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 "ctree.h"
  47 #include "disk-io.h"
  48 #include "transaction.h"
  49 #include "btrfs_inode.h"
  50 #include "print-tree.h"
  51 #include "ordered-data.h"
  52 #include "xattr.h"
  53 #include "tree-log.h"
  54 #include "volumes.h"
  55 #include "compression.h"
  56 #include "locking.h"
  57 #include "free-space-cache.h"
  58 #include "inode-map.h"
  59 #include "backref.h"
  60 #include "hash.h"
  61
  62 struct btrfs_iget_args {
  63         u64 ino;
  64         struct btrfs_root *root;
  65 };
  66
  67 static const struct inode_operations btrfs_dir_inode_operations;
  68 static const struct inode_operations btrfs_symlink_inode_operations;
  69 static const struct inode_operations btrfs_dir_ro_inode_operations;
  70 static const struct inode_operations btrfs_special_inode_operations;
  71 static const struct inode_operations btrfs_file_inode_operations;
  72 static const struct address_space_operations btrfs_aops;
  73 static const struct address_space_operations btrfs_symlink_aops;
  74 static const struct file_operations btrfs_dir_file_operations;
  75 static struct extent_io_ops btrfs_extent_io_ops;
  76
  77 static struct kmem_cache *btrfs_inode_cachep;
  78 static struct kmem_cache *btrfs_delalloc_work_cachep;
  79 struct kmem_cache *btrfs_trans_handle_cachep;
  80 struct kmem_cache *btrfs_transaction_cachep;
  81 struct kmem_cache *btrfs_path_cachep;
  82 struct kmem_cache *btrfs_free_space_cachep;
  83
  84 #define S_SHIFT 12
  85 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  86         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  87         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  88         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  89         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  90         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  91         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  92         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  93 };
  94
  95 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  96 static int btrfs_truncate(struct inode *inode);
  97 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  98 static noinline int cow_file_range(struct inode *inode,
  99                                    struct page *locked_page,
 100                                    u64 start, u64 end, int *page_started,
 101                                    unsigned long *nr_written, int unlock);
 102 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
 103                                            u64 len, u64 orig_start,
 104                                            u64 block_start, u64 block_len,
 105                                            u64 orig_block_len, u64 ram_bytes,
 106                                            int type);
 107
 108 static int btrfs_dirty_inode(struct inode *inode);
 109
 110 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 111                                      struct inode *inode,  struct inode *dir,
 112                                      const struct qstr *qstr)
 113 {
 114         int err;
 115
 116         err = btrfs_init_acl(trans, inode, dir);
 117         if (!err)
 118                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 119         return err;
 120 }
 121
 122 /*
 123  * this does all the hard work for inserting an inline extent into
 124  * the btree.  The caller should have done a btrfs_drop_extents so that
 125  * no overlapping inline items exist in the btree
 126  */
 127 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 128                                 struct btrfs_root *root, struct inode *inode,
 129                                 u64 start, size_t size, size_t compressed_size,
 130                                 int compress_type,
 131                                 struct page **compressed_pages)
 132 {
 133         struct btrfs_key key;
 134         struct btrfs_path *path;
 135         struct extent_buffer *leaf;
 136         struct page *page = NULL;
 137         char *kaddr;
 138         unsigned long ptr;
 139         struct btrfs_file_extent_item *ei;
 140         int err = 0;
 141         int ret;
 142         size_t cur_size = size;
 143         size_t datasize;
 144         unsigned long offset;
 145
 146         if (compressed_size && compressed_pages)
 147                 cur_size = compressed_size;
 148
 149         path = btrfs_alloc_path();
 150         if (!path)
 151                 return -ENOMEM;
 152
 153         path->leave_spinning = 1;
 154
 155         key.objectid = btrfs_ino(inode);
 156         key.offset = start;
 157         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 158         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 159
 160         inode_add_bytes(inode, size);
 161         ret = btrfs_insert_empty_item(trans, root, path, &key,
 162                                       datasize);
 163         if (ret) {
 164                 err = ret;
 165                 goto fail;
 166         }
 167         leaf = path->nodes[0];
 168         ei = btrfs_item_ptr(leaf, path->slots[0],
 169                             struct btrfs_file_extent_item);
 170         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 171         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 172         btrfs_set_file_extent_encryption(leaf, ei, 0);
 173         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 174         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 175         ptr = btrfs_file_extent_inline_start(ei);
 176
 177         if (compress_type != BTRFS_COMPRESS_NONE) {
 178                 struct page *cpage;
 179                 int i = 0;
 180                 while (compressed_size > 0) {
 181                         cpage = compressed_pages[i];
 182                         cur_size = min_t(unsigned long, compressed_size,
 183                                        PAGE_CACHE_SIZE);
 184
 185                         kaddr = kmap_atomic(cpage);
 186                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 187                         kunmap_atomic(kaddr);
 188
 189                         i++;
 190                         ptr += cur_size;
 191                         compressed_size -= cur_size;
 192                 }
 193                 btrfs_set_file_extent_compression(leaf, ei,
 194                                                   compress_type);
 195         } else {
 196                 page = find_get_page(inode->i_mapping,
 197                                      start >> PAGE_CACHE_SHIFT);
 198                 btrfs_set_file_extent_compression(leaf, ei, 0);
 199                 kaddr = kmap_atomic(page);
 200                 offset = start & (PAGE_CACHE_SIZE - 1);
 201                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 202                 kunmap_atomic(kaddr);
 203                 page_cache_release(page);
 204         }
 205         btrfs_mark_buffer_dirty(leaf);
 206         btrfs_free_path(path);
 207
 208         /*
 209          * we're an inline extent, so nobody can
 210          * extend the file past i_size without locking
 211          * a page we already have locked.
 212          *
 213          * We must do any isize and inode updates
 214          * before we unlock the pages.  Otherwise we
 215          * could end up racing with unlink.
 216          */
 217         BTRFS_I(inode)->disk_i_size = inode->i_size;
 218         ret = btrfs_update_inode(trans, root, inode);
 219
 220         return ret;
 221 fail:
 222         btrfs_free_path(path);
 223         return err;
 224 }
 225
 226
 227 /*
 228  * conditionally insert an inline extent into the file.  This
 229  * does the checks required to make sure the data is small enough
 230  * to fit as an inline extent.
 231  */
 232 static noinline int cow_file_range_inline(struct btrfs_root *root,
 233                                           struct inode *inode, u64 start,
 234                                           u64 end, size_t compressed_size,
 235                                           int compress_type,
 236                                           struct page **compressed_pages)
 237 {
 238         struct btrfs_trans_handle *trans;
 239         u64 isize = i_size_read(inode);
 240         u64 actual_end = min(end + 1, isize);
 241         u64 inline_len = actual_end - start;
 242         u64 aligned_end = ALIGN(end, root->sectorsize);
 243         u64 data_len = inline_len;
 244         int ret;
 245
 246         if (compressed_size)
 247                 data_len = compressed_size;
 248
 249         if (start > 0 ||
 250             actual_end >= PAGE_CACHE_SIZE ||
 251             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 252             (!compressed_size &&
 253             (actual_end & (root->sectorsize - 1)) == 0) ||
 254             end + 1 < isize ||
 255             data_len > root->fs_info->max_inline) {
 256                 return 1;
 257         }
 258
 259         trans = btrfs_join_transaction(root);
 260         if (IS_ERR(trans))
 261                 return PTR_ERR(trans);
 262         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 263
 264         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 265         if (ret) {
 266                 btrfs_abort_transaction(trans, root, ret);
 267                 goto out;
 268         }
 269
 270         if (isize > actual_end)
 271                 inline_len = min_t(u64, isize, actual_end);
 272         ret = insert_inline_extent(trans, root, inode, start,
 273                                    inline_len, compressed_size,
 274                                    compress_type, compressed_pages);
 275         if (ret && ret != -ENOSPC) {
 276                 btrfs_abort_transaction(trans, root, ret);
 277                 goto out;
 278         } else if (ret == -ENOSPC) {
 279                 ret = 1;
 280                 goto out;
 281         }
 282
 283         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
 284         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 285         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 286 out:
 287         btrfs_end_transaction(trans, root);
 288         return ret;
 289 }
 290
 291 struct async_extent {
 292         u64 start;
 293         u64 ram_size;
 294         u64 compressed_size;
 295         struct page **pages;
 296         unsigned long nr_pages;
 297         int compress_type;
 298         struct list_head list;
 299 };
 300
 301 struct async_cow {
 302         struct inode *inode;
 303         struct btrfs_root *root;
 304         struct page *locked_page;
 305         u64 start;
 306         u64 end;
 307         struct list_head extents;
 308         struct btrfs_work work;
 309 };
 310
 311 static noinline int add_async_extent(struct async_cow *cow,
 312                                      u64 start, u64 ram_size,
 313                                      u64 compressed_size,
 314                                      struct page **pages,
 315                                      unsigned long nr_pages,
 316                                      int compress_type)
 317 {
 318         struct async_extent *async_extent;
 319
 320         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 321         BUG_ON(!async_extent); /* -ENOMEM */
 322         async_extent->start = start;
 323         async_extent->ram_size = ram_size;
 324         async_extent->compressed_size = compressed_size;
 325         async_extent->pages = pages;
 326         async_extent->nr_pages = nr_pages;
 327         async_extent->compress_type = compress_type;
 328         list_add_tail(&async_extent->list, &cow->extents);
 329         return 0;
 330 }
 331
 332 /*
 333  * we create compressed extents in two phases.  The first
 334  * phase compresses a range of pages that have already been
 335  * locked (both pages and state bits are locked).
 336  *
 337  * This is done inside an ordered work queue, and the compression
 338  * is spread across many cpus.  The actual IO submission is step
 339  * two, and the ordered work queue takes care of making sure that
 340  * happens in the same order things were put onto the queue by
 341  * writepages and friends.
 342  *
 343  * If this code finds it can't get good compression, it puts an
 344  * entry onto the work queue to write the uncompressed bytes.  This
 345  * makes sure that both compressed inodes and uncompressed inodes
 346  * are written in the same order that the flusher thread sent them
 347  * down.
 348  */
 349 static noinline int compress_file_range(struct inode *inode,
 350                                         struct page *locked_page,
 351                                         u64 start, u64 end,
 352                                         struct async_cow *async_cow,
 353                                         int *num_added)
 354 {
 355         struct btrfs_root *root = BTRFS_I(inode)->root;
 356         u64 num_bytes;
 357         u64 blocksize = root->sectorsize;
 358         u64 actual_end;
 359         u64 isize = i_size_read(inode);
 360         int ret = 0;
 361         struct page **pages = NULL;
 362         unsigned long nr_pages;
 363         unsigned long nr_pages_ret = 0;
 364         unsigned long total_compressed = 0;
 365         unsigned long total_in = 0;
 366         unsigned long max_compressed = 128 * 1024;
 367         unsigned long max_uncompressed = 128 * 1024;
 368         int i;
 369         int will_compress;
 370         int compress_type = root->fs_info->compress_type;
 371         int redirty = 0;
 372
 373         /* if this is a small write inside eof, kick off a defrag */
 374         if ((end - start + 1) < 16 * 1024 &&
 375             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 376                 btrfs_add_inode_defrag(NULL, inode);
 377
 378         actual_end = min_t(u64, isize, end + 1);
 379 again:
 380         will_compress = 0;
 381         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 382         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 383
 384         /*
 385          * we don't want to send crud past the end of i_size through
 386          * compression, that's just a waste of CPU time.  So, if the
 387          * end of the file is before the start of our current
 388          * requested range of bytes, we bail out to the uncompressed
 389          * cleanup code that can deal with all of this.
 390          *
 391          * It isn't really the fastest way to fix things, but this is a
 392          * very uncommon corner.
 393          */
 394         if (actual_end <= start)
 395                 goto cleanup_and_bail_uncompressed;
 396
 397         total_compressed = actual_end - start;
 398
 399         /* we want to make sure that amount of ram required to uncompress
 400          * an extent is reasonable, so we limit the total size in ram
 401          * of a compressed extent to 128k.  This is a crucial number
 402          * because it also controls how easily we can spread reads across
 403          * cpus for decompression.
 404          *
 405          * We also want to make sure the amount of IO required to do
 406          * a random read is reasonably small, so we limit the size of
 407          * a compressed extent to 128k.
 408          */
 409         total_compressed = min(total_compressed, max_uncompressed);
 410         num_bytes = ALIGN(end - start + 1, blocksize);
 411         num_bytes = max(blocksize,  num_bytes);
 412         total_in = 0;
 413         ret = 0;
 414
 415         /*
 416          * we do compression for mount -o compress and when the
 417          * inode has not been flagged as nocompress.  This flag can
 418          * change at any time if we discover bad compression ratios.
 419          */
 420         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 421             (btrfs_test_opt(root, COMPRESS) ||
 422              (BTRFS_I(inode)->force_compress) ||
 423              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 424                 WARN_ON(pages);
 425                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 426                 if (!pages) {
 427                         /* just bail out to the uncompressed code */
 428                         goto cont;
 429                 }
 430
 431                 if (BTRFS_I(inode)->force_compress)
 432                         compress_type = BTRFS_I(inode)->force_compress;
 433
 434                 /*
 435                  * we need to call clear_page_dirty_for_io on each
 436                  * page in the range.  Otherwise applications with the file
 437                  * mmap'd can wander in and change the page contents while
 438                  * we are compressing them.
 439                  *
 440                  * If the compression fails for any reason, we set the pages
 441                  * dirty again later on.
 442                  */
 443                 extent_range_clear_dirty_for_io(inode, start, end);
 444                 redirty = 1;
 445                 ret = btrfs_compress_pages(compress_type,
 446                                            inode->i_mapping, start,
 447                                            total_compressed, pages,
 448                                            nr_pages, &nr_pages_ret,
 449                                            &total_in,
 450                                            &total_compressed,
 451                                            max_compressed);
 452
 453                 if (!ret) {
 454                         unsigned long offset = total_compressed &
 455                                 (PAGE_CACHE_SIZE - 1);
 456                         struct page *page = pages[nr_pages_ret - 1];
 457                         char *kaddr;
 458
 459                         /* zero the tail end of the last page, we might be
 460                          * sending it down to disk
 461                          */
 462                         if (offset) {
 463                                 kaddr = kmap_atomic(page);
 464                                 memset(kaddr + offset, 0,
 465                                        PAGE_CACHE_SIZE - offset);
 466                                 kunmap_atomic(kaddr);
 467                         }
 468                         will_compress = 1;
 469                 }
 470         }
 471 cont:
 472         if (start == 0) {
 473                 /* lets try to make an inline extent */
 474                 if (ret || total_in < (actual_end - start)) {
 475                         /* we didn't compress the entire range, try
 476                          * to make an uncompressed inline extent.
 477                          */
 478                         ret = cow_file_range_inline(root, inode, start, end,
 479                                                     0, 0, NULL);
 480                 } else {
 481                         /* try making a compressed inline extent */
 482                         ret = cow_file_range_inline(root, inode, start, end,
 483                                                     total_compressed,
 484                                                     compress_type, pages);
 485                 }
 486                 if (ret <= 0) {
 487                         unsigned long clear_flags = EXTENT_DELALLOC |
 488                                 EXTENT_DEFRAG;
 489                         clear_flags |= (ret < 0) ? EXTENT_DO_ACCOUNTING : 0;
 490
 491                         /*
 492                          * inline extent creation worked or returned error,
 493                          * we don't need to create any more async work items.
 494                          * Unlock and free up our temp pages.
 495                          */
 496                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 497                                                      clear_flags, PAGE_UNLOCK |
 498                                                      PAGE_CLEAR_DIRTY |
 499                                                      PAGE_SET_WRITEBACK |
 500                                                      PAGE_END_WRITEBACK);
 501                         goto free_pages_out;
 502                 }
 503         }
 504
 505         if (will_compress) {
 506                 /*
 507                  * we aren't doing an inline extent round the compressed size
 508                  * up to a block size boundary so the allocator does sane
 509                  * things
 510                  */
 511                 total_compressed = ALIGN(total_compressed, blocksize);
 512
 513                 /*
 514                  * one last check to make sure the compression is really a
 515                  * win, compare the page count read with the blocks on disk
 516                  */
 517                 total_in = ALIGN(total_in, PAGE_CACHE_SIZE);
 518                 if (total_compressed >= total_in) {
 519                         will_compress = 0;
 520                 } else {
 521                         num_bytes = total_in;
 522                 }
 523         }
 524         if (!will_compress && pages) {
 525                 /*
 526                  * the compression code ran but failed to make things smaller,
 527                  * free any pages it allocated and our page pointer array
 528                  */
 529                 for (i = 0; i < nr_pages_ret; i++) {
 530                         WARN_ON(pages[i]->mapping);
 531                         page_cache_release(pages[i]);
 532                 }
 533                 kfree(pages);
 534                 pages = NULL;
 535                 total_compressed = 0;
 536                 nr_pages_ret = 0;
 537
 538                 /* flag the file so we don't compress in the future */
 539                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 540                     !(BTRFS_I(inode)->force_compress)) {
 541                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 542                 }
 543         }
 544         if (will_compress) {
 545                 *num_added += 1;
 546
 547                 /* the async work queues will take care of doing actual
 548                  * allocation on disk for these compressed pages,
 549                  * and will submit them to the elevator.
 550                  */
 551                 add_async_extent(async_cow, start, num_bytes,
 552                                  total_compressed, pages, nr_pages_ret,
 553                                  compress_type);
 554
 555                 if (start + num_bytes < end) {
 556                         start += num_bytes;
 557                         pages = NULL;
 558                         cond_resched();
 559                         goto again;
 560                 }
 561         } else {
 562 cleanup_and_bail_uncompressed:
 563                 /*
 564                  * No compression, but we still need to write the pages in
 565                  * the file we've been given so far.  redirty the locked
 566                  * page if it corresponds to our extent and set things up
 567                  * for the async work queue to run cow_file_range to do
 568                  * the normal delalloc dance
 569                  */
 570                 if (page_offset(locked_page) >= start &&
 571                     page_offset(locked_page) <= end) {
 572                         __set_page_dirty_nobuffers(locked_page);
 573                         /* unlocked later on in the async handlers */
 574                 }
 575                 if (redirty)
 576                         extent_range_redirty_for_io(inode, start, end);
 577                 add_async_extent(async_cow, start, end - start + 1,
 578                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 579                 *num_added += 1;
 580         }
 581
 582 out:
 583         return ret;
 584
 585 free_pages_out:
 586         for (i = 0; i < nr_pages_ret; i++) {
 587                 WARN_ON(pages[i]->mapping);
 588                 page_cache_release(pages[i]);
 589         }
 590         kfree(pages);
 591
 592         goto out;
 593 }
 594
 595 /*
 596  * phase two of compressed writeback.  This is the ordered portion
 597  * of the code, which only gets called in the order the work was
 598  * queued.  We walk all the async extents created by compress_file_range
 599  * and send them down to the disk.
 600  */
 601 static noinline int submit_compressed_extents(struct inode *inode,
 602                                               struct async_cow *async_cow)
 603 {
 604         struct async_extent *async_extent;
 605         u64 alloc_hint = 0;
 606         struct btrfs_key ins;
 607         struct extent_map *em;
 608         struct btrfs_root *root = BTRFS_I(inode)->root;
 609         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 610         struct extent_io_tree *io_tree;
 611         int ret = 0;
 612
 613         if (list_empty(&async_cow->extents))
 614                 return 0;
 615
 616 again:
 617         while (!list_empty(&async_cow->extents)) {
 618                 async_extent = list_entry(async_cow->extents.next,
 619                                           struct async_extent, list);
 620                 list_del(&async_extent->list);
 621
 622                 io_tree = &BTRFS_I(inode)->io_tree;
 623
 624 retry:
 625                 /* did the compression code fall back to uncompressed IO? */
 626                 if (!async_extent->pages) {
 627                         int page_started = 0;
 628                         unsigned long nr_written = 0;
 629
 630                         lock_extent(io_tree, async_extent->start,
 631                                          async_extent->start +
 632                                          async_extent->ram_size - 1);
 633
 634                         /* allocate blocks */
 635                         ret = cow_file_range(inode, async_cow->locked_page,
 636                                              async_extent->start,
 637                                              async_extent->start +
 638                                              async_extent->ram_size - 1,
 639                                              &page_started, &nr_written, 0);
 640
 641                         /* JDM XXX */
 642
 643                         /*
 644                          * if page_started, cow_file_range inserted an
 645                          * inline extent and took care of all the unlocking
 646                          * and IO for us.  Otherwise, we need to submit
 647                          * all those pages down to the drive.
 648                          */
 649                         if (!page_started && !ret)
 650                                 extent_write_locked_range(io_tree,
 651                                                   inode, async_extent->start,
 652                                                   async_extent->start +
 653                                                   async_extent->ram_size - 1,
 654                                                   btrfs_get_extent,
 655                                                   WB_SYNC_ALL);
 656                         else if (ret)
 657                                 unlock_page(async_cow->locked_page);
 658                         kfree(async_extent);
 659                         cond_resched();
 660                         continue;
 661                 }
 662
 663                 lock_extent(io_tree, async_extent->start,
 664                             async_extent->start + async_extent->ram_size - 1);
 665
 666                 ret = btrfs_reserve_extent(root,
 667                                            async_extent->compressed_size,
 668                                            async_extent->compressed_size,
 669                                            0, alloc_hint, &ins, 1);
 670                 if (ret) {
 671                         int i;
 672
 673                         for (i = 0; i < async_extent->nr_pages; i++) {
 674                                 WARN_ON(async_extent->pages[i]->mapping);
 675                                 page_cache_release(async_extent->pages[i]);
 676                         }
 677                         kfree(async_extent->pages);
 678                         async_extent->nr_pages = 0;
 679                         async_extent->pages = NULL;
 680
 681                         if (ret == -ENOSPC) {
 682                                 unlock_extent(io_tree, async_extent->start,
 683                                               async_extent->start +
 684                                               async_extent->ram_size - 1);
 685                                 goto retry;
 686                         }
 687                         goto out_free;
 688                 }
 689
 690                 /*
 691                  * here we're doing allocation and writeback of the
 692                  * compressed pages
 693                  */
 694                 btrfs_drop_extent_cache(inode, async_extent->start,
 695                                         async_extent->start +
 696                                         async_extent->ram_size - 1, 0);
 697
 698                 em = alloc_extent_map();
 699                 if (!em) {
 700                         ret = -ENOMEM;
 701                         goto out_free_reserve;
 702                 }
 703                 em->start = async_extent->start;
 704                 em->len = async_extent->ram_size;
 705                 em->orig_start = em->start;
 706                 em->mod_start = em->start;
 707                 em->mod_len = em->len;
 708
 709                 em->block_start = ins.objectid;
 710                 em->block_len = ins.offset;
 711                 em->orig_block_len = ins.offset;
 712                 em->ram_bytes = async_extent->ram_size;
 713                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 714                 em->compress_type = async_extent->compress_type;
 715                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 716                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 717                 em->generation = -1;
 718
 719                 while (1) {
 720                         write_lock(&em_tree->lock);
 721                         ret = add_extent_mapping(em_tree, em, 1);
 722                         write_unlock(&em_tree->lock);
 723                         if (ret != -EEXIST) {
 724                                 free_extent_map(em);
 725                                 break;
 726                         }
 727                         btrfs_drop_extent_cache(inode, async_extent->start,
 728                                                 async_extent->start +
 729                                                 async_extent->ram_size - 1, 0);
 730                 }
 731
 732                 if (ret)
 733                         goto out_free_reserve;
 734
 735                 ret = btrfs_add_ordered_extent_compress(inode,
 736                                                 async_extent->start,
 737                                                 ins.objectid,
 738                                                 async_extent->ram_size,
 739                                                 ins.offset,
 740                                                 BTRFS_ORDERED_COMPRESSED,
 741                                                 async_extent->compress_type);
 742                 if (ret)
 743                         goto out_free_reserve;
 744
 745                 /*
 746                  * clear dirty, set writeback and unlock the pages.
 747                  */
 748                 extent_clear_unlock_delalloc(inode, async_extent->start,
 749                                 async_extent->start +
 750                                 async_extent->ram_size - 1,
 751                                 NULL, EXTENT_LOCKED | EXTENT_DELALLOC,
 752                                 PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 753                                 PAGE_SET_WRITEBACK);
 754                 ret = btrfs_submit_compressed_write(inode,
 755                                     async_extent->start,
 756                                     async_extent->ram_size,
 757                                     ins.objectid,
 758                                     ins.offset, async_extent->pages,
 759                                     async_extent->nr_pages);
 760                 alloc_hint = ins.objectid + ins.offset;
 761                 kfree(async_extent);
 762                 if (ret)
 763                         goto out;
 764                 cond_resched();
 765         }
 766         ret = 0;
 767 out:
 768         return ret;
 769 out_free_reserve:
 770         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 771 out_free:
 772         extent_clear_unlock_delalloc(inode, async_extent->start,
 773                                      async_extent->start +
 774                                      async_extent->ram_size - 1,
 775                                      NULL, EXTENT_LOCKED | EXTENT_DELALLOC |
 776                                      EXTENT_DEFRAG | EXTENT_DO_ACCOUNTING,
 777                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 778                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 779         kfree(async_extent);
 780         goto again;
 781 }
 782
 783 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 784                                       u64 num_bytes)
 785 {
 786         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 787         struct extent_map *em;
 788         u64 alloc_hint = 0;
 789
 790         read_lock(&em_tree->lock);
 791         em = search_extent_mapping(em_tree, start, num_bytes);
 792         if (em) {
 793                 /*
 794                  * if block start isn't an actual block number then find the
 795                  * first block in this inode and use that as a hint.  If that
 796                  * block is also bogus then just don't worry about it.
 797                  */
 798                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 799                         free_extent_map(em);
 800                         em = search_extent_mapping(em_tree, 0, 0);
 801                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 802                                 alloc_hint = em->block_start;
 803                         if (em)
 804                                 free_extent_map(em);
 805                 } else {
 806                         alloc_hint = em->block_start;
 807                         free_extent_map(em);
 808                 }
 809         }
 810         read_unlock(&em_tree->lock);
 811
 812         return alloc_hint;
 813 }
 814
 815 /*
 816  * when extent_io.c finds a delayed allocation range in the file,
 817  * the call backs end up in this code.  The basic idea is to
 818  * allocate extents on disk for the range, and create ordered data structs
 819  * in ram to track those extents.
 820  *
 821  * locked_page is the page that writepage had locked already.  We use
 822  * it to make sure we don't do extra locks or unlocks.
 823  *
 824  * *page_started is set to one if we unlock locked_page and do everything
 825  * required to start IO on it.  It may be clean and already done with
 826  * IO when we return.
 827  */
 828 static noinline int cow_file_range(struct inode *inode,
 829                                    struct page *locked_page,
 830                                    u64 start, u64 end, int *page_started,
 831                                    unsigned long *nr_written,
 832                                    int unlock)
 833 {
 834         struct btrfs_root *root = BTRFS_I(inode)->root;
 835         u64 alloc_hint = 0;
 836         u64 num_bytes;
 837         unsigned long ram_size;
 838         u64 disk_num_bytes;
 839         u64 cur_alloc_size;
 840         u64 blocksize = root->sectorsize;
 841         struct btrfs_key ins;
 842         struct extent_map *em;
 843         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 844         int ret = 0;
 845
 846         if (btrfs_is_free_space_inode(inode)) {
 847                 WARN_ON_ONCE(1);
 848                 return -EINVAL;
 849         }
 850
 851         num_bytes = ALIGN(end - start + 1, blocksize);
 852         num_bytes = max(blocksize,  num_bytes);
 853         disk_num_bytes = num_bytes;
 854
 855         /* if this is a small write inside eof, kick off defrag */
 856         if (num_bytes < 64 * 1024 &&
 857             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 858                 btrfs_add_inode_defrag(NULL, inode);
 859
 860         if (start == 0) {
 861                 /* lets try to make an inline extent */
 862                 ret = cow_file_range_inline(root, inode, start, end, 0, 0,
 863                                             NULL);
 864                 if (ret == 0) {
 865                         extent_clear_unlock_delalloc(inode, start, end, NULL,
 866                                      EXTENT_LOCKED | EXTENT_DELALLOC |
 867                                      EXTENT_DEFRAG, PAGE_UNLOCK |
 868                                      PAGE_CLEAR_DIRTY | PAGE_SET_WRITEBACK |
 869                                      PAGE_END_WRITEBACK);
 870
 871                         *nr_written = *nr_written +
 872                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 873                         *page_started = 1;
 874                         goto out;
 875                 } else if (ret < 0) {
 876                         goto out_unlock;
 877                 }
 878         }
 879
 880         BUG_ON(disk_num_bytes >
 881                btrfs_super_total_bytes(root->fs_info->super_copy));
 882
 883         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 884         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 885
 886         while (disk_num_bytes > 0) {
 887                 unsigned long op;
 888
 889                 cur_alloc_size = disk_num_bytes;
 890                 ret = btrfs_reserve_extent(root, cur_alloc_size,
 891                                            root->sectorsize, 0, alloc_hint,
 892                                            &ins, 1);
 893                 if (ret < 0)
 894                         goto out_unlock;
 895
 896                 em = alloc_extent_map();
 897                 if (!em) {
 898                         ret = -ENOMEM;
 899                         goto out_reserve;
 900                 }
 901                 em->start = start;
 902                 em->orig_start = em->start;
 903                 ram_size = ins.offset;
 904                 em->len = ins.offset;
 905                 em->mod_start = em->start;
 906                 em->mod_len = em->len;
 907
 908                 em->block_start = ins.objectid;
 909                 em->block_len = ins.offset;
 910                 em->orig_block_len = ins.offset;
 911                 em->ram_bytes = ram_size;
 912                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 913                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 914                 em->generation = -1;
 915
 916                 while (1) {
 917                         write_lock(&em_tree->lock);
 918                         ret = add_extent_mapping(em_tree, em, 1);
 919                         write_unlock(&em_tree->lock);
 920                         if (ret != -EEXIST) {
 921                                 free_extent_map(em);
 922                                 break;
 923                         }
 924                         btrfs_drop_extent_cache(inode, start,
 925                                                 start + ram_size - 1, 0);
 926                 }
 927                 if (ret)
 928                         goto out_reserve;
 929
 930                 cur_alloc_size = ins.offset;
 931                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 932                                                ram_size, cur_alloc_size, 0);
 933                 if (ret)
 934                         goto out_reserve;
 935
 936                 if (root->root_key.objectid ==
 937                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 938                         ret = btrfs_reloc_clone_csums(inode, start,
 939                                                       cur_alloc_size);
 940                         if (ret)
 941                                 goto out_reserve;
 942                 }
 943
 944                 if (disk_num_bytes < cur_alloc_size)
 945                         break;
 946
 947                 /* we're not doing compressed IO, don't unlock the first
 948                  * page (which the caller expects to stay locked), don't
 949                  * clear any dirty bits and don't set any writeback bits
 950                  *
 951                  * Do set the Private2 bit so we know this page was properly
 952                  * setup for writepage
 953                  */
 954                 op = unlock ? PAGE_UNLOCK : 0;
 955                 op |= PAGE_SET_PRIVATE2;
 956
 957                 extent_clear_unlock_delalloc(inode, start,
 958                                              start + ram_size - 1, locked_page,
 959                                              EXTENT_LOCKED | EXTENT_DELALLOC,
 960                                              op);
 961                 disk_num_bytes -= cur_alloc_size;
 962                 num_bytes -= cur_alloc_size;
 963                 alloc_hint = ins.objectid + ins.offset;
 964                 start += cur_alloc_size;
 965         }
 966 out:
 967         return ret;
 968
 969 out_reserve:
 970         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 971 out_unlock:
 972         extent_clear_unlock_delalloc(inode, start, end, locked_page,
 973                                      EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
 974                                      EXTENT_DELALLOC | EXTENT_DEFRAG,
 975                                      PAGE_UNLOCK | PAGE_CLEAR_DIRTY |
 976                                      PAGE_SET_WRITEBACK | PAGE_END_WRITEBACK);
 977         goto out;
 978 }
 979
 980 /*
 981  * work queue call back to started compression on a file and pages
 982  */
 983 static noinline void async_cow_start(struct btrfs_work *work)
 984 {
 985         struct async_cow *async_cow;
 986         int num_added = 0;
 987         async_cow = container_of(work, struct async_cow, work);
 988
 989         compress_file_range(async_cow->inode, async_cow->locked_page,
 990                             async_cow->start, async_cow->end, async_cow,
 991                             &num_added);
 992         if (num_added == 0) {
 993                 btrfs_add_delayed_iput(async_cow->inode);
 994                 async_cow->inode = NULL;
 995         }
 996 }
 997
 998 /*
 999  * work queue call back to submit previously compressed pages
1000  */
1001 static noinline void async_cow_submit(struct btrfs_work *work)
1002 {
1003         struct async_cow *async_cow;
1004         struct btrfs_root *root;
1005         unsigned long nr_pages;
1006
1007         async_cow = container_of(work, struct async_cow, work);
1008
1009         root = async_cow->root;
1010         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1011                 PAGE_CACHE_SHIFT;
1012
1013         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1014             5 * 1024 * 1024 &&
1015             waitqueue_active(&root->fs_info->async_submit_wait))
1016                 wake_up(&root->fs_info->async_submit_wait);
1017
1018         if (async_cow->inode)
1019                 submit_compressed_extents(async_cow->inode, async_cow);
1020 }
1021
1022 static noinline void async_cow_free(struct btrfs_work *work)
1023 {
1024         struct async_cow *async_cow;
1025         async_cow = container_of(work, struct async_cow, work);
1026         if (async_cow->inode)
1027                 btrfs_add_delayed_iput(async_cow->inode);
1028         kfree(async_cow);
1029 }
1030
1031 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1032                                 u64 start, u64 end, int *page_started,
1033                                 unsigned long *nr_written)
1034 {
1035         struct async_cow *async_cow;
1036         struct btrfs_root *root = BTRFS_I(inode)->root;
1037         unsigned long nr_pages;
1038         u64 cur_end;
1039         int limit = 10 * 1024 * 1024;
1040
1041         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1042                          1, 0, NULL, GFP_NOFS);
1043         while (start < end) {
1044                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1045                 BUG_ON(!async_cow); /* -ENOMEM */
1046                 async_cow->inode = igrab(inode);
1047                 async_cow->root = root;
1048                 async_cow->locked_page = locked_page;
1049                 async_cow->start = start;
1050
1051                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1052                         cur_end = end;
1053                 else
1054                         cur_end = min(end, start + 512 * 1024 - 1);
1055
1056                 async_cow->end = cur_end;
1057                 INIT_LIST_HEAD(&async_cow->extents);
1058
1059                 async_cow->work.func = async_cow_start;
1060                 async_cow->work.ordered_func = async_cow_submit;
1061                 async_cow->work.ordered_free = async_cow_free;
1062                 async_cow->work.flags = 0;
1063
1064                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1065                         PAGE_CACHE_SHIFT;
1066                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1067
1068                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1069                                    &async_cow->work);
1070
1071                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1072                         wait_event(root->fs_info->async_submit_wait,
1073                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1074                             limit));
1075                 }
1076
1077                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1078                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1079                         wait_event(root->fs_info->async_submit_wait,
1080                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1081                            0));
1082                 }
1083
1084                 *nr_written += nr_pages;
1085                 start = cur_end + 1;
1086         }
1087         *page_started = 1;
1088         return 0;
1089 }
1090
1091 static noinline int csum_exist_in_range(struct btrfs_root *root,
1092                                         u64 bytenr, u64 num_bytes)
1093 {
1094         int ret;
1095         struct btrfs_ordered_sum *sums;
1096         LIST_HEAD(list);
1097
1098         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1099                                        bytenr + num_bytes - 1, &list, 0);
1100         if (ret == 0 && list_empty(&list))
1101                 return 0;
1102
1103         while (!list_empty(&list)) {
1104                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1105                 list_del(&sums->list);
1106                 kfree(sums);
1107         }
1108         return 1;
1109 }
1110
1111 /*
1112  * when nowcow writeback call back.  This checks for snapshots or COW copies
1113  * of the extents that exist in the file, and COWs the file as required.
1114  *
1115  * If no cow copies or snapshots exist, we write directly to the existing
1116  * blocks on disk
1117  */
1118 static noinline int run_delalloc_nocow(struct inode *inode,
1119                                        struct page *locked_page,
1120                               u64 start, u64 end, int *page_started, int force,
1121                               unsigned long *nr_written)
1122 {
1123         struct btrfs_root *root = BTRFS_I(inode)->root;
1124         struct btrfs_trans_handle *trans;
1125         struct extent_buffer *leaf;
1126         struct btrfs_path *path;
1127         struct btrfs_file_extent_item *fi;
1128         struct btrfs_key found_key;
1129         u64 cow_start;
1130         u64 cur_offset;
1131         u64 extent_end;
1132         u64 extent_offset;
1133         u64 disk_bytenr;
1134         u64 num_bytes;
1135         u64 disk_num_bytes;
1136         u64 ram_bytes;
1137         int extent_type;
1138         int ret, err;
1139         int type;
1140         int nocow;
1141         int check_prev = 1;
1142         bool nolock;
1143         u64 ino = btrfs_ino(inode);
1144
1145         path = btrfs_alloc_path();
1146         if (!path) {
1147                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1148                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1149                                              EXTENT_DO_ACCOUNTING |
1150                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1151                                              PAGE_CLEAR_DIRTY |
1152                                              PAGE_SET_WRITEBACK |
1153                                              PAGE_END_WRITEBACK);
1154                 return -ENOMEM;
1155         }
1156
1157         nolock = btrfs_is_free_space_inode(inode);
1158
1159         if (nolock)
1160                 trans = btrfs_join_transaction_nolock(root);
1161         else
1162                 trans = btrfs_join_transaction(root);
1163
1164         if (IS_ERR(trans)) {
1165                 extent_clear_unlock_delalloc(inode, start, end, locked_page,
1166                                              EXTENT_LOCKED | EXTENT_DELALLOC |
1167                                              EXTENT_DO_ACCOUNTING |
1168                                              EXTENT_DEFRAG, PAGE_UNLOCK |
1169                                              PAGE_CLEAR_DIRTY |
1170                                              PAGE_SET_WRITEBACK |
1171                                              PAGE_END_WRITEBACK);
1172                 btrfs_free_path(path);
1173                 return PTR_ERR(trans);
1174         }
1175
1176         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1177
1178         cow_start = (u64)-1;
1179         cur_offset = start;
1180         while (1) {
1181                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1182                                                cur_offset, 0);
1183                 if (ret < 0)
1184                         goto error;
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                                 goto error;
1200                         if (ret > 0)
1201                                 break;
1202                         leaf = path->nodes[0];
1203                 }
1204
1205                 nocow = 0;
1206                 disk_bytenr = 0;
1207                 num_bytes = 0;
1208                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1209
1210                 if (found_key.objectid > ino ||
1211                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1212                     found_key.offset > end)
1213                         break;
1214
1215                 if (found_key.offset > cur_offset) {
1216                         extent_end = found_key.offset;
1217                         extent_type = 0;
1218                         goto out_check;
1219                 }
1220
1221                 fi = btrfs_item_ptr(leaf, path->slots[0],
1222                                     struct btrfs_file_extent_item);
1223                 extent_type = btrfs_file_extent_type(leaf, fi);
1224
1225                 ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
1226                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1227                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1228                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1229                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1230                         extent_end = found_key.offset +
1231                                 btrfs_file_extent_num_bytes(leaf, fi);
1232                         disk_num_bytes =
1233                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1234                         if (extent_end <= start) {
1235                                 path->slots[0]++;
1236                                 goto next_slot;
1237                         }
1238                         if (disk_bytenr == 0)
1239                                 goto out_check;
1240                         if (btrfs_file_extent_compression(leaf, fi) ||
1241                             btrfs_file_extent_encryption(leaf, fi) ||
1242                             btrfs_file_extent_other_encoding(leaf, fi))
1243                                 goto out_check;
1244                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1245                                 goto out_check;
1246                         if (btrfs_extent_readonly(root, disk_bytenr))
1247                                 goto out_check;
1248                         if (btrfs_cross_ref_exist(trans, root, ino,
1249                                                   found_key.offset -
1250                                                   extent_offset, disk_bytenr))
1251                                 goto out_check;
1252                         disk_bytenr += extent_offset;
1253                         disk_bytenr += cur_offset - found_key.offset;
1254                         num_bytes = min(end + 1, extent_end) - cur_offset;
1255                         /*
1256                          * force cow if csum exists in the range.
1257                          * this ensure that csum for a given extent are
1258                          * either valid or do not exist.
1259                          */
1260                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1261                                 goto out_check;
1262                         nocow = 1;
1263                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1264                         extent_end = found_key.offset +
1265                                 btrfs_file_extent_inline_len(leaf, fi);
1266                         extent_end = ALIGN(extent_end, root->sectorsize);
1267                 } else {
1268                         BUG_ON(1);
1269                 }
1270 out_check:
1271                 if (extent_end <= start) {
1272                         path->slots[0]++;
1273                         goto next_slot;
1274                 }
1275                 if (!nocow) {
1276                         if (cow_start == (u64)-1)
1277                                 cow_start = cur_offset;
1278                         cur_offset = extent_end;
1279                         if (cur_offset > end)
1280                                 break;
1281                         path->slots[0]++;
1282                         goto next_slot;
1283                 }
1284
1285                 btrfs_release_path(path);
1286                 if (cow_start != (u64)-1) {
1287                         ret = cow_file_range(inode, locked_page,
1288                                              cow_start, found_key.offset - 1,
1289                                              page_started, nr_written, 1);
1290                         if (ret)
1291                                 goto error;
1292                         cow_start = (u64)-1;
1293                 }
1294
1295                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1296                         struct extent_map *em;
1297                         struct extent_map_tree *em_tree;
1298                         em_tree = &BTRFS_I(inode)->extent_tree;
1299                         em = alloc_extent_map();
1300                         BUG_ON(!em); /* -ENOMEM */
1301                         em->start = cur_offset;
1302                         em->orig_start = found_key.offset - extent_offset;
1303                         em->len = num_bytes;
1304                         em->block_len = num_bytes;
1305                         em->block_start = disk_bytenr;
1306                         em->orig_block_len = disk_num_bytes;
1307                         em->ram_bytes = ram_bytes;
1308                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1309                         em->mod_start = em->start;
1310                         em->mod_len = em->len;
1311                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1312                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1313                         em->generation = -1;
1314                         while (1) {
1315                                 write_lock(&em_tree->lock);
1316                                 ret = add_extent_mapping(em_tree, em, 1);
1317                                 write_unlock(&em_tree->lock);
1318                                 if (ret != -EEXIST) {
1319                                         free_extent_map(em);
1320                                         break;
1321                                 }
1322                                 btrfs_drop_extent_cache(inode, em->start,
1323                                                 em->start + em->len - 1, 0);
1324                         }
1325                         type = BTRFS_ORDERED_PREALLOC;
1326                 } else {
1327                         type = BTRFS_ORDERED_NOCOW;
1328                 }
1329
1330                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1331                                                num_bytes, num_bytes, type);
1332                 BUG_ON(ret); /* -ENOMEM */
1333
1334                 if (root->root_key.objectid ==
1335                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1336                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1337                                                       num_bytes);
1338                         if (ret)
1339                                 goto error;
1340                 }
1341
1342                 extent_clear_unlock_delalloc(inode, cur_offset,
1343                                              cur_offset + num_bytes - 1,
1344                                              locked_page, EXTENT_LOCKED |
1345                                              EXTENT_DELALLOC, PAGE_UNLOCK |
1346                                              PAGE_SET_PRIVATE2);
1347                 cur_offset = extent_end;
1348                 if (cur_offset > end)
1349                         break;
1350         }
1351         btrfs_release_path(path);
1352
1353         if (cur_offset <= end && cow_start == (u64)-1) {
1354                 cow_start = cur_offset;
1355                 cur_offset = end;
1356         }
1357
1358         if (cow_start != (u64)-1) {
1359                 ret = cow_file_range(inode, locked_page, cow_start, end,
1360                                      page_started, nr_written, 1);
1361                 if (ret)
1362                         goto error;
1363         }
1364
1365 error:
1366         err = btrfs_end_transaction(trans, root);
1367         if (!ret)
1368                 ret = err;
1369
1370         if (ret && cur_offset < end)
1371                 extent_clear_unlock_delalloc(inode, cur_offset, end,
1372                                              locked_page, EXTENT_LOCKED |
1373                                              EXTENT_DELALLOC | EXTENT_DEFRAG |
1374                                              EXTENT_DO_ACCOUNTING, PAGE_UNLOCK |
1375                                              PAGE_CLEAR_DIRTY |
1376                                              PAGE_SET_WRITEBACK |
1377                                              PAGE_END_WRITEBACK);
1378         btrfs_free_path(path);
1379         return ret;
1380 }
1381
1382 /*
1383  * extent_io.c call back to do delayed allocation processing
1384  */
1385 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1386                               u64 start, u64 end, int *page_started,
1387                               unsigned long *nr_written)
1388 {
1389         int ret;
1390         struct btrfs_root *root = BTRFS_I(inode)->root;
1391
1392         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1393                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1394                                          page_started, 1, nr_written);
1395         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1396                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1397                                          page_started, 0, nr_written);
1398         } else if (!btrfs_test_opt(root, COMPRESS) &&
1399                    !(BTRFS_I(inode)->force_compress) &&
1400                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1401                 ret = cow_file_range(inode, locked_page, start, end,
1402                                       page_started, nr_written, 1);
1403         } else {
1404                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1405                         &BTRFS_I(inode)->runtime_flags);
1406                 ret = cow_file_range_async(inode, locked_page, start, end,
1407                                            page_started, nr_written);
1408         }
1409         return ret;
1410 }
1411
1412 static void btrfs_split_extent_hook(struct inode *inode,
1413                                     struct extent_state *orig, u64 split)
1414 {
1415         /* not delalloc, ignore it */
1416         if (!(orig->state & EXTENT_DELALLOC))
1417                 return;
1418
1419         spin_lock(&BTRFS_I(inode)->lock);
1420         BTRFS_I(inode)->outstanding_extents++;
1421         spin_unlock(&BTRFS_I(inode)->lock);
1422 }
1423
1424 /*
1425  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1426  * extents so we can keep track of new extents that are just merged onto old
1427  * extents, such as when we are doing sequential writes, so we can properly
1428  * account for the metadata space we'll need.
1429  */
1430 static void btrfs_merge_extent_hook(struct inode *inode,
1431                                     struct extent_state *new,
1432                                     struct extent_state *other)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(other->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents--;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 static void btrfs_add_delalloc_inodes(struct btrfs_root *root,
1444                                       struct inode *inode)
1445 {
1446         spin_lock(&root->delalloc_lock);
1447         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1448                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1449                               &root->delalloc_inodes);
1450                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1451                         &BTRFS_I(inode)->runtime_flags);
1452                 root->nr_delalloc_inodes++;
1453                 if (root->nr_delalloc_inodes == 1) {
1454                         spin_lock(&root->fs_info->delalloc_root_lock);
1455                         BUG_ON(!list_empty(&root->delalloc_root));
1456                         list_add_tail(&root->delalloc_root,
1457                                       &root->fs_info->delalloc_roots);
1458                         spin_unlock(&root->fs_info->delalloc_root_lock);
1459                 }
1460         }
1461         spin_unlock(&root->delalloc_lock);
1462 }
1463
1464 static void btrfs_del_delalloc_inode(struct btrfs_root *root,
1465                                      struct inode *inode)
1466 {
1467         spin_lock(&root->delalloc_lock);
1468         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1469                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1470                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1471                           &BTRFS_I(inode)->runtime_flags);
1472                 root->nr_delalloc_inodes--;
1473                 if (!root->nr_delalloc_inodes) {
1474                         spin_lock(&root->fs_info->delalloc_root_lock);
1475                         BUG_ON(list_empty(&root->delalloc_root));
1476                         list_del_init(&root->delalloc_root);
1477                         spin_unlock(&root->fs_info->delalloc_root_lock);
1478                 }
1479         }
1480         spin_unlock(&root->delalloc_lock);
1481 }
1482
1483 /*
1484  * extent_io.c set_bit_hook, used to track delayed allocation
1485  * bytes in this file, and to maintain the list of inodes that
1486  * have pending delalloc work to be done.
1487  */
1488 static void btrfs_set_bit_hook(struct inode *inode,
1489                                struct extent_state *state, unsigned long *bits)
1490 {
1491
1492         /*
1493          * set_bit and clear bit hooks normally require _irqsave/restore
1494          * but in this case, we are only testing for the DELALLOC
1495          * bit, which is only set or cleared with irqs on
1496          */
1497         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1498                 struct btrfs_root *root = BTRFS_I(inode)->root;
1499                 u64 len = state->end + 1 - state->start;
1500                 bool do_list = !btrfs_is_free_space_inode(inode);
1501
1502                 if (*bits & EXTENT_FIRST_DELALLOC) {
1503                         *bits &= ~EXTENT_FIRST_DELALLOC;
1504                 } else {
1505                         spin_lock(&BTRFS_I(inode)->lock);
1506                         BTRFS_I(inode)->outstanding_extents++;
1507                         spin_unlock(&BTRFS_I(inode)->lock);
1508                 }
1509
1510                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1511                                      root->fs_info->delalloc_batch);
1512                 spin_lock(&BTRFS_I(inode)->lock);
1513                 BTRFS_I(inode)->delalloc_bytes += len;
1514                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1515                                          &BTRFS_I(inode)->runtime_flags))
1516                         btrfs_add_delalloc_inodes(root, inode);
1517                 spin_unlock(&BTRFS_I(inode)->lock);
1518         }
1519 }
1520
1521 /*
1522  * extent_io.c clear_bit_hook, see set_bit_hook for why
1523  */
1524 static void btrfs_clear_bit_hook(struct inode *inode,
1525                                  struct extent_state *state,
1526                                  unsigned long *bits)
1527 {
1528         /*
1529          * set_bit and clear bit hooks normally require _irqsave/restore
1530          * but in this case, we are only testing for the DELALLOC
1531          * bit, which is only set or cleared with irqs on
1532          */
1533         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1534                 struct btrfs_root *root = BTRFS_I(inode)->root;
1535                 u64 len = state->end + 1 - state->start;
1536                 bool do_list = !btrfs_is_free_space_inode(inode);
1537
1538                 if (*bits & EXTENT_FIRST_DELALLOC) {
1539                         *bits &= ~EXTENT_FIRST_DELALLOC;
1540                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1541                         spin_lock(&BTRFS_I(inode)->lock);
1542                         BTRFS_I(inode)->outstanding_extents--;
1543                         spin_unlock(&BTRFS_I(inode)->lock);
1544                 }
1545
1546                 /*
1547                  * We don't reserve metadata space for space cache inodes so we
1548                  * don't need to call dellalloc_release_metadata if there is an
1549                  * error.
1550                  */
1551                 if (*bits & EXTENT_DO_ACCOUNTING &&
1552                     root != root->fs_info->tree_root)
1553                         btrfs_delalloc_release_metadata(inode, len);
1554
1555                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1556                     && do_list && !(state->state & EXTENT_NORESERVE))
1557                         btrfs_free_reserved_data_space(inode, len);
1558
1559                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1560                                      root->fs_info->delalloc_batch);
1561                 spin_lock(&BTRFS_I(inode)->lock);
1562                 BTRFS_I(inode)->delalloc_bytes -= len;
1563                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1564                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1565                              &BTRFS_I(inode)->runtime_flags))
1566                         btrfs_del_delalloc_inode(root, inode);
1567                 spin_unlock(&BTRFS_I(inode)->lock);
1568         }
1569 }
1570
1571 /*
1572  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1573  * we don't create bios that span stripes or chunks
1574  */
1575 int btrfs_merge_bio_hook(int rw, struct page *page, unsigned long offset,
1576                          size_t size, struct bio *bio,
1577                          unsigned long bio_flags)
1578 {
1579         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1580         u64 logical = (u64)bio->bi_sector << 9;
1581         u64 length = 0;
1582         u64 map_length;
1583         int ret;
1584
1585         if (bio_flags & EXTENT_BIO_COMPRESSED)
1586                 return 0;
1587
1588         length = bio->bi_size;
1589         map_length = length;
1590         ret = btrfs_map_block(root->fs_info, rw, logical,
1591                               &map_length, NULL, 0);
1592         /* Will always return 0 with map_multi == NULL */
1593         BUG_ON(ret < 0);
1594         if (map_length < length + size)
1595                 return 1;
1596         return 0;
1597 }
1598
1599 /*
1600  * in order to insert checksums into the metadata in large chunks,
1601  * we wait until bio submission time.   All the pages in the bio are
1602  * checksummed and sums are attached onto the ordered extent record.
1603  *
1604  * At IO completion time the cums attached on the ordered extent record
1605  * are inserted into the btree
1606  */
1607 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1608                                     struct bio *bio, int mirror_num,
1609                                     unsigned long bio_flags,
1610                                     u64 bio_offset)
1611 {
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613         int ret = 0;
1614
1615         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1616         BUG_ON(ret); /* -ENOMEM */
1617         return 0;
1618 }
1619
1620 /*
1621  * in order to insert checksums into the metadata in large chunks,
1622  * we wait until bio submission time.   All the pages in the bio are
1623  * checksummed and sums are attached onto the ordered extent record.
1624  *
1625  * At IO completion time the cums attached on the ordered extent record
1626  * are inserted into the btree
1627  */
1628 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1629                           int mirror_num, unsigned long bio_flags,
1630                           u64 bio_offset)
1631 {
1632         struct btrfs_root *root = BTRFS_I(inode)->root;
1633         int ret;
1634
1635         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1636         if (ret)
1637                 bio_endio(bio, ret);
1638         return ret;
1639 }
1640
1641 /*
1642  * extent_io.c submission hook. This does the right thing for csum calculation
1643  * on write, or reading the csums from the tree before a read
1644  */
1645 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1646                           int mirror_num, unsigned long bio_flags,
1647                           u64 bio_offset)
1648 {
1649         struct btrfs_root *root = BTRFS_I(inode)->root;
1650         int ret = 0;
1651         int skip_sum;
1652         int metadata = 0;
1653         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1654
1655         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1656
1657         if (btrfs_is_free_space_inode(inode))
1658                 metadata = 2;
1659
1660         if (!(rw & REQ_WRITE)) {
1661                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1662                 if (ret)
1663                         goto out;
1664
1665                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1666                         ret = btrfs_submit_compressed_read(inode, bio,
1667                                                            mirror_num,
1668                                                            bio_flags);
1669                         goto out;
1670                 } else if (!skip_sum) {
1671                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1672                         if (ret)
1673                                 goto out;
1674                 }
1675                 goto mapit;
1676         } else if (async && !skip_sum) {
1677                 /* csum items have already been cloned */
1678                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1679                         goto mapit;
1680                 /* we're doing a write, do the async checksumming */
1681                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1682                                    inode, rw, bio, mirror_num,
1683                                    bio_flags, bio_offset,
1684                                    __btrfs_submit_bio_start,
1685                                    __btrfs_submit_bio_done);
1686                 goto out;
1687         } else if (!skip_sum) {
1688                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1689                 if (ret)
1690                         goto out;
1691         }
1692
1693 mapit:
1694         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1695
1696 out:
1697         if (ret < 0)
1698                 bio_endio(bio, ret);
1699         return ret;
1700 }
1701
1702 /*
1703  * given a list of ordered sums record them in the inode.  This happens
1704  * at IO completion time based on sums calculated at bio submission time.
1705  */
1706 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1707                              struct inode *inode, u64 file_offset,
1708                              struct list_head *list)
1709 {
1710         struct btrfs_ordered_sum *sum;
1711
1712         list_for_each_entry(sum, list, list) {
1713                 trans->adding_csums = 1;
1714                 btrfs_csum_file_blocks(trans,
1715                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1716                 trans->adding_csums = 0;
1717         }
1718         return 0;
1719 }
1720
1721 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722                               struct extent_state **cached_state)
1723 {
1724         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1725         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1726                                    cached_state, GFP_NOFS);
1727 }
1728
1729 /* see btrfs_writepage_start_hook for details on why this is required */
1730 struct btrfs_writepage_fixup {
1731         struct page *page;
1732         struct btrfs_work work;
1733 };
1734
1735 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1736 {
1737         struct btrfs_writepage_fixup *fixup;
1738         struct btrfs_ordered_extent *ordered;
1739         struct extent_state *cached_state = NULL;
1740         struct page *page;
1741         struct inode *inode;
1742         u64 page_start;
1743         u64 page_end;
1744         int ret;
1745
1746         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1747         page = fixup->page;
1748 again:
1749         lock_page(page);
1750         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751                 ClearPageChecked(page);
1752                 goto out_page;
1753         }
1754
1755         inode = page->mapping->host;
1756         page_start = page_offset(page);
1757         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1758
1759         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1760                          &cached_state);
1761
1762         /* already ordered? We're done */
1763         if (PagePrivate2(page))
1764                 goto out;
1765
1766         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1767         if (ordered) {
1768                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769                                      page_end, &cached_state, GFP_NOFS);
1770                 unlock_page(page);
1771                 btrfs_start_ordered_extent(inode, ordered, 1);
1772                 btrfs_put_ordered_extent(ordered);
1773                 goto again;
1774         }
1775
1776         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1777         if (ret) {
1778                 mapping_set_error(page->mapping, ret);
1779                 end_extent_writepage(page, ret, page_start, page_end);
1780                 ClearPageChecked(page);
1781                 goto out;
1782          }
1783
1784         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1785         ClearPageChecked(page);
1786         set_page_dirty(page);
1787 out:
1788         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789                              &cached_state, GFP_NOFS);
1790 out_page:
1791         unlock_page(page);
1792         page_cache_release(page);
1793         kfree(fixup);
1794 }
1795
1796 /*
1797  * There are a few paths in the higher layers of the kernel that directly
1798  * set the page dirty bit without asking the filesystem if it is a
1799  * good idea.  This causes problems because we want to make sure COW
1800  * properly happens and the data=ordered rules are followed.
1801  *
1802  * In our case any range that doesn't have the ORDERED bit set
1803  * hasn't been properly setup for IO.  We kick off an async process
1804  * to fix it up.  The async helper will wait for ordered extents, set
1805  * the delalloc bit and make it safe to write the page.
1806  */
1807 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1808 {
1809         struct inode *inode = page->mapping->host;
1810         struct btrfs_writepage_fixup *fixup;
1811         struct btrfs_root *root = BTRFS_I(inode)->root;
1812
1813         /* this page is properly in the ordered list */
1814         if (TestClearPagePrivate2(page))
1815                 return 0;
1816
1817         if (PageChecked(page))
1818                 return -EAGAIN;
1819
1820         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1821         if (!fixup)
1822                 return -EAGAIN;
1823
1824         SetPageChecked(page);
1825         page_cache_get(page);
1826         fixup->work.func = btrfs_writepage_fixup_worker;
1827         fixup->page = page;
1828         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1829         return -EBUSY;
1830 }
1831
1832 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833                                        struct inode *inode, u64 file_pos,
1834                                        u64 disk_bytenr, u64 disk_num_bytes,
1835                                        u64 num_bytes, u64 ram_bytes,
1836                                        u8 compression, u8 encryption,
1837                                        u16 other_encoding, int extent_type)
1838 {
1839         struct btrfs_root *root = BTRFS_I(inode)->root;
1840         struct btrfs_file_extent_item *fi;
1841         struct btrfs_path *path;
1842         struct extent_buffer *leaf;
1843         struct btrfs_key ins;
1844         int ret;
1845
1846         path = btrfs_alloc_path();
1847         if (!path)
1848                 return -ENOMEM;
1849
1850         path->leave_spinning = 1;
1851
1852         /*
1853          * we may be replacing one extent in the tree with another.
1854          * The new extent is pinned in the extent map, and we don't want
1855          * to drop it from the cache until it is completely in the btree.
1856          *
1857          * So, tell btrfs_drop_extents to leave this extent in the cache.
1858          * the caller is expected to unpin it and allow it to be merged
1859          * with the others.
1860          */
1861         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1862                                  file_pos + num_bytes, 0);
1863         if (ret)
1864                 goto out;
1865
1866         ins.objectid = btrfs_ino(inode);
1867         ins.offset = file_pos;
1868         ins.type = BTRFS_EXTENT_DATA_KEY;
1869         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1870         if (ret)
1871                 goto out;
1872         leaf = path->nodes[0];
1873         fi = btrfs_item_ptr(leaf, path->slots[0],
1874                             struct btrfs_file_extent_item);
1875         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876         btrfs_set_file_extent_type(leaf, fi, extent_type);
1877         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879         btrfs_set_file_extent_offset(leaf, fi, 0);
1880         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882         btrfs_set_file_extent_compression(leaf, fi, compression);
1883         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1885
1886         btrfs_mark_buffer_dirty(leaf);
1887         btrfs_release_path(path);
1888
1889         inode_add_bytes(inode, num_bytes);
1890
1891         ins.objectid = disk_bytenr;
1892         ins.offset = disk_num_bytes;
1893         ins.type = BTRFS_EXTENT_ITEM_KEY;
1894         ret = btrfs_alloc_reserved_file_extent(trans, root,
1895                                         root->root_key.objectid,
1896                                         btrfs_ino(inode), file_pos, &ins);
1897 out:
1898         btrfs_free_path(path);
1899
1900         return ret;
1901 }
1902
1903 /* snapshot-aware defrag */
1904 struct sa_defrag_extent_backref {
1905         struct rb_node node;
1906         struct old_sa_defrag_extent *old;
1907         u64 root_id;
1908         u64 inum;
1909         u64 file_pos;
1910         u64 extent_offset;
1911         u64 num_bytes;
1912         u64 generation;
1913 };
1914
1915 struct old_sa_defrag_extent {
1916         struct list_head list;
1917         struct new_sa_defrag_extent *new;
1918
1919         u64 extent_offset;
1920         u64 bytenr;
1921         u64 offset;
1922         u64 len;
1923         int count;
1924 };
1925
1926 struct new_sa_defrag_extent {
1927         struct rb_root root;
1928         struct list_head head;
1929         struct btrfs_path *path;
1930         struct inode *inode;
1931         u64 file_pos;
1932         u64 len;
1933         u64 bytenr;
1934         u64 disk_len;
1935         u8 compress_type;
1936 };
1937
1938 static int backref_comp(struct sa_defrag_extent_backref *b1,
1939                         struct sa_defrag_extent_backref *b2)
1940 {
1941         if (b1->root_id < b2->root_id)
1942                 return -1;
1943         else if (b1->root_id > b2->root_id)
1944                 return 1;
1945
1946         if (b1->inum < b2->inum)
1947                 return -1;
1948         else if (b1->inum > b2->inum)
1949                 return 1;
1950
1951         if (b1->file_pos < b2->file_pos)
1952                 return -1;
1953         else if (b1->file_pos > b2->file_pos)
1954                 return 1;
1955
1956         /*
1957          * [------------------------------] ===> (a range of space)
1958          *     |<--->|   |<---->| =============> (fs/file tree A)
1959          * |<---------------------------->| ===> (fs/file tree B)
1960          *
1961          * A range of space can refer to two file extents in one tree while
1962          * refer to only one file extent in another tree.
1963          *
1964          * So we may process a disk offset more than one time(two extents in A)
1965          * and locate at the same extent(one extent in B), then insert two same
1966          * backrefs(both refer to the extent in B).
1967          */
1968         return 0;
1969 }
1970
1971 static void backref_insert(struct rb_root *root,
1972                            struct sa_defrag_extent_backref *backref)
1973 {
1974         struct rb_node **p = &root->rb_node;
1975         struct rb_node *parent = NULL;
1976         struct sa_defrag_extent_backref *entry;
1977         int ret;
1978
1979         while (*p) {
1980                 parent = *p;
1981                 entry = rb_entry(parent, struct sa_defrag_extent_backref, node);
1982
1983                 ret = backref_comp(backref, entry);
1984                 if (ret < 0)
1985                         p = &(*p)->rb_left;
1986                 else
1987                         p = &(*p)->rb_right;
1988         }
1989
1990         rb_link_node(&backref->node, parent, p);
1991         rb_insert_color(&backref->node, root);
1992 }
1993
1994 /*
1995  * Note the backref might has changed, and in this case we just return 0.
1996  */
1997 static noinline int record_one_backref(u64 inum, u64 offset, u64 root_id,
1998                                        void *ctx)
1999 {
2000         struct btrfs_file_extent_item *extent;
2001         struct btrfs_fs_info *fs_info;
2002         struct old_sa_defrag_extent *old = ctx;
2003         struct new_sa_defrag_extent *new = old->new;
2004         struct btrfs_path *path = new->path;
2005         struct btrfs_key key;
2006         struct btrfs_root *root;
2007         struct sa_defrag_extent_backref *backref;
2008         struct extent_buffer *leaf;
2009         struct inode *inode = new->inode;
2010         int slot;
2011         int ret;
2012         u64 extent_offset;
2013         u64 num_bytes;
2014
2015         if (BTRFS_I(inode)->root->root_key.objectid == root_id &&
2016             inum == btrfs_ino(inode))
2017                 return 0;
2018
2019         key.objectid = root_id;
2020         key.type = BTRFS_ROOT_ITEM_KEY;
2021         key.offset = (u64)-1;
2022
2023         fs_info = BTRFS_I(inode)->root->fs_info;
2024         root = btrfs_read_fs_root_no_name(fs_info, &key);
2025         if (IS_ERR(root)) {
2026                 if (PTR_ERR(root) == -ENOENT)
2027                         return 0;
2028                 WARN_ON(1);
2029                 pr_debug("inum=%llu, offset=%llu, root_id=%llu\n",
2030                          inum, offset, root_id);
2031                 return PTR_ERR(root);
2032         }
2033
2034         key.objectid = inum;
2035         key.type = BTRFS_EXTENT_DATA_KEY;
2036         if (offset > (u64)-1 << 32)
2037                 key.offset = 0;
2038         else
2039                 key.offset = offset;
2040
2041         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2042         if (WARN_ON(ret < 0))
2043                 return ret;
2044         ret = 0;
2045
2046         while (1) {
2047                 cond_resched();
2048
2049                 leaf = path->nodes[0];
2050                 slot = path->slots[0];
2051
2052                 if (slot >= btrfs_header_nritems(leaf)) {
2053                         ret = btrfs_next_leaf(root, path);
2054                         if (ret < 0) {
2055                                 goto out;
2056                         } else if (ret > 0) {
2057                                 ret = 0;
2058                                 goto out;
2059                         }
2060                         continue;
2061                 }
2062
2063                 path->slots[0]++;
2064
2065                 btrfs_item_key_to_cpu(leaf, &key, slot);
2066
2067                 if (key.objectid > inum)
2068                         goto out;
2069
2070                 if (key.objectid < inum || key.type != BTRFS_EXTENT_DATA_KEY)
2071                         continue;
2072
2073                 extent = btrfs_item_ptr(leaf, slot,
2074                                         struct btrfs_file_extent_item);
2075
2076                 if (btrfs_file_extent_disk_bytenr(leaf, extent) != old->bytenr)
2077                         continue;
2078
2079                 /*
2080                  * 'offset' refers to the exact key.offset,
2081                  * NOT the 'offset' field in btrfs_extent_data_ref, ie.
2082                  * (key.offset - extent_offset).
2083                  */
2084                 if (key.offset != offset)
2085                         continue;
2086
2087                 extent_offset = btrfs_file_extent_offset(leaf, extent);
2088                 num_bytes = btrfs_file_extent_num_bytes(leaf, extent);
2089
2090                 if (extent_offset >= old->extent_offset + old->offset +
2091                     old->len || extent_offset + num_bytes <=
2092                     old->extent_offset + old->offset)
2093                         continue;
2094                 break;
2095         }
2096
2097         backref = kmalloc(sizeof(*backref), GFP_NOFS);
2098         if (!backref) {
2099                 ret = -ENOENT;
2100                 goto out;
2101         }
2102
2103         backref->root_id = root_id;
2104         backref->inum = inum;
2105         backref->file_pos = offset;
2106         backref->num_bytes = num_bytes;
2107         backref->extent_offset = extent_offset;
2108         backref->generation = btrfs_file_extent_generation(leaf, extent);
2109         backref->old = old;
2110         backref_insert(&new->root, backref);
2111         old->count++;
2112 out:
2113         btrfs_release_path(path);
2114         WARN_ON(ret);
2115         return ret;
2116 }
2117
2118 static noinline bool record_extent_backrefs(struct btrfs_path *path,
2119                                    struct new_sa_defrag_extent *new)
2120 {
2121         struct btrfs_fs_info *fs_info = BTRFS_I(new->inode)->root->fs_info;
2122         struct old_sa_defrag_extent *old, *tmp;
2123         int ret;
2124
2125         new->path = path;
2126
2127         list_for_each_entry_safe(old, tmp, &new->head, list) {
2128                 ret = iterate_inodes_from_logical(old->bytenr +
2129                                                   old->extent_offset, fs_info,
2130                                                   path, record_one_backref,
2131                                                   old);
2132                 if (ret < 0 && ret != -ENOENT)
2133                         return false;
2134
2135                 /* no backref to be processed for this extent */
2136                 if (!old->count) {
2137                         list_del(&old->list);
2138                         kfree(old);
2139                 }
2140         }
2141
2142         if (list_empty(&new->head))
2143                 return false;
2144
2145         return true;
2146 }
2147
2148 static int relink_is_mergable(struct extent_buffer *leaf,
2149                               struct btrfs_file_extent_item *fi,
2150                               struct new_sa_defrag_extent *new)
2151 {
2152         if (btrfs_file_extent_disk_bytenr(leaf, fi) != new->bytenr)
2153                 return 0;
2154
2155         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2156                 return 0;
2157
2158         if (btrfs_file_extent_compression(leaf, fi) != new->compress_type)
2159                 return 0;
2160
2161         if (btrfs_file_extent_encryption(leaf, fi) ||
2162             btrfs_file_extent_other_encoding(leaf, fi))
2163                 return 0;
2164
2165         return 1;
2166 }
2167
2168 /*
2169  * Note the backref might has changed, and in this case we just return 0.
2170  */
2171 static noinline int relink_extent_backref(struct btrfs_path *path,
2172                                  struct sa_defrag_extent_backref *prev,
2173                                  struct sa_defrag_extent_backref *backref)
2174 {
2175         struct btrfs_file_extent_item *extent;
2176         struct btrfs_file_extent_item *item;
2177         struct btrfs_ordered_extent *ordered;
2178         struct btrfs_trans_handle *trans;
2179         struct btrfs_fs_info *fs_info;
2180         struct btrfs_root *root;
2181         struct btrfs_key key;
2182         struct extent_buffer *leaf;
2183         struct old_sa_defrag_extent *old = backref->old;
2184         struct new_sa_defrag_extent *new = old->new;
2185         struct inode *src_inode = new->inode;
2186         struct inode *inode;
2187         struct extent_state *cached = NULL;
2188         int ret = 0;
2189         u64 start;
2190         u64 len;
2191         u64 lock_start;
2192         u64 lock_end;
2193         bool merge = false;
2194         int index;
2195
2196         if (prev && prev->root_id == backref->root_id &&
2197             prev->inum == backref->inum &&
2198             prev->file_pos + prev->num_bytes == backref->file_pos)
2199                 merge = true;
2200
2201         /* step 1: get root */
2202         key.objectid = backref->root_id;
2203         key.type = BTRFS_ROOT_ITEM_KEY;
2204         key.offset = (u64)-1;
2205
2206         fs_info = BTRFS_I(src_inode)->root->fs_info;
2207         index = srcu_read_lock(&fs_info->subvol_srcu);
2208
2209         root = btrfs_read_fs_root_no_name(fs_info, &key);
2210         if (IS_ERR(root)) {
2211                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2212                 if (PTR_ERR(root) == -ENOENT)
2213                         return 0;
2214                 return PTR_ERR(root);
2215         }
2216
2217         /* step 2: get inode */
2218         key.objectid = backref->inum;
2219         key.type = BTRFS_INODE_ITEM_KEY;
2220         key.offset = 0;
2221
2222         inode = btrfs_iget(fs_info->sb, &key, root, NULL);
2223         if (IS_ERR(inode)) {
2224                 srcu_read_unlock(&fs_info->subvol_srcu, index);
2225                 return 0;
2226         }
2227
2228         srcu_read_unlock(&fs_info->subvol_srcu, index);
2229
2230         /* step 3: relink backref */
2231         lock_start = backref->file_pos;
2232         lock_end = backref->file_pos + backref->num_bytes - 1;
2233         lock_extent_bits(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2234                          0, &cached);
2235
2236         ordered = btrfs_lookup_first_ordered_extent(inode, lock_end);
2237         if (ordered) {
2238                 btrfs_put_ordered_extent(ordered);
2239                 goto out_unlock;
2240         }
2241
2242         trans = btrfs_join_transaction(root);
2243         if (IS_ERR(trans)) {
2244                 ret = PTR_ERR(trans);
2245                 goto out_unlock;
2246         }
2247
2248         key.objectid = backref->inum;
2249         key.type = BTRFS_EXTENT_DATA_KEY;
2250         key.offset = backref->file_pos;
2251
2252         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2253         if (ret < 0) {
2254                 goto out_free_path;
2255         } else if (ret > 0) {
2256                 ret = 0;
2257                 goto out_free_path;
2258         }
2259
2260         extent = btrfs_item_ptr(path->nodes[0], path->slots[0],
2261                                 struct btrfs_file_extent_item);
2262
2263         if (btrfs_file_extent_generation(path->nodes[0], extent) !=
2264             backref->generation)
2265                 goto out_free_path;
2266
2267         btrfs_release_path(path);
2268
2269         start = backref->file_pos;
2270         if (backref->extent_offset < old->extent_offset + old->offset)
2271                 start += old->extent_offset + old->offset -
2272                          backref->extent_offset;
2273
2274         len = min(backref->extent_offset + backref->num_bytes,
2275                   old->extent_offset + old->offset + old->len);
2276         len -= max(backref->extent_offset, old->extent_offset + old->offset);
2277
2278         ret = btrfs_drop_extents(trans, root, inode, start,
2279                                  start + len, 1);
2280         if (ret)
2281                 goto out_free_path;
2282 again:
2283         key.objectid = btrfs_ino(inode);
2284         key.type = BTRFS_EXTENT_DATA_KEY;
2285         key.offset = start;
2286
2287         path->leave_spinning = 1;
2288         if (merge) {
2289                 struct btrfs_file_extent_item *fi;
2290                 u64 extent_len;
2291                 struct btrfs_key found_key;
2292
2293                 ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
2294                 if (ret < 0)
2295                         goto out_free_path;
2296
2297                 path->slots[0]--;
2298                 leaf = path->nodes[0];
2299                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2300
2301                 fi = btrfs_item_ptr(leaf, path->slots[0],
2302                                     struct btrfs_file_extent_item);
2303                 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
2304
2305                 if (extent_len + found_key.offset == start &&
2306                     relink_is_mergable(leaf, fi, new)) {
2307                         btrfs_set_file_extent_num_bytes(leaf, fi,
2308                                                         extent_len + len);
2309                         btrfs_mark_buffer_dirty(leaf);
2310                         inode_add_bytes(inode, len);
2311
2312                         ret = 1;
2313                         goto out_free_path;
2314                 } else {
2315                         merge = false;
2316                         btrfs_release_path(path);
2317                         goto again;
2318                 }
2319         }
2320
2321         ret = btrfs_insert_empty_item(trans, root, path, &key,
2322                                         sizeof(*extent));
2323         if (ret) {
2324                 btrfs_abort_transaction(trans, root, ret);
2325                 goto out_free_path;
2326         }
2327
2328         leaf = path->nodes[0];
2329         item = btrfs_item_ptr(leaf, path->slots[0],
2330                                 struct btrfs_file_extent_item);
2331         btrfs_set_file_extent_disk_bytenr(leaf, item, new->bytenr);
2332         btrfs_set_file_extent_disk_num_bytes(leaf, item, new->disk_len);
2333         btrfs_set_file_extent_offset(leaf, item, start - new->file_pos);
2334         btrfs_set_file_extent_num_bytes(leaf, item, len);
2335         btrfs_set_file_extent_ram_bytes(leaf, item, new->len);
2336         btrfs_set_file_extent_generation(leaf, item, trans->transid);
2337         btrfs_set_file_extent_type(leaf, item, BTRFS_FILE_EXTENT_REG);
2338         btrfs_set_file_extent_compression(leaf, item, new->compress_type);
2339         btrfs_set_file_extent_encryption(leaf, item, 0);
2340         btrfs_set_file_extent_other_encoding(leaf, item, 0);
2341
2342         btrfs_mark_buffer_dirty(leaf);
2343         inode_add_bytes(inode, len);
2344         btrfs_release_path(path);
2345
2346         ret = btrfs_inc_extent_ref(trans, root, new->bytenr,
2347                         new->disk_len, 0,
2348                         backref->root_id, backref->inum,
2349                         new->file_pos, 0);      /* start - extent_offset */
2350         if (ret) {
2351                 btrfs_abort_transaction(trans, root, ret);
2352                 goto out_free_path;
2353         }
2354
2355         ret = 1;
2356 out_free_path:
2357         btrfs_release_path(path);
2358         path->leave_spinning = 0;
2359         btrfs_end_transaction(trans, root);
2360 out_unlock:
2361         unlock_extent_cached(&BTRFS_I(inode)->io_tree, lock_start, lock_end,
2362                              &cached, GFP_NOFS);
2363         iput(inode);
2364         return ret;
2365 }
2366
2367 static void free_sa_defrag_extent(struct new_sa_defrag_extent *new)
2368 {
2369         struct old_sa_defrag_extent *old, *tmp;
2370
2371         if (!new)
2372                 return;
2373
2374         list_for_each_entry_safe(old, tmp, &new->head, list) {
2375                 list_del(&old->list);
2376                 kfree(old);
2377         }
2378         kfree(new);
2379 }
2380
2381 static void relink_file_extents(struct new_sa_defrag_extent *new)
2382 {
2383         struct btrfs_path *path;
2384         struct sa_defrag_extent_backref *backref;
2385         struct sa_defrag_extent_backref *prev = NULL;
2386         struct inode *inode;
2387         struct btrfs_root *root;
2388         struct rb_node *node;
2389         int ret;
2390
2391         inode = new->inode;
2392         root = BTRFS_I(inode)->root;
2393
2394         path = btrfs_alloc_path();
2395         if (!path)
2396                 return;
2397
2398         if (!record_extent_backrefs(path, new)) {
2399                 btrfs_free_path(path);
2400                 goto out;
2401         }
2402         btrfs_release_path(path);
2403
2404         while (1) {
2405                 node = rb_first(&new->root);
2406                 if (!node)
2407                         break;
2408                 rb_erase(node, &new->root);
2409
2410                 backref = rb_entry(node, struct sa_defrag_extent_backref, node);
2411
2412                 ret = relink_extent_backref(path, prev, backref);
2413                 WARN_ON(ret < 0);
2414
2415                 kfree(prev);
2416
2417                 if (ret == 1)
2418                         prev = backref;
2419                 else
2420                         prev = NULL;
2421                 cond_resched();
2422         }
2423         kfree(prev);
2424
2425         btrfs_free_path(path);
2426 out:
2427         free_sa_defrag_extent(new);
2428
2429         atomic_dec(&root->fs_info->defrag_running);
2430         wake_up(&root->fs_info->transaction_wait);
2431 }
2432
2433 static struct new_sa_defrag_extent *
2434 record_old_file_extents(struct inode *inode,
2435                         struct btrfs_ordered_extent *ordered)
2436 {
2437         struct btrfs_root *root = BTRFS_I(inode)->root;
2438         struct btrfs_path *path;
2439         struct btrfs_key key;
2440         struct old_sa_defrag_extent *old;
2441         struct new_sa_defrag_extent *new;
2442         int ret;
2443
2444         new = kmalloc(sizeof(*new), GFP_NOFS);
2445         if (!new)
2446                 return NULL;
2447
2448         new->inode = inode;
2449         new->file_pos = ordered->file_offset;
2450         new->len = ordered->len;
2451         new->bytenr = ordered->start;
2452         new->disk_len = ordered->disk_len;
2453         new->compress_type = ordered->compress_type;
2454         new->root = RB_ROOT;
2455         INIT_LIST_HEAD(&new->head);
2456
2457         path = btrfs_alloc_path();
2458         if (!path)
2459                 goto out_kfree;
2460
2461         key.objectid = btrfs_ino(inode);
2462         key.type = BTRFS_EXTENT_DATA_KEY;
2463         key.offset = new->file_pos;
2464
2465         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2466         if (ret < 0)
2467                 goto out_free_path;
2468         if (ret > 0 && path->slots[0] > 0)
2469                 path->slots[0]--;
2470
2471         /* find out all the old extents for the file range */
2472         while (1) {
2473                 struct btrfs_file_extent_item *extent;
2474                 struct extent_buffer *l;
2475                 int slot;
2476                 u64 num_bytes;
2477                 u64 offset;
2478                 u64 end;
2479                 u64 disk_bytenr;
2480                 u64 extent_offset;
2481
2482                 l = path->nodes[0];
2483                 slot = path->slots[0];
2484
2485                 if (slot >= btrfs_header_nritems(l)) {
2486                         ret = btrfs_next_leaf(root, path);
2487                         if (ret < 0)
2488                                 goto out_free_path;
2489                         else if (ret > 0)
2490                                 break;
2491                         continue;
2492                 }
2493
2494                 btrfs_item_key_to_cpu(l, &key, slot);
2495
2496                 if (key.objectid != btrfs_ino(inode))
2497                         break;
2498                 if (key.type != BTRFS_EXTENT_DATA_KEY)
2499                         break;
2500                 if (key.offset >= new->file_pos + new->len)
2501                         break;
2502
2503                 extent = btrfs_item_ptr(l, slot, struct btrfs_file_extent_item);
2504
2505                 num_bytes = btrfs_file_extent_num_bytes(l, extent);
2506                 if (key.offset + num_bytes < new->file_pos)
2507                         goto next;
2508
2509                 disk_bytenr = btrfs_file_extent_disk_bytenr(l, extent);
2510                 if (!disk_bytenr)
2511                         goto next;
2512
2513                 extent_offset = btrfs_file_extent_offset(l, extent);
2514
2515                 old = kmalloc(sizeof(*old), GFP_NOFS);
2516                 if (!old)
2517                         goto out_free_path;
2518
2519                 offset = max(new->file_pos, key.offset);
2520                 end = min(new->file_pos + new->len, key.offset + num_bytes);
2521
2522                 old->bytenr = disk_bytenr;
2523                 old->extent_offset = extent_offset;
2524                 old->offset = offset - key.offset;
2525                 old->len = end - offset;
2526                 old->new = new;
2527                 old->count = 0;
2528                 list_add_tail(&old->list, &new->head);
2529 next:
2530                 path->slots[0]++;
2531                 cond_resched();
2532         }
2533
2534         btrfs_free_path(path);
2535         atomic_inc(&root->fs_info->defrag_running);
2536
2537         return new;
2538
2539 out_free_path:
2540         btrfs_free_path(path);
2541 out_kfree:
2542         free_sa_defrag_extent(new);
2543         return NULL;
2544 }
2545
2546 /* as ordered data IO finishes, this gets called so we can finish
2547  * an ordered extent if the range of bytes in the file it covers are
2548  * fully written.
2549  */
2550 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
2551 {
2552         struct inode *inode = ordered_extent->inode;
2553         struct btrfs_root *root = BTRFS_I(inode)->root;
2554         struct btrfs_trans_handle *trans = NULL;
2555         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2556         struct extent_state *cached_state = NULL;
2557         struct new_sa_defrag_extent *new = NULL;
2558         int compress_type = 0;
2559         int ret = 0;
2560         u64 logical_len = ordered_extent->len;
2561         bool nolock;
2562         bool truncated = false;
2563
2564         nolock = btrfs_is_free_space_inode(inode);
2565
2566         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
2567                 ret = -EIO;
2568                 goto out;
2569         }
2570
2571         if (test_bit(BTRFS_ORDERED_TRUNCATED, &ordered_extent->flags)) {
2572                 truncated = true;
2573                 logical_len = ordered_extent->truncated_len;
2574                 /* Truncated the entire extent, don't bother adding */
2575                 if (!logical_len)
2576                         goto out;
2577         }
2578
2579         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
2580                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
2581                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2582                 if (nolock)
2583                         trans = btrfs_join_transaction_nolock(root);
2584                 else
2585                         trans = btrfs_join_transaction(root);
2586                 if (IS_ERR(trans)) {
2587                         ret = PTR_ERR(trans);
2588                         trans = NULL;
2589                         goto out;
2590                 }
2591                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2592                 ret = btrfs_update_inode_fallback(trans, root, inode);
2593                 if (ret) /* -ENOMEM or corruption */
2594                         btrfs_abort_transaction(trans, root, ret);
2595                 goto out;
2596         }
2597
2598         lock_extent_bits(io_tree, ordered_extent->file_offset,
2599                          ordered_extent->file_offset + ordered_extent->len - 1,
2600                          0, &cached_state);
2601
2602         ret = test_range_bit(io_tree, ordered_extent->file_offset,
2603                         ordered_extent->file_offset + ordered_extent->len - 1,
2604                         EXTENT_DEFRAG, 1, cached_state);
2605         if (ret) {
2606                 u64 last_snapshot = btrfs_root_last_snapshot(&root->root_item);
2607                 if (last_snapshot >= BTRFS_I(inode)->generation)
2608                         /* the inode is shared */
2609                         new = record_old_file_extents(inode, ordered_extent);
2610
2611                 clear_extent_bit(io_tree, ordered_extent->file_offset,
2612                         ordered_extent->file_offset + ordered_extent->len - 1,
2613                         EXTENT_DEFRAG, 0, 0, &cached_state, GFP_NOFS);
2614         }
2615
2616         if (nolock)
2617                 trans = btrfs_join_transaction_nolock(root);
2618         else
2619                 trans = btrfs_join_transaction(root);
2620         if (IS_ERR(trans)) {
2621                 ret = PTR_ERR(trans);
2622                 trans = NULL;
2623                 goto out_unlock;
2624         }
2625         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
2626
2627         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
2628                 compress_type = ordered_extent->compress_type;
2629         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
2630                 BUG_ON(compress_type);
2631                 ret = btrfs_mark_extent_written(trans, inode,
2632                                                 ordered_extent->file_offset,
2633                                                 ordered_extent->file_offset +
2634                                                 logical_len);
2635         } else {
2636                 BUG_ON(root == root->fs_info->tree_root);
2637                 ret = insert_reserved_file_extent(trans, inode,
2638                                                 ordered_extent->file_offset,
2639                                                 ordered_extent->start,
2640                                                 ordered_extent->disk_len,
2641                                                 logical_len, logical_len,
2642                                                 compress_type, 0, 0,
2643                                                 BTRFS_FILE_EXTENT_REG);
2644         }
2645         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2646                            ordered_extent->file_offset, ordered_extent->len,
2647                            trans->transid);
2648         if (ret < 0) {
2649                 btrfs_abort_transaction(trans, root, ret);
2650                 goto out_unlock;
2651         }
2652
2653         add_pending_csums(trans, inode, ordered_extent->file_offset,
2654                           &ordered_extent->list);
2655
2656         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2657         ret = btrfs_update_inode_fallback(trans, root, inode);
2658         if (ret) { /* -ENOMEM or corruption */
2659                 btrfs_abort_transaction(trans, root, ret);
2660                 goto out_unlock;
2661         }
2662         ret = 0;
2663 out_unlock:
2664         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2665                              ordered_extent->file_offset +
2666                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2667 out:
2668         if (root != root->fs_info->tree_root)
2669                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2670         if (trans)
2671                 btrfs_end_transaction(trans, root);
2672
2673         if (ret || truncated) {
2674                 u64 start, end;
2675
2676                 if (truncated)
2677                         start = ordered_extent->file_offset + logical_len;
2678                 else
2679                         start = ordered_extent->file_offset;
2680                 end = ordered_extent->file_offset + ordered_extent->len - 1;
2681                 clear_extent_uptodate(io_tree, start, end, NULL, GFP_NOFS);
2682
2683                 /* Drop the cache for the part of the extent we didn't write. */
2684                 btrfs_drop_extent_cache(inode, start, end, 0);
2685
2686                 /*
2687                  * If the ordered extent had an IOERR or something else went
2688                  * wrong we need to return the space for this ordered extent
2689                  * back to the allocator.  We only free the extent in the
2690                  * truncated case if we didn't write out the extent at all.
2691                  */
2692                 if ((ret || !logical_len) &&
2693                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2694                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2695                         btrfs_free_reserved_extent(root, ordered_extent->start,
2696                                                    ordered_extent->disk_len);
2697         }
2698
2699
2700         /*
2701          * This needs to be done to make sure anybody waiting knows we are done
2702          * updating everything for this ordered extent.
2703          */
2704         btrfs_remove_ordered_extent(inode, ordered_extent);
2705
2706         /* for snapshot-aware defrag */
2707         if (new) {
2708                 if (ret) {
2709                         free_sa_defrag_extent(new);
2710                         atomic_dec(&root->fs_info->defrag_running);
2711                 } else {
2712                         relink_file_extents(new);
2713                 }
2714         }
2715
2716         /* once for us */
2717         btrfs_put_ordered_extent(ordered_extent);
2718         /* once for the tree */
2719         btrfs_put_ordered_extent(ordered_extent);
2720
2721         return ret;
2722 }
2723
2724 static void finish_ordered_fn(struct btrfs_work *work)
2725 {
2726         struct btrfs_ordered_extent *ordered_extent;
2727         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2728         btrfs_finish_ordered_io(ordered_extent);
2729 }
2730
2731 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2732                                 struct extent_state *state, int uptodate)
2733 {
2734         struct inode *inode = page->mapping->host;
2735         struct btrfs_root *root = BTRFS_I(inode)->root;
2736         struct btrfs_ordered_extent *ordered_extent = NULL;
2737         struct btrfs_workers *workers;
2738
2739         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2740
2741         ClearPagePrivate2(page);
2742         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2743                                             end - start + 1, uptodate))
2744                 return 0;
2745
2746         ordered_extent->work.func = finish_ordered_fn;
2747         ordered_extent->work.flags = 0;
2748
2749         if (btrfs_is_free_space_inode(inode))
2750                 workers = &root->fs_info->endio_freespace_worker;
2751         else
2752                 workers = &root->fs_info->endio_write_workers;
2753         btrfs_queue_worker(workers, &ordered_extent->work);
2754
2755         return 0;
2756 }
2757
2758 /*
2759  * when reads are done, we need to check csums to verify the data is correct
2760  * if there's a match, we allow the bio to finish.  If not, the code in
2761  * extent_io.c will try to find good copies for us.
2762  */
2763 static int btrfs_readpage_end_io_hook(struct btrfs_io_bio *io_bio,
2764                                       u64 phy_offset, struct page *page,
2765                                       u64 start, u64 end, int mirror)
2766 {
2767         size_t offset = start - page_offset(page);
2768         struct inode *inode = page->mapping->host;
2769         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2770         char *kaddr;
2771         struct btrfs_root *root = BTRFS_I(inode)->root;
2772         u32 csum_expected;
2773         u32 csum = ~(u32)0;
2774         static DEFINE_RATELIMIT_STATE(_rs, DEFAULT_RATELIMIT_INTERVAL,
2775                                       DEFAULT_RATELIMIT_BURST);
2776
2777         if (PageChecked(page)) {
2778                 ClearPageChecked(page);
2779                 goto good;
2780         }
2781
2782         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2783                 goto good;
2784
2785         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2786             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2787                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2788                                   GFP_NOFS);
2789                 return 0;
2790         }
2791
2792         phy_offset >>= inode->i_sb->s_blocksize_bits;
2793         csum_expected = *(((u32 *)io_bio->csum) + phy_offset);
2794
2795         kaddr = kmap_atomic(page);
2796         csum = btrfs_csum_data(kaddr + offset, csum,  end - start + 1);
2797         btrfs_csum_final(csum, (char *)&csum);
2798         if (csum != csum_expected)
2799                 goto zeroit;
2800
2801         kunmap_atomic(kaddr);
2802 good:
2803         return 0;
2804
2805 zeroit:
2806         if (__ratelimit(&_rs))
2807                 btrfs_info(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
2808                         btrfs_ino(page->mapping->host), start, csum, csum_expected);
2809         memset(kaddr + offset, 1, end - start + 1);
2810         flush_dcache_page(page);
2811         kunmap_atomic(kaddr);
2812         if (csum_expected == 0)
2813                 return 0;
2814         return -EIO;
2815 }
2816
2817 struct delayed_iput {
2818         struct list_head list;
2819         struct inode *inode;
2820 };
2821
2822 /* JDM: If this is fs-wide, why can't we add a pointer to
2823  * btrfs_inode instead and avoid the allocation? */
2824 void btrfs_add_delayed_iput(struct inode *inode)
2825 {
2826         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2827         struct delayed_iput *delayed;
2828
2829         if (atomic_add_unless(&inode->i_count, -1, 1))
2830                 return;
2831
2832         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2833         delayed->inode = inode;
2834
2835         spin_lock(&fs_info->delayed_iput_lock);
2836         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2837         spin_unlock(&fs_info->delayed_iput_lock);
2838 }
2839
2840 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2841 {
2842         LIST_HEAD(list);
2843         struct btrfs_fs_info *fs_info = root->fs_info;
2844         struct delayed_iput *delayed;
2845         int empty;
2846
2847         spin_lock(&fs_info->delayed_iput_lock);
2848         empty = list_empty(&fs_info->delayed_iputs);
2849         spin_unlock(&fs_info->delayed_iput_lock);
2850         if (empty)
2851                 return;
2852
2853         spin_lock(&fs_info->delayed_iput_lock);
2854         list_splice_init(&fs_info->delayed_iputs, &list);
2855         spin_unlock(&fs_info->delayed_iput_lock);
2856
2857         while (!list_empty(&list)) {
2858                 delayed = list_entry(list.next, struct delayed_iput, list);
2859                 list_del(&delayed->list);
2860                 iput(delayed->inode);
2861                 kfree(delayed);
2862         }
2863 }
2864
2865 /*
2866  * This is called in transaction commit time. If there are no orphan
2867  * files in the subvolume, it removes orphan item and frees block_rsv
2868  * structure.
2869  */
2870 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2871                               struct btrfs_root *root)
2872 {
2873         struct btrfs_block_rsv *block_rsv;
2874         int ret;
2875
2876         if (atomic_read(&root->orphan_inodes) ||
2877             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2878                 return;
2879
2880         spin_lock(&root->orphan_lock);
2881         if (atomic_read(&root->orphan_inodes)) {
2882                 spin_unlock(&root->orphan_lock);
2883                 return;
2884         }
2885
2886         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2887                 spin_unlock(&root->orphan_lock);
2888                 return;
2889         }
2890
2891         block_rsv = root->orphan_block_rsv;
2892         root->orphan_block_rsv = NULL;
2893         spin_unlock(&root->orphan_lock);
2894
2895         if (root->orphan_item_inserted &&
2896             btrfs_root_refs(&root->root_item) > 0) {
2897                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2898                                             root->root_key.objectid);
2899                 if (ret)
2900                         btrfs_abort_transaction(trans, root, ret);
2901                 else
2902                         root->orphan_item_inserted = 0;
2903         }
2904
2905         if (block_rsv) {
2906                 WARN_ON(block_rsv->size > 0);
2907                 btrfs_free_block_rsv(root, block_rsv);
2908         }
2909 }
2910
2911 /*
2912  * This creates an orphan entry for the given inode in case something goes
2913  * wrong in the middle of an unlink/truncate.
2914  *
2915  * NOTE: caller of this function should reserve 5 units of metadata for
2916  *       this function.
2917  */
2918 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2919 {
2920         struct btrfs_root *root = BTRFS_I(inode)->root;
2921         struct btrfs_block_rsv *block_rsv = NULL;
2922         int reserve = 0;
2923         int insert = 0;
2924         int ret;
2925
2926         if (!root->orphan_block_rsv) {
2927                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2928                 if (!block_rsv)
2929                         return -ENOMEM;
2930         }
2931
2932         spin_lock(&root->orphan_lock);
2933         if (!root->orphan_block_rsv) {
2934                 root->orphan_block_rsv = block_rsv;
2935         } else if (block_rsv) {
2936                 btrfs_free_block_rsv(root, block_rsv);
2937                 block_rsv = NULL;
2938         }
2939
2940         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2941                               &BTRFS_I(inode)->runtime_flags)) {
2942 #if 0
2943                 /*
2944                  * For proper ENOSPC handling, we should do orphan
2945                  * cleanup when mounting. But this introduces backward
2946                  * compatibility issue.
2947                  */
2948                 if (!xchg(&root->orphan_item_inserted, 1))
2949                         insert = 2;
2950                 else
2951                         insert = 1;
2952 #endif
2953                 insert = 1;
2954                 atomic_inc(&root->orphan_inodes);
2955         }
2956
2957         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2958                               &BTRFS_I(inode)->runtime_flags))
2959                 reserve = 1;
2960         spin_unlock(&root->orphan_lock);
2961
2962         /* grab metadata reservation from transaction handle */
2963         if (reserve) {
2964                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2965                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2966         }
2967
2968         /* insert an orphan item to track this unlinked/truncated file */
2969         if (insert >= 1) {
2970                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2971                 if (ret) {
2972                         atomic_dec(&root->orphan_inodes);
2973                         if (reserve) {
2974                                 clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2975                                           &BTRFS_I(inode)->runtime_flags);
2976                                 btrfs_orphan_release_metadata(inode);
2977                         }
2978                         if (ret != -EEXIST) {
2979                                 clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2980                                           &BTRFS_I(inode)->runtime_flags);
2981                                 btrfs_abort_transaction(trans, root, ret);
2982                                 return ret;
2983                         }
2984                 }
2985                 ret = 0;
2986         }
2987
2988         /* insert an orphan item to track subvolume contains orphan files */
2989         if (insert >= 2) {
2990                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2991                                                root->root_key.objectid);
2992                 if (ret && ret != -EEXIST) {
2993                         btrfs_abort_transaction(trans, root, ret);
2994                         return ret;
2995                 }
2996         }
2997         return 0;
2998 }
2999
3000 /*
3001  * We have done the truncate/delete so we can go ahead and remove the orphan
3002  * item for this particular inode.
3003  */
3004 static int btrfs_orphan_del(struct btrfs_trans_handle *trans,
3005                             struct inode *inode)
3006 {
3007         struct btrfs_root *root = BTRFS_I(inode)->root;
3008         int delete_item = 0;
3009         int release_rsv = 0;
3010         int ret = 0;
3011
3012         spin_lock(&root->orphan_lock);
3013         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3014                                &BTRFS_I(inode)->runtime_flags))
3015                 delete_item = 1;
3016
3017         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
3018                                &BTRFS_I(inode)->runtime_flags))
3019                 release_rsv = 1;
3020         spin_unlock(&root->orphan_lock);
3021
3022         if (delete_item) {
3023                 atomic_dec(&root->orphan_inodes);
3024                 if (trans)
3025                         ret = btrfs_del_orphan_item(trans, root,
3026                                                     btrfs_ino(inode));
3027         }
3028
3029         if (release_rsv)
3030                 btrfs_orphan_release_metadata(inode);
3031
3032         return ret;
3033 }
3034
3035 /*
3036  * this cleans up any orphans that may be left on the list from the last use
3037  * of this root.
3038  */
3039 int btrfs_orphan_cleanup(struct btrfs_root *root)
3040 {
3041         struct btrfs_path *path;
3042         struct extent_buffer *leaf;
3043         struct btrfs_key key, found_key;
3044         struct btrfs_trans_handle *trans;
3045         struct inode *inode;
3046         u64 last_objectid = 0;
3047         int ret = 0, nr_unlink = 0, nr_truncate = 0;
3048
3049         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
3050                 return 0;
3051
3052         path = btrfs_alloc_path();
3053         if (!path) {
3054                 ret = -ENOMEM;
3055                 goto out;
3056         }
3057         path->reada = -1;
3058
3059         key.objectid = BTRFS_ORPHAN_OBJECTID;
3060         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
3061         key.offset = (u64)-1;
3062
3063         while (1) {
3064                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3065                 if (ret < 0)
3066                         goto out;
3067
3068                 /*
3069                  * if ret == 0 means we found what we were searching for, which
3070                  * is weird, but possible, so only screw with path if we didn't
3071                  * find the key and see if we have stuff that matches
3072                  */
3073                 if (ret > 0) {
3074                         ret = 0;
3075                         if (path->slots[0] == 0)
3076                                 break;
3077                         path->slots[0]--;
3078                 }
3079
3080                 /* pull out the item */
3081                 leaf = path->nodes[0];
3082                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3083
3084                 /* make sure the item matches what we want */
3085                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
3086                         break;
3087                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
3088                         break;
3089
3090                 /* release the path since we're done with it */
3091                 btrfs_release_path(path);
3092
3093                 /*
3094                  * this is where we are basically btrfs_lookup, without the
3095                  * crossing root thing.  we store the inode number in the
3096                  * offset of the orphan item.
3097                  */
3098
3099                 if (found_key.offset == last_objectid) {
3100                         btrfs_err(root->fs_info,
3101                                 "Error removing orphan entry, stopping orphan cleanup");
3102                         ret = -EINVAL;
3103                         goto out;
3104                 }
3105
3106                 last_objectid = found_key.offset;
3107
3108                 found_key.objectid = found_key.offset;
3109                 found_key.type = BTRFS_INODE_ITEM_KEY;
3110                 found_key.offset = 0;
3111                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
3112                 ret = PTR_ERR_OR_ZERO(inode);
3113                 if (ret && ret != -ESTALE)
3114                         goto out;
3115
3116                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
3117                         struct btrfs_root *dead_root;
3118                         struct btrfs_fs_info *fs_info = root->fs_info;
3119                         int is_dead_root = 0;
3120
3121                         /*
3122                          * this is an orphan in the tree root. Currently these
3123                          * could come from 2 sources:
3124                          *  a) a snapshot deletion in progress
3125                          *  b) a free space cache inode
3126                          * We need to distinguish those two, as the snapshot
3127                          * orphan must not get deleted.
3128                          * find_dead_roots already ran before us, so if this
3129                          * is a snapshot deletion, we should find the root
3130                          * in the dead_roots list
3131                          */
3132                         spin_lock(&fs_info->trans_lock);
3133                         list_for_each_entry(dead_root, &fs_info->dead_roots,
3134                                             root_list) {
3135                                 if (dead_root->root_key.objectid ==
3136                                     found_key.objectid) {
3137                                         is_dead_root = 1;
3138                                         break;
3139                                 }
3140                         }
3141                         spin_unlock(&fs_info->trans_lock);
3142                         if (is_dead_root) {
3143                                 /* prevent this orphan from being found again */
3144                                 key.offset = found_key.objectid - 1;
3145                                 continue;
3146                         }
3147                 }
3148                 /*
3149                  * Inode is already gone but the orphan item is still there,
3150                  * kill the orphan item.
3151                  */
3152                 if (ret == -ESTALE) {
3153                         trans = btrfs_start_transaction(root, 1);
3154                         if (IS_ERR(trans)) {
3155                                 ret = PTR_ERR(trans);
3156                                 goto out;
3157                         }
3158                         btrfs_debug(root->fs_info, "auto deleting %Lu",
3159                                 found_key.objectid);
3160                         ret = btrfs_del_orphan_item(trans, root,
3161                                                     found_key.objectid);
3162                         btrfs_end_transaction(trans, root);
3163                         if (ret)
3164                                 goto out;
3165                         continue;
3166                 }
3167
3168                 /*
3169                  * add this inode to the orphan list so btrfs_orphan_del does
3170                  * the proper thing when we hit it
3171                  */
3172                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3173                         &BTRFS_I(inode)->runtime_flags);
3174                 atomic_inc(&root->orphan_inodes);
3175
3176                 /* if we have links, this was a truncate, lets do that */
3177                 if (inode->i_nlink) {
3178                         if (WARN_ON(!S_ISREG(inode->i_mode))) {
3179                                 iput(inode);
3180                                 continue;
3181                         }
3182                         nr_truncate++;
3183
3184                         /* 1 for the orphan item deletion. */
3185                         trans = btrfs_start_transaction(root, 1);
3186                         if (IS_ERR(trans)) {
3187                                 iput(inode);
3188                                 ret = PTR_ERR(trans);
3189                                 goto out;
3190                         }
3191                         ret = btrfs_orphan_add(trans, inode);
3192                         btrfs_end_transaction(trans, root);
3193                         if (ret) {
3194                                 iput(inode);
3195                                 goto out;
3196                         }
3197
3198                         ret = btrfs_truncate(inode);
3199                         if (ret)
3200                                 btrfs_orphan_del(NULL, inode);
3201                 } else {
3202                         nr_unlink++;
3203                 }
3204
3205                 /* this will do delete_inode and everything for us */
3206                 iput(inode);
3207                 if (ret)
3208                         goto out;
3209         }
3210         /* release the path since we're done with it */
3211         btrfs_release_path(path);
3212
3213         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
3214
3215         if (root->orphan_block_rsv)
3216                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
3217                                         (u64)-1);
3218
3219         if (root->orphan_block_rsv || root->orphan_item_inserted) {
3220                 trans = btrfs_join_transaction(root);
3221                 if (!IS_ERR(trans))
3222                         btrfs_end_transaction(trans, root);
3223         }
3224
3225         if (nr_unlink)
3226                 btrfs_debug(root->fs_info, "unlinked %d orphans", nr_unlink);
3227         if (nr_truncate)
3228                 btrfs_debug(root->fs_info, "truncated %d orphans", nr_truncate);
3229
3230 out:
3231         if (ret)
3232                 btrfs_crit(root->fs_info,
3233                         "could not do orphan cleanup %d", ret);
3234         btrfs_free_path(path);
3235         return ret;
3236 }
3237
3238 /*
3239  * very simple check to peek ahead in the leaf looking for xattrs.  If we
3240  * don't find any xattrs, we know there can't be any acls.
3241  *
3242  * slot is the slot the inode is in, objectid is the objectid of the inode
3243  */
3244 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
3245                                           int slot, u64 objectid)
3246 {
3247         u32 nritems = btrfs_header_nritems(leaf);
3248         struct btrfs_key found_key;
3249         static u64 xattr_access = 0;
3250         static u64 xattr_default = 0;
3251         int scanned = 0;
3252
3253         if (!xattr_access) {
3254                 xattr_access = btrfs_name_hash(POSIX_ACL_XATTR_ACCESS,
3255                                         strlen(POSIX_ACL_XATTR_ACCESS));
3256                 xattr_default = btrfs_name_hash(POSIX_ACL_XATTR_DEFAULT,
3257                                         strlen(POSIX_ACL_XATTR_DEFAULT));
3258         }
3259
3260         slot++;
3261         while (slot < nritems) {
3262                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3263
3264                 /* we found a different objectid, there must not be acls */
3265                 if (found_key.objectid != objectid)
3266                         return 0;
3267
3268                 /* we found an xattr, assume we've got an acl */
3269                 if (found_key.type == BTRFS_XATTR_ITEM_KEY) {
3270                         if (found_key.offset == xattr_access ||
3271                             found_key.offset == xattr_default)
3272                                 return 1;
3273                 }
3274
3275                 /*
3276                  * we found a key greater than an xattr key, there can't
3277                  * be any acls later on
3278                  */
3279                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
3280                         return 0;
3281
3282                 slot++;
3283                 scanned++;
3284
3285                 /*
3286                  * it goes inode, inode backrefs, xattrs, extents,
3287                  * so if there are a ton of hard links to an inode there can
3288                  * be a lot of backrefs.  Don't waste time searching too hard,
3289                  * this is just an optimization
3290                  */
3291                 if (scanned >= 8)
3292                         break;
3293         }
3294         /* we hit the end of the leaf before we found an xattr or
3295          * something larger than an xattr.  We have to assume the inode
3296          * has acls
3297          */
3298         return 1;
3299 }
3300
3301 /*
3302  * read an inode from the btree into the in-memory inode
3303  */
3304 static void btrfs_read_locked_inode(struct inode *inode)
3305 {
3306         struct btrfs_path *path;
3307         struct extent_buffer *leaf;
3308         struct btrfs_inode_item *inode_item;
3309         struct btrfs_timespec *tspec;
3310         struct btrfs_root *root = BTRFS_I(inode)->root;
3311         struct btrfs_key location;
3312         int maybe_acls;
3313         u32 rdev;
3314         int ret;
3315         bool filled = false;
3316
3317         ret = btrfs_fill_inode(inode, &rdev);
3318         if (!ret)
3319                 filled = true;
3320
3321         path = btrfs_alloc_path();
3322         if (!path)
3323                 goto make_bad;
3324
3325         path->leave_spinning = 1;
3326         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
3327
3328         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
3329         if (ret)
3330                 goto make_bad;
3331
3332         leaf = path->nodes[0];
3333
3334         if (filled)
3335                 goto cache_acl;
3336
3337         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3338                                     struct btrfs_inode_item);
3339         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
3340         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
3341         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
3342         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
3343         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
3344
3345         tspec = btrfs_inode_atime(inode_item);
3346         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3347         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3348
3349         tspec = btrfs_inode_mtime(inode_item);
3350         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3351         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3352
3353         tspec = btrfs_inode_ctime(inode_item);
3354         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
3355         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
3356
3357         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
3358         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
3359         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
3360
3361         /*
3362          * If we were modified in the current generation and evicted from memory
3363          * and then re-read we need to do a full sync since we don't have any
3364          * idea about which extents were modified before we were evicted from
3365          * cache.
3366          */
3367         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
3368                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3369                         &BTRFS_I(inode)->runtime_flags);
3370
3371         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
3372         inode->i_generation = BTRFS_I(inode)->generation;
3373         inode->i_rdev = 0;
3374         rdev = btrfs_inode_rdev(leaf, inode_item);
3375
3376         BTRFS_I(inode)->index_cnt = (u64)-1;
3377         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
3378 cache_acl:
3379         /*
3380          * try to precache a NULL acl entry for files that don't have
3381          * any xattrs or acls
3382          */
3383         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
3384                                            btrfs_ino(inode));
3385         if (!maybe_acls)
3386                 cache_no_acl(inode);
3387
3388         btrfs_free_path(path);
3389
3390         switch (inode->i_mode & S_IFMT) {
3391         case S_IFREG:
3392                 inode->i_mapping->a_ops = &btrfs_aops;
3393                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3394                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
3395                 inode->i_fop = &btrfs_file_operations;
3396                 inode->i_op = &btrfs_file_inode_operations;
3397                 break;
3398         case S_IFDIR:
3399                 inode->i_fop = &btrfs_dir_file_operations;
3400                 if (root == root->fs_info->tree_root)
3401                         inode->i_op = &btrfs_dir_ro_inode_operations;
3402                 else
3403                         inode->i_op = &btrfs_dir_inode_operations;
3404                 break;
3405         case S_IFLNK:
3406                 inode->i_op = &btrfs_symlink_inode_operations;
3407                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
3408                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
3409                 break;
3410         default:
3411                 inode->i_op = &btrfs_special_inode_operations;
3412                 init_special_inode(inode, inode->i_mode, rdev);
3413                 break;
3414         }
3415
3416         btrfs_update_iflags(inode);
3417         return;
3418
3419 make_bad:
3420         btrfs_free_path(path);
3421         make_bad_inode(inode);
3422 }
3423
3424 /*
3425  * given a leaf and an inode, copy the inode fields into the leaf
3426  */
3427 static void fill_inode_item(struct btrfs_trans_handle *trans,
3428                             struct extent_buffer *leaf,
3429                             struct btrfs_inode_item *item,
3430                             struct inode *inode)
3431 {
3432         struct btrfs_map_token token;
3433
3434         btrfs_init_map_token(&token);
3435
3436         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
3437         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
3438         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
3439                                    &token);
3440         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
3441         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
3442
3443         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
3444                                      inode->i_atime.tv_sec, &token);
3445         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
3446                                       inode->i_atime.tv_nsec, &token);
3447
3448         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
3449                                      inode->i_mtime.tv_sec, &token);
3450         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
3451                                       inode->i_mtime.tv_nsec, &token);
3452
3453         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
3454                                      inode->i_ctime.tv_sec, &token);
3455         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
3456                                       inode->i_ctime.tv_nsec, &token);
3457
3458         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
3459                                      &token);
3460         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
3461                                          &token);
3462         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
3463         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
3464         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
3465         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
3466         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
3467 }
3468
3469 /*
3470  * copy everything in the in-memory inode into the btree.
3471  */
3472 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
3473                                 struct btrfs_root *root, struct inode *inode)
3474 {
3475         struct btrfs_inode_item *inode_item;
3476         struct btrfs_path *path;
3477         struct extent_buffer *leaf;
3478         int ret;
3479
3480         path = btrfs_alloc_path();
3481         if (!path)
3482                 return -ENOMEM;
3483
3484         path->leave_spinning = 1;
3485         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
3486                                  1);
3487         if (ret) {
3488                 if (ret > 0)
3489                         ret = -ENOENT;
3490                 goto failed;
3491         }
3492
3493         btrfs_unlock_up_safe(path, 1);
3494         leaf = path->nodes[0];
3495         inode_item = btrfs_item_ptr(leaf, path->slots[0],
3496                                     struct btrfs_inode_item);
3497
3498         fill_inode_item(trans, leaf, inode_item, inode);
3499         btrfs_mark_buffer_dirty(leaf);
3500         btrfs_set_inode_last_trans(trans, inode);
3501         ret = 0;
3502 failed:
3503         btrfs_free_path(path);
3504         return ret;
3505 }
3506
3507 /*
3508  * copy everything in the in-memory inode into the btree.
3509  */
3510 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
3511                                 struct btrfs_root *root, struct inode *inode)
3512 {
3513         int ret;
3514
3515         /*
3516          * If the inode is a free space inode, we can deadlock during commit
3517          * if we put it into the delayed code.
3518          *
3519          * The data relocation inode should also be directly updated
3520          * without delay
3521          */
3522         if (!btrfs_is_free_space_inode(inode)
3523             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
3524                 btrfs_update_root_times(trans, root);
3525
3526                 ret = btrfs_delayed_update_inode(trans, root, inode);
3527                 if (!ret)
3528                         btrfs_set_inode_last_trans(trans, inode);
3529                 return ret;
3530         }
3531
3532         return btrfs_update_inode_item(trans, root, inode);
3533 }
3534
3535 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
3536                                          struct btrfs_root *root,
3537                                          struct inode *inode)
3538 {
3539         int ret;
3540
3541         ret = btrfs_update_inode(trans, root, inode);
3542         if (ret == -ENOSPC)
3543                 return btrfs_update_inode_item(trans, root, inode);
3544         return ret;
3545 }
3546
3547 /*
3548  * unlink helper that gets used here in inode.c and in the tree logging
3549  * recovery code.  It remove a link in a directory with a given name, and
3550  * also drops the back refs in the inode to the directory
3551  */
3552 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3553                                 struct btrfs_root *root,
3554                                 struct inode *dir, struct inode *inode,
3555                                 const char *name, int name_len)
3556 {
3557         struct btrfs_path *path;
3558         int ret = 0;
3559         struct extent_buffer *leaf;
3560         struct btrfs_dir_item *di;
3561         struct btrfs_key key;
3562         u64 index;
3563         u64 ino = btrfs_ino(inode);
3564         u64 dir_ino = btrfs_ino(dir);
3565
3566         path = btrfs_alloc_path();
3567         if (!path) {
3568                 ret = -ENOMEM;
3569                 goto out;
3570         }
3571
3572         path->leave_spinning = 1;
3573         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3574                                     name, name_len, -1);
3575         if (IS_ERR(di)) {
3576                 ret = PTR_ERR(di);
3577                 goto err;
3578         }
3579         if (!di) {
3580                 ret = -ENOENT;
3581                 goto err;
3582         }
3583         leaf = path->nodes[0];
3584         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3585         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3586         if (ret)
3587                 goto err;
3588         btrfs_release_path(path);
3589
3590         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
3591                                   dir_ino, &index);
3592         if (ret) {
3593                 btrfs_info(root->fs_info,
3594                         "failed to delete reference to %.*s, inode %llu parent %llu",
3595                         name_len, name, ino, dir_ino);
3596                 btrfs_abort_transaction(trans, root, ret);
3597                 goto err;
3598         }
3599
3600         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3601         if (ret) {
3602                 btrfs_abort_transaction(trans, root, ret);
3603                 goto err;
3604         }
3605
3606         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
3607                                          inode, dir_ino);
3608         if (ret != 0 && ret != -ENOENT) {
3609                 btrfs_abort_transaction(trans, root, ret);
3610                 goto err;
3611         }
3612
3613         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
3614                                            dir, index);
3615         if (ret == -ENOENT)
3616                 ret = 0;
3617         else if (ret)
3618                 btrfs_abort_transaction(trans, root, ret);
3619 err:
3620         btrfs_free_path(path);
3621         if (ret)
3622                 goto out;
3623
3624         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3625         inode_inc_iversion(inode);
3626         inode_inc_iversion(dir);
3627         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3628         ret = btrfs_update_inode(trans, root, dir);
3629 out:
3630         return ret;
3631 }
3632
3633 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
3634                        struct btrfs_root *root,
3635                        struct inode *dir, struct inode *inode,
3636                        const char *name, int name_len)
3637 {
3638         int ret;
3639         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
3640         if (!ret) {
3641                 drop_nlink(inode);
3642                 ret = btrfs_update_inode(trans, root, inode);
3643         }
3644         return ret;
3645 }
3646
3647 /*
3648  * helper to start transaction for unlink and rmdir.
3649  *
3650  * unlink and rmdir are special in btrfs, they do not always free space, so
3651  * if we cannot make our reservations the normal way try and see if there is
3652  * plenty of slack room in the global reserve to migrate, otherwise we cannot
3653  * allow the unlink to occur.
3654  */
3655 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir)
3656 {
3657         struct btrfs_trans_handle *trans;
3658         struct btrfs_root *root = BTRFS_I(dir)->root;
3659         int ret;
3660
3661         /*
3662          * 1 for the possible orphan item
3663          * 1 for the dir item
3664          * 1 for the dir index
3665          * 1 for the inode ref
3666          * 1 for the inode
3667          */
3668         trans = btrfs_start_transaction(root, 5);
3669         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3670                 return trans;
3671
3672         if (PTR_ERR(trans) == -ENOSPC) {
3673                 u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5);
3674
3675                 trans = btrfs_start_transaction(root, 0);
3676                 if (IS_ERR(trans))
3677                         return trans;
3678                 ret = btrfs_cond_migrate_bytes(root->fs_info,
3679                                                &root->fs_info->trans_block_rsv,
3680                                                num_bytes, 5);
3681                 if (ret) {
3682                         btrfs_end_transaction(trans, root);
3683                         return ERR_PTR(ret);
3684                 }
3685                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3686                 trans->bytes_reserved = num_bytes;
3687         }
3688         return trans;
3689 }
3690
3691 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3692 {
3693         struct btrfs_root *root = BTRFS_I(dir)->root;
3694         struct btrfs_trans_handle *trans;
3695         struct inode *inode = dentry->d_inode;
3696         int ret;
3697
3698         trans = __unlink_start_trans(dir);
3699         if (IS_ERR(trans))
3700                 return PTR_ERR(trans);
3701
3702         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3703
3704         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3705                                  dentry->d_name.name, dentry->d_name.len);
3706         if (ret)
3707                 goto out;
3708
3709         if (inode->i_nlink == 0) {
3710                 ret = btrfs_orphan_add(trans, inode);
3711                 if (ret)
3712                         goto out;
3713         }
3714
3715 out:
3716         btrfs_end_transaction(trans, root);
3717         btrfs_btree_balance_dirty(root);
3718         return ret;
3719 }
3720
3721 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3722                         struct btrfs_root *root,
3723                         struct inode *dir, u64 objectid,
3724                         const char *name, int name_len)
3725 {
3726         struct btrfs_path *path;
3727         struct extent_buffer *leaf;
3728         struct btrfs_dir_item *di;
3729         struct btrfs_key key;
3730         u64 index;
3731         int ret;
3732         u64 dir_ino = btrfs_ino(dir);
3733
3734         path = btrfs_alloc_path();
3735         if (!path)
3736                 return -ENOMEM;
3737
3738         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3739                                    name, name_len, -1);
3740         if (IS_ERR_OR_NULL(di)) {
3741                 if (!di)
3742                         ret = -ENOENT;
3743                 else
3744                         ret = PTR_ERR(di);
3745                 goto out;
3746         }
3747
3748         leaf = path->nodes[0];
3749         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3750         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3751         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3752         if (ret) {
3753                 btrfs_abort_transaction(trans, root, ret);
3754                 goto out;
3755         }
3756         btrfs_release_path(path);
3757
3758         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3759                                  objectid, root->root_key.objectid,
3760                                  dir_ino, &index, name, name_len);
3761         if (ret < 0) {
3762                 if (ret != -ENOENT) {
3763                         btrfs_abort_transaction(trans, root, ret);
3764                         goto out;
3765                 }
3766                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3767                                                  name, name_len);
3768                 if (IS_ERR_OR_NULL(di)) {
3769                         if (!di)
3770                                 ret = -ENOENT;
3771                         else
3772                                 ret = PTR_ERR(di);
3773                         btrfs_abort_transaction(trans, root, ret);
3774                         goto out;
3775                 }
3776
3777                 leaf = path->nodes[0];
3778                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3779                 btrfs_release_path(path);
3780                 index = key.offset;
3781         }
3782         btrfs_release_path(path);
3783
3784         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3785         if (ret) {
3786                 btrfs_abort_transaction(trans, root, ret);
3787                 goto out;
3788         }
3789
3790         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3791         inode_inc_iversion(dir);
3792         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3793         ret = btrfs_update_inode_fallback(trans, root, dir);
3794         if (ret)
3795                 btrfs_abort_transaction(trans, root, ret);
3796 out:
3797         btrfs_free_path(path);
3798         return ret;
3799 }
3800
3801 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3802 {
3803         struct inode *inode = dentry->d_inode;
3804         int err = 0;
3805         struct btrfs_root *root = BTRFS_I(dir)->root;
3806         struct btrfs_trans_handle *trans;
3807
3808         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3809                 return -ENOTEMPTY;
3810         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3811                 return -EPERM;
3812
3813         trans = __unlink_start_trans(dir);
3814         if (IS_ERR(trans))
3815                 return PTR_ERR(trans);
3816
3817         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3818                 err = btrfs_unlink_subvol(trans, root, dir,
3819                                           BTRFS_I(inode)->location.objectid,
3820                                           dentry->d_name.name,
3821                                           dentry->d_name.len);
3822                 goto out;
3823         }
3824
3825         err = btrfs_orphan_add(trans, inode);
3826         if (err)
3827                 goto out;
3828
3829         /* now the directory is empty */
3830         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3831                                  dentry->d_name.name, dentry->d_name.len);
3832         if (!err)
3833                 btrfs_i_size_write(inode, 0);
3834 out:
3835         btrfs_end_transaction(trans, root);
3836         btrfs_btree_balance_dirty(root);
3837
3838         return err;
3839 }
3840
3841 /*
3842  * this can truncate away extent items, csum items and directory items.
3843  * It starts at a high offset and removes keys until it can't find
3844  * any higher than new_size
3845  *
3846  * csum items that cross the new i_size are truncated to the new size
3847  * as well.
3848  *
3849  * min_type is the minimum key type to truncate down to.  If set to 0, this
3850  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3851  */
3852 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3853                                struct btrfs_root *root,
3854                                struct inode *inode,
3855                                u64 new_size, u32 min_type)
3856 {
3857         struct btrfs_path *path;
3858         struct extent_buffer *leaf;
3859         struct btrfs_file_extent_item *fi;
3860         struct btrfs_key key;
3861         struct btrfs_key found_key;
3862         u64 extent_start = 0;
3863         u64 extent_num_bytes = 0;
3864         u64 extent_offset = 0;
3865         u64 item_end = 0;
3866         u64 last_size = (u64)-1;
3867         u32 found_type = (u8)-1;
3868         int found_extent;
3869         int del_item;
3870         int pending_del_nr = 0;
3871         int pending_del_slot = 0;
3872         int extent_type = -1;
3873         int ret;
3874         int err = 0;
3875         u64 ino = btrfs_ino(inode);
3876
3877         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3878
3879         path = btrfs_alloc_path();
3880         if (!path)
3881                 return -ENOMEM;
3882         path->reada = -1;
3883
3884         /*
3885          * We want to drop from the next block forward in case this new size is
3886          * not block aligned since we will be keeping the last block of the
3887          * extent just the way it is.
3888          */
3889         if (root->ref_cows || root == root->fs_info->tree_root)
3890                 btrfs_drop_extent_cache(inode, ALIGN(new_size,
3891                                         root->sectorsize), (u64)-1, 0);
3892
3893         /*
3894          * This function is also used to drop the items in the log tree before
3895          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3896          * it is used to drop the loged items. So we shouldn't kill the delayed
3897          * items.
3898          */
3899         if (min_type == 0 && root == BTRFS_I(inode)->root)
3900                 btrfs_kill_delayed_inode_items(inode);
3901
3902         key.objectid = ino;
3903         key.offset = (u64)-1;
3904         key.type = (u8)-1;
3905
3906 search_again:
3907         path->leave_spinning = 1;
3908         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3909         if (ret < 0) {
3910                 err = ret;
3911                 goto out;
3912         }
3913
3914         if (ret > 0) {
3915                 /* there are no items in the tree for us to truncate, we're
3916                  * done
3917                  */
3918                 if (path->slots[0] == 0)
3919                         goto out;
3920                 path->slots[0]--;
3921         }
3922
3923         while (1) {
3924                 fi = NULL;
3925                 leaf = path->nodes[0];
3926                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3927                 found_type = btrfs_key_type(&found_key);
3928
3929                 if (found_key.objectid != ino)
3930                         break;
3931
3932                 if (found_type < min_type)
3933                         break;
3934
3935                 item_end = found_key.offset;
3936                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3937                         fi = btrfs_item_ptr(leaf, path->slots[0],
3938                                             struct btrfs_file_extent_item);
3939                         extent_type = btrfs_file_extent_type(leaf, fi);
3940                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3941                                 item_end +=
3942                                     btrfs_file_extent_num_bytes(leaf, fi);
3943                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3944                                 item_end += btrfs_file_extent_inline_len(leaf,
3945                                                                          fi);
3946                         }
3947                         item_end--;
3948                 }
3949                 if (found_type > min_type) {
3950                         del_item = 1;
3951                 } else {
3952                         if (item_end < new_size)
3953                                 break;
3954                         if (found_key.offset >= new_size)
3955                                 del_item = 1;
3956                         else
3957                                 del_item = 0;
3958                 }
3959                 found_extent = 0;
3960                 /* FIXME, shrink the extent if the ref count is only 1 */
3961                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3962                         goto delete;
3963
3964                 if (del_item)
3965                         last_size = found_key.offset;
3966                 else
3967                         last_size = new_size;
3968
3969                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3970                         u64 num_dec;
3971                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3972                         if (!del_item) {
3973                                 u64 orig_num_bytes =
3974                                         btrfs_file_extent_num_bytes(leaf, fi);
3975                                 extent_num_bytes = ALIGN(new_size -
3976                                                 found_key.offset,
3977                                                 root->sectorsize);
3978                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3979                                                          extent_num_bytes);
3980                                 num_dec = (orig_num_bytes -
3981                                            extent_num_bytes);
3982                                 if (root->ref_cows && extent_start != 0)
3983                                         inode_sub_bytes(inode, num_dec);
3984                                 btrfs_mark_buffer_dirty(leaf);
3985                         } else {
3986                                 extent_num_bytes =
3987                                         btrfs_file_extent_disk_num_bytes(leaf,
3988                                                                          fi);
3989                                 extent_offset = found_key.offset -
3990                                         btrfs_file_extent_offset(leaf, fi);
3991
3992                                 /* FIXME blocksize != 4096 */
3993                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3994                                 if (extent_start != 0) {
3995                                         found_extent = 1;
3996                                         if (root->ref_cows)
3997                                                 inode_sub_bytes(inode, num_dec);
3998                                 }
3999                         }
4000                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
4001                         /*
4002                          * we can't truncate inline items that have had
4003                          * special encodings
4004                          */
4005                         if (!del_item &&
4006                             btrfs_file_extent_compression(leaf, fi) == 0 &&
4007                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
4008                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
4009                                 u32 size = new_size - found_key.offset;
4010
4011                                 if (root->ref_cows) {
4012                                         inode_sub_bytes(inode, item_end + 1 -
4013                                                         new_size);
4014                                 }
4015                                 size =
4016                                     btrfs_file_extent_calc_inline_size(size);
4017                                 btrfs_truncate_item(root, path, size, 1);
4018                         } else if (root->ref_cows) {
4019                                 inode_sub_bytes(inode, item_end + 1 -
4020                                                 found_key.offset);
4021                         }
4022                 }
4023 delete:
4024                 if (del_item) {
4025                         if (!pending_del_nr) {
4026                                 /* no pending yet, add ourselves */
4027                                 pending_del_slot = path->slots[0];
4028                                 pending_del_nr = 1;
4029                         } else if (pending_del_nr &&
4030                                    path->slots[0] + 1 == pending_del_slot) {
4031                                 /* hop on the pending chunk */
4032                                 pending_del_nr++;
4033                                 pending_del_slot = path->slots[0];
4034                         } else {
4035                                 BUG();
4036                         }
4037                 } else {
4038                         break;
4039                 }
4040                 if (found_extent && (root->ref_cows ||
4041                                      root == root->fs_info->tree_root)) {
4042                         btrfs_set_path_blocking(path);
4043                         ret = btrfs_free_extent(trans, root, extent_start,
4044                                                 extent_num_bytes, 0,
4045                                                 btrfs_header_owner(leaf),
4046                                                 ino, extent_offset, 0);
4047                         BUG_ON(ret);
4048                 }
4049
4050                 if (found_type == BTRFS_INODE_ITEM_KEY)
4051                         break;
4052
4053                 if (path->slots[0] == 0 ||
4054                     path->slots[0] != pending_del_slot) {
4055                         if (pending_del_nr) {
4056                                 ret = btrfs_del_items(trans, root, path,
4057                                                 pending_del_slot,
4058                                                 pending_del_nr);
4059                                 if (ret) {
4060                                         btrfs_abort_transaction(trans,
4061                                                                 root, ret);
4062                                         goto error;
4063                                 }
4064                                 pending_del_nr = 0;
4065                         }
4066                         btrfs_release_path(path);
4067                         goto search_again;
4068                 } else {
4069                         path->slots[0]--;
4070                 }
4071         }
4072 out:
4073         if (pending_del_nr) {
4074                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
4075                                       pending_del_nr);
4076                 if (ret)
4077                         btrfs_abort_transaction(trans, root, ret);
4078         }
4079 error:
4080         if (last_size != (u64)-1)
4081                 btrfs_ordered_update_i_size(inode, last_size, NULL);
4082         btrfs_free_path(path);
4083         return err;
4084 }
4085
4086 /*
4087  * btrfs_truncate_page - read, zero a chunk and write a page
4088  * @inode - inode that we're zeroing
4089  * @from - the offset to start zeroing
4090  * @len - the length to zero, 0 to zero the entire range respective to the
4091  *      offset
4092  * @front - zero up to the offset instead of from the offset on
4093  *
4094  * This will find the page for the "from" offset and cow the page and zero the
4095  * part we want to zero.  This is used with truncate and hole punching.
4096  */
4097 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
4098                         int front)
4099 {
4100         struct address_space *mapping = inode->i_mapping;
4101         struct btrfs_root *root = BTRFS_I(inode)->root;
4102         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4103         struct btrfs_ordered_extent *ordered;
4104         struct extent_state *cached_state = NULL;
4105         char *kaddr;
4106         u32 blocksize = root->sectorsize;
4107         pgoff_t index = from >> PAGE_CACHE_SHIFT;
4108         unsigned offset = from & (PAGE_CACHE_SIZE-1);
4109         struct page *page;
4110         gfp_t mask = btrfs_alloc_write_mask(mapping);
4111         int ret = 0;
4112         u64 page_start;
4113         u64 page_end;
4114
4115         if ((offset & (blocksize - 1)) == 0 &&
4116             (!len || ((len & (blocksize - 1)) == 0)))
4117                 goto out;
4118         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
4119         if (ret)
4120                 goto out;
4121
4122 again:
4123         page = find_or_create_page(mapping, index, mask);
4124         if (!page) {
4125                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4126                 ret = -ENOMEM;
4127                 goto out;
4128         }
4129
4130         page_start = page_offset(page);
4131         page_end = page_start + PAGE_CACHE_SIZE - 1;
4132
4133         if (!PageUptodate(page)) {
4134                 ret = btrfs_readpage(NULL, page);
4135                 lock_page(page);
4136                 if (page->mapping != mapping) {
4137                         unlock_page(page);
4138                         page_cache_release(page);
4139                         goto again;
4140                 }
4141                 if (!PageUptodate(page)) {
4142                         ret = -EIO;
4143                         goto out_unlock;
4144                 }
4145         }
4146         wait_on_page_writeback(page);
4147
4148         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
4149         set_page_extent_mapped(page);
4150
4151         ordered = btrfs_lookup_ordered_extent(inode, page_start);
4152         if (ordered) {
4153                 unlock_extent_cached(io_tree, page_start, page_end,
4154                                      &cached_state, GFP_NOFS);
4155                 unlock_page(page);
4156                 page_cache_release(page);
4157                 btrfs_start_ordered_extent(inode, ordered, 1);
4158                 btrfs_put_ordered_extent(ordered);
4159                 goto again;
4160         }
4161
4162         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
4163                           EXTENT_DIRTY | EXTENT_DELALLOC |
4164                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
4165                           0, 0, &cached_state, GFP_NOFS);
4166
4167         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
4168                                         &cached_state);
4169         if (ret) {
4170                 unlock_extent_cached(io_tree, page_start, page_end,
4171                                      &cached_state, GFP_NOFS);
4172                 goto out_unlock;
4173         }
4174
4175         if (offset != PAGE_CACHE_SIZE) {
4176                 if (!len)
4177                         len = PAGE_CACHE_SIZE - offset;
4178                 kaddr = kmap(page);
4179                 if (front)
4180                         memset(kaddr, 0, offset);
4181                 else
4182                         memset(kaddr + offset, 0, len);
4183                 flush_dcache_page(page);
4184                 kunmap(page);
4185         }
4186         ClearPageChecked(page);
4187         set_page_dirty(page);
4188         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
4189                              GFP_NOFS);
4190
4191 out_unlock:
4192         if (ret)
4193                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
4194         unlock_page(page);
4195         page_cache_release(page);
4196 out:
4197         return ret;
4198 }
4199
4200 static int maybe_insert_hole(struct btrfs_root *root, struct inode *inode,
4201                              u64 offset, u64 len)
4202 {
4203         struct btrfs_trans_handle *trans;
4204         int ret;
4205
4206         /*
4207          * Still need to make sure the inode looks like it's been updated so
4208          * that any holes get logged if we fsync.
4209          */
4210         if (btrfs_fs_incompat(root->fs_info, NO_HOLES)) {
4211                 BTRFS_I(inode)->last_trans = root->fs_info->generation;
4212                 BTRFS_I(inode)->last_sub_trans = root->log_transid;
4213                 BTRFS_I(inode)->last_log_commit = root->last_log_commit;
4214                 return 0;
4215         }
4216
4217         /*
4218          * 1 - for the one we're dropping
4219          * 1 - for the one we're adding
4220          * 1 - for updating the inode.
4221          */
4222         trans = btrfs_start_transaction(root, 3);
4223         if (IS_ERR(trans))
4224                 return PTR_ERR(trans);
4225
4226         ret = btrfs_drop_extents(trans, root, inode, offset, offset + len, 1);
4227         if (ret) {
4228                 btrfs_abort_transaction(trans, root, ret);
4229                 btrfs_end_transaction(trans, root);
4230                 return ret;
4231         }
4232
4233         ret = btrfs_insert_file_extent(trans, root, btrfs_ino(inode), offset,
4234                                        0, 0, len, 0, len, 0, 0, 0);
4235         if (ret)
4236                 btrfs_abort_transaction(trans, root, ret);
4237         else
4238                 btrfs_update_inode(trans, root, inode);
4239         btrfs_end_transaction(trans, root);
4240         return ret;
4241 }
4242
4243 /*
4244  * This function puts in dummy file extents for the area we're creating a hole
4245  * for.  So if we are truncating this file to a larger size we need to insert
4246  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
4247  * the range between oldsize and size
4248  */
4249 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
4250 {
4251         struct btrfs_root *root = BTRFS_I(inode)->root;
4252         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4253         struct extent_map *em = NULL;
4254         struct extent_state *cached_state = NULL;
4255         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4256         u64 hole_start = ALIGN(oldsize, root->sectorsize);
4257         u64 block_end = ALIGN(size, root->sectorsize);
4258         u64 last_byte;
4259         u64 cur_offset;
4260         u64 hole_size;
4261         int err = 0;
4262
4263         /*
4264          * If our size started in the middle of a page we need to zero out the
4265          * rest of the page before we expand the i_size, otherwise we could
4266          * expose stale data.
4267          */
4268         err = btrfs_truncate_page(inode, oldsize, 0, 0);
4269         if (err)
4270                 return err;
4271
4272         if (size <= hole_start)
4273                 return 0;
4274
4275         while (1) {
4276                 struct btrfs_ordered_extent *ordered;
4277
4278                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
4279                                  &cached_state);
4280                 ordered = btrfs_lookup_ordered_range(inode, hole_start,
4281                                                      block_end - hole_start);
4282                 if (!ordered)
4283                         break;
4284                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
4285                                      &cached_state, GFP_NOFS);
4286                 btrfs_start_ordered_extent(inode, ordered, 1);
4287                 btrfs_put_ordered_extent(ordered);
4288         }
4289
4290         cur_offset = hole_start;
4291         while (1) {
4292                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
4293                                 block_end - cur_offset, 0);
4294                 if (IS_ERR(em)) {
4295                         err = PTR_ERR(em);
4296                         em = NULL;
4297                         break;
4298                 }
4299                 last_byte = min(extent_map_end(em), block_end);
4300                 last_byte = ALIGN(last_byte , root->sectorsize);
4301                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
4302                         struct extent_map *hole_em;
4303                         hole_size = last_byte - cur_offset;
4304
4305                         err = maybe_insert_hole(root, inode, cur_offset,
4306                                                 hole_size);
4307                         if (err)
4308                                 break;
4309                         btrfs_drop_extent_cache(inode, cur_offset,
4310                                                 cur_offset + hole_size - 1, 0);
4311                         hole_em = alloc_extent_map();
4312                         if (!hole_em) {
4313                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
4314                                         &BTRFS_I(inode)->runtime_flags);
4315                                 goto next;
4316                         }
4317                         hole_em->start = cur_offset;
4318                         hole_em->len = hole_size;
4319                         hole_em->orig_start = cur_offset;
4320
4321                         hole_em->block_start = EXTENT_MAP_HOLE;
4322                         hole_em->block_len = 0;
4323                         hole_em->orig_block_len = 0;
4324                         hole_em->ram_bytes = hole_size;
4325                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
4326                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
4327                         hole_em->generation = root->fs_info->generation;
4328
4329                         while (1) {
4330                                 write_lock(&em_tree->lock);
4331                                 err = add_extent_mapping(em_tree, hole_em, 1);
4332                                 write_unlock(&em_tree->lock);
4333                                 if (err != -EEXIST)
4334                                         break;
4335                                 btrfs_drop_extent_cache(inode, cur_offset,
4336                                                         cur_offset +
4337                                                         hole_size - 1, 0);
4338                         }
4339                         free_extent_map(hole_em);
4340                 }
4341 next:
4342                 free_extent_map(em);
4343                 em = NULL;
4344                 cur_offset = last_byte;
4345                 if (cur_offset >= block_end)
4346                         break;
4347         }
4348         free_extent_map(em);
4349         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
4350                              GFP_NOFS);
4351         return err;
4352 }
4353
4354 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
4355 {
4356         struct btrfs_root *root = BTRFS_I(inode)->root;
4357         struct btrfs_trans_handle *trans;
4358         loff_t oldsize = i_size_read(inode);
4359         loff_t newsize = attr->ia_size;
4360         int mask = attr->ia_valid;
4361         int ret;
4362
4363         /*
4364          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
4365          * special case where we need to update the times despite not having
4366          * these flags set.  For all other operations the VFS set these flags
4367          * explicitly if it wants a timestamp update.
4368          */
4369         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
4370                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
4371
4372         if (newsize > oldsize) {
4373                 truncate_pagecache(inode, newsize);
4374                 ret = btrfs_cont_expand(inode, oldsize, newsize);
4375                 if (ret)
4376                         return ret;
4377
4378                 trans = btrfs_start_transaction(root, 1);
4379                 if (IS_ERR(trans))
4380                         return PTR_ERR(trans);
4381
4382                 i_size_write(inode, newsize);
4383                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
4384                 ret = btrfs_update_inode(trans, root, inode);
4385                 btrfs_end_transaction(trans, root);
4386         } else {
4387
4388                 /*
4389                  * We're truncating a file that used to have good data down to
4390                  * zero. Make sure it gets into the ordered flush list so that
4391                  * any new writes get down to disk quickly.
4392                  */
4393                 if (newsize == 0)
4394                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
4395                                 &BTRFS_I(inode)->runtime_flags);
4396
4397                 /*
4398                  * 1 for the orphan item we're going to add
4399                  * 1 for the orphan item deletion.
4400                  */
4401                 trans = btrfs_start_transaction(root, 2);
4402                 if (IS_ERR(trans))
4403                         return PTR_ERR(trans);
4404
4405                 /*
4406                  * We need to do this in case we fail at _any_ point during the
4407                  * actual truncate.  Once we do the truncate_setsize we could
4408                  * invalidate pages which forces any outstanding ordered io to
4409                  * be instantly completed which will give us extents that need
4410                  * to be truncated.  If we fail to get an orphan inode down we
4411                  * could have left over extents that were never meant to live,
4412                  * so we need to garuntee from this point on that everything
4413                  * will be consistent.
4414                  */
4415                 ret = btrfs_orphan_add(trans, inode);
4416                 btrfs_end_transaction(trans, root);
4417                 if (ret)
4418                         return ret;
4419
4420                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
4421                 truncate_setsize(inode, newsize);
4422
4423                 /* Disable nonlocked read DIO to avoid the end less truncate */
4424                 btrfs_inode_block_unlocked_dio(inode);
4425                 inode_dio_wait(inode);
4426                 btrfs_inode_resume_unlocked_dio(inode);
4427
4428                 ret = btrfs_truncate(inode);
4429                 if (ret && inode->i_nlink) {
4430                         int err;
4431
4432                         /*
4433                          * failed to truncate, disk_i_size is only adjusted down
4434                          * as we remove extents, so it should represent the true
4435                          * size of the inode, so reset the in memory size and
4436                          * delete our orphan entry.
4437                          */
4438                         trans = btrfs_join_transaction(root);
4439                         if (IS_ERR(trans)) {
4440                                 btrfs_orphan_del(NULL, inode);
4441                                 return ret;
4442                         }
4443                         i_size_write(inode, BTRFS_I(inode)->disk_i_size);
4444                         err = btrfs_orphan_del(trans, inode);
4445                         if (err)
4446                                 btrfs_abort_transaction(trans, root, err);
4447                         btrfs_end_transaction(trans, root);
4448                 }
4449         }
4450
4451         return ret;
4452 }
4453
4454 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
4455 {
4456         struct inode *inode = dentry->d_inode;
4457         struct btrfs_root *root = BTRFS_I(inode)->root;
4458         int err;
4459
4460         if (btrfs_root_readonly(root))
4461                 return -EROFS;
4462
4463         err = inode_change_ok(inode, attr);
4464         if (err)
4465                 return err;
4466
4467         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
4468                 err = btrfs_setsize(inode, attr);
4469                 if (err)
4470                         return err;
4471         }
4472
4473         if (attr->ia_valid) {
4474                 setattr_copy(inode, attr);
4475                 inode_inc_iversion(inode);
4476                 err = btrfs_dirty_inode(inode);
4477
4478                 if (!err && attr->ia_valid & ATTR_MODE)
4479                         err = btrfs_acl_chmod(inode);
4480         }
4481
4482         return err;
4483 }
4484
4485 /*
4486  * While truncating the inode pages during eviction, we get the VFS calling
4487  * btrfs_invalidatepage() against each page of the inode. This is slow because
4488  * the calls to btrfs_invalidatepage() result in a huge amount of calls to
4489  * lock_extent_bits() and clear_extent_bit(), which keep merging and splitting
4490  * extent_state structures over and over, wasting lots of time.
4491  *
4492  * Therefore if the inode is being evicted, let btrfs_invalidatepage() skip all
4493  * those expensive operations on a per page basis and do only the ordered io
4494  * finishing, while we release here the extent_map and extent_state structures,
4495  * without the excessive merging and splitting.
4496  */
4497 static void evict_inode_truncate_pages(struct inode *inode)
4498 {
4499         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4500         struct extent_map_tree *map_tree = &BTRFS_I(inode)->extent_tree;
4501         struct rb_node *node;
4502
4503         ASSERT(inode->i_state & I_FREEING);
4504         truncate_inode_pages(&inode->i_data, 0);
4505
4506         write_lock(&map_tree->lock);
4507         while (!RB_EMPTY_ROOT(&map_tree->map)) {
4508                 struct extent_map *em;
4509
4510                 node = rb_first(&map_tree->map);
4511                 em = rb_entry(node, struct extent_map, rb_node);
4512                 remove_extent_mapping(map_tree, em);
4513                 free_extent_map(em);
4514         }
4515         write_unlock(&map_tree->lock);
4516
4517         spin_lock(&io_tree->lock);
4518         while (!RB_EMPTY_ROOT(&io_tree->state)) {
4519                 struct extent_state *state;
4520                 struct extent_state *cached_state = NULL;
4521
4522                 node = rb_first(&io_tree->state);
4523                 state = rb_entry(node, struct extent_state, rb_node);
4524                 atomic_inc(&state->refs);
4525                 spin_unlock(&io_tree->lock);
4526
4527                 lock_extent_bits(io_tree, state->start, state->end,
4528                                  0, &cached_state);
4529                 clear_extent_bit(io_tree, state->start, state->end,
4530                                  EXTENT_LOCKED | EXTENT_DIRTY |
4531                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
4532                                  EXTENT_DEFRAG, 1, 1,
4533                                  &cached_state, GFP_NOFS);
4534                 free_extent_state(state);
4535
4536                 spin_lock(&io_tree->lock);
4537         }
4538         spin_unlock(&io_tree->lock);
4539 }
4540
4541 void btrfs_evict_inode(struct inode *inode)
4542 {
4543         struct btrfs_trans_handle *trans;
4544         struct btrfs_root *root = BTRFS_I(inode)->root;
4545         struct btrfs_block_rsv *rsv, *global_rsv;
4546         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
4547         int ret;
4548
4549         trace_btrfs_inode_evict(inode);
4550
4551         evict_inode_truncate_pages(inode);
4552
4553         if (inode->i_nlink &&
4554             ((btrfs_root_refs(&root->root_item) != 0 &&
4555               root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID) ||
4556              btrfs_is_free_space_inode(inode)))
4557                 goto no_delete;
4558
4559         if (is_bad_inode(inode)) {
4560                 btrfs_orphan_del(NULL, inode);
4561                 goto no_delete;
4562         }
4563         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
4564         btrfs_wait_ordered_range(inode, 0, (u64)-1);
4565
4566         if (root->fs_info->log_root_recovering) {
4567                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
4568                                  &BTRFS_I(inode)->runtime_flags));
4569                 goto no_delete;
4570         }
4571
4572         if (inode->i_nlink > 0) {
4573                 BUG_ON(btrfs_root_refs(&root->root_item) != 0 &&
4574                        root->root_key.objectid != BTRFS_ROOT_TREE_OBJECTID);
4575                 goto no_delete;
4576         }
4577
4578         ret = btrfs_commit_inode_delayed_inode(inode);
4579         if (ret) {
4580                 btrfs_orphan_del(NULL, inode);
4581                 goto no_delete;
4582         }
4583
4584         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
4585         if (!rsv) {
4586                 btrfs_orphan_del(NULL, inode);
4587                 goto no_delete;
4588         }
4589         rsv->size = min_size;
4590         rsv->failfast = 1;
4591         global_rsv = &root->fs_info->global_block_rsv;
4592
4593         btrfs_i_size_write(inode, 0);
4594
4595         /*
4596          * This is a bit simpler than btrfs_truncate since we've already
4597          * reserved our space for our orphan item in the unlink, so we just
4598          * need to reserve some slack space in case we add bytes and update
4599          * inode item when doing the truncate.
4600          */
4601         while (1) {
4602                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
4603                                              BTRFS_RESERVE_FLUSH_LIMIT);
4604
4605                 /*
4606                  * Try and steal from the global reserve since we will
4607                  * likely not use this space anyway, we want to try as
4608                  * hard as possible to get this to work.
4609                  */
4610                 if (ret)
4611                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
4612
4613                 if (ret) {
4614                         btrfs_warn(root->fs_info,
4615                                 "Could not get space for a delete, will truncate on mount %d",
4616                                 ret);
4617                         btrfs_orphan_del(NULL, inode);
4618                         btrfs_free_block_rsv(root, rsv);
4619                         goto no_delete;
4620                 }
4621
4622                 trans = btrfs_join_transaction(root);
4623                 if (IS_ERR(trans)) {
4624                         btrfs_orphan_del(NULL, inode);
4625                         btrfs_free_block_rsv(root, rsv);
4626                         goto no_delete;
4627                 }
4628
4629                 trans->block_rsv = rsv;
4630
4631                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4632                 if (ret != -ENOSPC)
4633                         break;
4634
4635                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4636                 btrfs_end_transaction(trans, root);
4637                 trans = NULL;
4638                 btrfs_btree_balance_dirty(root);
4639         }
4640
4641         btrfs_free_block_rsv(root, rsv);
4642
4643         /*
4644          * Errors here aren't a big deal, it just means we leave orphan items
4645          * in the tree.  They will be cleaned up on the next mount.
4646          */
4647         if (ret == 0) {
4648                 trans->block_rsv = root->orphan_block_rsv;
4649                 btrfs_orphan_del(trans, inode);
4650         } else {
4651                 btrfs_orphan_del(NULL, inode);
4652         }
4653
4654         trans->block_rsv = &root->fs_info->trans_block_rsv;
4655         if (!(root == root->fs_info->tree_root ||
4656               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4657                 btrfs_return_ino(root, btrfs_ino(inode));
4658
4659         btrfs_end_transaction(trans, root);
4660         btrfs_btree_balance_dirty(root);
4661 no_delete:
4662         btrfs_remove_delayed_node(inode);
4663         clear_inode(inode);
4664         return;
4665 }
4666
4667 /*
4668  * this returns the key found in the dir entry in the location pointer.
4669  * If no dir entries were found, location->objectid is 0.
4670  */
4671 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4672                                struct btrfs_key *location)
4673 {
4674         const char *name = dentry->d_name.name;
4675         int namelen = dentry->d_name.len;
4676         struct btrfs_dir_item *di;
4677         struct btrfs_path *path;
4678         struct btrfs_root *root = BTRFS_I(dir)->root;
4679         int ret = 0;
4680
4681         path = btrfs_alloc_path();
4682         if (!path)
4683                 return -ENOMEM;
4684
4685         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4686                                     namelen, 0);
4687         if (IS_ERR(di))
4688                 ret = PTR_ERR(di);
4689
4690         if (IS_ERR_OR_NULL(di))
4691                 goto out_err;
4692
4693         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4694 out:
4695         btrfs_free_path(path);
4696         return ret;
4697 out_err:
4698         location->objectid = 0;
4699         goto out;
4700 }
4701
4702 /*
4703  * when we hit a tree root in a directory, the btrfs part of the inode
4704  * needs to be changed to reflect the root directory of the tree root.  This
4705  * is kind of like crossing a mount point.
4706  */
4707 static int fixup_tree_root_location(struct btrfs_root *root,
4708                                     struct inode *dir,
4709                                     struct dentry *dentry,
4710                                     struct btrfs_key *location,
4711                                     struct btrfs_root **sub_root)
4712 {
4713         struct btrfs_path *path;
4714         struct btrfs_root *new_root;
4715         struct btrfs_root_ref *ref;
4716         struct extent_buffer *leaf;
4717         int ret;
4718         int err = 0;
4719
4720         path = btrfs_alloc_path();
4721         if (!path) {
4722                 err = -ENOMEM;
4723                 goto out;
4724         }
4725
4726         err = -ENOENT;
4727         ret = btrfs_find_item(root->fs_info->tree_root, path,
4728                                 BTRFS_I(dir)->root->root_key.objectid,
4729                                 location->objectid, BTRFS_ROOT_REF_KEY, NULL);
4730         if (ret) {
4731                 if (ret < 0)
4732                         err = ret;
4733                 goto out;
4734         }
4735
4736         leaf = path->nodes[0];
4737         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4738         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4739             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4740                 goto out;
4741
4742         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4743                                    (unsigned long)(ref + 1),
4744                                    dentry->d_name.len);
4745         if (ret)
4746                 goto out;
4747
4748         btrfs_release_path(path);
4749
4750         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4751         if (IS_ERR(new_root)) {
4752                 err = PTR_ERR(new_root);
4753                 goto out;
4754         }
4755
4756         *sub_root = new_root;
4757         location->objectid = btrfs_root_dirid(&new_root->root_item);
4758         location->type = BTRFS_INODE_ITEM_KEY;
4759         location->offset = 0;
4760         err = 0;
4761 out:
4762         btrfs_free_path(path);
4763         return err;
4764 }
4765
4766 static void inode_tree_add(struct inode *inode)
4767 {
4768         struct btrfs_root *root = BTRFS_I(inode)->root;
4769         struct btrfs_inode *entry;
4770         struct rb_node **p;
4771         struct rb_node *parent;
4772         struct rb_node *new = &BTRFS_I(inode)->rb_node;
4773         u64 ino = btrfs_ino(inode);
4774
4775         if (inode_unhashed(inode))
4776                 return;
4777         parent = NULL;
4778         spin_lock(&root->inode_lock);
4779         p = &root->inode_tree.rb_node;
4780         while (*p) {
4781                 parent = *p;
4782                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4783
4784                 if (ino < btrfs_ino(&entry->vfs_inode))
4785                         p = &parent->rb_left;
4786                 else if (ino > btrfs_ino(&entry->vfs_inode))
4787                         p = &parent->rb_right;
4788                 else {
4789                         WARN_ON(!(entry->vfs_inode.i_state &
4790                                   (I_WILL_FREE | I_FREEING)));
4791                         rb_replace_node(parent, new, &root->inode_tree);
4792                         RB_CLEAR_NODE(parent);
4793                         spin_unlock(&root->inode_lock);
4794                         return;
4795                 }
4796         }
4797         rb_link_node(new, parent, p);
4798         rb_insert_color(new, &root->inode_tree);
4799         spin_unlock(&root->inode_lock);
4800 }
4801
4802 static void inode_tree_del(struct inode *inode)
4803 {
4804         struct btrfs_root *root = BTRFS_I(inode)->root;
4805         int empty = 0;
4806
4807         spin_lock(&root->inode_lock);
4808         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4809                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4810                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4811                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4812         }
4813         spin_unlock(&root->inode_lock);
4814
4815         if (empty && btrfs_root_refs(&root->root_item) == 0) {
4816                 synchronize_srcu(&root->fs_info->subvol_srcu);
4817                 spin_lock(&root->inode_lock);
4818                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4819                 spin_unlock(&root->inode_lock);
4820                 if (empty)
4821                         btrfs_add_dead_root(root);
4822         }
4823 }
4824
4825 void btrfs_invalidate_inodes(struct btrfs_root *root)
4826 {
4827         struct rb_node *node;
4828         struct rb_node *prev;
4829         struct btrfs_inode *entry;
4830         struct inode *inode;
4831         u64 objectid = 0;
4832
4833         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4834
4835         spin_lock(&root->inode_lock);
4836 again:
4837         node = root->inode_tree.rb_node;
4838         prev = NULL;
4839         while (node) {
4840                 prev = node;
4841                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4842
4843                 if (objectid < btrfs_ino(&entry->vfs_inode))
4844                         node = node->rb_left;
4845                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4846                         node = node->rb_right;
4847                 else
4848                         break;
4849         }
4850         if (!node) {
4851                 while (prev) {
4852                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4853                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4854                                 node = prev;
4855                                 break;
4856                         }
4857                         prev = rb_next(prev);
4858                 }
4859         }
4860         while (node) {
4861                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4862                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4863                 inode = igrab(&entry->vfs_inode);
4864                 if (inode) {
4865                         spin_unlock(&root->inode_lock);
4866                         if (atomic_read(&inode->i_count) > 1)
4867                                 d_prune_aliases(inode);
4868                         /*
4869                          * btrfs_drop_inode will have it removed from
4870                          * the inode cache when its usage count
4871                          * hits zero.
4872                          */
4873                         iput(inode);
4874                         cond_resched();
4875                         spin_lock(&root->inode_lock);
4876                         goto again;
4877                 }
4878
4879                 if (cond_resched_lock(&root->inode_lock))
4880                         goto again;
4881
4882                 node = rb_next(node);
4883         }
4884         spin_unlock(&root->inode_lock);
4885 }
4886
4887 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4888 {
4889         struct btrfs_iget_args *args = p;
4890         inode->i_ino = args->ino;
4891         BTRFS_I(inode)->root = args->root;
4892         return 0;
4893 }
4894
4895 static int btrfs_find_actor(struct inode *inode, void *opaque)
4896 {
4897         struct btrfs_iget_args *args = opaque;
4898         return args->ino == btrfs_ino(inode) &&
4899                 args->root == BTRFS_I(inode)->root;
4900 }
4901
4902 static struct inode *btrfs_iget_locked(struct super_block *s,
4903                                        u64 objectid,
4904                                        struct btrfs_root *root)
4905 {
4906         struct inode *inode;
4907         struct btrfs_iget_args args;
4908         unsigned long hashval = btrfs_inode_hash(objectid, root);
4909
4910         args.ino = objectid;
4911         args.root = root;
4912
4913         inode = iget5_locked(s, hashval, btrfs_find_actor,
4914                              btrfs_init_locked_inode,
4915                              (void *)&args);
4916         return inode;
4917 }
4918
4919 /* Get an inode object given its location and corresponding root.
4920  * Returns in *is_new if the inode was read from disk
4921  */
4922 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4923                          struct btrfs_root *root, int *new)
4924 {
4925         struct inode *inode;
4926
4927         inode = btrfs_iget_locked(s, location->objectid, root);
4928         if (!inode)
4929                 return ERR_PTR(-ENOMEM);
4930
4931         if (inode->i_state & I_NEW) {
4932                 BTRFS_I(inode)->root = root;
4933                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4934                 btrfs_read_locked_inode(inode);
4935                 if (!is_bad_inode(inode)) {
4936                         inode_tree_add(inode);
4937                         unlock_new_inode(inode);
4938                         if (new)
4939                                 *new = 1;
4940                 } else {
4941                         unlock_new_inode(inode);
4942                         iput(inode);
4943                         inode = ERR_PTR(-ESTALE);
4944                 }
4945         }
4946
4947         return inode;
4948 }
4949
4950 static struct inode *new_simple_dir(struct super_block *s,
4951                                     struct btrfs_key *key,
4952                                     struct btrfs_root *root)
4953 {
4954         struct inode *inode = new_inode(s);
4955
4956         if (!inode)
4957                 return ERR_PTR(-ENOMEM);
4958
4959         BTRFS_I(inode)->root = root;
4960         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4961         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4962
4963         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4964         inode->i_op = &btrfs_dir_ro_inode_operations;
4965         inode->i_fop = &simple_dir_operations;
4966         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4967         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4968
4969         return inode;
4970 }
4971
4972 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4973 {
4974         struct inode *inode;
4975         struct btrfs_root *root = BTRFS_I(dir)->root;
4976         struct btrfs_root *sub_root = root;
4977         struct btrfs_key location;
4978         int index;
4979         int ret = 0;
4980
4981         if (dentry->d_name.len > BTRFS_NAME_LEN)
4982                 return ERR_PTR(-ENAMETOOLONG);
4983
4984         ret = btrfs_inode_by_name(dir, dentry, &location);
4985         if (ret < 0)
4986                 return ERR_PTR(ret);
4987
4988         if (location.objectid == 0)
4989                 return ERR_PTR(-ENOENT);
4990
4991         if (location.type == BTRFS_INODE_ITEM_KEY) {
4992                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4993                 return inode;
4994         }
4995
4996         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4997
4998         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4999         ret = fixup_tree_root_location(root, dir, dentry,
5000                                        &location, &sub_root);
5001         if (ret < 0) {
5002                 if (ret != -ENOENT)
5003                         inode = ERR_PTR(ret);
5004                 else
5005                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
5006         } else {
5007                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
5008         }
5009         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
5010
5011         if (!IS_ERR(inode) && root != sub_root) {
5012                 down_read(&root->fs_info->cleanup_work_sem);
5013                 if (!(inode->i_sb->s_flags & MS_RDONLY))
5014                         ret = btrfs_orphan_cleanup(sub_root);
5015                 up_read(&root->fs_info->cleanup_work_sem);
5016                 if (ret) {
5017                         iput(inode);
5018                         inode = ERR_PTR(ret);
5019                 }
5020         }
5021
5022         return inode;
5023 }
5024
5025 static int btrfs_dentry_delete(const struct dentry *dentry)
5026 {
5027         struct btrfs_root *root;
5028         struct inode *inode = dentry->d_inode;
5029
5030         if (!inode && !IS_ROOT(dentry))
5031                 inode = dentry->d_parent->d_inode;
5032
5033         if (inode) {
5034                 root = BTRFS_I(inode)->root;
5035                 if (btrfs_root_refs(&root->root_item) == 0)
5036                         return 1;
5037
5038                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
5039                         return 1;
5040         }
5041         return 0;
5042 }
5043
5044 static void btrfs_dentry_release(struct dentry *dentry)
5045 {
5046         if (dentry->d_fsdata)
5047                 kfree(dentry->d_fsdata);
5048 }
5049
5050 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
5051                                    unsigned int flags)
5052 {
5053         struct inode *inode;
5054
5055         inode = btrfs_lookup_dentry(dir, dentry);
5056         if (IS_ERR(inode)) {
5057                 if (PTR_ERR(inode) == -ENOENT)
5058                         inode = NULL;
5059                 else
5060                         return ERR_CAST(inode);
5061         }
5062
5063         return d_splice_alias(inode, dentry);
5064 }
5065
5066 unsigned char btrfs_filetype_table[] = {
5067         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
5068 };
5069
5070 static int btrfs_real_readdir(struct file *file, struct dir_context *ctx)
5071 {
5072         struct inode *inode = file_inode(file);
5073         struct btrfs_root *root = BTRFS_I(inode)->root;
5074         struct btrfs_item *item;
5075         struct btrfs_dir_item *di;
5076         struct btrfs_key key;
5077         struct btrfs_key found_key;
5078         struct btrfs_path *path;
5079         struct list_head ins_list;
5080         struct list_head del_list;
5081         int ret;
5082         struct extent_buffer *leaf;
5083         int slot;
5084         unsigned char d_type;
5085         int over = 0;
5086         u32 di_cur;
5087         u32 di_total;
5088         u32 di_len;
5089         int key_type = BTRFS_DIR_INDEX_KEY;
5090         char tmp_name[32];
5091         char *name_ptr;
5092         int name_len;
5093         int is_curr = 0;        /* ctx->pos points to the current index? */
5094
5095         /* FIXME, use a real flag for deciding about the key type */
5096         if (root->fs_info->tree_root == root)
5097                 key_type = BTRFS_DIR_ITEM_KEY;
5098
5099         if (!dir_emit_dots(file, ctx))
5100                 return 0;
5101
5102         path = btrfs_alloc_path();
5103         if (!path)
5104                 return -ENOMEM;
5105
5106         path->reada = 1;
5107
5108         if (key_type == BTRFS_DIR_INDEX_KEY) {
5109                 INIT_LIST_HEAD(&ins_list);
5110                 INIT_LIST_HEAD(&del_list);
5111                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
5112         }
5113
5114         btrfs_set_key_type(&key, key_type);
5115         key.offset = ctx->pos;
5116         key.objectid = btrfs_ino(inode);
5117
5118         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5119         if (ret < 0)
5120                 goto err;
5121
5122         while (1) {
5123                 leaf = path->nodes[0];
5124                 slot = path->slots[0];
5125                 if (slot >= btrfs_header_nritems(leaf)) {
5126                         ret = btrfs_next_leaf(root, path);
5127                         if (ret < 0)
5128                                 goto err;
5129                         else if (ret > 0)
5130                                 break;
5131                         continue;
5132                 }
5133
5134                 item = btrfs_item_nr(slot);
5135                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5136
5137                 if (found_key.objectid != key.objectid)
5138                         break;
5139                 if (btrfs_key_type(&found_key) != key_type)
5140                         break;
5141                 if (found_key.offset < ctx->pos)
5142                         goto next;
5143                 if (key_type == BTRFS_DIR_INDEX_KEY &&
5144                     btrfs_should_delete_dir_index(&del_list,
5145                                                   found_key.offset))
5146                         goto next;
5147
5148                 ctx->pos = found_key.offset;
5149                 is_curr = 1;
5150
5151                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
5152                 di_cur = 0;
5153                 di_total = btrfs_item_size(leaf, item);
5154
5155                 while (di_cur < di_total) {
5156                         struct btrfs_key location;
5157
5158                         if (verify_dir_item(root, leaf, di))
5159                                 break;
5160
5161                         name_len = btrfs_dir_name_len(leaf, di);
5162                         if (name_len <= sizeof(tmp_name)) {
5163                                 name_ptr = tmp_name;
5164                         } else {
5165                                 name_ptr = kmalloc(name_len, GFP_NOFS);
5166                                 if (!name_ptr) {
5167                                         ret = -ENOMEM;
5168                                         goto err;
5169                                 }
5170                         }
5171                         read_extent_buffer(leaf, name_ptr,
5172                                            (unsigned long)(di + 1), name_len);
5173
5174                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
5175                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
5176
5177
5178                         /* is this a reference to our own snapshot? If so
5179                          * skip it.
5180                          *
5181                          * In contrast to old kernels, we insert the snapshot's
5182                          * dir item and dir index after it has been created, so
5183                          * we won't find a reference to our own snapshot. We
5184                          * still keep the following code for backward
5185                          * compatibility.
5186                          */
5187                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
5188                             location.objectid == root->root_key.objectid) {
5189                                 over = 0;
5190                                 goto skip;
5191                         }
5192                         over = !dir_emit(ctx, name_ptr, name_len,
5193                                        location.objectid, d_type);
5194
5195 skip:
5196                         if (name_ptr != tmp_name)
5197                                 kfree(name_ptr);
5198
5199                         if (over)
5200                                 goto nopos;
5201                         di_len = btrfs_dir_name_len(leaf, di) +
5202                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
5203                         di_cur += di_len;
5204                         di = (struct btrfs_dir_item *)((char *)di + di_len);
5205                 }
5206 next:
5207                 path->slots[0]++;
5208         }
5209
5210         if (key_type == BTRFS_DIR_INDEX_KEY) {
5211                 if (is_curr)
5212                         ctx->pos++;
5213                 ret = btrfs_readdir_delayed_dir_index(ctx, &ins_list);
5214                 if (ret)
5215                         goto nopos;
5216         }
5217
5218         /* Reached end of directory/root. Bump pos past the last item. */
5219         ctx->pos++;
5220
5221         /*
5222          * Stop new entries from being returned after we return the last
5223          * entry.
5224          *
5225          * New directory entries are assigned a strictly increasing
5226          * offset.  This means that new entries created during readdir
5227          * are *guaranteed* to be seen in the future by that readdir.
5228          * This has broken buggy programs which operate on names as
5229          * they're returned by readdir.  Until we re-use freed offsets
5230          * we have this hack to stop new entries from being returned
5231          * under the assumption that they'll never reach this huge
5232          * offset.
5233          *
5234          * This is being careful not to overflow 32bit loff_t unless the
5235          * last entry requires it because doing so has broken 32bit apps
5236          * in the past.
5237          */
5238         if (key_type == BTRFS_DIR_INDEX_KEY) {
5239                 if (ctx->pos >= INT_MAX)
5240                         ctx->pos = LLONG_MAX;
5241                 else
5242                         ctx->pos = INT_MAX;
5243         }
5244 nopos:
5245         ret = 0;
5246 err:
5247         if (key_type == BTRFS_DIR_INDEX_KEY)
5248                 btrfs_put_delayed_items(&ins_list, &del_list);
5249         btrfs_free_path(path);
5250         return ret;
5251 }
5252
5253 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
5254 {
5255         struct btrfs_root *root = BTRFS_I(inode)->root;
5256         struct btrfs_trans_handle *trans;
5257         int ret = 0;
5258         bool nolock = false;
5259
5260         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5261                 return 0;
5262
5263         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
5264                 nolock = true;
5265
5266         if (wbc->sync_mode == WB_SYNC_ALL) {
5267                 if (nolock)
5268                         trans = btrfs_join_transaction_nolock(root);
5269                 else
5270                         trans = btrfs_join_transaction(root);
5271                 if (IS_ERR(trans))
5272                         return PTR_ERR(trans);
5273                 ret = btrfs_commit_transaction(trans, root);
5274         }
5275         return ret;
5276 }
5277
5278 /*
5279  * This is somewhat expensive, updating the tree every time the
5280  * inode changes.  But, it is most likely to find the inode in cache.
5281  * FIXME, needs more benchmarking...there are no reasons other than performance
5282  * to keep or drop this code.
5283  */
5284 static int btrfs_dirty_inode(struct inode *inode)
5285 {
5286         struct btrfs_root *root = BTRFS_I(inode)->root;
5287         struct btrfs_trans_handle *trans;
5288         int ret;
5289
5290         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
5291                 return 0;
5292
5293         trans = btrfs_join_transaction(root);
5294         if (IS_ERR(trans))
5295                 return PTR_ERR(trans);
5296
5297         ret = btrfs_update_inode(trans, root, inode);
5298         if (ret && ret == -ENOSPC) {
5299                 /* whoops, lets try again with the full transaction */
5300                 btrfs_end_transaction(trans, root);
5301                 trans = btrfs_start_transaction(root, 1);
5302                 if (IS_ERR(trans))
5303                         return PTR_ERR(trans);
5304
5305                 ret = btrfs_update_inode(trans, root, inode);
5306         }
5307         btrfs_end_transaction(trans, root);
5308         if (BTRFS_I(inode)->delayed_node)
5309                 btrfs_balance_delayed_items(root);
5310
5311         return ret;
5312 }
5313
5314 /*
5315  * This is a copy of file_update_time.  We need this so we can return error on
5316  * ENOSPC for updating the inode in the case of file write and mmap writes.
5317  */
5318 static int btrfs_update_time(struct inode *inode, struct timespec *now,
5319                              int flags)
5320 {
5321         struct btrfs_root *root = BTRFS_I(inode)->root;
5322
5323         if (btrfs_root_readonly(root))
5324                 return -EROFS;
5325
5326         if (flags & S_VERSION)
5327                 inode_inc_iversion(inode);
5328         if (flags & S_CTIME)
5329                 inode->i_ctime = *now;
5330         if (flags & S_MTIME)
5331                 inode->i_mtime = *now;
5332         if (flags & S_ATIME)
5333                 inode->i_atime = *now;
5334         return btrfs_dirty_inode(inode);
5335 }
5336
5337 /*
5338  * find the highest existing sequence number in a directory
5339  * and then set the in-memory index_cnt variable to reflect
5340  * free sequence numbers
5341  */
5342 static int btrfs_set_inode_index_count(struct inode *inode)
5343 {
5344         struct btrfs_root *root = BTRFS_I(inode)->root;
5345         struct btrfs_key key, found_key;
5346         struct btrfs_path *path;
5347         struct extent_buffer *leaf;
5348         int ret;
5349
5350         key.objectid = btrfs_ino(inode);
5351         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
5352         key.offset = (u64)-1;
5353
5354         path = btrfs_alloc_path();
5355         if (!path)
5356                 return -ENOMEM;
5357
5358         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5359         if (ret < 0)
5360                 goto out;
5361         /* FIXME: we should be able to handle this */
5362         if (ret == 0)
5363                 goto out;
5364         ret = 0;
5365
5366         /*
5367          * MAGIC NUMBER EXPLANATION:
5368          * since we search a directory based on f_pos we have to start at 2
5369          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
5370          * else has to start at 2
5371          */
5372         if (path->slots[0] == 0) {
5373                 BTRFS_I(inode)->index_cnt = 2;
5374                 goto out;
5375         }
5376
5377         path->slots[0]--;
5378
5379         leaf = path->nodes[0];
5380         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5381
5382         if (found_key.objectid != btrfs_ino(inode) ||
5383             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
5384                 BTRFS_I(inode)->index_cnt = 2;
5385                 goto out;
5386         }
5387
5388         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
5389 out:
5390         btrfs_free_path(path);
5391         return ret;
5392 }
5393
5394 /*
5395  * helper to find a free sequence number in a given directory.  This current
5396  * code is very simple, later versions will do smarter things in the btree
5397  */
5398 int btrfs_set_inode_index(struct inode *dir, u64 *index)
5399 {
5400         int ret = 0;
5401
5402         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
5403                 ret = btrfs_inode_delayed_dir_index_count(dir);
5404                 if (ret) {
5405                         ret = btrfs_set_inode_index_count(dir);
5406                         if (ret)
5407                                 return ret;
5408                 }
5409         }
5410
5411         *index = BTRFS_I(dir)->index_cnt;
5412         BTRFS_I(dir)->index_cnt++;
5413
5414         return ret;
5415 }
5416
5417 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
5418                                      struct btrfs_root *root,
5419                                      struct inode *dir,
5420                                      const char *name, int name_len,
5421                                      u64 ref_objectid, u64 objectid,
5422                                      umode_t mode, u64 *index)
5423 {
5424         struct inode *inode;
5425         struct btrfs_inode_item *inode_item;
5426         struct btrfs_key *location;
5427         struct btrfs_path *path;
5428         struct btrfs_inode_ref *ref;
5429         struct btrfs_key key[2];
5430         u32 sizes[2];
5431         unsigned long ptr;
5432         int ret;
5433
5434         path = btrfs_alloc_path();
5435         if (!path)
5436                 return ERR_PTR(-ENOMEM);
5437
5438         inode = new_inode(root->fs_info->sb);
5439         if (!inode) {
5440                 btrfs_free_path(path);
5441                 return ERR_PTR(-ENOMEM);
5442         }
5443
5444         /*
5445          * we have to initialize this early, so we can reclaim the inode
5446          * number if we fail afterwards in this function.
5447          */
5448         inode->i_ino = objectid;
5449
5450         if (dir) {
5451                 trace_btrfs_inode_request(dir);
5452
5453                 ret = btrfs_set_inode_index(dir, index);
5454                 if (ret) {
5455                         btrfs_free_path(path);
5456                         iput(inode);
5457                         return ERR_PTR(ret);
5458                 }
5459         }
5460         /*
5461          * index_cnt is ignored for everything but a dir,
5462          * btrfs_get_inode_index_count has an explanation for the magic
5463          * number
5464          */
5465         BTRFS_I(inode)->index_cnt = 2;
5466         BTRFS_I(inode)->root = root;
5467         BTRFS_I(inode)->generation = trans->transid;
5468         inode->i_generation = BTRFS_I(inode)->generation;
5469
5470         /*
5471          * We could have gotten an inode number from somebody who was fsynced
5472          * and then removed in this same transaction, so let's just set full
5473          * sync since it will be a full sync anyway and this will blow away the
5474          * old info in the log.
5475          */
5476         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
5477
5478         key[0].objectid = objectid;
5479         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
5480         key[0].offset = 0;
5481
5482         /*
5483          * Start new inodes with an inode_ref. This is slightly more
5484          * efficient for small numbers of hard links since they will
5485          * be packed into one item. Extended refs will kick in if we
5486          * add more hard links than can fit in the ref item.
5487          */
5488         key[1].objectid = objectid;
5489         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
5490         key[1].offset = ref_objectid;
5491
5492         sizes[0] = sizeof(struct btrfs_inode_item);
5493         sizes[1] = name_len + sizeof(*ref);
5494
5495         path->leave_spinning = 1;
5496         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
5497         if (ret != 0)
5498                 goto fail;
5499
5500         inode_init_owner(inode, dir, mode);
5501         inode_set_bytes(inode, 0);
5502         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
5503         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
5504                                   struct btrfs_inode_item);
5505         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
5506                              sizeof(*inode_item));
5507         fill_inode_item(trans, path->nodes[0], inode_item, inode);
5508
5509         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
5510                              struct btrfs_inode_ref);
5511         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
5512         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
5513         ptr = (unsigned long)(ref + 1);
5514         write_extent_buffer(path->nodes[0], name, ptr, name_len);
5515
5516         btrfs_mark_buffer_dirty(path->nodes[0]);
5517         btrfs_free_path(path);
5518
5519         location = &BTRFS_I(inode)->location;
5520         location->objectid = objectid;
5521         location->offset = 0;
5522         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
5523
5524         btrfs_inherit_iflags(inode, dir);
5525
5526         if (S_ISREG(mode)) {
5527                 if (btrfs_test_opt(root, NODATASUM))
5528                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
5529                 if (btrfs_test_opt(root, NODATACOW))
5530                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW |
5531                                 BTRFS_INODE_NODATASUM;
5532         }
5533
5534         btrfs_insert_inode_hash(inode);
5535         inode_tree_add(inode);
5536
5537         trace_btrfs_inode_new(inode);
5538         btrfs_set_inode_last_trans(trans, inode);
5539
5540         btrfs_update_root_times(trans, root);
5541
5542         return inode;
5543 fail:
5544         if (dir)
5545                 BTRFS_I(dir)->index_cnt--;
5546         btrfs_free_path(path);
5547         iput(inode);
5548         return ERR_PTR(ret);
5549 }
5550
5551 static inline u8 btrfs_inode_type(struct inode *inode)
5552 {
5553         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
5554 }
5555
5556 /*
5557  * utility function to add 'inode' into 'parent_inode' with
5558  * a give name and a given sequence number.
5559  * if 'add_backref' is true, also insert a backref from the
5560  * inode to the parent directory.
5561  */
5562 int btrfs_add_link(struct btrfs_trans_handle *trans,
5563                    struct inode *parent_inode, struct inode *inode,
5564                    const char *name, int name_len, int add_backref, u64 index)
5565 {
5566         int ret = 0;
5567         struct btrfs_key key;
5568         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
5569         u64 ino = btrfs_ino(inode);
5570         u64 parent_ino = btrfs_ino(parent_inode);
5571
5572         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5573                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
5574         } else {
5575                 key.objectid = ino;
5576                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
5577                 key.offset = 0;
5578         }
5579
5580         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5581                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
5582                                          key.objectid, root->root_key.objectid,
5583                                          parent_ino, index, name, name_len);
5584         } else if (add_backref) {
5585                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
5586                                              parent_ino, index);
5587         }
5588
5589         /* Nothing to clean up yet */
5590         if (ret)
5591                 return ret;
5592
5593         ret = btrfs_insert_dir_item(trans, root, name, name_len,
5594                                     parent_inode, &key,
5595                                     btrfs_inode_type(inode), index);
5596         if (ret == -EEXIST || ret == -EOVERFLOW)
5597                 goto fail_dir_item;
5598         else if (ret) {
5599                 btrfs_abort_transaction(trans, root, ret);
5600                 return ret;
5601         }
5602
5603         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5604                            name_len * 2);
5605         inode_inc_iversion(parent_inode);
5606         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5607         ret = btrfs_update_inode(trans, root, parent_inode);
5608         if (ret)
5609                 btrfs_abort_transaction(trans, root, ret);
5610         return ret;
5611
5612 fail_dir_item:
5613         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5614                 u64 local_index;
5615                 int err;
5616                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5617                                  key.objectid, root->root_key.objectid,
5618                                  parent_ino, &local_index, name, name_len);
5619
5620         } else if (add_backref) {
5621                 u64 local_index;
5622                 int err;
5623
5624                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5625                                           ino, parent_ino, &local_index);
5626         }
5627         return ret;
5628 }
5629
5630 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5631                             struct inode *dir, struct dentry *dentry,
5632                             struct inode *inode, int backref, u64 index)
5633 {
5634         int err = btrfs_add_link(trans, dir, inode,
5635                                  dentry->d_name.name, dentry->d_name.len,
5636                                  backref, index);
5637         if (err > 0)
5638                 err = -EEXIST;
5639         return err;
5640 }
5641
5642 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5643                         umode_t mode, dev_t rdev)
5644 {
5645         struct btrfs_trans_handle *trans;
5646         struct btrfs_root *root = BTRFS_I(dir)->root;
5647         struct inode *inode = NULL;
5648         int err;
5649         int drop_inode = 0;
5650         u64 objectid;
5651         u64 index = 0;
5652
5653         if (!new_valid_dev(rdev))
5654                 return -EINVAL;
5655
5656         /*
5657          * 2 for inode item and ref
5658          * 2 for dir items
5659          * 1 for xattr if selinux is on
5660          */
5661         trans = btrfs_start_transaction(root, 5);
5662         if (IS_ERR(trans))
5663                 return PTR_ERR(trans);
5664
5665         err = btrfs_find_free_ino(root, &objectid);
5666         if (err)
5667                 goto out_unlock;
5668
5669         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5670                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5671                                 mode, &index);
5672         if (IS_ERR(inode)) {
5673                 err = PTR_ERR(inode);
5674                 goto out_unlock;
5675         }
5676
5677         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5678         if (err) {
5679                 drop_inode = 1;
5680                 goto out_unlock;
5681         }
5682
5683         /*
5684         * If the active LSM wants to access the inode during
5685         * d_instantiate it needs these. Smack checks to see
5686         * if the filesystem supports xattrs by looking at the
5687         * ops vector.
5688         */
5689
5690         inode->i_op = &btrfs_special_inode_operations;
5691         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5692         if (err)
5693                 drop_inode = 1;
5694         else {
5695                 init_special_inode(inode, inode->i_mode, rdev);
5696                 btrfs_update_inode(trans, root, inode);
5697                 d_instantiate(dentry, inode);
5698         }
5699 out_unlock:
5700         btrfs_end_transaction(trans, root);
5701         btrfs_btree_balance_dirty(root);
5702         if (drop_inode) {
5703                 inode_dec_link_count(inode);
5704                 iput(inode);
5705         }
5706         return err;
5707 }
5708
5709 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5710                         umode_t mode, bool excl)
5711 {
5712         struct btrfs_trans_handle *trans;
5713         struct btrfs_root *root = BTRFS_I(dir)->root;
5714         struct inode *inode = NULL;
5715         int drop_inode_on_err = 0;
5716         int err;
5717         u64 objectid;
5718         u64 index = 0;
5719
5720         /*
5721          * 2 for inode item and ref
5722          * 2 for dir items
5723          * 1 for xattr if selinux is on
5724          */
5725         trans = btrfs_start_transaction(root, 5);
5726         if (IS_ERR(trans))
5727                 return PTR_ERR(trans);
5728
5729         err = btrfs_find_free_ino(root, &objectid);
5730         if (err)
5731                 goto out_unlock;
5732
5733         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5734                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5735                                 mode, &index);
5736         if (IS_ERR(inode)) {
5737                 err = PTR_ERR(inode);
5738                 goto out_unlock;
5739         }
5740         drop_inode_on_err = 1;
5741
5742         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5743         if (err)
5744                 goto out_unlock;
5745
5746         err = btrfs_update_inode(trans, root, inode);
5747         if (err)
5748                 goto out_unlock;
5749
5750         /*
5751         * If the active LSM wants to access the inode during
5752         * d_instantiate it needs these. Smack checks to see
5753         * if the filesystem supports xattrs by looking at the
5754         * ops vector.
5755         */
5756         inode->i_fop = &btrfs_file_operations;
5757         inode->i_op = &btrfs_file_inode_operations;
5758
5759         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5760         if (err)
5761                 goto out_unlock;
5762
5763         inode->i_mapping->a_ops = &btrfs_aops;
5764         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5765         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5766         d_instantiate(dentry, inode);
5767
5768 out_unlock:
5769         btrfs_end_transaction(trans, root);
5770         if (err && drop_inode_on_err) {
5771                 inode_dec_link_count(inode);
5772                 iput(inode);
5773         }
5774         btrfs_btree_balance_dirty(root);
5775         return err;
5776 }
5777
5778 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5779                       struct dentry *dentry)
5780 {
5781         struct btrfs_trans_handle *trans;
5782         struct btrfs_root *root = BTRFS_I(dir)->root;
5783         struct inode *inode = old_dentry->d_inode;
5784         u64 index;
5785         int err;
5786         int drop_inode = 0;
5787
5788         /* do not allow sys_link's with other subvols of the same device */
5789         if (root->objectid != BTRFS_I(inode)->root->objectid)
5790                 return -EXDEV;
5791
5792         if (inode->i_nlink >= BTRFS_LINK_MAX)
5793                 return -EMLINK;
5794
5795         err = btrfs_set_inode_index(dir, &index);
5796         if (err)
5797                 goto fail;
5798
5799         /*
5800          * 2 items for inode and inode ref
5801          * 2 items for dir items
5802          * 1 item for parent inode
5803          */
5804         trans = btrfs_start_transaction(root, 5);
5805         if (IS_ERR(trans)) {
5806                 err = PTR_ERR(trans);
5807                 goto fail;
5808         }
5809
5810         inc_nlink(inode);
5811         inode_inc_iversion(inode);
5812         inode->i_ctime = CURRENT_TIME;
5813         ihold(inode);
5814         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5815
5816         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5817
5818         if (err) {
5819                 drop_inode = 1;
5820         } else {
5821                 struct dentry *parent = dentry->d_parent;
5822                 err = btrfs_update_inode(trans, root, inode);
5823                 if (err)
5824                         goto fail;
5825                 d_instantiate(dentry, inode);
5826                 btrfs_log_new_name(trans, inode, NULL, parent);
5827         }
5828
5829         btrfs_end_transaction(trans, root);
5830 fail:
5831         if (drop_inode) {
5832                 inode_dec_link_count(inode);
5833                 iput(inode);
5834         }
5835         btrfs_btree_balance_dirty(root);
5836         return err;
5837 }
5838
5839 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5840 {
5841         struct inode *inode = NULL;
5842         struct btrfs_trans_handle *trans;
5843         struct btrfs_root *root = BTRFS_I(dir)->root;
5844         int err = 0;
5845         int drop_on_err = 0;
5846         u64 objectid = 0;
5847         u64 index = 0;
5848
5849         /*
5850          * 2 items for inode and ref
5851          * 2 items for dir items
5852          * 1 for xattr if selinux is on
5853          */
5854         trans = btrfs_start_transaction(root, 5);
5855         if (IS_ERR(trans))
5856                 return PTR_ERR(trans);
5857
5858         err = btrfs_find_free_ino(root, &objectid);
5859         if (err)
5860                 goto out_fail;
5861
5862         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5863                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5864                                 S_IFDIR | mode, &index);
5865         if (IS_ERR(inode)) {
5866                 err = PTR_ERR(inode);
5867                 goto out_fail;
5868         }
5869
5870         drop_on_err = 1;
5871
5872         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5873         if (err)
5874                 goto out_fail;
5875
5876         inode->i_op = &btrfs_dir_inode_operations;
5877         inode->i_fop = &btrfs_dir_file_operations;
5878
5879         btrfs_i_size_write(inode, 0);
5880         err = btrfs_update_inode(trans, root, inode);
5881         if (err)
5882                 goto out_fail;
5883
5884         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5885                              dentry->d_name.len, 0, index);
5886         if (err)
5887                 goto out_fail;
5888
5889         d_instantiate(dentry, inode);
5890         drop_on_err = 0;
5891
5892 out_fail:
5893         btrfs_end_transaction(trans, root);
5894         if (drop_on_err)
5895                 iput(inode);
5896         btrfs_btree_balance_dirty(root);
5897         return err;
5898 }
5899
5900 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5901  * and an extent that you want to insert, deal with overlap and insert
5902  * the new extent into the tree.
5903  */
5904 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5905                                 struct extent_map *existing,
5906                                 struct extent_map *em,
5907                                 u64 map_start, u64 map_len)
5908 {
5909         u64 start_diff;
5910
5911         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5912         start_diff = map_start - em->start;
5913         em->start = map_start;
5914         em->len = map_len;
5915         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5916             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5917                 em->block_start += start_diff;
5918                 em->block_len -= start_diff;
5919         }
5920         return add_extent_mapping(em_tree, em, 0);
5921 }
5922
5923 static noinline int uncompress_inline(struct btrfs_path *path,
5924                                       struct inode *inode, struct page *page,
5925                                       size_t pg_offset, u64 extent_offset,
5926                                       struct btrfs_file_extent_item *item)
5927 {
5928         int ret;
5929         struct extent_buffer *leaf = path->nodes[0];
5930         char *tmp;
5931         size_t max_size;
5932         unsigned long inline_size;
5933         unsigned long ptr;
5934         int compress_type;
5935
5936         WARN_ON(pg_offset != 0);
5937         compress_type = btrfs_file_extent_compression(leaf, item);
5938         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5939         inline_size = btrfs_file_extent_inline_item_len(leaf,
5940                                         btrfs_item_nr(path->slots[0]));
5941         tmp = kmalloc(inline_size, GFP_NOFS);
5942         if (!tmp)
5943                 return -ENOMEM;
5944         ptr = btrfs_file_extent_inline_start(item);
5945
5946         read_extent_buffer(leaf, tmp, ptr, inline_size);
5947
5948         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5949         ret = btrfs_decompress(compress_type, tmp, page,
5950                                extent_offset, inline_size, max_size);
5951         if (ret) {
5952                 char *kaddr = kmap_atomic(page);
5953                 unsigned long copy_size = min_t(u64,
5954                                   PAGE_CACHE_SIZE - pg_offset,
5955                                   max_size - extent_offset);
5956                 memset(kaddr + pg_offset, 0, copy_size);
5957                 kunmap_atomic(kaddr);
5958         }
5959         kfree(tmp);
5960         return 0;
5961 }
5962
5963 /*
5964  * a bit scary, this does extent mapping from logical file offset to the disk.
5965  * the ugly parts come from merging extents from the disk with the in-ram
5966  * representation.  This gets more complex because of the data=ordered code,
5967  * where the in-ram extents might be locked pending data=ordered completion.
5968  *
5969  * This also copies inline extents directly into the page.
5970  */
5971
5972 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5973                                     size_t pg_offset, u64 start, u64 len,
5974                                     int create)
5975 {
5976         int ret;
5977         int err = 0;
5978         u64 bytenr;
5979         u64 extent_start = 0;
5980         u64 extent_end = 0;
5981         u64 objectid = btrfs_ino(inode);
5982         u32 found_type;
5983         struct btrfs_path *path = NULL;
5984         struct btrfs_root *root = BTRFS_I(inode)->root;
5985         struct btrfs_file_extent_item *item;
5986         struct extent_buffer *leaf;
5987         struct btrfs_key found_key;
5988         struct extent_map *em = NULL;
5989         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5990         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5991         struct btrfs_trans_handle *trans = NULL;
5992         int compress_type;
5993
5994 again:
5995         read_lock(&em_tree->lock);
5996         em = lookup_extent_mapping(em_tree, start, len);
5997         if (em)
5998                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5999         read_unlock(&em_tree->lock);
6000
6001         if (em) {
6002                 if (em->start > start || em->start + em->len <= start)
6003                         free_extent_map(em);
6004                 else if (em->block_start == EXTENT_MAP_INLINE && page)
6005                         free_extent_map(em);
6006                 else
6007                         goto out;
6008         }
6009         em = alloc_extent_map();
6010         if (!em) {
6011                 err = -ENOMEM;
6012                 goto out;
6013         }
6014         em->bdev = root->fs_info->fs_devices->latest_bdev;
6015         em->start = EXTENT_MAP_HOLE;
6016         em->orig_start = EXTENT_MAP_HOLE;
6017         em->len = (u64)-1;
6018         em->block_len = (u64)-1;
6019
6020         if (!path) {
6021                 path = btrfs_alloc_path();
6022                 if (!path) {
6023                         err = -ENOMEM;
6024                         goto out;
6025                 }
6026                 /*
6027                  * Chances are we'll be called again, so go ahead and do
6028                  * readahead
6029                  */
6030                 path->reada = 1;
6031         }
6032
6033         ret = btrfs_lookup_file_extent(trans, root, path,
6034                                        objectid, start, trans != NULL);
6035         if (ret < 0) {
6036                 err = ret;
6037                 goto out;
6038         }
6039
6040         if (ret != 0) {
6041                 if (path->slots[0] == 0)
6042                         goto not_found;
6043                 path->slots[0]--;
6044         }
6045
6046         leaf = path->nodes[0];
6047         item = btrfs_item_ptr(leaf, path->slots[0],
6048                               struct btrfs_file_extent_item);
6049         /* are we inside the extent that was found? */
6050         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6051         found_type = btrfs_key_type(&found_key);
6052         if (found_key.objectid != objectid ||
6053             found_type != BTRFS_EXTENT_DATA_KEY) {
6054                 /*
6055                  * If we backup past the first extent we want to move forward
6056                  * and see if there is an extent in front of us, otherwise we'll
6057                  * say there is a hole for our whole search range which can
6058                  * cause problems.
6059                  */
6060                 extent_end = start;
6061                 goto next;
6062         }
6063
6064         found_type = btrfs_file_extent_type(leaf, item);
6065         extent_start = found_key.offset;
6066         compress_type = btrfs_file_extent_compression(leaf, item);
6067         if (found_type == BTRFS_FILE_EXTENT_REG ||
6068             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6069                 extent_end = extent_start +
6070                        btrfs_file_extent_num_bytes(leaf, item);
6071         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6072                 size_t size;
6073                 size = btrfs_file_extent_inline_len(leaf, item);
6074                 extent_end = ALIGN(extent_start + size, root->sectorsize);
6075         }
6076 next:
6077         if (start >= extent_end) {
6078                 path->slots[0]++;
6079                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
6080                         ret = btrfs_next_leaf(root, path);
6081                         if (ret < 0) {
6082                                 err = ret;
6083                                 goto out;
6084                         }
6085                         if (ret > 0)
6086                                 goto not_found;
6087                         leaf = path->nodes[0];
6088                 }
6089                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
6090                 if (found_key.objectid != objectid ||
6091                     found_key.type != BTRFS_EXTENT_DATA_KEY)
6092                         goto not_found;
6093                 if (start + len <= found_key.offset)
6094                         goto not_found;
6095                 em->start = start;
6096                 em->orig_start = start;
6097                 em->len = found_key.offset - start;
6098                 goto not_found_em;
6099         }
6100
6101         em->ram_bytes = btrfs_file_extent_ram_bytes(leaf, item);
6102         if (found_type == BTRFS_FILE_EXTENT_REG ||
6103             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
6104                 em->start = extent_start;
6105                 em->len = extent_end - extent_start;
6106                 em->orig_start = extent_start -
6107                                  btrfs_file_extent_offset(leaf, item);
6108                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
6109                                                                       item);
6110                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
6111                 if (bytenr == 0) {
6112                         em->block_start = EXTENT_MAP_HOLE;
6113                         goto insert;
6114                 }
6115                 if (compress_type != BTRFS_COMPRESS_NONE) {
6116                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6117                         em->compress_type = compress_type;
6118                         em->block_start = bytenr;
6119                         em->block_len = em->orig_block_len;
6120                 } else {
6121                         bytenr += btrfs_file_extent_offset(leaf, item);
6122                         em->block_start = bytenr;
6123                         em->block_len = em->len;
6124                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
6125                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6126                 }
6127                 goto insert;
6128         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
6129                 unsigned long ptr;
6130                 char *map;
6131                 size_t size;
6132                 size_t extent_offset;
6133                 size_t copy_size;
6134
6135                 em->block_start = EXTENT_MAP_INLINE;
6136                 if (!page || create) {
6137                         em->start = extent_start;
6138                         em->len = extent_end - extent_start;
6139                         goto out;
6140                 }
6141
6142                 size = btrfs_file_extent_inline_len(leaf, item);
6143                 extent_offset = page_offset(page) + pg_offset - extent_start;
6144                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
6145                                 size - extent_offset);
6146                 em->start = extent_start + extent_offset;
6147                 em->len = ALIGN(copy_size, root->sectorsize);
6148                 em->orig_block_len = em->len;
6149                 em->orig_start = em->start;
6150                 if (compress_type) {
6151                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
6152                         em->compress_type = compress_type;
6153                 }
6154                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
6155                 if (create == 0 && !PageUptodate(page)) {
6156                         if (btrfs_file_extent_compression(leaf, item) !=
6157                             BTRFS_COMPRESS_NONE) {
6158                                 ret = uncompress_inline(path, inode, page,
6159                                                         pg_offset,
6160                                                         extent_offset, item);
6161                                 BUG_ON(ret); /* -ENOMEM */
6162                         } else {
6163                                 map = kmap(page);
6164                                 read_extent_buffer(leaf, map + pg_offset, ptr,
6165                                                    copy_size);
6166                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
6167                                         memset(map + pg_offset + copy_size, 0,
6168                                                PAGE_CACHE_SIZE - pg_offset -
6169                                                copy_size);
6170                                 }
6171                                 kunmap(page);
6172                         }
6173                         flush_dcache_page(page);
6174                 } else if (create && PageUptodate(page)) {
6175                         BUG();
6176                         if (!trans) {
6177                                 kunmap(page);
6178                                 free_extent_map(em);
6179                                 em = NULL;
6180
6181                                 btrfs_release_path(path);
6182                                 trans = btrfs_join_transaction(root);
6183
6184                                 if (IS_ERR(trans))
6185                                         return ERR_CAST(trans);
6186                                 goto again;
6187                         }
6188                         map = kmap(page);
6189                         write_extent_buffer(leaf, map + pg_offset, ptr,
6190                                             copy_size);
6191                         kunmap(page);
6192                         btrfs_mark_buffer_dirty(leaf);
6193                 }
6194                 set_extent_uptodate(io_tree, em->start,
6195                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
6196                 goto insert;
6197         } else {
6198                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
6199         }
6200 not_found:
6201         em->start = start;
6202         em->orig_start = start;
6203         em->len = len;
6204 not_found_em:
6205         em->block_start = EXTENT_MAP_HOLE;
6206         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
6207 insert:
6208         btrfs_release_path(path);
6209         if (em->start > start || extent_map_end(em) <= start) {
6210                 btrfs_err(root->fs_info, "bad extent! em: [%llu %llu] passed [%llu %llu]",
6211                         em->start, em->len, start, len);
6212                 err = -EIO;
6213                 goto out;
6214         }
6215
6216         err = 0;
6217         write_lock(&em_tree->lock);
6218         ret = add_extent_mapping(em_tree, em, 0);
6219         /* it is possible that someone inserted the extent into the tree
6220          * while we had the lock dropped.  It is also possible that
6221          * an overlapping map exists in the tree
6222          */
6223         if (ret == -EEXIST) {
6224                 struct extent_map *existing;
6225
6226                 ret = 0;
6227
6228                 existing = lookup_extent_mapping(em_tree, start, len);
6229                 if (existing && (existing->start > start ||
6230                     existing->start + existing->len <= start)) {
6231                         free_extent_map(existing);
6232                         existing = NULL;
6233                 }
6234                 if (!existing) {
6235                         existing = lookup_extent_mapping(em_tree, em->start,
6236                                                          em->len);
6237                         if (existing) {
6238                                 err = merge_extent_mapping(em_tree, existing,
6239                                                            em, start,
6240                                                            root->sectorsize);
6241                                 free_extent_map(existing);
6242                                 if (err) {
6243                                         free_extent_map(em);
6244                                         em = NULL;
6245                                 }
6246                         } else {
6247                                 err = -EIO;
6248                                 free_extent_map(em);
6249                                 em = NULL;
6250                         }
6251                 } else {
6252                         free_extent_map(em);
6253                         em = existing;
6254                         err = 0;
6255                 }
6256         }
6257         write_unlock(&em_tree->lock);
6258 out:
6259
6260         trace_btrfs_get_extent(root, em);
6261
6262         if (path)
6263                 btrfs_free_path(path);
6264         if (trans) {
6265                 ret = btrfs_end_transaction(trans, root);
6266                 if (!err)
6267                         err = ret;
6268         }
6269         if (err) {
6270                 free_extent_map(em);
6271                 return ERR_PTR(err);
6272         }
6273         BUG_ON(!em); /* Error is always set */
6274         return em;
6275 }
6276
6277 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
6278                                            size_t pg_offset, u64 start, u64 len,
6279                                            int create)
6280 {
6281         struct extent_map *em;
6282         struct extent_map *hole_em = NULL;
6283         u64 range_start = start;
6284         u64 end;
6285         u64 found;
6286         u64 found_end;
6287         int err = 0;
6288
6289         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
6290         if (IS_ERR(em))
6291                 return em;
6292         if (em) {
6293                 /*
6294                  * if our em maps to
6295                  * -  a hole or
6296                  * -  a pre-alloc extent,
6297                  * there might actually be delalloc bytes behind it.
6298                  */
6299                 if (em->block_start != EXTENT_MAP_HOLE &&
6300                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6301                         return em;
6302                 else
6303                         hole_em = em;
6304         }
6305
6306         /* check to see if we've wrapped (len == -1 or similar) */
6307         end = start + len;
6308         if (end < start)
6309                 end = (u64)-1;
6310         else
6311                 end -= 1;
6312
6313         em = NULL;
6314
6315         /* ok, we didn't find anything, lets look for delalloc */
6316         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
6317                                  end, len, EXTENT_DELALLOC, 1);
6318         found_end = range_start + found;
6319         if (found_end < range_start)
6320                 found_end = (u64)-1;
6321
6322         /*
6323          * we didn't find anything useful, return
6324          * the original results from get_extent()
6325          */
6326         if (range_start > end || found_end <= start) {
6327                 em = hole_em;
6328                 hole_em = NULL;
6329                 goto out;
6330         }
6331
6332         /* adjust the range_start to make sure it doesn't
6333          * go backwards from the start they passed in
6334          */
6335         range_start = max(start, range_start);
6336         found = found_end - range_start;
6337
6338         if (found > 0) {
6339                 u64 hole_start = start;
6340                 u64 hole_len = len;
6341
6342                 em = alloc_extent_map();
6343                 if (!em) {
6344                         err = -ENOMEM;
6345                         goto out;
6346                 }
6347                 /*
6348                  * when btrfs_get_extent can't find anything it
6349                  * returns one huge hole
6350                  *
6351                  * make sure what it found really fits our range, and
6352                  * adjust to make sure it is based on the start from
6353                  * the caller
6354                  */
6355                 if (hole_em) {
6356                         u64 calc_end = extent_map_end(hole_em);
6357
6358                         if (calc_end <= start || (hole_em->start > end)) {
6359                                 free_extent_map(hole_em);
6360                                 hole_em = NULL;
6361                         } else {
6362                                 hole_start = max(hole_em->start, start);
6363                                 hole_len = calc_end - hole_start;
6364                         }
6365                 }
6366                 em->bdev = NULL;
6367                 if (hole_em && range_start > hole_start) {
6368                         /* our hole starts before our delalloc, so we
6369                          * have to return just the parts of the hole
6370                          * that go until  the delalloc starts
6371                          */
6372                         em->len = min(hole_len,
6373                                       range_start - hole_start);
6374                         em->start = hole_start;
6375                         em->orig_start = hole_start;
6376                         /*
6377                          * don't adjust block start at all,
6378                          * it is fixed at EXTENT_MAP_HOLE
6379                          */
6380                         em->block_start = hole_em->block_start;
6381                         em->block_len = hole_len;
6382                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
6383                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
6384                 } else {
6385                         em->start = range_start;
6386                         em->len = found;
6387                         em->orig_start = range_start;
6388                         em->block_start = EXTENT_MAP_DELALLOC;
6389                         em->block_len = found;
6390                 }
6391         } else if (hole_em) {
6392                 return hole_em;
6393         }
6394 out:
6395
6396         free_extent_map(hole_em);
6397         if (err) {
6398                 free_extent_map(em);
6399                 return ERR_PTR(err);
6400         }
6401         return em;
6402 }
6403
6404 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
6405                                                   u64 start, u64 len)
6406 {
6407         struct btrfs_root *root = BTRFS_I(inode)->root;
6408         struct extent_map *em;
6409         struct btrfs_key ins;
6410         u64 alloc_hint;
6411         int ret;
6412
6413         alloc_hint = get_extent_allocation_hint(inode, start, len);
6414         ret = btrfs_reserve_extent(root, len, root->sectorsize, 0,
6415                                    alloc_hint, &ins, 1);
6416         if (ret)
6417                 return ERR_PTR(ret);
6418
6419         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
6420                               ins.offset, ins.offset, ins.offset, 0);
6421         if (IS_ERR(em)) {
6422                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6423                 return em;
6424         }
6425
6426         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
6427                                            ins.offset, ins.offset, 0);
6428         if (ret) {
6429                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
6430                 free_extent_map(em);
6431                 return ERR_PTR(ret);
6432         }
6433
6434         return em;
6435 }
6436
6437 /*
6438  * returns 1 when the nocow is safe, < 1 on error, 0 if the
6439  * block must be cow'd
6440  */
6441 noinline int can_nocow_extent(struct inode *inode, u64 offset, u64 *len,
6442                               u64 *orig_start, u64 *orig_block_len,
6443                               u64 *ram_bytes)
6444 {
6445         struct btrfs_trans_handle *trans;
6446         struct btrfs_path *path;
6447         int ret;
6448         struct extent_buffer *leaf;
6449         struct btrfs_root *root = BTRFS_I(inode)->root;
6450         struct btrfs_file_extent_item *fi;
6451         struct btrfs_key key;
6452         u64 disk_bytenr;
6453         u64 backref_offset;
6454         u64 extent_end;
6455         u64 num_bytes;
6456         int slot;
6457         int found_type;
6458         bool nocow = (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW);
6459         path = btrfs_alloc_path();
6460         if (!path)
6461                 return -ENOMEM;
6462
6463         ret = btrfs_lookup_file_extent(NULL, root, path, btrfs_ino(inode),
6464                                        offset, 0);
6465         if (ret < 0)
6466                 goto out;
6467
6468         slot = path->slots[0];
6469         if (ret == 1) {
6470                 if (slot == 0) {
6471                         /* can't find the item, must cow */
6472                         ret = 0;
6473                         goto out;
6474                 }
6475                 slot--;
6476         }
6477         ret = 0;
6478         leaf = path->nodes[0];
6479         btrfs_item_key_to_cpu(leaf, &key, slot);
6480         if (key.objectid != btrfs_ino(inode) ||
6481             key.type != BTRFS_EXTENT_DATA_KEY) {
6482                 /* not our file or wrong item type, must cow */
6483                 goto out;
6484         }
6485
6486         if (key.offset > offset) {
6487                 /* Wrong offset, must cow */
6488                 goto out;
6489         }
6490
6491         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
6492         found_type = btrfs_file_extent_type(leaf, fi);
6493         if (found_type != BTRFS_FILE_EXTENT_REG &&
6494             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
6495                 /* not a regular extent, must cow */
6496                 goto out;
6497         }
6498
6499         if (!nocow && found_type == BTRFS_FILE_EXTENT_REG)
6500                 goto out;
6501
6502         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
6503         if (disk_bytenr == 0)
6504                 goto out;
6505
6506         if (btrfs_file_extent_compression(leaf, fi) ||
6507             btrfs_file_extent_encryption(leaf, fi) ||
6508             btrfs_file_extent_other_encoding(leaf, fi))
6509                 goto out;
6510
6511         backref_offset = btrfs_file_extent_offset(leaf, fi);
6512
6513         if (orig_start) {
6514                 *orig_start = key.offset - backref_offset;
6515                 *orig_block_len = btrfs_file_extent_disk_num_bytes(leaf, fi);
6516                 *ram_bytes = btrfs_file_extent_ram_bytes(leaf, fi);
6517         }
6518
6519         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
6520
6521         if (btrfs_extent_readonly(root, disk_bytenr))
6522                 goto out;
6523         btrfs_release_path(path);
6524
6525         /*
6526          * look for other files referencing this extent, if we
6527          * find any we must cow
6528          */
6529         trans = btrfs_join_transaction(root);
6530         if (IS_ERR(trans)) {
6531                 ret = 0;
6532                 goto out;
6533         }
6534
6535         ret = btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
6536                                     key.offset - backref_offset, disk_bytenr);
6537         btrfs_end_transaction(trans, root);
6538         if (ret) {
6539                 ret = 0;
6540                 goto out;
6541         }
6542
6543         /*
6544          * adjust disk_bytenr and num_bytes to cover just the bytes
6545          * in this extent we are about to write.  If there
6546          * are any csums in that range we have to cow in order
6547          * to keep the csums correct
6548          */
6549         disk_bytenr += backref_offset;
6550         disk_bytenr += offset - key.offset;
6551         num_bytes = min(offset + *len, extent_end) - offset;
6552         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
6553                                 goto out;
6554         /*
6555          * all of the above have passed, it is safe to overwrite this extent
6556          * without cow
6557          */
6558         *len = num_bytes;
6559         ret = 1;
6560 out:
6561         btrfs_free_path(path);
6562         return ret;
6563 }
6564
6565 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
6566                               struct extent_state **cached_state, int writing)
6567 {
6568         struct btrfs_ordered_extent *ordered;
6569         int ret = 0;
6570
6571         while (1) {
6572                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6573                                  0, cached_state);
6574                 /*
6575                  * We're concerned with the entire range that we're going to be
6576                  * doing DIO to, so we need to make sure theres no ordered
6577                  * extents in this range.
6578                  */
6579                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6580                                                      lockend - lockstart + 1);
6581
6582                 /*
6583                  * We need to make sure there are no buffered pages in this
6584                  * range either, we could have raced between the invalidate in
6585                  * generic_file_direct_write and locking the extent.  The
6586                  * invalidate needs to happen so that reads after a write do not
6587                  * get stale data.
6588                  */
6589                 if (!ordered && (!writing ||
6590                     !test_range_bit(&BTRFS_I(inode)->io_tree,
6591                                     lockstart, lockend, EXTENT_UPTODATE, 0,
6592                                     *cached_state)))
6593                         break;
6594
6595                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6596                                      cached_state, GFP_NOFS);
6597
6598                 if (ordered) {
6599                         btrfs_start_ordered_extent(inode, ordered, 1);
6600                         btrfs_put_ordered_extent(ordered);
6601                 } else {
6602                         /* Screw you mmap */
6603                         ret = filemap_write_and_wait_range(inode->i_mapping,
6604                                                            lockstart,
6605                                                            lockend);
6606                         if (ret)
6607                                 break;
6608
6609                         /*
6610                          * If we found a page that couldn't be invalidated just
6611                          * fall back to buffered.
6612                          */
6613                         ret = invalidate_inode_pages2_range(inode->i_mapping,
6614                                         lockstart >> PAGE_CACHE_SHIFT,
6615                                         lockend >> PAGE_CACHE_SHIFT);
6616                         if (ret)
6617                                 break;
6618                 }
6619
6620                 cond_resched();
6621         }
6622
6623         return ret;
6624 }
6625
6626 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6627                                            u64 len, u64 orig_start,
6628                                            u64 block_start, u64 block_len,
6629                                            u64 orig_block_len, u64 ram_bytes,
6630                                            int type)
6631 {
6632         struct extent_map_tree *em_tree;
6633         struct extent_map *em;
6634         struct btrfs_root *root = BTRFS_I(inode)->root;
6635         int ret;
6636
6637         em_tree = &BTRFS_I(inode)->extent_tree;
6638         em = alloc_extent_map();
6639         if (!em)
6640                 return ERR_PTR(-ENOMEM);
6641
6642         em->start = start;
6643         em->orig_start = orig_start;
6644         em->mod_start = start;
6645         em->mod_len = len;
6646         em->len = len;
6647         em->block_len = block_len;
6648         em->block_start = block_start;
6649         em->bdev = root->fs_info->fs_devices->latest_bdev;
6650         em->orig_block_len = orig_block_len;
6651         em->ram_bytes = ram_bytes;
6652         em->generation = -1;
6653         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6654         if (type == BTRFS_ORDERED_PREALLOC)
6655                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6656
6657         do {
6658                 btrfs_drop_extent_cache(inode, em->start,
6659                                 em->start + em->len - 1, 0);
6660                 write_lock(&em_tree->lock);
6661                 ret = add_extent_mapping(em_tree, em, 1);
6662                 write_unlock(&em_tree->lock);
6663         } while (ret == -EEXIST);
6664
6665         if (ret) {
6666                 free_extent_map(em);
6667                 return ERR_PTR(ret);
6668         }
6669
6670         return em;
6671 }
6672
6673
6674 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6675                                    struct buffer_head *bh_result, int create)
6676 {
6677         struct extent_map *em;
6678         struct btrfs_root *root = BTRFS_I(inode)->root;
6679         struct extent_state *cached_state = NULL;
6680         u64 start = iblock << inode->i_blkbits;
6681         u64 lockstart, lockend;
6682         u64 len = bh_result->b_size;
6683         int unlock_bits = EXTENT_LOCKED;
6684         int ret = 0;
6685
6686         if (create)
6687                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6688         else
6689                 len = min_t(u64, len, root->sectorsize);
6690
6691         lockstart = start;
6692         lockend = start + len - 1;
6693
6694         /*
6695          * If this errors out it's because we couldn't invalidate pagecache for
6696          * this range and we need to fallback to buffered.
6697          */
6698         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6699                 return -ENOTBLK;
6700
6701         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6702         if (IS_ERR(em)) {
6703                 ret = PTR_ERR(em);
6704                 goto unlock_err;
6705         }
6706
6707         /*
6708          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6709          * io.  INLINE is special, and we could probably kludge it in here, but
6710          * it's still buffered so for safety lets just fall back to the generic
6711          * buffered path.
6712          *
6713          * For COMPRESSED we _have_ to read the entire extent in so we can
6714          * decompress it, so there will be buffering required no matter what we
6715          * do, so go ahead and fallback to buffered.
6716          *
6717          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6718          * to buffered IO.  Don't blame me, this is the price we pay for using
6719          * the generic code.
6720          */
6721         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6722             em->block_start == EXTENT_MAP_INLINE) {
6723                 free_extent_map(em);
6724                 ret = -ENOTBLK;
6725                 goto unlock_err;
6726         }
6727
6728         /* Just a good old fashioned hole, return */
6729         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6730                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6731                 free_extent_map(em);
6732                 goto unlock_err;
6733         }
6734
6735         /*
6736          * We don't allocate a new extent in the following cases
6737          *
6738          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6739          * existing extent.
6740          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6741          * just use the extent.
6742          *
6743          */
6744         if (!create) {
6745                 len = min(len, em->len - (start - em->start));
6746                 lockstart = start + len;
6747                 goto unlock;
6748         }
6749
6750         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6751             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6752              em->block_start != EXTENT_MAP_HOLE)) {
6753                 int type;
6754                 int ret;
6755                 u64 block_start, orig_start, orig_block_len, ram_bytes;
6756
6757                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6758                         type = BTRFS_ORDERED_PREALLOC;
6759                 else
6760                         type = BTRFS_ORDERED_NOCOW;
6761                 len = min(len, em->len - (start - em->start));
6762                 block_start = em->block_start + (start - em->start);
6763
6764                 if (can_nocow_extent(inode, start, &len, &orig_start,
6765                                      &orig_block_len, &ram_bytes) == 1) {
6766                         if (type == BTRFS_ORDERED_PREALLOC) {
6767                                 free_extent_map(em);
6768                                 em = create_pinned_em(inode, start, len,
6769                                                        orig_start,
6770                                                        block_start, len,
6771                                                        orig_block_len,
6772                                                        ram_bytes, type);
6773                                 if (IS_ERR(em))
6774                                         goto unlock_err;
6775                         }
6776
6777                         ret = btrfs_add_ordered_extent_dio(inode, start,
6778                                            block_start, len, len, type);
6779                         if (ret) {
6780                                 free_extent_map(em);
6781                                 goto unlock_err;
6782                         }
6783                         goto unlock;
6784                 }
6785         }
6786
6787         /*
6788          * this will cow the extent, reset the len in case we changed
6789          * it above
6790          */
6791         len = bh_result->b_size;
6792         free_extent_map(em);
6793         em = btrfs_new_extent_direct(inode, start, len);
6794         if (IS_ERR(em)) {
6795                 ret = PTR_ERR(em);
6796                 goto unlock_err;
6797         }
6798         len = min(len, em->len - (start - em->start));
6799 unlock:
6800         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6801                 inode->i_blkbits;
6802         bh_result->b_size = len;
6803         bh_result->b_bdev = em->bdev;
6804         set_buffer_mapped(bh_result);
6805         if (create) {
6806                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6807                         set_buffer_new(bh_result);
6808
6809                 /*
6810                  * Need to update the i_size under the extent lock so buffered
6811                  * readers will get the updated i_size when we unlock.
6812                  */
6813                 if (start + len > i_size_read(inode))
6814                         i_size_write(inode, start + len);
6815
6816                 spin_lock(&BTRFS_I(inode)->lock);
6817                 BTRFS_I(inode)->outstanding_extents++;
6818                 spin_unlock(&BTRFS_I(inode)->lock);
6819
6820                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6821                                      lockstart + len - 1, EXTENT_DELALLOC, NULL,
6822                                      &cached_state, GFP_NOFS);
6823                 BUG_ON(ret);
6824         }
6825
6826         /*
6827          * In the case of write we need to clear and unlock the entire range,
6828          * in the case of read we need to unlock only the end area that we
6829          * aren't using if there is any left over space.
6830          */
6831         if (lockstart < lockend) {
6832                 clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6833                                  lockend, unlock_bits, 1, 0,
6834                                  &cached_state, GFP_NOFS);
6835         } else {
6836                 free_extent_state(cached_state);
6837         }
6838
6839         free_extent_map(em);
6840
6841         return 0;
6842
6843 unlock_err:
6844         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6845                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6846         return ret;
6847 }
6848
6849 static void btrfs_endio_direct_read(struct bio *bio, int err)
6850 {
6851         struct btrfs_dio_private *dip = bio->bi_private;
6852         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6853         struct bio_vec *bvec = bio->bi_io_vec;
6854         struct inode *inode = dip->inode;
6855         struct btrfs_root *root = BTRFS_I(inode)->root;
6856         struct bio *dio_bio;
6857         u32 *csums = (u32 *)dip->csum;
6858         int index = 0;
6859         u64 start;
6860
6861         start = dip->logical_offset;
6862         do {
6863                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6864                         struct page *page = bvec->bv_page;
6865                         char *kaddr;
6866                         u32 csum = ~(u32)0;
6867                         unsigned long flags;
6868
6869                         local_irq_save(flags);
6870                         kaddr = kmap_atomic(page);
6871                         csum = btrfs_csum_data(kaddr + bvec->bv_offset,
6872                                                csum, bvec->bv_len);
6873                         btrfs_csum_final(csum, (char *)&csum);
6874                         kunmap_atomic(kaddr);
6875                         local_irq_restore(flags);
6876
6877                         flush_dcache_page(bvec->bv_page);
6878                         if (csum != csums[index]) {
6879                                 btrfs_err(root->fs_info, "csum failed ino %llu off %llu csum %u expected csum %u",
6880                                           btrfs_ino(inode), start, csum,
6881                                           csums[index]);
6882                                 err = -EIO;
6883                         }
6884                 }
6885
6886                 start += bvec->bv_len;
6887                 bvec++;
6888                 index++;
6889         } while (bvec <= bvec_end);
6890
6891         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6892                       dip->logical_offset + dip->bytes - 1);
6893         dio_bio = dip->dio_bio;
6894
6895         kfree(dip);
6896
6897         /* If we had a csum failure make sure to clear the uptodate flag */
6898         if (err)
6899                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6900         dio_end_io(dio_bio, err);
6901         bio_put(bio);
6902 }
6903
6904 static void btrfs_endio_direct_write(struct bio *bio, int err)
6905 {
6906         struct btrfs_dio_private *dip = bio->bi_private;
6907         struct inode *inode = dip->inode;
6908         struct btrfs_root *root = BTRFS_I(inode)->root;
6909         struct btrfs_ordered_extent *ordered = NULL;
6910         u64 ordered_offset = dip->logical_offset;
6911         u64 ordered_bytes = dip->bytes;
6912         struct bio *dio_bio;
6913         int ret;
6914
6915         if (err)
6916                 goto out_done;
6917 again:
6918         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6919                                                    &ordered_offset,
6920                                                    ordered_bytes, !err);
6921         if (!ret)
6922                 goto out_test;
6923
6924         ordered->work.func = finish_ordered_fn;
6925         ordered->work.flags = 0;
6926         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6927                            &ordered->work);
6928 out_test:
6929         /*
6930          * our bio might span multiple ordered extents.  If we haven't
6931          * completed the accounting for the whole dio, go back and try again
6932          */
6933         if (ordered_offset < dip->logical_offset + dip->bytes) {
6934                 ordered_bytes = dip->logical_offset + dip->bytes -
6935                         ordered_offset;
6936                 ordered = NULL;
6937                 goto again;
6938         }
6939 out_done:
6940         dio_bio = dip->dio_bio;
6941
6942         kfree(dip);
6943
6944         /* If we had an error make sure to clear the uptodate flag */
6945         if (err)
6946                 clear_bit(BIO_UPTODATE, &dio_bio->bi_flags);
6947         dio_end_io(dio_bio, err);
6948         bio_put(bio);
6949 }
6950
6951 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6952                                     struct bio *bio, int mirror_num,
6953                                     unsigned long bio_flags, u64 offset)
6954 {
6955         int ret;
6956         struct btrfs_root *root = BTRFS_I(inode)->root;
6957         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6958         BUG_ON(ret); /* -ENOMEM */
6959         return 0;
6960 }
6961
6962 static void btrfs_end_dio_bio(struct bio *bio, int err)
6963 {
6964         struct btrfs_dio_private *dip = bio->bi_private;
6965
6966         if (err) {
6967                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6968                       "sector %#Lx len %u err no %d\n",
6969                       btrfs_ino(dip->inode), bio->bi_rw,
6970                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6971                 dip->errors = 1;
6972
6973                 /*
6974                  * before atomic variable goto zero, we must make sure
6975                  * dip->errors is perceived to be set.
6976                  */
6977                 smp_mb__before_atomic_dec();
6978         }
6979
6980         /* if there are more bios still pending for this dio, just exit */
6981         if (!atomic_dec_and_test(&dip->pending_bios))
6982                 goto out;
6983
6984         if (dip->errors) {
6985                 bio_io_error(dip->orig_bio);
6986         } else {
6987                 set_bit(BIO_UPTODATE, &dip->dio_bio->bi_flags);
6988                 bio_endio(dip->orig_bio, 0);
6989         }
6990 out:
6991         bio_put(bio);
6992 }
6993
6994 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6995                                        u64 first_sector, gfp_t gfp_flags)
6996 {
6997         int nr_vecs = bio_get_nr_vecs(bdev);
6998         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6999 }
7000
7001 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
7002                                          int rw, u64 file_offset, int skip_sum,
7003                                          int async_submit)
7004 {
7005         struct btrfs_dio_private *dip = bio->bi_private;
7006         int write = rw & REQ_WRITE;
7007         struct btrfs_root *root = BTRFS_I(inode)->root;
7008         int ret;
7009
7010         if (async_submit)
7011                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
7012
7013         bio_get(bio);
7014
7015         if (!write) {
7016                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
7017                 if (ret)
7018                         goto err;
7019         }
7020
7021         if (skip_sum)
7022                 goto map;
7023
7024         if (write && async_submit) {
7025                 ret = btrfs_wq_submit_bio(root->fs_info,
7026                                    inode, rw, bio, 0, 0,
7027                                    file_offset,
7028                                    __btrfs_submit_bio_start_direct_io,
7029                                    __btrfs_submit_bio_done);
7030                 goto err;
7031         } else if (write) {
7032                 /*
7033                  * If we aren't doing async submit, calculate the csum of the
7034                  * bio now.
7035                  */
7036                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
7037                 if (ret)
7038                         goto err;
7039         } else if (!skip_sum) {
7040                 ret = btrfs_lookup_bio_sums_dio(root, inode, dip, bio,
7041                                                 file_offset);
7042                 if (ret)
7043                         goto err;
7044         }
7045
7046 map:
7047         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
7048 err:
7049         bio_put(bio);
7050         return ret;
7051 }
7052
7053 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
7054                                     int skip_sum)
7055 {
7056         struct inode *inode = dip->inode;
7057         struct btrfs_root *root = BTRFS_I(inode)->root;
7058         struct bio *bio;
7059         struct bio *orig_bio = dip->orig_bio;
7060         struct bio_vec *bvec = orig_bio->bi_io_vec;
7061         u64 start_sector = orig_bio->bi_sector;
7062         u64 file_offset = dip->logical_offset;
7063         u64 submit_len = 0;
7064         u64 map_length;
7065         int nr_pages = 0;
7066         int ret = 0;
7067         int async_submit = 0;
7068
7069         map_length = orig_bio->bi_size;
7070         ret = btrfs_map_block(root->fs_info, rw, start_sector << 9,
7071                               &map_length, NULL, 0);
7072         if (ret) {
7073                 bio_put(orig_bio);
7074                 return -EIO;
7075         }
7076
7077         if (map_length >= orig_bio->bi_size) {
7078                 bio = orig_bio;
7079                 goto submit;
7080         }
7081
7082         /* async crcs make it difficult to collect full stripe writes. */
7083         if (btrfs_get_alloc_profile(root, 1) &
7084             (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6))
7085                 async_submit = 0;
7086         else
7087                 async_submit = 1;
7088
7089         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
7090         if (!bio)
7091                 return -ENOMEM;
7092         bio->bi_private = dip;
7093         bio->bi_end_io = btrfs_end_dio_bio;
7094         atomic_inc(&dip->pending_bios);
7095
7096         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
7097                 if (unlikely(map_length < submit_len + bvec->bv_len ||
7098                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
7099                                  bvec->bv_offset) < bvec->bv_len)) {
7100                         /*
7101                          * inc the count before we submit the bio so
7102                          * we know the end IO handler won't happen before
7103                          * we inc the count. Otherwise, the dip might get freed
7104                          * before we're done setting it up
7105                          */
7106                         atomic_inc(&dip->pending_bios);
7107                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
7108                                                      file_offset, skip_sum,
7109                                                      async_submit);
7110                         if (ret) {
7111                                 bio_put(bio);
7112                                 atomic_dec(&dip->pending_bios);
7113                                 goto out_err;
7114                         }
7115
7116                         start_sector += submit_len >> 9;
7117                         file_offset += submit_len;
7118
7119                         submit_len = 0;
7120                         nr_pages = 0;
7121
7122                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
7123                                                   start_sector, GFP_NOFS);
7124                         if (!bio)
7125                                 goto out_err;
7126                         bio->bi_private = dip;
7127                         bio->bi_end_io = btrfs_end_dio_bio;
7128
7129                         map_length = orig_bio->bi_size;
7130                         ret = btrfs_map_block(root->fs_info, rw,
7131                                               start_sector << 9,
7132                                               &map_length, NULL, 0);
7133                         if (ret) {
7134                                 bio_put(bio);
7135                                 goto out_err;
7136                         }
7137                 } else {
7138                         submit_len += bvec->bv_len;
7139                         nr_pages++;
7140                         bvec++;
7141                 }
7142         }
7143
7144 submit:
7145         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
7146                                      async_submit);
7147         if (!ret)
7148                 return 0;
7149
7150         bio_put(bio);
7151 out_err:
7152         dip->errors = 1;
7153         /*
7154          * before atomic variable goto zero, we must
7155          * make sure dip->errors is perceived to be set.
7156          */
7157         smp_mb__before_atomic_dec();
7158         if (atomic_dec_and_test(&dip->pending_bios))
7159                 bio_io_error(dip->orig_bio);
7160
7161         /* bio_end_io() will handle error, so we needn't return it */
7162         return 0;
7163 }
7164
7165 static void btrfs_submit_direct(int rw, struct bio *dio_bio,
7166                                 struct inode *inode, loff_t file_offset)
7167 {
7168         struct btrfs_root *root = BTRFS_I(inode)->root;
7169         struct btrfs_dio_private *dip;
7170         struct bio *io_bio;
7171         int skip_sum;
7172         int sum_len;
7173         int write = rw & REQ_WRITE;
7174         int ret = 0;
7175         u16 csum_size;
7176
7177         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
7178
7179         io_bio = btrfs_bio_clone(dio_bio, GFP_NOFS);
7180         if (!io_bio) {
7181                 ret = -ENOMEM;
7182                 goto free_ordered;
7183         }
7184
7185         if (!skip_sum && !write) {
7186                 csum_size = btrfs_super_csum_size(root->fs_info->super_copy);
7187                 sum_len = dio_bio->bi_size >> inode->i_sb->s_blocksize_bits;
7188                 sum_len *= csum_size;
7189         } else {
7190                 sum_len = 0;
7191         }
7192
7193         dip = kmalloc(sizeof(*dip) + sum_len, GFP_NOFS);
7194         if (!dip) {
7195                 ret = -ENOMEM;
7196                 goto free_io_bio;
7197         }
7198
7199         dip->private = dio_bio->bi_private;
7200         dip->inode = inode;
7201         dip->logical_offset = file_offset;
7202         dip->bytes = dio_bio->bi_size;
7203         dip->disk_bytenr = (u64)dio_bio->bi_sector << 9;
7204         io_bio->bi_private = dip;
7205         dip->errors = 0;
7206         dip->orig_bio = io_bio;
7207         dip->dio_bio = dio_bio;
7208         atomic_set(&dip->pending_bios, 0);
7209
7210         if (write)
7211                 io_bio->bi_end_io = btrfs_endio_direct_write;
7212         else
7213                 io_bio->bi_end_io = btrfs_endio_direct_read;
7214
7215         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
7216         if (!ret)
7217                 return;
7218
7219 free_io_bio:
7220         bio_put(io_bio);
7221
7222 free_ordered:
7223         /*
7224          * If this is a write, we need to clean up the reserved space and kill
7225          * the ordered extent.
7226          */
7227         if (write) {
7228                 struct btrfs_ordered_extent *ordered;
7229                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
7230                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
7231                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
7232                         btrfs_free_reserved_extent(root, ordered->start,
7233                                                    ordered->disk_len);
7234                 btrfs_put_ordered_extent(ordered);
7235                 btrfs_put_ordered_extent(ordered);
7236         }
7237         bio_endio(dio_bio, ret);
7238 }
7239
7240 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
7241                         const struct iovec *iov, loff_t offset,
7242                         unsigned long nr_segs)
7243 {
7244         int seg;
7245         int i;
7246         size_t size;
7247         unsigned long addr;
7248         unsigned blocksize_mask = root->sectorsize - 1;
7249         ssize_t retval = -EINVAL;
7250         loff_t end = offset;
7251
7252         if (offset & blocksize_mask)
7253                 goto out;
7254
7255         /* Check the memory alignment.  Blocks cannot straddle pages */
7256         for (seg = 0; seg < nr_segs; seg++) {
7257                 addr = (unsigned long)iov[seg].iov_base;
7258                 size = iov[seg].iov_len;
7259                 end += size;
7260                 if ((addr & blocksize_mask) || (size & blocksize_mask))
7261                         goto out;
7262
7263                 /* If this is a write we don't need to check anymore */
7264                 if (rw & WRITE)
7265                         continue;
7266
7267                 /*
7268                  * Check to make sure we don't have duplicate iov_base's in this
7269                  * iovec, if so return EINVAL, otherwise we'll get csum errors
7270                  * when reading back.
7271                  */
7272                 for (i = seg + 1; i < nr_segs; i++) {
7273                         if (iov[seg].iov_base == iov[i].iov_base)
7274                                 goto out;
7275                 }
7276         }
7277         retval = 0;
7278 out:
7279         return retval;
7280 }
7281
7282 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
7283                         const struct iovec *iov, loff_t offset,
7284                         unsigned long nr_segs)
7285 {
7286         struct file *file = iocb->ki_filp;
7287         struct inode *inode = file->f_mapping->host;
7288         size_t count = 0;
7289         int flags = 0;
7290         bool wakeup = true;
7291         bool relock = false;
7292         ssize_t ret;
7293
7294         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
7295                             offset, nr_segs))
7296                 return 0;
7297
7298         atomic_inc(&inode->i_dio_count);
7299         smp_mb__after_atomic_inc();
7300
7301         /*
7302          * The generic stuff only does filemap_write_and_wait_range, which isn't
7303          * enough if we've written compressed pages to this area, so we need to
7304          * call btrfs_wait_ordered_range to make absolutely sure that any
7305          * outstanding dirty pages are on disk.
7306          */
7307         count = iov_length(iov, nr_segs);
7308         ret = btrfs_wait_ordered_range(inode, offset, count);
7309         if (ret)
7310                 return ret;
7311
7312         if (rw & WRITE) {
7313                 /*
7314                  * If the write DIO is beyond the EOF, we need update
7315                  * the isize, but it is protected by i_mutex. So we can
7316                  * not unlock the i_mutex at this case.
7317                  */
7318                 if (offset + count <= inode->i_size) {
7319                         mutex_unlock(&inode->i_mutex);
7320                         relock = true;
7321                 }
7322                 ret = btrfs_delalloc_reserve_space(inode, count);
7323                 if (ret)
7324                         goto out;
7325         } else if (unlikely(test_bit(BTRFS_INODE_READDIO_NEED_LOCK,
7326                                      &BTRFS_I(inode)->runtime_flags))) {
7327                 inode_dio_done(inode);
7328                 flags = DIO_LOCKING | DIO_SKIP_HOLES;
7329                 wakeup = false;
7330         }
7331
7332         ret = __blockdev_direct_IO(rw, iocb, inode,
7333                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
7334                         iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
7335                         btrfs_submit_direct, flags);
7336         if (rw & WRITE) {
7337                 if (ret < 0 && ret != -EIOCBQUEUED)
7338                         btrfs_delalloc_release_space(inode, count);
7339                 else if (ret >= 0 && (size_t)ret < count)
7340                         btrfs_delalloc_release_space(inode,
7341                                                      count - (size_t)ret);
7342                 else
7343                         btrfs_delalloc_release_metadata(inode, 0);
7344         }
7345 out:
7346         if (wakeup)
7347                 inode_dio_done(inode);
7348         if (relock)
7349                 mutex_lock(&inode->i_mutex);
7350
7351         return ret;
7352 }
7353
7354 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
7355
7356 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
7357                 __u64 start, __u64 len)
7358 {
7359         int     ret;
7360
7361         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
7362         if (ret)
7363                 return ret;
7364
7365         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
7366 }
7367
7368 int btrfs_readpage(struct file *file, struct page *page)
7369 {
7370         struct extent_io_tree *tree;
7371         tree = &BTRFS_I(page->mapping->host)->io_tree;
7372         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
7373 }
7374
7375 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
7376 {
7377         struct extent_io_tree *tree;
7378
7379
7380         if (current->flags & PF_MEMALLOC) {
7381                 redirty_page_for_writepage(wbc, page);
7382                 unlock_page(page);
7383                 return 0;
7384         }
7385         tree = &BTRFS_I(page->mapping->host)->io_tree;
7386         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
7387 }
7388
7389 static int btrfs_writepages(struct address_space *mapping,
7390                             struct writeback_control *wbc)
7391 {
7392         struct extent_io_tree *tree;
7393
7394         tree = &BTRFS_I(mapping->host)->io_tree;
7395         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
7396 }
7397
7398 static int
7399 btrfs_readpages(struct file *file, struct address_space *mapping,
7400                 struct list_head *pages, unsigned nr_pages)
7401 {
7402         struct extent_io_tree *tree;
7403         tree = &BTRFS_I(mapping->host)->io_tree;
7404         return extent_readpages(tree, mapping, pages, nr_pages,
7405                                 btrfs_get_extent);
7406 }
7407 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7408 {
7409         struct extent_io_tree *tree;
7410         struct extent_map_tree *map;
7411         int ret;
7412
7413         tree = &BTRFS_I(page->mapping->host)->io_tree;
7414         map = &BTRFS_I(page->mapping->host)->extent_tree;
7415         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
7416         if (ret == 1) {
7417                 ClearPagePrivate(page);
7418                 set_page_private(page, 0);
7419                 page_cache_release(page);
7420         }
7421         return ret;
7422 }
7423
7424 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
7425 {
7426         if (PageWriteback(page) || PageDirty(page))
7427                 return 0;
7428         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
7429 }
7430
7431 static void btrfs_invalidatepage(struct page *page, unsigned int offset,
7432                                  unsigned int length)
7433 {
7434         struct inode *inode = page->mapping->host;
7435         struct extent_io_tree *tree;
7436         struct btrfs_ordered_extent *ordered;
7437         struct extent_state *cached_state = NULL;
7438         u64 page_start = page_offset(page);
7439         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
7440         int inode_evicting = inode->i_state & I_FREEING;
7441
7442         /*
7443          * we have the page locked, so new writeback can't start,
7444          * and the dirty bit won't be cleared while we are here.
7445          *
7446          * Wait for IO on this page so that we can safely clear
7447          * the PagePrivate2 bit and do ordered accounting
7448          */
7449         wait_on_page_writeback(page);
7450
7451         tree = &BTRFS_I(inode)->io_tree;
7452         if (offset) {
7453                 btrfs_releasepage(page, GFP_NOFS);
7454                 return;
7455         }
7456
7457         if (!inode_evicting)
7458                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
7459         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7460         if (ordered) {
7461                 /*
7462                  * IO on this page will never be started, so we need
7463                  * to account for any ordered extents now
7464                  */
7465                 if (!inode_evicting)
7466                         clear_extent_bit(tree, page_start, page_end,
7467                                          EXTENT_DIRTY | EXTENT_DELALLOC |
7468                                          EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
7469                                          EXTENT_DEFRAG, 1, 0, &cached_state,
7470                                          GFP_NOFS);
7471                 /*
7472                  * whoever cleared the private bit is responsible
7473                  * for the finish_ordered_io
7474                  */
7475                 if (TestClearPagePrivate2(page)) {
7476                         struct btrfs_ordered_inode_tree *tree;
7477                         u64 new_len;
7478
7479                         tree = &BTRFS_I(inode)->ordered_tree;
7480
7481                         spin_lock_irq(&tree->lock);
7482                         set_bit(BTRFS_ORDERED_TRUNCATED, &ordered->flags);
7483                         new_len = page_start - ordered->file_offset;
7484                         if (new_len < ordered->truncated_len)
7485                                 ordered->truncated_len = new_len;
7486                         spin_unlock_irq(&tree->lock);
7487
7488                         if (btrfs_dec_test_ordered_pending(inode, &ordered,
7489                                                            page_start,
7490                                                            PAGE_CACHE_SIZE, 1))
7491                                 btrfs_finish_ordered_io(ordered);
7492                 }
7493                 btrfs_put_ordered_extent(ordered);
7494                 if (!inode_evicting) {
7495                         cached_state = NULL;
7496                         lock_extent_bits(tree, page_start, page_end, 0,
7497                                          &cached_state);
7498                 }
7499         }
7500
7501         if (!inode_evicting) {
7502                 clear_extent_bit(tree, page_start, page_end,
7503                                  EXTENT_LOCKED | EXTENT_DIRTY |
7504                                  EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING |
7505                                  EXTENT_DEFRAG, 1, 1,
7506                                  &cached_state, GFP_NOFS);
7507
7508                 __btrfs_releasepage(page, GFP_NOFS);
7509         }
7510
7511         ClearPageChecked(page);
7512         if (PagePrivate(page)) {
7513                 ClearPagePrivate(page);
7514                 set_page_private(page, 0);
7515                 page_cache_release(page);
7516         }
7517 }
7518
7519 /*
7520  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
7521  * called from a page fault handler when a page is first dirtied. Hence we must
7522  * be careful to check for EOF conditions here. We set the page up correctly
7523  * for a written page which means we get ENOSPC checking when writing into
7524  * holes and correct delalloc and unwritten extent mapping on filesystems that
7525  * support these features.
7526  *
7527  * We are not allowed to take the i_mutex here so we have to play games to
7528  * protect against truncate races as the page could now be beyond EOF.  Because
7529  * vmtruncate() writes the inode size before removing pages, once we have the
7530  * page lock we can determine safely if the page is beyond EOF. If it is not
7531  * beyond EOF, then the page is guaranteed safe against truncation until we
7532  * unlock the page.
7533  */
7534 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
7535 {
7536         struct page *page = vmf->page;
7537         struct inode *inode = file_inode(vma->vm_file);
7538         struct btrfs_root *root = BTRFS_I(inode)->root;
7539         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
7540         struct btrfs_ordered_extent *ordered;
7541         struct extent_state *cached_state = NULL;
7542         char *kaddr;
7543         unsigned long zero_start;
7544         loff_t size;
7545         int ret;
7546         int reserved = 0;
7547         u64 page_start;
7548         u64 page_end;
7549
7550         sb_start_pagefault(inode->i_sb);
7551         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
7552         if (!ret) {
7553                 ret = file_update_time(vma->vm_file);
7554                 reserved = 1;
7555         }
7556         if (ret) {
7557                 if (ret == -ENOMEM)
7558                         ret = VM_FAULT_OOM;
7559                 else /* -ENOSPC, -EIO, etc */
7560                         ret = VM_FAULT_SIGBUS;
7561                 if (reserved)
7562                         goto out;
7563                 goto out_noreserve;
7564         }
7565
7566         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
7567 again:
7568         lock_page(page);
7569         size = i_size_read(inode);
7570         page_start = page_offset(page);
7571         page_end = page_start + PAGE_CACHE_SIZE - 1;
7572
7573         if ((page->mapping != inode->i_mapping) ||
7574             (page_start >= size)) {
7575                 /* page got truncated out from underneath us */
7576                 goto out_unlock;
7577         }
7578         wait_on_page_writeback(page);
7579
7580         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
7581         set_page_extent_mapped(page);
7582
7583         /*
7584          * we can't set the delalloc bits if there are pending ordered
7585          * extents.  Drop our locks and wait for them to finish
7586          */
7587         ordered = btrfs_lookup_ordered_extent(inode, page_start);
7588         if (ordered) {
7589                 unlock_extent_cached(io_tree, page_start, page_end,
7590                                      &cached_state, GFP_NOFS);
7591                 unlock_page(page);
7592                 btrfs_start_ordered_extent(inode, ordered, 1);
7593                 btrfs_put_ordered_extent(ordered);
7594                 goto again;
7595         }
7596
7597         /*
7598          * XXX - page_mkwrite gets called every time the page is dirtied, even
7599          * if it was already dirty, so for space accounting reasons we need to
7600          * clear any delalloc bits for the range we are fixing to save.  There
7601          * is probably a better way to do this, but for now keep consistent with
7602          * prepare_pages in the normal write path.
7603          */
7604         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
7605                           EXTENT_DIRTY | EXTENT_DELALLOC |
7606                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
7607                           0, 0, &cached_state, GFP_NOFS);
7608
7609         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
7610                                         &cached_state);
7611         if (ret) {
7612                 unlock_extent_cached(io_tree, page_start, page_end,
7613                                      &cached_state, GFP_NOFS);
7614                 ret = VM_FAULT_SIGBUS;
7615                 goto out_unlock;
7616         }
7617         ret = 0;
7618
7619         /* page is wholly or partially inside EOF */
7620         if (page_start + PAGE_CACHE_SIZE > size)
7621                 zero_start = size & ~PAGE_CACHE_MASK;
7622         else
7623                 zero_start = PAGE_CACHE_SIZE;
7624
7625         if (zero_start != PAGE_CACHE_SIZE) {
7626                 kaddr = kmap(page);
7627                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
7628                 flush_dcache_page(page);
7629                 kunmap(page);
7630         }
7631         ClearPageChecked(page);
7632         set_page_dirty(page);
7633         SetPageUptodate(page);
7634
7635         BTRFS_I(inode)->last_trans = root->fs_info->generation;
7636         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
7637         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
7638
7639         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
7640
7641 out_unlock:
7642         if (!ret) {
7643                 sb_end_pagefault(inode->i_sb);
7644                 return VM_FAULT_LOCKED;
7645         }
7646         unlock_page(page);
7647 out:
7648         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
7649 out_noreserve:
7650         sb_end_pagefault(inode->i_sb);
7651         return ret;
7652 }
7653
7654 static int btrfs_truncate(struct inode *inode)
7655 {
7656         struct btrfs_root *root = BTRFS_I(inode)->root;
7657         struct btrfs_block_rsv *rsv;
7658         int ret = 0;
7659         int err = 0;
7660         struct btrfs_trans_handle *trans;
7661         u64 mask = root->sectorsize - 1;
7662         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
7663
7664         ret = btrfs_wait_ordered_range(inode, inode->i_size & (~mask),
7665                                        (u64)-1);
7666         if (ret)
7667                 return ret;
7668
7669         /*
7670          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
7671          * 3 things going on here
7672          *
7673          * 1) We need to reserve space for our orphan item and the space to
7674          * delete our orphan item.  Lord knows we don't want to have a dangling
7675          * orphan item because we didn't reserve space to remove it.
7676          *
7677          * 2) We need to reserve space to update our inode.
7678          *
7679          * 3) We need to have something to cache all the space that is going to
7680          * be free'd up by the truncate operation, but also have some slack
7681          * space reserved in case it uses space during the truncate (thank you
7682          * very much snapshotting).
7683          *
7684          * And we need these to all be seperate.  The fact is we can use alot of
7685          * space doing the truncate, and we have no earthly idea how much space
7686          * we will use, so we need the truncate reservation to be seperate so it
7687          * doesn't end up using space reserved for updating the inode or
7688          * removing the orphan item.  We also need to be able to stop the
7689          * transaction and start a new one, which means we need to be able to
7690          * update the inode several times, and we have no idea of knowing how
7691          * many times that will be, so we can't just reserve 1 item for the
7692          * entirety of the opration, so that has to be done seperately as well.
7693          * Then there is the orphan item, which does indeed need to be held on
7694          * to for the whole operation, and we need nobody to touch this reserved
7695          * space except the orphan code.
7696          *
7697          * So that leaves us with
7698          *
7699          * 1) root->orphan_block_rsv - for the orphan deletion.
7700          * 2) rsv - for the truncate reservation, which we will steal from the
7701          * transaction reservation.
7702          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7703          * updating the inode.
7704          */
7705         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7706         if (!rsv)
7707                 return -ENOMEM;
7708         rsv->size = min_size;
7709         rsv->failfast = 1;
7710
7711         /*
7712          * 1 for the truncate slack space
7713          * 1 for updating the inode.
7714          */
7715         trans = btrfs_start_transaction(root, 2);
7716         if (IS_ERR(trans)) {
7717                 err = PTR_ERR(trans);
7718                 goto out;
7719         }
7720
7721         /* Migrate the slack space for the truncate to our reserve */
7722         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7723                                       min_size);
7724         BUG_ON(ret);
7725
7726         /*
7727          * setattr is responsible for setting the ordered_data_close flag,
7728          * but that is only tested during the last file release.  That
7729          * could happen well after the next commit, leaving a great big
7730          * window where new writes may get lost if someone chooses to write
7731          * to this file after truncating to zero
7732          *
7733          * The inode doesn't have any dirty data here, and so if we commit
7734          * this is a noop.  If someone immediately starts writing to the inode
7735          * it is very likely we'll catch some of their writes in this
7736          * transaction, and the commit will find this file on the ordered
7737          * data list with good things to send down.
7738          *
7739          * This is a best effort solution, there is still a window where
7740          * using truncate to replace the contents of the file will
7741          * end up with a zero length file after a crash.
7742          */
7743         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7744                                            &BTRFS_I(inode)->runtime_flags))
7745                 btrfs_add_ordered_operation(trans, root, inode);
7746
7747         /*
7748          * So if we truncate and then write and fsync we normally would just
7749          * write the extents that changed, which is a problem if we need to
7750          * first truncate that entire inode.  So set this flag so we write out
7751          * all of the extents in the inode to the sync log so we're completely
7752          * safe.
7753          */
7754         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7755         trans->block_rsv = rsv;
7756
7757         while (1) {
7758                 ret = btrfs_truncate_inode_items(trans, root, inode,
7759                                                  inode->i_size,
7760                                                  BTRFS_EXTENT_DATA_KEY);
7761                 if (ret != -ENOSPC) {
7762                         err = ret;
7763                         break;
7764                 }
7765
7766                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7767                 ret = btrfs_update_inode(trans, root, inode);
7768                 if (ret) {
7769                         err = ret;
7770                         break;
7771                 }
7772
7773                 btrfs_end_transaction(trans, root);
7774                 btrfs_btree_balance_dirty(root);
7775
7776                 trans = btrfs_start_transaction(root, 2);
7777                 if (IS_ERR(trans)) {
7778                         ret = err = PTR_ERR(trans);
7779                         trans = NULL;
7780                         break;
7781                 }
7782
7783                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7784                                               rsv, min_size);
7785                 BUG_ON(ret);    /* shouldn't happen */
7786                 trans->block_rsv = rsv;
7787         }
7788
7789         if (ret == 0 && inode->i_nlink > 0) {
7790                 trans->block_rsv = root->orphan_block_rsv;
7791                 ret = btrfs_orphan_del(trans, inode);
7792                 if (ret)
7793                         err = ret;
7794         }
7795
7796         if (trans) {
7797                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7798                 ret = btrfs_update_inode(trans, root, inode);
7799                 if (ret && !err)
7800                         err = ret;
7801
7802                 ret = btrfs_end_transaction(trans, root);
7803                 btrfs_btree_balance_dirty(root);
7804         }
7805
7806 out:
7807         btrfs_free_block_rsv(root, rsv);
7808
7809         if (ret && !err)
7810                 err = ret;
7811
7812         return err;
7813 }
7814
7815 /*
7816  * create a new subvolume directory/inode (helper for the ioctl).
7817  */
7818 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7819                              struct btrfs_root *new_root, u64 new_dirid)
7820 {
7821         struct inode *inode;
7822         int err;
7823         u64 index = 0;
7824
7825         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7826                                 new_dirid, new_dirid,
7827                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7828                                 &index);
7829         if (IS_ERR(inode))
7830                 return PTR_ERR(inode);
7831         inode->i_op = &btrfs_dir_inode_operations;
7832         inode->i_fop = &btrfs_dir_file_operations;
7833
7834         set_nlink(inode, 1);
7835         btrfs_i_size_write(inode, 0);
7836
7837         err = btrfs_update_inode(trans, new_root, inode);
7838
7839         iput(inode);
7840         return err;
7841 }
7842
7843 struct inode *btrfs_alloc_inode(struct super_block *sb)
7844 {
7845         struct btrfs_inode *ei;
7846         struct inode *inode;
7847
7848         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7849         if (!ei)
7850                 return NULL;
7851
7852         ei->root = NULL;
7853         ei->generation = 0;
7854         ei->last_trans = 0;
7855         ei->last_sub_trans = 0;
7856         ei->logged_trans = 0;
7857         ei->delalloc_bytes = 0;
7858         ei->disk_i_size = 0;
7859         ei->flags = 0;
7860         ei->csum_bytes = 0;
7861         ei->index_cnt = (u64)-1;
7862         ei->last_unlink_trans = 0;
7863         ei->last_log_commit = 0;
7864
7865         spin_lock_init(&ei->lock);
7866         ei->outstanding_extents = 0;
7867         ei->reserved_extents = 0;
7868
7869         ei->runtime_flags = 0;
7870         ei->force_compress = BTRFS_COMPRESS_NONE;
7871
7872         ei->delayed_node = NULL;
7873
7874         inode = &ei->vfs_inode;
7875         extent_map_tree_init(&ei->extent_tree);
7876         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7877         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7878         ei->io_tree.track_uptodate = 1;
7879         ei->io_failure_tree.track_uptodate = 1;
7880         atomic_set(&ei->sync_writers, 0);
7881         mutex_init(&ei->log_mutex);
7882         mutex_init(&ei->delalloc_mutex);
7883         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7884         INIT_LIST_HEAD(&ei->delalloc_inodes);
7885         INIT_LIST_HEAD(&ei->ordered_operations);
7886         RB_CLEAR_NODE(&ei->rb_node);
7887
7888         return inode;
7889 }
7890
7891 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
7892 void btrfs_test_destroy_inode(struct inode *inode)
7893 {
7894         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7895         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7896 }
7897 #endif
7898
7899 static void btrfs_i_callback(struct rcu_head *head)
7900 {
7901         struct inode *inode = container_of(head, struct inode, i_rcu);
7902         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7903 }
7904
7905 void btrfs_destroy_inode(struct inode *inode)
7906 {
7907         struct btrfs_ordered_extent *ordered;
7908         struct btrfs_root *root = BTRFS_I(inode)->root;
7909
7910         WARN_ON(!hlist_empty(&inode->i_dentry));
7911         WARN_ON(inode->i_data.nrpages);
7912         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7913         WARN_ON(BTRFS_I(inode)->reserved_extents);
7914         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7915         WARN_ON(BTRFS_I(inode)->csum_bytes);
7916
7917         /*
7918          * This can happen where we create an inode, but somebody else also
7919          * created the same inode and we need to destroy the one we already
7920          * created.
7921          */
7922         if (!root)
7923                 goto free;
7924
7925         /*
7926          * Make sure we're properly removed from the ordered operation
7927          * lists.
7928          */
7929         smp_mb();
7930         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7931                 spin_lock(&root->fs_info->ordered_root_lock);
7932                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7933                 spin_unlock(&root->fs_info->ordered_root_lock);
7934         }
7935
7936         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7937                      &BTRFS_I(inode)->runtime_flags)) {
7938                 btrfs_info(root->fs_info, "inode %llu still on the orphan list",
7939                         btrfs_ino(inode));
7940                 atomic_dec(&root->orphan_inodes);
7941         }
7942
7943         while (1) {
7944                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7945                 if (!ordered)
7946                         break;
7947                 else {
7948                         btrfs_err(root->fs_info, "found ordered extent %llu %llu on inode cleanup",
7949                                 ordered->file_offset, ordered->len);
7950                         btrfs_remove_ordered_extent(inode, ordered);
7951                         btrfs_put_ordered_extent(ordered);
7952                         btrfs_put_ordered_extent(ordered);
7953                 }
7954         }
7955         inode_tree_del(inode);
7956         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7957 free:
7958         call_rcu(&inode->i_rcu, btrfs_i_callback);
7959 }
7960
7961 int btrfs_drop_inode(struct inode *inode)
7962 {
7963         struct btrfs_root *root = BTRFS_I(inode)->root;
7964
7965         if (root == NULL)
7966                 return 1;
7967
7968         /* the snap/subvol tree is on deleting */
7969         if (btrfs_root_refs(&root->root_item) == 0)
7970                 return 1;
7971         else
7972                 return generic_drop_inode(inode);
7973 }
7974
7975 static void init_once(void *foo)
7976 {
7977         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7978
7979         inode_init_once(&ei->vfs_inode);
7980 }
7981
7982 void btrfs_destroy_cachep(void)
7983 {
7984         /*
7985          * Make sure all delayed rcu free inodes are flushed before we
7986          * destroy cache.
7987          */
7988         rcu_barrier();
7989         if (btrfs_inode_cachep)
7990                 kmem_cache_destroy(btrfs_inode_cachep);
7991         if (btrfs_trans_handle_cachep)
7992                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7993         if (btrfs_transaction_cachep)
7994                 kmem_cache_destroy(btrfs_transaction_cachep);
7995         if (btrfs_path_cachep)
7996                 kmem_cache_destroy(btrfs_path_cachep);
7997         if (btrfs_free_space_cachep)
7998                 kmem_cache_destroy(btrfs_free_space_cachep);
7999         if (btrfs_delalloc_work_cachep)
8000                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
8001 }
8002
8003 int btrfs_init_cachep(void)
8004 {
8005         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
8006                         sizeof(struct btrfs_inode), 0,
8007                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
8008         if (!btrfs_inode_cachep)
8009                 goto fail;
8010
8011         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
8012                         sizeof(struct btrfs_trans_handle), 0,
8013                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8014         if (!btrfs_trans_handle_cachep)
8015                 goto fail;
8016
8017         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
8018                         sizeof(struct btrfs_transaction), 0,
8019                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8020         if (!btrfs_transaction_cachep)
8021                 goto fail;
8022
8023         btrfs_path_cachep = kmem_cache_create("btrfs_path",
8024                         sizeof(struct btrfs_path), 0,
8025                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8026         if (!btrfs_path_cachep)
8027                 goto fail;
8028
8029         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
8030                         sizeof(struct btrfs_free_space), 0,
8031                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
8032         if (!btrfs_free_space_cachep)
8033                 goto fail;
8034
8035         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
8036                         sizeof(struct btrfs_delalloc_work), 0,
8037                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
8038                         NULL);
8039         if (!btrfs_delalloc_work_cachep)
8040                 goto fail;
8041
8042         return 0;
8043 fail:
8044         btrfs_destroy_cachep();
8045         return -ENOMEM;
8046 }
8047
8048 static int btrfs_getattr(struct vfsmount *mnt,
8049                          struct dentry *dentry, struct kstat *stat)
8050 {
8051         u64 delalloc_bytes;
8052         struct inode *inode = dentry->d_inode;
8053         u32 blocksize = inode->i_sb->s_blocksize;
8054
8055         generic_fillattr(inode, stat);
8056         stat->dev = BTRFS_I(inode)->root->anon_dev;
8057         stat->blksize = PAGE_CACHE_SIZE;
8058
8059         spin_lock(&BTRFS_I(inode)->lock);
8060         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
8061         spin_unlock(&BTRFS_I(inode)->lock);
8062         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
8063                         ALIGN(delalloc_bytes, blocksize)) >> 9;
8064         return 0;
8065 }
8066
8067 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
8068                            struct inode *new_dir, struct dentry *new_dentry)
8069 {
8070         struct btrfs_trans_handle *trans;
8071         struct btrfs_root *root = BTRFS_I(old_dir)->root;
8072         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
8073         struct inode *new_inode = new_dentry->d_inode;
8074         struct inode *old_inode = old_dentry->d_inode;
8075         struct timespec ctime = CURRENT_TIME;
8076         u64 index = 0;
8077         u64 root_objectid;
8078         int ret;
8079         u64 old_ino = btrfs_ino(old_inode);
8080
8081         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
8082                 return -EPERM;
8083
8084         /* we only allow rename subvolume link between subvolumes */
8085         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
8086                 return -EXDEV;
8087
8088         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
8089             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
8090                 return -ENOTEMPTY;
8091
8092         if (S_ISDIR(old_inode->i_mode) && new_inode &&
8093             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
8094                 return -ENOTEMPTY;
8095
8096
8097         /* check for collisions, even if the  name isn't there */
8098         ret = btrfs_check_dir_item_collision(dest, new_dir->i_ino,
8099                              new_dentry->d_name.name,
8100                              new_dentry->d_name.len);
8101
8102         if (ret) {
8103                 if (ret == -EEXIST) {
8104                         /* we shouldn't get
8105                          * eexist without a new_inode */
8106                         if (WARN_ON(!new_inode)) {
8107                                 return ret;
8108                         }
8109                 } else {
8110                         /* maybe -EOVERFLOW */
8111                         return ret;
8112                 }
8113         }
8114         ret = 0;
8115
8116         /*
8117          * we're using rename to replace one file with another.
8118          * and the replacement file is large.  Start IO on it now so
8119          * we don't add too much work to the end of the transaction
8120          */
8121         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
8122             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
8123                 filemap_flush(old_inode->i_mapping);
8124
8125         /* close the racy window with snapshot create/destroy ioctl */
8126         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8127                 down_read(&root->fs_info->subvol_sem);
8128         /*
8129          * We want to reserve the absolute worst case amount of items.  So if
8130          * both inodes are subvols and we need to unlink them then that would
8131          * require 4 item modifications, but if they are both normal inodes it
8132          * would require 5 item modifications, so we'll assume their normal
8133          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
8134          * should cover the worst case number of items we'll modify.
8135          */
8136         trans = btrfs_start_transaction(root, 11);
8137         if (IS_ERR(trans)) {
8138                 ret = PTR_ERR(trans);
8139                 goto out_notrans;
8140         }
8141
8142         if (dest != root)
8143                 btrfs_record_root_in_trans(trans, dest);
8144
8145         ret = btrfs_set_inode_index(new_dir, &index);
8146         if (ret)
8147                 goto out_fail;
8148
8149         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8150                 /* force full log commit if subvolume involved. */
8151                 root->fs_info->last_trans_log_full_commit = trans->transid;
8152         } else {
8153                 ret = btrfs_insert_inode_ref(trans, dest,
8154                                              new_dentry->d_name.name,
8155                                              new_dentry->d_name.len,
8156                                              old_ino,
8157                                              btrfs_ino(new_dir), index);
8158                 if (ret)
8159                         goto out_fail;
8160                 /*
8161                  * this is an ugly little race, but the rename is required
8162                  * to make sure that if we crash, the inode is either at the
8163                  * old name or the new one.  pinning the log transaction lets
8164                  * us make sure we don't allow a log commit to come in after
8165                  * we unlink the name but before we add the new name back in.
8166                  */
8167                 btrfs_pin_log_trans(root);
8168         }
8169         /*
8170          * make sure the inode gets flushed if it is replacing
8171          * something.
8172          */
8173         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
8174                 btrfs_add_ordered_operation(trans, root, old_inode);
8175
8176         inode_inc_iversion(old_dir);
8177         inode_inc_iversion(new_dir);
8178         inode_inc_iversion(old_inode);
8179         old_dir->i_ctime = old_dir->i_mtime = ctime;
8180         new_dir->i_ctime = new_dir->i_mtime = ctime;
8181         old_inode->i_ctime = ctime;
8182
8183         if (old_dentry->d_parent != new_dentry->d_parent)
8184                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
8185
8186         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
8187                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
8188                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
8189                                         old_dentry->d_name.name,
8190                                         old_dentry->d_name.len);
8191         } else {
8192                 ret = __btrfs_unlink_inode(trans, root, old_dir,
8193                                         old_dentry->d_inode,
8194                                         old_dentry->d_name.name,
8195                                         old_dentry->d_name.len);
8196                 if (!ret)
8197                         ret = btrfs_update_inode(trans, root, old_inode);
8198         }
8199         if (ret) {
8200                 btrfs_abort_transaction(trans, root, ret);
8201                 goto out_fail;
8202         }
8203
8204         if (new_inode) {
8205                 inode_inc_iversion(new_inode);
8206                 new_inode->i_ctime = CURRENT_TIME;
8207                 if (unlikely(btrfs_ino(new_inode) ==
8208                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
8209                         root_objectid = BTRFS_I(new_inode)->location.objectid;
8210                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
8211                                                 root_objectid,
8212                                                 new_dentry->d_name.name,
8213                                                 new_dentry->d_name.len);
8214                         BUG_ON(new_inode->i_nlink == 0);
8215                 } else {
8216                         ret = btrfs_unlink_inode(trans, dest, new_dir,
8217                                                  new_dentry->d_inode,
8218                                                  new_dentry->d_name.name,
8219                                                  new_dentry->d_name.len);
8220                 }
8221                 if (!ret && new_inode->i_nlink == 0)
8222                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
8223                 if (ret) {
8224                         btrfs_abort_transaction(trans, root, ret);
8225                         goto out_fail;
8226                 }
8227         }
8228
8229         ret = btrfs_add_link(trans, new_dir, old_inode,
8230                              new_dentry->d_name.name,
8231                              new_dentry->d_name.len, 0, index);
8232         if (ret) {
8233                 btrfs_abort_transaction(trans, root, ret);
8234                 goto out_fail;
8235         }
8236
8237         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
8238                 struct dentry *parent = new_dentry->d_parent;
8239                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
8240                 btrfs_end_log_trans(root);
8241         }
8242 out_fail:
8243         btrfs_end_transaction(trans, root);
8244 out_notrans:
8245         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
8246                 up_read(&root->fs_info->subvol_sem);
8247
8248         return ret;
8249 }
8250
8251 static void btrfs_run_delalloc_work(struct btrfs_work *work)
8252 {
8253         struct btrfs_delalloc_work *delalloc_work;
8254         struct inode *inode;
8255
8256         delalloc_work = container_of(work, struct btrfs_delalloc_work,
8257                                      work);
8258         inode = delalloc_work->inode;
8259         if (delalloc_work->wait) {
8260                 btrfs_wait_ordered_range(inode, 0, (u64)-1);
8261         } else {
8262                 filemap_flush(inode->i_mapping);
8263                 if (test_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
8264                              &BTRFS_I(inode)->runtime_flags))
8265                         filemap_flush(inode->i_mapping);
8266         }
8267
8268         if (delalloc_work->delay_iput)
8269                 btrfs_add_delayed_iput(inode);
8270         else
8271                 iput(inode);
8272         complete(&delalloc_work->completion);
8273 }
8274
8275 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
8276                                                     int wait, int delay_iput)
8277 {
8278         struct btrfs_delalloc_work *work;
8279
8280         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
8281         if (!work)
8282                 return NULL;
8283
8284         init_completion(&work->completion);
8285         INIT_LIST_HEAD(&work->list);
8286         work->inode = inode;
8287         work->wait = wait;
8288         work->delay_iput = delay_iput;
8289         work->work.func = btrfs_run_delalloc_work;
8290
8291         return work;
8292 }
8293
8294 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
8295 {
8296         wait_for_completion(&work->completion);
8297         kmem_cache_free(btrfs_delalloc_work_cachep, work);
8298 }
8299
8300 /*
8301  * some fairly slow code that needs optimization. This walks the list
8302  * of all the inodes with pending delalloc and forces them to disk.
8303  */
8304 static int __start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8305 {
8306         struct btrfs_inode *binode;
8307         struct inode *inode;
8308         struct btrfs_delalloc_work *work, *next;
8309         struct list_head works;
8310         struct list_head splice;
8311         int ret = 0;
8312
8313         INIT_LIST_HEAD(&works);
8314         INIT_LIST_HEAD(&splice);
8315
8316         spin_lock(&root->delalloc_lock);
8317         list_splice_init(&root->delalloc_inodes, &splice);
8318         while (!list_empty(&splice)) {
8319                 binode = list_entry(splice.next, struct btrfs_inode,
8320                                     delalloc_inodes);
8321
8322                 list_move_tail(&binode->delalloc_inodes,
8323                                &root->delalloc_inodes);
8324                 inode = igrab(&binode->vfs_inode);
8325                 if (!inode) {
8326                         cond_resched_lock(&root->delalloc_lock);
8327                         continue;
8328                 }
8329                 spin_unlock(&root->delalloc_lock);
8330
8331                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
8332                 if (unlikely(!work)) {
8333                         if (delay_iput)
8334                                 btrfs_add_delayed_iput(inode);
8335                         else
8336                                 iput(inode);
8337                         ret = -ENOMEM;
8338                         goto out;
8339                 }
8340                 list_add_tail(&work->list, &works);
8341                 btrfs_queue_worker(&root->fs_info->flush_workers,
8342                                    &work->work);
8343
8344                 cond_resched();
8345                 spin_lock(&root->delalloc_lock);
8346         }
8347         spin_unlock(&root->delalloc_lock);
8348
8349         list_for_each_entry_safe(work, next, &works, list) {
8350                 list_del_init(&work->list);
8351                 btrfs_wait_and_free_delalloc_work(work);
8352         }
8353         return 0;
8354 out:
8355         list_for_each_entry_safe(work, next, &works, list) {
8356                 list_del_init(&work->list);
8357                 btrfs_wait_and_free_delalloc_work(work);
8358         }
8359
8360         if (!list_empty_careful(&splice)) {
8361                 spin_lock(&root->delalloc_lock);
8362                 list_splice_tail(&splice, &root->delalloc_inodes);
8363                 spin_unlock(&root->delalloc_lock);
8364         }
8365         return ret;
8366 }
8367
8368 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
8369 {
8370         int ret;
8371
8372         if (root->fs_info->sb->s_flags & MS_RDONLY)
8373                 return -EROFS;
8374
8375         ret = __start_delalloc_inodes(root, delay_iput);
8376         /*
8377          * the filemap_flush will queue IO into the worker threads, but
8378          * we have to make sure the IO is actually started and that
8379          * ordered extents get created before we return
8380          */
8381         atomic_inc(&root->fs_info->async_submit_draining);
8382         while (atomic_read(&root->fs_info->nr_async_submits) ||
8383               atomic_read(&root->fs_info->async_delalloc_pages)) {
8384                 wait_event(root->fs_info->async_submit_wait,
8385                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
8386                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
8387         }
8388         atomic_dec(&root->fs_info->async_submit_draining);
8389         return ret;
8390 }
8391
8392 int btrfs_start_delalloc_roots(struct btrfs_fs_info *fs_info, int delay_iput)
8393 {
8394         struct btrfs_root *root;
8395         struct list_head splice;
8396         int ret;
8397
8398         if (fs_info->sb->s_flags & MS_RDONLY)
8399                 return -EROFS;
8400
8401         INIT_LIST_HEAD(&splice);
8402
8403         spin_lock(&fs_info->delalloc_root_lock);
8404         list_splice_init(&fs_info->delalloc_roots, &splice);
8405         while (!list_empty(&splice)) {
8406                 root = list_first_entry(&splice, struct btrfs_root,
8407                                         delalloc_root);
8408                 root = btrfs_grab_fs_root(root);
8409                 BUG_ON(!root);
8410                 list_move_tail(&root->delalloc_root,
8411                                &fs_info->delalloc_roots);
8412                 spin_unlock(&fs_info->delalloc_root_lock);
8413
8414                 ret = __start_delalloc_inodes(root, delay_iput);
8415                 btrfs_put_fs_root(root);
8416                 if (ret)
8417                         goto out;
8418
8419                 spin_lock(&fs_info->delalloc_root_lock);
8420         }
8421         spin_unlock(&fs_info->delalloc_root_lock);
8422
8423         atomic_inc(&fs_info->async_submit_draining);
8424         while (atomic_read(&fs_info->nr_async_submits) ||
8425               atomic_read(&fs_info->async_delalloc_pages)) {
8426                 wait_event(fs_info->async_submit_wait,
8427                    (atomic_read(&fs_info->nr_async_submits) == 0 &&
8428                     atomic_read(&fs_info->async_delalloc_pages) == 0));
8429         }
8430         atomic_dec(&fs_info->async_submit_draining);
8431         return 0;
8432 out:
8433         if (!list_empty_careful(&splice)) {
8434                 spin_lock(&fs_info->delalloc_root_lock);
8435                 list_splice_tail(&splice, &fs_info->delalloc_roots);
8436                 spin_unlock(&fs_info->delalloc_root_lock);
8437         }
8438         return ret;
8439 }
8440
8441 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
8442                          const char *symname)
8443 {
8444         struct btrfs_trans_handle *trans;
8445         struct btrfs_root *root = BTRFS_I(dir)->root;
8446         struct btrfs_path *path;
8447         struct btrfs_key key;
8448         struct inode *inode = NULL;
8449         int err;
8450         int drop_inode = 0;
8451         u64 objectid;
8452         u64 index = 0;
8453         int name_len;
8454         int datasize;
8455         unsigned long ptr;
8456         struct btrfs_file_extent_item *ei;
8457         struct extent_buffer *leaf;
8458
8459         name_len = strlen(symname);
8460         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
8461                 return -ENAMETOOLONG;
8462
8463         /*
8464          * 2 items for inode item and ref
8465          * 2 items for dir items
8466          * 1 item for xattr if selinux is on
8467          */
8468         trans = btrfs_start_transaction(root, 5);
8469         if (IS_ERR(trans))
8470                 return PTR_ERR(trans);
8471
8472         err = btrfs_find_free_ino(root, &objectid);
8473         if (err)
8474                 goto out_unlock;
8475
8476         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
8477                                 dentry->d_name.len, btrfs_ino(dir), objectid,
8478                                 S_IFLNK|S_IRWXUGO, &index);
8479         if (IS_ERR(inode)) {
8480                 err = PTR_ERR(inode);
8481                 goto out_unlock;
8482         }
8483
8484         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
8485         if (err) {
8486                 drop_inode = 1;
8487                 goto out_unlock;
8488         }
8489
8490         /*
8491         * If the active LSM wants to access the inode during
8492         * d_instantiate it needs these. Smack checks to see
8493         * if the filesystem supports xattrs by looking at the
8494         * ops vector.
8495         */
8496         inode->i_fop = &btrfs_file_operations;
8497         inode->i_op = &btrfs_file_inode_operations;
8498
8499         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
8500         if (err)
8501                 drop_inode = 1;
8502         else {
8503                 inode->i_mapping->a_ops = &btrfs_aops;
8504                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8505                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
8506         }
8507         if (drop_inode)
8508                 goto out_unlock;
8509
8510         path = btrfs_alloc_path();
8511         if (!path) {
8512                 err = -ENOMEM;
8513                 drop_inode = 1;
8514                 goto out_unlock;
8515         }
8516         key.objectid = btrfs_ino(inode);
8517         key.offset = 0;
8518         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
8519         datasize = btrfs_file_extent_calc_inline_size(name_len);
8520         err = btrfs_insert_empty_item(trans, root, path, &key,
8521                                       datasize);
8522         if (err) {
8523                 drop_inode = 1;
8524                 btrfs_free_path(path);
8525                 goto out_unlock;
8526         }
8527         leaf = path->nodes[0];
8528         ei = btrfs_item_ptr(leaf, path->slots[0],
8529                             struct btrfs_file_extent_item);
8530         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
8531         btrfs_set_file_extent_type(leaf, ei,
8532                                    BTRFS_FILE_EXTENT_INLINE);
8533         btrfs_set_file_extent_encryption(leaf, ei, 0);
8534         btrfs_set_file_extent_compression(leaf, ei, 0);
8535         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
8536         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
8537
8538         ptr = btrfs_file_extent_inline_start(ei);
8539         write_extent_buffer(leaf, symname, ptr, name_len);
8540         btrfs_mark_buffer_dirty(leaf);
8541         btrfs_free_path(path);
8542
8543         inode->i_op = &btrfs_symlink_inode_operations;
8544         inode->i_mapping->a_ops = &btrfs_symlink_aops;
8545         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
8546         inode_set_bytes(inode, name_len);
8547         btrfs_i_size_write(inode, name_len);
8548         err = btrfs_update_inode(trans, root, inode);
8549         if (err)
8550                 drop_inode = 1;
8551
8552 out_unlock:
8553         if (!err)
8554                 d_instantiate(dentry, inode);
8555         btrfs_end_transaction(trans, root);
8556         if (drop_inode) {
8557                 inode_dec_link_count(inode);
8558                 iput(inode);
8559         }
8560         btrfs_btree_balance_dirty(root);
8561         return err;
8562 }
8563
8564 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
8565                                        u64 start, u64 num_bytes, u64 min_size,
8566                                        loff_t actual_len, u64 *alloc_hint,
8567                                        struct btrfs_trans_handle *trans)
8568 {
8569         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
8570         struct extent_map *em;
8571         struct btrfs_root *root = BTRFS_I(inode)->root;
8572         struct btrfs_key ins;
8573         u64 cur_offset = start;
8574         u64 i_size;
8575         u64 cur_bytes;
8576         int ret = 0;
8577         bool own_trans = true;
8578
8579         if (trans)
8580                 own_trans = false;
8581         while (num_bytes > 0) {
8582                 if (own_trans) {
8583                         trans = btrfs_start_transaction(root, 3);
8584                         if (IS_ERR(trans)) {
8585                                 ret = PTR_ERR(trans);
8586                                 break;
8587                         }
8588                 }
8589
8590                 cur_bytes = min(num_bytes, 256ULL * 1024 * 1024);
8591                 cur_bytes = max(cur_bytes, min_size);
8592                 ret = btrfs_reserve_extent(root, cur_bytes, min_size, 0,
8593                                            *alloc_hint, &ins, 1);
8594                 if (ret) {
8595                         if (own_trans)
8596                                 btrfs_end_transaction(trans, root);
8597                         break;
8598                 }
8599
8600                 ret = insert_reserved_file_extent(trans, inode,
8601                                                   cur_offset, ins.objectid,
8602                                                   ins.offset, ins.offset,
8603                                                   ins.offset, 0, 0, 0,
8604                                                   BTRFS_FILE_EXTENT_PREALLOC);
8605                 if (ret) {
8606                         btrfs_free_reserved_extent(root, ins.objectid,
8607                                                    ins.offset);
8608                         btrfs_abort_transaction(trans, root, ret);
8609                         if (own_trans)
8610                                 btrfs_end_transaction(trans, root);
8611                         break;
8612                 }
8613                 btrfs_drop_extent_cache(inode, cur_offset,
8614                                         cur_offset + ins.offset -1, 0);
8615
8616                 em = alloc_extent_map();
8617                 if (!em) {
8618                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
8619                                 &BTRFS_I(inode)->runtime_flags);
8620                         goto next;
8621                 }
8622
8623                 em->start = cur_offset;
8624                 em->orig_start = cur_offset;
8625                 em->len = ins.offset;
8626                 em->block_start = ins.objectid;
8627                 em->block_len = ins.offset;
8628                 em->orig_block_len = ins.offset;
8629                 em->ram_bytes = ins.offset;
8630                 em->bdev = root->fs_info->fs_devices->latest_bdev;
8631                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
8632                 em->generation = trans->transid;
8633
8634                 while (1) {
8635                         write_lock(&em_tree->lock);
8636                         ret = add_extent_mapping(em_tree, em, 1);
8637                         write_unlock(&em_tree->lock);
8638                         if (ret != -EEXIST)
8639                                 break;
8640                         btrfs_drop_extent_cache(inode, cur_offset,
8641                                                 cur_offset + ins.offset - 1,
8642                                                 0);
8643                 }
8644                 free_extent_map(em);
8645 next:
8646                 num_bytes -= ins.offset;
8647                 cur_offset += ins.offset;
8648                 *alloc_hint = ins.objectid + ins.offset;
8649
8650                 inode_inc_iversion(inode);
8651                 inode->i_ctime = CURRENT_TIME;
8652                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
8653                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
8654                     (actual_len > inode->i_size) &&
8655                     (cur_offset > inode->i_size)) {
8656                         if (cur_offset > actual_len)
8657                                 i_size = actual_len;
8658                         else
8659                                 i_size = cur_offset;
8660                         i_size_write(inode, i_size);
8661                         btrfs_ordered_update_i_size(inode, i_size, NULL);
8662                 }
8663
8664                 ret = btrfs_update_inode(trans, root, inode);
8665
8666                 if (ret) {
8667                         btrfs_abort_transaction(trans, root, ret);
8668                         if (own_trans)
8669                                 btrfs_end_transaction(trans, root);
8670                         break;
8671                 }
8672
8673                 if (own_trans)
8674                         btrfs_end_transaction(trans, root);
8675         }
8676         return ret;
8677 }
8678
8679 int btrfs_prealloc_file_range(struct inode *inode, int mode,
8680                               u64 start, u64 num_bytes, u64 min_size,
8681                               loff_t actual_len, u64 *alloc_hint)
8682 {
8683         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8684                                            min_size, actual_len, alloc_hint,
8685                                            NULL);
8686 }
8687
8688 int btrfs_prealloc_file_range_trans(struct inode *inode,
8689                                     struct btrfs_trans_handle *trans, int mode,
8690                                     u64 start, u64 num_bytes, u64 min_size,
8691                                     loff_t actual_len, u64 *alloc_hint)
8692 {
8693         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
8694                                            min_size, actual_len, alloc_hint, trans);
8695 }
8696
8697 static int btrfs_set_page_dirty(struct page *page)
8698 {
8699         return __set_page_dirty_nobuffers(page);
8700 }
8701
8702 static int btrfs_permission(struct inode *inode, int mask)
8703 {
8704         struct btrfs_root *root = BTRFS_I(inode)->root;
8705         umode_t mode = inode->i_mode;
8706
8707         if (mask & MAY_WRITE &&
8708             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
8709                 if (btrfs_root_readonly(root))
8710                         return -EROFS;
8711                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
8712                         return -EACCES;
8713         }
8714         return generic_permission(inode, mask);
8715 }
8716
8717 static const struct inode_operations btrfs_dir_inode_operations = {
8718         .getattr        = btrfs_getattr,
8719         .lookup         = btrfs_lookup,
8720         .create         = btrfs_create,
8721         .unlink         = btrfs_unlink,
8722         .link           = btrfs_link,
8723         .mkdir          = btrfs_mkdir,
8724         .rmdir          = btrfs_rmdir,
8725         .rename         = btrfs_rename,
8726         .symlink        = btrfs_symlink,
8727         .setattr        = btrfs_setattr,
8728         .mknod          = btrfs_mknod,
8729         .setxattr       = btrfs_setxattr,
8730         .getxattr       = btrfs_getxattr,
8731         .listxattr      = btrfs_listxattr,
8732         .removexattr    = btrfs_removexattr,
8733         .permission     = btrfs_permission,
8734         .get_acl        = btrfs_get_acl,
8735         .update_time    = btrfs_update_time,
8736 };
8737 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8738         .lookup         = btrfs_lookup,
8739         .permission     = btrfs_permission,
8740         .get_acl        = btrfs_get_acl,
8741         .update_time    = btrfs_update_time,
8742 };
8743
8744 static const struct file_operations btrfs_dir_file_operations = {
8745         .llseek         = generic_file_llseek,
8746         .read           = generic_read_dir,
8747         .iterate        = btrfs_real_readdir,
8748         .unlocked_ioctl = btrfs_ioctl,
8749 #ifdef CONFIG_COMPAT
8750         .compat_ioctl   = btrfs_ioctl,
8751 #endif
8752         .release        = btrfs_release_file,
8753         .fsync          = btrfs_sync_file,
8754 };
8755
8756 static struct extent_io_ops btrfs_extent_io_ops = {
8757         .fill_delalloc = run_delalloc_range,
8758         .submit_bio_hook = btrfs_submit_bio_hook,
8759         .merge_bio_hook = btrfs_merge_bio_hook,
8760         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8761         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8762         .writepage_start_hook = btrfs_writepage_start_hook,
8763         .set_bit_hook = btrfs_set_bit_hook,
8764         .clear_bit_hook = btrfs_clear_bit_hook,
8765         .merge_extent_hook = btrfs_merge_extent_hook,
8766         .split_extent_hook = btrfs_split_extent_hook,
8767 };
8768
8769 /*
8770  * btrfs doesn't support the bmap operation because swapfiles
8771  * use bmap to make a mapping of extents in the file.  They assume
8772  * these extents won't change over the life of the file and they
8773  * use the bmap result to do IO directly to the drive.
8774  *
8775  * the btrfs bmap call would return logical addresses that aren't
8776  * suitable for IO and they also will change frequently as COW
8777  * operations happen.  So, swapfile + btrfs == corruption.
8778  *
8779  * For now we're avoiding this by dropping bmap.
8780  */
8781 static const struct address_space_operations btrfs_aops = {
8782         .readpage       = btrfs_readpage,
8783         .writepage      = btrfs_writepage,
8784         .writepages     = btrfs_writepages,
8785         .readpages      = btrfs_readpages,
8786         .direct_IO      = btrfs_direct_IO,
8787         .invalidatepage = btrfs_invalidatepage,
8788         .releasepage    = btrfs_releasepage,
8789         .set_page_dirty = btrfs_set_page_dirty,
8790         .error_remove_page = generic_error_remove_page,
8791 };
8792
8793 static const struct address_space_operations btrfs_symlink_aops = {
8794         .readpage       = btrfs_readpage,
8795         .writepage      = btrfs_writepage,
8796         .invalidatepage = btrfs_invalidatepage,
8797         .releasepage    = btrfs_releasepage,
8798 };
8799
8800 static const struct inode_operations btrfs_file_inode_operations = {
8801         .getattr        = btrfs_getattr,
8802         .setattr        = btrfs_setattr,
8803         .setxattr       = btrfs_setxattr,
8804         .getxattr       = btrfs_getxattr,
8805         .listxattr      = btrfs_listxattr,
8806         .removexattr    = btrfs_removexattr,
8807         .permission     = btrfs_permission,
8808         .fiemap         = btrfs_fiemap,
8809         .get_acl        = btrfs_get_acl,
8810         .update_time    = btrfs_update_time,
8811 };
8812 static const struct inode_operations btrfs_special_inode_operations = {
8813         .getattr        = btrfs_getattr,
8814         .setattr        = btrfs_setattr,
8815         .permission     = btrfs_permission,
8816         .setxattr       = btrfs_setxattr,
8817         .getxattr       = btrfs_getxattr,
8818         .listxattr      = btrfs_listxattr,
8819         .removexattr    = btrfs_removexattr,
8820         .get_acl        = btrfs_get_acl,
8821         .update_time    = btrfs_update_time,
8822 };
8823 static const struct inode_operations btrfs_symlink_inode_operations = {
8824         .readlink       = generic_readlink,
8825         .follow_link    = page_follow_link_light,
8826         .put_link       = page_put_link,
8827         .getattr        = btrfs_getattr,
8828         .setattr        = btrfs_setattr,
8829         .permission     = btrfs_permission,
8830         .setxattr       = btrfs_setxattr,
8831         .getxattr       = btrfs_getxattr,
8832         .listxattr      = btrfs_listxattr,
8833         .removexattr    = btrfs_removexattr,
8834         .get_acl        = btrfs_get_acl,
8835         .update_time    = btrfs_update_time,
8836 };
8837
8838 const struct dentry_operations btrfs_dentry_operations = {
8839         .d_delete       = btrfs_dentry_delete,
8840         .d_release      = btrfs_dentry_release,
8841 };