fs/btrfs/inode.c

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