fs/btrfs/inode.c

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