_start, page_end,
7667                                      &cached_state, GFP_NOFS);
7668                 ret = VM_FAULT_SIGBUS;
7669                 goto out_unlock;
7670         }
7671         ret = 0;
7672
7673         /* page is wholly or partially inside EOF */
7674         if (page_start + PAGE_CACHE_SIZE > size)
7675                 zero_start = size & ~PAGE_CACHE_MASK;
7676         else
7677                 zero_start = PAGE_CACHE_SIZE;
7678
7679         if (zero_start != PAGE_CACHE_SIZE) {
7680                 kaddr = kmap(page);
7681                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7682                 flush_dcache_page(page);
7683                 kunmap(page);
7684         }
7685         ClearPageChecked(page);
7686         set_page_dirty(page);
7687         SetPageUptodate(page);
7688
7689         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7690         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7691         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7692
7693         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7694
7695 out_unlock:
7696         if (!ret) {
7697                 sb_end_pagefault(inode->i_sb);
7698                 return VM_FAULT_LOCKED;
7699         }
7700         unlock_page(page);
7701 out:
7702         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7703 out_noreserve:
7704         sb_end_pagefault(inode->i_sb);
7705         return ret;
7706 }
7707
7708 static int btrfs_truncate(struct inode *inode)
7709 {
7710         struct btrfs_root *root = BTRFS_I(inode)->root;
7711         struct btrfs_block_rsv *rsv;
7712         int ret;
7713         int err = 0;
7714         struct btrfs_trans_handle *trans;
7715         u64 mask = root->sectorsize - 1;
7716         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7717
7718         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
7719         if (ret)
7720                 return ret;
7721
7722         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7723         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
7724
7725         /*
7726          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7727          * 3 things going on here
7728          *
7729          * 1) We need to reserve space for our orphan item and the space to
7730          * delete our orphan item.  Lord knows we don't want to have a dangling
7731          * orphan item because we didn't reserve space to remove it.
7732          *
7733          * 2) We need to reserve space to update our inode.
7734          *
7735          * 3) We need to have something to cache all the space that is going to
7736          * be free'd up by the truncate operation, but also have some slack
7737          * space reserved in case it uses space during the truncate (thank you
7738          * very much snapshotting).
7739          *
7740          * And we need these to all be seperate.  The fact is we can use alot of
7741          * space doing the truncate, and we have no earthly idea how much space
7742          * we will use, so we need the truncate reservation to be seperate so it
7743          * doesn't end up using space reserved for updating the inode or
7744          * removing the orphan item.  We also need to be able to stop the
7745          * transaction and start a new one, which means we need to be able to
7746          * update the inode several times, and we have no idea of knowing how
7747          * many times that will be, so we can't just reserve 1 item for the
7748          * entirety of the opration, so that has to be done seperately as well.
7749          * Then there is the orphan item, which does indeed need to be held on
7750          * to for the whole operation, and we need nobody to touch this reserved
7751          * space except the orphan code.
7752          *
7753          * So that leaves us with
7754          *
7755          * 1) root->orphan_block_rsv - for the orphan deletion.
7756          * 2) rsv - for the truncate reservation, which we will steal from the
7757          * transaction reservation.
7758          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7759          * updating the inode.
7760          */
7761         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7762         if (!rsv)
7763                 return -ENOMEM;
7764         rsv->size = min_size;
7765         rsv->failfast = 1;
7766
7767         /*
7768          * 1 for the truncate slack space
7769          * 1 for updating the inode.
7770          */
7771         trans = btrfs_start_transaction(root, 2);
7772         if (IS_ERR(trans)) {
7773                 err = PTR_ERR(trans);
7774                 goto out;
7775         }
7776
7777         /* Migrate the slack space for the truncate to our reserve */
7778         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7779                                       min_size);
7780         BUG_ON(ret);
7781
7782         /*
7783          * setattr is responsible for setting the ordered_data_close flag,
7784          * but that is only tested during the last file release.  That
7785          * could happen well after the next commit, leaving a great big
7786          * window where new writes may get lost if someone chooses to write
7787          * to this file after truncating to zero
7788          *
7789          * The inode doesn't have any dirty data here, and so if we commit
7790          * this is a noop.  If someone immediately starts writing to the inode
7791          * it is very likely we'll catch some of their writes in this
7792          * transaction, and the commit will find this file on the ordered
7793          * data list with good things to send down.
7794          *
7795          * This is a best effort solution, there is still a window where
7796          * using truncate to replace the contents of the file will
7797          * end up with a zero length file after a crash.
7798          */
7799         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7800                                            &BTRFS_I(inode)->runtime_flags))
7801                 btrfs_add_ordered_operation(trans, root, inode);
7802
7803         /*
7804          * So if we truncate and then write and fsync we normally would just
7805          * write the extents that changed, which is a problem if we need to
7806          * first truncate that entire inode.  So set this flag so we write out
7807          * all of the extents in the inode to the sync log so we're completely
7808          * safe.
7809          */
7810         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7811         trans->block_rsv = rsv;
7812
7813         while (1) {
7814                 ret = btrfs_truncate_inode_items(trans, root, inode,
7815                                                  inode->i_size,
7816                                                  BTRFS_EXTENT_DATA_KEY);
7817                 if (ret != -ENOSPC) {
7818                         err = ret;
7819                         break;
7820                 }
7821
7822                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7823                 ret = btrfs_update_inode(trans, root, inode);
7824                 if (ret) {
7825                         err = ret;
7826                         break;
7827                 }
7828
7829                 btrfs_end_transaction(trans, root);
7830                 btrfs_btree_balance_dirty(root);
7831
7832                 trans = btrfs_start_transaction(root, 2);
7833                 if (IS_ERR(trans)) {
7834                         ret = err = PTR_ERR(trans);
7835                         trans = NULL;
7836                         break;
7837                 }
7838
7839                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7840                                               rsv, min_size);
7841                 BUG_ON(ret);    /* shouldn't happen */
7842                 trans->block_rsv = rsv;
7843         }
7844
7845         if (ret == 0 && inode->i_nlink > 0) {
7846                 trans->block_rsv = root->orphan_block_rsv;
7847                 ret = btrfs_orphan_del(trans, inode);
7848                 if (ret)
7849                         err = ret;
7850         }
7851
7852         if (trans) {
7853                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7854                 ret = btrfs_update_inode(trans, root, inode);
7855                 if (ret && !err)
7856                         err = ret;
7857
7858                 ret = btrfs_end_transaction(trans, root);
7859                 btrfs_btree_balance_dirty(root);
7860         }
7861
7862 out:
7863         btrfs_free_block_rsv(root, rsv);
7864
7865         if (ret && !err)
7866                 err = ret;
7867
7868         return err;
7869 }
7870
7871 /*
7872  * create a new subvolume directory/inode (helper for the ioctl).
7873  */
7874 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7875                              struct btrfs_root *new_root, u64 new_dirid)
7876 {
7877         struct inode *inode;
7878         int err;
7879         u64 index = 0;
7880
7881         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7882                                 new_dirid, new_dirid,
7883                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7884                                 &index);
7885         if (IS_ERR(inode))
7886                 return PTR_ERR(inode);
7887         inode->i_op = &btrfs_dir_inode_operations;
7888         inode->i_fop = &btrfs_dir_file_operations;
7889
7890         set_nlink(inode, 1);
7891         btrfs_i_size_write(inode, 0);
7892
7893         err = btrfs_update_inode(trans, new_root, inode);
7894
7895         iput(inode);
7896         return err;
7897 }
7898
7899 struct inode *btrfs_alloc_inode(struct super_block *sb)
7900 {
7901         struct btrfs_inode *ei;
7902         struct inode *inode;
7903
7904         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7905         if (!ei)
7906                 return NULL;
7907
7908         ei->root = NULL;
7909         ei->generation = 0;
7910         ei->last_trans = 0;
7911         ei->last_sub_trans = 0;
7912         ei->logged_trans = 0;
7913         ei->delalloc_bytes = 0;
7914         ei->disk_i_size = 0;
7915         ei->flags = 0;
7916         ei->csum_bytes = 0;
7917         ei->index_cnt = (u64)-1;
7918         ei->last_unlink_trans = 0;
7919         ei->last_log_commit = 0;
7920
7921         spin_lock_init(&ei->lock);
7922         ei->outstanding_extents = 0;
7923         ei->reserved_extents = 0;
7924
7925         ei->runtime_flags = 0;
7926         ei->force_compress = BTRFS_COMPRESS_NONE;
7927
7928         ei->delayed_node = NULL;
7929
7930         inode = &ei->vfs_inode;
7931         extent_map_tree_init(&ei->extent_tree);
7932         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7933         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7934         ei->io_tree.track_uptodate = 1;
7935         ei->io_failure_tree.track_uptodate = 1;
7936         atomic_set(&ei->sync_writers, 0);
7937         mutex_init(&ei->log_mutex);
7938         mutex_init(&ei->delalloc_mutex);
7939         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7940         INIT_LIST_HEAD(&ei->delalloc_inodes);
7941         INIT_LIST_HEAD(&ei->ordered_operations);
7942         RB_CLEAR_NODE(&ei->rb_node);
7943
7944         return inode;
7945 }
7946
7947 static void btrfs_i_callback(struct rcu_head *head)
7948 {
7949         struct inode *inode = container_of(head, struct inode, i_rcu);
7950         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7951 }
7952
7953 void btrfs_destroy_inode(struct inode *inode)
7954 {
7955         struct btrfs_ordered_extent *ordered;
7956         struct btrfs_root *root = BTRFS_I(inode)->root;
7957
7958         WARN_ON(!hlist_empty(&inode->i_dentry));
7959         WARN_ON(inode->i_data.nrpages);
7960         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7961         WARN_ON(BTRFS_I(inode)->reserved_extents);
7962         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7963         WARN_ON(BTRFS_I(inode)->csum_bytes);
7964
7965         /*
7966          * This can happen where we create an inode, but somebody else also
7967          * created the same inode and we need to destroy the one we already
7968          * created.
7969          */
7970         if (!root)
7971                 goto free;
7972
7973         /*
7974          * Make sure we're properly removed from the ordered operation
7975          * lists.
7976          */
7977         smp_mb();
7978         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7979                 spin_lock(&root->fs_info->ordered_extent_lock);
7980                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7981                 spin_unlock(&root->fs_info->ordered_extent_lock);
7982         }
7983
7984         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7985                      &BTRFS_I(inode)->runtime_flags)) {
7986                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7987                         (unsigned long long)btrfs_ino(inode));
7988                 atomic_dec(&root->orphan_inodes);
7989         }
7990
7991         while (1) {
7992                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7993                 if (!ordered)
7994                         break;
7995                 else {
7996                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7997                                 (unsigned long long)ordered->file_offset,
7998                                 (unsigned long long)ordered->len);
7999                         btrfs_remove_ordered_extent(inode, ordered);
8000                         btrfs_put_ordered_extent(ordered);
8001                         btrfs_put_ordered_extent(ordered);
8002                 }
8003         }
8004         inode_tree_del(inode);
8005         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
8006 free:
8007         call_rcu(&inode->i_rcu, btrfs_i_callback);
8008 }
8009
8010 int btrfs_drop_inode(struct inode *inode)
8011 {
8012         struct btrfs_root *root = BTRFS_I(inode)->root;
8013
8014         if (root == NULL)
8015                 return 1;
8016
8017         /* the snap/subvol tree is on deleting */
8018         if (btrfs_root_refs(&root->root_item) == 0 &&
8019             root != root->fs_info->tree_root)
8020                 return 1;
8021         else
8022                 return generic_drop_inode(inode);
8023 }
8024
8025 static void init_once(void *foo)
8026 {
8027         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
8028
8029         inode_init_once(&ei->vfs_inode);
8030 }
8031
8032 void btrfs_destroy_cachep(void)
8033 {
8034         /*
8035          * Make sure all delayed rcu free inodes are flushed before we
8036          * destroy cache.
8037          */
8038         rcu_barrier();
8039         if (btrfs_inode_cachep)
8040                 kmem_cache_destroy(btrfs_inode_cachep);
8041         if (btrfs_trans_handle_cachep)
8042                 kmem_cache_destroy(btrfs_trans_handle_cachep);
8043         if (btrfs_transaction_cachep)
8044                 kmem_cache_destroy(btrfs_transaction_cachep);
8045         if (btrfs_path_cachep)
8046                 kmem_cache_destroy(btrfs_path_cachep);
8047         if (btrfs_free_space_cachep)
8048                 kmem_cache_destroy(btrfs_free_space_cachep);
8049         if (btrfs_delalloc_work_cachep)
8050                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8051 }
8052
8053 int btrfs_init_cachep(void)
8054 {
8055         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8056                         sizeof(struct btrfs_inode), 0,
8057                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8058         if (!btrfs_inode_cachep)
8059                 goto fail;
8060
8061         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8062                         sizeof(struct btrfs_trans_handle), 0,
8063                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8064         if (!btrfs_trans_handle_cachep)
8065                 goto fail;
8066
8067         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8068                         sizeof(struct btrfs_transaction), 0,
8069                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8070         if (!btrfs_transaction_cachep)
8071                 goto fail;
8072
8073         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8074                         sizeof(struct btrfs_path), 0,
8075                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8076         if (!btrfs_path_cachep)
8077                 goto fail;
8078
8079         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8080                         sizeof(struct btrfs_free_space), 0,
8081                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8082         if (!btrfs_free_space_cachep)
8083                 goto fail;
8084
8085         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8086                         sizeof(struct btrfs_delalloc_work), 0,
8087                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8088                         NULL);
8089         if (!btrfs_delalloc_work_cachep)
8090                 goto fail;
8091
8092         return 0;
8093 fail:
8094         btrfs_destroy_cachep();
8095         return -ENOMEM;
8096 }
8097
8098 static int btrfs_getattr(struct vfsmount *mnt,
8099                          struct dentry *dentry, struct kstat *stat)
8100 {
8101         u64 delalloc_bytes;
8102         struct inode *inode = dentry->d_inode;
8103         u32 blocksize = inode->i_sb->s_blocksize;
8104
8105         generic_fillattr(inode, stat);
8106         stat->dev = BTRFS_I(inode)->root->anon_dev;
8107         stat->blksize = PAGE_CACHE_SIZE;
8108
8109         spin_lock(&BTRFS_I(inode)->lock);
8110         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8111         spin_unlock(&BTRFS_I(inode)->lock);
8112         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8113                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8114         return 0;
8115 }
8116
8117 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8118                            struct inode *new_dir, struct dentry *new_dentry)
8119 {
8120         struct btrfs_trans_handle *trans;
8121         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8122         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8123         struct inode *new_inode = new_dentry->d_inode;
8124         struct inode *old_inode = old_dentry->d_inode;
8125         struct timespec ctime = CURRENT_TIME;
8126         u64 index = 0;
8127         u64 root_objectid;
8128         int ret;
8129         u64 old_ino = btrfs_ino(old_inode);
8130
8131         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8132                 return -EPERM;
8133
8134         /* we only allow rename subvolume link between subvolumes */
8135         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8136                 return -EXDEV;
8137
8138         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8139             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8140                 return -ENOTEMPTY;
8141
8142         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8143             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8144                 return -ENOTEMPTY;
8145
8146
8147         /* check for collisions, even if the  name isn't there */
8148         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
8149                              new_dentry->d_name.name,
8150                              new_dentry->d_name.len);
8151
8152         if (ret) {
8153                 if (ret == -EEXIST) {
8154                         /* we shouldn't get
8155                          * eexist without a new_inode */
8156                         if (!new_inode) {
8157                                 WARN_ON(1);
8158                                 return ret;
8159                         }
8160                 } else {
8161                         /* maybe -EOVERFLOW */
8162                         return ret;
8163                 }
8164         }
8165         ret = 0;
8166
8167         /*
8168          * we're using rename to replace one file with another.
8169          * and the replacement file is large.  Start IO on it now so
8170          * we don't add too much work to the end of the transaction
8171          */
8172         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8173             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8174                 filemap_flush(old_inode->i_mapping);
8175
8176         /* close the racy window with snapshot create/destroy ioctl */
8177         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8178                 down_read(&root->fs_info->subvol_sem);
8179         /*
8180          * We want to reserve the absolute worst case amount of items.  So if
8181          * both inodes are subvols and we need to unlink them then that would
8182          * require 4 item modifications, but if they are both normal inodes it
8183          * would require 5 item modifications, so we'll assume their normal
8184          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8185          * should cover the worst case number of items we'll modify.
8186          */
8187         trans = btrfs_start_transaction(root, 11);
8188         if (IS_ERR(trans)) {
8189                 ret = PTR_ERR(trans);
8190                 goto out_notrans;
8191         }
8192
8193         if (dest != root)
8194                 btrfs_record_root_in_trans(trans, dest);
8195
8196         ret = btrfs_set_inode_index(new_dir, &index);
8197         if (ret)
8198                 goto out_fail;
8199
8200         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8201                 /* force full log commit if subvolume involved. */
8202                 root->fs_info->last_trans_log_full_commit = trans->transid;
8203         } else {
8204                 ret = btrfs_insert_inode_ref(trans, dest,
8205                                              new_dentry->d_name.name,
8206                                              new_dentry->d_name.len,
8207                                              old_ino,
8208                                              btrfs_ino(new_dir), index);
8209                 if (ret)
8210                         goto out_fail;
8211                 /*
8212                  * this is an ugly little race, but the rename is required
8213                  * to make sure that if we crash, the inode is either at the
8214                  * old name or the new one.  pinning the log transaction lets
8215                  * us make sure we don't allow a log commit to come in after
8216                  * we unlink the name but before we add the new name back in.
8217                  */
8218                 btrfs_pin_log_trans(root);
8219         }
8220         /*
8221          * make sure the inode gets flushed if it is replacing
8222          * something.
8223          */
8224         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8225                 btrfs_add_ordered_operation(trans, root, old_inode);
8226
8227         inode_inc_iversion(old_dir);
8228         inode_inc_iversion(new_dir);
8229         inode_inc_iversion(old_inode);
8230         old_dir->i_ctime = old_dir->i_mtime = ctime;
8231         new_dir->i_ctime = new_dir->i_mtime = ctime;
8232         old_inode->i_ctime = ctime;
8233
8234         if (old_dentry->d_parent != new_dentry->d_parent)
8235                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8236
8237         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8238                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8239                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8240                                         old_dentry->d_name.name,
8241                                         old_dentry->d_name.len);
8242         } else {
8243                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8244                                         old_dentry->d_inode,
8245                                         old_dentry->d_name.name,
8246                                         old_dentry->d_name.len);
8247                 if (!ret)
8248                         ret = btrfs_update_inode(trans, root, old_inode);
8249         }
8250         if (ret) {
8251                 btrfs_abort_transaction(trans, root, ret);
8252                 goto out_fail;
8253         }
8254
8255         if (new_inode) {
8256                 inode_inc_iversion(new_inode);
8257                 new_inode->i_ctime = CURRENT_TIME;
8258                 if (unlikely(btrfs_ino(new_inode) ==
8259                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8260                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8261                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8262                                                 root_objectid,
8263                                                 new_dentry->d_name.name,
8264                                                 new_dentry->d_name.len);
8265                         BUG_ON(new_inode->i_nlink == 0);
8266                 } else {
8267                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8268                                                  new_dentry->d_inode,
8269                                                  new_dentry->d_name.name,
8270                                                  new_dentry->d_name.len);
8271                 }
8272                 if (!ret && new_inode->i_nlink == 0) {
8273                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8274                         BUG_ON(ret);
8275                 }
8276                 if (ret) {
8277                         btrfs_abort_transaction(trans, root, ret);
8278                         goto out_fail;
8279                 }
8280         }
8281
8282         ret = btrfs_add_link(trans, new_dir, old_inode,
8283                              new_dentry->d_name.name,
8284                              new_dentry->d_name.len, 0, index);
8285         if (ret) {
8286                 btrfs_abort_transaction(trans, root, ret);
8287                 goto out_fail;
8288         }
8289
8290         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8291                 struct dentry *parent = new_dentry->d_parent;
8292                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8293                 btrfs_end_log_trans(root);
8294         }
8295 out_fail:
8296         btrfs_end_transaction(trans, root);
8297 out_notrans:
8298         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8299                 up_read(&root->fs_info->subvol_sem);
8300
8301         return ret;
8302 }
8303
8304 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8305 {
8306         struct btrfs_delalloc_work *delalloc_work;
8307
8308         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8309                                      work);
8310         if (delalloc_work->wait)
8311                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8312         else
8313                 filemap_flush(delalloc_work->inode->i_mapping);
8314
8315         if (delalloc_work->delay_iput)
8316                 btrfs_add_delayed_iput(delalloc_work->inode);
8317         else
8318                 iput(delalloc_work->inode);
8319         complete(&delalloc_work->completion);
8320 }
8321
8322 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8323                                                     int wait, int delay_iput)
8324 {
8325         struct btrfs_delalloc_work *work;
8326
8327         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8328         if (!work)
8329                 return NULL;
8330
8331         init_completion(&work->completion);
8332         INIT_LIST_HEAD(&work->list);
8333         work->inode = inode;
8334         work->wait = wait;
8335         work->delay_iput = delay_iput;
8336         work->work.func = btrfs_run_delalloc_work;
8337
8338         return work;
8339 }
8340
8341 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8342 {
8343         wait_for_completion(&work->completion);
8344         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8345 }
8346
8347 /*
8348  * some fairly slow code that needs optimization. This walks the list
8349  * of all the inodes with pending delalloc and forces them to disk.
8350  */
8351 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8352 {
8353         struct btrfs_inode *binode;
8354         struct inode *inode;
8355         struct btrfs_delalloc_work *work, *next;
8356         struct list_head works;
8357         struct list_head splice;
8358         int ret = 0;
8359
8360         if (root->fs_info->sb->s_flags & MS_RDONLY)
8361                 return -EROFS;
8362
8363         INIT_LIST_HEAD(&works);
8364         INIT_LIST_HEAD(&splice);
8365
8366         spin_lock(&root->fs_info->delalloc_lock);
8367         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
8368         while (!list_empty(&splice)) {
8369                 binode = list_entry(splice.next, struct btrfs_inode,
8370                                     delalloc_inodes);
8371
8372                 list_del_init(&binode->delalloc_inodes);
8373
8374                 inode = igrab(&binode->vfs_inode);
8375                 if (!inode) {
8376                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
8377                                   &binode->runtime_flags);
8378                         continue;
8379                 }
8380
8381                 list_add_tail(&binode->delalloc_inodes,
8382                               &root->fs_info->delalloc_inodes);
8383                 spin_unlock(&root->fs_info->delalloc_lock);
8384
8385                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8386                 if (unlikely(!work)) {
8387                         ret = -ENOMEM;
8388                         goto out;
8389                 }
8390                 list_add_tail(&work->list, &works);
8391                 btrfs_queue_worker(&root->fs_info->flush_workers,
8392                                    &work->work);
8393
8394                 cond_resched();
8395                 spin_lock(&root->fs_info->delalloc_lock);
8396         }
8397         spin_unlock(&root->fs_info->delalloc_lock);
8398
8399         list_for_each_entry_safe(work, next, &works, list) {
8400                 list_del_init(&work->list);
8401                 btrfs_wait_and_free_delalloc_work(work);
8402         }
8403
8404         /* the filemap_flush will queue IO into the worker threads, but
8405          * we have to make sure the IO is actually started and that
8406          * ordered extents get created before we return
8407          */
8408         atomic_inc(&root->fs_info->async_submit_draining);
8409         while (atomic_read(&root->fs_info->nr_async_submits) ||
8410               atomic_read(&root->fs_info->async_delalloc_pages)) {
8411                 wait_event(root->fs_info->async_submit_wait,
8412                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8413                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8414         }
8415         atomic_dec(&root->fs_info->async_submit_draining);
8416         return 0;
8417 out:
8418         list_for_each_entry_safe(work, next, &works, list) {
8419                 list_del_init(&work->list);
8420                 btrfs_wait_and_free_delalloc_work(work);
8421         }
8422
8423         if (!list_empty_careful(&splice)) {
8424                 spin_lock(&root->fs_info->delalloc_lock);
8425                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
8426                 spin_unlock(&root->fs_info->delalloc_lock);
8427         }
8428         return ret;
8429 }
8430
8431 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8432                          const char *symname)
8433 {
8434         struct btrfs_trans_handle *trans;
8435         struct btrfs_root *root = BTRFS_I(dir)->root;
8436         struct btrfs_path *path;
8437         struct btrfs_key key;
8438         struct inode *inode = NULL;
8439         int err;
8440         int drop_inode = 0;
8441         u64 objectid;
8442         u64 index = 0 ;
8443         int name_len;
8444         int datasize;
8445         unsigned long ptr;
8446         struct btrfs_file_extent_item *ei;
8447         struct extent_buffer *leaf;
8448
8449         name_len = strlen(symname) + 1;
8450         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8451                 return -ENAMETOOLONG;
8452
8453         /*
8454          * 2 items for inode item and ref
8455          * 2 items for dir items
8456          * 1 item for xattr if selinux is on
8457          */
8458         trans = btrfs_start_transaction(root, 5);
8459         if (IS_ERR(trans))
8460                 return PTR_ERR(trans);
8461
8462         err = btrfs_find_free_ino(root, &objectid);
8463         if (err)
8464                 goto out_unlock;
8465
8466         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8467                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8468                                 S_IFLNK|S_IRWXUGO, &index);
8469         if (IS_ERR(inode)) {
8470                 err = PTR_ERR(inode);
8471                 goto out_unlock;
8472         }
8473
8474         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8475         if (err) {
8476                 drop_inode = 1;
8477                 goto out_unlock;
8478         }
8479
8480         /*
8481         * If the active LSM wants to access the inode during
8482         * d_instantiate it needs these. Smack checks to see
8483         * if the filesystem supports xattrs by looking at the
8484         * ops vector.
8485         */
8486         inode->i_fop = &btrfs_file_operations;
8487         inode->i_op = &btrfs_file_inode_operations;
8488
8489         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8490         if (err)
8491                 drop_inode = 1;
8492         else {
8493                 inode->i_mapping->a_ops = &btrfs_aops;
8494                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8495                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8496         }
8497         if (drop_inode)
8498                 goto out_unlock;
8499
8500         path = btrfs_alloc_path();
8501         if (!path) {
8502                 err = -ENOMEM;
8503                 drop_inode = 1;
8504                 goto out_unlock;
8505         }
8506         key.objectid = btrfs_ino(inode);
8507         key.offset = 0;
8508         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8509         datasize = btrfs_file_extent_calc_inline_size(name_len);
8510         err = btrfs_insert_empty_item(trans, root, path, &key,
8511                                       datasize);
8512         if (err) {
8513                 drop_inode = 1;
8514                 btrfs_free_path(path);
8515                 goto out_unlock;
8516         }
8517         leaf = path->nodes[0];
8518         ei = btrfs_item_ptr(leaf, path->slots[0],
8519                             struct btrfs_file_extent_item);
8520         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8521         btrfs_set_file_extent_type(leaf, ei,
8522                                    BTRFS_FILE_EXTENT_INLINE);
8523         btrfs_set_file_extent_encryption(leaf, ei, 0);
8524         btrfs_set_file_extent_compression(leaf, ei, 0);
8525         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8526         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8527
8528         ptr = btrfs_file_extent_inline_start(ei);
8529         write_extent_buffer(leaf, symname, ptr, name_len);
8530         btrfs_mark_buffer_dirty(leaf);
8531         btrfs_free_path(path);
8532
8533         inode->i_op = &btrfs_symlink_inode_operations;
8534         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8535         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8536         inode_set_bytes(inode, name_len);
8537         btrfs_i_size_write(inode, name_len - 1);
8538         err = btrfs_update_inode(trans, root, inode);
8539         if (err)
8540                 drop_inode = 1;
8541
8542 out_unlock:
8543         if (!err)
8544                 d_instantiate(dentry, inode);
8545         btrfs_end_transaction(trans, root);
8546         if (drop_inode) {
8547                 inode_dec_link_count(inode);
8548                 iput(inode);
8549         }
8550         btrfs_btree_balance_dirty(root);
8551         return err;
8552 }
8553
8554 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8555                                        u64 start, u64 num_bytes, u64 min_size,
8556                                        loff_t actual_len, u64 *alloc_hint,
8557                                        struct btrfs_trans_handle *trans)
8558 {
8559         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8560         struct extent_map *em;
8561         struct btrfs_root *root = BTRFS_I(inode)->root;
8562         struct btrfs_key ins;
8563         u64 cur_offset = start;
8564         u64 i_size;
8565         u64 cur_bytes;
8566         int ret = 0;
8567         bool own_trans = true;
8568
8569         if (trans)
8570                 own_trans = false;
8571         while (num_bytes > 0) {
8572                 if (own_trans) {
8573                         trans = btrfs_start_transaction(root, 3);
8574                         if (IS_ERR(trans)) {
8575                                 ret = PTR_ERR(trans);
8576                                 break;
8577                         }
8578                 }
8579
8580                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8581                 cur_bytes = max(cur_bytes, min_size);
8582                 ret = btrfs_reserve_extent(trans, root, cur_bytes,
8583                                            min_size, 0, *alloc_hint, &ins, 1);
8584                 if (ret) {
8585                         if (own_trans)
8586                                 btrfs_end_transaction(trans, root);
8587                         break;
8588                 }
8589
8590                 ret = insert_reserved_file_extent(trans, inode,
8591                                                   cur_offset, ins.objectid,
8592                                                   ins.offset, ins.offset,
8593                                                   ins.offset, 0, 0, 0,
8594                                                   BTRFS_FILE_EXTENT_PREALLOC);
8595                 if (ret) {
8596                         btrfs_abort_transaction(trans, root, ret);
8597                         if (own_trans)
8598                                 btrfs_end_transaction(trans, root);
8599                         break;
8600                 }
8601                 btrfs_drop_extent_cache(inode, cur_offset,
8602                                         cur_offset + ins.offset -1, 0);
8603
8604                 em = alloc_extent_map();
8605                 if (!em) {
8606                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8607                                 &BTRFS_I(inode)->runtime_flags);
8608                         goto next;
8609                 }
8610
8611                 em->start = cur_offset;
8612                 em->orig_start = cur_offset;
8613                 em->len = ins.offset;
8614                 em->block_start = ins.objectid;
8615                 em->block_len = ins.offset;
8616                 em->orig_block_len = ins.offset;
8617                 em->ram_bytes = ins.offset;
8618                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8619                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8620                 em->generation = trans->transid;
8621
8622                 while (1) {
8623                         write_lock(&em_tree->lock);
8624                         ret = add_extent_mapping(em_tree, em, 1);
8625                         write_unlock(&em_tree->lock);
8626                         if (ret != -EEXIST)
8627                                 break;
8628                         btrfs_drop_extent_cache(inode, cur_offset,
8629                                                 cur_offset + ins.offset - 1,
8630                                                 0);
8631                 }
8632                 free_extent_map(em);
8633 next:
8634                 num_bytes -= ins.offset;
8635                 cur_offset += ins.offset;
8636                 *alloc_hint = ins.objectid + ins.offset;
8637
8638                 inode_inc_iversion(inode);
8639                 inode->i_ctime = CURRENT_TIME;
8640                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8641                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8642                     (actual_len > inode->i_size) &&
8643                     (cur_offset > inode->i_size)) {
8644                         if (cur_offset > actual_len)
8645                                 i_size = actual_len;
8646                         else
8647                                 i_size = cur_offset;
8648                         i_size_write(inode, i_size);
8649                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8650                 }
8651
8652                 ret = btrfs_update_inode(trans, root, inode);
8653
8654                 if (ret) {
8655                         btrfs_abort_transaction(trans, root, ret);
8656                         if (own_trans)
8657                                 btrfs_end_transaction(trans, root);
8658                         break;
8659                 }
8660
8661                 if (own_trans)
8662                         btrfs_end_transaction(trans, root);
8663         }
8664         return ret;
8665 }
8666
8667 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8668                               u64 start, u64 num_bytes, u64 min_size,
8669                               loff_t actual_len, u64 *alloc_hint)
8670 {
8671         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8672                                            min_size, actual_len, alloc_hint,
8673                                            NULL);
8674 }
8675
8676 int btrfs_prealloc_file_range_trans(struct inode *inode,
8677                                     struct btrfs_trans_handle *trans, int mode,
8678                                     u64 start, u64 num_bytes, u64 min_size,
8679                                     loff_t actual_len, u64 *alloc_hint)
8680 {
8681         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8682                                            min_size, actual_len, alloc_hint, trans);
8683 }
8684
8685 static int btrfs_set_page_dirty(struct page *page)
8686 {
8687         return __set_page_dirty_nobuffers(page);
8688 }
8689
8690 static int btrfs_permission(struct inode *inode, int mask)
8691 {
8692         struct btrfs_root *root = BTRFS_I(inode)->root;
8693         umode_t mode = inode->i_mode;
8694
8695         if (mask & MAY_WRITE &&
8696             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8697                 if (btrfs_root_readonly(root))
8698                         return -EROFS;
8699                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8700                         return -EACCES;
8701         }
8702         return generic_permission(inode, mask);
8703 }
8704
8705 static const struct inode_operations btrfs_dir_inode_operations = {
8706         .getattr        = btrfs_getattr,
8707         .lookup         = btrfs_lookup,
8708         .create         = btrfs_create,
8709         .unlink         = btrfs_unlink,
8710         .link           = btrfs_link,
8711         .mkdir          = btrfs_mkdir,
8712         .rmdir          = btrfs_rmdir,
8713         .rename         = btrfs_rename,
8714         .symlink        = btrfs_symlink,
8715         .setattr        = btrfs_setattr,
8716         .mknod          = btrfs_mknod,
8717         .setxattr       = btrfs_setxattr,
8718         .getxattr       = btrfs_getxattr,
8719         .listxattr      = btrfs_listxattr,
8720         .removexattr    = btrfs_removexattr,
8721         .permission     = btrfs_permission,
8722         .get_acl        = btrfs_get_acl,
8723 };
8724 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8725         .lookup         = btrfs_lookup,
8726         .permission     = btrfs_permission,
8727         .get_acl        = btrfs_get_acl,
8728 };
8729
8730 static const struct file_operations btrfs_dir_file_operations = {
8731         .llseek         = generic_file_llseek,
8732         .read           = generic_read_dir,
8733         .readdir        = btrfs_real_readdir,
8734         .unlocked_ioctl = btrfs_ioctl,
8735 #ifdef CONFIG_COMPAT
8736         .compat_ioctl   = btrfs_ioctl,
8737 #endif
8738         .release        = btrfs_release_file,
8739         .fsync          = btrfs_sync_file,
8740 };
8741
8742 static struct extent_io_ops btrfs_extent_io_ops = {
8743         .fill_delalloc = run_delalloc_range,
8744         .submit_bio_hook = btrfs_submit_bio_hook,
8745         .merge_bio_hook = btrfs_merge_bio_hook,
8746         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8747         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8748         .writepage_start_hook = btrfs_writepage_start_hook,
8749         .set_bit_hook = btrfs_set_bit_hook,
8750         .clear_bit_hook = btrfs_clear_bit_hook,
8751         .merge_extent_hook = btrfs_merge_extent_hook,
8752         .split_extent_hook = btrfs_split_extent_hook,
8753 };
8754
8755 /*
8756  * btrfs doesn't support the bmap operation because swapfiles
8757  * use bmap to make a mapping of extents in the file.  They assume
8758  * these extents won't change over the life of the file and they
8759  * use the bmap result to do IO directly to the drive.
8760  *
8761  * the btrfs bmap call would return logical addresses that aren't
8762  * suitable for IO and they also will change frequently as COW
8763  * operations happen.  So, swapfile + btrfs == corruption.
8764  *
8765  * For now we're avoiding this by dropping bmap.
8766  */
8767 static const struct address_space_operations btrfs_aops = {
8768         .readpage       = btrfs_readpage,
8769         .writepage      = btrfs_writepage,
8770         .writepages     = btrfs_writepages,
8771         .readpages      = btrfs_readpages,
8772         .direct_IO      = btrfs_direct_IO,
8773         .invalidatepage = btrfs_invalidatepage,
8774         .releasepage    = btrfs_releasepage,
8775         .set_page_dirty = btrfs_set_page_dirty,
8776         .error_remove_page = generic_error_remove_page,
8777 };
8778
8779 static const struct address_space_operations btrfs_symlink_aops = {
8780         .readpage       = btrfs_readpage,
8781         .writepage      = btrfs_writepage,
8782         .invalidatepage = btrfs_invalidatepage,
8783         .releasepage    = btrfs_releasepage,
8784 };
8785
8786 static const struct inode_operations btrfs_file_inode_operations = {
8787         .getattr        = btrfs_getattr,
8788         .setattr        = btrfs_setattr,
8789         .setxattr       = btrfs_setxattr,
8790         .getxattr       = btrfs_getxattr,
8791         .listxattr      = btrfs_listxattr,
8792         .removexattr    = btrfs_removexattr,
8793         .permission     = btrfs_permission,
8794         .fiemap         = btrfs_fiemap,
8795         .get_acl        = btrfs_get_acl,
8796         .update_time    = btrfs_update_time,
8797 };
8798 static const struct inode_operations btrfs_special_inode_operations = {
8799         .getattr        = btrfs_getattr,
8800         .setattr        = btrfs_setattr,
8801         .permission     = btrfs_permission,
8802         .setxattr       = btrfs_setxattr,
8803         .getxattr       = btrfs_getxattr,
8804         .listxattr      = btrfs_listxattr,
8805         .removexattr    = btrfs_removexattr,
8806         .get_acl        = btrfs_get_acl,
8807         .update_time    = btrfs_update_time,
8808 };
8809 static const struct inode_operations btrfs_symlink_inode_operations = {
8810         .readlink       = generic_readlink,
8811         .follow_link    = page_follow_link_light,
8812         .put_link       = page_put_link,
8813         .getattr        = btrfs_getattr,
8814         .setattr        = btrfs_setattr,
8815         .permission     = btrfs_permission,
8816         .setxattr       = btrfs_setxattr,
8817         .getxattr       = btrfs_getxattr,
8818         .listxattr      = btrfs_listxattr,
8819         .removexattr    = btrfs_removexattr,
8820         .get_acl        = btrfs_get_acl,
8821         .update_time    = btrfs_update_time,
8822 };
8823
8824 const struct dentry_operations btrfs_dentry_operations = {
8825         .d_delete       = btrfs_dentry_delete,
8826         .d_release      = btrfs_dentry_release,
8827 };