fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         ret = btrfs_update_inode(trans, root, inode);
 209
 210         return ret;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 data_len = inline_len;
 234         int ret;
 235
 236         if (compressed_size)
 237                 data_len = compressed_size;
 238
 239         if (start > 0 ||
 240             actual_end >= PAGE_CACHE_SIZE ||
 241             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 242             (!compressed_size &&
 243             (actual_end & (root->sectorsize - 1)) == 0) ||
 244             end + 1 < isize ||
 245             data_len > root->fs_info->max_inline) {
 246                 return 1;
 247         }
 248
 249         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 250         if (ret)
 251                 return ret;
 252
 253         if (isize > actual_end)
 254                 inline_len = min_t(u64, isize, actual_end);
 255         ret = insert_inline_extent(trans, root, inode, start,
 256                                    inline_len, compressed_size,
 257                                    compress_type, compressed_pages);
 258         if (ret && ret != -ENOSPC) {
 259                 btrfs_abort_transaction(trans, root, ret);
 260                 return ret;
 261         } else if (ret == -ENOSPC) {
 262                 return 1;
 263         }
 264
 265         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 266         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 267         return 0;
 268 }
 269
 270 struct async_extent {
 271         u64 start;
 272         u64 ram_size;
 273         u64 compressed_size;
 274         struct page **pages;
 275         unsigned long nr_pages;
 276         int compress_type;
 277         struct list_head list;
 278 };
 279
 280 struct async_cow {
 281         struct inode *inode;
 282         struct btrfs_root *root;
 283         struct page *locked_page;
 284         u64 start;
 285         u64 end;
 286         struct list_head extents;
 287         struct btrfs_work work;
 288 };
 289
 290 static noinline int add_async_extent(struct async_cow *cow,
 291                                      u64 start, u64 ram_size,
 292                                      u64 compressed_size,
 293                                      struct page **pages,
 294                                      unsigned long nr_pages,
 295                                      int compress_type)
 296 {
 297         struct async_extent *async_extent;
 298
 299         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 300         BUG_ON(!async_extent); /* -ENOMEM */
 301         async_extent->start = start;
 302         async_extent->ram_size = ram_size;
 303         async_extent->compressed_size = compressed_size;
 304         async_extent->pages = pages;
 305         async_extent->nr_pages = nr_pages;
 306         async_extent->compress_type = compress_type;
 307         list_add_tail(&async_extent->list, &cow->extents);
 308         return 0;
 309 }
 310
 311 /*
 312  * we create compressed extents in two phases.  The first
 313  * phase compresses a range of pages that have already been
 314  * locked (both pages and state bits are locked).
 315  *
 316  * This is done inside an ordered work queue, and the compression
 317  * is spread across many cpus.  The actual IO submission is step
 318  * two, and the ordered work queue takes care of making sure that
 319  * happens in the same order things were put onto the queue by
 320  * writepages and friends.
 321  *
 322  * If this code finds it can't get good compression, it puts an
 323  * entry onto the work queue to write the uncompressed bytes.  This
 324  * makes sure that both compressed inodes and uncompressed inodes
 325  * are written in the same order that the flusher thread sent them
 326  * down.
 327  */
 328 static noinline int compress_file_range(struct inode *inode,
 329                                         struct page *locked_page,
 330                                         u64 start, u64 end,
 331                                         struct async_cow *async_cow,
 332                                         int *num_added)
 333 {
 334         struct btrfs_root *root = BTRFS_I(inode)->root;
 335         struct btrfs_trans_handle *trans;
 336         u64 num_bytes;
 337         u64 blocksize = root->sectorsize;
 338         u64 actual_end;
 339         u64 isize = i_size_read(inode);
 340         int ret = 0;
 341         struct page **pages = NULL;
 342         unsigned long nr_pages;
 343         unsigned long nr_pages_ret = 0;
 344         unsigned long total_compressed = 0;
 345         unsigned long total_in = 0;
 346         unsigned long max_compressed = 128 * 1024;
 347         unsigned long max_uncompressed = 128 * 1024;
 348         int i;
 349         int will_compress;
 350         int compress_type = root->fs_info->compress_type;
 351
 352         /* if this is a small write inside eof, kick off a defrag */
 353         if ((end - start + 1) < 16 * 1024 &&
 354             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 355                 btrfs_add_inode_defrag(NULL, inode);
 356
 357         actual_end = min_t(u64, isize, end + 1);
 358 again:
 359         will_compress = 0;
 360         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 361         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 362
 363         /*
 364          * we don't want to send crud past the end of i_size through
 365          * compression, that's just a waste of CPU time.  So, if the
 366          * end of the file is before the start of our current
 367          * requested range of bytes, we bail out to the uncompressed
 368          * cleanup code that can deal with all of this.
 369          *
 370          * It isn't really the fastest way to fix things, but this is a
 371          * very uncommon corner.
 372          */
 373         if (actual_end <= start)
 374                 goto cleanup_and_bail_uncompressed;
 375
 376         total_compressed = actual_end - start;
 377
 378         /* we want to make sure that amount of ram required to uncompress
 379          * an extent is reasonable, so we limit the total size in ram
 380          * of a compressed extent to 128k.  This is a crucial number
 381          * because it also controls how easily we can spread reads across
 382          * cpus for decompression.
 383          *
 384          * We also want to make sure the amount of IO required to do
 385          * a random read is reasonably small, so we limit the size of
 386          * a compressed extent to 128k.
 387          */
 388         total_compressed = min(total_compressed, max_uncompressed);
 389         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 390         num_bytes = max(blocksize,  num_bytes);
 391         total_in = 0;
 392         ret = 0;
 393
 394         /*
 395          * we do compression for mount -o compress and when the
 396          * inode has not been flagged as nocompress.  This flag can
 397          * change at any time if we discover bad compression ratios.
 398          */
 399         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 400             (btrfs_test_opt(root, COMPRESS) ||
 401              (BTRFS_I(inode)->force_compress) ||
 402              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 403                 WARN_ON(pages);
 404                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 405                 if (!pages) {
 406                         /* just bail out to the uncompressed code */
 407                         goto cont;
 408                 }
 409
 410                 if (BTRFS_I(inode)->force_compress)
 411                         compress_type = BTRFS_I(inode)->force_compress;
 412
 413                 ret = btrfs_compress_pages(compress_type,
 414                                            inode->i_mapping, start,
 415                                            total_compressed, pages,
 416                                            nr_pages, &nr_pages_ret,
 417                                            &total_in,
 418                                            &total_compressed,
 419                                            max_compressed);
 420
 421                 if (!ret) {
 422                         unsigned long offset = total_compressed &
 423                                 (PAGE_CACHE_SIZE - 1);
 424                         struct page *page = pages[nr_pages_ret - 1];
 425                         char *kaddr;
 426
 427                         /* zero the tail end of the last page, we might be
 428                          * sending it down to disk
 429                          */
 430                         if (offset) {
 431                                 kaddr = kmap_atomic(page);
 432                                 memset(kaddr + offset, 0,
 433                                        PAGE_CACHE_SIZE - offset);
 434                                 kunmap_atomic(kaddr);
 435                         }
 436                         will_compress = 1;
 437                 }
 438         }
 439 cont:
 440         if (start == 0) {
 441                 trans = btrfs_join_transaction(root);
 442                 if (IS_ERR(trans)) {
 443                         ret = PTR_ERR(trans);
 444                         trans = NULL;
 445                         goto cleanup_and_out;
 446                 }
 447                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 448
 449                 /* lets try to make an inline extent */
 450                 if (ret || total_in < (actual_end - start)) {
 451                         /* we didn't compress the entire range, try
 452                          * to make an uncompressed inline extent.
 453                          */
 454                         ret = cow_file_range_inline(trans, root, inode,
 455                                                     start, end, 0, 0, NULL);
 456                 } else {
 457                         /* try making a compressed inline extent */
 458                         ret = cow_file_range_inline(trans, root, inode,
 459                                                     start, end,
 460                                                     total_compressed,
 461                                                     compress_type, pages);
 462                 }
 463                 if (ret <= 0) {
 464                         /*
 465                          * inline extent creation worked or returned error,
 466                          * we don't need to create any more async work items.
 467                          * Unlock and free up our temp pages.
 468                          */
 469                         extent_clear_unlock_delalloc(inode,
 470                              &BTRFS_I(inode)->io_tree,
 471                              start, end, NULL,
 472                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 473                              EXTENT_CLEAR_DELALLOC |
 474                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 475
 476                         btrfs_end_transaction(trans, root);
 477                         goto free_pages_out;
 478                 }
 479                 btrfs_end_transaction(trans, root);
 480         }
 481
 482         if (will_compress) {
 483                 /*
 484                  * we aren't doing an inline extent round the compressed size
 485                  * up to a block size boundary so the allocator does sane
 486                  * things
 487                  */
 488                 total_compressed = (total_compressed + blocksize - 1) &
 489                         ~(blocksize - 1);
 490
 491                 /*
 492                  * one last check to make sure the compression is really a
 493                  * win, compare the page count read with the blocks on disk
 494                  */
 495                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 496                         ~(PAGE_CACHE_SIZE - 1);
 497                 if (total_compressed >= total_in) {
 498                         will_compress = 0;
 499                 } else {
 500                         num_bytes = total_in;
 501                 }
 502         }
 503         if (!will_compress && pages) {
 504                 /*
 505                  * the compression code ran but failed to make things smaller,
 506                  * free any pages it allocated and our page pointer array
 507                  */
 508                 for (i = 0; i < nr_pages_ret; i++) {
 509                         WARN_ON(pages[i]->mapping);
 510                         page_cache_release(pages[i]);
 511                 }
 512                 kfree(pages);
 513                 pages = NULL;
 514                 total_compressed = 0;
 515                 nr_pages_ret = 0;
 516
 517                 /* flag the file so we don't compress in the future */
 518                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 519                     !(BTRFS_I(inode)->force_compress)) {
 520                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 521                 }
 522         }
 523         if (will_compress) {
 524                 *num_added += 1;
 525
 526                 /* the async work queues will take care of doing actual
 527                  * allocation on disk for these compressed pages,
 528                  * and will submit them to the elevator.
 529                  */
 530                 add_async_extent(async_cow, start, num_bytes,
 531                                  total_compressed, pages, nr_pages_ret,
 532                                  compress_type);
 533
 534                 if (start + num_bytes < end) {
 535                         start += num_bytes;
 536                         pages = NULL;
 537                         cond_resched();
 538                         goto again;
 539                 }
 540         } else {
 541 cleanup_and_bail_uncompressed:
 542                 /*
 543                  * No compression, but we still need to write the pages in
 544                  * the file we've been given so far.  redirty the locked
 545                  * page if it corresponds to our extent and set things up
 546                  * for the async work queue to run cow_file_range to do
 547                  * the normal delalloc dance
 548                  */
 549                 if (page_offset(locked_page) >= start &&
 550                     page_offset(locked_page) <= end) {
 551                         __set_page_dirty_nobuffers(locked_page);
 552                         /* unlocked later on in the async handlers */
 553                 }
 554                 add_async_extent(async_cow, start, end - start + 1,
 555                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 556                 *num_added += 1;
 557         }
 558
 559 out:
 560         return ret;
 561
 562 free_pages_out:
 563         for (i = 0; i < nr_pages_ret; i++) {
 564                 WARN_ON(pages[i]->mapping);
 565                 page_cache_release(pages[i]);
 566         }
 567         kfree(pages);
 568
 569         goto out;
 570
 571 cleanup_and_out:
 572         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 573                                      start, end, NULL,
 574                                      EXTENT_CLEAR_UNLOCK_PAGE |
 575                                      EXTENT_CLEAR_DIRTY |
 576                                      EXTENT_CLEAR_DELALLOC |
 577                                      EXTENT_SET_WRITEBACK |
 578                                      EXTENT_END_WRITEBACK);
 579         if (!trans || IS_ERR(trans))
 580                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 581         else
 582                 btrfs_abort_transaction(trans, root, ret);
 583         goto free_pages_out;
 584 }
 585
 586 /*
 587  * phase two of compressed writeback.  This is the ordered portion
 588  * of the code, which only gets called in the order the work was
 589  * queued.  We walk all the async extents created by compress_file_range
 590  * and send them down to the disk.
 591  */
 592 static noinline int submit_compressed_extents(struct inode *inode,
 593                                               struct async_cow *async_cow)
 594 {
 595         struct async_extent *async_extent;
 596         u64 alloc_hint = 0;
 597         struct btrfs_trans_handle *trans;
 598         struct btrfs_key ins;
 599         struct extent_map *em;
 600         struct btrfs_root *root = BTRFS_I(inode)->root;
 601         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 602         struct extent_io_tree *io_tree;
 603         int ret = 0;
 604
 605         if (list_empty(&async_cow->extents))
 606                 return 0;
 607
 608
 609         while (!list_empty(&async_cow->extents)) {
 610                 async_extent = list_entry(async_cow->extents.next,
 611                                           struct async_extent, list);
 612                 list_del(&async_extent->list);
 613
 614                 io_tree = &BTRFS_I(inode)->io_tree;
 615
 616 retry:
 617                 /* did the compression code fall back to uncompressed IO? */
 618                 if (!async_extent->pages) {
 619                         int page_started = 0;
 620                         unsigned long nr_written = 0;
 621
 622                         lock_extent(io_tree, async_extent->start,
 623                                          async_extent->start +
 624                                          async_extent->ram_size - 1);
 625
 626                         /* allocate blocks */
 627                         ret = cow_file_range(inode, async_cow->locked_page,
 628                                              async_extent->start,
 629                                              async_extent->start +
 630                                              async_extent->ram_size - 1,
 631                                              &page_started, &nr_written, 0);
 632
 633                         /* JDM XXX */
 634
 635                         /*
 636                          * if page_started, cow_file_range inserted an
 637                          * inline extent and took care of all the unlocking
 638                          * and IO for us.  Otherwise, we need to submit
 639                          * all those pages down to the drive.
 640                          */
 641                         if (!page_started && !ret)
 642                                 extent_write_locked_range(io_tree,
 643                                                   inode, async_extent->start,
 644                                                   async_extent->start +
 645                                                   async_extent->ram_size - 1,
 646                                                   btrfs_get_extent,
 647                                                   WB_SYNC_ALL);
 648                         kfree(async_extent);
 649                         cond_resched();
 650                         continue;
 651                 }
 652
 653                 lock_extent(io_tree, async_extent->start,
 654                             async_extent->start + async_extent->ram_size - 1);
 655
 656                 trans = btrfs_join_transaction(root);
 657                 if (IS_ERR(trans)) {
 658                         ret = PTR_ERR(trans);
 659                 } else {
 660                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 661                         ret = btrfs_reserve_extent(trans, root,
 662                                            async_extent->compressed_size,
 663                                            async_extent->compressed_size,
 664                                            0, alloc_hint, &ins, 1);
 665                         if (ret && ret != -ENOSPC)
 666                                 btrfs_abort_transaction(trans, root, ret);
 667                         btrfs_end_transaction(trans, root);
 668                 }
 669
 670                 if (ret) {
 671                         int i;
 672                         for (i = 0; i < async_extent->nr_pages; i++) {
 673                                 WARN_ON(async_extent->pages[i]->mapping);
 674                                 page_cache_release(async_extent->pages[i]);
 675                         }
 676                         kfree(async_extent->pages);
 677                         async_extent->nr_pages = 0;
 678                         async_extent->pages = NULL;
 679                         unlock_extent(io_tree, async_extent->start,
 680                                       async_extent->start +
 681                                       async_extent->ram_size - 1);
 682                         if (ret == -ENOSPC)
 683                                 goto retry;
 684                         goto out_free; /* JDM: Requeue? */
 685                 }
 686
 687                 /*
 688                  * here we're doing allocation and writeback of the
 689                  * compressed pages
 690                  */
 691                 btrfs_drop_extent_cache(inode, async_extent->start,
 692                                         async_extent->start +
 693                                         async_extent->ram_size - 1, 0);
 694
 695                 em = alloc_extent_map();
 696                 BUG_ON(!em); /* -ENOMEM */
 697                 em->start = async_extent->start;
 698                 em->len = async_extent->ram_size;
 699                 em->orig_start = em->start;
 700
 701                 em->block_start = ins.objectid;
 702                 em->block_len = ins.offset;
 703                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 704                 em->compress_type = async_extent->compress_type;
 705                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 706                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 707
 708                 while (1) {
 709                         write_lock(&em_tree->lock);
 710                         ret = add_extent_mapping(em_tree, em);
 711                         write_unlock(&em_tree->lock);
 712                         if (ret != -EEXIST) {
 713                                 free_extent_map(em);
 714                                 break;
 715                         }
 716                         btrfs_drop_extent_cache(inode, async_extent->start,
 717                                                 async_extent->start +
 718                                                 async_extent->ram_size - 1, 0);
 719                 }
 720
 721                 ret = btrfs_add_ordered_extent_compress(inode,
 722                                                 async_extent->start,
 723                                                 ins.objectid,
 724                                                 async_extent->ram_size,
 725                                                 ins.offset,
 726                                                 BTRFS_ORDERED_COMPRESSED,
 727                                                 async_extent->compress_type);
 728                 BUG_ON(ret); /* -ENOMEM */
 729
 730                 /*
 731                  * clear dirty, set writeback and unlock the pages.
 732                  */
 733                 extent_clear_unlock_delalloc(inode,
 734                                 &BTRFS_I(inode)->io_tree,
 735                                 async_extent->start,
 736                                 async_extent->start +
 737                                 async_extent->ram_size - 1,
 738                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 739                                 EXTENT_CLEAR_UNLOCK |
 740                                 EXTENT_CLEAR_DELALLOC |
 741                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 742
 743                 ret = btrfs_submit_compressed_write(inode,
 744                                     async_extent->start,
 745                                     async_extent->ram_size,
 746                                     ins.objectid,
 747                                     ins.offset, async_extent->pages,
 748                                     async_extent->nr_pages);
 749
 750                 BUG_ON(ret); /* -ENOMEM */
 751                 alloc_hint = ins.objectid + ins.offset;
 752                 kfree(async_extent);
 753                 cond_resched();
 754         }
 755         ret = 0;
 756 out:
 757         return ret;
 758 out_free:
 759         kfree(async_extent);
 760         goto out;
 761 }
 762
 763 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 764                                       u64 num_bytes)
 765 {
 766         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 767         struct extent_map *em;
 768         u64 alloc_hint = 0;
 769
 770         read_lock(&em_tree->lock);
 771         em = search_extent_mapping(em_tree, start, num_bytes);
 772         if (em) {
 773                 /*
 774                  * if block start isn't an actual block number then find the
 775                  * first block in this inode and use that as a hint.  If that
 776                  * block is also bogus then just don't worry about it.
 777                  */
 778                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 779                         free_extent_map(em);
 780                         em = search_extent_mapping(em_tree, 0, 0);
 781                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 782                                 alloc_hint = em->block_start;
 783                         if (em)
 784                                 free_extent_map(em);
 785                 } else {
 786                         alloc_hint = em->block_start;
 787                         free_extent_map(em);
 788                 }
 789         }
 790         read_unlock(&em_tree->lock);
 791
 792         return alloc_hint;
 793 }
 794
 795 /*
 796  * when extent_io.c finds a delayed allocation range in the file,
 797  * the call backs end up in this code.  The basic idea is to
 798  * allocate extents on disk for the range, and create ordered data structs
 799  * in ram to track those extents.
 800  *
 801  * locked_page is the page that writepage had locked already.  We use
 802  * it to make sure we don't do extra locks or unlocks.
 803  *
 804  * *page_started is set to one if we unlock locked_page and do everything
 805  * required to start IO on it.  It may be clean and already done with
 806  * IO when we return.
 807  */
 808 static noinline int cow_file_range(struct inode *inode,
 809                                    struct page *locked_page,
 810                                    u64 start, u64 end, int *page_started,
 811                                    unsigned long *nr_written,
 812                                    int unlock)
 813 {
 814         struct btrfs_root *root = BTRFS_I(inode)->root;
 815         struct btrfs_trans_handle *trans;
 816         u64 alloc_hint = 0;
 817         u64 num_bytes;
 818         unsigned long ram_size;
 819         u64 disk_num_bytes;
 820         u64 cur_alloc_size;
 821         u64 blocksize = root->sectorsize;
 822         struct btrfs_key ins;
 823         struct extent_map *em;
 824         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 825         int ret = 0;
 826
 827         BUG_ON(btrfs_is_free_space_inode(inode));
 828         trans = btrfs_join_transaction(root);
 829         if (IS_ERR(trans)) {
 830                 extent_clear_unlock_delalloc(inode,
 831                              &BTRFS_I(inode)->io_tree,
 832                              start, end, locked_page,
 833                              EXTENT_CLEAR_UNLOCK_PAGE |
 834                              EXTENT_CLEAR_UNLOCK |
 835                              EXTENT_CLEAR_DELALLOC |
 836                              EXTENT_CLEAR_DIRTY |
 837                              EXTENT_SET_WRITEBACK |
 838                              EXTENT_END_WRITEBACK);
 839                 return PTR_ERR(trans);
 840         }
 841         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 842
 843         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 844         num_bytes = max(blocksize,  num_bytes);
 845         disk_num_bytes = num_bytes;
 846         ret = 0;
 847
 848         /* if this is a small write inside eof, kick off defrag */
 849         if (num_bytes < 64 * 1024 &&
 850             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 851                 btrfs_add_inode_defrag(trans, inode);
 852
 853         if (start == 0) {
 854                 /* lets try to make an inline extent */
 855                 ret = cow_file_range_inline(trans, root, inode,
 856                                             start, end, 0, 0, NULL);
 857                 if (ret == 0) {
 858                         extent_clear_unlock_delalloc(inode,
 859                                      &BTRFS_I(inode)->io_tree,
 860                                      start, end, NULL,
 861                                      EXTENT_CLEAR_UNLOCK_PAGE |
 862                                      EXTENT_CLEAR_UNLOCK |
 863                                      EXTENT_CLEAR_DELALLOC |
 864                                      EXTENT_CLEAR_DIRTY |
 865                                      EXTENT_SET_WRITEBACK |
 866                                      EXTENT_END_WRITEBACK);
 867
 868                         *nr_written = *nr_written +
 869                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 870                         *page_started = 1;
 871                         goto out;
 872                 } else if (ret < 0) {
 873                         btrfs_abort_transaction(trans, root, ret);
 874                         goto out_unlock;
 875                 }
 876         }
 877
 878         BUG_ON(disk_num_bytes >
 879                btrfs_super_total_bytes(root->fs_info->super_copy));
 880
 881         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 882         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 883
 884         while (disk_num_bytes > 0) {
 885                 unsigned long op;
 886
 887                 cur_alloc_size = disk_num_bytes;
 888                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 889                                            root->sectorsize, 0, alloc_hint,
 890                                            &ins, 1);
 891                 if (ret < 0) {
 892                         btrfs_abort_transaction(trans, root, ret);
 893                         goto out_unlock;
 894                 }
 895
 896                 em = alloc_extent_map();
 897                 BUG_ON(!em); /* -ENOMEM */
 898                 em->start = start;
 899                 em->orig_start = em->start;
 900                 ram_size = ins.offset;
 901                 em->len = ins.offset;
 902
 903                 em->block_start = ins.objectid;
 904                 em->block_len = ins.offset;
 905                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 906                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 907
 908                 while (1) {
 909                         write_lock(&em_tree->lock);
 910                         ret = add_extent_mapping(em_tree, em);
 911                         write_unlock(&em_tree->lock);
 912                         if (ret != -EEXIST) {
 913                                 free_extent_map(em);
 914                                 break;
 915                         }
 916                         btrfs_drop_extent_cache(inode, start,
 917                                                 start + ram_size - 1, 0);
 918                 }
 919
 920                 cur_alloc_size = ins.offset;
 921                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 922                                                ram_size, cur_alloc_size, 0);
 923                 BUG_ON(ret); /* -ENOMEM */
 924
 925                 if (root->root_key.objectid ==
 926                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 927                         ret = btrfs_reloc_clone_csums(inode, start,
 928                                                       cur_alloc_size);
 929                         if (ret) {
 930                                 btrfs_abort_transaction(trans, root, ret);
 931                                 goto out_unlock;
 932                         }
 933                 }
 934
 935                 if (disk_num_bytes < cur_alloc_size)
 936                         break;
 937
 938                 /* we're not doing compressed IO, don't unlock the first
 939                  * page (which the caller expects to stay locked), don't
 940                  * clear any dirty bits and don't set any writeback bits
 941                  *
 942                  * Do set the Private2 bit so we know this page was properly
 943                  * setup for writepage
 944                  */
 945                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 946                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 947                         EXTENT_SET_PRIVATE2;
 948
 949                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 950                                              start, start + ram_size - 1,
 951                                              locked_page, op);
 952                 disk_num_bytes -= cur_alloc_size;
 953                 num_bytes -= cur_alloc_size;
 954                 alloc_hint = ins.objectid + ins.offset;
 955                 start += cur_alloc_size;
 956         }
 957         ret = 0;
 958 out:
 959         btrfs_end_transaction(trans, root);
 960
 961         return ret;
 962 out_unlock:
 963         extent_clear_unlock_delalloc(inode,
 964                      &BTRFS_I(inode)->io_tree,
 965                      start, end, locked_page,
 966                      EXTENT_CLEAR_UNLOCK_PAGE |
 967                      EXTENT_CLEAR_UNLOCK |
 968                      EXTENT_CLEAR_DELALLOC |
 969                      EXTENT_CLEAR_DIRTY |
 970                      EXTENT_SET_WRITEBACK |
 971                      EXTENT_END_WRITEBACK);
 972
 973         goto out;
 974 }
 975
 976 /*
 977  * work queue call back to started compression on a file and pages
 978  */
 979 static noinline void async_cow_start(struct btrfs_work *work)
 980 {
 981         struct async_cow *async_cow;
 982         int num_added = 0;
 983         async_cow = container_of(work, struct async_cow, work);
 984
 985         compress_file_range(async_cow->inode, async_cow->locked_page,
 986                             async_cow->start, async_cow->end, async_cow,
 987                             &num_added);
 988         if (num_added == 0) {
 989                 btrfs_add_delayed_iput(async_cow->inode);
 990                 async_cow->inode = NULL;
 991         }
 992 }
 993
 994 /*
 995  * work queue call back to submit previously compressed pages
 996  */
 997 static noinline void async_cow_submit(struct btrfs_work *work)
 998 {
 999         struct async_cow *async_cow;
1000         struct btrfs_root *root;
1001         unsigned long nr_pages;
1002
1003         async_cow = container_of(work, struct async_cow, work);
1004
1005         root = async_cow->root;
1006         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1007                 PAGE_CACHE_SHIFT;
1008
1009         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1010             5 * 1024 * 1024 &&
1011             waitqueue_active(&root->fs_info->async_submit_wait))
1012                 wake_up(&root->fs_info->async_submit_wait);
1013
1014         if (async_cow->inode)
1015                 submit_compressed_extents(async_cow->inode, async_cow);
1016 }
1017
1018 static noinline void async_cow_free(struct btrfs_work *work)
1019 {
1020         struct async_cow *async_cow;
1021         async_cow = container_of(work, struct async_cow, work);
1022         if (async_cow->inode)
1023                 btrfs_add_delayed_iput(async_cow->inode);
1024         kfree(async_cow);
1025 }
1026
1027 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1028                                 u64 start, u64 end, int *page_started,
1029                                 unsigned long *nr_written)
1030 {
1031         struct async_cow *async_cow;
1032         struct btrfs_root *root = BTRFS_I(inode)->root;
1033         unsigned long nr_pages;
1034         u64 cur_end;
1035         int limit = 10 * 1024 * 1024;
1036
1037         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1038                          1, 0, NULL, GFP_NOFS);
1039         while (start < end) {
1040                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1041                 BUG_ON(!async_cow); /* -ENOMEM */
1042                 async_cow->inode = igrab(inode);
1043                 async_cow->root = root;
1044                 async_cow->locked_page = locked_page;
1045                 async_cow->start = start;
1046
1047                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1048                         cur_end = end;
1049                 else
1050                         cur_end = min(end, start + 512 * 1024 - 1);
1051
1052                 async_cow->end = cur_end;
1053                 INIT_LIST_HEAD(&async_cow->extents);
1054
1055                 async_cow->work.func = async_cow_start;
1056                 async_cow->work.ordered_func = async_cow_submit;
1057                 async_cow->work.ordered_free = async_cow_free;
1058                 async_cow->work.flags = 0;
1059
1060                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1061                         PAGE_CACHE_SHIFT;
1062                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1063
1064                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1065                                    &async_cow->work);
1066
1067                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1068                         wait_event(root->fs_info->async_submit_wait,
1069                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1070                             limit));
1071                 }
1072
1073                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1074                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1075                         wait_event(root->fs_info->async_submit_wait,
1076                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1077                            0));
1078                 }
1079
1080                 *nr_written += nr_pages;
1081                 start = cur_end + 1;
1082         }
1083         *page_started = 1;
1084         return 0;
1085 }
1086
1087 static noinline int csum_exist_in_range(struct btrfs_root *root,
1088                                         u64 bytenr, u64 num_bytes)
1089 {
1090         int ret;
1091         struct btrfs_ordered_sum *sums;
1092         LIST_HEAD(list);
1093
1094         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1095                                        bytenr + num_bytes - 1, &list, 0);
1096         if (ret == 0 && list_empty(&list))
1097                 return 0;
1098
1099         while (!list_empty(&list)) {
1100                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1101                 list_del(&sums->list);
1102                 kfree(sums);
1103         }
1104         return 1;
1105 }
1106
1107 /*
1108  * when nowcow writeback call back.  This checks for snapshots or COW copies
1109  * of the extents that exist in the file, and COWs the file as required.
1110  *
1111  * If no cow copies or snapshots exist, we write directly to the existing
1112  * blocks on disk
1113  */
1114 static noinline int run_delalloc_nocow(struct inode *inode,
1115                                        struct page *locked_page,
1116                               u64 start, u64 end, int *page_started, int force,
1117                               unsigned long *nr_written)
1118 {
1119         struct btrfs_root *root = BTRFS_I(inode)->root;
1120         struct btrfs_trans_handle *trans;
1121         struct extent_buffer *leaf;
1122         struct btrfs_path *path;
1123         struct btrfs_file_extent_item *fi;
1124         struct btrfs_key found_key;
1125         u64 cow_start;
1126         u64 cur_offset;
1127         u64 extent_end;
1128         u64 extent_offset;
1129         u64 disk_bytenr;
1130         u64 num_bytes;
1131         int extent_type;
1132         int ret, err;
1133         int type;
1134         int nocow;
1135         int check_prev = 1;
1136         bool nolock;
1137         u64 ino = btrfs_ino(inode);
1138
1139         path = btrfs_alloc_path();
1140         if (!path) {
1141                 extent_clear_unlock_delalloc(inode,
1142                              &BTRFS_I(inode)->io_tree,
1143                              start, end, locked_page,
1144                              EXTENT_CLEAR_UNLOCK_PAGE |
1145                              EXTENT_CLEAR_UNLOCK |
1146                              EXTENT_CLEAR_DELALLOC |
1147                              EXTENT_CLEAR_DIRTY |
1148                              EXTENT_SET_WRITEBACK |
1149                              EXTENT_END_WRITEBACK);
1150                 return -ENOMEM;
1151         }
1152
1153         nolock = btrfs_is_free_space_inode(inode);
1154
1155         if (nolock)
1156                 trans = btrfs_join_transaction_nolock(root);
1157         else
1158                 trans = btrfs_join_transaction(root);
1159
1160         if (IS_ERR(trans)) {
1161                 extent_clear_unlock_delalloc(inode,
1162                              &BTRFS_I(inode)->io_tree,
1163                              start, end, locked_page,
1164                              EXTENT_CLEAR_UNLOCK_PAGE |
1165                              EXTENT_CLEAR_UNLOCK |
1166                              EXTENT_CLEAR_DELALLOC |
1167                              EXTENT_CLEAR_DIRTY |
1168                              EXTENT_SET_WRITEBACK |
1169                              EXTENT_END_WRITEBACK);
1170                 btrfs_free_path(path);
1171                 return PTR_ERR(trans);
1172         }
1173
1174         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1175
1176         cow_start = (u64)-1;
1177         cur_offset = start;
1178         while (1) {
1179                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1180                                                cur_offset, 0);
1181                 if (ret < 0) {
1182                         btrfs_abort_transaction(trans, root, ret);
1183                         goto error;
1184                 }
1185                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1186                         leaf = path->nodes[0];
1187                         btrfs_item_key_to_cpu(leaf, &found_key,
1188                                               path->slots[0] - 1);
1189                         if (found_key.objectid == ino &&
1190                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1191                                 path->slots[0]--;
1192                 }
1193                 check_prev = 0;
1194 next_slot:
1195                 leaf = path->nodes[0];
1196                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1197                         ret = btrfs_next_leaf(root, path);
1198                         if (ret < 0) {
1199                                 btrfs_abort_transaction(trans, root, ret);
1200                                 goto error;
1201                         }
1202                         if (ret > 0)
1203                                 break;
1204                         leaf = path->nodes[0];
1205                 }
1206
1207                 nocow = 0;
1208                 disk_bytenr = 0;
1209                 num_bytes = 0;
1210                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1211
1212                 if (found_key.objectid > ino ||
1213                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1214                     found_key.offset > end)
1215                         break;
1216
1217                 if (found_key.offset > cur_offset) {
1218                         extent_end = found_key.offset;
1219                         extent_type = 0;
1220                         goto out_check;
1221                 }
1222
1223                 fi = btrfs_item_ptr(leaf, path->slots[0],
1224                                     struct btrfs_file_extent_item);
1225                 extent_type = btrfs_file_extent_type(leaf, fi);
1226
1227                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1228                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1229                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1230                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1231                         extent_end = found_key.offset +
1232                                 btrfs_file_extent_num_bytes(leaf, fi);
1233                         if (extent_end <= start) {
1234                                 path->slots[0]++;
1235                                 goto next_slot;
1236                         }
1237                         if (disk_bytenr == 0)
1238                                 goto out_check;
1239                         if (btrfs_file_extent_compression(leaf, fi) ||
1240                             btrfs_file_extent_encryption(leaf, fi) ||
1241                             btrfs_file_extent_other_encoding(leaf, fi))
1242                                 goto out_check;
1243                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1244                                 goto out_check;
1245                         if (btrfs_extent_readonly(root, disk_bytenr))
1246                                 goto out_check;
1247                         if (btrfs_cross_ref_exist(trans, root, ino,
1248                                                   found_key.offset -
1249                                                   extent_offset, disk_bytenr))
1250                                 goto out_check;
1251                         disk_bytenr += extent_offset;
1252                         disk_bytenr += cur_offset - found_key.offset;
1253                         num_bytes = min(end + 1, extent_end) - cur_offset;
1254                         /*
1255                          * force cow if csum exists in the range.
1256                          * this ensure that csum for a given extent are
1257                          * either valid or do not exist.
1258                          */
1259                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1260                                 goto out_check;
1261                         nocow = 1;
1262                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1263                         extent_end = found_key.offset +
1264                                 btrfs_file_extent_inline_len(leaf, fi);
1265                         extent_end = ALIGN(extent_end, root->sectorsize);
1266                 } else {
1267                         BUG_ON(1);
1268                 }
1269 out_check:
1270                 if (extent_end <= start) {
1271                         path->slots[0]++;
1272                         goto next_slot;
1273                 }
1274                 if (!nocow) {
1275                         if (cow_start == (u64)-1)
1276                                 cow_start = cur_offset;
1277                         cur_offset = extent_end;
1278                         if (cur_offset > end)
1279                                 break;
1280                         path->slots[0]++;
1281                         goto next_slot;
1282                 }
1283
1284                 btrfs_release_path(path);
1285                 if (cow_start != (u64)-1) {
1286                         ret = cow_file_range(inode, locked_page, cow_start,
1287                                         found_key.offset - 1, page_started,
1288                                         nr_written, 1);
1289                         if (ret) {
1290                                 btrfs_abort_transaction(trans, root, ret);
1291                                 goto error;
1292                         }
1293                         cow_start = (u64)-1;
1294                 }
1295
1296                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1297                         struct extent_map *em;
1298                         struct extent_map_tree *em_tree;
1299                         em_tree = &BTRFS_I(inode)->extent_tree;
1300                         em = alloc_extent_map();
1301                         BUG_ON(!em); /* -ENOMEM */
1302                         em->start = cur_offset;
1303                         em->orig_start = em->start;
1304                         em->len = num_bytes;
1305                         em->block_len = num_bytes;
1306                         em->block_start = disk_bytenr;
1307                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1308                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1309                         set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1310                         while (1) {
1311                                 write_lock(&em_tree->lock);
1312                                 ret = add_extent_mapping(em_tree, em);
1313                                 write_unlock(&em_tree->lock);
1314                                 if (ret != -EEXIST) {
1315                                         free_extent_map(em);
1316                                         break;
1317                                 }
1318                                 btrfs_drop_extent_cache(inode, em->start,
1319                                                 em->start + em->len - 1, 0);
1320                         }
1321                         type = BTRFS_ORDERED_PREALLOC;
1322                 } else {
1323                         type = BTRFS_ORDERED_NOCOW;
1324                 }
1325
1326                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1327                                                num_bytes, num_bytes, type);
1328                 BUG_ON(ret); /* -ENOMEM */
1329
1330                 if (root->root_key.objectid ==
1331                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1332                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1333                                                       num_bytes);
1334                         if (ret) {
1335                                 btrfs_abort_transaction(trans, root, ret);
1336                                 goto error;
1337                         }
1338                 }
1339
1340                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1341                                 cur_offset, cur_offset + num_bytes - 1,
1342                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1343                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1344                                 EXTENT_SET_PRIVATE2);
1345                 cur_offset = extent_end;
1346                 if (cur_offset > end)
1347                         break;
1348         }
1349         btrfs_release_path(path);
1350
1351         if (cur_offset <= end && cow_start == (u64)-1) {
1352                 cow_start = cur_offset;
1353                 cur_offset = end;
1354         }
1355
1356         if (cow_start != (u64)-1) {
1357                 ret = cow_file_range(inode, locked_page, cow_start, end,
1358                                      page_started, nr_written, 1);
1359                 if (ret) {
1360                         btrfs_abort_transaction(trans, root, ret);
1361                         goto error;
1362                 }
1363         }
1364
1365 error:
1366         err = btrfs_end_transaction(trans, root);
1367         if (!ret)
1368                 ret = err;
1369
1370         if (ret && cur_offset < end)
1371                 extent_clear_unlock_delalloc(inode,
1372                              &BTRFS_I(inode)->io_tree,
1373                              cur_offset, end, locked_page,
1374                              EXTENT_CLEAR_UNLOCK_PAGE |
1375                              EXTENT_CLEAR_UNLOCK |
1376                              EXTENT_CLEAR_DELALLOC |
1377                              EXTENT_CLEAR_DIRTY |
1378                              EXTENT_SET_WRITEBACK |
1379                              EXTENT_END_WRITEBACK);
1380
1381         btrfs_free_path(path);
1382         return ret;
1383 }
1384
1385 /*
1386  * extent_io.c call back to do delayed allocation processing
1387  */
1388 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1389                               u64 start, u64 end, int *page_started,
1390                               unsigned long *nr_written)
1391 {
1392         int ret;
1393         struct btrfs_root *root = BTRFS_I(inode)->root;
1394
1395         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1396                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1397                                          page_started, 1, nr_written);
1398         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1399                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1400                                          page_started, 0, nr_written);
1401         } else if (!btrfs_test_opt(root, COMPRESS) &&
1402                    !(BTRFS_I(inode)->force_compress) &&
1403                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1404                 ret = cow_file_range(inode, locked_page, start, end,
1405                                       page_started, nr_written, 1);
1406         } else {
1407                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1408                         &BTRFS_I(inode)->runtime_flags);
1409                 ret = cow_file_range_async(inode, locked_page, start, end,
1410                                            page_started, nr_written);
1411         }
1412         return ret;
1413 }
1414
1415 static void btrfs_split_extent_hook(struct inode *inode,
1416                                     struct extent_state *orig, u64 split)
1417 {
1418         /* not delalloc, ignore it */
1419         if (!(orig->state & EXTENT_DELALLOC))
1420                 return;
1421
1422         spin_lock(&BTRFS_I(inode)->lock);
1423         BTRFS_I(inode)->outstanding_extents++;
1424         spin_unlock(&BTRFS_I(inode)->lock);
1425 }
1426
1427 /*
1428  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1429  * extents so we can keep track of new extents that are just merged onto old
1430  * extents, such as when we are doing sequential writes, so we can properly
1431  * account for the metadata space we'll need.
1432  */
1433 static void btrfs_merge_extent_hook(struct inode *inode,
1434                                     struct extent_state *new,
1435                                     struct extent_state *other)
1436 {
1437         /* not delalloc, ignore it */
1438         if (!(other->state & EXTENT_DELALLOC))
1439                 return;
1440
1441         spin_lock(&BTRFS_I(inode)->lock);
1442         BTRFS_I(inode)->outstanding_extents--;
1443         spin_unlock(&BTRFS_I(inode)->lock);
1444 }
1445
1446 /*
1447  * extent_io.c set_bit_hook, used to track delayed allocation
1448  * bytes in this file, and to maintain the list of inodes that
1449  * have pending delalloc work to be done.
1450  */
1451 static void btrfs_set_bit_hook(struct inode *inode,
1452                                struct extent_state *state, int *bits)
1453 {
1454
1455         /*
1456          * set_bit and clear bit hooks normally require _irqsave/restore
1457          * but in this case, we are only testing for the DELALLOC
1458          * bit, which is only set or cleared with irqs on
1459          */
1460         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1461                 struct btrfs_root *root = BTRFS_I(inode)->root;
1462                 u64 len = state->end + 1 - state->start;
1463                 bool do_list = !btrfs_is_free_space_inode(inode);
1464
1465                 if (*bits & EXTENT_FIRST_DELALLOC) {
1466                         *bits &= ~EXTENT_FIRST_DELALLOC;
1467                 } else {
1468                         spin_lock(&BTRFS_I(inode)->lock);
1469                         BTRFS_I(inode)->outstanding_extents++;
1470                         spin_unlock(&BTRFS_I(inode)->lock);
1471                 }
1472
1473                 spin_lock(&root->fs_info->delalloc_lock);
1474                 BTRFS_I(inode)->delalloc_bytes += len;
1475                 root->fs_info->delalloc_bytes += len;
1476                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1477                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1478                                       &root->fs_info->delalloc_inodes);
1479                 }
1480                 spin_unlock(&root->fs_info->delalloc_lock);
1481         }
1482 }
1483
1484 /*
1485  * extent_io.c clear_bit_hook, see set_bit_hook for why
1486  */
1487 static void btrfs_clear_bit_hook(struct inode *inode,
1488                                  struct extent_state *state, int *bits)
1489 {
1490         /*
1491          * set_bit and clear bit hooks normally require _irqsave/restore
1492          * but in this case, we are only testing for the DELALLOC
1493          * bit, which is only set or cleared with irqs on
1494          */
1495         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1496                 struct btrfs_root *root = BTRFS_I(inode)->root;
1497                 u64 len = state->end + 1 - state->start;
1498                 bool do_list = !btrfs_is_free_space_inode(inode);
1499
1500                 if (*bits & EXTENT_FIRST_DELALLOC) {
1501                         *bits &= ~EXTENT_FIRST_DELALLOC;
1502                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1503                         spin_lock(&BTRFS_I(inode)->lock);
1504                         BTRFS_I(inode)->outstanding_extents--;
1505                         spin_unlock(&BTRFS_I(inode)->lock);
1506                 }
1507
1508                 if (*bits & EXTENT_DO_ACCOUNTING)
1509                         btrfs_delalloc_release_metadata(inode, len);
1510
1511                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1512                     && do_list)
1513                         btrfs_free_reserved_data_space(inode, len);
1514
1515                 spin_lock(&root->fs_info->delalloc_lock);
1516                 root->fs_info->delalloc_bytes -= len;
1517                 BTRFS_I(inode)->delalloc_bytes -= len;
1518
1519                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1520                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1521                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1522                 }
1523                 spin_unlock(&root->fs_info->delalloc_lock);
1524         }
1525 }
1526
1527 /*
1528  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1529  * we don't create bios that span stripes or chunks
1530  */
1531 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1532                          size_t size, struct bio *bio,
1533                          unsigned long bio_flags)
1534 {
1535         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1536         struct btrfs_mapping_tree *map_tree;
1537         u64 logical = (u64)bio->bi_sector << 9;
1538         u64 length = 0;
1539         u64 map_length;
1540         int ret;
1541
1542         if (bio_flags & EXTENT_BIO_COMPRESSED)
1543                 return 0;
1544
1545         length = bio->bi_size;
1546         map_tree = &root->fs_info->mapping_tree;
1547         map_length = length;
1548         ret = btrfs_map_block(map_tree, READ, logical,
1549                               &map_length, NULL, 0);
1550         /* Will always return 0 or 1 with map_multi == NULL */
1551         BUG_ON(ret < 0);
1552         if (map_length < length + size)
1553                 return 1;
1554         return 0;
1555 }
1556
1557 /*
1558  * in order to insert checksums into the metadata in large chunks,
1559  * we wait until bio submission time.   All the pages in the bio are
1560  * checksummed and sums are attached onto the ordered extent record.
1561  *
1562  * At IO completion time the cums attached on the ordered extent record
1563  * are inserted into the btree
1564  */
1565 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1566                                     struct bio *bio, int mirror_num,
1567                                     unsigned long bio_flags,
1568                                     u64 bio_offset)
1569 {
1570         struct btrfs_root *root = BTRFS_I(inode)->root;
1571         int ret = 0;
1572
1573         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1574         BUG_ON(ret); /* -ENOMEM */
1575         return 0;
1576 }
1577
1578 /*
1579  * in order to insert checksums into the metadata in large chunks,
1580  * we wait until bio submission time.   All the pages in the bio are
1581  * checksummed and sums are attached onto the ordered extent record.
1582  *
1583  * At IO completion time the cums attached on the ordered extent record
1584  * are inserted into the btree
1585  */
1586 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1587                           int mirror_num, unsigned long bio_flags,
1588                           u64 bio_offset)
1589 {
1590         struct btrfs_root *root = BTRFS_I(inode)->root;
1591         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1592 }
1593
1594 /*
1595  * extent_io.c submission hook. This does the right thing for csum calculation
1596  * on write, or reading the csums from the tree before a read
1597  */
1598 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1599                           int mirror_num, unsigned long bio_flags,
1600                           u64 bio_offset)
1601 {
1602         struct btrfs_root *root = BTRFS_I(inode)->root;
1603         int ret = 0;
1604         int skip_sum;
1605         int metadata = 0;
1606
1607         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1608
1609         if (btrfs_is_free_space_inode(inode))
1610                 metadata = 2;
1611
1612         if (!(rw & REQ_WRITE)) {
1613                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1614                 if (ret)
1615                         return ret;
1616
1617                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1618                         return btrfs_submit_compressed_read(inode, bio,
1619                                                     mirror_num, bio_flags);
1620                 } else if (!skip_sum) {
1621                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1622                         if (ret)
1623                                 return ret;
1624                 }
1625                 goto mapit;
1626         } else if (!skip_sum) {
1627                 /* csum items have already been cloned */
1628                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1629                         goto mapit;
1630                 /* we're doing a write, do the async checksumming */
1631                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1632                                    inode, rw, bio, mirror_num,
1633                                    bio_flags, bio_offset,
1634                                    __btrfs_submit_bio_start,
1635                                    __btrfs_submit_bio_done);
1636         }
1637
1638 mapit:
1639         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1640 }
1641
1642 /*
1643  * given a list of ordered sums record them in the inode.  This happens
1644  * at IO completion time based on sums calculated at bio submission time.
1645  */
1646 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1647                              struct inode *inode, u64 file_offset,
1648                              struct list_head *list)
1649 {
1650         struct btrfs_ordered_sum *sum;
1651
1652         list_for_each_entry(sum, list, list) {
1653                 btrfs_csum_file_blocks(trans,
1654                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1655         }
1656         return 0;
1657 }
1658
1659 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1660                               struct extent_state **cached_state)
1661 {
1662         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1663                 WARN_ON(1);
1664         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1665                                    cached_state, GFP_NOFS);
1666 }
1667
1668 /* see btrfs_writepage_start_hook for details on why this is required */
1669 struct btrfs_writepage_fixup {
1670         struct page *page;
1671         struct btrfs_work work;
1672 };
1673
1674 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1675 {
1676         struct btrfs_writepage_fixup *fixup;
1677         struct btrfs_ordered_extent *ordered;
1678         struct extent_state *cached_state = NULL;
1679         struct page *page;
1680         struct inode *inode;
1681         u64 page_start;
1682         u64 page_end;
1683         int ret;
1684
1685         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1686         page = fixup->page;
1687 again:
1688         lock_page(page);
1689         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1690                 ClearPageChecked(page);
1691                 goto out_page;
1692         }
1693
1694         inode = page->mapping->host;
1695         page_start = page_offset(page);
1696         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1697
1698         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1699                          &cached_state);
1700
1701         /* already ordered? We're done */
1702         if (PagePrivate2(page))
1703                 goto out;
1704
1705         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1706         if (ordered) {
1707                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1708                                      page_end, &cached_state, GFP_NOFS);
1709                 unlock_page(page);
1710                 btrfs_start_ordered_extent(inode, ordered, 1);
1711                 btrfs_put_ordered_extent(ordered);
1712                 goto again;
1713         }
1714
1715         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1716         if (ret) {
1717                 mapping_set_error(page->mapping, ret);
1718                 end_extent_writepage(page, ret, page_start, page_end);
1719                 ClearPageChecked(page);
1720                 goto out;
1721          }
1722
1723         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1724         ClearPageChecked(page);
1725         set_page_dirty(page);
1726 out:
1727         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1728                              &cached_state, GFP_NOFS);
1729 out_page:
1730         unlock_page(page);
1731         page_cache_release(page);
1732         kfree(fixup);
1733 }
1734
1735 /*
1736  * There are a few paths in the higher layers of the kernel that directly
1737  * set the page dirty bit without asking the filesystem if it is a
1738  * good idea.  This causes problems because we want to make sure COW
1739  * properly happens and the data=ordered rules are followed.
1740  *
1741  * In our case any range that doesn't have the ORDERED bit set
1742  * hasn't been properly setup for IO.  We kick off an async process
1743  * to fix it up.  The async helper will wait for ordered extents, set
1744  * the delalloc bit and make it safe to write the page.
1745  */
1746 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1747 {
1748         struct inode *inode = page->mapping->host;
1749         struct btrfs_writepage_fixup *fixup;
1750         struct btrfs_root *root = BTRFS_I(inode)->root;
1751
1752         /* this page is properly in the ordered list */
1753         if (TestClearPagePrivate2(page))
1754                 return 0;
1755
1756         if (PageChecked(page))
1757                 return -EAGAIN;
1758
1759         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1760         if (!fixup)
1761                 return -EAGAIN;
1762
1763         SetPageChecked(page);
1764         page_cache_get(page);
1765         fixup->work.func = btrfs_writepage_fixup_worker;
1766         fixup->page = page;
1767         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1768         return -EBUSY;
1769 }
1770
1771 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1772                                        struct inode *inode, u64 file_pos,
1773                                        u64 disk_bytenr, u64 disk_num_bytes,
1774                                        u64 num_bytes, u64 ram_bytes,
1775                                        u8 compression, u8 encryption,
1776                                        u16 other_encoding, int extent_type)
1777 {
1778         struct btrfs_root *root = BTRFS_I(inode)->root;
1779         struct btrfs_file_extent_item *fi;
1780         struct btrfs_path *path;
1781         struct extent_buffer *leaf;
1782         struct btrfs_key ins;
1783         int ret;
1784
1785         path = btrfs_alloc_path();
1786         if (!path)
1787                 return -ENOMEM;
1788
1789         path->leave_spinning = 1;
1790
1791         /*
1792          * we may be replacing one extent in the tree with another.
1793          * The new extent is pinned in the extent map, and we don't want
1794          * to drop it from the cache until it is completely in the btree.
1795          *
1796          * So, tell btrfs_drop_extents to leave this extent in the cache.
1797          * the caller is expected to unpin it and allow it to be merged
1798          * with the others.
1799          */
1800         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1801                                  file_pos + num_bytes, 0);
1802         if (ret)
1803                 goto out;
1804
1805         ins.objectid = btrfs_ino(inode);
1806         ins.offset = file_pos;
1807         ins.type = BTRFS_EXTENT_DATA_KEY;
1808         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1809         if (ret)
1810                 goto out;
1811         leaf = path->nodes[0];
1812         fi = btrfs_item_ptr(leaf, path->slots[0],
1813                             struct btrfs_file_extent_item);
1814         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1815         btrfs_set_file_extent_type(leaf, fi, extent_type);
1816         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1817         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1818         btrfs_set_file_extent_offset(leaf, fi, 0);
1819         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1820         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1821         btrfs_set_file_extent_compression(leaf, fi, compression);
1822         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1823         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1824
1825         btrfs_unlock_up_safe(path, 1);
1826         btrfs_set_lock_blocking(leaf);
1827
1828         btrfs_mark_buffer_dirty(leaf);
1829
1830         inode_add_bytes(inode, num_bytes);
1831
1832         ins.objectid = disk_bytenr;
1833         ins.offset = disk_num_bytes;
1834         ins.type = BTRFS_EXTENT_ITEM_KEY;
1835         ret = btrfs_alloc_reserved_file_extent(trans, root,
1836                                         root->root_key.objectid,
1837                                         btrfs_ino(inode), file_pos, &ins);
1838 out:
1839         btrfs_free_path(path);
1840
1841         return ret;
1842 }
1843
1844 /*
1845  * helper function for btrfs_finish_ordered_io, this
1846  * just reads in some of the csum leaves to prime them into ram
1847  * before we start the transaction.  It limits the amount of btree
1848  * reads required while inside the transaction.
1849  */
1850 /* as ordered data IO finishes, this gets called so we can finish
1851  * an ordered extent if the range of bytes in the file it covers are
1852  * fully written.
1853  */
1854 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1855 {
1856         struct inode *inode = ordered_extent->inode;
1857         struct btrfs_root *root = BTRFS_I(inode)->root;
1858         struct btrfs_trans_handle *trans = NULL;
1859         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1860         struct extent_state *cached_state = NULL;
1861         int compress_type = 0;
1862         int ret;
1863         bool nolock;
1864
1865         nolock = btrfs_is_free_space_inode(inode);
1866
1867         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1868                 ret = -EIO;
1869                 goto out;
1870         }
1871
1872         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1873                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1874                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1875                 if (!ret) {
1876                         if (nolock)
1877                                 trans = btrfs_join_transaction_nolock(root);
1878                         else
1879                                 trans = btrfs_join_transaction(root);
1880                         if (IS_ERR(trans)) {
1881                                 ret = PTR_ERR(trans);
1882                                 trans = NULL;
1883                                 goto out;
1884                         }
1885                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1886                         ret = btrfs_update_inode_fallback(trans, root, inode);
1887                         if (ret) /* -ENOMEM or corruption */
1888                                 btrfs_abort_transaction(trans, root, ret);
1889                 }
1890                 goto out;
1891         }
1892
1893         lock_extent_bits(io_tree, ordered_extent->file_offset,
1894                          ordered_extent->file_offset + ordered_extent->len - 1,
1895                          0, &cached_state);
1896
1897         if (nolock)
1898                 trans = btrfs_join_transaction_nolock(root);
1899         else
1900                 trans = btrfs_join_transaction(root);
1901         if (IS_ERR(trans)) {
1902                 ret = PTR_ERR(trans);
1903                 trans = NULL;
1904                 goto out_unlock;
1905         }
1906         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1907
1908         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1909                 compress_type = ordered_extent->compress_type;
1910         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1911                 BUG_ON(compress_type);
1912                 ret = btrfs_mark_extent_written(trans, inode,
1913                                                 ordered_extent->file_offset,
1914                                                 ordered_extent->file_offset +
1915                                                 ordered_extent->len);
1916         } else {
1917                 BUG_ON(root == root->fs_info->tree_root);
1918                 ret = insert_reserved_file_extent(trans, inode,
1919                                                 ordered_extent->file_offset,
1920                                                 ordered_extent->start,
1921                                                 ordered_extent->disk_len,
1922                                                 ordered_extent->len,
1923                                                 ordered_extent->len,
1924                                                 compress_type, 0, 0,
1925                                                 BTRFS_FILE_EXTENT_REG);
1926         }
1927         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1928                            ordered_extent->file_offset, ordered_extent->len,
1929                            trans->transid);
1930         if (ret < 0) {
1931                 btrfs_abort_transaction(trans, root, ret);
1932                 goto out_unlock;
1933         }
1934
1935         add_pending_csums(trans, inode, ordered_extent->file_offset,
1936                           &ordered_extent->list);
1937
1938         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1939         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1940                 ret = btrfs_update_inode_fallback(trans, root, inode);
1941                 if (ret) { /* -ENOMEM or corruption */
1942                         btrfs_abort_transaction(trans, root, ret);
1943                         goto out_unlock;
1944                 }
1945         } else {
1946                 btrfs_set_inode_last_trans(trans, inode);
1947         }
1948         ret = 0;
1949 out_unlock:
1950         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1951                              ordered_extent->file_offset +
1952                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1953 out:
1954         if (root != root->fs_info->tree_root)
1955                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1956         if (trans)
1957                 btrfs_end_transaction(trans, root);
1958
1959         if (ret)
1960                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1961                                       ordered_extent->file_offset +
1962                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1963
1964         /*
1965          * This needs to be dont to make sure anybody waiting knows we are done
1966          * upating everything for this ordered extent.
1967          */
1968         btrfs_remove_ordered_extent(inode, ordered_extent);
1969
1970         /* once for us */
1971         btrfs_put_ordered_extent(ordered_extent);
1972         /* once for the tree */
1973         btrfs_put_ordered_extent(ordered_extent);
1974
1975         return ret;
1976 }
1977
1978 static void finish_ordered_fn(struct btrfs_work *work)
1979 {
1980         struct btrfs_ordered_extent *ordered_extent;
1981         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
1982         btrfs_finish_ordered_io(ordered_extent);
1983 }
1984
1985 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1986                                 struct extent_state *state, int uptodate)
1987 {
1988         struct inode *inode = page->mapping->host;
1989         struct btrfs_root *root = BTRFS_I(inode)->root;
1990         struct btrfs_ordered_extent *ordered_extent = NULL;
1991         struct btrfs_workers *workers;
1992
1993         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1994
1995         ClearPagePrivate2(page);
1996         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1997                                             end - start + 1, uptodate))
1998                 return 0;
1999
2000         ordered_extent->work.func = finish_ordered_fn;
2001         ordered_extent->work.flags = 0;
2002
2003         if (btrfs_is_free_space_inode(inode))
2004                 workers = &root->fs_info->endio_freespace_worker;
2005         else
2006                 workers = &root->fs_info->endio_write_workers;
2007         btrfs_queue_worker(workers, &ordered_extent->work);
2008
2009         return 0;
2010 }
2011
2012 /*
2013  * when reads are done, we need to check csums to verify the data is correct
2014  * if there's a match, we allow the bio to finish.  If not, the code in
2015  * extent_io.c will try to find good copies for us.
2016  */
2017 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2018                                struct extent_state *state, int mirror)
2019 {
2020         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2021         struct inode *inode = page->mapping->host;
2022         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2023         char *kaddr;
2024         u64 private = ~(u32)0;
2025         int ret;
2026         struct btrfs_root *root = BTRFS_I(inode)->root;
2027         u32 csum = ~(u32)0;
2028
2029         if (PageChecked(page)) {
2030                 ClearPageChecked(page);
2031                 goto good;
2032         }
2033
2034         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2035                 goto good;
2036
2037         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2038             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2039                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2040                                   GFP_NOFS);
2041                 return 0;
2042         }
2043
2044         if (state && state->start == start) {
2045                 private = state->private;
2046                 ret = 0;
2047         } else {
2048                 ret = get_state_private(io_tree, start, &private);
2049         }
2050         kaddr = kmap_atomic(page);
2051         if (ret)
2052                 goto zeroit;
2053
2054         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2055         btrfs_csum_final(csum, (char *)&csum);
2056         if (csum != private)
2057                 goto zeroit;
2058
2059         kunmap_atomic(kaddr);
2060 good:
2061         return 0;
2062
2063 zeroit:
2064         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2065                        "private %llu\n",
2066                        (unsigned long long)btrfs_ino(page->mapping->host),
2067                        (unsigned long long)start, csum,
2068                        (unsigned long long)private);
2069         memset(kaddr + offset, 1, end - start + 1);
2070         flush_dcache_page(page);
2071         kunmap_atomic(kaddr);
2072         if (private == 0)
2073                 return 0;
2074         return -EIO;
2075 }
2076
2077 struct delayed_iput {
2078         struct list_head list;
2079         struct inode *inode;
2080 };
2081
2082 /* JDM: If this is fs-wide, why can't we add a pointer to
2083  * btrfs_inode instead and avoid the allocation? */
2084 void btrfs_add_delayed_iput(struct inode *inode)
2085 {
2086         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2087         struct delayed_iput *delayed;
2088
2089         if (atomic_add_unless(&inode->i_count, -1, 1))
2090                 return;
2091
2092         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2093         delayed->inode = inode;
2094
2095         spin_lock(&fs_info->delayed_iput_lock);
2096         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2097         spin_unlock(&fs_info->delayed_iput_lock);
2098 }
2099
2100 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2101 {
2102         LIST_HEAD(list);
2103         struct btrfs_fs_info *fs_info = root->fs_info;
2104         struct delayed_iput *delayed;
2105         int empty;
2106
2107         spin_lock(&fs_info->delayed_iput_lock);
2108         empty = list_empty(&fs_info->delayed_iputs);
2109         spin_unlock(&fs_info->delayed_iput_lock);
2110         if (empty)
2111                 return;
2112
2113         spin_lock(&fs_info->delayed_iput_lock);
2114         list_splice_init(&fs_info->delayed_iputs, &list);
2115         spin_unlock(&fs_info->delayed_iput_lock);
2116
2117         while (!list_empty(&list)) {
2118                 delayed = list_entry(list.next, struct delayed_iput, list);
2119                 list_del(&delayed->list);
2120                 iput(delayed->inode);
2121                 kfree(delayed);
2122         }
2123 }
2124
2125 enum btrfs_orphan_cleanup_state {
2126         ORPHAN_CLEANUP_STARTED  = 1,
2127         ORPHAN_CLEANUP_DONE     = 2,
2128 };
2129
2130 /*
2131  * This is called in transaction commit time. If there are no orphan
2132  * files in the subvolume, it removes orphan item and frees block_rsv
2133  * structure.
2134  */
2135 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2136                               struct btrfs_root *root)
2137 {
2138         struct btrfs_block_rsv *block_rsv;
2139         int ret;
2140
2141         if (atomic_read(&root->orphan_inodes) ||
2142             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2143                 return;
2144
2145         spin_lock(&root->orphan_lock);
2146         if (atomic_read(&root->orphan_inodes)) {
2147                 spin_unlock(&root->orphan_lock);
2148                 return;
2149         }
2150
2151         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2152                 spin_unlock(&root->orphan_lock);
2153                 return;
2154         }
2155
2156         block_rsv = root->orphan_block_rsv;
2157         root->orphan_block_rsv = NULL;
2158         spin_unlock(&root->orphan_lock);
2159
2160         if (root->orphan_item_inserted &&
2161             btrfs_root_refs(&root->root_item) > 0) {
2162                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2163                                             root->root_key.objectid);
2164                 BUG_ON(ret);
2165                 root->orphan_item_inserted = 0;
2166         }
2167
2168         if (block_rsv) {
2169                 WARN_ON(block_rsv->size > 0);
2170                 btrfs_free_block_rsv(root, block_rsv);
2171         }
2172 }
2173
2174 /*
2175  * This creates an orphan entry for the given inode in case something goes
2176  * wrong in the middle of an unlink/truncate.
2177  *
2178  * NOTE: caller of this function should reserve 5 units of metadata for
2179  *       this function.
2180  */
2181 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2182 {
2183         struct btrfs_root *root = BTRFS_I(inode)->root;
2184         struct btrfs_block_rsv *block_rsv = NULL;
2185         int reserve = 0;
2186         int insert = 0;
2187         int ret;
2188
2189         if (!root->orphan_block_rsv) {
2190                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2191                 if (!block_rsv)
2192                         return -ENOMEM;
2193         }
2194
2195         spin_lock(&root->orphan_lock);
2196         if (!root->orphan_block_rsv) {
2197                 root->orphan_block_rsv = block_rsv;
2198         } else if (block_rsv) {
2199                 btrfs_free_block_rsv(root, block_rsv);
2200                 block_rsv = NULL;
2201         }
2202
2203         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2204                               &BTRFS_I(inode)->runtime_flags)) {
2205 #if 0
2206                 /*
2207                  * For proper ENOSPC handling, we should do orphan
2208                  * cleanup when mounting. But this introduces backward
2209                  * compatibility issue.
2210                  */
2211                 if (!xchg(&root->orphan_item_inserted, 1))
2212                         insert = 2;
2213                 else
2214                         insert = 1;
2215 #endif
2216                 insert = 1;
2217                 atomic_inc(&root->orphan_inodes);
2218         }
2219
2220         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2221                               &BTRFS_I(inode)->runtime_flags))
2222                 reserve = 1;
2223         spin_unlock(&root->orphan_lock);
2224
2225         /* grab metadata reservation from transaction handle */
2226         if (reserve) {
2227                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2228                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2229         }
2230
2231         /* insert an orphan item to track this unlinked/truncated file */
2232         if (insert >= 1) {
2233                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2234                 if (ret && ret != -EEXIST) {
2235                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2236                                   &BTRFS_I(inode)->runtime_flags);
2237                         btrfs_abort_transaction(trans, root, ret);
2238                         return ret;
2239                 }
2240                 ret = 0;
2241         }
2242
2243         /* insert an orphan item to track subvolume contains orphan files */
2244         if (insert >= 2) {
2245                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2246                                                root->root_key.objectid);
2247                 if (ret && ret != -EEXIST) {
2248                         btrfs_abort_transaction(trans, root, ret);
2249                         return ret;
2250                 }
2251         }
2252         return 0;
2253 }
2254
2255 /*
2256  * We have done the truncate/delete so we can go ahead and remove the orphan
2257  * item for this particular inode.
2258  */
2259 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2260 {
2261         struct btrfs_root *root = BTRFS_I(inode)->root;
2262         int delete_item = 0;
2263         int release_rsv = 0;
2264         int ret = 0;
2265
2266         spin_lock(&root->orphan_lock);
2267         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2268                                &BTRFS_I(inode)->runtime_flags))
2269                 delete_item = 1;
2270
2271         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2272                                &BTRFS_I(inode)->runtime_flags))
2273                 release_rsv = 1;
2274         spin_unlock(&root->orphan_lock);
2275
2276         if (trans && delete_item) {
2277                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2278                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2279         }
2280
2281         if (release_rsv) {
2282                 btrfs_orphan_release_metadata(inode);
2283                 atomic_dec(&root->orphan_inodes);
2284         }
2285
2286         return 0;
2287 }
2288
2289 /*
2290  * this cleans up any orphans that may be left on the list from the last use
2291  * of this root.
2292  */
2293 int btrfs_orphan_cleanup(struct btrfs_root *root)
2294 {
2295         struct btrfs_path *path;
2296         struct extent_buffer *leaf;
2297         struct btrfs_key key, found_key;
2298         struct btrfs_trans_handle *trans;
2299         struct inode *inode;
2300         u64 last_objectid = 0;
2301         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2302
2303         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2304                 return 0;
2305
2306         path = btrfs_alloc_path();
2307         if (!path) {
2308                 ret = -ENOMEM;
2309                 goto out;
2310         }
2311         path->reada = -1;
2312
2313         key.objectid = BTRFS_ORPHAN_OBJECTID;
2314         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2315         key.offset = (u64)-1;
2316
2317         while (1) {
2318                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2319                 if (ret < 0)
2320                         goto out;
2321
2322                 /*
2323                  * if ret == 0 means we found what we were searching for, which
2324                  * is weird, but possible, so only screw with path if we didn't
2325                  * find the key and see if we have stuff that matches
2326                  */
2327                 if (ret > 0) {
2328                         ret = 0;
2329                         if (path->slots[0] == 0)
2330                                 break;
2331                         path->slots[0]--;
2332                 }
2333
2334                 /* pull out the item */
2335                 leaf = path->nodes[0];
2336                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2337
2338                 /* make sure the item matches what we want */
2339                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2340                         break;
2341                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2342                         break;
2343
2344                 /* release the path since we're done with it */
2345                 btrfs_release_path(path);
2346
2347                 /*
2348                  * this is where we are basically btrfs_lookup, without the
2349                  * crossing root thing.  we store the inode number in the
2350                  * offset of the orphan item.
2351                  */
2352
2353                 if (found_key.offset == last_objectid) {
2354                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2355                                "stopping orphan cleanup\n");
2356                         ret = -EINVAL;
2357                         goto out;
2358                 }
2359
2360                 last_objectid = found_key.offset;
2361
2362                 found_key.objectid = found_key.offset;
2363                 found_key.type = BTRFS_INODE_ITEM_KEY;
2364                 found_key.offset = 0;
2365                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2366                 ret = PTR_RET(inode);
2367                 if (ret && ret != -ESTALE)
2368                         goto out;
2369
2370                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2371                         struct btrfs_root *dead_root;
2372                         struct btrfs_fs_info *fs_info = root->fs_info;
2373                         int is_dead_root = 0;
2374
2375                         /*
2376                          * this is an orphan in the tree root. Currently these
2377                          * could come from 2 sources:
2378                          *  a) a snapshot deletion in progress
2379                          *  b) a free space cache inode
2380                          * We need to distinguish those two, as the snapshot
2381                          * orphan must not get deleted.
2382                          * find_dead_roots already ran before us, so if this
2383                          * is a snapshot deletion, we should find the root
2384                          * in the dead_roots list
2385                          */
2386                         spin_lock(&fs_info->trans_lock);
2387                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2388                                             root_list) {
2389                                 if (dead_root->root_key.objectid ==
2390                                     found_key.objectid) {
2391                                         is_dead_root = 1;
2392                                         break;
2393                                 }
2394                         }
2395                         spin_unlock(&fs_info->trans_lock);
2396                         if (is_dead_root) {
2397                                 /* prevent this orphan from being found again */
2398                                 key.offset = found_key.objectid - 1;
2399                                 continue;
2400                         }
2401                 }
2402                 /*
2403                  * Inode is already gone but the orphan item is still there,
2404                  * kill the orphan item.
2405                  */
2406                 if (ret == -ESTALE) {
2407                         trans = btrfs_start_transaction(root, 1);
2408                         if (IS_ERR(trans)) {
2409                                 ret = PTR_ERR(trans);
2410                                 goto out;
2411                         }
2412                         printk(KERN_ERR "auto deleting %Lu\n",
2413                                found_key.objectid);
2414                         ret = btrfs_del_orphan_item(trans, root,
2415                                                     found_key.objectid);
2416                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2417                         btrfs_end_transaction(trans, root);
2418                         continue;
2419                 }
2420
2421                 /*
2422                  * add this inode to the orphan list so btrfs_orphan_del does
2423                  * the proper thing when we hit it
2424                  */
2425                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2426                         &BTRFS_I(inode)->runtime_flags);
2427
2428                 /* if we have links, this was a truncate, lets do that */
2429                 if (inode->i_nlink) {
2430                         if (!S_ISREG(inode->i_mode)) {
2431                                 WARN_ON(1);
2432                                 iput(inode);
2433                                 continue;
2434                         }
2435                         nr_truncate++;
2436                         ret = btrfs_truncate(inode);
2437                 } else {
2438                         nr_unlink++;
2439                 }
2440
2441                 /* this will do delete_inode and everything for us */
2442                 iput(inode);
2443                 if (ret)
2444                         goto out;
2445         }
2446         /* release the path since we're done with it */
2447         btrfs_release_path(path);
2448
2449         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2450
2451         if (root->orphan_block_rsv)
2452                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2453                                         (u64)-1);
2454
2455         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2456                 trans = btrfs_join_transaction(root);
2457                 if (!IS_ERR(trans))
2458                         btrfs_end_transaction(trans, root);
2459         }
2460
2461         if (nr_unlink)
2462                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2463         if (nr_truncate)
2464                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2465
2466 out:
2467         if (ret)
2468                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2469         btrfs_free_path(path);
2470         return ret;
2471 }
2472
2473 /*
2474  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2475  * don't find any xattrs, we know there can't be any acls.
2476  *
2477  * slot is the slot the inode is in, objectid is the objectid of the inode
2478  */
2479 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2480                                           int slot, u64 objectid)
2481 {
2482         u32 nritems = btrfs_header_nritems(leaf);
2483         struct btrfs_key found_key;
2484         int scanned = 0;
2485
2486         slot++;
2487         while (slot < nritems) {
2488                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2489
2490                 /* we found a different objectid, there must not be acls */
2491                 if (found_key.objectid != objectid)
2492                         return 0;
2493
2494                 /* we found an xattr, assume we've got an acl */
2495                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2496                         return 1;
2497
2498                 /*
2499                  * we found a key greater than an xattr key, there can't
2500                  * be any acls later on
2501                  */
2502                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2503                         return 0;
2504
2505                 slot++;
2506                 scanned++;
2507
2508                 /*
2509                  * it goes inode, inode backrefs, xattrs, extents,
2510                  * so if there are a ton of hard links to an inode there can
2511                  * be a lot of backrefs.  Don't waste time searching too hard,
2512                  * this is just an optimization
2513                  */
2514                 if (scanned >= 8)
2515                         break;
2516         }
2517         /* we hit the end of the leaf before we found an xattr or
2518          * something larger than an xattr.  We have to assume the inode
2519          * has acls
2520          */
2521         return 1;
2522 }
2523
2524 /*
2525  * read an inode from the btree into the in-memory inode
2526  */
2527 static void btrfs_read_locked_inode(struct inode *inode)
2528 {
2529         struct btrfs_path *path;
2530         struct extent_buffer *leaf;
2531         struct btrfs_inode_item *inode_item;
2532         struct btrfs_timespec *tspec;
2533         struct btrfs_root *root = BTRFS_I(inode)->root;
2534         struct btrfs_key location;
2535         int maybe_acls;
2536         u32 rdev;
2537         int ret;
2538         bool filled = false;
2539
2540         ret = btrfs_fill_inode(inode, &rdev);
2541         if (!ret)
2542                 filled = true;
2543
2544         path = btrfs_alloc_path();
2545         if (!path)
2546                 goto make_bad;
2547
2548         path->leave_spinning = 1;
2549         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2550
2551         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2552         if (ret)
2553                 goto make_bad;
2554
2555         leaf = path->nodes[0];
2556
2557         if (filled)
2558                 goto cache_acl;
2559
2560         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2561                                     struct btrfs_inode_item);
2562         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2563         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2564         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2565         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2566         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2567
2568         tspec = btrfs_inode_atime(inode_item);
2569         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2570         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2571
2572         tspec = btrfs_inode_mtime(inode_item);
2573         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2574         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2575
2576         tspec = btrfs_inode_ctime(inode_item);
2577         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2578         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2579
2580         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2581         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2582         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2583
2584         /*
2585          * If we were modified in the current generation and evicted from memory
2586          * and then re-read we need to do a full sync since we don't have any
2587          * idea about which extents were modified before we were evicted from
2588          * cache.
2589          */
2590         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2591                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2592                         &BTRFS_I(inode)->runtime_flags);
2593
2594         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2595         inode->i_generation = BTRFS_I(inode)->generation;
2596         inode->i_rdev = 0;
2597         rdev = btrfs_inode_rdev(leaf, inode_item);
2598
2599         BTRFS_I(inode)->index_cnt = (u64)-1;
2600         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2601 cache_acl:
2602         /*
2603          * try to precache a NULL acl entry for files that don't have
2604          * any xattrs or acls
2605          */
2606         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2607                                            btrfs_ino(inode));
2608         if (!maybe_acls)
2609                 cache_no_acl(inode);
2610
2611         btrfs_free_path(path);
2612
2613         switch (inode->i_mode & S_IFMT) {
2614         case S_IFREG:
2615                 inode->i_mapping->a_ops = &btrfs_aops;
2616                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2617                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2618                 inode->i_fop = &btrfs_file_operations;
2619                 inode->i_op = &btrfs_file_inode_operations;
2620                 break;
2621         case S_IFDIR:
2622                 inode->i_fop = &btrfs_dir_file_operations;
2623                 if (root == root->fs_info->tree_root)
2624                         inode->i_op = &btrfs_dir_ro_inode_operations;
2625                 else
2626                         inode->i_op = &btrfs_dir_inode_operations;
2627                 break;
2628         case S_IFLNK:
2629                 inode->i_op = &btrfs_symlink_inode_operations;
2630                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2631                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2632                 break;
2633         default:
2634                 inode->i_op = &btrfs_special_inode_operations;
2635                 init_special_inode(inode, inode->i_mode, rdev);
2636                 break;
2637         }
2638
2639         btrfs_update_iflags(inode);
2640         return;
2641
2642 make_bad:
2643         btrfs_free_path(path);
2644         make_bad_inode(inode);
2645 }
2646
2647 /*
2648  * given a leaf and an inode, copy the inode fields into the leaf
2649  */
2650 static void fill_inode_item(struct btrfs_trans_handle *trans,
2651                             struct extent_buffer *leaf,
2652                             struct btrfs_inode_item *item,
2653                             struct inode *inode)
2654 {
2655         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2656         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2657         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2658         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2659         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2660
2661         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2662                                inode->i_atime.tv_sec);
2663         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2664                                 inode->i_atime.tv_nsec);
2665
2666         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2667                                inode->i_mtime.tv_sec);
2668         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2669                                 inode->i_mtime.tv_nsec);
2670
2671         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2672                                inode->i_ctime.tv_sec);
2673         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2674                                 inode->i_ctime.tv_nsec);
2675
2676         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2677         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2678         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2679         btrfs_set_inode_transid(leaf, item, trans->transid);
2680         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2681         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2682         btrfs_set_inode_block_group(leaf, item, 0);
2683 }
2684
2685 /*
2686  * copy everything in the in-memory inode into the btree.
2687  */
2688 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2689                                 struct btrfs_root *root, struct inode *inode)
2690 {
2691         struct btrfs_inode_item *inode_item;
2692         struct btrfs_path *path;
2693         struct extent_buffer *leaf;
2694         int ret;
2695
2696         path = btrfs_alloc_path();
2697         if (!path)
2698                 return -ENOMEM;
2699
2700         path->leave_spinning = 1;
2701         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2702                                  1);
2703         if (ret) {
2704                 if (ret > 0)
2705                         ret = -ENOENT;
2706                 goto failed;
2707         }
2708
2709         btrfs_unlock_up_safe(path, 1);
2710         leaf = path->nodes[0];
2711         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2712                                     struct btrfs_inode_item);
2713
2714         fill_inode_item(trans, leaf, inode_item, inode);
2715         btrfs_mark_buffer_dirty(leaf);
2716         btrfs_set_inode_last_trans(trans, inode);
2717         ret = 0;
2718 failed:
2719         btrfs_free_path(path);
2720         return ret;
2721 }
2722
2723 /*
2724  * copy everything in the in-memory inode into the btree.
2725  */
2726 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2727                                 struct btrfs_root *root, struct inode *inode)
2728 {
2729         int ret;
2730
2731         /*
2732          * If the inode is a free space inode, we can deadlock during commit
2733          * if we put it into the delayed code.
2734          *
2735          * The data relocation inode should also be directly updated
2736          * without delay
2737          */
2738         if (!btrfs_is_free_space_inode(inode)
2739             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2740                 btrfs_update_root_times(trans, root);
2741
2742                 ret = btrfs_delayed_update_inode(trans, root, inode);
2743                 if (!ret)
2744                         btrfs_set_inode_last_trans(trans, inode);
2745                 return ret;
2746         }
2747
2748         return btrfs_update_inode_item(trans, root, inode);
2749 }
2750
2751 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2752                                 struct btrfs_root *root, struct inode *inode)
2753 {
2754         int ret;
2755
2756         ret = btrfs_update_inode(trans, root, inode);
2757         if (ret == -ENOSPC)
2758                 return btrfs_update_inode_item(trans, root, inode);
2759         return ret;
2760 }
2761
2762 /*
2763  * unlink helper that gets used here in inode.c and in the tree logging
2764  * recovery code.  It remove a link in a directory with a given name, and
2765  * also drops the back refs in the inode to the directory
2766  */
2767 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2768                                 struct btrfs_root *root,
2769                                 struct inode *dir, struct inode *inode,
2770                                 const char *name, int name_len)
2771 {
2772         struct btrfs_path *path;
2773         int ret = 0;
2774         struct extent_buffer *leaf;
2775         struct btrfs_dir_item *di;
2776         struct btrfs_key key;
2777         u64 index;
2778         u64 ino = btrfs_ino(inode);
2779         u64 dir_ino = btrfs_ino(dir);
2780
2781         path = btrfs_alloc_path();
2782         if (!path) {
2783                 ret = -ENOMEM;
2784                 goto out;
2785         }
2786
2787         path->leave_spinning = 1;
2788         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2789                                     name, name_len, -1);
2790         if (IS_ERR(di)) {
2791                 ret = PTR_ERR(di);
2792                 goto err;
2793         }
2794         if (!di) {
2795                 ret = -ENOENT;
2796                 goto err;
2797         }
2798         leaf = path->nodes[0];
2799         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2800         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2801         if (ret)
2802                 goto err;
2803         btrfs_release_path(path);
2804
2805         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2806                                   dir_ino, &index);
2807         if (ret) {
2808                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2809                        "inode %llu parent %llu\n", name_len, name,
2810                        (unsigned long long)ino, (unsigned long long)dir_ino);
2811                 btrfs_abort_transaction(trans, root, ret);
2812                 goto err;
2813         }
2814
2815         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2816         if (ret) {
2817                 btrfs_abort_transaction(trans, root, ret);
2818                 goto err;
2819         }
2820
2821         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2822                                          inode, dir_ino);
2823         if (ret != 0 && ret != -ENOENT) {
2824                 btrfs_abort_transaction(trans, root, ret);
2825                 goto err;
2826         }
2827
2828         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2829                                            dir, index);
2830         if (ret == -ENOENT)
2831                 ret = 0;
2832 err:
2833         btrfs_free_path(path);
2834         if (ret)
2835                 goto out;
2836
2837         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2838         inode_inc_iversion(inode);
2839         inode_inc_iversion(dir);
2840         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2841         ret = btrfs_update_inode(trans, root, dir);
2842 out:
2843         return ret;
2844 }
2845
2846 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2847                        struct btrfs_root *root,
2848                        struct inode *dir, struct inode *inode,
2849                        const char *name, int name_len)
2850 {
2851         int ret;
2852         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2853         if (!ret) {
2854                 btrfs_drop_nlink(inode);
2855                 ret = btrfs_update_inode(trans, root, inode);
2856         }
2857         return ret;
2858 }
2859
2860
2861 /* helper to check if there is any shared block in the path */
2862 static int check_path_shared(struct btrfs_root *root,
2863                              struct btrfs_path *path)
2864 {
2865         struct extent_buffer *eb;
2866         int level;
2867         u64 refs = 1;
2868
2869         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2870                 int ret;
2871
2872                 if (!path->nodes[level])
2873                         break;
2874                 eb = path->nodes[level];
2875                 if (!btrfs_block_can_be_shared(root, eb))
2876                         continue;
2877                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2878                                                &refs, NULL);
2879                 if (refs > 1)
2880                         return 1;
2881         }
2882         return 0;
2883 }
2884
2885 /*
2886  * helper to start transaction for unlink and rmdir.
2887  *
2888  * unlink and rmdir are special in btrfs, they do not always free space.
2889  * so in enospc case, we should make sure they will free space before
2890  * allowing them to use the global metadata reservation.
2891  */
2892 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2893                                                        struct dentry *dentry)
2894 {
2895         struct btrfs_trans_handle *trans;
2896         struct btrfs_root *root = BTRFS_I(dir)->root;
2897         struct btrfs_path *path;
2898         struct btrfs_dir_item *di;
2899         struct inode *inode = dentry->d_inode;
2900         u64 index;
2901         int check_link = 1;
2902         int err = -ENOSPC;
2903         int ret;
2904         u64 ino = btrfs_ino(inode);
2905         u64 dir_ino = btrfs_ino(dir);
2906
2907         /*
2908          * 1 for the possible orphan item
2909          * 1 for the dir item
2910          * 1 for the dir index
2911          * 1 for the inode ref
2912          * 1 for the inode ref in the tree log
2913          * 2 for the dir entries in the log
2914          * 1 for the inode
2915          */
2916         trans = btrfs_start_transaction(root, 8);
2917         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2918                 return trans;
2919
2920         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2921                 return ERR_PTR(-ENOSPC);
2922
2923         /* check if there is someone else holds reference */
2924         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2925                 return ERR_PTR(-ENOSPC);
2926
2927         if (atomic_read(&inode->i_count) > 2)
2928                 return ERR_PTR(-ENOSPC);
2929
2930         if (xchg(&root->fs_info->enospc_unlink, 1))
2931                 return ERR_PTR(-ENOSPC);
2932
2933         path = btrfs_alloc_path();
2934         if (!path) {
2935                 root->fs_info->enospc_unlink = 0;
2936                 return ERR_PTR(-ENOMEM);
2937         }
2938
2939         /* 1 for the orphan item */
2940         trans = btrfs_start_transaction(root, 1);
2941         if (IS_ERR(trans)) {
2942                 btrfs_free_path(path);
2943                 root->fs_info->enospc_unlink = 0;
2944                 return trans;
2945         }
2946
2947         path->skip_locking = 1;
2948         path->search_commit_root = 1;
2949
2950         ret = btrfs_lookup_inode(trans, root, path,
2951                                 &BTRFS_I(dir)->location, 0);
2952         if (ret < 0) {
2953                 err = ret;
2954                 goto out;
2955         }
2956         if (ret == 0) {
2957                 if (check_path_shared(root, path))
2958                         goto out;
2959         } else {
2960                 check_link = 0;
2961         }
2962         btrfs_release_path(path);
2963
2964         ret = btrfs_lookup_inode(trans, root, path,
2965                                 &BTRFS_I(inode)->location, 0);
2966         if (ret < 0) {
2967                 err = ret;
2968                 goto out;
2969         }
2970         if (ret == 0) {
2971                 if (check_path_shared(root, path))
2972                         goto out;
2973         } else {
2974                 check_link = 0;
2975         }
2976         btrfs_release_path(path);
2977
2978         if (ret == 0 && S_ISREG(inode->i_mode)) {
2979                 ret = btrfs_lookup_file_extent(trans, root, path,
2980                                                ino, (u64)-1, 0);
2981                 if (ret < 0) {
2982                         err = ret;
2983                         goto out;
2984                 }
2985                 BUG_ON(ret == 0); /* Corruption */
2986                 if (check_path_shared(root, path))
2987                         goto out;
2988                 btrfs_release_path(path);
2989         }
2990
2991         if (!check_link) {
2992                 err = 0;
2993                 goto out;
2994         }
2995
2996         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2997                                 dentry->d_name.name, dentry->d_name.len, 0);
2998         if (IS_ERR(di)) {
2999                 err = PTR_ERR(di);
3000                 goto out;
3001         }
3002         if (di) {
3003                 if (check_path_shared(root, path))
3004                         goto out;
3005         } else {
3006                 err = 0;
3007                 goto out;
3008         }
3009         btrfs_release_path(path);
3010
3011         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3012                                         dentry->d_name.len, ino, dir_ino, 0,
3013                                         &index);
3014         if (ret) {
3015                 err = ret;
3016                 goto out;
3017         }
3018
3019         if (check_path_shared(root, path))
3020                 goto out;
3021
3022         btrfs_release_path(path);
3023
3024         /*
3025          * This is a commit root search, if we can lookup inode item and other
3026          * relative items in the commit root, it means the transaction of
3027          * dir/file creation has been committed, and the dir index item that we
3028          * delay to insert has also been inserted into the commit root. So
3029          * we needn't worry about the delayed insertion of the dir index item
3030          * here.
3031          */
3032         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3033                                 dentry->d_name.name, dentry->d_name.len, 0);
3034         if (IS_ERR(di)) {
3035                 err = PTR_ERR(di);
3036                 goto out;
3037         }
3038         BUG_ON(ret == -ENOENT);
3039         if (check_path_shared(root, path))
3040                 goto out;
3041
3042         err = 0;
3043 out:
3044         btrfs_free_path(path);
3045         /* Migrate the orphan reservation over */
3046         if (!err)
3047                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3048                                 &root->fs_info->global_block_rsv,
3049                                 trans->bytes_reserved);
3050
3051         if (err) {
3052                 btrfs_end_transaction(trans, root);
3053                 root->fs_info->enospc_unlink = 0;
3054                 return ERR_PTR(err);
3055         }
3056
3057         trans->block_rsv = &root->fs_info->global_block_rsv;
3058         return trans;
3059 }
3060
3061 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3062                                struct btrfs_root *root)
3063 {
3064         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3065                 btrfs_block_rsv_release(root, trans->block_rsv,
3066                                         trans->bytes_reserved);
3067                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3068                 BUG_ON(!root->fs_info->enospc_unlink);
3069                 root->fs_info->enospc_unlink = 0;
3070         }
3071         btrfs_end_transaction(trans, root);
3072 }
3073
3074 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3075 {
3076         struct btrfs_root *root = BTRFS_I(dir)->root;
3077         struct btrfs_trans_handle *trans;
3078         struct inode *inode = dentry->d_inode;
3079         int ret;
3080         unsigned long nr = 0;
3081
3082         trans = __unlink_start_trans(dir, dentry);
3083         if (IS_ERR(trans))
3084                 return PTR_ERR(trans);
3085
3086         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3087
3088         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3089                                  dentry->d_name.name, dentry->d_name.len);
3090         if (ret)
3091                 goto out;
3092
3093         if (inode->i_nlink == 0) {
3094                 ret = btrfs_orphan_add(trans, inode);
3095                 if (ret)
3096                         goto out;
3097         }
3098
3099 out:
3100         nr = trans->blocks_used;
3101         __unlink_end_trans(trans, root);
3102         btrfs_btree_balance_dirty(root, nr);
3103         return ret;
3104 }
3105
3106 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3107                         struct btrfs_root *root,
3108                         struct inode *dir, u64 objectid,
3109                         const char *name, int name_len)
3110 {
3111         struct btrfs_path *path;
3112         struct extent_buffer *leaf;
3113         struct btrfs_dir_item *di;
3114         struct btrfs_key key;
3115         u64 index;
3116         int ret;
3117         u64 dir_ino = btrfs_ino(dir);
3118
3119         path = btrfs_alloc_path();
3120         if (!path)
3121                 return -ENOMEM;
3122
3123         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3124                                    name, name_len, -1);
3125         if (IS_ERR_OR_NULL(di)) {
3126                 if (!di)
3127                         ret = -ENOENT;
3128                 else
3129                         ret = PTR_ERR(di);
3130                 goto out;
3131         }
3132
3133         leaf = path->nodes[0];
3134         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3135         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3136         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3137         if (ret) {
3138                 btrfs_abort_transaction(trans, root, ret);
3139                 goto out;
3140         }
3141         btrfs_release_path(path);
3142
3143         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3144                                  objectid, root->root_key.objectid,
3145                                  dir_ino, &index, name, name_len);
3146         if (ret < 0) {
3147                 if (ret != -ENOENT) {
3148                         btrfs_abort_transaction(trans, root, ret);
3149                         goto out;
3150                 }
3151                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3152                                                  name, name_len);
3153                 if (IS_ERR_OR_NULL(di)) {
3154                         if (!di)
3155                                 ret = -ENOENT;
3156                         else
3157                                 ret = PTR_ERR(di);
3158                         btrfs_abort_transaction(trans, root, ret);
3159                         goto out;
3160                 }
3161
3162                 leaf = path->nodes[0];
3163                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3164                 btrfs_release_path(path);
3165                 index = key.offset;
3166         }
3167         btrfs_release_path(path);
3168
3169         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3170         if (ret) {
3171                 btrfs_abort_transaction(trans, root, ret);
3172                 goto out;
3173         }
3174
3175         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3176         inode_inc_iversion(dir);
3177         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3178         ret = btrfs_update_inode_fallback(trans, root, dir);
3179         if (ret)
3180                 btrfs_abort_transaction(trans, root, ret);
3181 out:
3182         btrfs_free_path(path);
3183         return ret;
3184 }
3185
3186 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3187 {
3188         struct inode *inode = dentry->d_inode;
3189         int err = 0;
3190         struct btrfs_root *root = BTRFS_I(dir)->root;
3191         struct btrfs_trans_handle *trans;
3192         unsigned long nr = 0;
3193
3194         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3195                 return -ENOTEMPTY;
3196         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3197                 return -EPERM;
3198
3199         trans = __unlink_start_trans(dir, dentry);
3200         if (IS_ERR(trans))
3201                 return PTR_ERR(trans);
3202
3203         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3204                 err = btrfs_unlink_subvol(trans, root, dir,
3205                                           BTRFS_I(inode)->location.objectid,
3206                                           dentry->d_name.name,
3207                                           dentry->d_name.len);
3208                 goto out;
3209         }
3210
3211         err = btrfs_orphan_add(trans, inode);
3212         if (err)
3213                 goto out;
3214
3215         /* now the directory is empty */
3216         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3217                                  dentry->d_name.name, dentry->d_name.len);
3218         if (!err)
3219                 btrfs_i_size_write(inode, 0);
3220 out:
3221         nr = trans->blocks_used;
3222         __unlink_end_trans(trans, root);
3223         btrfs_btree_balance_dirty(root, nr);
3224
3225         return err;
3226 }
3227
3228 /*
3229  * this can truncate away extent items, csum items and directory items.
3230  * It starts at a high offset and removes keys until it can't find
3231  * any higher than new_size
3232  *
3233  * csum items that cross the new i_size are truncated to the new size
3234  * as well.
3235  *
3236  * min_type is the minimum key type to truncate down to.  If set to 0, this
3237  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3238  */
3239 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3240                                struct btrfs_root *root,
3241                                struct inode *inode,
3242                                u64 new_size, u32 min_type)
3243 {
3244         struct btrfs_path *path;
3245         struct extent_buffer *leaf;
3246         struct btrfs_file_extent_item *fi;
3247         struct btrfs_key key;
3248         struct btrfs_key found_key;
3249         u64 extent_start = 0;
3250         u64 extent_num_bytes = 0;
3251         u64 extent_offset = 0;
3252         u64 item_end = 0;
3253         u64 mask = root->sectorsize - 1;
3254         u32 found_type = (u8)-1;
3255         int found_extent;
3256         int del_item;
3257         int pending_del_nr = 0;
3258         int pending_del_slot = 0;
3259         int extent_type = -1;
3260         int ret;
3261         int err = 0;
3262         u64 ino = btrfs_ino(inode);
3263
3264         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3265
3266         path = btrfs_alloc_path();
3267         if (!path)
3268                 return -ENOMEM;
3269         path->reada = -1;
3270
3271         /*
3272          * We want to drop from the next block forward in case this new size is
3273          * not block aligned since we will be keeping the last block of the
3274          * extent just the way it is.
3275          */
3276         if (root->ref_cows || root == root->fs_info->tree_root)
3277                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3278
3279         /*
3280          * This function is also used to drop the items in the log tree before
3281          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3282          * it is used to drop the loged items. So we shouldn't kill the delayed
3283          * items.
3284          */
3285         if (min_type == 0 && root == BTRFS_I(inode)->root)
3286                 btrfs_kill_delayed_inode_items(inode);
3287
3288         key.objectid = ino;
3289         key.offset = (u64)-1;
3290         key.type = (u8)-1;
3291
3292 search_again:
3293         path->leave_spinning = 1;
3294         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3295         if (ret < 0) {
3296                 err = ret;
3297                 goto out;
3298         }
3299
3300         if (ret > 0) {
3301                 /* there are no items in the tree for us to truncate, we're
3302                  * done
3303                  */
3304                 if (path->slots[0] == 0)
3305                         goto out;
3306                 path->slots[0]--;
3307         }
3308
3309         while (1) {
3310                 fi = NULL;
3311                 leaf = path->nodes[0];
3312                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3313                 found_type = btrfs_key_type(&found_key);
3314
3315                 if (found_key.objectid != ino)
3316                         break;
3317
3318                 if (found_type < min_type)
3319                         break;
3320
3321                 item_end = found_key.offset;
3322                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3323                         fi = btrfs_item_ptr(leaf, path->slots[0],
3324                                             struct btrfs_file_extent_item);
3325                         extent_type = btrfs_file_extent_type(leaf, fi);
3326                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3327                                 item_end +=
3328                                     btrfs_file_extent_num_bytes(leaf, fi);
3329                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3330                                 item_end += btrfs_file_extent_inline_len(leaf,
3331                                                                          fi);
3332                         }
3333                         item_end--;
3334                 }
3335                 if (found_type > min_type) {
3336                         del_item = 1;
3337                 } else {
3338                         if (item_end < new_size)
3339                                 break;
3340                         if (found_key.offset >= new_size)
3341                                 del_item = 1;
3342                         else
3343                                 del_item = 0;
3344                 }
3345                 found_extent = 0;
3346                 /* FIXME, shrink the extent if the ref count is only 1 */
3347                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3348                         goto delete;
3349
3350                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3351                         u64 num_dec;
3352                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3353                         if (!del_item) {
3354                                 u64 orig_num_bytes =
3355                                         btrfs_file_extent_num_bytes(leaf, fi);
3356                                 extent_num_bytes = new_size -
3357                                         found_key.offset + root->sectorsize - 1;
3358                                 extent_num_bytes = extent_num_bytes &
3359                                         ~((u64)root->sectorsize - 1);
3360                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3361                                                          extent_num_bytes);
3362                                 num_dec = (orig_num_bytes -
3363                                            extent_num_bytes);
3364                                 if (root->ref_cows && extent_start != 0)
3365                                         inode_sub_bytes(inode, num_dec);
3366                                 btrfs_mark_buffer_dirty(leaf);
3367                         } else {
3368                                 extent_num_bytes =
3369                                         btrfs_file_extent_disk_num_bytes(leaf,
3370                                                                          fi);
3371                                 extent_offset = found_key.offset -
3372                                         btrfs_file_extent_offset(leaf, fi);
3373
3374                                 /* FIXME blocksize != 4096 */
3375                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3376                                 if (extent_start != 0) {
3377                                         found_extent = 1;
3378                                         if (root->ref_cows)
3379                                                 inode_sub_bytes(inode, num_dec);
3380                                 }
3381                         }
3382                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3383                         /*
3384                          * we can't truncate inline items that have had
3385                          * special encodings
3386                          */
3387                         if (!del_item &&
3388                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3389                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3390                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3391                                 u32 size = new_size - found_key.offset;
3392
3393                                 if (root->ref_cows) {
3394                                         inode_sub_bytes(inode, item_end + 1 -
3395                                                         new_size);
3396                                 }
3397                                 size =
3398                                     btrfs_file_extent_calc_inline_size(size);
3399                                 btrfs_truncate_item(trans, root, path,
3400                                                     size, 1);
3401                         } else if (root->ref_cows) {
3402                                 inode_sub_bytes(inode, item_end + 1 -
3403                                                 found_key.offset);
3404                         }
3405                 }
3406 delete:
3407                 if (del_item) {
3408                         if (!pending_del_nr) {
3409                                 /* no pending yet, add ourselves */
3410                                 pending_del_slot = path->slots[0];
3411                                 pending_del_nr = 1;
3412                         } else if (pending_del_nr &&
3413                                    path->slots[0] + 1 == pending_del_slot) {
3414                                 /* hop on the pending chunk */
3415                                 pending_del_nr++;
3416                                 pending_del_slot = path->slots[0];
3417                         } else {
3418                                 BUG();
3419                         }
3420                 } else {
3421                         break;
3422                 }
3423                 if (found_extent && (root->ref_cows ||
3424                                      root == root->fs_info->tree_root)) {
3425                         btrfs_set_path_blocking(path);
3426                         ret = btrfs_free_extent(trans, root, extent_start,
3427                                                 extent_num_bytes, 0,
3428                                                 btrfs_header_owner(leaf),
3429                                                 ino, extent_offset, 0);
3430                         BUG_ON(ret);
3431                 }
3432
3433                 if (found_type == BTRFS_INODE_ITEM_KEY)
3434                         break;
3435
3436                 if (path->slots[0] == 0 ||
3437                     path->slots[0] != pending_del_slot) {
3438                         if (pending_del_nr) {
3439                                 ret = btrfs_del_items(trans, root, path,
3440                                                 pending_del_slot,
3441                                                 pending_del_nr);
3442                                 if (ret) {
3443                                         btrfs_abort_transaction(trans,
3444                                                                 root, ret);
3445                                         goto error;
3446                                 }
3447                                 pending_del_nr = 0;
3448                         }
3449                         btrfs_release_path(path);
3450                         goto search_again;
3451                 } else {
3452                         path->slots[0]--;
3453                 }
3454         }
3455 out:
3456         if (pending_del_nr) {
3457                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3458                                       pending_del_nr);
3459                 if (ret)
3460                         btrfs_abort_transaction(trans, root, ret);
3461         }
3462 error:
3463         btrfs_free_path(path);
3464         return err;
3465 }
3466
3467 /*
3468  * btrfs_truncate_page - read, zero a chunk and write a page
3469  * @inode - inode that we're zeroing
3470  * @from - the offset to start zeroing
3471  * @len - the length to zero, 0 to zero the entire range respective to the
3472  *      offset
3473  * @front - zero up to the offset instead of from the offset on
3474  *
3475  * This will find the page for the "from" offset and cow the page and zero the
3476  * part we want to zero.  This is used with truncate and hole punching.
3477  */
3478 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3479                         int front)
3480 {
3481         struct address_space *mapping = inode->i_mapping;
3482         struct btrfs_root *root = BTRFS_I(inode)->root;
3483         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3484         struct btrfs_ordered_extent *ordered;
3485         struct extent_state *cached_state = NULL;
3486         char *kaddr;
3487         u32 blocksize = root->sectorsize;
3488         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3489         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3490         struct page *page;
3491         gfp_t mask = btrfs_alloc_write_mask(mapping);
3492         int ret = 0;
3493         u64 page_start;
3494         u64 page_end;
3495
3496         if ((offset & (blocksize - 1)) == 0 &&
3497             (!len || ((len & (blocksize - 1)) == 0)))
3498                 goto out;
3499         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3500         if (ret)
3501                 goto out;
3502
3503         ret = -ENOMEM;
3504 again:
3505         page = find_or_create_page(mapping, index, mask);
3506         if (!page) {
3507                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3508                 goto out;
3509         }
3510
3511         page_start = page_offset(page);
3512         page_end = page_start + PAGE_CACHE_SIZE - 1;
3513
3514         if (!PageUptodate(page)) {
3515                 ret = btrfs_readpage(NULL, page);
3516                 lock_page(page);
3517                 if (page->mapping != mapping) {
3518                         unlock_page(page);
3519                         page_cache_release(page);
3520                         goto again;
3521                 }
3522                 if (!PageUptodate(page)) {
3523                         ret = -EIO;
3524                         goto out_unlock;
3525                 }
3526         }
3527         wait_on_page_writeback(page);
3528
3529         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3530         set_page_extent_mapped(page);
3531
3532         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3533         if (ordered) {
3534                 unlock_extent_cached(io_tree, page_start, page_end,
3535                                      &cached_state, GFP_NOFS);
3536                 unlock_page(page);
3537                 page_cache_release(page);
3538                 btrfs_start_ordered_extent(inode, ordered, 1);
3539                 btrfs_put_ordered_extent(ordered);
3540                 goto again;
3541         }
3542
3543         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3544                           EXTENT_DIRTY | EXTENT_DELALLOC |
3545                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3546                           0, 0, &cached_state, GFP_NOFS);
3547
3548         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3549                                         &cached_state);
3550         if (ret) {
3551                 unlock_extent_cached(io_tree, page_start, page_end,
3552                                      &cached_state, GFP_NOFS);
3553                 goto out_unlock;
3554         }
3555
3556         ret = 0;
3557         if (offset != PAGE_CACHE_SIZE) {
3558                 if (!len)
3559                         len = PAGE_CACHE_SIZE - offset;
3560                 kaddr = kmap(page);
3561                 if (front)
3562                         memset(kaddr, 0, offset);
3563                 else
3564                         memset(kaddr + offset, 0, len);
3565                 flush_dcache_page(page);
3566                 kunmap(page);
3567         }
3568         ClearPageChecked(page);
3569         set_page_dirty(page);
3570         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3571                              GFP_NOFS);
3572
3573 out_unlock:
3574         if (ret)
3575                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3576         unlock_page(page);
3577         page_cache_release(page);
3578 out:
3579         return ret;
3580 }
3581
3582 /*
3583  * This function puts in dummy file extents for the area we're creating a hole
3584  * for.  So if we are truncating this file to a larger size we need to insert
3585  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3586  * the range between oldsize and size
3587  */
3588 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3589 {
3590         struct btrfs_trans_handle *trans;
3591         struct btrfs_root *root = BTRFS_I(inode)->root;
3592         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3593         struct extent_map *em = NULL;
3594         struct extent_state *cached_state = NULL;
3595         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3596         u64 mask = root->sectorsize - 1;
3597         u64 hole_start = (oldsize + mask) & ~mask;
3598         u64 block_end = (size + mask) & ~mask;
3599         u64 last_byte;
3600         u64 cur_offset;
3601         u64 hole_size;
3602         int err = 0;
3603
3604         if (size <= hole_start)
3605                 return 0;
3606
3607         while (1) {
3608                 struct btrfs_ordered_extent *ordered;
3609                 btrfs_wait_ordered_range(inode, hole_start,
3610                                          block_end - hole_start);
3611                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3612                                  &cached_state);
3613                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3614                 if (!ordered)
3615                         break;
3616                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3617                                      &cached_state, GFP_NOFS);
3618                 btrfs_put_ordered_extent(ordered);
3619         }
3620
3621         cur_offset = hole_start;
3622         while (1) {
3623                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3624                                 block_end - cur_offset, 0);
3625                 if (IS_ERR(em)) {
3626                         err = PTR_ERR(em);
3627                         break;
3628                 }
3629                 last_byte = min(extent_map_end(em), block_end);
3630                 last_byte = (last_byte + mask) & ~mask;
3631                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3632                         struct extent_map *hole_em;
3633                         hole_size = last_byte - cur_offset;
3634
3635                         trans = btrfs_start_transaction(root, 3);
3636                         if (IS_ERR(trans)) {
3637                                 err = PTR_ERR(trans);
3638                                 break;
3639                         }
3640
3641                         err = btrfs_drop_extents(trans, root, inode,
3642                                                  cur_offset,
3643                                                  cur_offset + hole_size, 1);
3644                         if (err) {
3645                                 btrfs_abort_transaction(trans, root, err);
3646                                 btrfs_end_transaction(trans, root);
3647                                 break;
3648                         }
3649
3650                         err = btrfs_insert_file_extent(trans, root,
3651                                         btrfs_ino(inode), cur_offset, 0,
3652                                         0, hole_size, 0, hole_size,
3653                                         0, 0, 0);
3654                         if (err) {
3655                                 btrfs_abort_transaction(trans, root, err);
3656                                 btrfs_end_transaction(trans, root);
3657                                 break;
3658                         }
3659
3660                         btrfs_drop_extent_cache(inode, cur_offset,
3661                                                 cur_offset + hole_size - 1, 0);
3662                         hole_em = alloc_extent_map();
3663                         if (!hole_em) {
3664                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3665                                         &BTRFS_I(inode)->runtime_flags);
3666                                 goto next;
3667                         }
3668                         hole_em->start = cur_offset;
3669                         hole_em->len = hole_size;
3670                         hole_em->orig_start = cur_offset;
3671
3672                         hole_em->block_start = EXTENT_MAP_HOLE;
3673                         hole_em->block_len = 0;
3674                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3675                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3676                         hole_em->generation = trans->transid;
3677
3678                         while (1) {
3679                                 write_lock(&em_tree->lock);
3680                                 err = add_extent_mapping(em_tree, hole_em);
3681                                 if (!err)
3682                                         list_move(&hole_em->list,
3683                                                   &em_tree->modified_extents);
3684                                 write_unlock(&em_tree->lock);
3685                                 if (err != -EEXIST)
3686                                         break;
3687                                 btrfs_drop_extent_cache(inode, cur_offset,
3688                                                         cur_offset +
3689                                                         hole_size - 1, 0);
3690                         }
3691                         free_extent_map(hole_em);
3692 next:
3693                         btrfs_update_inode(trans, root, inode);
3694                         btrfs_end_transaction(trans, root);
3695                 }
3696                 free_extent_map(em);
3697                 em = NULL;
3698                 cur_offset = last_byte;
3699                 if (cur_offset >= block_end)
3700                         break;
3701         }
3702
3703         free_extent_map(em);
3704         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3705                              GFP_NOFS);
3706         return err;
3707 }
3708
3709 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3710 {
3711         struct btrfs_root *root = BTRFS_I(inode)->root;
3712         struct btrfs_trans_handle *trans;
3713         loff_t oldsize = i_size_read(inode);
3714         int ret;
3715
3716         if (newsize == oldsize)
3717                 return 0;
3718
3719         if (newsize > oldsize) {
3720                 truncate_pagecache(inode, oldsize, newsize);
3721                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3722                 if (ret)
3723                         return ret;
3724
3725                 trans = btrfs_start_transaction(root, 1);
3726                 if (IS_ERR(trans))
3727                         return PTR_ERR(trans);
3728
3729                 i_size_write(inode, newsize);
3730                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3731                 ret = btrfs_update_inode(trans, root, inode);
3732                 btrfs_end_transaction(trans, root);
3733         } else {
3734
3735                 /*
3736                  * We're truncating a file that used to have good data down to
3737                  * zero. Make sure it gets into the ordered flush list so that
3738                  * any new writes get down to disk quickly.
3739                  */
3740                 if (newsize == 0)
3741                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3742                                 &BTRFS_I(inode)->runtime_flags);
3743
3744                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3745                 truncate_setsize(inode, newsize);
3746                 ret = btrfs_truncate(inode);
3747         }
3748
3749         return ret;
3750 }
3751
3752 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3753 {
3754         struct inode *inode = dentry->d_inode;
3755         struct btrfs_root *root = BTRFS_I(inode)->root;
3756         int err;
3757
3758         if (btrfs_root_readonly(root))
3759                 return -EROFS;
3760
3761         err = inode_change_ok(inode, attr);
3762         if (err)
3763                 return err;
3764
3765         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3766                 err = btrfs_setsize(inode, attr->ia_size);
3767                 if (err)
3768                         return err;
3769         }
3770
3771         if (attr->ia_valid) {
3772                 setattr_copy(inode, attr);
3773                 inode_inc_iversion(inode);
3774                 err = btrfs_dirty_inode(inode);
3775
3776                 if (!err && attr->ia_valid & ATTR_MODE)
3777                         err = btrfs_acl_chmod(inode);
3778         }
3779
3780         return err;
3781 }
3782
3783 void btrfs_evict_inode(struct inode *inode)
3784 {
3785         struct btrfs_trans_handle *trans;
3786         struct btrfs_root *root = BTRFS_I(inode)->root;
3787         struct btrfs_block_rsv *rsv, *global_rsv;
3788         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3789         unsigned long nr;
3790         int ret;
3791
3792         trace_btrfs_inode_evict(inode);
3793
3794         truncate_inode_pages(&inode->i_data, 0);
3795         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3796                                btrfs_is_free_space_inode(inode)))
3797                 goto no_delete;
3798
3799         if (is_bad_inode(inode)) {
3800                 btrfs_orphan_del(NULL, inode);
3801                 goto no_delete;
3802         }
3803         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3804         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3805
3806         if (root->fs_info->log_root_recovering) {
3807                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3808                                  &BTRFS_I(inode)->runtime_flags));
3809                 goto no_delete;
3810         }
3811
3812         if (inode->i_nlink > 0) {
3813                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3814                 goto no_delete;
3815         }
3816
3817         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3818         if (!rsv) {
3819                 btrfs_orphan_del(NULL, inode);
3820                 goto no_delete;
3821         }
3822         rsv->size = min_size;
3823         rsv->failfast = 1;
3824         global_rsv = &root->fs_info->global_block_rsv;
3825
3826         btrfs_i_size_write(inode, 0);
3827
3828         /*
3829          * This is a bit simpler than btrfs_truncate since we've already
3830          * reserved our space for our orphan item in the unlink, so we just
3831          * need to reserve some slack space in case we add bytes and update
3832          * inode item when doing the truncate.
3833          */
3834         while (1) {
3835                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3836
3837                 /*
3838                  * Try and steal from the global reserve since we will
3839                  * likely not use this space anyway, we want to try as
3840                  * hard as possible to get this to work.
3841                  */
3842                 if (ret)
3843                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3844
3845                 if (ret) {
3846                         printk(KERN_WARNING "Could not get space for a "
3847                                "delete, will truncate on mount %d\n", ret);
3848                         btrfs_orphan_del(NULL, inode);
3849                         btrfs_free_block_rsv(root, rsv);
3850                         goto no_delete;
3851                 }
3852
3853                 trans = btrfs_start_transaction_noflush(root, 1);
3854                 if (IS_ERR(trans)) {
3855                         btrfs_orphan_del(NULL, inode);
3856                         btrfs_free_block_rsv(root, rsv);
3857                         goto no_delete;
3858                 }
3859
3860                 trans->block_rsv = rsv;
3861
3862                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3863                 if (ret != -ENOSPC)
3864                         break;
3865
3866                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3867                 ret = btrfs_update_inode(trans, root, inode);
3868                 BUG_ON(ret);
3869
3870                 nr = trans->blocks_used;
3871                 btrfs_end_transaction(trans, root);
3872                 trans = NULL;
3873                 btrfs_btree_balance_dirty(root, nr);
3874         }
3875
3876         btrfs_free_block_rsv(root, rsv);
3877
3878         if (ret == 0) {
3879                 trans->block_rsv = root->orphan_block_rsv;
3880                 ret = btrfs_orphan_del(trans, inode);
3881                 BUG_ON(ret);
3882         }
3883
3884         trans->block_rsv = &root->fs_info->trans_block_rsv;
3885         if (!(root == root->fs_info->tree_root ||
3886               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3887                 btrfs_return_ino(root, btrfs_ino(inode));
3888
3889         nr = trans->blocks_used;
3890         btrfs_end_transaction(trans, root);
3891         btrfs_btree_balance_dirty(root, nr);
3892 no_delete:
3893         clear_inode(inode);
3894         return;
3895 }
3896
3897 /*
3898  * this returns the key found in the dir entry in the location pointer.
3899  * If no dir entries were found, location->objectid is 0.
3900  */
3901 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3902                                struct btrfs_key *location)
3903 {
3904         const char *name = dentry->d_name.name;
3905         int namelen = dentry->d_name.len;
3906         struct btrfs_dir_item *di;
3907         struct btrfs_path *path;
3908         struct btrfs_root *root = BTRFS_I(dir)->root;
3909         int ret = 0;
3910
3911         path = btrfs_alloc_path();
3912         if (!path)
3913                 return -ENOMEM;
3914
3915         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3916                                     namelen, 0);
3917         if (IS_ERR(di))
3918                 ret = PTR_ERR(di);
3919
3920         if (IS_ERR_OR_NULL(di))
3921                 goto out_err;
3922
3923         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3924 out:
3925         btrfs_free_path(path);
3926         return ret;
3927 out_err:
3928         location->objectid = 0;
3929         goto out;
3930 }
3931
3932 /*
3933  * when we hit a tree root in a directory, the btrfs part of the inode
3934  * needs to be changed to reflect the root directory of the tree root.  This
3935  * is kind of like crossing a mount point.
3936  */
3937 static int fixup_tree_root_location(struct btrfs_root *root,
3938                                     struct inode *dir,
3939                                     struct dentry *dentry,
3940                                     struct btrfs_key *location,
3941                                     struct btrfs_root **sub_root)
3942 {
3943         struct btrfs_path *path;
3944         struct btrfs_root *new_root;
3945         struct btrfs_root_ref *ref;
3946         struct extent_buffer *leaf;
3947         int ret;
3948         int err = 0;
3949
3950         path = btrfs_alloc_path();
3951         if (!path) {
3952                 err = -ENOMEM;
3953                 goto out;
3954         }
3955
3956         err = -ENOENT;
3957         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3958                                   BTRFS_I(dir)->root->root_key.objectid,
3959                                   location->objectid);
3960         if (ret) {
3961                 if (ret < 0)
3962                         err = ret;
3963                 goto out;
3964         }
3965
3966         leaf = path->nodes[0];
3967         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3968         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3969             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3970                 goto out;
3971
3972         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3973                                    (unsigned long)(ref + 1),
3974                                    dentry->d_name.len);
3975         if (ret)
3976                 goto out;
3977
3978         btrfs_release_path(path);
3979
3980         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3981         if (IS_ERR(new_root)) {
3982                 err = PTR_ERR(new_root);
3983                 goto out;
3984         }
3985
3986         if (btrfs_root_refs(&new_root->root_item) == 0) {
3987                 err = -ENOENT;
3988                 goto out;
3989         }
3990
3991         *sub_root = new_root;
3992         location->objectid = btrfs_root_dirid(&new_root->root_item);
3993         location->type = BTRFS_INODE_ITEM_KEY;
3994         location->offset = 0;
3995         err = 0;
3996 out:
3997         btrfs_free_path(path);
3998         return err;
3999 }
4000
4001 static void inode_tree_add(struct inode *inode)
4002 {
4003         struct btrfs_root *root = BTRFS_I(inode)->root;
4004         struct btrfs_inode *entry;
4005         struct rb_node **p;
4006         struct rb_node *parent;
4007         u64 ino = btrfs_ino(inode);
4008 again:
4009         p = &root->inode_tree.rb_node;
4010         parent = NULL;
4011
4012         if (inode_unhashed(inode))
4013                 return;
4014
4015         spin_lock(&root->inode_lock);
4016         while (*p) {
4017                 parent = *p;
4018                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4019
4020                 if (ino < btrfs_ino(&entry->vfs_inode))
4021                         p = &parent->rb_left;
4022                 else if (ino > btrfs_ino(&entry->vfs_inode))
4023                         p = &parent->rb_right;
4024                 else {
4025                         WARN_ON(!(entry->vfs_inode.i_state &
4026                                   (I_WILL_FREE | I_FREEING)));
4027                         rb_erase(parent, &root->inode_tree);
4028                         RB_CLEAR_NODE(parent);
4029                         spin_unlock(&root->inode_lock);
4030                         goto again;
4031                 }
4032         }
4033         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4034         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4035         spin_unlock(&root->inode_lock);
4036 }
4037
4038 static void inode_tree_del(struct inode *inode)
4039 {
4040         struct btrfs_root *root = BTRFS_I(inode)->root;
4041         int empty = 0;
4042
4043         spin_lock(&root->inode_lock);
4044         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4045                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4046                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4047                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4048         }
4049         spin_unlock(&root->inode_lock);
4050
4051         /*
4052          * Free space cache has inodes in the tree root, but the tree root has a
4053          * root_refs of 0, so this could end up dropping the tree root as a
4054          * snapshot, so we need the extra !root->fs_info->tree_root check to
4055          * make sure we don't drop it.
4056          */
4057         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4058             root != root->fs_info->tree_root) {
4059                 synchronize_srcu(&root->fs_info->subvol_srcu);
4060                 spin_lock(&root->inode_lock);
4061                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4062                 spin_unlock(&root->inode_lock);
4063                 if (empty)
4064                         btrfs_add_dead_root(root);
4065         }
4066 }
4067
4068 void btrfs_invalidate_inodes(struct btrfs_root *root)
4069 {
4070         struct rb_node *node;
4071         struct rb_node *prev;
4072         struct btrfs_inode *entry;
4073         struct inode *inode;
4074         u64 objectid = 0;
4075
4076         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4077
4078         spin_lock(&root->inode_lock);
4079 again:
4080         node = root->inode_tree.rb_node;
4081         prev = NULL;
4082         while (node) {
4083                 prev = node;
4084                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4085
4086                 if (objectid < btrfs_ino(&entry->vfs_inode))
4087                         node = node->rb_left;
4088                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4089                         node = node->rb_right;
4090                 else
4091                         break;
4092         }
4093         if (!node) {
4094                 while (prev) {
4095                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4096                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4097                                 node = prev;
4098                                 break;
4099                         }
4100                         prev = rb_next(prev);
4101                 }
4102         }
4103         while (node) {
4104                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4105                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4106                 inode = igrab(&entry->vfs_inode);
4107                 if (inode) {
4108                         spin_unlock(&root->inode_lock);
4109                         if (atomic_read(&inode->i_count) > 1)
4110                                 d_prune_aliases(inode);
4111                         /*
4112                          * btrfs_drop_inode will have it removed from
4113                          * the inode cache when its usage count
4114                          * hits zero.
4115                          */
4116                         iput(inode);
4117                         cond_resched();
4118                         spin_lock(&root->inode_lock);
4119                         goto again;
4120                 }
4121
4122                 if (cond_resched_lock(&root->inode_lock))
4123                         goto again;
4124
4125                 node = rb_next(node);
4126         }
4127         spin_unlock(&root->inode_lock);
4128 }
4129
4130 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4131 {
4132         struct btrfs_iget_args *args = p;
4133         inode->i_ino = args->ino;
4134         BTRFS_I(inode)->root = args->root;
4135         return 0;
4136 }
4137
4138 static int btrfs_find_actor(struct inode *inode, void *opaque)
4139 {
4140         struct btrfs_iget_args *args = opaque;
4141         return args->ino == btrfs_ino(inode) &&
4142                 args->root == BTRFS_I(inode)->root;
4143 }
4144
4145 static struct inode *btrfs_iget_locked(struct super_block *s,
4146                                        u64 objectid,
4147                                        struct btrfs_root *root)
4148 {
4149         struct inode *inode;
4150         struct btrfs_iget_args args;
4151         args.ino = objectid;
4152         args.root = root;
4153
4154         inode = iget5_locked(s, objectid, btrfs_find_actor,
4155                              btrfs_init_locked_inode,
4156                              (void *)&args);
4157         return inode;
4158 }
4159
4160 /* Get an inode object given its location and corresponding root.
4161  * Returns in *is_new if the inode was read from disk
4162  */
4163 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4164                          struct btrfs_root *root, int *new)
4165 {
4166         struct inode *inode;
4167
4168         inode = btrfs_iget_locked(s, location->objectid, root);
4169         if (!inode)
4170                 return ERR_PTR(-ENOMEM);
4171
4172         if (inode->i_state & I_NEW) {
4173                 BTRFS_I(inode)->root = root;
4174                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4175                 btrfs_read_locked_inode(inode);
4176                 if (!is_bad_inode(inode)) {
4177                         inode_tree_add(inode);
4178                         unlock_new_inode(inode);
4179                         if (new)
4180                                 *new = 1;
4181                 } else {
4182                         unlock_new_inode(inode);
4183                         iput(inode);
4184                         inode = ERR_PTR(-ESTALE);
4185                 }
4186         }
4187
4188         return inode;
4189 }
4190
4191 static struct inode *new_simple_dir(struct super_block *s,
4192                                     struct btrfs_key *key,
4193                                     struct btrfs_root *root)
4194 {
4195         struct inode *inode = new_inode(s);
4196
4197         if (!inode)
4198                 return ERR_PTR(-ENOMEM);
4199
4200         BTRFS_I(inode)->root = root;
4201         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4202         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4203
4204         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4205         inode->i_op = &btrfs_dir_ro_inode_operations;
4206         inode->i_fop = &simple_dir_operations;
4207         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4208         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4209
4210         return inode;
4211 }
4212
4213 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4214 {
4215         struct inode *inode;
4216         struct btrfs_root *root = BTRFS_I(dir)->root;
4217         struct btrfs_root *sub_root = root;
4218         struct btrfs_key location;
4219         int index;
4220         int ret = 0;
4221
4222         if (dentry->d_name.len > BTRFS_NAME_LEN)
4223                 return ERR_PTR(-ENAMETOOLONG);
4224
4225         if (unlikely(d_need_lookup(dentry))) {
4226                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4227                 kfree(dentry->d_fsdata);
4228                 dentry->d_fsdata = NULL;
4229                 /* This thing is hashed, drop it for now */
4230                 d_drop(dentry);
4231         } else {
4232                 ret = btrfs_inode_by_name(dir, dentry, &location);
4233         }
4234
4235         if (ret < 0)
4236                 return ERR_PTR(ret);
4237
4238         if (location.objectid == 0)
4239                 return NULL;
4240
4241         if (location.type == BTRFS_INODE_ITEM_KEY) {
4242                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4243                 return inode;
4244         }
4245
4246         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4247
4248         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4249         ret = fixup_tree_root_location(root, dir, dentry,
4250                                        &location, &sub_root);
4251         if (ret < 0) {
4252                 if (ret != -ENOENT)
4253                         inode = ERR_PTR(ret);
4254                 else
4255                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4256         } else {
4257                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4258         }
4259         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4260
4261         if (!IS_ERR(inode) && root != sub_root) {
4262                 down_read(&root->fs_info->cleanup_work_sem);
4263                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4264                         ret = btrfs_orphan_cleanup(sub_root);
4265                 up_read(&root->fs_info->cleanup_work_sem);
4266                 if (ret)
4267                         inode = ERR_PTR(ret);
4268         }
4269
4270         return inode;
4271 }
4272
4273 static int btrfs_dentry_delete(const struct dentry *dentry)
4274 {
4275         struct btrfs_root *root;
4276         struct inode *inode = dentry->d_inode;
4277
4278         if (!inode && !IS_ROOT(dentry))
4279                 inode = dentry->d_parent->d_inode;
4280
4281         if (inode) {
4282                 root = BTRFS_I(inode)->root;
4283                 if (btrfs_root_refs(&root->root_item) == 0)
4284                         return 1;
4285
4286                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4287                         return 1;
4288         }
4289         return 0;
4290 }
4291
4292 static void btrfs_dentry_release(struct dentry *dentry)
4293 {
4294         if (dentry->d_fsdata)
4295                 kfree(dentry->d_fsdata);
4296 }
4297
4298 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4299                                    unsigned int flags)
4300 {
4301         struct dentry *ret;
4302
4303         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4304         if (unlikely(d_need_lookup(dentry))) {
4305                 spin_lock(&dentry->d_lock);
4306                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4307                 spin_unlock(&dentry->d_lock);
4308         }
4309         return ret;
4310 }
4311
4312 unsigned char btrfs_filetype_table[] = {
4313         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4314 };
4315
4316 static int btrfs_real_readdir(struct file *filp, void *dirent,
4317                               filldir_t filldir)
4318 {
4319         struct inode *inode = filp->f_dentry->d_inode;
4320         struct btrfs_root *root = BTRFS_I(inode)->root;
4321         struct btrfs_item *item;
4322         struct btrfs_dir_item *di;
4323         struct btrfs_key key;
4324         struct btrfs_key found_key;
4325         struct btrfs_path *path;
4326         struct list_head ins_list;
4327         struct list_head del_list;
4328         int ret;
4329         struct extent_buffer *leaf;
4330         int slot;
4331         unsigned char d_type;
4332         int over = 0;
4333         u32 di_cur;
4334         u32 di_total;
4335         u32 di_len;
4336         int key_type = BTRFS_DIR_INDEX_KEY;
4337         char tmp_name[32];
4338         char *name_ptr;
4339         int name_len;
4340         int is_curr = 0;        /* filp->f_pos points to the current index? */
4341
4342         /* FIXME, use a real flag for deciding about the key type */
4343         if (root->fs_info->tree_root == root)
4344                 key_type = BTRFS_DIR_ITEM_KEY;
4345
4346         /* special case for "." */
4347         if (filp->f_pos == 0) {
4348                 over = filldir(dirent, ".", 1,
4349                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4350                 if (over)
4351                         return 0;
4352                 filp->f_pos = 1;
4353         }
4354         /* special case for .., just use the back ref */
4355         if (filp->f_pos == 1) {
4356                 u64 pino = parent_ino(filp->f_path.dentry);
4357                 over = filldir(dirent, "..", 2,
4358                                filp->f_pos, pino, DT_DIR);
4359                 if (over)
4360                         return 0;
4361                 filp->f_pos = 2;
4362         }
4363         path = btrfs_alloc_path();
4364         if (!path)
4365                 return -ENOMEM;
4366
4367         path->reada = 1;
4368
4369         if (key_type == BTRFS_DIR_INDEX_KEY) {
4370                 INIT_LIST_HEAD(&ins_list);
4371                 INIT_LIST_HEAD(&del_list);
4372                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4373         }
4374
4375         btrfs_set_key_type(&key, key_type);
4376         key.offset = filp->f_pos;
4377         key.objectid = btrfs_ino(inode);
4378
4379         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4380         if (ret < 0)
4381                 goto err;
4382
4383         while (1) {
4384                 leaf = path->nodes[0];
4385                 slot = path->slots[0];
4386                 if (slot >= btrfs_header_nritems(leaf)) {
4387                         ret = btrfs_next_leaf(root, path);
4388                         if (ret < 0)
4389                                 goto err;
4390                         else if (ret > 0)
4391                                 break;
4392                         continue;
4393                 }
4394
4395                 item = btrfs_item_nr(leaf, slot);
4396                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4397
4398                 if (found_key.objectid != key.objectid)
4399                         break;
4400                 if (btrfs_key_type(&found_key) != key_type)
4401                         break;
4402                 if (found_key.offset < filp->f_pos)
4403                         goto next;
4404                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4405                     btrfs_should_delete_dir_index(&del_list,
4406                                                   found_key.offset))
4407                         goto next;
4408
4409                 filp->f_pos = found_key.offset;
4410                 is_curr = 1;
4411
4412                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4413                 di_cur = 0;
4414                 di_total = btrfs_item_size(leaf, item);
4415
4416                 while (di_cur < di_total) {
4417                         struct btrfs_key location;
4418
4419                         if (verify_dir_item(root, leaf, di))
4420                                 break;
4421
4422                         name_len = btrfs_dir_name_len(leaf, di);
4423                         if (name_len <= sizeof(tmp_name)) {
4424                                 name_ptr = tmp_name;
4425                         } else {
4426                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4427                                 if (!name_ptr) {
4428                                         ret = -ENOMEM;
4429                                         goto err;
4430                                 }
4431                         }
4432                         read_extent_buffer(leaf, name_ptr,
4433                                            (unsigned long)(di + 1), name_len);
4434
4435                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4436                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4437
4438
4439                         /* is this a reference to our own snapshot? If so
4440                          * skip it.
4441                          *
4442                          * In contrast to old kernels, we insert the snapshot's
4443                          * dir item and dir index after it has been created, so
4444                          * we won't find a reference to our own snapshot. We
4445                          * still keep the following code for backward
4446                          * compatibility.
4447                          */
4448                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4449                             location.objectid == root->root_key.objectid) {
4450                                 over = 0;
4451                                 goto skip;
4452                         }
4453                         over = filldir(dirent, name_ptr, name_len,
4454                                        found_key.offset, location.objectid,
4455                                        d_type);
4456
4457 skip:
4458                         if (name_ptr != tmp_name)
4459                                 kfree(name_ptr);
4460
4461                         if (over)
4462                                 goto nopos;
4463                         di_len = btrfs_dir_name_len(leaf, di) +
4464                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4465                         di_cur += di_len;
4466                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4467                 }
4468 next:
4469                 path->slots[0]++;
4470         }
4471
4472         if (key_type == BTRFS_DIR_INDEX_KEY) {
4473                 if (is_curr)
4474                         filp->f_pos++;
4475                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4476                                                       &ins_list);
4477                 if (ret)
4478                         goto nopos;
4479         }
4480
4481         /* Reached end of directory/root. Bump pos past the last item. */
4482         if (key_type == BTRFS_DIR_INDEX_KEY)
4483                 /*
4484                  * 32-bit glibc will use getdents64, but then strtol -
4485                  * so the last number we can serve is this.
4486                  */
4487                 filp->f_pos = 0x7fffffff;
4488         else
4489                 filp->f_pos++;
4490 nopos:
4491         ret = 0;
4492 err:
4493         if (key_type == BTRFS_DIR_INDEX_KEY)
4494                 btrfs_put_delayed_items(&ins_list, &del_list);
4495         btrfs_free_path(path);
4496         return ret;
4497 }
4498
4499 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4500 {
4501         struct btrfs_root *root = BTRFS_I(inode)->root;
4502         struct btrfs_trans_handle *trans;
4503         int ret = 0;
4504         bool nolock = false;
4505
4506         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4507                 return 0;
4508
4509         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4510                 nolock = true;
4511
4512         if (wbc->sync_mode == WB_SYNC_ALL) {
4513                 if (nolock)
4514                         trans = btrfs_join_transaction_nolock(root);
4515                 else
4516                         trans = btrfs_join_transaction(root);
4517                 if (IS_ERR(trans))
4518                         return PTR_ERR(trans);
4519                 ret = btrfs_commit_transaction(trans, root);
4520         }
4521         return ret;
4522 }
4523
4524 /*
4525  * This is somewhat expensive, updating the tree every time the
4526  * inode changes.  But, it is most likely to find the inode in cache.
4527  * FIXME, needs more benchmarking...there are no reasons other than performance
4528  * to keep or drop this code.
4529  */
4530 int btrfs_dirty_inode(struct inode *inode)
4531 {
4532         struct btrfs_root *root = BTRFS_I(inode)->root;
4533         struct btrfs_trans_handle *trans;
4534         int ret;
4535
4536         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4537                 return 0;
4538
4539         trans = btrfs_join_transaction(root);
4540         if (IS_ERR(trans))
4541                 return PTR_ERR(trans);
4542
4543         ret = btrfs_update_inode(trans, root, inode);
4544         if (ret && ret == -ENOSPC) {
4545                 /* whoops, lets try again with the full transaction */
4546                 btrfs_end_transaction(trans, root);
4547                 trans = btrfs_start_transaction(root, 1);
4548                 if (IS_ERR(trans))
4549                         return PTR_ERR(trans);
4550
4551                 ret = btrfs_update_inode(trans, root, inode);
4552         }
4553         btrfs_end_transaction(trans, root);
4554         if (BTRFS_I(inode)->delayed_node)
4555                 btrfs_balance_delayed_items(root);
4556
4557         return ret;
4558 }
4559
4560 /*
4561  * This is a copy of file_update_time.  We need this so we can return error on
4562  * ENOSPC for updating the inode in the case of file write and mmap writes.
4563  */
4564 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4565                              int flags)
4566 {
4567         struct btrfs_root *root = BTRFS_I(inode)->root;
4568
4569         if (btrfs_root_readonly(root))
4570                 return -EROFS;
4571
4572         if (flags & S_VERSION)
4573                 inode_inc_iversion(inode);
4574         if (flags & S_CTIME)
4575                 inode->i_ctime = *now;
4576         if (flags & S_MTIME)
4577                 inode->i_mtime = *now;
4578         if (flags & S_ATIME)
4579                 inode->i_atime = *now;
4580         return btrfs_dirty_inode(inode);
4581 }
4582
4583 /*
4584  * find the highest existing sequence number in a directory
4585  * and then set the in-memory index_cnt variable to reflect
4586  * free sequence numbers
4587  */
4588 static int btrfs_set_inode_index_count(struct inode *inode)
4589 {
4590         struct btrfs_root *root = BTRFS_I(inode)->root;
4591         struct btrfs_key key, found_key;
4592         struct btrfs_path *path;
4593         struct extent_buffer *leaf;
4594         int ret;
4595
4596         key.objectid = btrfs_ino(inode);
4597         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4598         key.offset = (u64)-1;
4599
4600         path = btrfs_alloc_path();
4601         if (!path)
4602                 return -ENOMEM;
4603
4604         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4605         if (ret < 0)
4606                 goto out;
4607         /* FIXME: we should be able to handle this */
4608         if (ret == 0)
4609                 goto out;
4610         ret = 0;
4611
4612         /*
4613          * MAGIC NUMBER EXPLANATION:
4614          * since we search a directory based on f_pos we have to start at 2
4615          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4616          * else has to start at 2
4617          */
4618         if (path->slots[0] == 0) {
4619                 BTRFS_I(inode)->index_cnt = 2;
4620                 goto out;
4621         }
4622
4623         path->slots[0]--;
4624
4625         leaf = path->nodes[0];
4626         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4627
4628         if (found_key.objectid != btrfs_ino(inode) ||
4629             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4630                 BTRFS_I(inode)->index_cnt = 2;
4631                 goto out;
4632         }
4633
4634         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4635 out:
4636         btrfs_free_path(path);
4637         return ret;
4638 }
4639
4640 /*
4641  * helper to find a free sequence number in a given directory.  This current
4642  * code is very simple, later versions will do smarter things in the btree
4643  */
4644 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4645 {
4646         int ret = 0;
4647
4648         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4649                 ret = btrfs_inode_delayed_dir_index_count(dir);
4650                 if (ret) {
4651                         ret = btrfs_set_inode_index_count(dir);
4652                         if (ret)
4653                                 return ret;
4654                 }
4655         }
4656
4657         *index = BTRFS_I(dir)->index_cnt;
4658         BTRFS_I(dir)->index_cnt++;
4659
4660         return ret;
4661 }
4662
4663 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4664                                      struct btrfs_root *root,
4665                                      struct inode *dir,
4666                                      const char *name, int name_len,
4667                                      u64 ref_objectid, u64 objectid,
4668                                      umode_t mode, u64 *index)
4669 {
4670         struct inode *inode;
4671         struct btrfs_inode_item *inode_item;
4672         struct btrfs_key *location;
4673         struct btrfs_path *path;
4674         struct btrfs_inode_ref *ref;
4675         struct btrfs_key key[2];
4676         u32 sizes[2];
4677         unsigned long ptr;
4678         int ret;
4679         int owner;
4680
4681         path = btrfs_alloc_path();
4682         if (!path)
4683                 return ERR_PTR(-ENOMEM);
4684
4685         inode = new_inode(root->fs_info->sb);
4686         if (!inode) {
4687                 btrfs_free_path(path);
4688                 return ERR_PTR(-ENOMEM);
4689         }
4690
4691         /*
4692          * we have to initialize this early, so we can reclaim the inode
4693          * number if we fail afterwards in this function.
4694          */
4695         inode->i_ino = objectid;
4696
4697         if (dir) {
4698                 trace_btrfs_inode_request(dir);
4699
4700                 ret = btrfs_set_inode_index(dir, index);
4701                 if (ret) {
4702                         btrfs_free_path(path);
4703                         iput(inode);
4704                         return ERR_PTR(ret);
4705                 }
4706         }
4707         /*
4708          * index_cnt is ignored for everything but a dir,
4709          * btrfs_get_inode_index_count has an explanation for the magic
4710          * number
4711          */
4712         BTRFS_I(inode)->index_cnt = 2;
4713         BTRFS_I(inode)->root = root;
4714         BTRFS_I(inode)->generation = trans->transid;
4715         inode->i_generation = BTRFS_I(inode)->generation;
4716
4717         /*
4718          * We could have gotten an inode number from somebody who was fsynced
4719          * and then removed in this same transaction, so let's just set full
4720          * sync since it will be a full sync anyway and this will blow away the
4721          * old info in the log.
4722          */
4723         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4724
4725         if (S_ISDIR(mode))
4726                 owner = 0;
4727         else
4728                 owner = 1;
4729
4730         key[0].objectid = objectid;
4731         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4732         key[0].offset = 0;
4733
4734         /*
4735          * Start new inodes with an inode_ref. This is slightly more
4736          * efficient for small numbers of hard links since they will
4737          * be packed into one item. Extended refs will kick in if we
4738          * add more hard links than can fit in the ref item.
4739          */
4740         key[1].objectid = objectid;
4741         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4742         key[1].offset = ref_objectid;
4743
4744         sizes[0] = sizeof(struct btrfs_inode_item);
4745         sizes[1] = name_len + sizeof(*ref);
4746
4747         path->leave_spinning = 1;
4748         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4749         if (ret != 0)
4750                 goto fail;
4751
4752         inode_init_owner(inode, dir, mode);
4753         inode_set_bytes(inode, 0);
4754         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4755         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4756                                   struct btrfs_inode_item);
4757         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4758                              sizeof(*inode_item));
4759         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4760
4761         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4762                              struct btrfs_inode_ref);
4763         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4764         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4765         ptr = (unsigned long)(ref + 1);
4766         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4767
4768         btrfs_mark_buffer_dirty(path->nodes[0]);
4769         btrfs_free_path(path);
4770
4771         location = &BTRFS_I(inode)->location;
4772         location->objectid = objectid;
4773         location->offset = 0;
4774         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4775
4776         btrfs_inherit_iflags(inode, dir);
4777
4778         if (S_ISREG(mode)) {
4779                 if (btrfs_test_opt(root, NODATASUM))
4780                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4781                 if (btrfs_test_opt(root, NODATACOW) ||
4782                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4783                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4784         }
4785
4786         insert_inode_hash(inode);
4787         inode_tree_add(inode);
4788
4789         trace_btrfs_inode_new(inode);
4790         btrfs_set_inode_last_trans(trans, inode);
4791
4792         btrfs_update_root_times(trans, root);
4793
4794         return inode;
4795 fail:
4796         if (dir)
4797                 BTRFS_I(dir)->index_cnt--;
4798         btrfs_free_path(path);
4799         iput(inode);
4800         return ERR_PTR(ret);
4801 }
4802
4803 static inline u8 btrfs_inode_type(struct inode *inode)
4804 {
4805         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4806 }
4807
4808 /*
4809  * utility function to add 'inode' into 'parent_inode' with
4810  * a give name and a given sequence number.
4811  * if 'add_backref' is true, also insert a backref from the
4812  * inode to the parent directory.
4813  */
4814 int btrfs_add_link(struct btrfs_trans_handle *trans,
4815                    struct inode *parent_inode, struct inode *inode,
4816                    const char *name, int name_len, int add_backref, u64 index)
4817 {
4818         int ret = 0;
4819         struct btrfs_key key;
4820         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4821         u64 ino = btrfs_ino(inode);
4822         u64 parent_ino = btrfs_ino(parent_inode);
4823
4824         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4825                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4826         } else {
4827                 key.objectid = ino;
4828                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4829                 key.offset = 0;
4830         }
4831
4832         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4833                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4834                                          key.objectid, root->root_key.objectid,
4835                                          parent_ino, index, name, name_len);
4836         } else if (add_backref) {
4837                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4838                                              parent_ino, index);
4839         }
4840
4841         /* Nothing to clean up yet */
4842         if (ret)
4843                 return ret;
4844
4845         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4846                                     parent_inode, &key,
4847                                     btrfs_inode_type(inode), index);
4848         if (ret == -EEXIST)
4849                 goto fail_dir_item;
4850         else if (ret) {
4851                 btrfs_abort_transaction(trans, root, ret);
4852                 return ret;
4853         }
4854
4855         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4856                            name_len * 2);
4857         inode_inc_iversion(parent_inode);
4858         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4859         ret = btrfs_update_inode(trans, root, parent_inode);
4860         if (ret)
4861                 btrfs_abort_transaction(trans, root, ret);
4862         return ret;
4863
4864 fail_dir_item:
4865         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4866                 u64 local_index;
4867                 int err;
4868                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4869                                  key.objectid, root->root_key.objectid,
4870                                  parent_ino, &local_index, name, name_len);
4871
4872         } else if (add_backref) {
4873                 u64 local_index;
4874                 int err;
4875
4876                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4877                                           ino, parent_ino, &local_index);
4878         }
4879         return ret;
4880 }
4881
4882 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4883                             struct inode *dir, struct dentry *dentry,
4884                             struct inode *inode, int backref, u64 index)
4885 {
4886         int err = btrfs_add_link(trans, dir, inode,
4887                                  dentry->d_name.name, dentry->d_name.len,
4888                                  backref, index);
4889         if (err > 0)
4890                 err = -EEXIST;
4891         return err;
4892 }
4893
4894 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4895                         umode_t mode, dev_t rdev)
4896 {
4897         struct btrfs_trans_handle *trans;
4898         struct btrfs_root *root = BTRFS_I(dir)->root;
4899         struct inode *inode = NULL;
4900         int err;
4901         int drop_inode = 0;
4902         u64 objectid;
4903         unsigned long nr = 0;
4904         u64 index = 0;
4905
4906         if (!new_valid_dev(rdev))
4907                 return -EINVAL;
4908
4909         /*
4910          * 2 for inode item and ref
4911          * 2 for dir items
4912          * 1 for xattr if selinux is on
4913          */
4914         trans = btrfs_start_transaction(root, 5);
4915         if (IS_ERR(trans))
4916                 return PTR_ERR(trans);
4917
4918         err = btrfs_find_free_ino(root, &objectid);
4919         if (err)
4920                 goto out_unlock;
4921
4922         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4923                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4924                                 mode, &index);
4925         if (IS_ERR(inode)) {
4926                 err = PTR_ERR(inode);
4927                 goto out_unlock;
4928         }
4929
4930         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4931         if (err) {
4932                 drop_inode = 1;
4933                 goto out_unlock;
4934         }
4935
4936         /*
4937         * If the active LSM wants to access the inode during
4938         * d_instantiate it needs these. Smack checks to see
4939         * if the filesystem supports xattrs by looking at the
4940         * ops vector.
4941         */
4942
4943         inode->i_op = &btrfs_special_inode_operations;
4944         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4945         if (err)
4946                 drop_inode = 1;
4947         else {
4948                 init_special_inode(inode, inode->i_mode, rdev);
4949                 btrfs_update_inode(trans, root, inode);
4950                 d_instantiate(dentry, inode);
4951         }
4952 out_unlock:
4953         nr = trans->blocks_used;
4954         btrfs_end_transaction(trans, root);
4955         btrfs_btree_balance_dirty(root, nr);
4956         if (drop_inode) {
4957                 inode_dec_link_count(inode);
4958                 iput(inode);
4959         }
4960         return err;
4961 }
4962
4963 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4964                         umode_t mode, bool excl)
4965 {
4966         struct btrfs_trans_handle *trans;
4967         struct btrfs_root *root = BTRFS_I(dir)->root;
4968         struct inode *inode = NULL;
4969         int drop_inode = 0;
4970         int err;
4971         unsigned long nr = 0;
4972         u64 objectid;
4973         u64 index = 0;
4974
4975         /*
4976          * 2 for inode item and ref
4977          * 2 for dir items
4978          * 1 for xattr if selinux is on
4979          */
4980         trans = btrfs_start_transaction(root, 5);
4981         if (IS_ERR(trans))
4982                 return PTR_ERR(trans);
4983
4984         err = btrfs_find_free_ino(root, &objectid);
4985         if (err)
4986                 goto out_unlock;
4987
4988         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4989                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4990                                 mode, &index);
4991         if (IS_ERR(inode)) {
4992                 err = PTR_ERR(inode);
4993                 goto out_unlock;
4994         }
4995
4996         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4997         if (err) {
4998                 drop_inode = 1;
4999                 goto out_unlock;
5000         }
5001
5002         /*
5003         * If the active LSM wants to access the inode during
5004         * d_instantiate it needs these. Smack checks to see
5005         * if the filesystem supports xattrs by looking at the
5006         * ops vector.
5007         */
5008         inode->i_fop = &btrfs_file_operations;
5009         inode->i_op = &btrfs_file_inode_operations;
5010
5011         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5012         if (err)
5013                 drop_inode = 1;
5014         else {
5015                 inode->i_mapping->a_ops = &btrfs_aops;
5016                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5017                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5018                 d_instantiate(dentry, inode);
5019         }
5020 out_unlock:
5021         nr = trans->blocks_used;
5022         btrfs_end_transaction(trans, root);
5023         if (drop_inode) {
5024                 inode_dec_link_count(inode);
5025                 iput(inode);
5026         }
5027         btrfs_btree_balance_dirty(root, nr);
5028         return err;
5029 }
5030
5031 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5032                       struct dentry *dentry)
5033 {
5034         struct btrfs_trans_handle *trans;
5035         struct btrfs_root *root = BTRFS_I(dir)->root;
5036         struct inode *inode = old_dentry->d_inode;
5037         u64 index;
5038         unsigned long nr = 0;
5039         int err;
5040         int drop_inode = 0;
5041
5042         /* do not allow sys_link's with other subvols of the same device */
5043         if (root->objectid != BTRFS_I(inode)->root->objectid)
5044                 return -EXDEV;
5045
5046         if (inode->i_nlink >= BTRFS_LINK_MAX)
5047                 return -EMLINK;
5048
5049         err = btrfs_set_inode_index(dir, &index);
5050         if (err)
5051                 goto fail;
5052
5053         /*
5054          * 2 items for inode and inode ref
5055          * 2 items for dir items
5056          * 1 item for parent inode
5057          */
5058         trans = btrfs_start_transaction(root, 5);
5059         if (IS_ERR(trans)) {
5060                 err = PTR_ERR(trans);
5061                 goto fail;
5062         }
5063
5064         btrfs_inc_nlink(inode);
5065         inode_inc_iversion(inode);
5066         inode->i_ctime = CURRENT_TIME;
5067         ihold(inode);
5068
5069         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5070
5071         if (err) {
5072                 drop_inode = 1;
5073         } else {
5074                 struct dentry *parent = dentry->d_parent;
5075                 err = btrfs_update_inode(trans, root, inode);
5076                 if (err)
5077                         goto fail;
5078                 d_instantiate(dentry, inode);
5079                 btrfs_log_new_name(trans, inode, NULL, parent);
5080         }
5081
5082         nr = trans->blocks_used;
5083         btrfs_end_transaction(trans, root);
5084 fail:
5085         if (drop_inode) {
5086                 inode_dec_link_count(inode);
5087                 iput(inode);
5088         }
5089         btrfs_btree_balance_dirty(root, nr);
5090         return err;
5091 }
5092
5093 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5094 {
5095         struct inode *inode = NULL;
5096         struct btrfs_trans_handle *trans;
5097         struct btrfs_root *root = BTRFS_I(dir)->root;
5098         int err = 0;
5099         int drop_on_err = 0;
5100         u64 objectid = 0;
5101         u64 index = 0;
5102         unsigned long nr = 1;
5103
5104         /*
5105          * 2 items for inode and ref
5106          * 2 items for dir items
5107          * 1 for xattr if selinux is on
5108          */
5109         trans = btrfs_start_transaction(root, 5);
5110         if (IS_ERR(trans))
5111                 return PTR_ERR(trans);
5112
5113         err = btrfs_find_free_ino(root, &objectid);
5114         if (err)
5115                 goto out_fail;
5116
5117         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5118                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5119                                 S_IFDIR | mode, &index);
5120         if (IS_ERR(inode)) {
5121                 err = PTR_ERR(inode);
5122                 goto out_fail;
5123         }
5124
5125         drop_on_err = 1;
5126
5127         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5128         if (err)
5129                 goto out_fail;
5130
5131         inode->i_op = &btrfs_dir_inode_operations;
5132         inode->i_fop = &btrfs_dir_file_operations;
5133
5134         btrfs_i_size_write(inode, 0);
5135         err = btrfs_update_inode(trans, root, inode);
5136         if (err)
5137                 goto out_fail;
5138
5139         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5140                              dentry->d_name.len, 0, index);
5141         if (err)
5142                 goto out_fail;
5143
5144         d_instantiate(dentry, inode);
5145         drop_on_err = 0;
5146
5147 out_fail:
5148         nr = trans->blocks_used;
5149         btrfs_end_transaction(trans, root);
5150         if (drop_on_err)
5151                 iput(inode);
5152         btrfs_btree_balance_dirty(root, nr);
5153         return err;
5154 }
5155
5156 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5157  * and an extent that you want to insert, deal with overlap and insert
5158  * the new extent into the tree.
5159  */
5160 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5161                                 struct extent_map *existing,
5162                                 struct extent_map *em,
5163                                 u64 map_start, u64 map_len)
5164 {
5165         u64 start_diff;
5166
5167         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5168         start_diff = map_start - em->start;
5169         em->start = map_start;
5170         em->len = map_len;
5171         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5172             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5173                 em->block_start += start_diff;
5174                 em->block_len -= start_diff;
5175         }
5176         return add_extent_mapping(em_tree, em);
5177 }
5178
5179 static noinline int uncompress_inline(struct btrfs_path *path,
5180                                       struct inode *inode, struct page *page,
5181                                       size_t pg_offset, u64 extent_offset,
5182                                       struct btrfs_file_extent_item *item)
5183 {
5184         int ret;
5185         struct extent_buffer *leaf = path->nodes[0];
5186         char *tmp;
5187         size_t max_size;
5188         unsigned long inline_size;
5189         unsigned long ptr;
5190         int compress_type;
5191
5192         WARN_ON(pg_offset != 0);
5193         compress_type = btrfs_file_extent_compression(leaf, item);
5194         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5195         inline_size = btrfs_file_extent_inline_item_len(leaf,
5196                                         btrfs_item_nr(leaf, path->slots[0]));
5197         tmp = kmalloc(inline_size, GFP_NOFS);
5198         if (!tmp)
5199                 return -ENOMEM;
5200         ptr = btrfs_file_extent_inline_start(item);
5201
5202         read_extent_buffer(leaf, tmp, ptr, inline_size);
5203
5204         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5205         ret = btrfs_decompress(compress_type, tmp, page,
5206                                extent_offset, inline_size, max_size);
5207         if (ret) {
5208                 char *kaddr = kmap_atomic(page);
5209                 unsigned long copy_size = min_t(u64,
5210                                   PAGE_CACHE_SIZE - pg_offset,
5211                                   max_size - extent_offset);
5212                 memset(kaddr + pg_offset, 0, copy_size);
5213                 kunmap_atomic(kaddr);
5214         }
5215         kfree(tmp);
5216         return 0;
5217 }
5218
5219 /*
5220  * a bit scary, this does extent mapping from logical file offset to the disk.
5221  * the ugly parts come from merging extents from the disk with the in-ram
5222  * representation.  This gets more complex because of the data=ordered code,
5223  * where the in-ram extents might be locked pending data=ordered completion.
5224  *
5225  * This also copies inline extents directly into the page.
5226  */
5227
5228 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5229                                     size_t pg_offset, u64 start, u64 len,
5230                                     int create)
5231 {
5232         int ret;
5233         int err = 0;
5234         u64 bytenr;
5235         u64 extent_start = 0;
5236         u64 extent_end = 0;
5237         u64 objectid = btrfs_ino(inode);
5238         u32 found_type;
5239         struct btrfs_path *path = NULL;
5240         struct btrfs_root *root = BTRFS_I(inode)->root;
5241         struct btrfs_file_extent_item *item;
5242         struct extent_buffer *leaf;
5243         struct btrfs_key found_key;
5244         struct extent_map *em = NULL;
5245         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5246         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5247         struct btrfs_trans_handle *trans = NULL;
5248         int compress_type;
5249
5250 again:
5251         read_lock(&em_tree->lock);
5252         em = lookup_extent_mapping(em_tree, start, len);
5253         if (em)
5254                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5255         read_unlock(&em_tree->lock);
5256
5257         if (em) {
5258                 if (em->start > start || em->start + em->len <= start)
5259                         free_extent_map(em);
5260                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5261                         free_extent_map(em);
5262                 else
5263                         goto out;
5264         }
5265         em = alloc_extent_map();
5266         if (!em) {
5267                 err = -ENOMEM;
5268                 goto out;
5269         }
5270         em->bdev = root->fs_info->fs_devices->latest_bdev;
5271         em->start = EXTENT_MAP_HOLE;
5272         em->orig_start = EXTENT_MAP_HOLE;
5273         em->len = (u64)-1;
5274         em->block_len = (u64)-1;
5275
5276         if (!path) {
5277                 path = btrfs_alloc_path();
5278                 if (!path) {
5279                         err = -ENOMEM;
5280                         goto out;
5281                 }
5282                 /*
5283                  * Chances are we'll be called again, so go ahead and do
5284                  * readahead
5285                  */
5286                 path->reada = 1;
5287         }
5288
5289         ret = btrfs_lookup_file_extent(trans, root, path,
5290                                        objectid, start, trans != NULL);
5291         if (ret < 0) {
5292                 err = ret;
5293                 goto out;
5294         }
5295
5296         if (ret != 0) {
5297                 if (path->slots[0] == 0)
5298                         goto not_found;
5299                 path->slots[0]--;
5300         }
5301
5302         leaf = path->nodes[0];
5303         item = btrfs_item_ptr(leaf, path->slots[0],
5304                               struct btrfs_file_extent_item);
5305         /* are we inside the extent that was found? */
5306         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5307         found_type = btrfs_key_type(&found_key);
5308         if (found_key.objectid != objectid ||
5309             found_type != BTRFS_EXTENT_DATA_KEY) {
5310                 goto not_found;
5311         }
5312
5313         found_type = btrfs_file_extent_type(leaf, item);
5314         extent_start = found_key.offset;
5315         compress_type = btrfs_file_extent_compression(leaf, item);
5316         if (found_type == BTRFS_FILE_EXTENT_REG ||
5317             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5318                 extent_end = extent_start +
5319                        btrfs_file_extent_num_bytes(leaf, item);
5320         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5321                 size_t size;
5322                 size = btrfs_file_extent_inline_len(leaf, item);
5323                 extent_end = (extent_start + size + root->sectorsize - 1) &
5324                         ~((u64)root->sectorsize - 1);
5325         }
5326
5327         if (start >= extent_end) {
5328                 path->slots[0]++;
5329                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5330                         ret = btrfs_next_leaf(root, path);
5331                         if (ret < 0) {
5332                                 err = ret;
5333                                 goto out;
5334                         }
5335                         if (ret > 0)
5336                                 goto not_found;
5337                         leaf = path->nodes[0];
5338                 }
5339                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5340                 if (found_key.objectid != objectid ||
5341                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5342                         goto not_found;
5343                 if (start + len <= found_key.offset)
5344                         goto not_found;
5345                 em->start = start;
5346                 em->len = found_key.offset - start;
5347                 goto not_found_em;
5348         }
5349
5350         if (found_type == BTRFS_FILE_EXTENT_REG ||
5351             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5352                 em->start = extent_start;
5353                 em->len = extent_end - extent_start;
5354                 em->orig_start = extent_start -
5355                                  btrfs_file_extent_offset(leaf, item);
5356                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5357                 if (bytenr == 0) {
5358                         em->block_start = EXTENT_MAP_HOLE;
5359                         goto insert;
5360                 }
5361                 if (compress_type != BTRFS_COMPRESS_NONE) {
5362                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5363                         em->compress_type = compress_type;
5364                         em->block_start = bytenr;
5365                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5366                                                                          item);
5367                 } else {
5368                         bytenr += btrfs_file_extent_offset(leaf, item);
5369                         em->block_start = bytenr;
5370                         em->block_len = em->len;
5371                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5372                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5373                 }
5374                 goto insert;
5375         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5376                 unsigned long ptr;
5377                 char *map;
5378                 size_t size;
5379                 size_t extent_offset;
5380                 size_t copy_size;
5381
5382                 em->block_start = EXTENT_MAP_INLINE;
5383                 if (!page || create) {
5384                         em->start = extent_start;
5385                         em->len = extent_end - extent_start;
5386                         goto out;
5387                 }
5388
5389                 size = btrfs_file_extent_inline_len(leaf, item);
5390                 extent_offset = page_offset(page) + pg_offset - extent_start;
5391                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5392                                 size - extent_offset);
5393                 em->start = extent_start + extent_offset;
5394                 em->len = (copy_size + root->sectorsize - 1) &
5395                         ~((u64)root->sectorsize - 1);
5396                 em->orig_start = EXTENT_MAP_INLINE;
5397                 if (compress_type) {
5398                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5399                         em->compress_type = compress_type;
5400                 }
5401                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5402                 if (create == 0 && !PageUptodate(page)) {
5403                         if (btrfs_file_extent_compression(leaf, item) !=
5404                             BTRFS_COMPRESS_NONE) {
5405                                 ret = uncompress_inline(path, inode, page,
5406                                                         pg_offset,
5407                                                         extent_offset, item);
5408                                 BUG_ON(ret); /* -ENOMEM */
5409                         } else {
5410                                 map = kmap(page);
5411                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5412                                                    copy_size);
5413                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5414                                         memset(map + pg_offset + copy_size, 0,
5415                                                PAGE_CACHE_SIZE - pg_offset -
5416                                                copy_size);
5417                                 }
5418                                 kunmap(page);
5419                         }
5420                         flush_dcache_page(page);
5421                 } else if (create && PageUptodate(page)) {
5422                         BUG();
5423                         if (!trans) {
5424                                 kunmap(page);
5425                                 free_extent_map(em);
5426                                 em = NULL;
5427
5428                                 btrfs_release_path(path);
5429                                 trans = btrfs_join_transaction(root);
5430
5431                                 if (IS_ERR(trans))
5432                                         return ERR_CAST(trans);
5433                                 goto again;
5434                         }
5435                         map = kmap(page);
5436                         write_extent_buffer(leaf, map + pg_offset, ptr,
5437                                             copy_size);
5438                         kunmap(page);
5439                         btrfs_mark_buffer_dirty(leaf);
5440                 }
5441                 set_extent_uptodate(io_tree, em->start,
5442                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5443                 goto insert;
5444         } else {
5445                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5446                 WARN_ON(1);
5447         }
5448 not_found:
5449         em->start = start;
5450         em->len = len;
5451 not_found_em:
5452         em->block_start = EXTENT_MAP_HOLE;
5453         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5454 insert:
5455         btrfs_release_path(path);
5456         if (em->start > start || extent_map_end(em) <= start) {
5457                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5458                        "[%llu %llu]\n", (unsigned long long)em->start,
5459                        (unsigned long long)em->len,
5460                        (unsigned long long)start,
5461                        (unsigned long long)len);
5462                 err = -EIO;
5463                 goto out;
5464         }
5465
5466         err = 0;
5467         write_lock(&em_tree->lock);
5468         ret = add_extent_mapping(em_tree, em);
5469         /* it is possible that someone inserted the extent into the tree
5470          * while we had the lock dropped.  It is also possible that
5471          * an overlapping map exists in the tree
5472          */
5473         if (ret == -EEXIST) {
5474                 struct extent_map *existing;
5475
5476                 ret = 0;
5477
5478                 existing = lookup_extent_mapping(em_tree, start, len);
5479                 if (existing && (existing->start > start ||
5480                     existing->start + existing->len <= start)) {
5481                         free_extent_map(existing);
5482                         existing = NULL;
5483                 }
5484                 if (!existing) {
5485                         existing = lookup_extent_mapping(em_tree, em->start,
5486                                                          em->len);
5487                         if (existing) {
5488                                 err = merge_extent_mapping(em_tree, existing,
5489                                                            em, start,
5490                                                            root->sectorsize);
5491                                 free_extent_map(existing);
5492                                 if (err) {
5493                                         free_extent_map(em);
5494                                         em = NULL;
5495                                 }
5496                         } else {
5497                                 err = -EIO;
5498                                 free_extent_map(em);
5499                                 em = NULL;
5500                         }
5501                 } else {
5502                         free_extent_map(em);
5503                         em = existing;
5504                         err = 0;
5505                 }
5506         }
5507         write_unlock(&em_tree->lock);
5508 out:
5509
5510         trace_btrfs_get_extent(root, em);
5511
5512         if (path)
5513                 btrfs_free_path(path);
5514         if (trans) {
5515                 ret = btrfs_end_transaction(trans, root);
5516                 if (!err)
5517                         err = ret;
5518         }
5519         if (err) {
5520                 free_extent_map(em);
5521                 return ERR_PTR(err);
5522         }
5523         BUG_ON(!em); /* Error is always set */
5524         return em;
5525 }
5526
5527 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5528                                            size_t pg_offset, u64 start, u64 len,
5529                                            int create)
5530 {
5531         struct extent_map *em;
5532         struct extent_map *hole_em = NULL;
5533         u64 range_start = start;
5534         u64 end;
5535         u64 found;
5536         u64 found_end;
5537         int err = 0;
5538
5539         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5540         if (IS_ERR(em))
5541                 return em;
5542         if (em) {
5543                 /*
5544                  * if our em maps to a hole, there might
5545                  * actually be delalloc bytes behind it
5546                  */
5547                 if (em->block_start != EXTENT_MAP_HOLE)
5548                         return em;
5549                 else
5550                         hole_em = em;
5551         }
5552
5553         /* check to see if we've wrapped (len == -1 or similar) */
5554         end = start + len;
5555         if (end < start)
5556                 end = (u64)-1;
5557         else
5558                 end -= 1;
5559
5560         em = NULL;
5561
5562         /* ok, we didn't find anything, lets look for delalloc */
5563         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5564                                  end, len, EXTENT_DELALLOC, 1);
5565         found_end = range_start + found;
5566         if (found_end < range_start)
5567                 found_end = (u64)-1;
5568
5569         /*
5570          * we didn't find anything useful, return
5571          * the original results from get_extent()
5572          */
5573         if (range_start > end || found_end <= start) {
5574                 em = hole_em;
5575                 hole_em = NULL;
5576                 goto out;
5577         }
5578
5579         /* adjust the range_start to make sure it doesn't
5580          * go backwards from the start they passed in
5581          */
5582         range_start = max(start,range_start);
5583         found = found_end - range_start;
5584
5585         if (found > 0) {
5586                 u64 hole_start = start;
5587                 u64 hole_len = len;
5588
5589                 em = alloc_extent_map();
5590                 if (!em) {
5591                         err = -ENOMEM;
5592                         goto out;
5593                 }
5594                 /*
5595                  * when btrfs_get_extent can't find anything it
5596                  * returns one huge hole
5597                  *
5598                  * make sure what it found really fits our range, and
5599                  * adjust to make sure it is based on the start from
5600                  * the caller
5601                  */
5602                 if (hole_em) {
5603                         u64 calc_end = extent_map_end(hole_em);
5604
5605                         if (calc_end <= start || (hole_em->start > end)) {
5606                                 free_extent_map(hole_em);
5607                                 hole_em = NULL;
5608                         } else {
5609                                 hole_start = max(hole_em->start, start);
5610                                 hole_len = calc_end - hole_start;
5611                         }
5612                 }
5613                 em->bdev = NULL;
5614                 if (hole_em && range_start > hole_start) {
5615                         /* our hole starts before our delalloc, so we
5616                          * have to return just the parts of the hole
5617                          * that go until  the delalloc starts
5618                          */
5619                         em->len = min(hole_len,
5620                                       range_start - hole_start);
5621                         em->start = hole_start;
5622                         em->orig_start = hole_start;
5623                         /*
5624                          * don't adjust block start at all,
5625                          * it is fixed at EXTENT_MAP_HOLE
5626                          */
5627                         em->block_start = hole_em->block_start;
5628                         em->block_len = hole_len;
5629                 } else {
5630                         em->start = range_start;
5631                         em->len = found;
5632                         em->orig_start = range_start;
5633                         em->block_start = EXTENT_MAP_DELALLOC;
5634                         em->block_len = found;
5635                 }
5636         } else if (hole_em) {
5637                 return hole_em;
5638         }
5639 out:
5640
5641         free_extent_map(hole_em);
5642         if (err) {
5643                 free_extent_map(em);
5644                 return ERR_PTR(err);
5645         }
5646         return em;
5647 }
5648
5649 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5650                                                   struct extent_map *em,
5651                                                   u64 start, u64 len)
5652 {
5653         struct btrfs_root *root = BTRFS_I(inode)->root;
5654         struct btrfs_trans_handle *trans;
5655         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5656         struct btrfs_key ins;
5657         u64 alloc_hint;
5658         int ret;
5659         bool insert = false;
5660
5661         /*
5662          * Ok if the extent map we looked up is a hole and is for the exact
5663          * range we want, there is no reason to allocate a new one, however if
5664          * it is not right then we need to free this one and drop the cache for
5665          * our range.
5666          */
5667         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5668             em->len != len) {
5669                 free_extent_map(em);
5670                 em = NULL;
5671                 insert = true;
5672                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5673         }
5674
5675         trans = btrfs_join_transaction(root);
5676         if (IS_ERR(trans))
5677                 return ERR_CAST(trans);
5678
5679         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5680                 btrfs_add_inode_defrag(trans, inode);
5681
5682         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5683
5684         alloc_hint = get_extent_allocation_hint(inode, start, len);
5685         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5686                                    alloc_hint, &ins, 1);
5687         if (ret) {
5688                 em = ERR_PTR(ret);
5689                 goto out;
5690         }
5691
5692         if (!em) {
5693                 em = alloc_extent_map();
5694                 if (!em) {
5695                         em = ERR_PTR(-ENOMEM);
5696                         goto out;
5697                 }
5698         }
5699
5700         em->start = start;
5701         em->orig_start = em->start;
5702         em->len = ins.offset;
5703
5704         em->block_start = ins.objectid;
5705         em->block_len = ins.offset;
5706         em->bdev = root->fs_info->fs_devices->latest_bdev;
5707
5708         /*
5709          * We need to do this because if we're using the original em we searched
5710          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5711          */
5712         em->flags = 0;
5713         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5714
5715         while (insert) {
5716                 write_lock(&em_tree->lock);
5717                 ret = add_extent_mapping(em_tree, em);
5718                 write_unlock(&em_tree->lock);
5719                 if (ret != -EEXIST)
5720                         break;
5721                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5722         }
5723
5724         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5725                                            ins.offset, ins.offset, 0);
5726         if (ret) {
5727                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5728                 em = ERR_PTR(ret);
5729         }
5730 out:
5731         btrfs_end_transaction(trans, root);
5732         return em;
5733 }
5734
5735 /*
5736  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5737  * block must be cow'd
5738  */
5739 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5740                                       struct inode *inode, u64 offset, u64 len)
5741 {
5742         struct btrfs_path *path;
5743         int ret;
5744         struct extent_buffer *leaf;
5745         struct btrfs_root *root = BTRFS_I(inode)->root;
5746         struct btrfs_file_extent_item *fi;
5747         struct btrfs_key key;
5748         u64 disk_bytenr;
5749         u64 backref_offset;
5750         u64 extent_end;
5751         u64 num_bytes;
5752         int slot;
5753         int found_type;
5754
5755         path = btrfs_alloc_path();
5756         if (!path)
5757                 return -ENOMEM;
5758
5759         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5760                                        offset, 0);
5761         if (ret < 0)
5762                 goto out;
5763
5764         slot = path->slots[0];
5765         if (ret == 1) {
5766                 if (slot == 0) {
5767                         /* can't find the item, must cow */
5768                         ret = 0;
5769                         goto out;
5770                 }
5771                 slot--;
5772         }
5773         ret = 0;
5774         leaf = path->nodes[0];
5775         btrfs_item_key_to_cpu(leaf, &key, slot);
5776         if (key.objectid != btrfs_ino(inode) ||
5777             key.type != BTRFS_EXTENT_DATA_KEY) {
5778                 /* not our file or wrong item type, must cow */
5779                 goto out;
5780         }
5781
5782         if (key.offset > offset) {
5783                 /* Wrong offset, must cow */
5784                 goto out;
5785         }
5786
5787         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5788         found_type = btrfs_file_extent_type(leaf, fi);
5789         if (found_type != BTRFS_FILE_EXTENT_REG &&
5790             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5791                 /* not a regular extent, must cow */
5792                 goto out;
5793         }
5794         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5795         backref_offset = btrfs_file_extent_offset(leaf, fi);
5796
5797         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5798         if (extent_end < offset + len) {
5799                 /* extent doesn't include our full range, must cow */
5800                 goto out;
5801         }
5802
5803         if (btrfs_extent_readonly(root, disk_bytenr))
5804                 goto out;
5805
5806         /*
5807          * look for other files referencing this extent, if we
5808          * find any we must cow
5809          */
5810         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5811                                   key.offset - backref_offset, disk_bytenr))
5812                 goto out;
5813
5814         /*
5815          * adjust disk_bytenr and num_bytes to cover just the bytes
5816          * in this extent we are about to write.  If there
5817          * are any csums in that range we have to cow in order
5818          * to keep the csums correct
5819          */
5820         disk_bytenr += backref_offset;
5821         disk_bytenr += offset - key.offset;
5822         num_bytes = min(offset + len, extent_end) - offset;
5823         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5824                                 goto out;
5825         /*
5826          * all of the above have passed, it is safe to overwrite this extent
5827          * without cow
5828          */
5829         ret = 1;
5830 out:
5831         btrfs_free_path(path);
5832         return ret;
5833 }
5834
5835 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5836                               struct extent_state **cached_state, int writing)
5837 {
5838         struct btrfs_ordered_extent *ordered;
5839         int ret = 0;
5840
5841         while (1) {
5842                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5843                                  0, cached_state);
5844                 /*
5845                  * We're concerned with the entire range that we're going to be
5846                  * doing DIO to, so we need to make sure theres no ordered
5847                  * extents in this range.
5848                  */
5849                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5850                                                      lockend - lockstart + 1);
5851
5852                 /*
5853                  * We need to make sure there are no buffered pages in this
5854                  * range either, we could have raced between the invalidate in
5855                  * generic_file_direct_write and locking the extent.  The
5856                  * invalidate needs to happen so that reads after a write do not
5857                  * get stale data.
5858                  */
5859                 if (!ordered && (!writing ||
5860                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5861                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5862                                     *cached_state)))
5863                         break;
5864
5865                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5866                                      cached_state, GFP_NOFS);
5867
5868                 if (ordered) {
5869                         btrfs_start_ordered_extent(inode, ordered, 1);
5870                         btrfs_put_ordered_extent(ordered);
5871                 } else {
5872                         /* Screw you mmap */
5873                         ret = filemap_write_and_wait_range(inode->i_mapping,
5874                                                            lockstart,
5875                                                            lockend);
5876                         if (ret)
5877                                 break;
5878
5879                         /*
5880                          * If we found a page that couldn't be invalidated just
5881                          * fall back to buffered.
5882                          */
5883                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5884                                         lockstart >> PAGE_CACHE_SHIFT,
5885                                         lockend >> PAGE_CACHE_SHIFT);
5886                         if (ret)
5887                                 break;
5888                 }
5889
5890                 cond_resched();
5891         }
5892
5893         return ret;
5894 }
5895
5896 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5897                                            u64 len, u64 orig_start,
5898                                            u64 block_start, u64 block_len,
5899                                            int type)
5900 {
5901         struct extent_map_tree *em_tree;
5902         struct extent_map *em;
5903         struct btrfs_root *root = BTRFS_I(inode)->root;
5904         int ret;
5905
5906         em_tree = &BTRFS_I(inode)->extent_tree;
5907         em = alloc_extent_map();
5908         if (!em)
5909                 return ERR_PTR(-ENOMEM);
5910
5911         em->start = start;
5912         em->orig_start = orig_start;
5913         em->len = len;
5914         em->block_len = block_len;
5915         em->block_start = block_start;
5916         em->bdev = root->fs_info->fs_devices->latest_bdev;
5917         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5918         if (type == BTRFS_ORDERED_PREALLOC)
5919                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5920
5921         do {
5922                 btrfs_drop_extent_cache(inode, em->start,
5923                                 em->start + em->len - 1, 0);
5924                 write_lock(&em_tree->lock);
5925                 ret = add_extent_mapping(em_tree, em);
5926                 write_unlock(&em_tree->lock);
5927         } while (ret == -EEXIST);
5928
5929         if (ret) {
5930                 free_extent_map(em);
5931                 return ERR_PTR(ret);
5932         }
5933
5934         return em;
5935 }
5936
5937
5938 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5939                                    struct buffer_head *bh_result, int create)
5940 {
5941         struct extent_map *em;
5942         struct btrfs_root *root = BTRFS_I(inode)->root;
5943         struct extent_state *cached_state = NULL;
5944         u64 start = iblock << inode->i_blkbits;
5945         u64 lockstart, lockend;
5946         u64 len = bh_result->b_size;
5947         struct btrfs_trans_handle *trans;
5948         int unlock_bits = EXTENT_LOCKED;
5949         int ret;
5950
5951         if (create) {
5952                 ret = btrfs_delalloc_reserve_space(inode, len);
5953                 if (ret)
5954                         return ret;
5955                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5956         } else {
5957                 len = min_t(u64, len, root->sectorsize);
5958         }
5959
5960         lockstart = start;
5961         lockend = start + len - 1;
5962
5963         /*
5964          * If this errors out it's because we couldn't invalidate pagecache for
5965          * this range and we need to fallback to buffered.
5966          */
5967         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5968                 return -ENOTBLK;
5969
5970         if (create) {
5971                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5972                                      lockend, EXTENT_DELALLOC, NULL,
5973                                      &cached_state, GFP_NOFS);
5974                 if (ret)
5975                         goto unlock_err;
5976         }
5977
5978         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5979         if (IS_ERR(em)) {
5980                 ret = PTR_ERR(em);
5981                 goto unlock_err;
5982         }
5983
5984         /*
5985          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5986          * io.  INLINE is special, and we could probably kludge it in here, but
5987          * it's still buffered so for safety lets just fall back to the generic
5988          * buffered path.
5989          *
5990          * For COMPRESSED we _have_ to read the entire extent in so we can
5991          * decompress it, so there will be buffering required no matter what we
5992          * do, so go ahead and fallback to buffered.
5993          *
5994          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5995          * to buffered IO.  Don't blame me, this is the price we pay for using
5996          * the generic code.
5997          */
5998         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5999             em->block_start == EXTENT_MAP_INLINE) {
6000                 free_extent_map(em);
6001                 ret = -ENOTBLK;
6002                 goto unlock_err;
6003         }
6004
6005         /* Just a good old fashioned hole, return */
6006         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6007                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6008                 free_extent_map(em);
6009                 ret = 0;
6010                 goto unlock_err;
6011         }
6012
6013         /*
6014          * We don't allocate a new extent in the following cases
6015          *
6016          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6017          * existing extent.
6018          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6019          * just use the extent.
6020          *
6021          */
6022         if (!create) {
6023                 len = min(len, em->len - (start - em->start));
6024                 lockstart = start + len;
6025                 goto unlock;
6026         }
6027
6028         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6029             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6030              em->block_start != EXTENT_MAP_HOLE)) {
6031                 int type;
6032                 int ret;
6033                 u64 block_start;
6034
6035                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6036                         type = BTRFS_ORDERED_PREALLOC;
6037                 else
6038                         type = BTRFS_ORDERED_NOCOW;
6039                 len = min(len, em->len - (start - em->start));
6040                 block_start = em->block_start + (start - em->start);
6041
6042                 /*
6043                  * we're not going to log anything, but we do need
6044                  * to make sure the current transaction stays open
6045                  * while we look for nocow cross refs
6046                  */
6047                 trans = btrfs_join_transaction(root);
6048                 if (IS_ERR(trans))
6049                         goto must_cow;
6050
6051                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6052                         u64 orig_start = em->start;
6053
6054                         if (type == BTRFS_ORDERED_PREALLOC) {
6055                                 free_extent_map(em);
6056                                 em = create_pinned_em(inode, start, len,
6057                                                        orig_start,
6058                                                        block_start, len, type);
6059                                 if (IS_ERR(em)) {
6060                                         btrfs_end_transaction(trans, root);
6061                                         goto unlock_err;
6062                                 }
6063                         }
6064
6065                         ret = btrfs_add_ordered_extent_dio(inode, start,
6066                                            block_start, len, len, type);
6067                         btrfs_end_transaction(trans, root);
6068                         if (ret) {
6069                                 free_extent_map(em);
6070                                 goto unlock_err;
6071                         }
6072                         goto unlock;
6073                 }
6074                 btrfs_end_transaction(trans, root);
6075         }
6076 must_cow:
6077         /*
6078          * this will cow the extent, reset the len in case we changed
6079          * it above
6080          */
6081         len = bh_result->b_size;
6082         em = btrfs_new_extent_direct(inode, em, start, len);
6083         if (IS_ERR(em)) {
6084                 ret = PTR_ERR(em);
6085                 goto unlock_err;
6086         }
6087         len = min(len, em->len - (start - em->start));
6088 unlock:
6089         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6090                 inode->i_blkbits;
6091         bh_result->b_size = len;
6092         bh_result->b_bdev = em->bdev;
6093         set_buffer_mapped(bh_result);
6094         if (create) {
6095                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6096                         set_buffer_new(bh_result);
6097
6098                 /*
6099                  * Need to update the i_size under the extent lock so buffered
6100                  * readers will get the updated i_size when we unlock.
6101                  */
6102                 if (start + len > i_size_read(inode))
6103                         i_size_write(inode, start + len);
6104         }
6105
6106         /*
6107          * In the case of write we need to clear and unlock the entire range,
6108          * in the case of read we need to unlock only the end area that we
6109          * aren't using if there is any left over space.
6110          */
6111         if (lockstart < lockend) {
6112                 if (create && len < lockend - lockstart) {
6113                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6114                                          lockstart + len - 1,
6115                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6116                                          &cached_state, GFP_NOFS);
6117                         /*
6118                          * Beside unlock, we also need to cleanup reserved space
6119                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6120                          */
6121                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6122                                          lockstart + len, lockend,
6123                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6124                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6125                 } else {
6126                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6127                                          lockend, unlock_bits, 1, 0,
6128                                          &cached_state, GFP_NOFS);
6129                 }
6130         } else {
6131                 free_extent_state(cached_state);
6132         }
6133
6134         free_extent_map(em);
6135
6136         return 0;
6137
6138 unlock_err:
6139         if (create)
6140                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6141
6142         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6143                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6144         return ret;
6145 }
6146
6147 struct btrfs_dio_private {
6148         struct inode *inode;
6149         u64 logical_offset;
6150         u64 disk_bytenr;
6151         u64 bytes;
6152         void *private;
6153
6154         /* number of bios pending for this dio */
6155         atomic_t pending_bios;
6156
6157         /* IO errors */
6158         int errors;
6159
6160         struct bio *orig_bio;
6161 };
6162
6163 static void btrfs_endio_direct_read(struct bio *bio, int err)
6164 {
6165         struct btrfs_dio_private *dip = bio->bi_private;
6166         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6167         struct bio_vec *bvec = bio->bi_io_vec;
6168         struct inode *inode = dip->inode;
6169         struct btrfs_root *root = BTRFS_I(inode)->root;
6170         u64 start;
6171
6172         start = dip->logical_offset;
6173         do {
6174                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6175                         struct page *page = bvec->bv_page;
6176                         char *kaddr;
6177                         u32 csum = ~(u32)0;
6178                         u64 private = ~(u32)0;
6179                         unsigned long flags;
6180
6181                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6182                                               start, &private))
6183                                 goto failed;
6184                         local_irq_save(flags);
6185                         kaddr = kmap_atomic(page);
6186                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6187                                                csum, bvec->bv_len);
6188                         btrfs_csum_final(csum, (char *)&csum);
6189                         kunmap_atomic(kaddr);
6190                         local_irq_restore(flags);
6191
6192                         flush_dcache_page(bvec->bv_page);
6193                         if (csum != private) {
6194 failed:
6195                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6196                                       " %llu csum %u private %u\n",
6197                                       (unsigned long long)btrfs_ino(inode),
6198                                       (unsigned long long)start,
6199                                       csum, (unsigned)private);
6200                                 err = -EIO;
6201                         }
6202                 }
6203
6204                 start += bvec->bv_len;
6205                 bvec++;
6206         } while (bvec <= bvec_end);
6207
6208         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6209                       dip->logical_offset + dip->bytes - 1);
6210         bio->bi_private = dip->private;
6211
6212         kfree(dip);
6213
6214         /* If we had a csum failure make sure to clear the uptodate flag */
6215         if (err)
6216                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6217         dio_end_io(bio, err);
6218 }
6219
6220 static void btrfs_endio_direct_write(struct bio *bio, int err)
6221 {
6222         struct btrfs_dio_private *dip = bio->bi_private;
6223         struct inode *inode = dip->inode;
6224         struct btrfs_root *root = BTRFS_I(inode)->root;
6225         struct btrfs_ordered_extent *ordered = NULL;
6226         u64 ordered_offset = dip->logical_offset;
6227         u64 ordered_bytes = dip->bytes;
6228         int ret;
6229
6230         if (err)
6231                 goto out_done;
6232 again:
6233         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6234                                                    &ordered_offset,
6235                                                    ordered_bytes, !err);
6236         if (!ret)
6237                 goto out_test;
6238
6239         ordered->work.func = finish_ordered_fn;
6240         ordered->work.flags = 0;
6241         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6242                            &ordered->work);
6243 out_test:
6244         /*
6245          * our bio might span multiple ordered extents.  If we haven't
6246          * completed the accounting for the whole dio, go back and try again
6247          */
6248         if (ordered_offset < dip->logical_offset + dip->bytes) {
6249                 ordered_bytes = dip->logical_offset + dip->bytes -
6250                         ordered_offset;
6251                 ordered = NULL;
6252                 goto again;
6253         }
6254 out_done:
6255         bio->bi_private = dip->private;
6256
6257         kfree(dip);
6258
6259         /* If we had an error make sure to clear the uptodate flag */
6260         if (err)
6261                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6262         dio_end_io(bio, err);
6263 }
6264
6265 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6266                                     struct bio *bio, int mirror_num,
6267                                     unsigned long bio_flags, u64 offset)
6268 {
6269         int ret;
6270         struct btrfs_root *root = BTRFS_I(inode)->root;
6271         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6272         BUG_ON(ret); /* -ENOMEM */
6273         return 0;
6274 }
6275
6276 static void btrfs_end_dio_bio(struct bio *bio, int err)
6277 {
6278         struct btrfs_dio_private *dip = bio->bi_private;
6279
6280         if (err) {
6281                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6282                       "sector %#Lx len %u err no %d\n",
6283                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6284                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6285                 dip->errors = 1;
6286
6287                 /*
6288                  * before atomic variable goto zero, we must make sure
6289                  * dip->errors is perceived to be set.
6290                  */
6291                 smp_mb__before_atomic_dec();
6292         }
6293
6294         /* if there are more bios still pending for this dio, just exit */
6295         if (!atomic_dec_and_test(&dip->pending_bios))
6296                 goto out;
6297
6298         if (dip->errors)
6299                 bio_io_error(dip->orig_bio);
6300         else {
6301                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6302                 bio_endio(dip->orig_bio, 0);
6303         }
6304 out:
6305         bio_put(bio);
6306 }
6307
6308 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6309                                        u64 first_sector, gfp_t gfp_flags)
6310 {
6311         int nr_vecs = bio_get_nr_vecs(bdev);
6312         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6313 }
6314
6315 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6316                                          int rw, u64 file_offset, int skip_sum,
6317                                          int async_submit)
6318 {
6319         int write = rw & REQ_WRITE;
6320         struct btrfs_root *root = BTRFS_I(inode)->root;
6321         int ret;
6322
6323         bio_get(bio);
6324
6325         if (!write) {
6326                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6327                 if (ret)
6328                         goto err;
6329         }
6330
6331         if (skip_sum)
6332                 goto map;
6333
6334         if (write && async_submit) {
6335                 ret = btrfs_wq_submit_bio(root->fs_info,
6336                                    inode, rw, bio, 0, 0,
6337                                    file_offset,
6338                                    __btrfs_submit_bio_start_direct_io,
6339                                    __btrfs_submit_bio_done);
6340                 goto err;
6341         } else if (write) {
6342                 /*
6343                  * If we aren't doing async submit, calculate the csum of the
6344                  * bio now.
6345                  */
6346                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6347                 if (ret)
6348                         goto err;
6349         } else if (!skip_sum) {
6350                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6351                 if (ret)
6352                         goto err;
6353         }
6354
6355 map:
6356         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6357 err:
6358         bio_put(bio);
6359         return ret;
6360 }
6361
6362 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6363                                     int skip_sum)
6364 {
6365         struct inode *inode = dip->inode;
6366         struct btrfs_root *root = BTRFS_I(inode)->root;
6367         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6368         struct bio *bio;
6369         struct bio *orig_bio = dip->orig_bio;
6370         struct bio_vec *bvec = orig_bio->bi_io_vec;
6371         u64 start_sector = orig_bio->bi_sector;
6372         u64 file_offset = dip->logical_offset;
6373         u64 submit_len = 0;
6374         u64 map_length;
6375         int nr_pages = 0;
6376         int ret = 0;
6377         int async_submit = 0;
6378
6379         map_length = orig_bio->bi_size;
6380         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6381                               &map_length, NULL, 0);
6382         if (ret) {
6383                 bio_put(orig_bio);
6384                 return -EIO;
6385         }
6386
6387         if (map_length >= orig_bio->bi_size) {
6388                 bio = orig_bio;
6389                 goto submit;
6390         }
6391
6392         async_submit = 1;
6393         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6394         if (!bio)
6395                 return -ENOMEM;
6396         bio->bi_private = dip;
6397         bio->bi_end_io = btrfs_end_dio_bio;
6398         atomic_inc(&dip->pending_bios);
6399
6400         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6401                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6402                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6403                                  bvec->bv_offset) < bvec->bv_len)) {
6404                         /*
6405                          * inc the count before we submit the bio so
6406                          * we know the end IO handler won't happen before
6407                          * we inc the count. Otherwise, the dip might get freed
6408                          * before we're done setting it up
6409                          */
6410                         atomic_inc(&dip->pending_bios);
6411                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6412                                                      file_offset, skip_sum,
6413                                                      async_submit);
6414                         if (ret) {
6415                                 bio_put(bio);
6416                                 atomic_dec(&dip->pending_bios);
6417                                 goto out_err;
6418                         }
6419
6420                         start_sector += submit_len >> 9;
6421                         file_offset += submit_len;
6422
6423                         submit_len = 0;
6424                         nr_pages = 0;
6425
6426                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6427                                                   start_sector, GFP_NOFS);
6428                         if (!bio)
6429                                 goto out_err;
6430                         bio->bi_private = dip;
6431                         bio->bi_end_io = btrfs_end_dio_bio;
6432
6433                         map_length = orig_bio->bi_size;
6434                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6435                                               &map_length, NULL, 0);
6436                         if (ret) {
6437                                 bio_put(bio);
6438                                 goto out_err;
6439                         }
6440                 } else {
6441                         submit_len += bvec->bv_len;
6442                         nr_pages ++;
6443                         bvec++;
6444                 }
6445         }
6446
6447 submit:
6448         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6449                                      async_submit);
6450         if (!ret)
6451                 return 0;
6452
6453         bio_put(bio);
6454 out_err:
6455         dip->errors = 1;
6456         /*
6457          * before atomic variable goto zero, we must
6458          * make sure dip->errors is perceived to be set.
6459          */
6460         smp_mb__before_atomic_dec();
6461         if (atomic_dec_and_test(&dip->pending_bios))
6462                 bio_io_error(dip->orig_bio);
6463
6464         /* bio_end_io() will handle error, so we needn't return it */
6465         return 0;
6466 }
6467
6468 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6469                                 loff_t file_offset)
6470 {
6471         struct btrfs_root *root = BTRFS_I(inode)->root;
6472         struct btrfs_dio_private *dip;
6473         struct bio_vec *bvec = bio->bi_io_vec;
6474         int skip_sum;
6475         int write = rw & REQ_WRITE;
6476         int ret = 0;
6477
6478         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6479
6480         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6481         if (!dip) {
6482                 ret = -ENOMEM;
6483                 goto free_ordered;
6484         }
6485
6486         dip->private = bio->bi_private;
6487         dip->inode = inode;
6488         dip->logical_offset = file_offset;
6489
6490         dip->bytes = 0;
6491         do {
6492                 dip->bytes += bvec->bv_len;
6493                 bvec++;
6494         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6495
6496         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6497         bio->bi_private = dip;
6498         dip->errors = 0;
6499         dip->orig_bio = bio;
6500         atomic_set(&dip->pending_bios, 0);
6501
6502         if (write)
6503                 bio->bi_end_io = btrfs_endio_direct_write;
6504         else
6505                 bio->bi_end_io = btrfs_endio_direct_read;
6506
6507         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6508         if (!ret)
6509                 return;
6510 free_ordered:
6511         /*
6512          * If this is a write, we need to clean up the reserved space and kill
6513          * the ordered extent.
6514          */
6515         if (write) {
6516                 struct btrfs_ordered_extent *ordered;
6517                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6518                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6519                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6520                         btrfs_free_reserved_extent(root, ordered->start,
6521                                                    ordered->disk_len);
6522                 btrfs_put_ordered_extent(ordered);
6523                 btrfs_put_ordered_extent(ordered);
6524         }
6525         bio_endio(bio, ret);
6526 }
6527
6528 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6529                         const struct iovec *iov, loff_t offset,
6530                         unsigned long nr_segs)
6531 {
6532         int seg;
6533         int i;
6534         size_t size;
6535         unsigned long addr;
6536         unsigned blocksize_mask = root->sectorsize - 1;
6537         ssize_t retval = -EINVAL;
6538         loff_t end = offset;
6539
6540         if (offset & blocksize_mask)
6541                 goto out;
6542
6543         /* Check the memory alignment.  Blocks cannot straddle pages */
6544         for (seg = 0; seg < nr_segs; seg++) {
6545                 addr = (unsigned long)iov[seg].iov_base;
6546                 size = iov[seg].iov_len;
6547                 end += size;
6548                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6549                         goto out;
6550
6551                 /* If this is a write we don't need to check anymore */
6552                 if (rw & WRITE)
6553                         continue;
6554
6555                 /*
6556                  * Check to make sure we don't have duplicate iov_base's in this
6557                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6558                  * when reading back.
6559                  */
6560                 for (i = seg + 1; i < nr_segs; i++) {
6561                         if (iov[seg].iov_base == iov[i].iov_base)
6562                                 goto out;
6563                 }
6564         }
6565         retval = 0;
6566 out:
6567         return retval;
6568 }
6569
6570 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6571                         const struct iovec *iov, loff_t offset,
6572                         unsigned long nr_segs)
6573 {
6574         struct file *file = iocb->ki_filp;
6575         struct inode *inode = file->f_mapping->host;
6576
6577         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6578                             offset, nr_segs))
6579                 return 0;
6580
6581         return __blockdev_direct_IO(rw, iocb, inode,
6582                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6583                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6584                    btrfs_submit_direct, 0);
6585 }
6586
6587 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6588                 __u64 start, __u64 len)
6589 {
6590         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6591 }
6592
6593 int btrfs_readpage(struct file *file, struct page *page)
6594 {
6595         struct extent_io_tree *tree;
6596         tree = &BTRFS_I(page->mapping->host)->io_tree;
6597         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6598 }
6599
6600 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6601 {
6602         struct extent_io_tree *tree;
6603
6604
6605         if (current->flags & PF_MEMALLOC) {
6606                 redirty_page_for_writepage(wbc, page);
6607                 unlock_page(page);
6608                 return 0;
6609         }
6610         tree = &BTRFS_I(page->mapping->host)->io_tree;
6611         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6612 }
6613
6614 int btrfs_writepages(struct address_space *mapping,
6615                      struct writeback_control *wbc)
6616 {
6617         struct extent_io_tree *tree;
6618
6619         tree = &BTRFS_I(mapping->host)->io_tree;
6620         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6621 }
6622
6623 static int
6624 btrfs_readpages(struct file *file, struct address_space *mapping,
6625                 struct list_head *pages, unsigned nr_pages)
6626 {
6627         struct extent_io_tree *tree;
6628         tree = &BTRFS_I(mapping->host)->io_tree;
6629         return extent_readpages(tree, mapping, pages, nr_pages,
6630                                 btrfs_get_extent);
6631 }
6632 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6633 {
6634         struct extent_io_tree *tree;
6635         struct extent_map_tree *map;
6636         int ret;
6637
6638         tree = &BTRFS_I(page->mapping->host)->io_tree;
6639         map = &BTRFS_I(page->mapping->host)->extent_tree;
6640         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6641         if (ret == 1) {
6642                 ClearPagePrivate(page);
6643                 set_page_private(page, 0);
6644                 page_cache_release(page);
6645         }
6646         return ret;
6647 }
6648
6649 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6650 {
6651         if (PageWriteback(page) || PageDirty(page))
6652                 return 0;
6653         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6654 }
6655
6656 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6657 {
6658         struct inode *inode = page->mapping->host;
6659         struct extent_io_tree *tree;
6660         struct btrfs_ordered_extent *ordered;
6661         struct extent_state *cached_state = NULL;
6662         u64 page_start = page_offset(page);
6663         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6664
6665         /*
6666          * we have the page locked, so new writeback can't start,
6667          * and the dirty bit won't be cleared while we are here.
6668          *
6669          * Wait for IO on this page so that we can safely clear
6670          * the PagePrivate2 bit and do ordered accounting
6671          */
6672         wait_on_page_writeback(page);
6673
6674         tree = &BTRFS_I(inode)->io_tree;
6675         if (offset) {
6676                 btrfs_releasepage(page, GFP_NOFS);
6677                 return;
6678         }
6679         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6680         ordered = btrfs_lookup_ordered_extent(inode,
6681                                            page_offset(page));
6682         if (ordered) {
6683                 /*
6684                  * IO on this page will never be started, so we need
6685                  * to account for any ordered extents now
6686                  */
6687                 clear_extent_bit(tree, page_start, page_end,
6688                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6689                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6690                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6691                 /*
6692                  * whoever cleared the private bit is responsible
6693                  * for the finish_ordered_io
6694                  */
6695                 if (TestClearPagePrivate2(page) &&
6696                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6697                                                    PAGE_CACHE_SIZE, 1)) {
6698                         btrfs_finish_ordered_io(ordered);
6699                 }
6700                 btrfs_put_ordered_extent(ordered);
6701                 cached_state = NULL;
6702                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6703         }
6704         clear_extent_bit(tree, page_start, page_end,
6705                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6706                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6707                  &cached_state, GFP_NOFS);
6708         __btrfs_releasepage(page, GFP_NOFS);
6709
6710         ClearPageChecked(page);
6711         if (PagePrivate(page)) {
6712                 ClearPagePrivate(page);
6713                 set_page_private(page, 0);
6714                 page_cache_release(page);
6715         }
6716 }
6717
6718 /*
6719  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6720  * called from a page fault handler when a page is first dirtied. Hence we must
6721  * be careful to check for EOF conditions here. We set the page up correctly
6722  * for a written page which means we get ENOSPC checking when writing into
6723  * holes and correct delalloc and unwritten extent mapping on filesystems that
6724  * support these features.
6725  *
6726  * We are not allowed to take the i_mutex here so we have to play games to
6727  * protect against truncate races as the page could now be beyond EOF.  Because
6728  * vmtruncate() writes the inode size before removing pages, once we have the
6729  * page lock we can determine safely if the page is beyond EOF. If it is not
6730  * beyond EOF, then the page is guaranteed safe against truncation until we
6731  * unlock the page.
6732  */
6733 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6734 {
6735         struct page *page = vmf->page;
6736         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6737         struct btrfs_root *root = BTRFS_I(inode)->root;
6738         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6739         struct btrfs_ordered_extent *ordered;
6740         struct extent_state *cached_state = NULL;
6741         char *kaddr;
6742         unsigned long zero_start;
6743         loff_t size;
6744         int ret;
6745         int reserved = 0;
6746         u64 page_start;
6747         u64 page_end;
6748
6749         sb_start_pagefault(inode->i_sb);
6750         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6751         if (!ret) {
6752                 ret = file_update_time(vma->vm_file);
6753                 reserved = 1;
6754         }
6755         if (ret) {
6756                 if (ret == -ENOMEM)
6757                         ret = VM_FAULT_OOM;
6758                 else /* -ENOSPC, -EIO, etc */
6759                         ret = VM_FAULT_SIGBUS;
6760                 if (reserved)
6761                         goto out;
6762                 goto out_noreserve;
6763         }
6764
6765         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6766 again:
6767         lock_page(page);
6768         size = i_size_read(inode);
6769         page_start = page_offset(page);
6770         page_end = page_start + PAGE_CACHE_SIZE - 1;
6771
6772         if ((page->mapping != inode->i_mapping) ||
6773             (page_start >= size)) {
6774                 /* page got truncated out from underneath us */
6775                 goto out_unlock;
6776         }
6777         wait_on_page_writeback(page);
6778
6779         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6780         set_page_extent_mapped(page);
6781
6782         /*
6783          * we can't set the delalloc bits if there are pending ordered
6784          * extents.  Drop our locks and wait for them to finish
6785          */
6786         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6787         if (ordered) {
6788                 unlock_extent_cached(io_tree, page_start, page_end,
6789                                      &cached_state, GFP_NOFS);
6790                 unlock_page(page);
6791                 btrfs_start_ordered_extent(inode, ordered, 1);
6792                 btrfs_put_ordered_extent(ordered);
6793                 goto again;
6794         }
6795
6796         /*
6797          * XXX - page_mkwrite gets called every time the page is dirtied, even
6798          * if it was already dirty, so for space accounting reasons we need to
6799          * clear any delalloc bits for the range we are fixing to save.  There
6800          * is probably a better way to do this, but for now keep consistent with
6801          * prepare_pages in the normal write path.
6802          */
6803         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6804                           EXTENT_DIRTY | EXTENT_DELALLOC |
6805                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6806                           0, 0, &cached_state, GFP_NOFS);
6807
6808         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6809                                         &cached_state);
6810         if (ret) {
6811                 unlock_extent_cached(io_tree, page_start, page_end,
6812                                      &cached_state, GFP_NOFS);
6813                 ret = VM_FAULT_SIGBUS;
6814                 goto out_unlock;
6815         }
6816         ret = 0;
6817
6818         /* page is wholly or partially inside EOF */
6819         if (page_start + PAGE_CACHE_SIZE > size)
6820                 zero_start = size & ~PAGE_CACHE_MASK;
6821         else
6822                 zero_start = PAGE_CACHE_SIZE;
6823
6824         if (zero_start != PAGE_CACHE_SIZE) {
6825                 kaddr = kmap(page);
6826                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6827                 flush_dcache_page(page);
6828                 kunmap(page);
6829         }
6830         ClearPageChecked(page);
6831         set_page_dirty(page);
6832         SetPageUptodate(page);
6833
6834         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6835         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6836         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6837
6838         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6839
6840 out_unlock:
6841         if (!ret) {
6842                 sb_end_pagefault(inode->i_sb);
6843                 return VM_FAULT_LOCKED;
6844         }
6845         unlock_page(page);
6846 out:
6847         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6848 out_noreserve:
6849         sb_end_pagefault(inode->i_sb);
6850         return ret;
6851 }
6852
6853 static int btrfs_truncate(struct inode *inode)
6854 {
6855         struct btrfs_root *root = BTRFS_I(inode)->root;
6856         struct btrfs_block_rsv *rsv;
6857         int ret;
6858         int err = 0;
6859         struct btrfs_trans_handle *trans;
6860         unsigned long nr;
6861         u64 mask = root->sectorsize - 1;
6862         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6863
6864         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6865         if (ret)
6866                 return ret;
6867
6868         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6869         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6870
6871         /*
6872          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6873          * 3 things going on here
6874          *
6875          * 1) We need to reserve space for our orphan item and the space to
6876          * delete our orphan item.  Lord knows we don't want to have a dangling
6877          * orphan item because we didn't reserve space to remove it.
6878          *
6879          * 2) We need to reserve space to update our inode.
6880          *
6881          * 3) We need to have something to cache all the space that is going to
6882          * be free'd up by the truncate operation, but also have some slack
6883          * space reserved in case it uses space during the truncate (thank you
6884          * very much snapshotting).
6885          *
6886          * And we need these to all be seperate.  The fact is we can use alot of
6887          * space doing the truncate, and we have no earthly idea how much space
6888          * we will use, so we need the truncate reservation to be seperate so it
6889          * doesn't end up using space reserved for updating the inode or
6890          * removing the orphan item.  We also need to be able to stop the
6891          * transaction and start a new one, which means we need to be able to
6892          * update the inode several times, and we have no idea of knowing how
6893          * many times that will be, so we can't just reserve 1 item for the
6894          * entirety of the opration, so that has to be done seperately as well.
6895          * Then there is the orphan item, which does indeed need to be held on
6896          * to for the whole operation, and we need nobody to touch this reserved
6897          * space except the orphan code.
6898          *
6899          * So that leaves us with
6900          *
6901          * 1) root->orphan_block_rsv - for the orphan deletion.
6902          * 2) rsv - for the truncate reservation, which we will steal from the
6903          * transaction reservation.
6904          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6905          * updating the inode.
6906          */
6907         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6908         if (!rsv)
6909                 return -ENOMEM;
6910         rsv->size = min_size;
6911         rsv->failfast = 1;
6912
6913         /*
6914          * 1 for the truncate slack space
6915          * 1 for the orphan item we're going to add
6916          * 1 for the orphan item deletion
6917          * 1 for updating the inode.
6918          */
6919         trans = btrfs_start_transaction(root, 4);
6920         if (IS_ERR(trans)) {
6921                 err = PTR_ERR(trans);
6922                 goto out;
6923         }
6924
6925         /* Migrate the slack space for the truncate to our reserve */
6926         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6927                                       min_size);
6928         BUG_ON(ret);
6929
6930         ret = btrfs_orphan_add(trans, inode);
6931         if (ret) {
6932                 btrfs_end_transaction(trans, root);
6933                 goto out;
6934         }
6935
6936         /*
6937          * setattr is responsible for setting the ordered_data_close flag,
6938          * but that is only tested during the last file release.  That
6939          * could happen well after the next commit, leaving a great big
6940          * window where new writes may get lost if someone chooses to write
6941          * to this file after truncating to zero
6942          *
6943          * The inode doesn't have any dirty data here, and so if we commit
6944          * this is a noop.  If someone immediately starts writing to the inode
6945          * it is very likely we'll catch some of their writes in this
6946          * transaction, and the commit will find this file on the ordered
6947          * data list with good things to send down.
6948          *
6949          * This is a best effort solution, there is still a window where
6950          * using truncate to replace the contents of the file will
6951          * end up with a zero length file after a crash.
6952          */
6953         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6954                                            &BTRFS_I(inode)->runtime_flags))
6955                 btrfs_add_ordered_operation(trans, root, inode);
6956
6957         /*
6958          * So if we truncate and then write and fsync we normally would just
6959          * write the extents that changed, which is a problem if we need to
6960          * first truncate that entire inode.  So set this flag so we write out
6961          * all of the extents in the inode to the sync log so we're completely
6962          * safe.
6963          */
6964         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6965         trans->block_rsv = rsv;
6966
6967         while (1) {
6968                 ret = btrfs_truncate_inode_items(trans, root, inode,
6969                                                  inode->i_size,
6970                                                  BTRFS_EXTENT_DATA_KEY);
6971                 if (ret != -ENOSPC) {
6972                         err = ret;
6973                         break;
6974                 }
6975
6976                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6977                 ret = btrfs_update_inode(trans, root, inode);
6978                 if (ret) {
6979                         err = ret;
6980                         break;
6981                 }
6982
6983                 nr = trans->blocks_used;
6984                 btrfs_end_transaction(trans, root);
6985                 btrfs_btree_balance_dirty(root, nr);
6986
6987                 trans = btrfs_start_transaction(root, 2);
6988                 if (IS_ERR(trans)) {
6989                         ret = err = PTR_ERR(trans);
6990                         trans = NULL;
6991                         break;
6992                 }
6993
6994                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
6995                                               rsv, min_size);
6996                 BUG_ON(ret);    /* shouldn't happen */
6997                 trans->block_rsv = rsv;
6998         }
6999
7000         if (ret == 0 && inode->i_nlink > 0) {
7001                 trans->block_rsv = root->orphan_block_rsv;
7002                 ret = btrfs_orphan_del(trans, inode);
7003                 if (ret)
7004                         err = ret;
7005         } else if (ret && inode->i_nlink > 0) {
7006                 /*
7007                  * Failed to do the truncate, remove us from the in memory
7008                  * orphan list.
7009                  */
7010                 ret = btrfs_orphan_del(NULL, inode);
7011         }
7012
7013         if (trans) {
7014                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7015                 ret = btrfs_update_inode(trans, root, inode);
7016                 if (ret && !err)
7017                         err = ret;
7018
7019                 nr = trans->blocks_used;
7020                 ret = btrfs_end_transaction(trans, root);
7021                 btrfs_btree_balance_dirty(root, nr);
7022         }
7023
7024 out:
7025         btrfs_free_block_rsv(root, rsv);
7026
7027         if (ret && !err)
7028                 err = ret;
7029
7030         return err;
7031 }
7032
7033 /*
7034  * create a new subvolume directory/inode (helper for the ioctl).
7035  */
7036 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7037                              struct btrfs_root *new_root, u64 new_dirid)
7038 {
7039         struct inode *inode;
7040         int err;
7041         u64 index = 0;
7042
7043         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7044                                 new_dirid, new_dirid,
7045                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7046                                 &index);
7047         if (IS_ERR(inode))
7048                 return PTR_ERR(inode);
7049         inode->i_op = &btrfs_dir_inode_operations;
7050         inode->i_fop = &btrfs_dir_file_operations;
7051
7052         set_nlink(inode, 1);
7053         btrfs_i_size_write(inode, 0);
7054
7055         err = btrfs_update_inode(trans, new_root, inode);
7056
7057         iput(inode);
7058         return err;
7059 }
7060
7061 struct inode *btrfs_alloc_inode(struct super_block *sb)
7062 {
7063         struct btrfs_inode *ei;
7064         struct inode *inode;
7065
7066         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7067         if (!ei)
7068                 return NULL;
7069
7070         ei->root = NULL;
7071         ei->generation = 0;
7072         ei->last_trans = 0;
7073         ei->last_sub_trans = 0;
7074         ei->logged_trans = 0;
7075         ei->delalloc_bytes = 0;
7076         ei->disk_i_size = 0;
7077         ei->flags = 0;
7078         ei->csum_bytes = 0;
7079         ei->index_cnt = (u64)-1;
7080         ei->last_unlink_trans = 0;
7081         ei->last_log_commit = 0;
7082
7083         spin_lock_init(&ei->lock);
7084         ei->outstanding_extents = 0;
7085         ei->reserved_extents = 0;
7086
7087         ei->runtime_flags = 0;
7088         ei->force_compress = BTRFS_COMPRESS_NONE;
7089
7090         ei->delayed_node = NULL;
7091
7092         inode = &ei->vfs_inode;
7093         extent_map_tree_init(&ei->extent_tree);
7094         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7095         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7096         ei->io_tree.track_uptodate = 1;
7097         ei->io_failure_tree.track_uptodate = 1;
7098         mutex_init(&ei->log_mutex);
7099         mutex_init(&ei->delalloc_mutex);
7100         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7101         INIT_LIST_HEAD(&ei->delalloc_inodes);
7102         INIT_LIST_HEAD(&ei->ordered_operations);
7103         RB_CLEAR_NODE(&ei->rb_node);
7104
7105         return inode;
7106 }
7107
7108 static void btrfs_i_callback(struct rcu_head *head)
7109 {
7110         struct inode *inode = container_of(head, struct inode, i_rcu);
7111         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7112 }
7113
7114 void btrfs_destroy_inode(struct inode *inode)
7115 {
7116         struct btrfs_ordered_extent *ordered;
7117         struct btrfs_root *root = BTRFS_I(inode)->root;
7118
7119         WARN_ON(!hlist_empty(&inode->i_dentry));
7120         WARN_ON(inode->i_data.nrpages);
7121         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7122         WARN_ON(BTRFS_I(inode)->reserved_extents);
7123         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7124         WARN_ON(BTRFS_I(inode)->csum_bytes);
7125
7126         /*
7127          * This can happen where we create an inode, but somebody else also
7128          * created the same inode and we need to destroy the one we already
7129          * created.
7130          */
7131         if (!root)
7132                 goto free;
7133
7134         /*
7135          * Make sure we're properly removed from the ordered operation
7136          * lists.
7137          */
7138         smp_mb();
7139         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7140                 spin_lock(&root->fs_info->ordered_extent_lock);
7141                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7142                 spin_unlock(&root->fs_info->ordered_extent_lock);
7143         }
7144
7145         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7146                      &BTRFS_I(inode)->runtime_flags)) {
7147                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7148                        (unsigned long long)btrfs_ino(inode));
7149                 atomic_dec(&root->orphan_inodes);
7150         }
7151
7152         while (1) {
7153                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7154                 if (!ordered)
7155                         break;
7156                 else {
7157                         printk(KERN_ERR "btrfs found ordered "
7158                                "extent %llu %llu on inode cleanup\n",
7159                                (unsigned long long)ordered->file_offset,
7160                                (unsigned long long)ordered->len);
7161                         btrfs_remove_ordered_extent(inode, ordered);
7162                         btrfs_put_ordered_extent(ordered);
7163                         btrfs_put_ordered_extent(ordered);
7164                 }
7165         }
7166         inode_tree_del(inode);
7167         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7168 free:
7169         btrfs_remove_delayed_node(inode);
7170         call_rcu(&inode->i_rcu, btrfs_i_callback);
7171 }
7172
7173 int btrfs_drop_inode(struct inode *inode)
7174 {
7175         struct btrfs_root *root = BTRFS_I(inode)->root;
7176
7177         if (btrfs_root_refs(&root->root_item) == 0 &&
7178             !btrfs_is_free_space_inode(inode))
7179                 return 1;
7180         else
7181                 return generic_drop_inode(inode);
7182 }
7183
7184 static void init_once(void *foo)
7185 {
7186         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7187
7188         inode_init_once(&ei->vfs_inode);
7189 }
7190
7191 void btrfs_destroy_cachep(void)
7192 {
7193         if (btrfs_inode_cachep)
7194                 kmem_cache_destroy(btrfs_inode_cachep);
7195         if (btrfs_trans_handle_cachep)
7196                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7197         if (btrfs_transaction_cachep)
7198                 kmem_cache_destroy(btrfs_transaction_cachep);
7199         if (btrfs_path_cachep)
7200                 kmem_cache_destroy(btrfs_path_cachep);
7201         if (btrfs_free_space_cachep)
7202                 kmem_cache_destroy(btrfs_free_space_cachep);
7203 }
7204
7205 int btrfs_init_cachep(void)
7206 {
7207         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7208                         sizeof(struct btrfs_inode), 0,
7209                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7210         if (!btrfs_inode_cachep)
7211                 goto fail;
7212
7213         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7214                         sizeof(struct btrfs_trans_handle), 0,
7215                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7216         if (!btrfs_trans_handle_cachep)
7217                 goto fail;
7218
7219         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7220                         sizeof(struct btrfs_transaction), 0,
7221                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7222         if (!btrfs_transaction_cachep)
7223                 goto fail;
7224
7225         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7226                         sizeof(struct btrfs_path), 0,
7227                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7228         if (!btrfs_path_cachep)
7229                 goto fail;
7230
7231         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7232                         sizeof(struct btrfs_free_space), 0,
7233                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7234         if (!btrfs_free_space_cachep)
7235                 goto fail;
7236
7237         return 0;
7238 fail:
7239         btrfs_destroy_cachep();
7240         return -ENOMEM;
7241 }
7242
7243 static int btrfs_getattr(struct vfsmount *mnt,
7244                          struct dentry *dentry, struct kstat *stat)
7245 {
7246         struct inode *inode = dentry->d_inode;
7247         u32 blocksize = inode->i_sb->s_blocksize;
7248
7249         generic_fillattr(inode, stat);
7250         stat->dev = BTRFS_I(inode)->root->anon_dev;
7251         stat->blksize = PAGE_CACHE_SIZE;
7252         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7253                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7254         return 0;
7255 }
7256
7257 /*
7258  * If a file is moved, it will inherit the cow and compression flags of the new
7259  * directory.
7260  */
7261 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7262 {
7263         struct btrfs_inode *b_dir = BTRFS_I(dir);
7264         struct btrfs_inode *b_inode = BTRFS_I(inode);
7265
7266         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7267                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7268         else
7269                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7270
7271         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7272                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7273                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7274         } else {
7275                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7276                                     BTRFS_INODE_NOCOMPRESS);
7277         }
7278 }
7279
7280 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7281                            struct inode *new_dir, struct dentry *new_dentry)
7282 {
7283         struct btrfs_trans_handle *trans;
7284         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7285         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7286         struct inode *new_inode = new_dentry->d_inode;
7287         struct inode *old_inode = old_dentry->d_inode;
7288         struct timespec ctime = CURRENT_TIME;
7289         u64 index = 0;
7290         u64 root_objectid;
7291         int ret;
7292         u64 old_ino = btrfs_ino(old_inode);
7293
7294         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7295                 return -EPERM;
7296
7297         /* we only allow rename subvolume link between subvolumes */
7298         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7299                 return -EXDEV;
7300
7301         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7302             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7303                 return -ENOTEMPTY;
7304
7305         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7306             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7307                 return -ENOTEMPTY;
7308         /*
7309          * we're using rename to replace one file with another.
7310          * and the replacement file is large.  Start IO on it now so
7311          * we don't add too much work to the end of the transaction
7312          */
7313         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7314             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7315                 filemap_flush(old_inode->i_mapping);
7316
7317         /* close the racy window with snapshot create/destroy ioctl */
7318         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7319                 down_read(&root->fs_info->subvol_sem);
7320         /*
7321          * We want to reserve the absolute worst case amount of items.  So if
7322          * both inodes are subvols and we need to unlink them then that would
7323          * require 4 item modifications, but if they are both normal inodes it
7324          * would require 5 item modifications, so we'll assume their normal
7325          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7326          * should cover the worst case number of items we'll modify.
7327          */
7328         trans = btrfs_start_transaction(root, 20);
7329         if (IS_ERR(trans)) {
7330                 ret = PTR_ERR(trans);
7331                 goto out_notrans;
7332         }
7333
7334         if (dest != root)
7335                 btrfs_record_root_in_trans(trans, dest);
7336
7337         ret = btrfs_set_inode_index(new_dir, &index);
7338         if (ret)
7339                 goto out_fail;
7340
7341         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7342                 /* force full log commit if subvolume involved. */
7343                 root->fs_info->last_trans_log_full_commit = trans->transid;
7344         } else {
7345                 ret = btrfs_insert_inode_ref(trans, dest,
7346                                              new_dentry->d_name.name,
7347                                              new_dentry->d_name.len,
7348                                              old_ino,
7349                                              btrfs_ino(new_dir), index);
7350                 if (ret)
7351                         goto out_fail;
7352                 /*
7353                  * this is an ugly little race, but the rename is required
7354                  * to make sure that if we crash, the inode is either at the
7355                  * old name or the new one.  pinning the log transaction lets
7356                  * us make sure we don't allow a log commit to come in after
7357                  * we unlink the name but before we add the new name back in.
7358                  */
7359                 btrfs_pin_log_trans(root);
7360         }
7361         /*
7362          * make sure the inode gets flushed if it is replacing
7363          * something.
7364          */
7365         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7366                 btrfs_add_ordered_operation(trans, root, old_inode);
7367
7368         inode_inc_iversion(old_dir);
7369         inode_inc_iversion(new_dir);
7370         inode_inc_iversion(old_inode);
7371         old_dir->i_ctime = old_dir->i_mtime = ctime;
7372         new_dir->i_ctime = new_dir->i_mtime = ctime;
7373         old_inode->i_ctime = ctime;
7374
7375         if (old_dentry->d_parent != new_dentry->d_parent)
7376                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7377
7378         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7379                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7380                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7381                                         old_dentry->d_name.name,
7382                                         old_dentry->d_name.len);
7383         } else {
7384                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7385                                         old_dentry->d_inode,
7386                                         old_dentry->d_name.name,
7387                                         old_dentry->d_name.len);
7388                 if (!ret)
7389                         ret = btrfs_update_inode(trans, root, old_inode);
7390         }
7391         if (ret) {
7392                 btrfs_abort_transaction(trans, root, ret);
7393                 goto out_fail;
7394         }
7395
7396         if (new_inode) {
7397                 inode_inc_iversion(new_inode);
7398                 new_inode->i_ctime = CURRENT_TIME;
7399                 if (unlikely(btrfs_ino(new_inode) ==
7400                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7401                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7402                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7403                                                 root_objectid,
7404                                                 new_dentry->d_name.name,
7405                                                 new_dentry->d_name.len);
7406                         BUG_ON(new_inode->i_nlink == 0);
7407                 } else {
7408                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7409                                                  new_dentry->d_inode,
7410                                                  new_dentry->d_name.name,
7411                                                  new_dentry->d_name.len);
7412                 }
7413                 if (!ret && new_inode->i_nlink == 0) {
7414                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7415                         BUG_ON(ret);
7416                 }
7417                 if (ret) {
7418                         btrfs_abort_transaction(trans, root, ret);
7419                         goto out_fail;
7420                 }
7421         }
7422
7423         fixup_inode_flags(new_dir, old_inode);
7424
7425         ret = btrfs_add_link(trans, new_dir, old_inode,
7426                              new_dentry->d_name.name,
7427                              new_dentry->d_name.len, 0, index);
7428         if (ret) {
7429                 btrfs_abort_transaction(trans, root, ret);
7430                 goto out_fail;
7431         }
7432
7433         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7434                 struct dentry *parent = new_dentry->d_parent;
7435                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7436                 btrfs_end_log_trans(root);
7437         }
7438 out_fail:
7439         btrfs_end_transaction(trans, root);
7440 out_notrans:
7441         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7442                 up_read(&root->fs_info->subvol_sem);
7443
7444         return ret;
7445 }
7446
7447 /*
7448  * some fairly slow code that needs optimization. This walks the list
7449  * of all the inodes with pending delalloc and forces them to disk.
7450  */
7451 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7452 {
7453         struct list_head *head = &root->fs_info->delalloc_inodes;
7454         struct btrfs_inode *binode;
7455         struct inode *inode;
7456
7457         if (root->fs_info->sb->s_flags & MS_RDONLY)
7458                 return -EROFS;
7459
7460         spin_lock(&root->fs_info->delalloc_lock);
7461         while (!list_empty(head)) {
7462                 binode = list_entry(head->next, struct btrfs_inode,
7463                                     delalloc_inodes);
7464                 inode = igrab(&binode->vfs_inode);
7465                 if (!inode)
7466                         list_del_init(&binode->delalloc_inodes);
7467                 spin_unlock(&root->fs_info->delalloc_lock);
7468                 if (inode) {
7469                         filemap_flush(inode->i_mapping);
7470                         if (delay_iput)
7471                                 btrfs_add_delayed_iput(inode);
7472                         else
7473                                 iput(inode);
7474                 }
7475                 cond_resched();
7476                 spin_lock(&root->fs_info->delalloc_lock);
7477         }
7478         spin_unlock(&root->fs_info->delalloc_lock);
7479
7480         /* the filemap_flush will queue IO into the worker threads, but
7481          * we have to make sure the IO is actually started and that
7482          * ordered extents get created before we return
7483          */
7484         atomic_inc(&root->fs_info->async_submit_draining);
7485         while (atomic_read(&root->fs_info->nr_async_submits) ||
7486               atomic_read(&root->fs_info->async_delalloc_pages)) {
7487                 wait_event(root->fs_info->async_submit_wait,
7488                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7489                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7490         }
7491         atomic_dec(&root->fs_info->async_submit_draining);
7492         return 0;
7493 }
7494
7495 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7496                          const char *symname)
7497 {
7498         struct btrfs_trans_handle *trans;
7499         struct btrfs_root *root = BTRFS_I(dir)->root;
7500         struct btrfs_path *path;
7501         struct btrfs_key key;
7502         struct inode *inode = NULL;
7503         int err;
7504         int drop_inode = 0;
7505         u64 objectid;
7506         u64 index = 0 ;
7507         int name_len;
7508         int datasize;
7509         unsigned long ptr;
7510         struct btrfs_file_extent_item *ei;
7511         struct extent_buffer *leaf;
7512         unsigned long nr = 0;
7513
7514         name_len = strlen(symname) + 1;
7515         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7516                 return -ENAMETOOLONG;
7517
7518         /*
7519          * 2 items for inode item and ref
7520          * 2 items for dir items
7521          * 1 item for xattr if selinux is on
7522          */
7523         trans = btrfs_start_transaction(root, 5);
7524         if (IS_ERR(trans))
7525                 return PTR_ERR(trans);
7526
7527         err = btrfs_find_free_ino(root, &objectid);
7528         if (err)
7529                 goto out_unlock;
7530
7531         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7532                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7533                                 S_IFLNK|S_IRWXUGO, &index);
7534         if (IS_ERR(inode)) {
7535                 err = PTR_ERR(inode);
7536                 goto out_unlock;
7537         }
7538
7539         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7540         if (err) {
7541                 drop_inode = 1;
7542                 goto out_unlock;
7543         }
7544
7545         /*
7546         * If the active LSM wants to access the inode during
7547         * d_instantiate it needs these. Smack checks to see
7548         * if the filesystem supports xattrs by looking at the
7549         * ops vector.
7550         */
7551         inode->i_fop = &btrfs_file_operations;
7552         inode->i_op = &btrfs_file_inode_operations;
7553
7554         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7555         if (err)
7556                 drop_inode = 1;
7557         else {
7558                 inode->i_mapping->a_ops = &btrfs_aops;
7559                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7560                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7561         }
7562         if (drop_inode)
7563                 goto out_unlock;
7564
7565         path = btrfs_alloc_path();
7566         if (!path) {
7567                 err = -ENOMEM;
7568                 drop_inode = 1;
7569                 goto out_unlock;
7570         }
7571         key.objectid = btrfs_ino(inode);
7572         key.offset = 0;
7573         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7574         datasize = btrfs_file_extent_calc_inline_size(name_len);
7575         err = btrfs_insert_empty_item(trans, root, path, &key,
7576                                       datasize);
7577         if (err) {
7578                 drop_inode = 1;
7579                 btrfs_free_path(path);
7580                 goto out_unlock;
7581         }
7582         leaf = path->nodes[0];
7583         ei = btrfs_item_ptr(leaf, path->slots[0],
7584                             struct btrfs_file_extent_item);
7585         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7586         btrfs_set_file_extent_type(leaf, ei,
7587                                    BTRFS_FILE_EXTENT_INLINE);
7588         btrfs_set_file_extent_encryption(leaf, ei, 0);
7589         btrfs_set_file_extent_compression(leaf, ei, 0);
7590         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7591         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7592
7593         ptr = btrfs_file_extent_inline_start(ei);
7594         write_extent_buffer(leaf, symname, ptr, name_len);
7595         btrfs_mark_buffer_dirty(leaf);
7596         btrfs_free_path(path);
7597
7598         inode->i_op = &btrfs_symlink_inode_operations;
7599         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7600         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7601         inode_set_bytes(inode, name_len);
7602         btrfs_i_size_write(inode, name_len - 1);
7603         err = btrfs_update_inode(trans, root, inode);
7604         if (err)
7605                 drop_inode = 1;
7606
7607 out_unlock:
7608         if (!err)
7609                 d_instantiate(dentry, inode);
7610         nr = trans->blocks_used;
7611         btrfs_end_transaction(trans, root);
7612         if (drop_inode) {
7613                 inode_dec_link_count(inode);
7614                 iput(inode);
7615         }
7616         btrfs_btree_balance_dirty(root, nr);
7617         return err;
7618 }
7619
7620 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7621                                        u64 start, u64 num_bytes, u64 min_size,
7622                                        loff_t actual_len, u64 *alloc_hint,
7623                                        struct btrfs_trans_handle *trans)
7624 {
7625         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7626         struct extent_map *em;
7627         struct btrfs_root *root = BTRFS_I(inode)->root;
7628         struct btrfs_key ins;
7629         u64 cur_offset = start;
7630         u64 i_size;
7631         int ret = 0;
7632         bool own_trans = true;
7633
7634         if (trans)
7635                 own_trans = false;
7636         while (num_bytes > 0) {
7637                 if (own_trans) {
7638                         trans = btrfs_start_transaction(root, 3);
7639                         if (IS_ERR(trans)) {
7640                                 ret = PTR_ERR(trans);
7641                                 break;
7642                         }
7643                 }
7644
7645                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7646                                            0, *alloc_hint, &ins, 1);
7647                 if (ret) {
7648                         if (own_trans)
7649                                 btrfs_end_transaction(trans, root);
7650                         break;
7651                 }
7652
7653                 ret = insert_reserved_file_extent(trans, inode,
7654                                                   cur_offset, ins.objectid,
7655                                                   ins.offset, ins.offset,
7656                                                   ins.offset, 0, 0, 0,
7657                                                   BTRFS_FILE_EXTENT_PREALLOC);
7658                 if (ret) {
7659                         btrfs_abort_transaction(trans, root, ret);
7660                         if (own_trans)
7661                                 btrfs_end_transaction(trans, root);
7662                         break;
7663                 }
7664                 btrfs_drop_extent_cache(inode, cur_offset,
7665                                         cur_offset + ins.offset -1, 0);
7666
7667                 em = alloc_extent_map();
7668                 if (!em) {
7669                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7670                                 &BTRFS_I(inode)->runtime_flags);
7671                         goto next;
7672                 }
7673
7674                 em->start = cur_offset;
7675                 em->orig_start = cur_offset;
7676                 em->len = ins.offset;
7677                 em->block_start = ins.objectid;
7678                 em->block_len = ins.offset;
7679                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7680                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7681                 em->generation = trans->transid;
7682
7683                 while (1) {
7684                         write_lock(&em_tree->lock);
7685                         ret = add_extent_mapping(em_tree, em);
7686                         if (!ret)
7687                                 list_move(&em->list,
7688                                           &em_tree->modified_extents);
7689                         write_unlock(&em_tree->lock);
7690                         if (ret != -EEXIST)
7691                                 break;
7692                         btrfs_drop_extent_cache(inode, cur_offset,
7693                                                 cur_offset + ins.offset - 1,
7694                                                 0);
7695                 }
7696                 free_extent_map(em);
7697 next:
7698                 num_bytes -= ins.offset;
7699                 cur_offset += ins.offset;
7700                 *alloc_hint = ins.objectid + ins.offset;
7701
7702                 inode_inc_iversion(inode);
7703                 inode->i_ctime = CURRENT_TIME;
7704                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7705                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7706                     (actual_len > inode->i_size) &&
7707                     (cur_offset > inode->i_size)) {
7708                         if (cur_offset > actual_len)
7709                                 i_size = actual_len;
7710                         else
7711                                 i_size = cur_offset;
7712                         i_size_write(inode, i_size);
7713                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7714                 }
7715
7716                 ret = btrfs_update_inode(trans, root, inode);
7717
7718                 if (ret) {
7719                         btrfs_abort_transaction(trans, root, ret);
7720                         if (own_trans)
7721                                 btrfs_end_transaction(trans, root);
7722                         break;
7723                 }
7724
7725                 if (own_trans)
7726                         btrfs_end_transaction(trans, root);
7727         }
7728         return ret;
7729 }
7730
7731 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7732                               u64 start, u64 num_bytes, u64 min_size,
7733                               loff_t actual_len, u64 *alloc_hint)
7734 {
7735         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7736                                            min_size, actual_len, alloc_hint,
7737                                            NULL);
7738 }
7739
7740 int btrfs_prealloc_file_range_trans(struct inode *inode,
7741                                     struct btrfs_trans_handle *trans, int mode,
7742                                     u64 start, u64 num_bytes, u64 min_size,
7743                                     loff_t actual_len, u64 *alloc_hint)
7744 {
7745         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7746                                            min_size, actual_len, alloc_hint, trans);
7747 }
7748
7749 static int btrfs_set_page_dirty(struct page *page)
7750 {
7751         return __set_page_dirty_nobuffers(page);
7752 }
7753
7754 static int btrfs_permission(struct inode *inode, int mask)
7755 {
7756         struct btrfs_root *root = BTRFS_I(inode)->root;
7757         umode_t mode = inode->i_mode;
7758
7759         if (mask & MAY_WRITE &&
7760             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7761                 if (btrfs_root_readonly(root))
7762                         return -EROFS;
7763                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7764                         return -EACCES;
7765         }
7766         return generic_permission(inode, mask);
7767 }
7768
7769 static const struct inode_operations btrfs_dir_inode_operations = {
7770         .getattr        = btrfs_getattr,
7771         .lookup         = btrfs_lookup,
7772         .create         = btrfs_create,
7773         .unlink         = btrfs_unlink,
7774         .link           = btrfs_link,
7775         .mkdir          = btrfs_mkdir,
7776         .rmdir          = btrfs_rmdir,
7777         .rename         = btrfs_rename,
7778         .symlink        = btrfs_symlink,
7779         .setattr        = btrfs_setattr,
7780         .mknod          = btrfs_mknod,
7781         .setxattr       = btrfs_setxattr,
7782         .getxattr       = btrfs_getxattr,
7783         .listxattr      = btrfs_listxattr,
7784         .removexattr    = btrfs_removexattr,
7785         .permission     = btrfs_permission,
7786         .get_acl        = btrfs_get_acl,
7787 };
7788 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7789         .lookup         = btrfs_lookup,
7790         .permission     = btrfs_permission,
7791         .get_acl        = btrfs_get_acl,
7792 };
7793
7794 static const struct file_operations btrfs_dir_file_operations = {
7795         .llseek         = generic_file_llseek,
7796         .read           = generic_read_dir,
7797         .readdir        = btrfs_real_readdir,
7798         .unlocked_ioctl = btrfs_ioctl,
7799 #ifdef CONFIG_COMPAT
7800         .compat_ioctl   = btrfs_ioctl,
7801 #endif
7802         .release        = btrfs_release_file,
7803         .fsync          = btrfs_sync_file,
7804 };
7805
7806 static struct extent_io_ops btrfs_extent_io_ops = {
7807         .fill_delalloc = run_delalloc_range,
7808         .submit_bio_hook = btrfs_submit_bio_hook,
7809         .merge_bio_hook = btrfs_merge_bio_hook,
7810         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7811         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7812         .writepage_start_hook = btrfs_writepage_start_hook,
7813         .set_bit_hook = btrfs_set_bit_hook,
7814         .clear_bit_hook = btrfs_clear_bit_hook,
7815         .merge_extent_hook = btrfs_merge_extent_hook,
7816         .split_extent_hook = btrfs_split_extent_hook,
7817 };
7818
7819 /*
7820  * btrfs doesn't support the bmap operation because swapfiles
7821  * use bmap to make a mapping of extents in the file.  They assume
7822  * these extents won't change over the life of the file and they
7823  * use the bmap result to do IO directly to the drive.
7824  *
7825  * the btrfs bmap call would return logical addresses that aren't
7826  * suitable for IO and they also will change frequently as COW
7827  * operations happen.  So, swapfile + btrfs == corruption.
7828  *
7829  * For now we're avoiding this by dropping bmap.
7830  */
7831 static const struct address_space_operations btrfs_aops = {
7832         .readpage       = btrfs_readpage,
7833         .writepage      = btrfs_writepage,
7834         .writepages     = btrfs_writepages,
7835         .readpages      = btrfs_readpages,
7836         .direct_IO      = btrfs_direct_IO,
7837         .invalidatepage = btrfs_invalidatepage,
7838         .releasepage    = btrfs_releasepage,
7839         .set_page_dirty = btrfs_set_page_dirty,
7840         .error_remove_page = generic_error_remove_page,
7841 };
7842
7843 static const struct address_space_operations btrfs_symlink_aops = {
7844         .readpage       = btrfs_readpage,
7845         .writepage      = btrfs_writepage,
7846         .invalidatepage = btrfs_invalidatepage,
7847         .releasepage    = btrfs_releasepage,
7848 };
7849
7850 static const struct inode_operations btrfs_file_inode_operations = {
7851         .getattr        = btrfs_getattr,
7852         .setattr        = btrfs_setattr,
7853         .setxattr       = btrfs_setxattr,
7854         .getxattr       = btrfs_getxattr,
7855         .listxattr      = btrfs_listxattr,
7856         .removexattr    = btrfs_removexattr,
7857         .permission     = btrfs_permission,
7858         .fiemap         = btrfs_fiemap,
7859         .get_acl        = btrfs_get_acl,
7860         .update_time    = btrfs_update_time,
7861 };
7862 static const struct inode_operations btrfs_special_inode_operations = {
7863         .getattr        = btrfs_getattr,
7864         .setattr        = btrfs_setattr,
7865         .permission     = btrfs_permission,
7866         .setxattr       = btrfs_setxattr,
7867         .getxattr       = btrfs_getxattr,
7868         .listxattr      = btrfs_listxattr,
7869         .removexattr    = btrfs_removexattr,
7870         .get_acl        = btrfs_get_acl,
7871         .update_time    = btrfs_update_time,
7872 };
7873 static const struct inode_operations btrfs_symlink_inode_operations = {
7874         .readlink       = generic_readlink,
7875         .follow_link    = page_follow_link_light,
7876         .put_link       = page_put_link,
7877         .getattr        = btrfs_getattr,
7878         .setattr        = btrfs_setattr,
7879         .permission     = btrfs_permission,
7880         .setxattr       = btrfs_setxattr,
7881         .getxattr       = btrfs_getxattr,
7882         .listxattr      = btrfs_listxattr,
7883         .removexattr    = btrfs_removexattr,
7884         .get_acl        = btrfs_get_acl,
7885         .update_time    = btrfs_update_time,
7886 };
7887
7888 const struct dentry_operations btrfs_dentry_operations = {
7889         .d_delete       = btrfs_dentry_delete,
7890         .d_release      = btrfs_dentry_release,
7891 };