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