fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/aio.h>
  36 #include <linux/bit_spinlock.h>
  37 #include <linux/xattr.h>
  38 #include <linux/posix_acl.h>
  39 #include <linux/falloc.h>
  40 #include <linux/slab.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/mount.h>
  43 #include <linux/btrfs.h>
  44 #include <linux/blkdev.h>
  45 #include <linux/posix_acl_xattr.h>
  46 #include "compat.h"
  47 #include "ctree.h"
  48 #include "disk-io.h"
  49 #include "transaction.h"
  50 #include "btrfs_inode.h"
  51 #include "print-tree.h"
  52 #include "ordered-data.h"
  53 #include "xattr.h"
  54 #include "tree-log.h"
  55 #include "volumes.h"
  56 #include "compression.h"
  57 #include "locking.h"
  58 #include "free-space-cache.h"
  59 #include "inode-map.h"
  60 #include "backref.h"
  61 #include "hash.h"
  62
  63 struct btrfs_iget_args {
  64         u64 ino;
  65         struct btrfs_root *root;
  66 };
  67
  68 static const struct inode_operations btrfs_dir_inode_operations;
  69 static const struct inode_operations btrfs_symlink_inode_operations;
  70 static const struct inode_operations btrfs_dir_ro_inode_operations;
  71 static const struct inode_operations btrfs_special_inode_operations;
  72 static const struct inode_operations btrfs_file_inode_operations;
  73 static const struct address_space_operations btrfs_aops;
  74 static const struct address_space_operations btrfs_symlink_aops;
  75 static const struct file_operations btrfs_dir_file_operations;
  76 static struct extent_io_ops btrfs_extent_io_ops;
  77
  78 static struct kmem_cache *btrfs_inode_cachep;
  79 static struct kmem_cache *btrfs_delalloc_work_cachep;
  80 struct kmem_cache *btrfs_trans_handle_cachep;
  81 struct kmem_cache *btrfs_transaction_cachep;
  82 struct kmem_cache *btrfs_path_cachep;
  83 struct kmem_cache *btrfs_free_space_cachep;
  84
  85 #define S_SHIFT 12
  86 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  87         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  88         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  89         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  90         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  91         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  92         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  93         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  94 };
  95
  96 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  97 static int btrfs_truncate(struct inode *inode);
  98 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  99 static noinline int cow_file_range(struct inode *inode,
 100                                    struct page *locked_page,
 101                                    u64 start, u64 end, int *page_started,
 102                                    unsigned long *nr_written, int unlock);
 103 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 104                                            u64 len, u64 orig_start,
 105                                            u64 block_start, u64 block_len,
 106                                            u64 orig_block_len, u64 ram_bytes,
 107                                            int type);
 108
 109 static int btrfs_dirty_inode(struct inode *inode);
 110
 111 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 112                                      struct inode *inode,  struct inode *dir,
 113                                      const struct qstr *qstr)
 114 {
 115         int err;
 116
 117         err = btrfs_init_acl(trans, inode, dir);
 118         if (!err)
 119                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 120         return err;
 121 }
 122
 123 /*
 124  * this does all the hard work for inserting an inline extent into
 125  * the btree.  The caller should have done a btrfs_drop_extents so that
 126  * no overlapping inline items exist in the btree
 127  */
 128 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 129                                 struct btrfs_root *root, struct inode *inode,
 130                                 u64 start, size_t size, size_t compressed_size,
 131                                 int compress_type,
 132                                 struct page **compressed_pages)
 133 {
 134         struct btrfs_key key;
 135         struct btrfs_path *path;
 136         struct extent_buffer *leaf;
 137         struct page *page = NULL;
 138         char *kaddr;
 139         unsigned long ptr;
 140         struct btrfs_file_extent_item *ei;
 141         int err = 0;
 142         int ret;
 143         size_t cur_size = size;
 144         size_t datasize;
 145         unsigned long offset;
 146
 147         if (compressed_size && compressed_pages)
 148                 cur_size = compressed_size;
 149
 150         path = btrfs_alloc_path();
 151         if (!path)
 152                 return -ENOMEM;
 153
 154         path->leave_spinning = 1;
 155
 156         key.objectid = btrfs_ino(inode);
 157         key.offset = start;
 158         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 159         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 160
 161         inode_add_bytes(inode, size);
 162         ret = btrfs_insert_empty_item(trans, root, path, &key,
 163                                       datasize);
 164         if (ret) {
 165                 err = ret;
 166                 goto fail;
 167         }
 168         leaf = path->nodes[0];
 169         ei = btrfs_item_ptr(leaf, path->slots[0],
 170                             struct btrfs_file_extent_item);
 171         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 172         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 173         btrfs_set_file_extent_encryption(leaf, ei, 0);
 174         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 175         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 176         ptr = btrfs_file_extent_inline_start(ei);
 177
 178         if (compress_type != BTRFS_COMPRESS_NONE) {
 179                 struct page *cpage;
 180                 int i = 0;
 181                 while (compressed_size > 0) {
 182                         cpage = compressed_pages[i];
 183                         cur_size = min_t(unsigned long, compressed_size,
 184                                        PAGE_CACHE_SIZE);
 185
 186                         kaddr = kmap_atomic(cpage);
 187                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 188                         kunmap_atomic(kaddr);
 189
 190                         i++;
 191                         ptr += cur_size;
 192                         compressed_size -= cur_size;
 193                 }
 194                 btrfs_set_file_extent_compression(leaf, ei,
 195                                                   compress_type);
 196         } else {
 197                 page = find_get_page(inode->i_mapping,
 198                                      start >> PAGE_CACHE_SHIFT);
 199                 btrfs_set_file_extent_compression(leaf, ei, 0);
 200                 kaddr = kmap_atomic(page);
 201                 offset = start & (PAGE_CACHE_SIZE - 1);
 202                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 203                 kunmap_atomic(kaddr);
 204                 page_cache_release(page);
 205         }
 206         btrfs_mark_buffer_dirty(leaf);
 207         btrfs_free_path(path);
 208
 209         /*
 210          * we're an inline extent, so nobody can
 211          * extend the file past i_size without locking
 212          * a page we already have locked.
 213          *
 214          * We must do any isize and inode updates
 215          * before we unlock the pages.  Otherwise we
 216          * could end up racing with unlink.
 217          */
 218         BTRFS_I(inode)->disk_i_size = inode->i_size;
 219         ret = btrfs_update_inode(trans, root, inode);
 220
 221         return ret;
 222 fail:
 223         btrfs_free_path(path);
 224         return err;
 225 }
 226
 227
 228 /*
 229  * conditionally insert an inline extent into the file.  This
 230  * does the checks required to make sure the data is small enough
 231  * to fit as an inline extent.
 232  */
 233 static noinline int cow_file_range_inline(struct btrfs_root *root,
 234                                           struct inode *inode, u64 start,
 235                                           u64 end, size_t compressed_size,
 236                                           int compress_type,
 237                                           struct page **compressed_pages)
 238 {
 239         struct btrfs_trans_handle *trans;
 240         u64 isize = i_size_read(inode);
 241         u64 actual_end = min(end + 1, isize);
 242         u64 inline_len = actual_end - start;
 243         u64 aligned_end = ALIGN(end, root->sectorsize);
 244         u64 data_len = inline_len;
 245         int ret;
 246
 247         if (compressed_size)
 248                 data_len = compressed_size;
 249
 250         if (start > 0 ||
 251             actual_end >= PAGE_CACHE_SIZE ||
 252             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 253             (!compressed_size &&
 254             (actual_end & (root->sectorsize - 1)) == 0) ||
 255             end + 1 < isize ||
 256             data_len > root->fs_info->max_inline) {
 257                 return 1;
 258         }
 259
 260         trans = btrfs_join_transaction(root);
 261         if (IS_ERR(trans))
 262                 return PTR_ERR(trans);
 263         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 264
 265         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 266         if (ret) {
 267                 btrfs_abort_transaction(trans, root, ret);
 268                 goto out;
 269         }
 270
 271         if (isize > actual_end)
 272                 inline_len = min_t(u64, isize, actual_end);
 273         ret = insert_inline_extent(trans, root, inode, start,
 274                                    inline_len, compressed_size,
 275                                    compress_type, compressed_pages);
 276         if (ret && ret != -ENOSPC) {
 277                 btrfs_abort_transaction(trans, root, ret);
 278                 goto out;
 279         } else if (ret == -ENOSPC) {
 280                 ret = 1;
 281                 goto out;
 282         }
 283
 284         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 285         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 286         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 287 out:
 288         btrfs_end_transaction(trans, root);
 289         return ret;
 290 }
 291
 292 struct async_extent {
 293         u64 start;
 294         u64 ram_size;
 295         u64 compressed_size;
 296         struct page **pages;
 297         unsigned long nr_pages;
 298         int compress_type;
 299         struct list_head list;
 300 };
 301
 302 struct async_cow {
 303         struct inode *inode;
 304         struct btrfs_root *root;
 305         struct page *locked_page;
 306         u64 start;
 307         u64 end;
 308         struct list_head extents;
 309         struct btrfs_work work;
 310 };
 311
 312 static noinline int add_async_extent(struct async_cow *cow,
 313                                      u64 start, u64 ram_size,
 314                                      u64 compressed_size,
 315                                      struct page **pages,
 316                                      unsigned long nr_pages,
 317                                      int compress_type)
 318 {
 319         struct async_extent *async_extent;
 320
 321         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 322         BUG_ON(!async_extent); /* -ENOMEM */
 323         async_extent->start = start;
 324         async_extent->ram_size = ram_size;
 325         async_extent->compressed_size = compressed_size;
 326         async_extent->pages = pages;
 327         async_extent->nr_pages = nr_pages;
 328         async_extent->compress_type = compress_type;
 329         list_add_tail(&async_extent->list, &cow->extents);
 330         return 0;
 331 }
 332
 333 /*
 334  * we create compressed extents in two phases.  The first
 335  * phase compresses a range of pages that have already been
 336  * locked (both pages and state bits are locked).
 337  *
 338  * This is done inside an ordered work queue, and the compression
 339  * is spread across many cpus.  The actual IO submission is step
 340  * two, and the ordered work queue takes care of making sure that
 341  * happens in the same order things were put onto the queue by
 342  * writepages and friends.
 343  *
 344  * If this code finds it can't get good compression, it puts an
 345  * entry onto the work queue to write the uncompressed bytes.  This
 346  * makes sure that both compressed inodes and uncompressed inodes
 347  * are written in the same order that the flusher thread sent them
 348  * down.
 349  */
 350 static noinline int compress_file_range(struct inode *inode,
 351                                         struct page *locked_page,
 352                                         u64 start, u64 end,
 353                                         struct async_cow *async_cow,
 354                                         int *num_added)
 355 {
 356         struct btrfs_root *root = BTRFS_I(inode)->root;
 357         u64 num_bytes;
 358         u64 blocksize = root->sectorsize;
 359         u64 actual_end;
 360         u64 isize = i_size_read(inode);
 361         int ret = 0;
 362         struct page **pages = NULL;
 363         unsigned long nr_pages;
 364         unsigned long nr_pages_ret = 0;
 365         unsigned long total_compressed = 0;
 366         unsigned long total_in = 0;
 367         unsigned long max_compressed = 128 * 1024;
 368         unsigned long max_uncompressed = 128 * 1024;
 369         int i;
 370         int will_compress;
 371         int compress_type = root->fs_info->compress_type;
 372         int redirty = 0;
 373
 374         /* if this is a small write inside eof, kick off a defrag */
 375         if ((end - start + 1) < 16 * 1024 &&
 376             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 377                 btrfs_add_inode_defrag(NULL, inode);
 378
 379         actual_end = min_t(u64, isize, end + 1);
 380 again:
 381         will_compress = 0;
 382         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 383         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 384
 385         /*
 386          * we don't want to send crud past the end of i_size through
 387          * compression, that's just a waste of CPU time.  So, if the
 388          * end of the file is before the start of our current
 389          * requested range of bytes, we bail out to the uncompressed
 390          * cleanup code that can deal with all of this.
 391          *
 392          * It isn't really the fastest way to fix things, but this is a
 393          * very uncommon corner.
 394          */
 395         if (actual_end <= start)
 396                 goto cleanup_and_bail_uncompressed;
 397
 398         total_compressed = actual_end - start;
 399
 400         /* we want to make sure that amount of ram required to uncompress
 401          * an extent is reasonable, so we limit the total size in ram
 402          * of a compressed extent to 128k.  This is a crucial number
 403          * because it also controls how easily we can spread reads across
 404          * cpus for decompression.
 405          *
 406          * We also want to make sure the amount of IO required to do
 407          * a random read is reasonably small, so we limit the size of
 408          * a compressed extent to 128k.
 409          */
 410         total_compressed = min(total_compressed, max_uncompressed);
 411         num_bytes = ALIGN(end - start + 1, blocksize);
 412         num_bytes = max(blocksize,  num_bytes);
 413         total_in = 0;
 414         ret = 0;
 415
 416         /*
 417          * we do compression for mount -o compress and when the
 418          * inode has not been flagged as nocompress.  This flag can
 419          * change at any time if we discover bad compression ratios.
 420          */
 421         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 422             (btrfs_test_opt(root, COMPRESS) ||
 423              (BTRFS_I(inode)->force_compress) ||
 424              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 425                 WARN_ON(pages);
 426                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 427                 if (!pages) {
 428                         /* just bail out to the uncompressed code */
 429                         goto cont;
 430                 }
 431
 432                 if (BTRFS_I(inode)->force_compress)
 433                         compress_type = BTRFS_I(inode)->force_compress;
 434
 435                 /*
 436                  * we need to call clear_page_dirty_for_io on each
 437                  * page in the range.  Otherwise applications with the file
 438                  * mmap'd can wander in and change the page contents while
 439                  * we are compressing them.
 440                  *
 441                  * If the compression fails for any reason, we set the pages
 442                  * dirty again later on.
 443                  */
 444                 extent_range_clear_dirty_for_io(inode, start, end);
 445                 redirty = 1;
 446                 ret = btrfs_compress_pages(compress_type,
 447                                            inode->i_mapping, start,
 448                                            total_compressed, pages,
 449                                            nr_pages, &nr_pages_ret,
 450                                            &total_in,
 451                                            &total_compressed,
 452                                            max_compressed);
 453
 454                 if (!ret) {
 455                         unsigned long offset = total_compressed &
 456                                 (PAGE_CACHE_SIZE - 1);
 457                         struct page *page = pages[nr_pages_ret - 1];
 458                         char *kaddr;
 459
 460                         /* zero the tail end of the last page, we might be
 461                          * sending it down to disk
 462                          */
 463                         if (offset) {
 464                                 kaddr = kmap_atomic(page);
 465                                 memset(kaddr + offset, 0,
 466                                        PAGE_CACHE_SIZE - offset);
 467                                 kunmap_atomic(kaddr);
 468                         }
 469                         will_compress = 1;
 470                 }
 471         }
 472 cont:
 473         if (start == 0) {
 474                 /* lets try to make an inline extent */
 475                 if (ret || total_in < (actual_end - start)) {
 476                         /* we didn't compress the entire range, try
 477                          * to make an uncompressed inline extent.
 478                          */
 479                         ret = cow_file_range_inline(root, inode, start, end,
 480                                                     0, 0, NULL);
 481                 } else {
 482                         /* try making a compressed inline extent */
 483                         ret = cow_file_range_inline(root, inode, start, end,
 484                                                     total_compressed,
 485                                                     compress_type, pages);
 486                 }
 487                 if (ret <= 0) {
 488                         unsigned long clear_flags = EXTENT_DELALLOC |
 489                                 EXTENT_DEFRAG;
 490                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 491
 492                         /*
 493                          * inline extent creation worked or returned error,
 494                          * we don't need to create any more async work items.
 495                          * Unlock and free up our temp pages.
 496                          */
 497                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 498                                                      clear_flags, PAGE_UNLOCK |
 499                                                      PAGE_CLEAR_DIRTY |
 500                                                      PAGE_SET_WRITEBACK |
 501                                                      PAGE_END_WRITEBACK);
 502                         goto free_pages_out;
 503                 }
 504         }
 505
 506         if (will_compress) {
 507                 /*
 508                  * we aren't doing an inline extent round the compressed size
 509                  * up to a block size boundary so the allocator does sane
 510                  * things
 511                  */
 512                 total_compressed = ALIGN(total_compressed, blocksize);
 513
 514                 /*
 515                  * one last check to make sure the compression is really a
 516                  * win, compare the page count read with the blocks on disk
 517                  */
 518                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 519                 if (total_compressed >= total_in) {
 520                         will_compress = 0;
 521                 } else {
 522                         num_bytes = total_in;
 523                 }
 524         }
 525         if (!will_compress && pages) {
 526                 /*
 527                  * the compression code ran but failed to make things smaller,
 528                  * free any pages it allocated and our page pointer array
 529                  */
 530                 for (i = 0; i < nr_pages_ret; i++) {
 531                         WARN_ON(pages[i]->mapping);
 532                         page_cache_release(pages[i]);
 533                 }
 534                 kfree(pages);
 535                 pages = NULL;
 536                 total_compressed = 0;
 537                 nr_pages_ret = 0;
 538
 539                 /* flag the file so we don't compress in the future */
 540                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 541                     !(BTRFS_I(inode)->force_compress)) {
 542                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 543                 }
 544         }
 545         if (will_compress) {
 546                 *num_added += 1;
 547
 548                 /* the async work queues will take care of doing actual
 549                  * allocation on disk for these compressed pages,
 550                  * and will submit them to the elevator.
 551                  */
 552                 add_async_extent(async_cow, start, num_bytes,
 553                                  total_compressed, pages, nr_pages_ret,
 554                                  compress_type);
 555
 556                 if (start + num_bytes < end) {
 557                         start += num_bytes;
 558                         pages = NULL;
 559                         cond_resched();
 560                         goto again;
 561                 }
 562         } else {
 563 cleanup_and_bail_uncompressed:
 564                 /*
 565                  * No compression, but we still need to write the pages in
 566                  * the file we've been given so far.  redirty the locked
 567                  * page if it corresponds to our extent and set things up
 568                  * for the async work queue to run cow_file_range to do
 569                  * the normal delalloc dance
 570                  */
 571                 if (page_offset(locked_page) >= start &&
 572                     page_offset(locked_page) <= end) {
 573                         __set_page_dirty_nobuffers(locked_page);
 574                         /* unlocked later on in the async handlers */
 575                 }
 576                 if (redirty)
 577                         extent_range_redirty_for_io(inode, start, end);
 578                 add_async_extent(async_cow, start, end - start + 1,
 579                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 580                 *num_added += 1;
 581         }
 582
 583 out:
 584         return ret;
 585
 586 free_pages_out:
 587         for (i = 0; i < nr_pages_ret; i++) {
 588                 WARN_ON(pages[i]->mapping);
 589                 page_cache_release(pages[i]);
 590         }
 591         kfree(pages);
 592
 593         goto out;
 594 }
 595
 596 /*
 597  * phase two of compressed writeback.  This is the ordered portion
 598  * of the code, which only gets called in the order the work was
 599  * queued.  We walk all the async extents created by compress_file_range
 600  * and send them down to the disk.
 601  */
 602 static noinline int submit_compressed_extents(struct inode *inode,
 603                                               struct async_cow *async_cow)
 604 {
 605         struct async_extent *async_extent;
 606         u64 alloc_hint = 0;
 607         struct btrfs_key ins;
 608         struct extent_map *em;
 609         struct btrfs_root *root = BTRFS_I(inode)->root;
 610         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 611         struct extent_io_tree *io_tree;
 612         int ret = 0;
 613
 614         if (list_empty(&async_cow->extents))
 615                 return 0;
 616
 617 again:
 618         while (!list_empty(&async_cow->extents)) {
 619                 async_extent = list_entry(async_cow->extents.next,
 620                                           struct async_extent, list);
 621                 list_del(&async_extent->list);
 622
 623                 io_tree = &BTRFS_I(inode)->io_tree;
 624
 625 retry:
 626                 /* did the compression code fall back to uncompressed IO? */
 627                 if (!async_extent->pages) {
 628                         int page_started = 0;
 629                         unsigned long nr_written = 0;
 630
 631                         lock_extent(io_tree, async_extent->start,
 632                                          async_extent->start +
 633                                          async_extent->ram_size - 1);
 634
 635                         /* allocate blocks */
 636                         ret = cow_file_range(inode, async_cow->locked_page,
 637                                              async_extent->start,
 638                                              async_extent->start +
 639                                              async_extent->ram_size - 1,
 640                                              &page_started, &nr_written, 0);
 641
 642                         /* JDM XXX */
 643
 644                         /*
 645                          * if page_started, cow_file_range inserted an
 646                          * inline extent and took care of all the unlocking
 647                          * and IO for us.  Otherwise, we need to submit
 648                          * all those pages down to the drive.
 649                          */
 650                         if (!page_started && !ret)
 651                                 extent_write_locked_range(io_tree,
 652                                                   inode, async_extent->start,
 653                                                   async_extent->start +
 654                                                   async_extent->ram_size - 1,
 655                                                   btrfs_get_extent,
 656                                                   WB_SYNC_ALL);
 657                         else if (ret)
 658                                 unlock_page(async_cow->locked_page);
 659                         kfree(async_extent);
 660                         cond_resched();
 661                         continue;
 662                 }
 663
 664                 lock_extent(io_tree, async_extent->start,
 665                             async_extent->start + async_extent->ram_size - 1);
 666
 667                 ret = btrfs_reserve_extent(root,
 668                                            async_extent->compressed_size,
 669                                            async_extent->compressed_size,
 670                                            0, alloc_hint, &ins, 1);
 671                 if (ret) {
 672                         int i;
 673
 674                         for (i = 0; i < async_extent->nr_pages; i++) {
 675                                 WARN_ON(async_extent->pages[i]->mapping);
 676                                 page_cache_release(async_extent->pages[i]);
 677                         }
 678                         kfree(async_extent->pages);
 679                         async_extent->nr_pages = 0;
 680                         async_extent->pages = NULL;
 681
 682                         if (ret == -ENOSPC) {
 683                                 unlock_extent(io_tree, async_extent->start,
 684                                               async_extent->start +
 685                                               async_extent->ram_size - 1);
 686                                 goto retry;
 687                         }
 688                         goto out_free;
 689                 }
 690
 691                 /*
 692                  * here we're doing allocation and writeback of the
 693                  * compressed pages
 694                  */
 695                 btrfs_drop_extent_cache(inode, async_extent->start,
 696                                         async_extent->start +
 697                                         async_extent->ram_size - 1, 0);
 698
 699                 em = alloc_extent_map();
 700                 if (!em) {
 701                         ret = -ENOMEM;
 702                         goto out_free_reserve;
 703                 }
 704                 em->start = async_extent->start;
 705                 em->len = async_extent->ram_size;
 706                 em->orig_start = em->start;
 707                 em->mod_start = em->start;
 708                 em->mod_len = em->len;
 709
 710                 em->block_start = ins.objectid;
 711                 em->block_len = ins.offset;
 712                 em->orig_block_len = ins.offset;
 713                 em->ram_bytes = async_extent->ram_size;
 714                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 715                 em->compress_type = async_extent->compress_type;
 716                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 717                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 718                 em->generation = -1;
 719
 720                 while (1) {
 721                         write_lock(&em_tree->lock);
 722                         ret = add_extent_mapping(em_tree, em, 1);
 723                         write_unlock(&em_tree->lock);
 724                         if (ret != -EEXIST) {
 725                                 free_extent_map(em);
 726                                 break;
 727                         }
 728                         btrfs_drop_extent_cache(inode, async_extent->start,
 729                                                 async_extent->start +
 730                                                 async_extent->ram_size - 1, 0);
 731                 }
 732
 733                 if (ret)
 734                         goto out_free_reserve;
 735
 736                 ret = btrfs_add_ordered_extent_compress(inode,
 737                                                 async_extent->start,
 738                                                 ins.objectid,
 739                                                 async_extent->ram_size,
 740                                                 ins.offset,
 741                                                 BTRFS_ORDERED_COMPRESSED,
 742                                                 async_extent->compress_type);
 743                 if (ret)
 744                         goto out_free_reserve;
 745
 746                 /*
 747                  * clear dirty, set writeback and unlock the pages.
 748                  */
 749                 extent_clear_unlock_delalloc(inode, async_extent->start,
 750                                 async_extent->start +
 751                                 async_extent->ram_size - 1,
 752                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 753                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 754                                 PAGE_SET_WRITEBACK);
 755                 ret = btrfs_submit_compressed_write(inode,
 756                                     async_extent->start,
 757                                     async_extent->ram_size,
 758                                     ins.objectid,
 759                                     ins.offset, async_extent->pages,
 760                                     async_extent->nr_pages);
 761                 alloc_hint = ins.objectid + ins.offset;
 762                 kfree(async_extent);
 763                 if (ret)
 764                         goto out;
 765                 cond_resched();
 766         }
 767         ret = 0;
 768 out:
 769         return ret;
 770 out_free_reserve:
 771         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 772 out_free:
 773         extent_clear_unlock_delalloc(inode, async_extent->start,
 774                                      async_extent->start +
 775                                      async_extent->ram_size - 1,
 776                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 777                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 778                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 779                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 780         kfree(async_extent);
 781         goto again;
 782 }
 783
 784 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 785                                       u64 num_bytes)
 786 {
 787         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 788         struct extent_map *em;
 789         u64 alloc_hint = 0;
 790
 791         read_lock(&em_tree->lock);
 792         em = search_extent_mapping(em_tree, start, num_bytes);
 793         if (em) {
 794                 /*
 795                  * if block start isn't an actual block number then find the
 796                  * first block in this inode and use that as a hint.  If that
 797                  * block is also bogus then just don't worry about it.
 798                  */
 799                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 800                         free_extent_map(em);
 801                         em = search_extent_mapping(em_tree, 0, 0);
 802                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 803                                 alloc_hint = em->block_start;
 804                         if (em)
 805                                 free_extent_map(em);
 806                 } else {
 807                         alloc_hint = em->block_start;
 808                         free_extent_map(em);
 809                 }
 810         }
 811         read_unlock(&em_tree->lock);
 812
 813         return alloc_hint;
 814 }
 815
 816 /*
 817  * when extent_io.c finds a delayed allocation range in the file,
 818  * the call backs end up in this code.  The basic idea is to
 819  * allocate extents on disk for the range, and create ordered data structs
 820  * in ram to track those extents.
 821  *
 822  * locked_page is the page that writepage had locked already.  We use
 823  * it to make sure we don't do extra locks or unlocks.
 824  *
 825  * *page_started is set to one if we unlock locked_page and do everything
 826  * required to start IO on it.  It may be clean and already done with
 827  * IO when we return.
 828  */
 829 static noinline int cow_file_range(struct inode *inode,
 830                                    struct page *locked_page,
 831                                    u64 start, u64 end, int *page_started,
 832                                    unsigned long *nr_written,
 833                                    int unlock)
 834 {
 835         struct btrfs_root *root = BTRFS_I(inode)->root;
 836         u64 alloc_hint = 0;
 837         u64 num_bytes;
 838         unsigned long ram_size;
 839         u64 disk_num_bytes;
 840         u64 cur_alloc_size;
 841         u64 blocksize = root->sectorsize;
 842         struct btrfs_key ins;
 843         struct extent_map *em;
 844         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 845         int ret = 0;
 846
 847         BUG_ON(btrfs_is_free_space_inode(inode));
 848
 849         num_bytes = ALIGN(end - start + 1, blocksize);
 850         num_bytes = max(blocksize,  num_bytes);
 851         disk_num_bytes = num_bytes;
 852
 853         /* if this is a small write inside eof, kick off defrag */
 854         if (num_bytes < 64 * 1024 &&
 855             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 856                 btrfs_add_inode_defrag(NULL, inode);
 857
 858         if (start == 0) {
 859                 /* lets try to make an inline extent */
 860                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 861                                             NULL);
 862                 if (ret == 0) {
 863                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 864                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 865                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 866                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 867                                      PAGE_END_WRITEBACK);
 868
 869                         *nr_written = *nr_written +
 870                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 871                         *page_started = 1;
 872                         goto out;
 873                 } else if (ret < 0) {
 874                         goto out_unlock;
 875                 }
 876         }
 877
 878         BUG_ON(disk_num_bytes >
 879                btrfs_super_total_bytes(root->fs_info->super_copy));
 880
 881         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 882         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 883
 884         while (disk_num_bytes > 0) {
 885                 unsigned long op;
 886
 887                 cur_alloc_size = disk_num_bytes;
 888                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 889                                            root->sectorsize, 0, alloc_hint,
 890                                            &ins, 1);
 891                 if (ret < 0)
 892                         goto out_unlock;
 893
 894                 em = alloc_extent_map();
 895                 if (!em) {
 896                         ret = -ENOMEM;
 897                         goto out_reserve;
 898                 }
 899                 em->start = start;
 900                 em->orig_start = em->start;
 901                 ram_size = ins.offset;
 902                 em->len = ins.offset;
 903                 em->mod_start = em->start;
 904                 em->mod_len = em->len;
 905
 906                 em->block_start = ins.objectid;
 907                 em->block_len = ins.offset;
 908                 em->orig_block_len = ins.offset;
 909                 em->ram_bytes = ram_size;
 910                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 911                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 912                 em->generation = -1;
 913
 914                 while (1) {
 915                         write_lock(&em_tree->lock);
 916                         ret = add_extent_mapping(em_tree, em, 1);
 917                         write_unlock(&em_tree->lock);
 918                         if (ret != -EEXIST) {
 919                                 free_extent_map(em);
 920                                 break;
 921                         }
 922                         btrfs_drop_extent_cache(inode, start,
 923                                                 start + ram_size - 1, 0);
 924                 }
 925                 if (ret)
 926                         goto out_reserve;
 927
 928                 cur_alloc_size = ins.offset;
 929                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 930                                                ram_size, cur_alloc_size, 0);
 931                 if (ret)
 932                         goto out_reserve;
 933
 934                 if (root->root_key.objectid ==
 935                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 936                         ret = btrfs_reloc_clone_csums(inode, start,
 937                                                       cur_alloc_size);
 938                         if (ret)
 939                                 goto out_reserve;
 940                 }
 941
 942                 if (disk_num_bytes < cur_alloc_size)
 943                         break;
 944
 945                 /* we're not doing compressed IO, don't unlock the first
 946                  * page (which the caller expects to stay locked), don't
 947                  * clear any dirty bits and don't set any writeback bits
 948                  *
 949                  * Do set the Private2 bit so we know this page was properly
 950                  * setup for writepage
 951                  */
 952                 op = unlock ? PAGE_UNLOCK : 0;
 953                 op |= PAGE_SET_PRIVATE2;
 954
 955                 extent_clear_unlock_delalloc(inode, start,
 956                                              start + ram_size - 1, locked_page,
 957                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 958                                              op);
 959                 disk_num_bytes -= cur_alloc_size;
 960                 num_bytes -= cur_alloc_size;
 961                 alloc_hint = ins.objectid + ins.offset;
 962                 start += cur_alloc_size;
 963         }
 964 out:
 965         return ret;
 966
 967 out_reserve:
 968         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 969 out_unlock:
 970         extent_clear_unlock_delalloc(inode, start, end, locked_page,
 971                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
 972                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
 973                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 974                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 975         goto out;
 976 }
 977
 978 /*
 979  * work queue call back to started compression on a file and pages
 980  */
 981 static noinline void async_cow_start(struct btrfs_work *work)
 982 {
 983         struct async_cow *async_cow;
 984         int num_added = 0;
 985         async_cow = container_of(work, struct async_cow, work);
 986
 987         compress_file_range(async_cow->inode, async_cow->locked_page,
 988                             async_cow->start, async_cow->end, async_cow,
 989                             &num_added);
 990         if (num_added == 0) {
 991                 btrfs_add_delayed_iput(async_cow->inode);
 992                 async_cow->inode = NULL;
 993         }
 994 }
 995
 996 /*
 997  * work queue call back to submit previously compressed pages
 998  */
 999 static noinline void async_cow_submit(struct btrfs_work *work)
1000 {
1001         struct async_cow *async_cow;
1002         struct btrfs_root *root;
1003         unsigned long nr_pages;
1004
1005         async_cow = container_of(work, struct async_cow, work);
1006
1007         root = async_cow->root;
1008         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1009                 PAGE_CACHE_SHIFT;
1010
1011         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1012             5 * 1024 * 1024 &&
1013             waitqueue_active(&root->fs_info->async_submit_wait))
1014                 wake_up(&root->fs_info->async_submit_wait);
1015
1016         if (async_cow->inode)
1017                 submit_compressed_extents(async_cow->inode, async_cow);
1018 }
1019
1020 static noinline void async_cow_free(struct btrfs_work *work)
1021 {
1022         struct async_cow *async_cow;
1023         async_cow = container_of(work, struct async_cow, work);
1024         if (async_cow->inode)
1025                 btrfs_add_delayed_iput(async_cow->inode);
1026         kfree(async_cow);
1027 }
1028
1029 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1030                                 u64 start, u64 end, int *page_started,
1031                                 unsigned long *nr_written)
1032 {
1033         struct async_cow *async_cow;
1034         struct btrfs_root *root = BTRFS_I(inode)->root;
1035         unsigned long nr_pages;
1036         u64 cur_end;
1037         int limit = 10 * 1024 * 1024;
1038
1039         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1040                          1, 0, NULL, GFP_NOFS);
1041         while (start < end) {
1042                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1043                 BUG_ON(!async_cow); /* -ENOMEM */
1044                 async_cow->inode = igrab(inode);
1045                 async_cow->root = root;
1046                 async_cow->locked_page = locked_page;
1047                 async_cow->start = start;
1048
1049                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1050                         cur_end = end;
1051                 else
1052                         cur_end = min(end, start + 512 * 1024 - 1);
1053
1054                 async_cow->end = cur_end;
1055                 INIT_LIST_HEAD(&async_cow->extents);
1056
1057                 async_cow->work.func = async_cow_start;
1058                 async_cow->work.ordered_func = async_cow_submit;
1059                 async_cow->work.ordered_free = async_cow_free;
1060                 async_cow->work.flags = 0;
1061
1062                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1063                         PAGE_CACHE_SHIFT;
1064                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1065
1066                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1067                                    &async_cow->work);
1068
1069                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1070                         wait_event(root->fs_info->async_submit_wait,
1071                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1072                             limit));
1073                 }
1074
1075                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1076                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1077                         wait_event(root->fs_info->async_submit_wait,
1078                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1079                            0));
1080                 }
1081
1082                 *nr_written += nr_pages;
1083                 start = cur_end + 1;
1084         }
1085         *page_started = 1;
1086         return 0;
1087 }
1088
1089 static noinline int csum_exist_in_range(struct btrfs_root *root,
1090                                         u64 bytenr, u64 num_bytes)
1091 {
1092         int ret;
1093         struct btrfs_ordered_sum *sums;
1094         LIST_HEAD(list);
1095
1096         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1097                                        bytenr + num_bytes - 1, &list, 0);
1098         if (ret == 0 && list_empty(&list))
1099                 return 0;
1100
1101         while (!list_empty(&list)) {
1102                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1103                 list_del(&sums->list);
1104                 kfree(sums);
1105         }
1106         return 1;
1107 }
1108
1109 /*
1110  * when nowcow writeback call back.  This checks for snapshots or COW copies
1111  * of the extents that exist in the file, and COWs the file as required.
1112  *
1113  * If no cow copies or snapshots exist, we write directly to the existing
1114  * blocks on disk
1115  */
1116 static noinline int run_delalloc_nocow(struct inode *inode,
1117                                        struct page *locked_page,
1118                               u64 start, u64 end, int *page_started, int force,
1119                               unsigned long *nr_written)
1120 {
1121         struct btrfs_root *root = BTRFS_I(inode)->root;
1122         struct btrfs_trans_handle *trans;
1123         struct extent_buffer *leaf;
1124         struct btrfs_path *path;
1125         struct btrfs_file_extent_item *fi;
1126         struct btrfs_key found_key;
1127         u64 cow_start;
1128         u64 cur_offset;
1129         u64 extent_end;
1130         u64 extent_offset;
1131         u64 disk_bytenr;
1132         u64 num_bytes;
1133         u64 disk_num_bytes;
1134         u64 ram_bytes;
1135         int extent_type;
1136         int ret, err;
1137         int type;
1138         int nocow;
1139         int check_prev = 1;
1140         bool nolock;
1141         u64 ino = btrfs_ino(inode);
1142
1143         path = btrfs_alloc_path();
1144         if (!path) {
1145                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1146                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1147                                              EXTENT_DO_ACCOUNTING |
1148                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1149                                              PAGE_CLEAR_DIRTY |
1150                                              PAGE_SET_WRITEBACK |
1151                                              PAGE_END_WRITEBACK);
1152                 return -ENOMEM;
1153         }
1154
1155         nolock = btrfs_is_free_space_inode(inode);
1156
1157         if (nolock)
1158                 trans = btrfs_join_transaction_nolock(root);
1159         else
1160                 trans = btrfs_join_transaction(root);
1161
1162         if (IS_ERR(trans)) {
1163                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1164                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1165                                              EXTENT_DO_ACCOUNTING |
1166                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1167                                              PAGE_CLEAR_DIRTY |
1168                                              PAGE_SET_WRITEBACK |
1169                                              PAGE_END_WRITEBACK);
1170                 btrfs_free_path(path);
1171                 return PTR_ERR(trans);
1172         }
1173
1174         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1175
1176         cow_start = (u64)-1;
1177         cur_offset = start;
1178         while (1) {
1179                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1180                                                cur_offset, 0);
1181                 if (ret < 0) {
1182                         btrfs_abort_transaction(trans, root, ret);
1183                         goto error;
1184                 }
1185                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186                         leaf = path->nodes[0];
1187                         btrfs_item_key_to_cpu(leaf, &found_key,
1188                                               path->slots[0] - 1);
1189                         if (found_key.objectid == ino &&
1190                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1191                                 path->slots[0]--;
1192                 }
1193                 check_prev = 0;
1194 next_slot:
1195                 leaf = path->nodes[0];
1196                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197                         ret = btrfs_next_leaf(root, path);
1198                         if (ret < 0) {
1199                                 btrfs_abort_transaction(trans, root, ret);
1200                                 goto error;
1201                         }
1202                         if (ret > 0)
1203                                 break;
1204                         leaf = path->nodes[0];
1205                 }
1206
1207                 nocow = 0;
1208                 disk_bytenr = 0;
1209                 num_bytes = 0;
1210                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1211
1212                 if (found_key.objectid > ino ||
1213                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1214                     found_key.offset > end)
1215                         break;
1216
1217                 if (found_key.offset > cur_offset) {
1218                         extent_end = found_key.offset;
1219                         extent_type = 0;
1220                         goto out_check;
1221                 }
1222
1223                 fi = btrfs_item_ptr(leaf, path->slots[0],
1224                                     struct btrfs_file_extent_item);
1225                 extent_type = btrfs_file_extent_type(leaf, fi);
1226
1227                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1228                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1229                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1230                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1231                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1232                         extent_end = found_key.offset +
1233                                 btrfs_file_extent_num_bytes(leaf, fi);
1234                         disk_num_bytes =
1235                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1236                         if (extent_end <= start) {
1237                                 path->slots[0]++;
1238                                 goto next_slot;
1239                         }
1240                         if (disk_bytenr == 0)
1241                                 goto out_check;
1242                         if (btrfs_file_extent_compression(leaf, fi) ||
1243                             btrfs_file_extent_encryption(leaf, fi) ||
1244                             btrfs_file_extent_other_encoding(leaf, fi))
1245                                 goto out_check;
1246                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1247                                 goto out_check;
1248                         if (btrfs_extent_readonly(root, disk_bytenr))
1249                                 goto out_check;
1250                         if (btrfs_cross_ref_exist(trans, root, ino,
1251                                                   found_key.offset -
1252                                                   extent_offset, disk_bytenr))
1253                                 goto out_check;
1254                         disk_bytenr += extent_offset;
1255                         disk_bytenr += cur_offset - found_key.offset;
1256                         num_bytes = min(end + 1, extent_end) - cur_offset;
1257                         /*
1258                          * force cow if csum exists in the range.
1259                          * this ensure that csum for a given extent are
1260                          * either valid or do not exist.
1261                          */
1262                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1263                                 goto out_check;
1264                         nocow = 1;
1265                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1266                         extent_end = found_key.offset +
1267                                 btrfs_file_extent_inline_len(leaf, fi);
1268                         extent_end = ALIGN(extent_end, root->sectorsize);
1269                 } else {
1270                         BUG_ON(1);
1271                 }
1272 out_check:
1273                 if (extent_end <= start) {
1274                         path->slots[0]++;
1275                         goto next_slot;
1276                 }
1277                 if (!nocow) {
1278                         if (cow_start == (u64)-1)
1279                                 cow_start = cur_offset;
1280                         cur_offset = extent_end;
1281                         if (cur_offset > end)
1282                                 break;
1283                         path->slots[0]++;
1284                         goto next_slot;
1285                 }
1286
1287                 btrfs_release_path(path);
1288                 if (cow_start != (u64)-1) {
1289                         ret = cow_file_range(inode, locked_page,
1290                                              cow_start, found_key.offset - 1,
1291                                              page_started, nr_written, 1);
1292                         if (ret) {
1293                                 btrfs_abort_transaction(trans, root, ret);
1294                                 goto error;
1295                         }
1296                         cow_start = (u64)-1;
1297                 }
1298
1299                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1300                         struct extent_map *em;
1301                         struct extent_map_tree *em_tree;
1302                         em_tree = &BTRFS_I(inode)->extent_tree;
1303                         em = alloc_extent_map();
1304                         BUG_ON(!em); /* -ENOMEM */
1305                         em->start = cur_offset;
1306                         em->orig_start = found_key.offset - extent_offset;
1307                         em->len = num_bytes;
1308                         em->block_len = num_bytes;
1309                         em->block_start = disk_bytenr;
1310                         em->orig_block_len = disk_num_bytes;
1311                         em->ram_bytes = ram_bytes;
1312                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1313                         em->mod_start = em->start;
1314                         em->mod_len = em->len;
1315                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1316                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1317                         em->generation = -1;
1318                         while (1) {
1319                                 write_lock(&em_tree->lock);
1320                                 ret = add_extent_mapping(em_tree, em, 1);
1321                                 write_unlock(&em_tree->lock);
1322                                 if (ret != -EEXIST) {
1323                                         free_extent_map(em);
1324                                         break;
1325                                 }
1326                                 btrfs_drop_extent_cache(inode, em->start,
1327                                                 em->start + em->len - 1, 0);
1328                         }
1329                         type = BTRFS_ORDERED_PREALLOC;
1330                 } else {
1331                         type = BTRFS_ORDERED_NOCOW;
1332                 }
1333
1334                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1335                                                num_bytes, num_bytes, type);
1336                 BUG_ON(ret); /* -ENOMEM */
1337
1338                 if (root->root_key.objectid ==
1339                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1340                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1341                                                       num_bytes);
1342                         if (ret) {
1343                                 btrfs_abort_transaction(trans, root, ret);
1344                                 goto error;
1345                         }
1346                 }
1347
1348                 extent_clear_unlock_delalloc(inode, cur_offset,
1349                                              cur_offset + num_bytes - 1,
1350                                              locked_page, EXTENT_LOCKED |
1351                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1352                                              PAGE_SET_PRIVATE2);
1353                 cur_offset = extent_end;
1354                 if (cur_offset > end)
1355                         break;
1356         }
1357         btrfs_release_path(path);
1358
1359         if (cur_offset <= end && cow_start == (u64)-1) {
1360                 cow_start = cur_offset;
1361                 cur_offset = end;
1362         }
1363
1364         if (cow_start != (u64)-1) {
1365                 ret = cow_file_range(inode, locked_page, cow_start, end,
1366                                      page_started, nr_written, 1);
1367                 if (ret) {
1368                         btrfs_abort_transaction(trans, root, ret);
1369                         goto error;
1370                 }
1371         }
1372
1373 error:
1374         err = btrfs_end_transaction(trans, root);
1375         if (!ret)
1376                 ret = err;
1377
1378         if (ret && cur_offset < end)
1379                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1380                                              locked_page, EXTENT_LOCKED |
1381                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1382                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1383                                              PAGE_CLEAR_DIRTY |
1384                                              PAGE_SET_WRITEBACK |
1385                                              PAGE_END_WRITEBACK);
1386         btrfs_free_path(path);
1387         return ret;
1388 }
1389
1390 /*
1391  * extent_io.c call back to do delayed allocation processing
1392  */
1393 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1394                               u64 start, u64 end, int *page_started,
1395                               unsigned long *nr_written)
1396 {
1397         int ret;
1398         struct btrfs_root *root = BTRFS_I(inode)->root;
1399
1400         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1401                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1402                                          page_started, 1, nr_written);
1403         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1404                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1405                                          page_started, 0, nr_written);
1406         } else if (!btrfs_test_opt(root, COMPRESS) &&
1407                    !(BTRFS_I(inode)->force_compress) &&
1408                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1409                 ret = cow_file_range(inode, locked_page, start, end,
1410                                       page_started, nr_written, 1);
1411         } else {
1412                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1413                         &BTRFS_I(inode)->runtime_flags);
1414                 ret = cow_file_range_async(inode, locked_page, start, end,
1415                                            page_started, nr_written);
1416         }
1417         return ret;
1418 }
1419
1420 static void btrfs_split_extent_hook(struct inode *inode,
1421                                     struct extent_state *orig, u64 split)
1422 {
1423         /* not delalloc, ignore it */
1424         if (!(orig->state & EXTENT_DELALLOC))
1425                 return;
1426
1427         spin_lock(&BTRFS_I(inode)->lock);
1428         BTRFS_I(inode)->outstanding_extents++;
1429         spin_unlock(&BTRFS_I(inode)->lock);
1430 }
1431
1432 /*
1433  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1434  * extents so we can keep track of new extents that are just merged onto old
1435  * extents, such as when we are doing sequential writes, so we can properly
1436  * account for the metadata space we'll need.
1437  */
1438 static void btrfs_merge_extent_hook(struct inode *inode,
1439                                     struct extent_state *new,
1440                                     struct extent_state *other)
1441 {
1442         /* not delalloc, ignore it */
1443         if (!(other->state & EXTENT_DELALLOC))
1444                 return;
1445
1446         spin_lock(&BTRFS_I(inode)->lock);
1447         BTRFS_I(inode)->outstanding_extents--;
1448         spin_unlock(&BTRFS_I(inode)->lock);
1449 }
1450
1451 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1452                                       struct inode *inode)
1453 {
1454         spin_lock(&root->delalloc_lock);
1455         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1456                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1457                               &root->delalloc_inodes);
1458                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1459                         &BTRFS_I(inode)->runtime_flags);
1460                 root->nr_delalloc_inodes++;
1461                 if (root->nr_delalloc_inodes == 1) {
1462                         spin_lock(&root->fs_info->delalloc_root_lock);
1463                         BUG_ON(!list_empty(&root->delalloc_root));
1464                         list_add_tail(&root->delalloc_root,
1465                                       &root->fs_info->delalloc_roots);
1466                         spin_unlock(&root->fs_info->delalloc_root_lock);
1467                 }
1468         }
1469         spin_unlock(&root->delalloc_lock);
1470 }
1471
1472 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1473                                      struct inode *inode)
1474 {
1475         spin_lock(&root->delalloc_lock);
1476         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1477                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1478                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1479                           &BTRFS_I(inode)->runtime_flags);
1480                 root->nr_delalloc_inodes--;
1481                 if (!root->nr_delalloc_inodes) {
1482                         spin_lock(&root->fs_info->delalloc_root_lock);
1483                         BUG_ON(list_empty(&root->delalloc_root));
1484                         list_del_init(&root->delalloc_root);
1485                         spin_unlock(&root->fs_info->delalloc_root_lock);
1486                 }
1487         }
1488         spin_unlock(&root->delalloc_lock);
1489 }
1490
1491 /*
1492  * extent_io.c set_bit_hook, used to track delayed allocation
1493  * bytes in this file, and to maintain the list of inodes that
1494  * have pending delalloc work to be done.
1495  */
1496 static void btrfs_set_bit_hook(struct inode *inode,
1497                                struct extent_state *state, unsigned long *bits)
1498 {
1499
1500         /*
1501          * set_bit and clear bit hooks normally require _irqsave/restore
1502          * but in this case, we are only testing for the DELALLOC
1503          * bit, which is only set or cleared with irqs on
1504          */
1505         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1506                 struct btrfs_root *root = BTRFS_I(inode)->root;
1507                 u64 len = state->end + 1 - state->start;
1508                 bool do_list = !btrfs_is_free_space_inode(inode);
1509
1510                 if (*bits & EXTENT_FIRST_DELALLOC) {
1511                         *bits &= ~EXTENT_FIRST_DELALLOC;
1512                 } else {
1513                         spin_lock(&BTRFS_I(inode)->lock);
1514                         BTRFS_I(inode)->outstanding_extents++;
1515                         spin_unlock(&BTRFS_I(inode)->lock);
1516                 }
1517
1518                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1519                                      root->fs_info->delalloc_batch);
1520                 spin_lock(&BTRFS_I(inode)->lock);
1521                 BTRFS_I(inode)->delalloc_bytes += len;
1522                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1523                                          &BTRFS_I(inode)->runtime_flags))
1524                         btrfs_add_delalloc_inodes(root, inode);
1525                 spin_unlock(&BTRFS_I(inode)->lock);
1526         }
1527 }
1528
1529 /*
1530  * extent_io.c clear_bit_hook, see set_bit_hook for why
1531  */
1532 static void btrfs_clear_bit_hook(struct inode *inode,
1533                                  struct extent_state *state,
1534                                  unsigned long *bits)
1535 {
1536         /*
1537          * set_bit and clear bit hooks normally require _irqsave/restore
1538          * but in this case, we are only testing for the DELALLOC
1539          * bit, which is only set or cleared with irqs on
1540          */
1541         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1542                 struct btrfs_root *root = BTRFS_I(inode)->root;
1543                 u64 len = state->end + 1 - state->start;
1544                 bool do_list = !btrfs_is_free_space_inode(inode);
1545
1546                 if (*bits & EXTENT_FIRST_DELALLOC) {
1547                         *bits &= ~EXTENT_FIRST_DELALLOC;
1548                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1549                         spin_lock(&BTRFS_I(inode)->lock);
1550                         BTRFS_I(inode)->outstanding_extents--;
1551                         spin_unlock(&BTRFS_I(inode)->lock);
1552                 }
1553
1554                 /*
1555                  * We don't reserve metadata space for space cache inodes so we
1556                  * don't need to call dellalloc_release_metadata if there is an
1557                  * error.
1558                  */
1559                 if (*bits & EXTENT_DO_ACCOUNTING &&
1560                     root != root->fs_info->tree_root)
1561                         btrfs_delalloc_release_metadata(inode, len);
1562
1563                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1564                     && do_list && !(state->state & EXTENT_NORESERVE))
1565                         btrfs_free_reserved_data_space(inode, len);
1566
1567                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1568                                      root->fs_info->delalloc_batch);
1569                 spin_lock(&BTRFS_I(inode)->lock);
1570                 BTRFS_I(inode)->delalloc_bytes -= len;
1571                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1572                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1573                              &BTRFS_I(inode)->runtime_flags))
1574                         btrfs_del_delalloc_inode(root, inode);
1575                 spin_unlock(&BTRFS_I(inode)->lock);
1576         }
1577 }
1578
1579 /*
1580  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1581  * we don't create bios that span stripes or chunks
1582  */
1583 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1584                          size_t size, struct bio *bio,
1585                          unsigned long bio_flags)
1586 {
1587         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1588         u64 logical = (u64)bio->bi_sector << 9;
1589         u64 length = 0;
1590         u64 map_length;
1591         int ret;
1592
1593         if (bio_flags & EXTENT_BIO_COMPRESSED)
1594                 return 0;
1595
1596         length = bio->bi_size;
1597         map_length = length;
1598         ret = btrfs_map_block(root->fs_info, rw, logical,
1599                               &map_length, NULL, 0);
1600         /* Will always return 0 with map_multi == NULL */
1601         BUG_ON(ret < 0);
1602         if (map_length < length + size)
1603                 return 1;
1604         return 0;
1605 }
1606
1607 /*
1608  * in order to insert checksums into the metadata in large chunks,
1609  * we wait until bio submission time.   All the pages in the bio are
1610  * checksummed and sums are attached onto the ordered extent record.
1611  *
1612  * At IO completion time the cums attached on the ordered extent record
1613  * are inserted into the btree
1614  */
1615 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1616                                     struct bio *bio, int mirror_num,
1617                                     unsigned long bio_flags,
1618                                     u64 bio_offset)
1619 {
1620         struct btrfs_root *root = BTRFS_I(inode)->root;
1621         int ret = 0;
1622
1623         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1624         BUG_ON(ret); /* -ENOMEM */
1625         return 0;
1626 }
1627
1628 /*
1629  * in order to insert checksums into the metadata in large chunks,
1630  * we wait until bio submission time.   All the pages in the bio are
1631  * checksummed and sums are attached onto the ordered extent record.
1632  *
1633  * At IO completion time the cums attached on the ordered extent record
1634  * are inserted into the btree
1635  */
1636 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1637                           int mirror_num, unsigned long bio_flags,
1638                           u64 bio_offset)
1639 {
1640         struct btrfs_root *root = BTRFS_I(inode)->root;
1641         int ret;
1642
1643         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1644         if (ret)
1645                 bio_endio(bio, ret);
1646         return ret;
1647 }
1648
1649 /*
1650  * extent_io.c submission hook. This does the right thing for csum calculation
1651  * on write, or reading the csums from the tree before a read
1652  */
1653 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1654                           int mirror_num, unsigned long bio_flags,
1655                           u64 bio_offset)
1656 {
1657         struct btrfs_root *root = BTRFS_I(inode)->root;
1658         int ret = 0;
1659         int skip_sum;
1660         int metadata = 0;
1661         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1662
1663         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1664
1665         if (btrfs_is_free_space_inode(inode))
1666                 metadata = 2;
1667
1668         if (!(rw & REQ_WRITE)) {
1669                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1670                 if (ret)
1671                         goto out;
1672
1673                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1674                         ret = btrfs_submit_compressed_read(inode, bio,
1675                                                            mirror_num,
1676                                                            bio_flags);
1677                         goto out;
1678                 } else if (!skip_sum) {
1679                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1680                         if (ret)
1681                                 goto out;
1682                 }
1683                 goto mapit;
1684         } else if (async && !skip_sum) {
1685                 /* csum items have already been cloned */
1686                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1687                         goto mapit;
1688                 /* we're doing a write, do the async checksumming */
1689                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1690                                    inode, rw, bio, mirror_num,
1691                                    bio_flags, bio_offset,
1692                                    __btrfs_submit_bio_start,
1693                                    __btrfs_submit_bio_done);
1694                 goto out;
1695         } else if (!skip_sum) {
1696                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1697                 if (ret)
1698                         goto out;
1699         }
1700
1701 mapit:
1702         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1703
1704 out:
1705         if (ret < 0)
1706                 bio_endio(bio, ret);
1707         return ret;
1708 }
1709
1710 /*
1711  * given a list of ordered sums record them in the inode.  This happens
1712  * at IO completion time based on sums calculated at bio submission time.
1713  */
1714 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1715                              struct inode *inode, u64 file_offset,
1716                              struct list_head *list)
1717 {
1718         struct btrfs_ordered_sum *sum;
1719
1720         list_for_each_entry(sum, list, list) {
1721                 trans->adding_csums = 1;
1722                 btrfs_csum_file_blocks(trans,
1723                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1724                 trans->adding_csums = 0;
1725         }
1726         return 0;
1727 }
1728
1729 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1730                               struct extent_state **cached_state)
1731 {
1732         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1733         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1734                                    cached_state, GFP_NOFS);
1735 }
1736
1737 /* see btrfs_writepage_start_hook for details on why this is required */
1738 struct btrfs_writepage_fixup {
1739         struct page *page;
1740         struct btrfs_work work;
1741 };
1742
1743 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1744 {
1745         struct btrfs_writepage_fixup *fixup;
1746         struct btrfs_ordered_extent *ordered;
1747         struct extent_state *cached_state = NULL;
1748         struct page *page;
1749         struct inode *inode;
1750         u64 page_start;
1751         u64 page_end;
1752         int ret;
1753
1754         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1755         page = fixup->page;
1756 again:
1757         lock_page(page);
1758         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1759                 ClearPageChecked(page);
1760                 goto out_page;
1761         }
1762
1763         inode = page->mapping->host;
1764         page_start = page_offset(page);
1765         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1766
1767         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1768                          &cached_state);
1769
1770         /* already ordered? We're done */
1771         if (PagePrivate2(page))
1772                 goto out;
1773
1774         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1775         if (ordered) {
1776                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1777                                      page_end, &cached_state, GFP_NOFS);
1778                 unlock_page(page);
1779                 btrfs_start_ordered_extent(inode, ordered, 1);
1780                 btrfs_put_ordered_extent(ordered);
1781                 goto again;
1782         }
1783
1784         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1785         if (ret) {
1786                 mapping_set_error(page->mapping, ret);
1787                 end_extent_writepage(page, ret, page_start, page_end);
1788                 ClearPageChecked(page);
1789                 goto out;
1790          }
1791
1792         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1793         ClearPageChecked(page);
1794         set_page_dirty(page);
1795 out:
1796         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1797                              &cached_state, GFP_NOFS);
1798 out_page:
1799         unlock_page(page);
1800         page_cache_release(page);
1801         kfree(fixup);
1802 }
1803
1804 /*
1805  * There are a few paths in the higher layers of the kernel that directly
1806  * set the page dirty bit without asking the filesystem if it is a
1807  * good idea.  This causes problems because we want to make sure COW
1808  * properly happens and the data=ordered rules are followed.
1809  *
1810  * In our case any range that doesn't have the ORDERED bit set
1811  * hasn't been properly setup for IO.  We kick off an async process
1812  * to fix it up.  The async helper will wait for ordered extents, set
1813  * the delalloc bit and make it safe to write the page.
1814  */
1815 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1816 {
1817         struct inode *inode = page->mapping->host;
1818         struct btrfs_writepage_fixup *fixup;
1819         struct btrfs_root *root = BTRFS_I(inode)->root;
1820
1821         /* this page is properly in the ordered list */
1822         if (TestClearPagePrivate2(page))
1823                 return 0;
1824
1825         if (PageChecked(page))
1826                 return -EAGAIN;
1827
1828         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1829         if (!fixup)
1830                 return -EAGAIN;
1831
1832         SetPageChecked(page);
1833         page_cache_get(page);
1834         fixup->work.func = btrfs_writepage_fixup_worker;
1835         fixup->page = page;
1836         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1837         return -EBUSY;
1838 }
1839
1840 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1841                                        struct inode *inode, u64 file_pos,
1842                                        u64 disk_bytenr, u64 disk_num_bytes,
1843                                        u64 num_bytes, u64 ram_bytes,
1844                                        u8 compression, u8 encryption,
1845                                        u16 other_encoding, int extent_type)
1846 {
1847         struct btrfs_root *root = BTRFS_I(inode)->root;
1848         struct btrfs_file_extent_item *fi;
1849         struct btrfs_path *path;
1850         struct extent_buffer *leaf;
1851         struct btrfs_key ins;
1852         int ret;
1853
1854         path = btrfs_alloc_path();
1855         if (!path)
1856                 return -ENOMEM;
1857
1858         path->leave_spinning = 1;
1859
1860         /*
1861          * we may be replacing one extent in the tree with another.
1862          * The new extent is pinned in the extent map, and we don't want
1863          * to drop it from the cache until it is completely in the btree.
1864          *
1865          * So, tell btrfs_drop_extents to leave this extent in the cache.
1866          * the caller is expected to unpin it and allow it to be merged
1867          * with the others.
1868          */
1869         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1870                                  file_pos + num_bytes, 0);
1871         if (ret)
1872                 goto out;
1873
1874         ins.objectid = btrfs_ino(inode);
1875         ins.offset = file_pos;
1876         ins.type = BTRFS_EXTENT_DATA_KEY;
1877         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1878         if (ret)
1879                 goto out;
1880         leaf = path->nodes[0];
1881         fi = btrfs_item_ptr(leaf, path->slots[0],
1882                             struct btrfs_file_extent_item);
1883         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1884         btrfs_set_file_extent_type(leaf, fi, extent_type);
1885         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1886         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1887         btrfs_set_file_extent_offset(leaf, fi, 0);
1888         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1889         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1890         btrfs_set_file_extent_compression(leaf, fi, compression);
1891         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1892         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1893
1894         btrfs_mark_buffer_dirty(leaf);
1895         btrfs_release_path(path);
1896
1897         inode_add_bytes(inode, num_bytes);
1898
1899         ins.objectid = disk_bytenr;
1900         ins.offset = disk_num_bytes;
1901         ins.type = BTRFS_EXTENT_ITEM_KEY;
1902         ret = btrfs_alloc_reserved_file_extent(trans, root,
1903                                         root->root_key.objectid,
1904                                         btrfs_ino(inode), file_pos, &ins);
1905 out:
1906         btrfs_free_path(path);
1907
1908         return ret;
1909 }
1910
1911 /* snapshot-aware defrag */
1912 struct sa_defrag_extent_backref {
1913         struct rb_node node;
1914         struct old_sa_defrag_extent *old;
1915         u64 root_id;
1916         u64 inum;
1917         u64 file_pos;
1918         u64 extent_offset;
1919         u64 num_bytes;
1920         u64 generation;
1921 };
1922
1923 struct old_sa_defrag_extent {
1924         struct list_head list;
1925         struct new_sa_defrag_extent *new;
1926
1927         u64 extent_offset;
1928         u64 bytenr;
1929         u64 offset;
1930         u64 len;
1931         int count;
1932 };
1933
1934 struct new_sa_defrag_extent {
1935         struct rb_root root;
1936         struct list_head head;
1937         struct btrfs_path *path;
1938         struct inode *inode;
1939         u64 file_pos;
1940         u64 len;
1941         u64 bytenr;
1942         u64 disk_len;
1943         u8 compress_type;
1944 };
1945
1946 static int backref_comp(struct sa_defrag_extent_backref *b1,
1947                         struct sa_defrag_extent_backref *b2)
1948 {
1949         if (b1->root_id < b2->root_id)
1950                 return -1;
1951         else if (b1->root_id > b2->root_id)
1952                 return 1;
1953
1954         if (b1->inum < b2->inum)
1955                 return -1;
1956         else if (b1->inum > b2->inum)
1957                 return 1;
1958
1959         if (b1->file_pos < b2->file_pos)
1960                 return -1;
1961         else if (b1->file_pos > b2->file_pos)
1962                 return 1;
1963
1964         /*
1965          * [------------------------------] ===> (a range of space)
1966          *     |<--->|   |<---->| =============> (fs/file tree A)
1967          * |<---------------------------->| ===> (fs/file tree B)
1968          *
1969          * A range of space can refer to two file extents in one tree while
1970          * refer to only one file extent in another tree.
1971          *
1972          * So we may process a disk offset more than one time(two extents in A)
1973          * and locate at the same extent(one extent in B), then insert two same
1974          * backrefs(both refer to the extent in B).
1975          */
1976         return 0;
1977 }
1978
1979 static void backref_insert(struct rb_root *root,
1980                            struct sa_defrag_extent_backref *backref)
1981 {
1982         struct rb_node **p = &root->rb_node;
1983         struct rb_node *parent = NULL;
1984         struct sa_defrag_extent_backref *entry;
1985         int ret;
1986
1987         while (*p) {
1988                 parent = *p;
1989                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1990
1991                 ret = backref_comp(backref, entry);
1992                 if (ret < 0)
1993                         p = &(*p)->rb_left;
1994                 else
1995                         p = &(*p)->rb_right;
1996         }
1997
1998         rb_link_node(&backref->node, parent, p);
1999         rb_insert_color(&backref->node, root);
2000 }
2001
2002 /*
2003  * Note the backref might has changed, and in this case we just return 0.
2004  */
2005 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
2006                                        void *ctx)
2007 {
2008         struct btrfs_file_extent_item *extent;
2009         struct btrfs_fs_info *fs_info;
2010         struct old_sa_defrag_extent *old = ctx;
2011         struct new_sa_defrag_extent *new = old->new;
2012         struct btrfs_path *path = new->path;
2013         struct btrfs_key key;
2014         struct btrfs_root *root;
2015         struct sa_defrag_extent_backref *backref;
2016         struct extent_buffer *leaf;
2017         struct inode *inode = new->inode;
2018         int slot;
2019         int ret;
2020         u64 extent_offset;
2021         u64 num_bytes;
2022
2023         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2024             inum == btrfs_ino(inode))
2025                 return 0;
2026
2027         key.objectid = root_id;
2028         key.type = BTRFS_ROOT_ITEM_KEY;
2029         key.offset = (u64)-1;
2030
2031         fs_info = BTRFS_I(inode)->root->fs_info;
2032         root = btrfs_read_fs_root_no_name(fs_info, &key);
2033         if (IS_ERR(root)) {
2034                 if (PTR_ERR(root) == -ENOENT)
2035                         return 0;
2036                 WARN_ON(1);
2037                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2038                          inum, offset, root_id);
2039                 return PTR_ERR(root);
2040         }
2041
2042         key.objectid = inum;
2043         key.type = BTRFS_EXTENT_DATA_KEY;
2044         if (offset > (u64)-1 << 32)
2045                 key.offset = 0;
2046         else
2047                 key.offset = offset;
2048
2049         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2050         if (ret < 0) {
2051                 WARN_ON(1);
2052                 return ret;
2053         }
2054         ret = 0;
2055
2056         while (1) {
2057                 cond_resched();
2058
2059                 leaf = path->nodes[0];
2060                 slot = path->slots[0];
2061
2062                 if (slot >= btrfs_header_nritems(leaf)) {
2063                         ret = btrfs_next_leaf(root, path);
2064                         if (ret < 0) {
2065                                 goto out;
2066                         } else if (ret > 0) {
2067                                 ret = 0;
2068                                 goto out;
2069                         }
2070                         continue;
2071                 }
2072
2073                 path->slots[0]++;
2074
2075                 btrfs_item_key_to_cpu(leaf, &key, slot);
2076
2077                 if (key.objectid > inum)
2078                         goto out;
2079
2080                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2081                         continue;
2082
2083                 extent = btrfs_item_ptr(leaf, slot,
2084                                         struct btrfs_file_extent_item);
2085
2086                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2087                         continue;
2088
2089                 /*
2090                  * 'offset' refers to the exact key.offset,
2091                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2092                  * (key.offset - extent_offset).
2093                  */
2094                 if (key.offset != offset)
2095                         continue;
2096
2097                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2098                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2099
2100                 if (extent_offset >= old->extent_offset + old->offset +
2101                     old->len || extent_offset + num_bytes <=
2102                     old->extent_offset + old->offset)
2103                         continue;
2104                 break;
2105         }
2106
2107         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2108         if (!backref) {
2109                 ret = -ENOENT;
2110                 goto out;
2111         }
2112
2113         backref->root_id = root_id;
2114         backref->inum = inum;
2115         backref->file_pos = offset;
2116         backref->num_bytes = num_bytes;
2117         backref->extent_offset = extent_offset;
2118         backref->generation = btrfs_file_extent_generation(leaf, extent);
2119         backref->old = old;
2120         backref_insert(&new->root, backref);
2121         old->count++;
2122 out:
2123         btrfs_release_path(path);
2124         WARN_ON(ret);
2125         return ret;
2126 }
2127
2128 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2129                                    struct new_sa_defrag_extent *new)
2130 {
2131         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2132         struct old_sa_defrag_extent *old, *tmp;
2133         int ret;
2134
2135         new->path = path;
2136
2137         list_for_each_entry_safe(old, tmp, &new->head, list) {
2138                 ret = iterate_inodes_from_logical(old->bytenr +
2139                                                   old->extent_offset, fs_info,
2140                                                   path, record_one_backref,
2141                                                   old);
2142                 BUG_ON(ret < 0 && ret != -ENOENT);
2143
2144                 /* no backref to be processed for this extent */
2145                 if (!old->count) {
2146                         list_del(&old->list);
2147                         kfree(old);
2148                 }
2149         }
2150
2151         if (list_empty(&new->head))
2152                 return false;
2153
2154         return true;
2155 }
2156
2157 static int relink_is_mergable(struct extent_buffer *leaf,
2158                               struct btrfs_file_extent_item *fi,
2159                               struct new_sa_defrag_extent *new)
2160 {
2161         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2162                 return 0;
2163
2164         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2165                 return 0;
2166
2167         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2168                 return 0;
2169
2170         if (btrfs_file_extent_encryption(leaf, fi) ||
2171             btrfs_file_extent_other_encoding(leaf, fi))
2172                 return 0;
2173
2174         return 1;
2175 }
2176
2177 /*
2178  * Note the backref might has changed, and in this case we just return 0.
2179  */
2180 static noinline int relink_extent_backref(struct btrfs_path *path,
2181                                  struct sa_defrag_extent_backref *prev,
2182                                  struct sa_defrag_extent_backref *backref)
2183 {
2184         struct btrfs_file_extent_item *extent;
2185         struct btrfs_file_extent_item *item;
2186         struct btrfs_ordered_extent *ordered;
2187         struct btrfs_trans_handle *trans;
2188         struct btrfs_fs_info *fs_info;
2189         struct btrfs_root *root;
2190         struct btrfs_key key;
2191         struct extent_buffer *leaf;
2192         struct old_sa_defrag_extent *old = backref->old;
2193         struct new_sa_defrag_extent *new = old->new;
2194         struct inode *src_inode = new->inode;
2195         struct inode *inode;
2196         struct extent_state *cached = NULL;
2197         int ret = 0;
2198         u64 start;
2199         u64 len;
2200         u64 lock_start;
2201         u64 lock_end;
2202         bool merge = false;
2203         int index;
2204
2205         if (prev && prev->root_id == backref->root_id &&
2206             prev->inum == backref->inum &&
2207             prev->file_pos + prev->num_bytes == backref->file_pos)
2208                 merge = true;
2209
2210         /* step 1: get root */
2211         key.objectid = backref->root_id;
2212         key.type = BTRFS_ROOT_ITEM_KEY;
2213         key.offset = (u64)-1;
2214
2215         fs_info = BTRFS_I(src_inode)->root->fs_info;
2216         index = srcu_read_lock(&fs_info->subvol_srcu);
2217
2218         root = btrfs_read_fs_root_no_name(fs_info, &key);
2219         if (IS_ERR(root)) {
2220                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2221                 if (PTR_ERR(root) == -ENOENT)
2222                         return 0;
2223                 return PTR_ERR(root);
2224         }
2225
2226         /* step 2: get inode */
2227         key.objectid = backref->inum;
2228         key.type = BTRFS_INODE_ITEM_KEY;
2229         key.offset = 0;
2230
2231         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2232         if (IS_ERR(inode)) {
2233                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2234                 return 0;
2235         }
2236
2237         srcu_read_unlock(&fs_info->subvol_srcu, index);
2238
2239         /* step 3: relink backref */
2240         lock_start = backref->file_pos;
2241         lock_end = backref->file_pos + backref->num_bytes - 1;
2242         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2243                          0, &cached);
2244
2245         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2246         if (ordered) {
2247                 btrfs_put_ordered_extent(ordered);
2248                 goto out_unlock;
2249         }
2250
2251         trans = btrfs_join_transaction(root);
2252         if (IS_ERR(trans)) {
2253                 ret = PTR_ERR(trans);
2254                 goto out_unlock;
2255         }
2256
2257         key.objectid = backref->inum;
2258         key.type = BTRFS_EXTENT_DATA_KEY;
2259         key.offset = backref->file_pos;
2260
2261         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2262         if (ret < 0) {
2263                 goto out_free_path;
2264         } else if (ret > 0) {
2265                 ret = 0;
2266                 goto out_free_path;
2267         }
2268
2269         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2270                                 struct btrfs_file_extent_item);
2271
2272         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2273             backref->generation)
2274                 goto out_free_path;
2275
2276         btrfs_release_path(path);
2277
2278         start = backref->file_pos;
2279         if (backref->extent_offset < old->extent_offset + old->offset)
2280                 start += old->extent_offset + old->offset -
2281                          backref->extent_offset;
2282
2283         len = min(backref->extent_offset + backref->num_bytes,
2284                   old->extent_offset + old->offset + old->len);
2285         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2286
2287         ret = btrfs_drop_extents(trans, root, inode, start,
2288                                  start + len, 1);
2289         if (ret)
2290                 goto out_free_path;
2291 again:
2292         key.objectid = btrfs_ino(inode);
2293         key.type = BTRFS_EXTENT_DATA_KEY;
2294         key.offset = start;
2295
2296         path->leave_spinning = 1;
2297         if (merge) {
2298                 struct btrfs_file_extent_item *fi;
2299                 u64 extent_len;
2300                 struct btrfs_key found_key;
2301
2302                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2303                 if (ret < 0)
2304                         goto out_free_path;
2305
2306                 path->slots[0]--;
2307                 leaf = path->nodes[0];
2308                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2309
2310                 fi = btrfs_item_ptr(leaf, path->slots[0],
2311                                     struct btrfs_file_extent_item);
2312                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2313
2314                 if (extent_len + found_key.offset == start &&
2315                     relink_is_mergable(leaf, fi, new)) {
2316                         btrfs_set_file_extent_num_bytes(leaf, fi,
2317                                                         extent_len + len);
2318                         btrfs_mark_buffer_dirty(leaf);
2319                         inode_add_bytes(inode, len);
2320
2321                         ret = 1;
2322                         goto out_free_path;
2323                 } else {
2324                         merge = false;
2325                         btrfs_release_path(path);
2326                         goto again;
2327                 }
2328         }
2329
2330         ret = btrfs_insert_empty_item(trans, root, path, &key,
2331                                         sizeof(*extent));
2332         if (ret) {
2333                 btrfs_abort_transaction(trans, root, ret);
2334                 goto out_free_path;
2335         }
2336
2337         leaf = path->nodes[0];
2338         item = btrfs_item_ptr(leaf, path->slots[0],
2339                                 struct btrfs_file_extent_item);
2340         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2341         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2342         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2343         btrfs_set_file_extent_num_bytes(leaf, item, len);
2344         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2345         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2346         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2347         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2348         btrfs_set_file_extent_encryption(leaf, item, 0);
2349         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2350
2351         btrfs_mark_buffer_dirty(leaf);
2352         inode_add_bytes(inode, len);
2353         btrfs_release_path(path);
2354
2355         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2356                         new->disk_len, 0,
2357                         backref->root_id, backref->inum,
2358                         new->file_pos, 0);      /* start - extent_offset */
2359         if (ret) {
2360                 btrfs_abort_transaction(trans, root, ret);
2361                 goto out_free_path;
2362         }
2363
2364         ret = 1;
2365 out_free_path:
2366         btrfs_release_path(path);
2367         path->leave_spinning = 0;
2368         btrfs_end_transaction(trans, root);
2369 out_unlock:
2370         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2371                              &cached, GFP_NOFS);
2372         iput(inode);
2373         return ret;
2374 }
2375
2376 static void relink_file_extents(struct new_sa_defrag_extent *new)
2377 {
2378         struct btrfs_path *path;
2379         struct old_sa_defrag_extent *old, *tmp;
2380         struct sa_defrag_extent_backref *backref;
2381         struct sa_defrag_extent_backref *prev = NULL;
2382         struct inode *inode;
2383         struct btrfs_root *root;
2384         struct rb_node *node;
2385         int ret;
2386
2387         inode = new->inode;
2388         root = BTRFS_I(inode)->root;
2389
2390         path = btrfs_alloc_path();
2391         if (!path)
2392                 return;
2393
2394         if (!record_extent_backrefs(path, new)) {
2395                 btrfs_free_path(path);
2396                 goto out;
2397         }
2398         btrfs_release_path(path);
2399
2400         while (1) {
2401                 node = rb_first(&new->root);
2402                 if (!node)
2403                         break;
2404                 rb_erase(node, &new->root);
2405
2406                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2407
2408                 ret = relink_extent_backref(path, prev, backref);
2409                 WARN_ON(ret < 0);
2410
2411                 kfree(prev);
2412
2413                 if (ret == 1)
2414                         prev = backref;
2415                 else
2416                         prev = NULL;
2417                 cond_resched();
2418         }
2419         kfree(prev);
2420
2421         btrfs_free_path(path);
2422
2423         list_for_each_entry_safe(old, tmp, &new->head, list) {
2424                 list_del(&old->list);
2425                 kfree(old);
2426         }
2427 out:
2428         atomic_dec(&root->fs_info->defrag_running);
2429         wake_up(&root->fs_info->transaction_wait);
2430
2431         kfree(new);
2432 }
2433
2434 static struct new_sa_defrag_extent *
2435 record_old_file_extents(struct inode *inode,
2436                         struct btrfs_ordered_extent *ordered)
2437 {
2438         struct btrfs_root *root = BTRFS_I(inode)->root;
2439         struct btrfs_path *path;
2440         struct btrfs_key key;
2441         struct old_sa_defrag_extent *old, *tmp;
2442         struct new_sa_defrag_extent *new;
2443         int ret;
2444
2445         new = kmalloc(sizeof(*new), GFP_NOFS);
2446         if (!new)
2447                 return NULL;
2448
2449         new->inode = inode;
2450         new->file_pos = ordered->file_offset;
2451         new->len = ordered->len;
2452         new->bytenr = ordered->start;
2453         new->disk_len = ordered->disk_len;
2454         new->compress_type = ordered->compress_type;
2455         new->root = RB_ROOT;
2456         INIT_LIST_HEAD(&new->head);
2457
2458         path = btrfs_alloc_path();
2459         if (!path)
2460                 goto out_kfree;
2461
2462         key.objectid = btrfs_ino(inode);
2463         key.type = BTRFS_EXTENT_DATA_KEY;
2464         key.offset = new->file_pos;
2465
2466         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2467         if (ret < 0)
2468                 goto out_free_path;
2469         if (ret > 0 && path->slots[0] > 0)
2470                 path->slots[0]--;
2471
2472         /* find out all the old extents for the file range */
2473         while (1) {
2474                 struct btrfs_file_extent_item *extent;
2475                 struct extent_buffer *l;
2476                 int slot;
2477                 u64 num_bytes;
2478                 u64 offset;
2479                 u64 end;
2480                 u64 disk_bytenr;
2481                 u64 extent_offset;
2482
2483                 l = path->nodes[0];
2484                 slot = path->slots[0];
2485
2486                 if (slot >= btrfs_header_nritems(l)) {
2487                         ret = btrfs_next_leaf(root, path);
2488                         if (ret < 0)
2489                                 goto out_free_list;
2490                         else if (ret > 0)
2491                                 break;
2492                         continue;
2493                 }
2494
2495                 btrfs_item_key_to_cpu(l, &key, slot);
2496
2497                 if (key.objectid != btrfs_ino(inode))
2498                         break;
2499                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2500                         break;
2501                 if (key.offset >= new->file_pos + new->len)
2502                         break;
2503
2504                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2505
2506                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2507                 if (key.offset + num_bytes < new->file_pos)
2508                         goto next;
2509
2510                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2511                 if (!disk_bytenr)
2512                         goto next;
2513
2514                 extent_offset = btrfs_file_extent_offset(l, extent);
2515
2516                 old = kmalloc(sizeof(*old), GFP_NOFS);
2517                 if (!old)
2518                         goto out_free_list;
2519
2520                 offset = max(new->file_pos, key.offset);
2521                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2522
2523                 old->bytenr = disk_bytenr;
2524                 old->extent_offset = extent_offset;
2525                 old->offset = offset - key.offset;
2526                 old->len = end - offset;
2527                 old->new = new;
2528                 old->count = 0;
2529                 list_add_tail(&old->list, &new->head);
2530 next:
2531                 path->slots[0]++;
2532                 cond_resched();
2533         }
2534
2535         btrfs_free_path(path);
2536         atomic_inc(&root->fs_info->defrag_running);
2537
2538         return new;
2539
2540 out_free_list:
2541         list_for_each_entry_safe(old, tmp, &new->head, list) {
2542                 list_del(&old->list);
2543                 kfree(old);
2544         }
2545 out_free_path:
2546         btrfs_free_path(path);
2547 out_kfree:
2548         kfree(new);
2549         return NULL;
2550 }
2551
2552 /*
2553  * helper function for btrfs_finish_ordered_io, this
2554  * just reads in some of the csum leaves to prime them into ram
2555  * before we start the transaction.  It limits the amount of btree
2556  * reads required while inside the transaction.
2557  */
2558 /* as ordered data IO finishes, this gets called so we can finish
2559  * an ordered extent if the range of bytes in the file it covers are
2560  * fully written.
2561  */
2562 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2563 {
2564         struct inode *inode = ordered_extent->inode;
2565         struct btrfs_root *root = BTRFS_I(inode)->root;
2566         struct btrfs_trans_handle *trans = NULL;
2567         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2568         struct extent_state *cached_state = NULL;
2569         struct new_sa_defrag_extent *new = NULL;
2570         int compress_type = 0;
2571         int ret = 0;
2572         u64 logical_len = ordered_extent->len;
2573         bool nolock;
2574         bool truncated = false;
2575
2576         nolock = btrfs_is_free_space_inode(inode);
2577
2578         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2579                 ret = -EIO;
2580                 goto out;
2581         }
2582
2583         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2584                 truncated = true;
2585                 logical_len = ordered_extent->truncated_len;
2586                 /* Truncated the entire extent, don't bother adding */
2587                 if (!logical_len)
2588                         goto out;
2589         }
2590
2591         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2592                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2593                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2594                 if (nolock)
2595                         trans = btrfs_join_transaction_nolock(root);
2596                 else
2597                         trans = btrfs_join_transaction(root);
2598                 if (IS_ERR(trans)) {
2599                         ret = PTR_ERR(trans);
2600                         trans = NULL;
2601                         goto out;
2602                 }
2603                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2604                 ret = btrfs_update_inode_fallback(trans, root, inode);
2605                 if (ret) /* -ENOMEM or corruption */
2606                         btrfs_abort_transaction(trans, root, ret);
2607                 goto out;
2608         }
2609
2610         lock_extent_bits(io_tree, ordered_extent->file_offset,
2611                          ordered_extent->file_offset + ordered_extent->len - 1,
2612                          0, &cached_state);
2613
2614         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2615                         ordered_extent->file_offset + ordered_extent->len - 1,
2616                         EXTENT_DEFRAG, 1, cached_state);
2617         if (ret) {
2618                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2619                 if (last_snapshot >= BTRFS_I(inode)->generation)
2620                         /* the inode is shared */
2621                         new = record_old_file_extents(inode, ordered_extent);
2622
2623                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2624                         ordered_extent->file_offset + ordered_extent->len - 1,
2625                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2626         }
2627
2628         if (nolock)
2629                 trans = btrfs_join_transaction_nolock(root);
2630         else
2631                 trans = btrfs_join_transaction(root);
2632         if (IS_ERR(trans)) {
2633                 ret = PTR_ERR(trans);
2634                 trans = NULL;
2635                 goto out_unlock;
2636         }
2637         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2638
2639         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2640                 compress_type = ordered_extent->compress_type;
2641         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2642                 BUG_ON(compress_type);
2643                 ret = btrfs_mark_extent_written(trans, inode,
2644                                                 ordered_extent->file_offset,
2645                                                 ordered_extent->file_offset +
2646                                                 logical_len);
2647         } else {
2648                 BUG_ON(root == root->fs_info->tree_root);
2649                 ret = insert_reserved_file_extent(trans, inode,
2650                                                 ordered_extent->file_offset,
2651                                                 ordered_extent->start,
2652                                                 ordered_extent->disk_len,
2653                                                 logical_len, logical_len,
2654                                                 compress_type, 0, 0,
2655                                                 BTRFS_FILE_EXTENT_REG);
2656         }
2657         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2658                            ordered_extent->file_offset, ordered_extent->len,
2659                            trans->transid);
2660         if (ret < 0) {
2661                 btrfs_abort_transaction(trans, root, ret);
2662                 goto out_unlock;
2663         }
2664
2665         add_pending_csums(trans, inode, ordered_extent->file_offset,
2666                           &ordered_extent->list);
2667
2668         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2669         ret = btrfs_update_inode_fallback(trans, root, inode);
2670         if (ret) { /* -ENOMEM or corruption */
2671                 btrfs_abort_transaction(trans, root, ret);
2672                 goto out_unlock;
2673         }
2674         ret = 0;
2675 out_unlock:
2676         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2677                              ordered_extent->file_offset +
2678                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2679 out:
2680         if (root != root->fs_info->tree_root)
2681                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2682         if (trans)
2683                 btrfs_end_transaction(trans, root);
2684
2685         if (ret || truncated) {
2686                 u64 start, end;
2687
2688                 if (truncated)
2689                         start = ordered_extent->file_offset + logical_len;
2690                 else
2691                         start = ordered_extent->file_offset;
2692                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2693                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2694
2695                 /* Drop the cache for the part of the extent we didn't write. */
2696                 btrfs_drop_extent_cache(inode, start, end, 0);
2697
2698                 /*
2699                  * If the ordered extent had an IOERR or something else went
2700                  * wrong we need to return the space for this ordered extent
2701                  * back to the allocator.  We only free the extent in the
2702                  * truncated case if we didn't write out the extent at all.
2703                  */
2704                 if ((ret || !logical_len) &&
2705                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2706                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2707                         btrfs_free_reserved_extent(root, ordered_extent->start,
2708                                                    ordered_extent->disk_len);
2709         }
2710
2711
2712         /*
2713          * This needs to be done to make sure anybody waiting knows we are done
2714          * updating everything for this ordered extent.
2715          */
2716         btrfs_remove_ordered_extent(inode, ordered_extent);
2717
2718         /* for snapshot-aware defrag */
2719         if (new)
2720                 relink_file_extents(new);
2721
2722         /* once for us */
2723         btrfs_put_ordered_extent(ordered_extent);
2724         /* once for the tree */
2725         btrfs_put_ordered_extent(ordered_extent);
2726
2727         return ret;
2728 }
2729
2730 static void finish_ordered_fn(struct btrfs_work *work)
2731 {
2732         struct btrfs_ordered_extent *ordered_extent;
2733         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2734         btrfs_finish_ordered_io(ordered_extent);
2735 }
2736
2737 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2738                                 struct extent_state *state, int uptodate)
2739 {
2740         struct inode *inode = page->mapping->host;
2741         struct btrfs_root *root = BTRFS_I(inode)->root;
2742         struct btrfs_ordered_extent *ordered_extent = NULL;
2743         struct btrfs_workers *workers;
2744
2745         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2746
2747         ClearPagePrivate2(page);
2748         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2749                                             end - start + 1, uptodate))
2750                 return 0;
2751
2752         ordered_extent->work.func = finish_ordered_fn;
2753         ordered_extent->work.flags = 0;
2754
2755         if (btrfs_is_free_space_inode(inode))
2756                 workers = &root->fs_info->endio_freespace_worker;
2757         else
2758                 workers = &root->fs_info->endio_write_workers;
2759         btrfs_queue_worker(workers, &ordered_extent->work);
2760
2761         return 0;
2762 }
2763
2764 /*
2765  * when reads are done, we need to check csums to verify the data is correct
2766  * if there's a match, we allow the bio to finish.  If not, the code in
2767  * extent_io.c will try to find good copies for us.
2768  */
2769 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2770                                       u64 phy_offset, struct page *page,
2771                                       u64 start, u64 end, int mirror)
2772 {
2773         size_t offset = start - page_offset(page);
2774         struct inode *inode = page->mapping->host;
2775         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2776         char *kaddr;
2777         struct btrfs_root *root = BTRFS_I(inode)->root;
2778         u32 csum_expected;
2779         u32 csum = ~(u32)0;
2780         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2781                                       DEFAULT_RATELIMIT_BURST);
2782
2783         if (PageChecked(page)) {
2784                 ClearPageChecked(page);
2785                 goto good;
2786         }
2787
2788         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2789                 goto good;
2790
2791         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2792             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2793                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2794                                   GFP_NOFS);
2795                 return 0;
2796         }
2797
2798         phy_offset >>= inode->i_sb->s_blocksize_bits;
2799         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2800
2801         kaddr = kmap_atomic(page);
2802         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2803         btrfs_csum_final(csum, (char *)&csum);
2804         if (csum != csum_expected)
2805                 goto zeroit;
2806
2807         kunmap_atomic(kaddr);
2808 good:
2809         return 0;
2810
2811 zeroit:
2812         if (__ratelimit(&_rs))
2813                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2814                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2815         memset(kaddr + offset, 1, end - start + 1);
2816         flush_dcache_page(page);
2817         kunmap_atomic(kaddr);
2818         if (csum_expected == 0)
2819                 return 0;
2820         return -EIO;
2821 }
2822
2823 struct delayed_iput {
2824         struct list_head list;
2825         struct inode *inode;
2826 };
2827
2828 /* JDM: If this is fs-wide, why can't we add a pointer to
2829  * btrfs_inode instead and avoid the allocation? */
2830 void btrfs_add_delayed_iput(struct inode *inode)
2831 {
2832         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2833         struct delayed_iput *delayed;
2834
2835         if (atomic_add_unless(&inode->i_count, -1, 1))
2836                 return;
2837
2838         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2839         delayed->inode = inode;
2840
2841         spin_lock(&fs_info->delayed_iput_lock);
2842         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2843         spin_unlock(&fs_info->delayed_iput_lock);
2844 }
2845
2846 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2847 {
2848         LIST_HEAD(list);
2849         struct btrfs_fs_info *fs_info = root->fs_info;
2850         struct delayed_iput *delayed;
2851         int empty;
2852
2853         spin_lock(&fs_info->delayed_iput_lock);
2854         empty = list_empty(&fs_info->delayed_iputs);
2855         spin_unlock(&fs_info->delayed_iput_lock);
2856         if (empty)
2857                 return;
2858
2859         spin_lock(&fs_info->delayed_iput_lock);
2860         list_splice_init(&fs_info->delayed_iputs, &list);
2861         spin_unlock(&fs_info->delayed_iput_lock);
2862
2863         while (!list_empty(&list)) {
2864                 delayed = list_entry(list.next, struct delayed_iput, list);
2865                 list_del(&delayed->list);
2866                 iput(delayed->inode);
2867                 kfree(delayed);
2868         }
2869 }
2870
2871 /*
2872  * This is called in transaction commit time. If there are no orphan
2873  * files in the subvolume, it removes orphan item and frees block_rsv
2874  * structure.
2875  */
2876 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2877                               struct btrfs_root *root)
2878 {
2879         struct btrfs_block_rsv *block_rsv;
2880         int ret;
2881
2882         if (atomic_read(&root->orphan_inodes) ||
2883             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2884                 return;
2885
2886         spin_lock(&root->orphan_lock);
2887         if (atomic_read(&root->orphan_inodes)) {
2888                 spin_unlock(&root->orphan_lock);
2889                 return;
2890         }
2891
2892         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2893                 spin_unlock(&root->orphan_lock);
2894                 return;
2895         }
2896
2897         block_rsv = root->orphan_block_rsv;
2898         root->orphan_block_rsv = NULL;
2899         spin_unlock(&root->orphan_lock);
2900
2901         if (root->orphan_item_inserted &&
2902             btrfs_root_refs(&root->root_item) > 0) {
2903                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2904                                             root->root_key.objectid);
2905                 if (ret)
2906                         btrfs_abort_transaction(trans, root, ret);
2907                 else
2908                         root->orphan_item_inserted = 0;
2909         }
2910
2911         if (block_rsv) {
2912                 WARN_ON(block_rsv->size > 0);
2913                 btrfs_free_block_rsv(root, block_rsv);
2914         }
2915 }
2916
2917 /*
2918  * This creates an orphan entry for the given inode in case something goes
2919  * wrong in the middle of an unlink/truncate.
2920  *
2921  * NOTE: caller of this function should reserve 5 units of metadata for
2922  *       this function.
2923  */
2924 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2925 {
2926         struct btrfs_root *root = BTRFS_I(inode)->root;
2927         struct btrfs_block_rsv *block_rsv = NULL;
2928         int reserve = 0;
2929         int insert = 0;
2930         int ret;
2931
2932         if (!root->orphan_block_rsv) {
2933                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2934                 if (!block_rsv)
2935                         return -ENOMEM;
2936         }
2937
2938         spin_lock(&root->orphan_lock);
2939         if (!root->orphan_block_rsv) {
2940                 root->orphan_block_rsv = block_rsv;
2941         } else if (block_rsv) {
2942                 btrfs_free_block_rsv(root, block_rsv);
2943                 block_rsv = NULL;
2944         }
2945
2946         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2947                               &BTRFS_I(inode)->runtime_flags)) {
2948 #if 0
2949                 /*
2950                  * For proper ENOSPC handling, we should do orphan
2951                  * cleanup when mounting. But this introduces backward
2952                  * compatibility issue.
2953                  */
2954                 if (!xchg(&root->orphan_item_inserted, 1))
2955                         insert = 2;
2956                 else
2957                         insert = 1;
2958 #endif
2959                 insert = 1;
2960                 atomic_inc(&root->orphan_inodes);
2961         }
2962
2963         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2964                               &BTRFS_I(inode)->runtime_flags))
2965                 reserve = 1;
2966         spin_unlock(&root->orphan_lock);
2967
2968         /* grab metadata reservation from transaction handle */
2969         if (reserve) {
2970                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2971                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2972         }
2973
2974         /* insert an orphan item to track this unlinked/truncated file */
2975         if (insert >= 1) {
2976                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2977                 if (ret) {
2978                         atomic_dec(&root->orphan_inodes);
2979                         if (reserve) {
2980                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2981                                           &BTRFS_I(inode)->runtime_flags);
2982                                 btrfs_orphan_release_metadata(inode);
2983                         }
2984                         if (ret != -EEXIST) {
2985                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2986                                           &BTRFS_I(inode)->runtime_flags);
2987                                 btrfs_abort_transaction(trans, root, ret);
2988                                 return ret;
2989                         }
2990                 }
2991                 ret = 0;
2992         }
2993
2994         /* insert an orphan item to track subvolume contains orphan files */
2995         if (insert >= 2) {
2996                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2997                                                root->root_key.objectid);
2998                 if (ret && ret != -EEXIST) {
2999                         btrfs_abort_transaction(trans, root, ret);
3000                         return ret;
3001                 }
3002         }
3003         return 0;
3004 }
3005
3006 /*
3007  * We have done the truncate/delete so we can go ahead and remove the orphan
3008  * item for this particular inode.
3009  */
3010 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3011                             struct inode *inode)
3012 {
3013         struct btrfs_root *root = BTRFS_I(inode)->root;
3014         int delete_item = 0;
3015         int release_rsv = 0;
3016         int ret = 0;
3017
3018         spin_lock(&root->orphan_lock);
3019         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3020                                &BTRFS_I(inode)->runtime_flags))
3021                 delete_item = 1;
3022
3023         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3024                                &BTRFS_I(inode)->runtime_flags))
3025                 release_rsv = 1;
3026         spin_unlock(&root->orphan_lock);
3027
3028         if (delete_item) {
3029                 atomic_dec(&root->orphan_inodes);
3030                 if (trans)
3031                         ret = btrfs_del_orphan_item(trans, root,
3032                                                     btrfs_ino(inode));
3033         }
3034
3035         if (release_rsv)
3036                 btrfs_orphan_release_metadata(inode);
3037
3038         return ret;
3039 }
3040
3041 /*
3042  * this cleans up any orphans that may be left on the list from the last use
3043  * of this root.
3044  */
3045 int btrfs_orphan_cleanup(struct btrfs_root *root)
3046 {
3047         struct btrfs_path *path;
3048         struct extent_buffer *leaf;
3049         struct btrfs_key key, found_key;
3050         struct btrfs_trans_handle *trans;
3051         struct inode *inode;
3052         u64 last_objectid = 0;
3053         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3054
3055         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3056                 return 0;
3057
3058         path = btrfs_alloc_path();
3059         if (!path) {
3060                 ret = -ENOMEM;
3061                 goto out;
3062         }
3063         path->reada = -1;
3064
3065         key.objectid = BTRFS_ORPHAN_OBJECTID;
3066         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3067         key.offset = (u64)-1;
3068
3069         while (1) {
3070                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3071                 if (ret < 0)
3072                         goto out;
3073
3074                 /*
3075                  * if ret == 0 means we found what we were searching for, which
3076                  * is weird, but possible, so only screw with path if we didn't
3077                  * find the key and see if we have stuff that matches
3078                  */
3079                 if (ret > 0) {
3080                         ret = 0;
3081                         if (path->slots[0] == 0)
3082                                 break;
3083                         path->slots[0]--;
3084                 }
3085
3086                 /* pull out the item */
3087                 leaf = path->nodes[0];
3088                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3089
3090                 /* make sure the item matches what we want */
3091                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3092                         break;
3093                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3094                         break;
3095
3096                 /* release the path since we're done with it */
3097                 btrfs_release_path(path);
3098
3099                 /*
3100                  * this is where we are basically btrfs_lookup, without the
3101                  * crossing root thing.  we store the inode number in the
3102                  * offset of the orphan item.
3103                  */
3104
3105                 if (found_key.offset == last_objectid) {
3106                         btrfs_err(root->fs_info,
3107                                 "Error removing orphan entry, stopping orphan cleanup");
3108                         ret = -EINVAL;
3109                         goto out;
3110                 }
3111
3112                 last_objectid = found_key.offset;
3113
3114                 found_key.objectid = found_key.offset;
3115                 found_key.type = BTRFS_INODE_ITEM_KEY;
3116                 found_key.offset = 0;
3117                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3118                 ret = PTR_ERR_OR_ZERO(inode);
3119                 if (ret && ret != -ESTALE)
3120                         goto out;
3121
3122                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3123                         struct btrfs_root *dead_root;
3124                         struct btrfs_fs_info *fs_info = root->fs_info;
3125                         int is_dead_root = 0;
3126
3127                         /*
3128                          * this is an orphan in the tree root. Currently these
3129                          * could come from 2 sources:
3130                          *  a) a snapshot deletion in progress
3131                          *  b) a free space cache inode
3132                          * We need to distinguish those two, as the snapshot
3133                          * orphan must not get deleted.
3134                          * find_dead_roots already ran before us, so if this
3135                          * is a snapshot deletion, we should find the root
3136                          * in the dead_roots list
3137                          */
3138                         spin_lock(&fs_info->trans_lock);
3139                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3140                                             root_list) {
3141                                 if (dead_root->root_key.objectid ==
3142                                     found_key.objectid) {
3143                                         is_dead_root = 1;
3144                                         break;
3145                                 }
3146                         }
3147                         spin_unlock(&fs_info->trans_lock);
3148                         if (is_dead_root) {
3149                                 /* prevent this orphan from being found again */
3150                                 key.offset = found_key.objectid - 1;
3151                                 continue;
3152                         }
3153                 }
3154                 /*
3155                  * Inode is already gone but the orphan item is still there,
3156                  * kill the orphan item.
3157                  */
3158                 if (ret == -ESTALE) {
3159                         trans = btrfs_start_transaction(root, 1);
3160                         if (IS_ERR(trans)) {
3161                                 ret = PTR_ERR(trans);
3162                                 goto out;
3163                         }
3164                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3165                                 found_key.objectid);
3166                         ret = btrfs_del_orphan_item(trans, root,
3167                                                     found_key.objectid);
3168                         btrfs_end_transaction(trans, root);
3169                         if (ret)
3170                                 goto out;
3171                         continue;
3172                 }
3173
3174                 /*
3175                  * add this inode to the orphan list so btrfs_orphan_del does
3176                  * the proper thing when we hit it
3177                  */
3178                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3179                         &BTRFS_I(inode)->runtime_flags);
3180                 atomic_inc(&root->orphan_inodes);
3181
3182                 /* if we have links, this was a truncate, lets do that */
3183                 if (inode->i_nlink) {
3184                         if (!S_ISREG(inode->i_mode)) {
3185                                 WARN_ON(1);
3186                                 iput(inode);
3187                                 continue;
3188                         }
3189                         nr_truncate++;
3190
3191                         /* 1 for the orphan item deletion. */
3192                         trans = btrfs_start_transaction(root, 1);
3193                         if (IS_ERR(trans)) {
3194                                 iput(inode);
3195                                 ret = PTR_ERR(trans);
3196                                 goto out;
3197                         }
3198                         ret = btrfs_orphan_add(trans, inode);
3199                         btrfs_end_transaction(trans, root);
3200                         if (ret) {
3201                                 iput(inode);
3202                                 goto out;
3203                         }
3204
3205                         ret = btrfs_truncate(inode);
3206                         if (ret)
3207                                 btrfs_orphan_del(NULL, inode);
3208                 } else {
3209                         nr_unlink++;
3210                 }
3211
3212                 /* this will do delete_inode and everything for us */
3213                 iput(inode);
3214                 if (ret)
3215                         goto out;
3216         }
3217         /* release the path since we're done with it */
3218         btrfs_release_path(path);
3219
3220         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3221
3222         if (root->orphan_block_rsv)
3223                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3224                                         (u64)-1);
3225
3226         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3227                 trans = btrfs_join_transaction(root);
3228                 if (!IS_ERR(trans))
3229                         btrfs_end_transaction(trans, root);
3230         }
3231
3232         if (nr_unlink)
3233                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3234         if (nr_truncate)
3235                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3236
3237 out:
3238         if (ret)
3239                 btrfs_crit(root->fs_info,
3240                         "could not do orphan cleanup %d", ret);
3241         btrfs_free_path(path);
3242         return ret;
3243 }
3244
3245 /*
3246  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3247  * don't find any xattrs, we know there can't be any acls.
3248  *
3249  * slot is the slot the inode is in, objectid is the objectid of the inode
3250  */
3251 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3252                                           int slot, u64 objectid)
3253 {
3254         u32 nritems = btrfs_header_nritems(leaf);
3255         struct btrfs_key found_key;
3256         static u64 xattr_access = 0;
3257         static u64 xattr_default = 0;
3258         int scanned = 0;
3259
3260         if (!xattr_access) {
3261                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3262                                         strlen(POSIX_ACL_XATTR_ACCESS));
3263                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3264                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3265         }
3266
3267         slot++;
3268         while (slot < nritems) {
3269                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3270
3271                 /* we found a different objectid, there must not be acls */
3272                 if (found_key.objectid != objectid)
3273                         return 0;
3274
3275                 /* we found an xattr, assume we've got an acl */
3276                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3277                         if (found_key.offset == xattr_access ||
3278                             found_key.offset == xattr_default)
3279                                 return 1;
3280                 }
3281
3282                 /*
3283                  * we found a key greater than an xattr key, there can't
3284                  * be any acls later on
3285                  */
3286                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3287                         return 0;
3288
3289                 slot++;
3290                 scanned++;
3291
3292                 /*
3293                  * it goes inode, inode backrefs, xattrs, extents,
3294                  * so if there are a ton of hard links to an inode there can
3295                  * be a lot of backrefs.  Don't waste time searching too hard,
3296                  * this is just an optimization
3297                  */
3298                 if (scanned >= 8)
3299                         break;
3300         }
3301         /* we hit the end of the leaf before we found an xattr or
3302          * something larger than an xattr.  We have to assume the inode
3303          * has acls
3304          */
3305         return 1;
3306 }
3307
3308 /*
3309  * read an inode from the btree into the in-memory inode
3310  */
3311 static void btrfs_read_locked_inode(struct inode *inode)
3312 {
3313         struct btrfs_path *path;
3314         struct extent_buffer *leaf;
3315         struct btrfs_inode_item *inode_item;
3316         struct btrfs_timespec *tspec;
3317         struct btrfs_root *root = BTRFS_I(inode)->root;
3318         struct btrfs_key location;
3319         int maybe_acls;
3320         u32 rdev;
3321         int ret;
3322         bool filled = false;
3323
3324         ret = btrfs_fill_inode(inode, &rdev);
3325         if (!ret)
3326                 filled = true;
3327
3328         path = btrfs_alloc_path();
3329         if (!path)
3330                 goto make_bad;
3331
3332         path->leave_spinning = 1;
3333         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3334
3335         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3336         if (ret)
3337                 goto make_bad;
3338
3339         leaf = path->nodes[0];
3340
3341         if (filled)
3342                 goto cache_acl;
3343
3344         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3345                                     struct btrfs_inode_item);
3346         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3347         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3348         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3349         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3350         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3351
3352         tspec = btrfs_inode_atime(inode_item);
3353         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3354         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3355
3356         tspec = btrfs_inode_mtime(inode_item);
3357         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3358         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3359
3360         tspec = btrfs_inode_ctime(inode_item);
3361         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3362         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3363
3364         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3365         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3366         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3367
3368         /*
3369          * If we were modified in the current generation and evicted from memory
3370          * and then re-read we need to do a full sync since we don't have any
3371          * idea about which extents were modified before we were evicted from
3372          * cache.
3373          */
3374         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3375                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3376                         &BTRFS_I(inode)->runtime_flags);
3377
3378         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3379         inode->i_generation = BTRFS_I(inode)->generation;
3380         inode->i_rdev = 0;
3381         rdev = btrfs_inode_rdev(leaf, inode_item);
3382
3383         BTRFS_I(inode)->index_cnt = (u64)-1;
3384         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3385 cache_acl:
3386         /*
3387          * try to precache a NULL acl entry for files that don't have
3388          * any xattrs or acls
3389          */
3390         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3391                                            btrfs_ino(inode));
3392         if (!maybe_acls)
3393                 cache_no_acl(inode);
3394
3395         btrfs_free_path(path);
3396
3397         switch (inode->i_mode & S_IFMT) {
3398         case S_IFREG:
3399                 inode->i_mapping->a_ops = &btrfs_aops;
3400                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3401                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3402                 inode->i_fop = &btrfs_file_operations;
3403                 inode->i_op = &btrfs_file_inode_operations;
3404                 break;
3405         case S_IFDIR:
3406                 inode->i_fop = &btrfs_dir_file_operations;
3407                 if (root == root->fs_info->tree_root)
3408                         inode->i_op = &btrfs_dir_ro_inode_operations;
3409                 else
3410                         inode->i_op = &btrfs_dir_inode_operations;
3411                 break;
3412         case S_IFLNK:
3413                 inode->i_op = &btrfs_symlink_inode_operations;
3414                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3415                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3416                 break;
3417         default:
3418                 inode->i_op = &btrfs_special_inode_operations;
3419                 init_special_inode(inode, inode->i_mode, rdev);
3420                 break;
3421         }
3422
3423         btrfs_update_iflags(inode);
3424         return;
3425
3426 make_bad:
3427         btrfs_free_path(path);
3428         make_bad_inode(inode);
3429 }
3430
3431 /*
3432  * given a leaf and an inode, copy the inode fields into the leaf
3433  */
3434 static void fill_inode_item(struct btrfs_trans_handle *trans,
3435                             struct extent_buffer *leaf,
3436                             struct btrfs_inode_item *item,
3437                             struct inode *inode)
3438 {
3439         struct btrfs_map_token token;
3440
3441         btrfs_init_map_token(&token);
3442
3443         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3444         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3445         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3446                                    &token);
3447         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3448         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3449
3450         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3451                                      inode->i_atime.tv_sec, &token);
3452         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3453                                       inode->i_atime.tv_nsec, &token);
3454
3455         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3456                                      inode->i_mtime.tv_sec, &token);
3457         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3458                                       inode->i_mtime.tv_nsec, &token);
3459
3460         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3461                                      inode->i_ctime.tv_sec, &token);
3462         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3463                                       inode->i_ctime.tv_nsec, &token);
3464
3465         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3466                                      &token);
3467         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3468                                          &token);
3469         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3470         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3471         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3472         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3473         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3474 }
3475
3476 /*
3477  * copy everything in the in-memory inode into the btree.
3478  */
3479 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3480                                 struct btrfs_root *root, struct inode *inode)
3481 {
3482         struct btrfs_inode_item *inode_item;
3483         struct btrfs_path *path;
3484         struct extent_buffer *leaf;
3485         int ret;
3486
3487         path = btrfs_alloc_path();
3488         if (!path)
3489                 return -ENOMEM;
3490
3491         path->leave_spinning = 1;
3492         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3493                                  1);
3494         if (ret) {
3495                 if (ret > 0)
3496                         ret = -ENOENT;
3497                 goto failed;
3498         }
3499
3500         btrfs_unlock_up_safe(path, 1);
3501         leaf = path->nodes[0];
3502         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3503                                     struct btrfs_inode_item);
3504
3505         fill_inode_item(trans, leaf, inode_item, inode);
3506         btrfs_mark_buffer_dirty(leaf);
3507         btrfs_set_inode_last_trans(trans, inode);
3508         ret = 0;
3509 failed:
3510         btrfs_free_path(path);
3511         return ret;
3512 }
3513
3514 /*
3515  * copy everything in the in-memory inode into the btree.
3516  */
3517 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3518                                 struct btrfs_root *root, struct inode *inode)
3519 {
3520         int ret;
3521
3522         /*
3523          * If the inode is a free space inode, we can deadlock during commit
3524          * if we put it into the delayed code.
3525          *
3526          * The data relocation inode should also be directly updated
3527          * without delay
3528          */
3529         if (!btrfs_is_free_space_inode(inode)
3530             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3531                 btrfs_update_root_times(trans, root);
3532
3533                 ret = btrfs_delayed_update_inode(trans, root, inode);
3534                 if (!ret)
3535                         btrfs_set_inode_last_trans(trans, inode);
3536                 return ret;
3537         }
3538
3539         return btrfs_update_inode_item(trans, root, inode);
3540 }
3541
3542 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3543                                          struct btrfs_root *root,
3544                                          struct inode *inode)
3545 {
3546         int ret;
3547
3548         ret = btrfs_update_inode(trans, root, inode);
3549         if (ret == -ENOSPC)
3550                 return btrfs_update_inode_item(trans, root, inode);
3551         return ret;
3552 }
3553
3554 /*
3555  * unlink helper that gets used here in inode.c and in the tree logging
3556  * recovery code.  It remove a link in a directory with a given name, and
3557  * also drops the back refs in the inode to the directory
3558  */
3559 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3560                                 struct btrfs_root *root,
3561                                 struct inode *dir, struct inode *inode,
3562                                 const char *name, int name_len)
3563 {
3564         struct btrfs_path *path;
3565         int ret = 0;
3566         struct extent_buffer *leaf;
3567         struct btrfs_dir_item *di;
3568         struct btrfs_key key;
3569         u64 index;
3570         u64 ino = btrfs_ino(inode);
3571         u64 dir_ino = btrfs_ino(dir);
3572
3573         path = btrfs_alloc_path();
3574         if (!path) {
3575                 ret = -ENOMEM;
3576                 goto out;
3577         }
3578
3579         path->leave_spinning = 1;
3580         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3581                                     name, name_len, -1);
3582         if (IS_ERR(di)) {
3583                 ret = PTR_ERR(di);
3584                 goto err;
3585         }
3586         if (!di) {
3587                 ret = -ENOENT;
3588                 goto err;
3589         }
3590         leaf = path->nodes[0];
3591         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3592         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3593         if (ret)
3594                 goto err;
3595         btrfs_release_path(path);
3596
3597         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3598                                   dir_ino, &index);
3599         if (ret) {
3600                 btrfs_info(root->fs_info,
3601                         "failed to delete reference to %.*s, inode %llu parent %llu",
3602                         name_len, name, ino, dir_ino);
3603                 btrfs_abort_transaction(trans, root, ret);
3604                 goto err;
3605         }
3606
3607         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3608         if (ret) {
3609                 btrfs_abort_transaction(trans, root, ret);
3610                 goto err;
3611         }
3612
3613         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3614                                          inode, dir_ino);
3615         if (ret != 0 && ret != -ENOENT) {
3616                 btrfs_abort_transaction(trans, root, ret);
3617                 goto err;
3618         }
3619
3620         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3621                                            dir, index);
3622         if (ret == -ENOENT)
3623                 ret = 0;
3624         else if (ret)
3625                 btrfs_abort_transaction(trans, root, ret);
3626 err:
3627         btrfs_free_path(path);
3628         if (ret)
3629                 goto out;
3630
3631         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3632         inode_inc_iversion(inode);
3633         inode_inc_iversion(dir);
3634         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3635         ret = btrfs_update_inode(trans, root, dir);
3636 out:
3637         return ret;
3638 }
3639
3640 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3641                        struct btrfs_root *root,
3642                        struct inode *dir, struct inode *inode,
3643                        const char *name, int name_len)
3644 {
3645         int ret;
3646         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3647         if (!ret) {
3648                 btrfs_drop_nlink(inode);
3649                 ret = btrfs_update_inode(trans, root, inode);
3650         }
3651         return ret;
3652 }
3653
3654 /*
3655  * helper to start transaction for unlink and rmdir.
3656  *
3657  * unlink and rmdir are special in btrfs, they do not always free space, so
3658  * if we cannot make our reservations the normal way try and see if there is
3659  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3660  * allow the unlink to occur.
3661  */
3662 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3663 {
3664         struct btrfs_trans_handle *trans;
3665         struct btrfs_root *root = BTRFS_I(dir)->root;
3666         int ret;
3667
3668         /*
3669          * 1 for the possible orphan item
3670          * 1 for the dir item
3671          * 1 for the dir index
3672          * 1 for the inode ref
3673          * 1 for the inode
3674          */
3675         trans = btrfs_start_transaction(root, 5);
3676         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3677                 return trans;
3678
3679         if (PTR_ERR(trans) == -ENOSPC) {
3680                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3681
3682                 trans = btrfs_start_transaction(root, 0);
3683                 if (IS_ERR(trans))
3684                         return trans;
3685                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3686                                                &root->fs_info->trans_block_rsv,
3687                                                num_bytes, 5);
3688                 if (ret) {
3689                         btrfs_end_transaction(trans, root);
3690                         return ERR_PTR(ret);
3691                 }
3692                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3693                 trans->bytes_reserved = num_bytes;
3694         }
3695         return trans;
3696 }
3697
3698 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3699 {
3700         struct btrfs_root *root = BTRFS_I(dir)->root;
3701         struct btrfs_trans_handle *trans;
3702         struct inode *inode = dentry->d_inode;
3703         int ret;
3704
3705         trans = __unlink_start_trans(dir);
3706         if (IS_ERR(trans))
3707                 return PTR_ERR(trans);
3708
3709         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3710
3711         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3712                                  dentry->d_name.name, dentry->d_name.len);
3713         if (ret)
3714                 goto out;
3715
3716         if (inode->i_nlink == 0) {
3717                 ret = btrfs_orphan_add(trans, inode);
3718                 if (ret)
3719                         goto out;
3720         }
3721
3722 out:
3723         btrfs_end_transaction(trans, root);
3724         btrfs_btree_balance_dirty(root);
3725         return ret;
3726 }
3727
3728 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3729                         struct btrfs_root *root,
3730                         struct inode *dir, u64 objectid,
3731                         const char *name, int name_len)
3732 {
3733         struct btrfs_path *path;
3734         struct extent_buffer *leaf;
3735         struct btrfs_dir_item *di;
3736         struct btrfs_key key;
3737         u64 index;
3738         int ret;
3739         u64 dir_ino = btrfs_ino(dir);
3740
3741         path = btrfs_alloc_path();
3742         if (!path)
3743                 return -ENOMEM;
3744
3745         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3746                                    name, name_len, -1);
3747         if (IS_ERR_OR_NULL(di)) {
3748                 if (!di)
3749                         ret = -ENOENT;
3750                 else
3751                         ret = PTR_ERR(di);
3752                 goto out;
3753         }
3754
3755         leaf = path->nodes[0];
3756         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3757         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3758         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3759         if (ret) {
3760                 btrfs_abort_transaction(trans, root, ret);
3761                 goto out;
3762         }
3763         btrfs_release_path(path);
3764
3765         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3766                                  objectid, root->root_key.objectid,
3767                                  dir_ino, &index, name, name_len);
3768         if (ret < 0) {
3769                 if (ret != -ENOENT) {
3770                         btrfs_abort_transaction(trans, root, ret);
3771                         goto out;
3772                 }
3773                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3774                                                  name, name_len);
3775                 if (IS_ERR_OR_NULL(di)) {
3776                         if (!di)
3777                                 ret = -ENOENT;
3778                         else
3779                                 ret = PTR_ERR(di);
3780                         btrfs_abort_transaction(trans, root, ret);
3781                         goto out;
3782                 }
3783
3784                 leaf = path->nodes[0];
3785                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3786                 btrfs_release_path(path);
3787                 index = key.offset;
3788         }
3789         btrfs_release_path(path);
3790
3791         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3792         if (ret) {
3793                 btrfs_abort_transaction(trans, root, ret);
3794                 goto out;
3795         }
3796
3797         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3798         inode_inc_iversion(dir);
3799         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3800         ret = btrfs_update_inode_fallback(trans, root, dir);
3801         if (ret)
3802                 btrfs_abort_transaction(trans, root, ret);
3803 out:
3804         btrfs_free_path(path);
3805         return ret;
3806 }
3807
3808 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3809 {
3810         struct inode *inode = dentry->d_inode;
3811         int err = 0;
3812         struct btrfs_root *root = BTRFS_I(dir)->root;
3813         struct btrfs_trans_handle *trans;
3814
3815         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3816                 return -ENOTEMPTY;
3817         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3818                 return -EPERM;
3819
3820         trans = __unlink_start_trans(dir);
3821         if (IS_ERR(trans))
3822                 return PTR_ERR(trans);
3823
3824         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3825                 err = btrfs_unlink_subvol(trans, root, dir,
3826                                           BTRFS_I(inode)->location.objectid,
3827                                           dentry->d_name.name,
3828                                           dentry->d_name.len);
3829                 goto out;
3830         }
3831
3832         err = btrfs_orphan_add(trans, inode);
3833         if (err)
3834                 goto out;
3835
3836         /* now the directory is empty */
3837         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3838                                  dentry->d_name.name, dentry->d_name.len);
3839         if (!err)
3840                 btrfs_i_size_write(inode, 0);
3841 out:
3842         btrfs_end_transaction(trans, root);
3843         btrfs_btree_balance_dirty(root);
3844
3845         return err;
3846 }
3847
3848 /*
3849  * this can truncate away extent items, csum items and directory items.
3850  * It starts at a high offset and removes keys until it can't find
3851  * any higher than new_size
3852  *
3853  * csum items that cross the new i_size are truncated to the new size
3854  * as well.
3855  *
3856  * min_type is the minimum key type to truncate down to.  If set to 0, this
3857  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3858  */
3859 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3860                                struct btrfs_root *root,
3861                                struct inode *inode,
3862                                u64 new_size, u32 min_type)
3863 {
3864         struct btrfs_path *path;
3865         struct extent_buffer *leaf;
3866         struct btrfs_file_extent_item *fi;
3867         struct btrfs_key key;
3868         struct btrfs_key found_key;
3869         u64 extent_start = 0;
3870         u64 extent_num_bytes = 0;
3871         u64 extent_offset = 0;
3872         u64 item_end = 0;
3873         u64 last_size = (u64)-1;
3874         u32 found_type = (u8)-1;
3875         int found_extent;
3876         int del_item;
3877         int pending_del_nr = 0;
3878         int pending_del_slot = 0;
3879         int extent_type = -1;
3880         int ret;
3881         int err = 0;
3882         u64 ino = btrfs_ino(inode);
3883
3884         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3885
3886         path = btrfs_alloc_path();
3887         if (!path)
3888                 return -ENOMEM;
3889         path->reada = -1;
3890
3891         /*
3892          * We want to drop from the next block forward in case this new size is
3893          * not block aligned since we will be keeping the last block of the
3894          * extent just the way it is.
3895          */
3896         if (root->ref_cows || root == root->fs_info->tree_root)
3897                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3898                                         root->sectorsize), (u64)-1, 0);
3899
3900         /*
3901          * This function is also used to drop the items in the log tree before
3902          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3903          * it is used to drop the loged items. So we shouldn't kill the delayed
3904          * items.
3905          */
3906         if (min_type == 0 && root == BTRFS_I(inode)->root)
3907                 btrfs_kill_delayed_inode_items(inode);
3908
3909         key.objectid = ino;
3910         key.offset = (u64)-1;
3911         key.type = (u8)-1;
3912
3913 search_again:
3914         path->leave_spinning = 1;
3915         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3916         if (ret < 0) {
3917                 err = ret;
3918                 goto out;
3919         }
3920
3921         if (ret > 0) {
3922                 /* there are no items in the tree for us to truncate, we're
3923                  * done
3924                  */
3925                 if (path->slots[0] == 0)
3926                         goto out;
3927                 path->slots[0]--;
3928         }
3929
3930         while (1) {
3931                 fi = NULL;
3932                 leaf = path->nodes[0];
3933                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3934                 found_type = btrfs_key_type(&found_key);
3935
3936                 if (found_key.objectid != ino)
3937                         break;
3938
3939                 if (found_type < min_type)
3940                         break;
3941
3942                 item_end = found_key.offset;
3943                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3944                         fi = btrfs_item_ptr(leaf, path->slots[0],
3945                                             struct btrfs_file_extent_item);
3946                         extent_type = btrfs_file_extent_type(leaf, fi);
3947                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3948                                 item_end +=
3949                                     btrfs_file_extent_num_bytes(leaf, fi);
3950                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3951                                 item_end += btrfs_file_extent_inline_len(leaf,
3952                                                                          fi);
3953                         }
3954                         item_end--;
3955                 }
3956                 if (found_type > min_type) {
3957                         del_item = 1;
3958                 } else {
3959                         if (item_end < new_size)
3960                                 break;
3961                         if (found_key.offset >= new_size)
3962                                 del_item = 1;
3963                         else
3964                                 del_item = 0;
3965                 }
3966                 found_extent = 0;
3967                 /* FIXME, shrink the extent if the ref count is only 1 */
3968                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3969                         goto delete;
3970
3971                 if (del_item)
3972                         last_size = found_key.offset;
3973                 else
3974                         last_size = new_size;
3975
3976                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3977                         u64 num_dec;
3978                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3979                         if (!del_item) {
3980                                 u64 orig_num_bytes =
3981                                         btrfs_file_extent_num_bytes(leaf, fi);
3982                                 extent_num_bytes = ALIGN(new_size -
3983                                                 found_key.offset,
3984                                                 root->sectorsize);
3985                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3986                                                          extent_num_bytes);
3987                                 num_dec = (orig_num_bytes -
3988                                            extent_num_bytes);
3989                                 if (root->ref_cows && extent_start != 0)
3990                                         inode_sub_bytes(inode, num_dec);
3991                                 btrfs_mark_buffer_dirty(leaf);
3992                         } else {
3993                                 extent_num_bytes =
3994                                         btrfs_file_extent_disk_num_bytes(leaf,
3995                                                                          fi);
3996                                 extent_offset = found_key.offset -
3997                                         btrfs_file_extent_offset(leaf, fi);
3998
3999                                 /* FIXME blocksize != 4096 */
4000                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
4001                                 if (extent_start != 0) {
4002                                         found_extent = 1;
4003                                         if (root->ref_cows)
4004                                                 inode_sub_bytes(inode, num_dec);
4005                                 }
4006                         }
4007                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4008                         /*
4009                          * we can't truncate inline items that have had
4010                          * special encodings
4011                          */
4012                         if (!del_item &&
4013                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4014                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4015                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4016                                 u32 size = new_size - found_key.offset;
4017
4018                                 if (root->ref_cows) {
4019                                         inode_sub_bytes(inode, item_end + 1 -
4020                                                         new_size);
4021                                 }
4022                                 size =
4023                                     btrfs_file_extent_calc_inline_size(size);
4024                                 btrfs_truncate_item(root, path, size, 1);
4025                         } else if (root->ref_cows) {
4026                                 inode_sub_bytes(inode, item_end + 1 -
4027                                                 found_key.offset);
4028                         }
4029                 }
4030 delete:
4031                 if (del_item) {
4032                         if (!pending_del_nr) {
4033                                 /* no pending yet, add ourselves */
4034                                 pending_del_slot = path->slots[0];
4035                                 pending_del_nr = 1;
4036                         } else if (pending_del_nr &&
4037                                    path->slots[0] + 1 == pending_del_slot) {
4038                                 /* hop on the pending chunk */
4039                                 pending_del_nr++;
4040                                 pending_del_slot = path->slots[0];
4041                         } else {
4042                                 BUG();
4043                         }
4044                 } else {
4045                         break;
4046                 }
4047                 if (found_extent && (root->ref_cows ||
4048                                      root == root->fs_info->tree_root)) {
4049                         btrfs_set_path_blocking(path);
4050                         ret = btrfs_free_extent(trans, root, extent_start,
4051                                                 extent_num_bytes, 0,
4052                                                 btrfs_header_owner(leaf),
4053                                                 ino, extent_offset, 0);
4054                         BUG_ON(ret);
4055                 }
4056
4057                 if (found_type == BTRFS_INODE_ITEM_KEY)
4058                         break;
4059
4060                 if (path->slots[0] == 0 ||
4061                     path->slots[0] != pending_del_slot) {
4062                         if (pending_del_nr) {
4063                                 ret = btrfs_del_items(trans, root, path,
4064                                                 pending_del_slot,
4065                                                 pending_del_nr);
4066                                 if (ret) {
4067                                         btrfs_abort_transaction(trans,
4068                                                                 root, ret);
4069                                         goto error;
4070                                 }
4071                                 pending_del_nr = 0;
4072                         }
4073                         btrfs_release_path(path);
4074                         goto search_again;
4075                 } else {
4076                         path->slots[0]--;
4077                 }
4078         }
4079 out:
4080         if (pending_del_nr) {
4081                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4082                                       pending_del_nr);
4083                 if (ret)
4084                         btrfs_abort_transaction(trans, root, ret);
4085         }
4086 error:
4087         if (last_size != (u64)-1)
4088                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4089         btrfs_free_path(path);
4090         return err;
4091 }
4092
4093 /*
4094  * btrfs_truncate_page - read, zero a chunk and write a page
4095  * @inode - inode that we're zeroing
4096  * @from - the offset to start zeroing
4097  * @len - the length to zero, 0 to zero the entire range respective to the
4098  *      offset
4099  * @front - zero up to the offset instead of from the offset on
4100  *
4101  * This will find the page for the "from" offset and cow the page and zero the
4102  * part we want to zero.  This is used with truncate and hole punching.
4103  */
4104 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4105                         int front)
4106 {
4107         struct address_space *mapping = inode->i_mapping;
4108         struct btrfs_root *root = BTRFS_I(inode)->root;
4109         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4110         struct btrfs_ordered_extent *ordered;
4111         struct extent_state *cached_state = NULL;
4112         char *kaddr;
4113         u32 blocksize = root->sectorsize;
4114         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4115         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4116         struct page *page;
4117         gfp_t mask = btrfs_alloc_write_mask(mapping);
4118         int ret = 0;
4119         u64 page_start;
4120         u64 page_end;
4121
4122         if ((offset & (blocksize - 1)) == 0 &&
4123             (!len || ((len & (blocksize - 1)) == 0)))
4124                 goto out;
4125         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4126         if (ret)
4127                 goto out;
4128
4129 again:
4130         page = find_or_create_page(mapping, index, mask);
4131         if (!page) {
4132                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4133                 ret = -ENOMEM;
4134                 goto out;
4135         }
4136
4137         page_start = page_offset(page);
4138         page_end = page_start + PAGE_CACHE_SIZE - 1;
4139
4140         if (!PageUptodate(page)) {
4141                 ret = btrfs_readpage(NULL, page);
4142                 lock_page(page);
4143                 if (page->mapping != mapping) {
4144                         unlock_page(page);
4145                         page_cache_release(page);
4146                         goto again;
4147                 }
4148                 if (!PageUptodate(page)) {
4149                         ret = -EIO;
4150                         goto out_unlock;
4151                 }
4152         }
4153         wait_on_page_writeback(page);
4154
4155         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4156         set_page_extent_mapped(page);
4157
4158         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4159         if (ordered) {
4160                 unlock_extent_cached(io_tree, page_start, page_end,
4161                                      &cached_state, GFP_NOFS);
4162                 unlock_page(page);
4163                 page_cache_release(page);
4164                 btrfs_start_ordered_extent(inode, ordered, 1);
4165                 btrfs_put_ordered_extent(ordered);
4166                 goto again;
4167         }
4168
4169         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4170                           EXTENT_DIRTY | EXTENT_DELALLOC |
4171                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4172                           0, 0, &cached_state, GFP_NOFS);
4173
4174         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4175                                         &cached_state);
4176         if (ret) {
4177                 unlock_extent_cached(io_tree, page_start, page_end,
4178                                      &cached_state, GFP_NOFS);
4179                 goto out_unlock;
4180         }
4181
4182         if (offset != PAGE_CACHE_SIZE) {
4183                 if (!len)
4184                         len = PAGE_CACHE_SIZE - offset;
4185                 kaddr = kmap(page);
4186                 if (front)
4187                         memset(kaddr, 0, offset);
4188                 else
4189                         memset(kaddr + offset, 0, len);
4190                 flush_dcache_page(page);
4191                 kunmap(page);
4192         }
4193         ClearPageChecked(page);
4194         set_page_dirty(page);
4195         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4196                              GFP_NOFS);
4197
4198 out_unlock:
4199         if (ret)
4200                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4201         unlock_page(page);
4202         page_cache_release(page);
4203 out:
4204         return ret;
4205 }
4206
4207 /*
4208  * This function puts in dummy file extents for the area we're creating a hole
4209  * for.  So if we are truncating this file to a larger size we need to insert
4210  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4211  * the range between oldsize and size
4212  */
4213 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4214 {
4215         struct btrfs_trans_handle *trans;
4216         struct btrfs_root *root = BTRFS_I(inode)->root;
4217         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4218         struct extent_map *em = NULL;
4219         struct extent_state *cached_state = NULL;
4220         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4221         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4222         u64 block_end = ALIGN(size, root->sectorsize);
4223         u64 last_byte;
4224         u64 cur_offset;
4225         u64 hole_size;
4226         int err = 0;
4227
4228         /*
4229          * If our size started in the middle of a page we need to zero out the
4230          * rest of the page before we expand the i_size, otherwise we could
4231          * expose stale data.
4232          */
4233         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4234         if (err)
4235                 return err;
4236
4237         if (size <= hole_start)
4238                 return 0;
4239
4240         while (1) {
4241                 struct btrfs_ordered_extent *ordered;
4242
4243                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4244                                  &cached_state);
4245                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4246                                                      block_end - hole_start);
4247                 if (!ordered)
4248                         break;
4249                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4250                                      &cached_state, GFP_NOFS);
4251                 btrfs_start_ordered_extent(inode, ordered, 1);
4252                 btrfs_put_ordered_extent(ordered);
4253         }
4254
4255         cur_offset = hole_start;
4256         while (1) {
4257                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4258                                 block_end - cur_offset, 0);
4259                 if (IS_ERR(em)) {
4260                         err = PTR_ERR(em);
4261                         em = NULL;
4262                         break;
4263                 }
4264                 last_byte = min(extent_map_end(em), block_end);
4265                 last_byte = ALIGN(last_byte , root->sectorsize);
4266                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4267                         struct extent_map *hole_em;
4268                         hole_size = last_byte - cur_offset;
4269
4270                         trans = btrfs_start_transaction(root, 3);
4271                         if (IS_ERR(trans)) {
4272                                 err = PTR_ERR(trans);
4273                                 break;
4274                         }
4275
4276                         err = btrfs_drop_extents(trans, root, inode,
4277                                                  cur_offset,
4278                                                  cur_offset + hole_size, 1);
4279                         if (err) {
4280                                 btrfs_abort_transaction(trans, root, err);
4281                                 btrfs_end_transaction(trans, root);
4282                                 break;
4283                         }
4284
4285                         err = btrfs_insert_file_extent(trans, root,
4286                                         btrfs_ino(inode), cur_offset, 0,
4287                                         0, hole_size, 0, hole_size,
4288                                         0, 0, 0);
4289                         if (err) {
4290                                 btrfs_abort_transaction(trans, root, err);
4291                                 btrfs_end_transaction(trans, root);
4292                                 break;
4293                         }
4294
4295                         btrfs_drop_extent_cache(inode, cur_offset,
4296                                                 cur_offset + hole_size - 1, 0);
4297                         hole_em = alloc_extent_map();
4298                         if (!hole_em) {
4299                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4300                                         &BTRFS_I(inode)->runtime_flags);
4301                                 goto next;
4302                         }
4303                         hole_em->start = cur_offset;
4304                         hole_em->len = hole_size;
4305                         hole_em->orig_start = cur_offset;
4306
4307                         hole_em->block_start = EXTENT_MAP_HOLE;
4308                         hole_em->block_len = 0;
4309                         hole_em->orig_block_len = 0;
4310                         hole_em->ram_bytes = hole_size;
4311                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4312                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4313                         hole_em->generation = trans->transid;
4314
4315                         while (1) {
4316                                 write_lock(&em_tree->lock);
4317                                 err = add_extent_mapping(em_tree, hole_em, 1);
4318                                 write_unlock(&em_tree->lock);
4319                                 if (err != -EEXIST)
4320                                         break;
4321                                 btrfs_drop_extent_cache(inode, cur_offset,
4322                                                         cur_offset +
4323                                                         hole_size - 1, 0);
4324                         }
4325                         free_extent_map(hole_em);
4326 next:
4327                         btrfs_update_inode(trans, root, inode);
4328                         btrfs_end_transaction(trans, root);
4329                 }
4330                 free_extent_map(em);
4331                 em = NULL;
4332                 cur_offset = last_byte;
4333                 if (cur_offset >= block_end)
4334                         break;
4335         }
4336
4337         free_extent_map(em);
4338         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4339                              GFP_NOFS);
4340         return err;
4341 }
4342
4343 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4344 {
4345         struct btrfs_root *root = BTRFS_I(inode)->root;
4346         struct btrfs_trans_handle *trans;
4347         loff_t oldsize = i_size_read(inode);
4348         loff_t newsize = attr->ia_size;
4349         int mask = attr->ia_valid;
4350         int ret;
4351
4352         /*
4353          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4354          * special case where we need to update the times despite not having
4355          * these flags set.  For all other operations the VFS set these flags
4356          * explicitly if it wants a timestamp update.
4357          */
4358         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4359                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4360
4361         if (newsize > oldsize) {
4362                 truncate_pagecache(inode, newsize);
4363                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4364                 if (ret)
4365                         return ret;
4366
4367                 trans = btrfs_start_transaction(root, 1);
4368                 if (IS_ERR(trans))
4369                         return PTR_ERR(trans);
4370
4371                 i_size_write(inode, newsize);
4372                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4373                 ret = btrfs_update_inode(trans, root, inode);
4374                 btrfs_end_transaction(trans, root);
4375         } else {
4376
4377                 /*
4378                  * We're truncating a file that used to have good data down to
4379                  * zero. Make sure it gets into the ordered flush list so that
4380                  * any new writes get down to disk quickly.
4381                  */
4382                 if (newsize == 0)
4383                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4384                                 &BTRFS_I(inode)->runtime_flags);
4385
4386                 /*
4387                  * 1 for the orphan item we're going to add
4388                  * 1 for the orphan item deletion.
4389                  */
4390                 trans = btrfs_start_transaction(root, 2);
4391                 if (IS_ERR(trans))
4392                         return PTR_ERR(trans);
4393
4394                 /*
4395                  * We need to do this in case we fail at _any_ point during the
4396                  * actual truncate.  Once we do the truncate_setsize we could
4397                  * invalidate pages which forces any outstanding ordered io to
4398                  * be instantly completed which will give us extents that need
4399                  * to be truncated.  If we fail to get an orphan inode down we
4400                  * could have left over extents that were never meant to live,
4401                  * so we need to garuntee from this point on that everything
4402                  * will be consistent.
4403                  */
4404                 ret = btrfs_orphan_add(trans, inode);
4405                 btrfs_end_transaction(trans, root);
4406                 if (ret)
4407                         return ret;
4408
4409                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4410                 truncate_setsize(inode, newsize);
4411
4412                 /* Disable nonlocked read DIO to avoid the end less truncate */
4413                 btrfs_inode_block_unlocked_dio(inode);
4414                 inode_dio_wait(inode);
4415                 btrfs_inode_resume_unlocked_dio(inode);
4416
4417                 ret = btrfs_truncate(inode);
4418                 if (ret && inode->i_nlink) {
4419                         int err;
4420
4421                         /*
4422                          * failed to truncate, disk_i_size is only adjusted down
4423                          * as we remove extents, so it should represent the true
4424                          * size of the inode, so reset the in memory size and
4425                          * delete our orphan entry.
4426                          */
4427                         trans = btrfs_join_transaction(root);
4428                         if (IS_ERR(trans)) {
4429                                 btrfs_orphan_del(NULL, inode);
4430                                 return ret;
4431                         }
4432                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4433                         err = btrfs_orphan_del(trans, inode);
4434                         if (err)
4435                                 btrfs_abort_transaction(trans, root, err);
4436                         btrfs_end_transaction(trans, root);
4437                 }
4438         }
4439
4440         return ret;
4441 }
4442
4443 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4444 {
4445         struct inode *inode = dentry->d_inode;
4446         struct btrfs_root *root = BTRFS_I(inode)->root;
4447         int err;
4448
4449         if (btrfs_root_readonly(root))
4450                 return -EROFS;
4451
4452         err = inode_change_ok(inode, attr);
4453         if (err)
4454                 return err;
4455
4456         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4457                 err = btrfs_setsize(inode, attr);
4458                 if (err)
4459                         return err;
4460         }
4461
4462         if (attr->ia_valid) {
4463                 setattr_copy(inode, attr);
4464                 inode_inc_iversion(inode);
4465                 err = btrfs_dirty_inode(inode);
4466
4467                 if (!err && attr->ia_valid & ATTR_MODE)
4468                         err = btrfs_acl_chmod(inode);
4469         }
4470
4471         return err;
4472 }
4473
4474 void btrfs_evict_inode(struct inode *inode)
4475 {
4476         struct btrfs_trans_handle *trans;
4477         struct btrfs_root *root = BTRFS_I(inode)->root;
4478         struct btrfs_block_rsv *rsv, *global_rsv;
4479         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4480         int ret;
4481
4482         trace_btrfs_inode_evict(inode);
4483
4484         truncate_inode_pages(&inode->i_data, 0);
4485         if (inode->i_nlink &&
4486             ((btrfs_root_refs(&root->root_item) != 0 &&
4487               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4488              btrfs_is_free_space_inode(inode)))
4489                 goto no_delete;
4490
4491         if (is_bad_inode(inode)) {
4492                 btrfs_orphan_del(NULL, inode);
4493                 goto no_delete;
4494         }
4495         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4496         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4497
4498         if (root->fs_info->log_root_recovering) {
4499                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4500                                  &BTRFS_I(inode)->runtime_flags));
4501                 goto no_delete;
4502         }
4503
4504         if (inode->i_nlink > 0) {
4505                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4506                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
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         /*
4576          * Errors here aren't a big deal, it just means we leave orphan items
4577          * in the tree.  They will be cleaned up on the next mount.
4578          */
4579         if (ret == 0) {
4580                 trans->block_rsv = root->orphan_block_rsv;
4581                 btrfs_orphan_del(trans, inode);
4582         } else {
4583                 btrfs_orphan_del(NULL, inode);
4584         }
4585
4586         trans->block_rsv = &root->fs_info->trans_block_rsv;
4587         if (!(root == root->fs_info->tree_root ||
4588               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4589                 btrfs_return_ino(root, btrfs_ino(inode));
4590
4591         btrfs_end_transaction(trans, root);
4592         btrfs_btree_balance_dirty(root);
4593 no_delete:
4594         btrfs_remove_delayed_node(inode);
4595         clear_inode(inode);
4596         return;
4597 }
4598
4599 /*
4600  * this returns the key found in the dir entry in the location pointer.
4601  * If no dir entries were found, location->objectid is 0.
4602  */
4603 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4604                                struct btrfs_key *location)
4605 {
4606         const char *name = dentry->d_name.name;
4607         int namelen = dentry->d_name.len;
4608         struct btrfs_dir_item *di;
4609         struct btrfs_path *path;
4610         struct btrfs_root *root = BTRFS_I(dir)->root;
4611         int ret = 0;
4612
4613         path = btrfs_alloc_path();
4614         if (!path)
4615                 return -ENOMEM;
4616
4617         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4618                                     namelen, 0);
4619         if (IS_ERR(di))
4620                 ret = PTR_ERR(di);
4621
4622         if (IS_ERR_OR_NULL(di))
4623                 goto out_err;
4624
4625         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4626 out:
4627         btrfs_free_path(path);
4628         return ret;
4629 out_err:
4630         location->objectid = 0;
4631         goto out;
4632 }
4633
4634 /*
4635  * when we hit a tree root in a directory, the btrfs part of the inode
4636  * needs to be changed to reflect the root directory of the tree root.  This
4637  * is kind of like crossing a mount point.
4638  */
4639 static int fixup_tree_root_location(struct btrfs_root *root,
4640                                     struct inode *dir,
4641                                     struct dentry *dentry,
4642                                     struct btrfs_key *location,
4643                                     struct btrfs_root **sub_root)
4644 {
4645         struct btrfs_path *path;
4646         struct btrfs_root *new_root;
4647         struct btrfs_root_ref *ref;
4648         struct extent_buffer *leaf;
4649         int ret;
4650         int err = 0;
4651
4652         path = btrfs_alloc_path();
4653         if (!path) {
4654                 err = -ENOMEM;
4655                 goto out;
4656         }
4657
4658         err = -ENOENT;
4659         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4660                                   BTRFS_I(dir)->root->root_key.objectid,
4661                                   location->objectid);
4662         if (ret) {
4663                 if (ret < 0)
4664                         err = ret;
4665                 goto out;
4666         }
4667
4668         leaf = path->nodes[0];
4669         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4670         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4671             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4672                 goto out;
4673
4674         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4675                                    (unsigned long)(ref + 1),
4676                                    dentry->d_name.len);
4677         if (ret)
4678                 goto out;
4679
4680         btrfs_release_path(path);
4681
4682         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4683         if (IS_ERR(new_root)) {
4684                 err = PTR_ERR(new_root);
4685                 goto out;
4686         }
4687
4688         *sub_root = new_root;
4689         location->objectid = btrfs_root_dirid(&new_root->root_item);
4690         location->type = BTRFS_INODE_ITEM_KEY;
4691         location->offset = 0;
4692         err = 0;
4693 out:
4694         btrfs_free_path(path);
4695         return err;
4696 }
4697
4698 static void inode_tree_add(struct inode *inode)
4699 {
4700         struct btrfs_root *root = BTRFS_I(inode)->root;
4701         struct btrfs_inode *entry;
4702         struct rb_node **p;
4703         struct rb_node *parent;
4704         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4705         u64 ino = btrfs_ino(inode);
4706
4707         if (inode_unhashed(inode))
4708                 return;
4709         parent = NULL;
4710         spin_lock(&root->inode_lock);
4711         p = &root->inode_tree.rb_node;
4712         while (*p) {
4713                 parent = *p;
4714                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4715
4716                 if (ino < btrfs_ino(&entry->vfs_inode))
4717                         p = &parent->rb_left;
4718                 else if (ino > btrfs_ino(&entry->vfs_inode))
4719                         p = &parent->rb_right;
4720                 else {
4721                         WARN_ON(!(entry->vfs_inode.i_state &
4722                                   (I_WILL_FREE | I_FREEING)));
4723                         rb_replace_node(parent, new, &root->inode_tree);
4724                         RB_CLEAR_NODE(parent);
4725                         spin_unlock(&root->inode_lock);
4726                         return;
4727                 }
4728         }
4729         rb_link_node(new, parent, p);
4730         rb_insert_color(new, &root->inode_tree);
4731         spin_unlock(&root->inode_lock);
4732 }
4733
4734 static void inode_tree_del(struct inode *inode)
4735 {
4736         struct btrfs_root *root = BTRFS_I(inode)->root;
4737         int empty = 0;
4738
4739         spin_lock(&root->inode_lock);
4740         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4741                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4742                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4743                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4744         }
4745         spin_unlock(&root->inode_lock);
4746
4747         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4748                 synchronize_srcu(&root->fs_info->subvol_srcu);
4749                 spin_lock(&root->inode_lock);
4750                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4751                 spin_unlock(&root->inode_lock);
4752                 if (empty)
4753                         btrfs_add_dead_root(root);
4754         }
4755 }
4756
4757 void btrfs_invalidate_inodes(struct btrfs_root *root)
4758 {
4759         struct rb_node *node;
4760         struct rb_node *prev;
4761         struct btrfs_inode *entry;
4762         struct inode *inode;
4763         u64 objectid = 0;
4764
4765         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4766
4767         spin_lock(&root->inode_lock);
4768 again:
4769         node = root->inode_tree.rb_node;
4770         prev = NULL;
4771         while (node) {
4772                 prev = node;
4773                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4774
4775                 if (objectid < btrfs_ino(&entry->vfs_inode))
4776                         node = node->rb_left;
4777                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4778                         node = node->rb_right;
4779                 else
4780                         break;
4781         }
4782         if (!node) {
4783                 while (prev) {
4784                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4785                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4786                                 node = prev;
4787                                 break;
4788                         }
4789                         prev = rb_next(prev);
4790                 }
4791         }
4792         while (node) {
4793                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4794                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4795                 inode = igrab(&entry->vfs_inode);
4796                 if (inode) {
4797                         spin_unlock(&root->inode_lock);
4798                         if (atomic_read(&inode->i_count) > 1)
4799                                 d_prune_aliases(inode);
4800                         /*
4801                          * btrfs_drop_inode will have it removed from
4802                          * the inode cache when its usage count
4803                          * hits zero.
4804                          */
4805                         iput(inode);
4806                         cond_resched();
4807                         spin_lock(&root->inode_lock);
4808                         goto again;
4809                 }
4810
4811                 if (cond_resched_lock(&root->inode_lock))
4812                         goto again;
4813
4814                 node = rb_next(node);
4815         }
4816         spin_unlock(&root->inode_lock);
4817 }
4818
4819 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4820 {
4821         struct btrfs_iget_args *args = p;
4822         inode->i_ino = args->ino;
4823         BTRFS_I(inode)->root = args->root;
4824         return 0;
4825 }
4826
4827 static int btrfs_find_actor(struct inode *inode, void *opaque)
4828 {
4829         struct btrfs_iget_args *args = opaque;
4830         return args->ino == btrfs_ino(inode) &&
4831                 args->root == BTRFS_I(inode)->root;
4832 }
4833
4834 static struct inode *btrfs_iget_locked(struct super_block *s,
4835                                        u64 objectid,
4836                                        struct btrfs_root *root)
4837 {
4838         struct inode *inode;
4839         struct btrfs_iget_args args;
4840         unsigned long hashval = btrfs_inode_hash(objectid, root);
4841
4842         args.ino = objectid;
4843         args.root = root;
4844
4845         inode = iget5_locked(s, hashval, 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 *file, struct dir_context *ctx)
4996 {
4997         struct inode *inode = file_inode(file);
4998         struct btrfs_root *root = BTRFS_I(inode)->root;
4999         struct btrfs_item *item;
5000         struct btrfs_dir_item *di;
5001         struct btrfs_key key;
5002         struct btrfs_key found_key;
5003         struct btrfs_path *path;
5004         struct list_head ins_list;
5005         struct list_head del_list;
5006         int ret;
5007         struct extent_buffer *leaf;
5008         int slot;
5009         unsigned char d_type;
5010         int over = 0;
5011         u32 di_cur;
5012         u32 di_total;
5013         u32 di_len;
5014         int key_type = BTRFS_DIR_INDEX_KEY;
5015         char tmp_name[32];
5016         char *name_ptr;
5017         int name_len;
5018         int is_curr = 0;        /* ctx->pos points to the current index? */
5019
5020         /* FIXME, use a real flag for deciding about the key type */
5021         if (root->fs_info->tree_root == root)
5022                 key_type = BTRFS_DIR_ITEM_KEY;
5023
5024         if (!dir_emit_dots(file, ctx))
5025                 return 0;
5026
5027         path = btrfs_alloc_path();
5028         if (!path)
5029                 return -ENOMEM;
5030
5031         path->reada = 1;
5032
5033         if (key_type == BTRFS_DIR_INDEX_KEY) {
5034                 INIT_LIST_HEAD(&ins_list);
5035                 INIT_LIST_HEAD(&del_list);
5036                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5037         }
5038
5039         btrfs_set_key_type(&key, key_type);
5040         key.offset = ctx->pos;
5041         key.objectid = btrfs_ino(inode);
5042
5043         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5044         if (ret < 0)
5045                 goto err;
5046
5047         while (1) {
5048                 leaf = path->nodes[0];
5049                 slot = path->slots[0];
5050                 if (slot >= btrfs_header_nritems(leaf)) {
5051                         ret = btrfs_next_leaf(root, path);
5052                         if (ret < 0)
5053                                 goto err;
5054                         else if (ret > 0)
5055                                 break;
5056                         continue;
5057                 }
5058
5059                 item = btrfs_item_nr(slot);
5060                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5061
5062                 if (found_key.objectid != key.objectid)
5063                         break;
5064                 if (btrfs_key_type(&found_key) != key_type)
5065                         break;
5066                 if (found_key.offset < ctx->pos)
5067                         goto next;
5068                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5069                     btrfs_should_delete_dir_index(&del_list,
5070                                                   found_key.offset))
5071                         goto next;
5072
5073                 ctx->pos = found_key.offset;
5074                 is_curr = 1;
5075
5076                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5077                 di_cur = 0;
5078                 di_total = btrfs_item_size(leaf, item);
5079
5080                 while (di_cur < di_total) {
5081                         struct btrfs_key location;
5082
5083                         if (verify_dir_item(root, leaf, di))
5084                                 break;
5085
5086                         name_len = btrfs_dir_name_len(leaf, di);
5087                         if (name_len <= sizeof(tmp_name)) {
5088                                 name_ptr = tmp_name;
5089                         } else {
5090                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5091                                 if (!name_ptr) {
5092                                         ret = -ENOMEM;
5093                                         goto err;
5094                                 }
5095                         }
5096                         read_extent_buffer(leaf, name_ptr,
5097                                            (unsigned long)(di + 1), name_len);
5098
5099                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5100                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5101
5102
5103                         /* is this a reference to our own snapshot? If so
5104                          * skip it.
5105                          *
5106                          * In contrast to old kernels, we insert the snapshot's
5107                          * dir item and dir index after it has been created, so
5108                          * we won't find a reference to our own snapshot. We
5109                          * still keep the following code for backward
5110                          * compatibility.
5111                          */
5112                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5113                             location.objectid == root->root_key.objectid) {
5114                                 over = 0;
5115                                 goto skip;
5116                         }
5117                         over = !dir_emit(ctx, name_ptr, name_len,
5118                                        location.objectid, d_type);
5119
5120 skip:
5121                         if (name_ptr != tmp_name)
5122                                 kfree(name_ptr);
5123
5124                         if (over)
5125                                 goto nopos;
5126                         di_len = btrfs_dir_name_len(leaf, di) +
5127                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5128                         di_cur += di_len;
5129                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5130                 }
5131 next:
5132                 path->slots[0]++;
5133         }
5134
5135         if (key_type == BTRFS_DIR_INDEX_KEY) {
5136                 if (is_curr)
5137                         ctx->pos++;
5138                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5139                 if (ret)
5140                         goto nopos;
5141         }
5142
5143         /* Reached end of directory/root. Bump pos past the last item. */
5144         ctx->pos++;
5145
5146         /*
5147          * Stop new entries from being returned after we return the last
5148          * entry.
5149          *
5150          * New directory entries are assigned a strictly increasing
5151          * offset.  This means that new entries created during readdir
5152          * are *guaranteed* to be seen in the future by that readdir.
5153          * This has broken buggy programs which operate on names as
5154          * they're returned by readdir.  Until we re-use freed offsets
5155          * we have this hack to stop new entries from being returned
5156          * under the assumption that they'll never reach this huge
5157          * offset.
5158          *
5159          * This is being careful not to overflow 32bit loff_t unless the
5160          * last entry requires it because doing so has broken 32bit apps
5161          * in the past.
5162          */
5163         if (key_type == BTRFS_DIR_INDEX_KEY) {
5164                 if (ctx->pos >= INT_MAX)
5165                         ctx->pos = LLONG_MAX;
5166                 else
5167                         ctx->pos = INT_MAX;
5168         }
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         btrfs_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(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                 /*
5986                  * If we backup past the first extent we want to move forward
5987                  * and see if there is an extent in front of us, otherwise we'll
5988                  * say there is a hole for our whole search range which can
5989                  * cause problems.
5990                  */
5991                 extent_end = start;
5992                 goto next;
5993         }
5994
5995         found_type = btrfs_file_extent_type(leaf, item);
5996         extent_start = found_key.offset;
5997         compress_type = btrfs_file_extent_compression(leaf, item);
5998         if (found_type == BTRFS_FILE_EXTENT_REG ||
5999             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6000                 extent_end = extent_start +
6001                        btrfs_file_extent_num_bytes(leaf, item);
6002         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6003                 size_t size;
6004                 size = btrfs_file_extent_inline_len(leaf, item);
6005                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6006         }
6007 next:
6008         if (start >= extent_end) {
6009                 path->slots[0]++;
6010                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6011                         ret = btrfs_next_leaf(root, path);
6012                         if (ret < 0) {
6013                                 err = ret;
6014                                 goto out;
6015                         }
6016                         if (ret > 0)
6017                                 goto not_found;
6018                         leaf = path->nodes[0];
6019                 }
6020                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6021                 if (found_key.objectid != objectid ||
6022                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6023                         goto not_found;
6024                 if (start + len <= found_key.offset)
6025                         goto not_found;
6026                 em->start = start;
6027                 em->orig_start = start;
6028                 em->len = found_key.offset - start;
6029                 goto not_found_em;
6030         }
6031
6032         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6033         if (found_type == BTRFS_FILE_EXTENT_REG ||
6034             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6035                 em->start = extent_start;
6036                 em->len = extent_end - extent_start;
6037                 em->orig_start = extent_start -
6038                                  btrfs_file_extent_offset(leaf, item);
6039                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6040                                                                       item);
6041                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6042                 if (bytenr == 0) {
6043                         em->block_start = EXTENT_MAP_HOLE;
6044                         goto insert;
6045                 }
6046                 if (compress_type != BTRFS_COMPRESS_NONE) {
6047                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6048                         em->compress_type = compress_type;
6049                         em->block_start = bytenr;
6050                         em->block_len = em->orig_block_len;
6051                 } else {
6052                         bytenr += btrfs_file_extent_offset(leaf, item);
6053                         em->block_start = bytenr;
6054                         em->block_len = em->len;
6055                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6056                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6057                 }
6058                 goto insert;
6059         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6060                 unsigned long ptr;
6061                 char *map;
6062                 size_t size;
6063                 size_t extent_offset;
6064                 size_t copy_size;
6065
6066                 em->block_start = EXTENT_MAP_INLINE;
6067                 if (!page || create) {
6068                         em->start = extent_start;
6069                         em->len = extent_end - extent_start;
6070                         goto out;
6071                 }
6072
6073                 size = btrfs_file_extent_inline_len(leaf, item);
6074                 extent_offset = page_offset(page) + pg_offset - extent_start;
6075                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6076                                 size - extent_offset);
6077                 em->start = extent_start + extent_offset;
6078                 em->len = ALIGN(copy_size, root->sectorsize);
6079                 em->orig_block_len = em->len;
6080                 em->orig_start = em->start;
6081                 if (compress_type) {
6082                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6083                         em->compress_type = compress_type;
6084                 }
6085                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6086                 if (create == 0 && !PageUptodate(page)) {
6087                         if (btrfs_file_extent_compression(leaf, item) !=
6088                             BTRFS_COMPRESS_NONE) {
6089                                 ret = uncompress_inline(path, inode, page,
6090                                                         pg_offset,
6091                                                         extent_offset, item);
6092                                 BUG_ON(ret); /* -ENOMEM */
6093                         } else {
6094                                 map = kmap(page);
6095                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6096                                                    copy_size);
6097                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6098                                         memset(map + pg_offset + copy_size, 0,
6099                                                PAGE_CACHE_SIZE - pg_offset -
6100                                                copy_size);
6101                                 }
6102                                 kunmap(page);
6103                         }
6104                         flush_dcache_page(page);
6105                 } else if (create && PageUptodate(page)) {
6106                         BUG();
6107                         if (!trans) {
6108                                 kunmap(page);
6109                                 free_extent_map(em);
6110                                 em = NULL;
6111
6112                                 btrfs_release_path(path);
6113                                 trans = btrfs_join_transaction(root);
6114
6115                                 if (IS_ERR(trans))
6116                                         return ERR_CAST(trans);
6117                                 goto again;
6118                         }
6119                         map = kmap(page);
6120                         write_extent_buffer(leaf, map + pg_offset, ptr,
6121                                             copy_size);
6122                         kunmap(page);
6123                         btrfs_mark_buffer_dirty(leaf);
6124                 }
6125                 set_extent_uptodate(io_tree, em->start,
6126                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6127                 goto insert;
6128         } else {
6129                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6130         }
6131 not_found:
6132         em->start = start;
6133         em->orig_start = start;
6134         em->len = len;
6135 not_found_em:
6136         em->block_start = EXTENT_MAP_HOLE;
6137         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6138 insert:
6139         btrfs_release_path(path);
6140         if (em->start > start || extent_map_end(em) <= start) {
6141                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6142                         em->start, em->len, start, len);
6143                 err = -EIO;
6144                 goto out;
6145         }
6146
6147         err = 0;
6148         write_lock(&em_tree->lock);
6149         ret = add_extent_mapping(em_tree, em, 0);
6150         /* it is possible that someone inserted the extent into the tree
6151          * while we had the lock dropped.  It is also possible that
6152          * an overlapping map exists in the tree
6153          */
6154         if (ret == -EEXIST) {
6155                 struct extent_map *existing;
6156
6157                 ret = 0;
6158
6159                 existing = lookup_extent_mapping(em_tree, start, len);
6160                 if (existing && (existing->start > start ||
6161                     existing->start + existing->len <= start)) {
6162                         free_extent_map(existing);
6163                         existing = NULL;
6164                 }
6165                 if (!existing) {
6166                         existing = lookup_extent_mapping(em_tree, em->start,
6167                                                          em->len);
6168                         if (existing) {
6169                                 err = merge_extent_mapping(em_tree, existing,
6170                                                            em, start,
6171                                                            root->sectorsize);
6172                                 free_extent_map(existing);
6173                                 if (err) {
6174                                         free_extent_map(em);
6175                                         em = NULL;
6176                                 }
6177                         } else {
6178                                 err = -EIO;
6179                                 free_extent_map(em);
6180                                 em = NULL;
6181                         }
6182                 } else {
6183                         free_extent_map(em);
6184                         em = existing;
6185                         err = 0;
6186                 }
6187         }
6188         write_unlock(&em_tree->lock);
6189 out:
6190
6191         if (em)
6192                 trace_btrfs_get_extent(root, em);
6193
6194         if (path)
6195                 btrfs_free_path(path);
6196         if (trans) {
6197                 ret = btrfs_end_transaction(trans, root);
6198                 if (!err)
6199                         err = ret;
6200         }
6201         if (err) {
6202                 free_extent_map(em);
6203                 return ERR_PTR(err);
6204         }
6205         BUG_ON(!em); /* Error is always set */
6206         return em;
6207 }
6208
6209 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6210                                            size_t pg_offset, u64 start, u64 len,
6211                                            int create)
6212 {
6213         struct extent_map *em;
6214         struct extent_map *hole_em = NULL;
6215         u64 range_start = start;
6216         u64 end;
6217         u64 found;
6218         u64 found_end;
6219         int err = 0;
6220
6221         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6222         if (IS_ERR(em))
6223                 return em;
6224         if (em) {
6225                 /*
6226                  * if our em maps to
6227                  * -  a hole or
6228                  * -  a pre-alloc extent,
6229                  * there might actually be delalloc bytes behind it.
6230                  */
6231                 if (em->block_start != EXTENT_MAP_HOLE &&
6232                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6233                         return em;
6234                 else
6235                         hole_em = em;
6236         }
6237
6238         /* check to see if we've wrapped (len == -1 or similar) */
6239         end = start + len;
6240         if (end < start)
6241                 end = (u64)-1;
6242         else
6243                 end -= 1;
6244
6245         em = NULL;
6246
6247         /* ok, we didn't find anything, lets look for delalloc */
6248         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6249                                  end, len, EXTENT_DELALLOC, 1);
6250         found_end = range_start + found;
6251         if (found_end < range_start)
6252                 found_end = (u64)-1;
6253
6254         /*
6255          * we didn't find anything useful, return
6256          * the original results from get_extent()
6257          */
6258         if (range_start > end || found_end <= start) {
6259                 em = hole_em;
6260                 hole_em = NULL;
6261                 goto out;
6262         }
6263
6264         /* adjust the range_start to make sure it doesn't
6265          * go backwards from the start they passed in
6266          */
6267         range_start = max(start,range_start);
6268         found = found_end - range_start;
6269
6270         if (found > 0) {
6271                 u64 hole_start = start;
6272                 u64 hole_len = len;
6273
6274                 em = alloc_extent_map();
6275                 if (!em) {
6276                         err = -ENOMEM;
6277                         goto out;
6278                 }
6279                 /*
6280                  * when btrfs_get_extent can't find anything it
6281                  * returns one huge hole
6282                  *
6283                  * make sure what it found really fits our range, and
6284                  * adjust to make sure it is based on the start from
6285                  * the caller
6286                  */
6287                 if (hole_em) {
6288                         u64 calc_end = extent_map_end(hole_em);
6289
6290                         if (calc_end <= start || (hole_em->start > end)) {
6291                                 free_extent_map(hole_em);
6292                                 hole_em = NULL;
6293                         } else {
6294                                 hole_start = max(hole_em->start, start);
6295                                 hole_len = calc_end - hole_start;
6296                         }
6297                 }
6298                 em->bdev = NULL;
6299                 if (hole_em && range_start > hole_start) {
6300                         /* our hole starts before our delalloc, so we
6301                          * have to return just the parts of the hole
6302                          * that go until  the delalloc starts
6303                          */
6304                         em->len = min(hole_len,
6305                                       range_start - hole_start);
6306                         em->start = hole_start;
6307                         em->orig_start = hole_start;
6308                         /*
6309                          * don't adjust block start at all,
6310                          * it is fixed at EXTENT_MAP_HOLE
6311                          */
6312                         em->block_start = hole_em->block_start;
6313                         em->block_len = hole_len;
6314                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6315                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6316                 } else {
6317                         em->start = range_start;
6318                         em->len = found;
6319                         em->orig_start = range_start;
6320                         em->block_start = EXTENT_MAP_DELALLOC;
6321                         em->block_len = found;
6322                 }
6323         } else if (hole_em) {
6324                 return hole_em;
6325         }
6326 out:
6327
6328         free_extent_map(hole_em);
6329         if (err) {
6330                 free_extent_map(em);
6331                 return ERR_PTR(err);
6332         }
6333         return em;
6334 }
6335
6336 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6337                                                   u64 start, u64 len)
6338 {
6339         struct btrfs_root *root = BTRFS_I(inode)->root;
6340         struct extent_map *em;
6341         struct btrfs_key ins;
6342         u64 alloc_hint;
6343         int ret;
6344
6345         alloc_hint = get_extent_allocation_hint(inode, start, len);
6346         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6347                                    alloc_hint, &ins, 1);
6348         if (ret)
6349                 return ERR_PTR(ret);
6350
6351         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6352                               ins.offset, ins.offset, ins.offset, 0);
6353         if (IS_ERR(em)) {
6354                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6355                 return em;
6356         }
6357
6358         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6359                                            ins.offset, ins.offset, 0);
6360         if (ret) {
6361                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6362                 free_extent_map(em);
6363                 return ERR_PTR(ret);
6364         }
6365
6366         return em;
6367 }
6368
6369 /*
6370  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6371  * block must be cow'd
6372  */
6373 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6374                               u64 *orig_start, u64 *orig_block_len,
6375                               u64 *ram_bytes)
6376 {
6377         struct btrfs_trans_handle *trans;
6378         struct btrfs_path *path;
6379         int ret;
6380         struct extent_buffer *leaf;
6381         struct btrfs_root *root = BTRFS_I(inode)->root;
6382         struct btrfs_file_extent_item *fi;
6383         struct btrfs_key key;
6384         u64 disk_bytenr;
6385         u64 backref_offset;
6386         u64 extent_end;
6387         u64 num_bytes;
6388         int slot;
6389         int found_type;
6390         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6391         path = btrfs_alloc_path();
6392         if (!path)
6393                 return -ENOMEM;
6394
6395         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6396                                        offset, 0);
6397         if (ret < 0)
6398                 goto out;
6399
6400         slot = path->slots[0];
6401         if (ret == 1) {
6402                 if (slot == 0) {
6403                         /* can't find the item, must cow */
6404                         ret = 0;
6405                         goto out;
6406                 }
6407                 slot--;
6408         }
6409         ret = 0;
6410         leaf = path->nodes[0];
6411         btrfs_item_key_to_cpu(leaf, &key, slot);
6412         if (key.objectid != btrfs_ino(inode) ||
6413             key.type != BTRFS_EXTENT_DATA_KEY) {
6414                 /* not our file or wrong item type, must cow */
6415                 goto out;
6416         }
6417
6418         if (key.offset > offset) {
6419                 /* Wrong offset, must cow */
6420                 goto out;
6421         }
6422
6423         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6424         found_type = btrfs_file_extent_type(leaf, fi);
6425         if (found_type != BTRFS_FILE_EXTENT_REG &&
6426             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6427                 /* not a regular extent, must cow */
6428                 goto out;
6429         }
6430
6431         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6432                 goto out;
6433
6434         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6435         if (disk_bytenr == 0)
6436                 goto out;
6437
6438         if (btrfs_file_extent_compression(leaf, fi) ||
6439             btrfs_file_extent_encryption(leaf, fi) ||
6440             btrfs_file_extent_other_encoding(leaf, fi))
6441                 goto out;
6442
6443         backref_offset = btrfs_file_extent_offset(leaf, fi);
6444
6445         if (orig_start) {
6446                 *orig_start = key.offset - backref_offset;
6447                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6448                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6449         }
6450
6451         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6452
6453         if (btrfs_extent_readonly(root, disk_bytenr))
6454                 goto out;
6455         btrfs_release_path(path);
6456
6457         /*
6458          * look for other files referencing this extent, if we
6459          * find any we must cow
6460          */
6461         trans = btrfs_join_transaction(root);
6462         if (IS_ERR(trans)) {
6463                 ret = 0;
6464                 goto out;
6465         }
6466
6467         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6468                                     key.offset - backref_offset, disk_bytenr);
6469         btrfs_end_transaction(trans, root);
6470         if (ret) {
6471                 ret = 0;
6472                 goto out;
6473         }
6474
6475         /*
6476          * adjust disk_bytenr and num_bytes to cover just the bytes
6477          * in this extent we are about to write.  If there
6478          * are any csums in that range we have to cow in order
6479          * to keep the csums correct
6480          */
6481         disk_bytenr += backref_offset;
6482         disk_bytenr += offset - key.offset;
6483         num_bytes = min(offset + *len, extent_end) - offset;
6484         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6485                                 goto out;
6486         /*
6487          * all of the above have passed, it is safe to overwrite this extent
6488          * without cow
6489          */
6490         *len = num_bytes;
6491         ret = 1;
6492 out:
6493         btrfs_free_path(path);
6494         return ret;
6495 }
6496
6497 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6498                               struct extent_state **cached_state, int writing)
6499 {
6500         struct btrfs_ordered_extent *ordered;
6501         int ret = 0;
6502
6503         while (1) {
6504                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6505                                  0, cached_state);
6506                 /*
6507                  * We're concerned with the entire range that we're going to be
6508                  * doing DIO to, so we need to make sure theres no ordered
6509                  * extents in this range.
6510                  */
6511                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6512                                                      lockend - lockstart + 1);
6513
6514                 /*
6515                  * We need to make sure there are no buffered pages in this
6516                  * range either, we could have raced between the invalidate in
6517                  * generic_file_direct_write and locking the extent.  The
6518                  * invalidate needs to happen so that reads after a write do not
6519                  * get stale data.
6520                  */
6521                 if (!ordered && (!writing ||
6522                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6523                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6524                                     *cached_state)))
6525                         break;
6526
6527                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6528                                      cached_state, GFP_NOFS);
6529
6530                 if (ordered) {
6531                         btrfs_start_ordered_extent(inode, ordered, 1);
6532                         btrfs_put_ordered_extent(ordered);
6533                 } else {
6534                         /* Screw you mmap */
6535                         ret = filemap_write_and_wait_range(inode->i_mapping,
6536                                                            lockstart,
6537                                                            lockend);
6538                         if (ret)
6539                                 break;
6540
6541                         /*
6542                          * If we found a page that couldn't be invalidated just
6543                          * fall back to buffered.
6544                          */
6545                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6546                                         lockstart >> PAGE_CACHE_SHIFT,
6547                                         lockend >> PAGE_CACHE_SHIFT);
6548                         if (ret)
6549                                 break;
6550                 }
6551
6552                 cond_resched();
6553         }
6554
6555         return ret;
6556 }
6557
6558 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6559                                            u64 len, u64 orig_start,
6560                                            u64 block_start, u64 block_len,
6561                                            u64 orig_block_len, u64 ram_bytes,
6562                                            int type)
6563 {
6564         struct extent_map_tree *em_tree;
6565         struct extent_map *em;
6566         struct btrfs_root *root = BTRFS_I(inode)->root;
6567         int ret;
6568
6569         em_tree = &BTRFS_I(inode)->extent_tree;
6570         em = alloc_extent_map();
6571         if (!em)
6572                 return ERR_PTR(-ENOMEM);
6573
6574         em->start = start;
6575         em->orig_start = orig_start;
6576         em->mod_start = start;
6577         em->mod_len = len;
6578         em->len = len;
6579         em->block_len = block_len;
6580         em->block_start = block_start;
6581         em->bdev = root->fs_info->fs_devices->latest_bdev;
6582         em->orig_block_len = orig_block_len;
6583         em->ram_bytes = ram_bytes;
6584         em->generation = -1;
6585         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6586         if (type == BTRFS_ORDERED_PREALLOC)
6587                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6588
6589         do {
6590                 btrfs_drop_extent_cache(inode, em->start,
6591                                 em->start + em->len - 1, 0);
6592                 write_lock(&em_tree->lock);
6593                 ret = add_extent_mapping(em_tree, em, 1);
6594                 write_unlock(&em_tree->lock);
6595         } while (ret == -EEXIST);
6596
6597         if (ret) {
6598                 free_extent_map(em);
6599                 return ERR_PTR(ret);
6600         }
6601
6602         return em;
6603 }
6604
6605
6606 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6607                                    struct buffer_head *bh_result, int create)
6608 {
6609         struct extent_map *em;
6610         struct btrfs_root *root = BTRFS_I(inode)->root;
6611         struct extent_state *cached_state = NULL;
6612         u64 start = iblock << inode->i_blkbits;
6613         u64 lockstart, lockend;
6614         u64 len = bh_result->b_size;
6615         int unlock_bits = EXTENT_LOCKED;
6616         int ret = 0;
6617
6618         if (create)
6619                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6620         else
6621                 len = min_t(u64, len, root->sectorsize);
6622
6623         lockstart = start;
6624         lockend = start + len - 1;
6625
6626         /*
6627          * If this errors out it's because we couldn't invalidate pagecache for
6628          * this range and we need to fallback to buffered.
6629          */
6630         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6631                 return -ENOTBLK;
6632
6633         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6634         if (IS_ERR(em)) {
6635                 ret = PTR_ERR(em);
6636                 goto unlock_err;
6637         }
6638
6639         /*
6640          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6641          * io.  INLINE is special, and we could probably kludge it in here, but
6642          * it's still buffered so for safety lets just fall back to the generic
6643          * buffered path.
6644          *
6645          * For COMPRESSED we _have_ to read the entire extent in so we can
6646          * decompress it, so there will be buffering required no matter what we
6647          * do, so go ahead and fallback to buffered.
6648          *
6649          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6650          * to buffered IO.  Don't blame me, this is the price we pay for using
6651          * the generic code.
6652          */
6653         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6654             em->block_start == EXTENT_MAP_INLINE) {
6655                 free_extent_map(em);
6656                 ret = -ENOTBLK;
6657                 goto unlock_err;
6658         }
6659
6660         /* Just a good old fashioned hole, return */
6661         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6662                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6663                 free_extent_map(em);
6664                 goto unlock_err;
6665         }
6666
6667         /*
6668          * We don't allocate a new extent in the following cases
6669          *
6670          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6671          * existing extent.
6672          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6673          * just use the extent.
6674          *
6675          */
6676         if (!create) {
6677                 len = min(len, em->len - (start - em->start));
6678                 lockstart = start + len;
6679                 goto unlock;
6680         }
6681
6682         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6683             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6684              em->block_start != EXTENT_MAP_HOLE)) {
6685                 int type;
6686                 int ret;
6687                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6688
6689                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6690                         type = BTRFS_ORDERED_PREALLOC;
6691                 else
6692                         type = BTRFS_ORDERED_NOCOW;
6693                 len = min(len, em->len - (start - em->start));
6694                 block_start = em->block_start + (start - em->start);
6695
6696                 if (can_nocow_extent(inode, start, &len, &orig_start,
6697                                      &orig_block_len, &ram_bytes) == 1) {
6698                         if (type == BTRFS_ORDERED_PREALLOC) {
6699                                 free_extent_map(em);
6700                                 em = create_pinned_em(inode, start, len,
6701                                                        orig_start,
6702                                                        block_start, len,
6703                                                        orig_block_len,
6704                                                        ram_bytes, type);
6705                                 if (IS_ERR(em))
6706                                         goto unlock_err;
6707                         }
6708
6709                         ret = btrfs_add_ordered_extent_dio(inode, start,
6710                                            block_start, len, len, type);
6711                         if (ret) {
6712                                 free_extent_map(em);
6713                                 goto unlock_err;
6714                         }
6715                         goto unlock;
6716                 }
6717         }
6718
6719         /*
6720          * this will cow the extent, reset the len in case we changed
6721          * it above
6722          */
6723         len = bh_result->b_size;
6724         free_extent_map(em);
6725         em = btrfs_new_extent_direct(inode, start, len);
6726         if (IS_ERR(em)) {
6727                 ret = PTR_ERR(em);
6728                 goto unlock_err;
6729         }
6730         len = min(len, em->len - (start - em->start));
6731 unlock:
6732         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6733                 inode->i_blkbits;
6734         bh_result->b_size = len;
6735         bh_result->b_bdev = em->bdev;
6736         set_buffer_mapped(bh_result);
6737         if (create) {
6738                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6739                         set_buffer_new(bh_result);
6740
6741                 /*
6742                  * Need to update the i_size under the extent lock so buffered
6743                  * readers will get the updated i_size when we unlock.
6744                  */
6745                 if (start + len > i_size_read(inode))
6746                         i_size_write(inode, start + len);
6747
6748                 spin_lock(&BTRFS_I(inode)->lock);
6749                 BTRFS_I(inode)->outstanding_extents++;
6750                 spin_unlock(&BTRFS_I(inode)->lock);
6751
6752                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6753                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6754                                      &cached_state, GFP_NOFS);
6755                 BUG_ON(ret);
6756         }
6757
6758         /*
6759          * In the case of write we need to clear and unlock the entire range,
6760          * in the case of read we need to unlock only the end area that we
6761          * aren't using if there is any left over space.
6762          */
6763         if (lockstart < lockend) {
6764                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6765                                  lockend, unlock_bits, 1, 0,
6766                                  &cached_state, GFP_NOFS);
6767         } else {
6768                 free_extent_state(cached_state);
6769         }
6770
6771         free_extent_map(em);
6772
6773         return 0;
6774
6775 unlock_err:
6776         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6777                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6778         return ret;
6779 }
6780
6781 static void btrfs_endio_direct_read(struct bio *bio, int err)
6782 {
6783         struct btrfs_dio_private *dip = bio->bi_private;
6784         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6785         struct bio_vec *bvec = bio->bi_io_vec;
6786         struct inode *inode = dip->inode;
6787         struct btrfs_root *root = BTRFS_I(inode)->root;
6788         struct bio *dio_bio;
6789         u32 *csums = (u32 *)dip->csum;
6790         int index = 0;
6791         u64 start;
6792
6793         start = dip->logical_offset;
6794         do {
6795                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6796                         struct page *page = bvec->bv_page;
6797                         char *kaddr;
6798                         u32 csum = ~(u32)0;
6799                         unsigned long flags;
6800
6801                         local_irq_save(flags);
6802                         kaddr = kmap_atomic(page);
6803                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6804                                                csum, bvec->bv_len);
6805                         btrfs_csum_final(csum, (char *)&csum);
6806                         kunmap_atomic(kaddr);
6807                         local_irq_restore(flags);
6808
6809                         flush_dcache_page(bvec->bv_page);
6810                         if (csum != csums[index]) {
6811                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6812                                           btrfs_ino(inode), start, csum,
6813                                           csums[index]);
6814                                 err = -EIO;
6815                         }
6816                 }
6817
6818                 start += bvec->bv_len;
6819                 bvec++;
6820                 index++;
6821         } while (bvec <= bvec_end);
6822
6823         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6824                       dip->logical_offset + dip->bytes - 1);
6825         dio_bio = dip->dio_bio;
6826
6827         kfree(dip);
6828
6829         /* If we had a csum failure make sure to clear the uptodate flag */
6830         if (err)
6831                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6832         dio_end_io(dio_bio, err);
6833         bio_put(bio);
6834 }
6835
6836 static void btrfs_endio_direct_write(struct bio *bio, int err)
6837 {
6838         struct btrfs_dio_private *dip = bio->bi_private;
6839         struct inode *inode = dip->inode;
6840         struct btrfs_root *root = BTRFS_I(inode)->root;
6841         struct btrfs_ordered_extent *ordered = NULL;
6842         u64 ordered_offset = dip->logical_offset;
6843         u64 ordered_bytes = dip->bytes;
6844         struct bio *dio_bio;
6845         int ret;
6846
6847         if (err)
6848                 goto out_done;
6849 again:
6850         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6851                                                    &ordered_offset,
6852                                                    ordered_bytes, !err);
6853         if (!ret)
6854                 goto out_test;
6855
6856         ordered->work.func = finish_ordered_fn;
6857         ordered->work.flags = 0;
6858         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6859                            &ordered->work);
6860 out_test:
6861         /*
6862          * our bio might span multiple ordered extents.  If we haven't
6863          * completed the accounting for the whole dio, go back and try again
6864          */
6865         if (ordered_offset < dip->logical_offset + dip->bytes) {
6866                 ordered_bytes = dip->logical_offset + dip->bytes -
6867                         ordered_offset;
6868                 ordered = NULL;
6869                 goto again;
6870         }
6871 out_done:
6872         dio_bio = dip->dio_bio;
6873
6874         kfree(dip);
6875
6876         /* If we had an error make sure to clear the uptodate flag */
6877         if (err)
6878                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6879         dio_end_io(dio_bio, err);
6880         bio_put(bio);
6881 }
6882
6883 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6884                                     struct bio *bio, int mirror_num,
6885                                     unsigned long bio_flags, u64 offset)
6886 {
6887         int ret;
6888         struct btrfs_root *root = BTRFS_I(inode)->root;
6889         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6890         BUG_ON(ret); /* -ENOMEM */
6891         return 0;
6892 }
6893
6894 static void btrfs_end_dio_bio(struct bio *bio, int err)
6895 {
6896         struct btrfs_dio_private *dip = bio->bi_private;
6897
6898         if (err) {
6899                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6900                       "sector %#Lx len %u err no %d\n",
6901                       btrfs_ino(dip->inode), bio->bi_rw,
6902                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6903                 dip->errors = 1;
6904
6905                 /*
6906                  * before atomic variable goto zero, we must make sure
6907                  * dip->errors is perceived to be set.
6908                  */
6909                 smp_mb__before_atomic_dec();
6910         }
6911
6912         /* if there are more bios still pending for this dio, just exit */
6913         if (!atomic_dec_and_test(&dip->pending_bios))
6914                 goto out;
6915
6916         if (dip->errors) {
6917                 bio_io_error(dip->orig_bio);
6918         } else {
6919                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6920                 bio_endio(dip->orig_bio, 0);
6921         }
6922 out:
6923         bio_put(bio);
6924 }
6925
6926 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6927                                        u64 first_sector, gfp_t gfp_flags)
6928 {
6929         int nr_vecs = bio_get_nr_vecs(bdev);
6930         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6931 }
6932
6933 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6934                                          int rw, u64 file_offset, int skip_sum,
6935                                          int async_submit)
6936 {
6937         struct btrfs_dio_private *dip = bio->bi_private;
6938         int write = rw & REQ_WRITE;
6939         struct btrfs_root *root = BTRFS_I(inode)->root;
6940         int ret;
6941
6942         if (async_submit)
6943                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6944
6945         bio_get(bio);
6946
6947         if (!write) {
6948                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6949                 if (ret)
6950                         goto err;
6951         }
6952
6953         if (skip_sum)
6954                 goto map;
6955
6956         if (write && async_submit) {
6957                 ret = btrfs_wq_submit_bio(root->fs_info,
6958                                    inode, rw, bio, 0, 0,
6959                                    file_offset,
6960                                    __btrfs_submit_bio_start_direct_io,
6961                                    __btrfs_submit_bio_done);
6962                 goto err;
6963         } else if (write) {
6964                 /*
6965                  * If we aren't doing async submit, calculate the csum of the
6966                  * bio now.
6967                  */
6968                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6969                 if (ret)
6970                         goto err;
6971         } else if (!skip_sum) {
6972                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
6973                                                 file_offset);
6974                 if (ret)
6975                         goto err;
6976         }
6977
6978 map:
6979         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6980 err:
6981         bio_put(bio);
6982         return ret;
6983 }
6984
6985 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6986                                     int skip_sum)
6987 {
6988         struct inode *inode = dip->inode;
6989         struct btrfs_root *root = BTRFS_I(inode)->root;
6990         struct bio *bio;
6991         struct bio *orig_bio = dip->orig_bio;
6992         struct bio_vec *bvec = orig_bio->bi_io_vec;
6993         u64 start_sector = orig_bio->bi_sector;
6994         u64 file_offset = dip->logical_offset;
6995         u64 submit_len = 0;
6996         u64 map_length;
6997         int nr_pages = 0;
6998         int ret = 0;
6999         int async_submit = 0;
7000
7001         map_length = orig_bio->bi_size;
7002         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7003                               &map_length, NULL, 0);
7004         if (ret) {
7005                 bio_put(orig_bio);
7006                 return -EIO;
7007         }
7008
7009         if (map_length >= orig_bio->bi_size) {
7010                 bio = orig_bio;
7011                 goto submit;
7012         }
7013
7014         /* async crcs make it difficult to collect full stripe writes. */
7015         if (btrfs_get_alloc_profile(root, 1) &
7016             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7017                 async_submit = 0;
7018         else
7019                 async_submit = 1;
7020
7021         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7022         if (!bio)
7023                 return -ENOMEM;
7024         bio->bi_private = dip;
7025         bio->bi_end_io = btrfs_end_dio_bio;
7026         atomic_inc(&dip->pending_bios);
7027
7028         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7029                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7030                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7031                                  bvec->bv_offset) < bvec->bv_len)) {
7032                         /*
7033                          * inc the count before we submit the bio so
7034                          * we know the end IO handler won't happen before
7035                          * we inc the count. Otherwise, the dip might get freed
7036                          * before we're done setting it up
7037                          */
7038                         atomic_inc(&dip->pending_bios);
7039                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7040                                                      file_offset, skip_sum,
7041                                                      async_submit);
7042                         if (ret) {
7043                                 bio_put(bio);
7044                                 atomic_dec(&dip->pending_bios);
7045                                 goto out_err;
7046                         }
7047
7048                         start_sector += submit_len >> 9;
7049                         file_offset += submit_len;
7050
7051                         submit_len = 0;
7052                         nr_pages = 0;
7053
7054                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7055                                                   start_sector, GFP_NOFS);
7056                         if (!bio)
7057                                 goto out_err;
7058                         bio->bi_private = dip;
7059                         bio->bi_end_io = btrfs_end_dio_bio;
7060
7061                         map_length = orig_bio->bi_size;
7062                         ret = btrfs_map_block(root->fs_info, rw,
7063                                               start_sector << 9,
7064                                               &map_length, NULL, 0);
7065                         if (ret) {
7066                                 bio_put(bio);
7067                                 goto out_err;
7068                         }
7069                 } else {
7070                         submit_len += bvec->bv_len;
7071                         nr_pages ++;
7072                         bvec++;
7073                 }
7074         }
7075
7076 submit:
7077         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7078                                      async_submit);
7079         if (!ret)
7080                 return 0;
7081
7082         bio_put(bio);
7083 out_err:
7084         dip->errors = 1;
7085         /*
7086          * before atomic variable goto zero, we must
7087          * make sure dip->errors is perceived to be set.
7088          */
7089         smp_mb__before_atomic_dec();
7090         if (atomic_dec_and_test(&dip->pending_bios))
7091                 bio_io_error(dip->orig_bio);
7092
7093         /* bio_end_io() will handle error, so we needn't return it */
7094         return 0;
7095 }
7096
7097 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7098                                 struct inode *inode, loff_t file_offset)
7099 {
7100         struct btrfs_root *root = BTRFS_I(inode)->root;
7101         struct btrfs_dio_private *dip;
7102         struct bio *io_bio;
7103         int skip_sum;
7104         int sum_len;
7105         int write = rw & REQ_WRITE;
7106         int ret = 0;
7107         u16 csum_size;
7108
7109         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7110
7111         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7112         if (!io_bio) {
7113                 ret = -ENOMEM;
7114                 goto free_ordered;
7115         }
7116
7117         if (!skip_sum && !write) {
7118                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7119                 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7120                 sum_len *= csum_size;
7121         } else {
7122                 sum_len = 0;
7123         }
7124
7125         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7126         if (!dip) {
7127                 ret = -ENOMEM;
7128                 goto free_io_bio;
7129         }
7130
7131         dip->private = dio_bio->bi_private;
7132         dip->inode = inode;
7133         dip->logical_offset = file_offset;
7134         dip->bytes = dio_bio->bi_size;
7135         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7136         io_bio->bi_private = dip;
7137         dip->errors = 0;
7138         dip->orig_bio = io_bio;
7139         dip->dio_bio = dio_bio;
7140         atomic_set(&dip->pending_bios, 0);
7141
7142         if (write)
7143                 io_bio->bi_end_io = btrfs_endio_direct_write;
7144         else
7145                 io_bio->bi_end_io = btrfs_endio_direct_read;
7146
7147         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7148         if (!ret)
7149                 return;
7150
7151 free_io_bio:
7152         bio_put(io_bio);
7153
7154 free_ordered:
7155         /*
7156          * If this is a write, we need to clean up the reserved space and kill
7157          * the ordered extent.
7158          */
7159         if (write) {
7160                 struct btrfs_ordered_extent *ordered;
7161                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7162                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7163                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7164                         btrfs_free_reserved_extent(root, ordered->start,
7165                                                    ordered->disk_len);
7166                 btrfs_put_ordered_extent(ordered);
7167                 btrfs_put_ordered_extent(ordered);
7168         }
7169         bio_endio(dio_bio, ret);
7170 }
7171
7172 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7173                         const struct iovec *iov, loff_t offset,
7174                         unsigned long nr_segs)
7175 {
7176         int seg;
7177         int i;
7178         size_t size;
7179         unsigned long addr;
7180         unsigned blocksize_mask = root->sectorsize - 1;
7181         ssize_t retval = -EINVAL;
7182         loff_t end = offset;
7183
7184         if (offset & blocksize_mask)
7185                 goto out;
7186
7187         /* Check the memory alignment.  Blocks cannot straddle pages */
7188         for (seg = 0; seg < nr_segs; seg++) {
7189                 addr = (unsigned long)iov[seg].iov_base;
7190                 size = iov[seg].iov_len;
7191                 end += size;
7192                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7193                         goto out;
7194
7195                 /* If this is a write we don't need to check anymore */
7196                 if (rw & WRITE)
7197                         continue;
7198
7199                 /*
7200                  * Check to make sure we don't have duplicate iov_base's in this
7201                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7202                  * when reading back.
7203                  */
7204                 for (i = seg + 1; i < nr_segs; i++) {
7205                         if (iov[seg].iov_base == iov[i].iov_base)
7206                                 goto out;
7207                 }
7208         }
7209         retval = 0;
7210 out:
7211         return retval;
7212 }
7213
7214 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7215                         const struct iovec *iov, loff_t offset,
7216                         unsigned long nr_segs)
7217 {
7218         struct file *file = iocb->ki_filp;
7219         struct inode *inode = file->f_mapping->host;
7220         size_t count = 0;
7221         int flags = 0;
7222         bool wakeup = true;
7223         bool relock = false;
7224         ssize_t ret;
7225
7226         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7227                             offset, nr_segs))
7228                 return 0;
7229
7230         atomic_inc(&inode->i_dio_count);
7231         smp_mb__after_atomic_inc();
7232
7233         /*
7234          * The generic stuff only does filemap_write_and_wait_range, which isn't
7235          * enough if we've written compressed pages to this area, so we need to
7236          * call btrfs_wait_ordered_range to make absolutely sure that any
7237          * outstanding dirty pages are on disk.
7238          */
7239         count = iov_length(iov, nr_segs);
7240         btrfs_wait_ordered_range(inode, offset, count);
7241
7242         if (rw & WRITE) {
7243                 /*
7244                  * If the write DIO is beyond the EOF, we need update
7245                  * the isize, but it is protected by i_mutex. So we can
7246                  * not unlock the i_mutex at this case.
7247                  */
7248                 if (offset + count <= inode->i_size) {
7249                         mutex_unlock(&inode->i_mutex);
7250                         relock = true;
7251                 }
7252                 ret = btrfs_delalloc_reserve_space(inode, count);
7253                 if (ret)
7254                         goto out;
7255         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7256                                      &BTRFS_I(inode)->runtime_flags))) {
7257                 inode_dio_done(inode);
7258                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7259                 wakeup = false;
7260         }
7261
7262         ret = __blockdev_direct_IO(rw, iocb, inode,
7263                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7264                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7265                         btrfs_submit_direct, flags);
7266         if (rw & WRITE) {
7267                 if (ret < 0 && ret != -EIOCBQUEUED)
7268                         btrfs_delalloc_release_space(inode, count);
7269                 else if (ret >= 0 && (size_t)ret < count)
7270                         btrfs_delalloc_release_space(inode,
7271                                                      count - (size_t)ret);
7272                 else
7273                         btrfs_delalloc_release_metadata(inode, 0);
7274         }
7275 out:
7276         if (wakeup)
7277                 inode_dio_done(inode);
7278         if (relock)
7279                 mutex_lock(&inode->i_mutex);
7280
7281         return ret;
7282 }
7283
7284 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7285
7286 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7287                 __u64 start, __u64 len)
7288 {
7289         int     ret;
7290
7291         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7292         if (ret)
7293                 return ret;
7294
7295         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7296 }
7297
7298 int btrfs_readpage(struct file *file, struct page *page)
7299 {
7300         struct extent_io_tree *tree;
7301         tree = &BTRFS_I(page->mapping->host)->io_tree;
7302         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7303 }
7304
7305 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7306 {
7307         struct extent_io_tree *tree;
7308
7309
7310         if (current->flags & PF_MEMALLOC) {
7311                 redirty_page_for_writepage(wbc, page);
7312                 unlock_page(page);
7313                 return 0;
7314         }
7315         tree = &BTRFS_I(page->mapping->host)->io_tree;
7316         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7317 }
7318
7319 static int btrfs_writepages(struct address_space *mapping,
7320                             struct writeback_control *wbc)
7321 {
7322         struct extent_io_tree *tree;
7323
7324         tree = &BTRFS_I(mapping->host)->io_tree;
7325         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7326 }
7327
7328 static int
7329 btrfs_readpages(struct file *file, struct address_space *mapping,
7330                 struct list_head *pages, unsigned nr_pages)
7331 {
7332         struct extent_io_tree *tree;
7333         tree = &BTRFS_I(mapping->host)->io_tree;
7334         return extent_readpages(tree, mapping, pages, nr_pages,
7335                                 btrfs_get_extent);
7336 }
7337 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7338 {
7339         struct extent_io_tree *tree;
7340         struct extent_map_tree *map;
7341         int ret;
7342
7343         tree = &BTRFS_I(page->mapping->host)->io_tree;
7344         map = &BTRFS_I(page->mapping->host)->extent_tree;
7345         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7346         if (ret == 1) {
7347                 ClearPagePrivate(page);
7348                 set_page_private(page, 0);
7349                 page_cache_release(page);
7350         }
7351         return ret;
7352 }
7353
7354 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7355 {
7356         if (PageWriteback(page) || PageDirty(page))
7357                 return 0;
7358         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7359 }
7360
7361 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7362                                  unsigned int length)
7363 {
7364         struct inode *inode = page->mapping->host;
7365         struct extent_io_tree *tree;
7366         struct btrfs_ordered_extent *ordered;
7367         struct extent_state *cached_state = NULL;
7368         u64 page_start = page_offset(page);
7369         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7370
7371         /*
7372          * we have the page locked, so new writeback can't start,
7373          * and the dirty bit won't be cleared while we are here.
7374          *
7375          * Wait for IO on this page so that we can safely clear
7376          * the PagePrivate2 bit and do ordered accounting
7377          */
7378         wait_on_page_writeback(page);
7379
7380         tree = &BTRFS_I(inode)->io_tree;
7381         if (offset) {
7382                 btrfs_releasepage(page, GFP_NOFS);
7383                 return;
7384         }
7385         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7386         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
7387         if (ordered) {
7388                 /*
7389                  * IO on this page will never be started, so we need
7390                  * to account for any ordered extents now
7391                  */
7392                 clear_extent_bit(tree, page_start, page_end,
7393                                  EXTENT_DIRTY | EXTENT_DELALLOC |
7394                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7395                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
7396                 /*
7397                  * whoever cleared the private bit is responsible
7398                  * for the finish_ordered_io
7399                  */
7400                 if (TestClearPagePrivate2(page)) {
7401                         struct btrfs_ordered_inode_tree *tree;
7402                         u64 new_len;
7403
7404                         tree = &BTRFS_I(inode)->ordered_tree;
7405
7406                         spin_lock_irq(&tree->lock);
7407                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7408                         new_len = page_start - ordered->file_offset;
7409                         if (new_len < ordered->truncated_len)
7410                                 ordered->truncated_len = new_len;
7411                         spin_unlock_irq(&tree->lock);
7412
7413                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7414                                                            page_start,
7415                                                            PAGE_CACHE_SIZE, 1))
7416                                 btrfs_finish_ordered_io(ordered);
7417                 }
7418                 btrfs_put_ordered_extent(ordered);
7419                 cached_state = NULL;
7420                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7421         }
7422         clear_extent_bit(tree, page_start, page_end,
7423                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
7424                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
7425                  &cached_state, GFP_NOFS);
7426         __btrfs_releasepage(page, GFP_NOFS);
7427
7428         ClearPageChecked(page);
7429         if (PagePrivate(page)) {
7430                 ClearPagePrivate(page);
7431                 set_page_private(page, 0);
7432                 page_cache_release(page);
7433         }
7434 }
7435
7436 /*
7437  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7438  * called from a page fault handler when a page is first dirtied. Hence we must
7439  * be careful to check for EOF conditions here. We set the page up correctly
7440  * for a written page which means we get ENOSPC checking when writing into
7441  * holes and correct delalloc and unwritten extent mapping on filesystems that
7442  * support these features.
7443  *
7444  * We are not allowed to take the i_mutex here so we have to play games to
7445  * protect against truncate races as the page could now be beyond EOF.  Because
7446  * vmtruncate() writes the inode size before removing pages, once we have the
7447  * page lock we can determine safely if the page is beyond EOF. If it is not
7448  * beyond EOF, then the page is guaranteed safe against truncation until we
7449  * unlock the page.
7450  */
7451 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7452 {
7453         struct page *page = vmf->page;
7454         struct inode *inode = file_inode(vma->vm_file);
7455         struct btrfs_root *root = BTRFS_I(inode)->root;
7456         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7457         struct btrfs_ordered_extent *ordered;
7458         struct extent_state *cached_state = NULL;
7459         char *kaddr;
7460         unsigned long zero_start;
7461         loff_t size;
7462         int ret;
7463         int reserved = 0;
7464         u64 page_start;
7465         u64 page_end;
7466
7467         sb_start_pagefault(inode->i_sb);
7468         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7469         if (!ret) {
7470                 ret = file_update_time(vma->vm_file);
7471                 reserved = 1;
7472         }
7473         if (ret) {
7474                 if (ret == -ENOMEM)
7475                         ret = VM_FAULT_OOM;
7476                 else /* -ENOSPC, -EIO, etc */
7477                         ret = VM_FAULT_SIGBUS;
7478                 if (reserved)
7479                         goto out;
7480                 goto out_noreserve;
7481         }
7482
7483         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7484 again:
7485         lock_page(page);
7486         size = i_size_read(inode);
7487         page_start = page_offset(page);
7488         page_end = page_start + PAGE_CACHE_SIZE - 1;
7489
7490         if ((page->mapping != inode->i_mapping) ||
7491             (page_start >= size)) {
7492                 /* page got truncated out from underneath us */
7493                 goto out_unlock;
7494         }
7495         wait_on_page_writeback(page);
7496
7497         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7498         set_page_extent_mapped(page);
7499
7500         /*
7501          * we can't set the delalloc bits if there are pending ordered
7502          * extents.  Drop our locks and wait for them to finish
7503          */
7504         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7505         if (ordered) {
7506                 unlock_extent_cached(io_tree, page_start, page_end,
7507                                      &cached_state, GFP_NOFS);
7508                 unlock_page(page);
7509                 btrfs_start_ordered_extent(inode, ordered, 1);
7510                 btrfs_put_ordered_extent(ordered);
7511                 goto again;
7512         }
7513
7514         /*
7515          * XXX - page_mkwrite gets called every time the page is dirtied, even
7516          * if it was already dirty, so for space accounting reasons we need to
7517          * clear any delalloc bits for the range we are fixing to save.  There
7518          * is probably a better way to do this, but for now keep consistent with
7519          * prepare_pages in the normal write path.
7520          */
7521         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7522                           EXTENT_DIRTY | EXTENT_DELALLOC |
7523                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7524                           0, 0, &cached_state, GFP_NOFS);
7525
7526         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7527                                         &cached_state);
7528         if (ret) {
7529                 unlock_extent_cached(io_tree, page_start, page_end,
7530                                      &cached_state, GFP_NOFS);
7531                 ret = VM_FAULT_SIGBUS;
7532                 goto out_unlock;
7533         }
7534         ret = 0;
7535
7536         /* page is wholly or partially inside EOF */
7537         if (page_start + PAGE_CACHE_SIZE > size)
7538                 zero_start = size & ~PAGE_CACHE_MASK;
7539         else
7540                 zero_start = PAGE_CACHE_SIZE;
7541
7542         if (zero_start != PAGE_CACHE_SIZE) {
7543                 kaddr = kmap(page);
7544                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7545                 flush_dcache_page(page);
7546                 kunmap(page);
7547         }
7548         ClearPageChecked(page);
7549         set_page_dirty(page);
7550         SetPageUptodate(page);
7551
7552         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7553         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7554         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7555
7556         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7557
7558 out_unlock:
7559         if (!ret) {
7560                 sb_end_pagefault(inode->i_sb);
7561                 return VM_FAULT_LOCKED;
7562         }
7563         unlock_page(page);
7564 out:
7565         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7566 out_noreserve:
7567         sb_end_pagefault(inode->i_sb);
7568         return ret;
7569 }
7570
7571 static int btrfs_truncate(struct inode *inode)
7572 {
7573         struct btrfs_root *root = BTRFS_I(inode)->root;
7574         struct btrfs_block_rsv *rsv;
7575         int ret = 0;
7576         int err = 0;
7577         struct btrfs_trans_handle *trans;
7578         u64 mask = root->sectorsize - 1;
7579         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7580
7581         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
7582
7583         /*
7584          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7585          * 3 things going on here
7586          *
7587          * 1) We need to reserve space for our orphan item and the space to
7588          * delete our orphan item.  Lord knows we don't want to have a dangling
7589          * orphan item because we didn't reserve space to remove it.
7590          *
7591          * 2) We need to reserve space to update our inode.
7592          *
7593          * 3) We need to have something to cache all the space that is going to
7594          * be free'd up by the truncate operation, but also have some slack
7595          * space reserved in case it uses space during the truncate (thank you
7596          * very much snapshotting).
7597          *
7598          * And we need these to all be seperate.  The fact is we can use alot of
7599          * space doing the truncate, and we have no earthly idea how much space
7600          * we will use, so we need the truncate reservation to be seperate so it
7601          * doesn't end up using space reserved for updating the inode or
7602          * removing the orphan item.  We also need to be able to stop the
7603          * transaction and start a new one, which means we need to be able to
7604          * update the inode several times, and we have no idea of knowing how
7605          * many times that will be, so we can't just reserve 1 item for the
7606          * entirety of the opration, so that has to be done seperately as well.
7607          * Then there is the orphan item, which does indeed need to be held on
7608          * to for the whole operation, and we need nobody to touch this reserved
7609          * space except the orphan code.
7610          *
7611          * So that leaves us with
7612          *
7613          * 1) root->orphan_block_rsv - for the orphan deletion.
7614          * 2) rsv - for the truncate reservation, which we will steal from the
7615          * transaction reservation.
7616          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7617          * updating the inode.
7618          */
7619         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7620         if (!rsv)
7621                 return -ENOMEM;
7622         rsv->size = min_size;
7623         rsv->failfast = 1;
7624
7625         /*
7626          * 1 for the truncate slack space
7627          * 1 for updating the inode.
7628          */
7629         trans = btrfs_start_transaction(root, 2);
7630         if (IS_ERR(trans)) {
7631                 err = PTR_ERR(trans);
7632                 goto out;
7633         }
7634
7635         /* Migrate the slack space for the truncate to our reserve */
7636         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7637                                       min_size);
7638         BUG_ON(ret);
7639
7640         /*
7641          * setattr is responsible for setting the ordered_data_close flag,
7642          * but that is only tested during the last file release.  That
7643          * could happen well after the next commit, leaving a great big
7644          * window where new writes may get lost if someone chooses to write
7645          * to this file after truncating to zero
7646          *
7647          * The inode doesn't have any dirty data here, and so if we commit
7648          * this is a noop.  If someone immediately starts writing to the inode
7649          * it is very likely we'll catch some of their writes in this
7650          * transaction, and the commit will find this file on the ordered
7651          * data list with good things to send down.
7652          *
7653          * This is a best effort solution, there is still a window where
7654          * using truncate to replace the contents of the file will
7655          * end up with a zero length file after a crash.
7656          */
7657         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7658                                            &BTRFS_I(inode)->runtime_flags))
7659                 btrfs_add_ordered_operation(trans, root, inode);
7660
7661         /*
7662          * So if we truncate and then write and fsync we normally would just
7663          * write the extents that changed, which is a problem if we need to
7664          * first truncate that entire inode.  So set this flag so we write out
7665          * all of the extents in the inode to the sync log so we're completely
7666          * safe.
7667          */
7668         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7669         trans->block_rsv = rsv;
7670
7671         while (1) {
7672                 ret = btrfs_truncate_inode_items(trans, root, inode,
7673                                                  inode->i_size,
7674                                                  BTRFS_EXTENT_DATA_KEY);
7675                 if (ret != -ENOSPC) {
7676                         err = ret;
7677                         break;
7678                 }
7679
7680                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7681                 ret = btrfs_update_inode(trans, root, inode);
7682                 if (ret) {
7683                         err = ret;
7684                         break;
7685                 }
7686
7687                 btrfs_end_transaction(trans, root);
7688                 btrfs_btree_balance_dirty(root);
7689
7690                 trans = btrfs_start_transaction(root, 2);
7691                 if (IS_ERR(trans)) {
7692                         ret = err = PTR_ERR(trans);
7693                         trans = NULL;
7694                         break;
7695                 }
7696
7697                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7698                                               rsv, min_size);
7699                 BUG_ON(ret);    /* shouldn't happen */
7700                 trans->block_rsv = rsv;
7701         }
7702
7703         if (ret == 0 && inode->i_nlink > 0) {
7704                 trans->block_rsv = root->orphan_block_rsv;
7705                 ret = btrfs_orphan_del(trans, inode);
7706                 if (ret)
7707                         err = ret;
7708         }
7709
7710         if (trans) {
7711                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7712                 ret = btrfs_update_inode(trans, root, inode);
7713                 if (ret && !err)
7714                         err = ret;
7715
7716                 ret = btrfs_end_transaction(trans, root);
7717                 btrfs_btree_balance_dirty(root);
7718         }
7719
7720 out:
7721         btrfs_free_block_rsv(root, rsv);
7722
7723         if (ret && !err)
7724                 err = ret;
7725
7726         return err;
7727 }
7728
7729 /*
7730  * create a new subvolume directory/inode (helper for the ioctl).
7731  */
7732 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7733                              struct btrfs_root *new_root, u64 new_dirid)
7734 {
7735         struct inode *inode;
7736         int err;
7737         u64 index = 0;
7738
7739         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7740                                 new_dirid, new_dirid,
7741                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7742                                 &index);
7743         if (IS_ERR(inode))
7744                 return PTR_ERR(inode);
7745         inode->i_op = &btrfs_dir_inode_operations;
7746         inode->i_fop = &btrfs_dir_file_operations;
7747
7748         set_nlink(inode, 1);
7749         btrfs_i_size_write(inode, 0);
7750
7751         err = btrfs_update_inode(trans, new_root, inode);
7752
7753         iput(inode);
7754         return err;
7755 }
7756
7757 struct inode *btrfs_alloc_inode(struct super_block *sb)
7758 {
7759         struct btrfs_inode *ei;
7760         struct inode *inode;
7761
7762         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7763         if (!ei)
7764                 return NULL;
7765
7766         ei->root = NULL;
7767         ei->generation = 0;
7768         ei->last_trans = 0;
7769         ei->last_sub_trans = 0;
7770         ei->logged_trans = 0;
7771         ei->delalloc_bytes = 0;
7772         ei->disk_i_size = 0;
7773         ei->flags = 0;
7774         ei->csum_bytes = 0;
7775         ei->index_cnt = (u64)-1;
7776         ei->last_unlink_trans = 0;
7777         ei->last_log_commit = 0;
7778
7779         spin_lock_init(&ei->lock);
7780         ei->outstanding_extents = 0;
7781         ei->reserved_extents = 0;
7782
7783         ei->runtime_flags = 0;
7784         ei->force_compress = BTRFS_COMPRESS_NONE;
7785
7786         ei->delayed_node = NULL;
7787
7788         inode = &ei->vfs_inode;
7789         extent_map_tree_init(&ei->extent_tree);
7790         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7791         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7792         ei->io_tree.track_uptodate = 1;
7793         ei->io_failure_tree.track_uptodate = 1;
7794         atomic_set(&ei->sync_writers, 0);
7795         mutex_init(&ei->log_mutex);
7796         mutex_init(&ei->delalloc_mutex);
7797         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7798         INIT_LIST_HEAD(&ei->delalloc_inodes);
7799         INIT_LIST_HEAD(&ei->ordered_operations);
7800         RB_CLEAR_NODE(&ei->rb_node);
7801
7802         return inode;
7803 }
7804
7805 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7806 void btrfs_test_destroy_inode(struct inode *inode)
7807 {
7808         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7809         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7810 }
7811 #endif
7812
7813 static void btrfs_i_callback(struct rcu_head *head)
7814 {
7815         struct inode *inode = container_of(head, struct inode, i_rcu);
7816         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7817 }
7818
7819 void btrfs_destroy_inode(struct inode *inode)
7820 {
7821         struct btrfs_ordered_extent *ordered;
7822         struct btrfs_root *root = BTRFS_I(inode)->root;
7823
7824         WARN_ON(!hlist_empty(&inode->i_dentry));
7825         WARN_ON(inode->i_data.nrpages);
7826         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7827         WARN_ON(BTRFS_I(inode)->reserved_extents);
7828         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7829         WARN_ON(BTRFS_I(inode)->csum_bytes);
7830
7831         /*
7832          * This can happen where we create an inode, but somebody else also
7833          * created the same inode and we need to destroy the one we already
7834          * created.
7835          */
7836         if (!root)
7837                 goto free;
7838
7839         /*
7840          * Make sure we're properly removed from the ordered operation
7841          * lists.
7842          */
7843         smp_mb();
7844         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7845                 spin_lock(&root->fs_info->ordered_root_lock);
7846                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7847                 spin_unlock(&root->fs_info->ordered_root_lock);
7848         }
7849
7850         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7851                      &BTRFS_I(inode)->runtime_flags)) {
7852                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7853                         btrfs_ino(inode));
7854                 atomic_dec(&root->orphan_inodes);
7855         }
7856
7857         while (1) {
7858                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7859                 if (!ordered)
7860                         break;
7861                 else {
7862                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7863                                 ordered->file_offset, ordered->len);
7864                         btrfs_remove_ordered_extent(inode, ordered);
7865                         btrfs_put_ordered_extent(ordered);
7866                         btrfs_put_ordered_extent(ordered);
7867                 }
7868         }
7869         inode_tree_del(inode);
7870         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7871 free:
7872         call_rcu(&inode->i_rcu, btrfs_i_callback);
7873 }
7874
7875 int btrfs_drop_inode(struct inode *inode)
7876 {
7877         struct btrfs_root *root = BTRFS_I(inode)->root;
7878
7879         if (root == NULL)
7880                 return 1;
7881
7882         /* the snap/subvol tree is on deleting */
7883         if (btrfs_root_refs(&root->root_item) == 0)
7884                 return 1;
7885         else
7886                 return generic_drop_inode(inode);
7887 }
7888
7889 static void init_once(void *foo)
7890 {
7891         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7892
7893         inode_init_once(&ei->vfs_inode);
7894 }
7895
7896 void btrfs_destroy_cachep(void)
7897 {
7898         /*
7899          * Make sure all delayed rcu free inodes are flushed before we
7900          * destroy cache.
7901          */
7902         rcu_barrier();
7903         if (btrfs_inode_cachep)
7904                 kmem_cache_destroy(btrfs_inode_cachep);
7905         if (btrfs_trans_handle_cachep)
7906                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7907         if (btrfs_transaction_cachep)
7908                 kmem_cache_destroy(btrfs_transaction_cachep);
7909         if (btrfs_path_cachep)
7910                 kmem_cache_destroy(btrfs_path_cachep);
7911         if (btrfs_free_space_cachep)
7912                 kmem_cache_destroy(btrfs_free_space_cachep);
7913         if (btrfs_delalloc_work_cachep)
7914                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7915 }
7916
7917 int btrfs_init_cachep(void)
7918 {
7919         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7920                         sizeof(struct btrfs_inode), 0,
7921                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7922         if (!btrfs_inode_cachep)
7923                 goto fail;
7924
7925         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7926                         sizeof(struct btrfs_trans_handle), 0,
7927                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7928         if (!btrfs_trans_handle_cachep)
7929                 goto fail;
7930
7931         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7932                         sizeof(struct btrfs_transaction), 0,
7933                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7934         if (!btrfs_transaction_cachep)
7935                 goto fail;
7936
7937         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7938                         sizeof(struct btrfs_path), 0,
7939                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7940         if (!btrfs_path_cachep)
7941                 goto fail;
7942
7943         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7944                         sizeof(struct btrfs_free_space), 0,
7945                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7946         if (!btrfs_free_space_cachep)
7947                 goto fail;
7948
7949         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7950                         sizeof(struct btrfs_delalloc_work), 0,
7951                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7952                         NULL);
7953         if (!btrfs_delalloc_work_cachep)
7954                 goto fail;
7955
7956         return 0;
7957 fail:
7958         btrfs_destroy_cachep();
7959         return -ENOMEM;
7960 }
7961
7962 static int btrfs_getattr(struct vfsmount *mnt,
7963                          struct dentry *dentry, struct kstat *stat)
7964 {
7965         u64 delalloc_bytes;
7966         struct inode *inode = dentry->d_inode;
7967         u32 blocksize = inode->i_sb->s_blocksize;
7968
7969         generic_fillattr(inode, stat);
7970         stat->dev = BTRFS_I(inode)->root->anon_dev;
7971         stat->blksize = PAGE_CACHE_SIZE;
7972
7973         spin_lock(&BTRFS_I(inode)->lock);
7974         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7975         spin_unlock(&BTRFS_I(inode)->lock);
7976         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7977                         ALIGN(delalloc_bytes, blocksize)) >> 9;
7978         return 0;
7979 }
7980
7981 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7982                            struct inode *new_dir, struct dentry *new_dentry)
7983 {
7984         struct btrfs_trans_handle *trans;
7985         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7986         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7987         struct inode *new_inode = new_dentry->d_inode;
7988         struct inode *old_inode = old_dentry->d_inode;
7989         struct timespec ctime = CURRENT_TIME;
7990         u64 index = 0;
7991         u64 root_objectid;
7992         int ret;
7993         u64 old_ino = btrfs_ino(old_inode);
7994
7995         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7996                 return -EPERM;
7997
7998         /* we only allow rename subvolume link between subvolumes */
7999         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8000                 return -EXDEV;
8001
8002         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8003             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8004                 return -ENOTEMPTY;
8005
8006         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8007             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8008                 return -ENOTEMPTY;
8009
8010
8011         /* check for collisions, even if the  name isn't there */
8012         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8013                              new_dentry->d_name.name,
8014                              new_dentry->d_name.len);
8015
8016         if (ret) {
8017                 if (ret == -EEXIST) {
8018                         /* we shouldn't get
8019                          * eexist without a new_inode */
8020                         if (!new_inode) {
8021                                 WARN_ON(1);
8022                                 return ret;
8023                         }
8024                 } else {
8025                         /* maybe -EOVERFLOW */
8026                         return ret;
8027                 }
8028         }
8029         ret = 0;
8030
8031         /*
8032          * we're using rename to replace one file with another.
8033          * and the replacement file is large.  Start IO on it now so
8034          * we don't add too much work to the end of the transaction
8035          */
8036         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8037             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8038                 filemap_flush(old_inode->i_mapping);
8039
8040         /* close the racy window with snapshot create/destroy ioctl */
8041         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8042                 down_read(&root->fs_info->subvol_sem);
8043         /*
8044          * We want to reserve the absolute worst case amount of items.  So if
8045          * both inodes are subvols and we need to unlink them then that would
8046          * require 4 item modifications, but if they are both normal inodes it
8047          * would require 5 item modifications, so we'll assume their normal
8048          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8049          * should cover the worst case number of items we'll modify.
8050          */
8051         trans = btrfs_start_transaction(root, 11);
8052         if (IS_ERR(trans)) {
8053                 ret = PTR_ERR(trans);
8054                 goto out_notrans;
8055         }
8056
8057         if (dest != root)
8058                 btrfs_record_root_in_trans(trans, dest);
8059
8060         ret = btrfs_set_inode_index(new_dir, &index);
8061         if (ret)
8062                 goto out_fail;
8063
8064         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8065                 /* force full log commit if subvolume involved. */
8066                 root->fs_info->last_trans_log_full_commit = trans->transid;
8067         } else {
8068                 ret = btrfs_insert_inode_ref(trans, dest,
8069                                              new_dentry->d_name.name,
8070                                              new_dentry->d_name.len,
8071                                              old_ino,
8072                                              btrfs_ino(new_dir), index);
8073                 if (ret)
8074                         goto out_fail;
8075                 /*
8076                  * this is an ugly little race, but the rename is required
8077                  * to make sure that if we crash, the inode is either at the
8078                  * old name or the new one.  pinning the log transaction lets
8079                  * us make sure we don't allow a log commit to come in after
8080                  * we unlink the name but before we add the new name back in.
8081                  */
8082                 btrfs_pin_log_trans(root);
8083         }
8084         /*
8085          * make sure the inode gets flushed if it is replacing
8086          * something.
8087          */
8088         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8089                 btrfs_add_ordered_operation(trans, root, old_inode);
8090
8091         inode_inc_iversion(old_dir);
8092         inode_inc_iversion(new_dir);
8093         inode_inc_iversion(old_inode);
8094         old_dir->i_ctime = old_dir->i_mtime = ctime;
8095         new_dir->i_ctime = new_dir->i_mtime = ctime;
8096         old_inode->i_ctime = ctime;
8097
8098         if (old_dentry->d_parent != new_dentry->d_parent)
8099                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8100
8101         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8102                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8103                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8104                                         old_dentry->d_name.name,
8105                                         old_dentry->d_name.len);
8106         } else {
8107                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8108                                         old_dentry->d_inode,
8109                                         old_dentry->d_name.name,
8110                                         old_dentry->d_name.len);
8111                 if (!ret)
8112                         ret = btrfs_update_inode(trans, root, old_inode);
8113         }
8114         if (ret) {
8115                 btrfs_abort_transaction(trans, root, ret);
8116                 goto out_fail;
8117         }
8118
8119         if (new_inode) {
8120                 inode_inc_iversion(new_inode);
8121                 new_inode->i_ctime = CURRENT_TIME;
8122                 if (unlikely(btrfs_ino(new_inode) ==
8123                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8124                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8125                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8126                                                 root_objectid,
8127                                                 new_dentry->d_name.name,
8128                                                 new_dentry->d_name.len);
8129                         BUG_ON(new_inode->i_nlink == 0);
8130                 } else {
8131                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8132                                                  new_dentry->d_inode,
8133                                                  new_dentry->d_name.name,
8134                                                  new_dentry->d_name.len);
8135                 }
8136                 if (!ret && new_inode->i_nlink == 0)
8137                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8138                 if (ret) {
8139                         btrfs_abort_transaction(trans, root, ret);
8140                         goto out_fail;
8141                 }
8142         }
8143
8144         ret = btrfs_add_link(trans, new_dir, old_inode,
8145                              new_dentry->d_name.name,
8146                              new_dentry->d_name.len, 0, index);
8147         if (ret) {
8148                 btrfs_abort_transaction(trans, root, ret);
8149                 goto out_fail;
8150         }
8151
8152         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8153                 struct dentry *parent = new_dentry->d_parent;
8154                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8155                 btrfs_end_log_trans(root);
8156         }
8157 out_fail:
8158         btrfs_end_transaction(trans, root);
8159 out_notrans:
8160         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8161                 up_read(&root->fs_info->subvol_sem);
8162
8163         return ret;
8164 }
8165
8166 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8167 {
8168         struct btrfs_delalloc_work *delalloc_work;
8169
8170         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8171                                      work);
8172         if (delalloc_work->wait)
8173                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
8174         else
8175                 filemap_flush(delalloc_work->inode->i_mapping);
8176
8177         if (delalloc_work->delay_iput)
8178                 btrfs_add_delayed_iput(delalloc_work->inode);
8179         else
8180                 iput(delalloc_work->inode);
8181         complete(&delalloc_work->completion);
8182 }
8183
8184 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8185                                                     int wait, int delay_iput)
8186 {
8187         struct btrfs_delalloc_work *work;
8188
8189         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8190         if (!work)
8191                 return NULL;
8192
8193         init_completion(&work->completion);
8194         INIT_LIST_HEAD(&work->list);
8195         work->inode = inode;
8196         work->wait = wait;
8197         work->delay_iput = delay_iput;
8198         work->work.func = btrfs_run_delalloc_work;
8199
8200         return work;
8201 }
8202
8203 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8204 {
8205         wait_for_completion(&work->completion);
8206         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8207 }
8208
8209 /*
8210  * some fairly slow code that needs optimization. This walks the list
8211  * of all the inodes with pending delalloc and forces them to disk.
8212  */
8213 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8214 {
8215         struct btrfs_inode *binode;
8216         struct inode *inode;
8217         struct btrfs_delalloc_work *work, *next;
8218         struct list_head works;
8219         struct list_head splice;
8220         int ret = 0;
8221
8222         INIT_LIST_HEAD(&works);
8223         INIT_LIST_HEAD(&splice);
8224
8225         spin_lock(&root->delalloc_lock);
8226         list_splice_init(&root->delalloc_inodes, &splice);
8227         while (!list_empty(&splice)) {
8228                 binode = list_entry(splice.next, struct btrfs_inode,
8229                                     delalloc_inodes);
8230
8231                 list_move_tail(&binode->delalloc_inodes,
8232                                &root->delalloc_inodes);
8233                 inode = igrab(&binode->vfs_inode);
8234                 if (!inode) {
8235                         cond_resched_lock(&root->delalloc_lock);
8236                         continue;
8237                 }
8238                 spin_unlock(&root->delalloc_lock);
8239
8240                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8241                 if (unlikely(!work)) {
8242                         if (delay_iput)
8243                                 btrfs_add_delayed_iput(inode);
8244                         else
8245                                 iput(inode);
8246                         ret = -ENOMEM;
8247                         goto out;
8248                 }
8249                 list_add_tail(&work->list, &works);
8250                 btrfs_queue_worker(&root->fs_info->flush_workers,
8251                                    &work->work);
8252
8253                 cond_resched();
8254                 spin_lock(&root->delalloc_lock);
8255         }
8256         spin_unlock(&root->delalloc_lock);
8257
8258         list_for_each_entry_safe(work, next, &works, list) {
8259                 list_del_init(&work->list);
8260                 btrfs_wait_and_free_delalloc_work(work);
8261         }
8262         return 0;
8263 out:
8264         list_for_each_entry_safe(work, next, &works, list) {
8265                 list_del_init(&work->list);
8266                 btrfs_wait_and_free_delalloc_work(work);
8267         }
8268
8269         if (!list_empty_careful(&splice)) {
8270                 spin_lock(&root->delalloc_lock);
8271                 list_splice_tail(&splice, &root->delalloc_inodes);
8272                 spin_unlock(&root->delalloc_lock);
8273         }
8274         return ret;
8275 }
8276
8277 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8278 {
8279         int ret;
8280
8281         if (root->fs_info->sb->s_flags & MS_RDONLY)
8282                 return -EROFS;
8283
8284         ret = __start_delalloc_inodes(root, delay_iput);
8285         /*
8286          * the filemap_flush will queue IO into the worker threads, but
8287          * we have to make sure the IO is actually started and that
8288          * ordered extents get created before we return
8289          */
8290         atomic_inc(&root->fs_info->async_submit_draining);
8291         while (atomic_read(&root->fs_info->nr_async_submits) ||
8292               atomic_read(&root->fs_info->async_delalloc_pages)) {
8293                 wait_event(root->fs_info->async_submit_wait,
8294                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8295                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8296         }
8297         atomic_dec(&root->fs_info->async_submit_draining);
8298         return ret;
8299 }
8300
8301 int btrfs_start_all_delalloc_inodes(struct btrfs_fs_info *fs_info,
8302                                     int delay_iput)
8303 {
8304         struct btrfs_root *root;
8305         struct list_head splice;
8306         int ret;
8307
8308         if (fs_info->sb->s_flags & MS_RDONLY)
8309                 return -EROFS;
8310
8311         INIT_LIST_HEAD(&splice);
8312
8313         spin_lock(&fs_info->delalloc_root_lock);
8314         list_splice_init(&fs_info->delalloc_roots, &splice);
8315         while (!list_empty(&splice)) {
8316                 root = list_first_entry(&splice, struct btrfs_root,
8317                                         delalloc_root);
8318                 root = btrfs_grab_fs_root(root);
8319                 BUG_ON(!root);
8320                 list_move_tail(&root->delalloc_root,
8321                                &fs_info->delalloc_roots);
8322                 spin_unlock(&fs_info->delalloc_root_lock);
8323
8324                 ret = __start_delalloc_inodes(root, delay_iput);
8325                 btrfs_put_fs_root(root);
8326                 if (ret)
8327                         goto out;
8328
8329                 spin_lock(&fs_info->delalloc_root_lock);
8330         }
8331         spin_unlock(&fs_info->delalloc_root_lock);
8332
8333         atomic_inc(&fs_info->async_submit_draining);
8334         while (atomic_read(&fs_info->nr_async_submits) ||
8335               atomic_read(&fs_info->async_delalloc_pages)) {
8336                 wait_event(fs_info->async_submit_wait,
8337                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8338                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8339         }
8340         atomic_dec(&fs_info->async_submit_draining);
8341         return 0;
8342 out:
8343         if (!list_empty_careful(&splice)) {
8344                 spin_lock(&fs_info->delalloc_root_lock);
8345                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8346                 spin_unlock(&fs_info->delalloc_root_lock);
8347         }
8348         return ret;
8349 }
8350
8351 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8352                          const char *symname)
8353 {
8354         struct btrfs_trans_handle *trans;
8355         struct btrfs_root *root = BTRFS_I(dir)->root;
8356         struct btrfs_path *path;
8357         struct btrfs_key key;
8358         struct inode *inode = NULL;
8359         int err;
8360         int drop_inode = 0;
8361         u64 objectid;
8362         u64 index = 0 ;
8363         int name_len;
8364         int datasize;
8365         unsigned long ptr;
8366         struct btrfs_file_extent_item *ei;
8367         struct extent_buffer *leaf;
8368
8369         name_len = strlen(symname);
8370         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8371                 return -ENAMETOOLONG;
8372
8373         /*
8374          * 2 items for inode item and ref
8375          * 2 items for dir items
8376          * 1 item for xattr if selinux is on
8377          */
8378         trans = btrfs_start_transaction(root, 5);
8379         if (IS_ERR(trans))
8380                 return PTR_ERR(trans);
8381
8382         err = btrfs_find_free_ino(root, &objectid);
8383         if (err)
8384                 goto out_unlock;
8385
8386         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8387                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8388                                 S_IFLNK|S_IRWXUGO, &index);
8389         if (IS_ERR(inode)) {
8390                 err = PTR_ERR(inode);
8391                 goto out_unlock;
8392         }
8393
8394         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8395         if (err) {
8396                 drop_inode = 1;
8397                 goto out_unlock;
8398         }
8399
8400         /*
8401         * If the active LSM wants to access the inode during
8402         * d_instantiate it needs these. Smack checks to see
8403         * if the filesystem supports xattrs by looking at the
8404         * ops vector.
8405         */
8406         inode->i_fop = &btrfs_file_operations;
8407         inode->i_op = &btrfs_file_inode_operations;
8408
8409         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8410         if (err)
8411                 drop_inode = 1;
8412         else {
8413                 inode->i_mapping->a_ops = &btrfs_aops;
8414                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8415                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8416         }
8417         if (drop_inode)
8418                 goto out_unlock;
8419
8420         path = btrfs_alloc_path();
8421         if (!path) {
8422                 err = -ENOMEM;
8423                 drop_inode = 1;
8424                 goto out_unlock;
8425         }
8426         key.objectid = btrfs_ino(inode);
8427         key.offset = 0;
8428         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8429         datasize = btrfs_file_extent_calc_inline_size(name_len);
8430         err = btrfs_insert_empty_item(trans, root, path, &key,
8431                                       datasize);
8432         if (err) {
8433                 drop_inode = 1;
8434                 btrfs_free_path(path);
8435                 goto out_unlock;
8436         }
8437         leaf = path->nodes[0];
8438         ei = btrfs_item_ptr(leaf, path->slots[0],
8439                             struct btrfs_file_extent_item);
8440         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8441         btrfs_set_file_extent_type(leaf, ei,
8442                                    BTRFS_FILE_EXTENT_INLINE);
8443         btrfs_set_file_extent_encryption(leaf, ei, 0);
8444         btrfs_set_file_extent_compression(leaf, ei, 0);
8445         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8446         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8447
8448         ptr = btrfs_file_extent_inline_start(ei);
8449         write_extent_buffer(leaf, symname, ptr, name_len);
8450         btrfs_mark_buffer_dirty(leaf);
8451         btrfs_free_path(path);
8452
8453         inode->i_op = &btrfs_symlink_inode_operations;
8454         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8455         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8456         inode_set_bytes(inode, name_len);
8457         btrfs_i_size_write(inode, name_len);
8458         err = btrfs_update_inode(trans, root, inode);
8459         if (err)
8460                 drop_inode = 1;
8461
8462 out_unlock:
8463         if (!err)
8464                 d_instantiate(dentry, inode);
8465         btrfs_end_transaction(trans, root);
8466         if (drop_inode) {
8467                 inode_dec_link_count(inode);
8468                 iput(inode);
8469         }
8470         btrfs_btree_balance_dirty(root);
8471         return err;
8472 }
8473
8474 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8475                                        u64 start, u64 num_bytes, u64 min_size,
8476                                        loff_t actual_len, u64 *alloc_hint,
8477                                        struct btrfs_trans_handle *trans)
8478 {
8479         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8480         struct extent_map *em;
8481         struct btrfs_root *root = BTRFS_I(inode)->root;
8482         struct btrfs_key ins;
8483         u64 cur_offset = start;
8484         u64 i_size;
8485         u64 cur_bytes;
8486         int ret = 0;
8487         bool own_trans = true;
8488
8489         if (trans)
8490                 own_trans = false;
8491         while (num_bytes > 0) {
8492                 if (own_trans) {
8493                         trans = btrfs_start_transaction(root, 3);
8494                         if (IS_ERR(trans)) {
8495                                 ret = PTR_ERR(trans);
8496                                 break;
8497                         }
8498                 }
8499
8500                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8501                 cur_bytes = max(cur_bytes, min_size);
8502                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8503                                            *alloc_hint, &ins, 1);
8504                 if (ret) {
8505                         if (own_trans)
8506                                 btrfs_end_transaction(trans, root);
8507                         break;
8508                 }
8509
8510                 ret = insert_reserved_file_extent(trans, inode,
8511                                                   cur_offset, ins.objectid,
8512                                                   ins.offset, ins.offset,
8513                                                   ins.offset, 0, 0, 0,
8514                                                   BTRFS_FILE_EXTENT_PREALLOC);
8515                 if (ret) {
8516                         btrfs_free_reserved_extent(root, ins.objectid,
8517                                                    ins.offset);
8518                         btrfs_abort_transaction(trans, root, ret);
8519                         if (own_trans)
8520                                 btrfs_end_transaction(trans, root);
8521                         break;
8522                 }
8523                 btrfs_drop_extent_cache(inode, cur_offset,
8524                                         cur_offset + ins.offset -1, 0);
8525
8526                 em = alloc_extent_map();
8527                 if (!em) {
8528                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8529                                 &BTRFS_I(inode)->runtime_flags);
8530                         goto next;
8531                 }
8532
8533                 em->start = cur_offset;
8534                 em->orig_start = cur_offset;
8535                 em->len = ins.offset;
8536                 em->block_start = ins.objectid;
8537                 em->block_len = ins.offset;
8538                 em->orig_block_len = ins.offset;
8539                 em->ram_bytes = ins.offset;
8540                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8541                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8542                 em->generation = trans->transid;
8543
8544                 while (1) {
8545                         write_lock(&em_tree->lock);
8546                         ret = add_extent_mapping(em_tree, em, 1);
8547                         write_unlock(&em_tree->lock);
8548                         if (ret != -EEXIST)
8549                                 break;
8550                         btrfs_drop_extent_cache(inode, cur_offset,
8551                                                 cur_offset + ins.offset - 1,
8552                                                 0);
8553                 }
8554                 free_extent_map(em);
8555 next:
8556                 num_bytes -= ins.offset;
8557                 cur_offset += ins.offset;
8558                 *alloc_hint = ins.objectid + ins.offset;
8559
8560                 inode_inc_iversion(inode);
8561                 inode->i_ctime = CURRENT_TIME;
8562                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8563                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8564                     (actual_len > inode->i_size) &&
8565                     (cur_offset > inode->i_size)) {
8566                         if (cur_offset > actual_len)
8567                                 i_size = actual_len;
8568                         else
8569                                 i_size = cur_offset;
8570                         i_size_write(inode, i_size);
8571                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8572                 }
8573
8574                 ret = btrfs_update_inode(trans, root, inode);
8575
8576                 if (ret) {
8577                         btrfs_abort_transaction(trans, root, ret);
8578                         if (own_trans)
8579                                 btrfs_end_transaction(trans, root);
8580                         break;
8581                 }
8582
8583                 if (own_trans)
8584                         btrfs_end_transaction(trans, root);
8585         }
8586         return ret;
8587 }
8588
8589 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8590                               u64 start, u64 num_bytes, u64 min_size,
8591                               loff_t actual_len, u64 *alloc_hint)
8592 {
8593         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8594                                            min_size, actual_len, alloc_hint,
8595                                            NULL);
8596 }
8597
8598 int btrfs_prealloc_file_range_trans(struct inode *inode,
8599                                     struct btrfs_trans_handle *trans, int mode,
8600                                     u64 start, u64 num_bytes, u64 min_size,
8601                                     loff_t actual_len, u64 *alloc_hint)
8602 {
8603         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8604                                            min_size, actual_len, alloc_hint, trans);
8605 }
8606
8607 static int btrfs_set_page_dirty(struct page *page)
8608 {
8609         return __set_page_dirty_nobuffers(page);
8610 }
8611
8612 static int btrfs_permission(struct inode *inode, int mask)
8613 {
8614         struct btrfs_root *root = BTRFS_I(inode)->root;
8615         umode_t mode = inode->i_mode;
8616
8617         if (mask & MAY_WRITE &&
8618             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8619                 if (btrfs_root_readonly(root))
8620                         return -EROFS;
8621                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8622                         return -EACCES;
8623         }
8624         return generic_permission(inode, mask);
8625 }
8626
8627 static const struct inode_operations btrfs_dir_inode_operations = {
8628         .getattr        = btrfs_getattr,
8629         .lookup         = btrfs_lookup,
8630         .create         = btrfs_create,
8631         .unlink         = btrfs_unlink,
8632         .link           = btrfs_link,
8633         .mkdir          = btrfs_mkdir,
8634         .rmdir          = btrfs_rmdir,
8635         .rename         = btrfs_rename,
8636         .symlink        = btrfs_symlink,
8637         .setattr        = btrfs_setattr,
8638         .mknod          = btrfs_mknod,
8639         .setxattr       = btrfs_setxattr,
8640         .getxattr       = btrfs_getxattr,
8641         .listxattr      = btrfs_listxattr,
8642         .removexattr    = btrfs_removexattr,
8643         .permission     = btrfs_permission,
8644         .get_acl        = btrfs_get_acl,
8645         .update_time    = btrfs_update_time,
8646 };
8647 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8648         .lookup         = btrfs_lookup,
8649         .permission     = btrfs_permission,
8650         .get_acl        = btrfs_get_acl,
8651         .update_time    = btrfs_update_time,
8652 };
8653
8654 static const struct file_operations btrfs_dir_file_operations = {
8655         .llseek         = generic_file_llseek,
8656         .read           = generic_read_dir,
8657         .iterate        = btrfs_real_readdir,
8658         .unlocked_ioctl = btrfs_ioctl,
8659 #ifdef CONFIG_COMPAT
8660         .compat_ioctl   = btrfs_ioctl,
8661 #endif
8662         .release        = btrfs_release_file,
8663         .fsync          = btrfs_sync_file,
8664 };
8665
8666 static struct extent_io_ops btrfs_extent_io_ops = {
8667         .fill_delalloc = run_delalloc_range,
8668         .submit_bio_hook = btrfs_submit_bio_hook,
8669         .merge_bio_hook = btrfs_merge_bio_hook,
8670         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8671         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8672         .writepage_start_hook = btrfs_writepage_start_hook,
8673         .set_bit_hook = btrfs_set_bit_hook,
8674         .clear_bit_hook = btrfs_clear_bit_hook,
8675         .merge_extent_hook = btrfs_merge_extent_hook,
8676         .split_extent_hook = btrfs_split_extent_hook,
8677 };
8678
8679 /*
8680  * btrfs doesn't support the bmap operation because swapfiles
8681  * use bmap to make a mapping of extents in the file.  They assume
8682  * these extents won't change over the life of the file and they
8683  * use the bmap result to do IO directly to the drive.
8684  *
8685  * the btrfs bmap call would return logical addresses that aren't
8686  * suitable for IO and they also will change frequently as COW
8687  * operations happen.  So, swapfile + btrfs == corruption.
8688  *
8689  * For now we're avoiding this by dropping bmap.
8690  */
8691 static const struct address_space_operations btrfs_aops = {
8692         .readpage       = btrfs_readpage,
8693         .writepage      = btrfs_writepage,
8694         .writepages     = btrfs_writepages,
8695         .readpages      = btrfs_readpages,
8696         .direct_IO      = btrfs_direct_IO,
8697         .invalidatepage = btrfs_invalidatepage,
8698         .releasepage    = btrfs_releasepage,
8699         .set_page_dirty = btrfs_set_page_dirty,
8700         .error_remove_page = generic_error_remove_page,
8701 };
8702
8703 static const struct address_space_operations btrfs_symlink_aops = {
8704         .readpage       = btrfs_readpage,
8705         .writepage      = btrfs_writepage,
8706         .invalidatepage = btrfs_invalidatepage,
8707         .releasepage    = btrfs_releasepage,
8708 };
8709
8710 static const struct inode_operations btrfs_file_inode_operations = {
8711         .getattr        = btrfs_getattr,
8712         .setattr        = btrfs_setattr,
8713         .setxattr       = btrfs_setxattr,
8714         .getxattr       = btrfs_getxattr,
8715         .listxattr      = btrfs_listxattr,
8716         .removexattr    = btrfs_removexattr,
8717         .permission     = btrfs_permission,
8718         .fiemap         = btrfs_fiemap,
8719         .get_acl        = btrfs_get_acl,
8720         .update_time    = btrfs_update_time,
8721 };
8722 static const struct inode_operations btrfs_special_inode_operations = {
8723         .getattr        = btrfs_getattr,
8724         .setattr        = btrfs_setattr,
8725         .permission     = btrfs_permission,
8726         .setxattr       = btrfs_setxattr,
8727         .getxattr       = btrfs_getxattr,
8728         .listxattr      = btrfs_listxattr,
8729         .removexattr    = btrfs_removexattr,
8730         .get_acl        = btrfs_get_acl,
8731         .update_time    = btrfs_update_time,
8732 };
8733 static const struct inode_operations btrfs_symlink_inode_operations = {
8734         .readlink       = generic_readlink,
8735         .follow_link    = page_follow_link_light,
8736         .put_link       = page_put_link,
8737         .getattr        = btrfs_getattr,
8738         .setattr        = btrfs_setattr,
8739         .permission     = btrfs_permission,
8740         .setxattr       = btrfs_setxattr,
8741         .getxattr       = btrfs_getxattr,
8742         .listxattr      = btrfs_listxattr,
8743         .removexattr    = btrfs_removexattr,
8744         .get_acl        = btrfs_get_acl,
8745         .update_time    = btrfs_update_time,
8746 };
8747
8748 const struct dentry_operations btrfs_dentry_operations = {
8749         .d_delete       = btrfs_dentry_delete,
8750         .d_release      = btrfs_dentry_release,
8751 };