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