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