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 inode *inode, u64 start, u64 end);
  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         BUG_ON(ret);
 154         if (ret) {
 155                 err = ret;
 156                 goto fail;
 157         }
 158         leaf = path->nodes[0];
 159         ei = btrfs_item_ptr(leaf, path->slots[0],
 160                             struct btrfs_file_extent_item);
 161         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 162         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 163         btrfs_set_file_extent_encryption(leaf, ei, 0);
 164         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 165         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 166         ptr = btrfs_file_extent_inline_start(ei);
 167
 168         if (compress_type != BTRFS_COMPRESS_NONE) {
 169                 struct page *cpage;
 170                 int i = 0;
 171                 while (compressed_size > 0) {
 172                         cpage = compressed_pages[i];
 173                         cur_size = min_t(unsigned long, compressed_size,
 174                                        PAGE_CACHE_SIZE);
 175
 176                         kaddr = kmap_atomic(cpage, KM_USER0);
 177                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 178                         kunmap_atomic(kaddr, KM_USER0);
 179
 180                         i++;
 181                         ptr += cur_size;
 182                         compressed_size -= cur_size;
 183                 }
 184                 btrfs_set_file_extent_compression(leaf, ei,
 185                                                   compress_type);
 186         } else {
 187                 page = find_get_page(inode->i_mapping,
 188                                      start >> PAGE_CACHE_SHIFT);
 189                 btrfs_set_file_extent_compression(leaf, ei, 0);
 190                 kaddr = kmap_atomic(page, KM_USER0);
 191                 offset = start & (PAGE_CACHE_SIZE - 1);
 192                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 193                 kunmap_atomic(kaddr, KM_USER0);
 194                 page_cache_release(page);
 195         }
 196         btrfs_mark_buffer_dirty(leaf);
 197         btrfs_free_path(path);
 198
 199         /*
 200          * we're an inline extent, so nobody can
 201          * extend the file past i_size without locking
 202          * a page we already have locked.
 203          *
 204          * We must do any isize and inode updates
 205          * before we unlock the pages.  Otherwise we
 206          * could end up racing with unlink.
 207          */
 208         BTRFS_I(inode)->disk_i_size = inode->i_size;
 209         btrfs_update_inode(trans, root, inode);
 210
 211         return 0;
 212 fail:
 213         btrfs_free_path(path);
 214         return err;
 215 }
 216
 217
 218 /*
 219  * conditionally insert an inline extent into the file.  This
 220  * does the checks required to make sure the data is small enough
 221  * to fit as an inline extent.
 222  */
 223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 224                                  struct btrfs_root *root,
 225                                  struct inode *inode, u64 start, u64 end,
 226                                  size_t compressed_size, int compress_type,
 227                                  struct page **compressed_pages)
 228 {
 229         u64 isize = i_size_read(inode);
 230         u64 actual_end = min(end + 1, isize);
 231         u64 inline_len = actual_end - start;
 232         u64 aligned_end = (end + root->sectorsize - 1) &
 233                         ~((u64)root->sectorsize - 1);
 234         u64 hint_byte;
 235         u64 data_len = inline_len;
 236         int ret;
 237
 238         if (compressed_size)
 239                 data_len = compressed_size;
 240
 241         if (start > 0 ||
 242             actual_end >= PAGE_CACHE_SIZE ||
 243             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 244             (!compressed_size &&
 245             (actual_end & (root->sectorsize - 1)) == 0) ||
 246             end + 1 < isize ||
 247             data_len > root->fs_info->max_inline) {
 248                 return 1;
 249         }
 250
 251         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 252                                  &hint_byte, 1);
 253         BUG_ON(ret);
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         BUG_ON(ret);
 261         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 262         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 263         return 0;
 264 }
 265
 266 struct async_extent {
 267         u64 start;
 268         u64 ram_size;
 269         u64 compressed_size;
 270         struct page **pages;
 271         unsigned long nr_pages;
 272         int compress_type;
 273         struct list_head list;
 274 };
 275
 276 struct async_cow {
 277         struct inode *inode;
 278         struct btrfs_root *root;
 279         struct page *locked_page;
 280         u64 start;
 281         u64 end;
 282         struct list_head extents;
 283         struct btrfs_work work;
 284 };
 285
 286 static noinline int add_async_extent(struct async_cow *cow,
 287                                      u64 start, u64 ram_size,
 288                                      u64 compressed_size,
 289                                      struct page **pages,
 290                                      unsigned long nr_pages,
 291                                      int compress_type)
 292 {
 293         struct async_extent *async_extent;
 294
 295         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 296         BUG_ON(!async_extent);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         async_extent->compress_type = compress_type;
 303         list_add_tail(&async_extent->list, &cow->extents);
 304         return 0;
 305 }
 306
 307 /*
 308  * we create compressed extents in two phases.  The first
 309  * phase compresses a range of pages that have already been
 310  * locked (both pages and state bits are locked).
 311  *
 312  * This is done inside an ordered work queue, and the compression
 313  * is spread across many cpus.  The actual IO submission is step
 314  * two, and the ordered work queue takes care of making sure that
 315  * happens in the same order things were put onto the queue by
 316  * writepages and friends.
 317  *
 318  * If this code finds it can't get good compression, it puts an
 319  * entry onto the work queue to write the uncompressed bytes.  This
 320  * makes sure that both compressed inodes and uncompressed inodes
 321  * are written in the same order that pdflush sent them down.
 322  */
 323 static noinline int compress_file_range(struct inode *inode,
 324                                         struct page *locked_page,
 325                                         u64 start, u64 end,
 326                                         struct async_cow *async_cow,
 327                                         int *num_added)
 328 {
 329         struct btrfs_root *root = BTRFS_I(inode)->root;
 330         struct btrfs_trans_handle *trans;
 331         u64 num_bytes;
 332         u64 blocksize = root->sectorsize;
 333         u64 actual_end;
 334         u64 isize = i_size_read(inode);
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345         int compress_type = root->fs_info->compress_type;
 346
 347         /* if this is a small write inside eof, kick off a defragbot */
 348         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 349                 btrfs_add_inode_defrag(NULL, inode);
 350
 351         actual_end = min_t(u64, isize, end + 1);
 352 again:
 353         will_compress = 0;
 354         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 355         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 356
 357         /*
 358          * we don't want to send crud past the end of i_size through
 359          * compression, that's just a waste of CPU time.  So, if the
 360          * end of the file is before the start of our current
 361          * requested range of bytes, we bail out to the uncompressed
 362          * cleanup code that can deal with all of this.
 363          *
 364          * It isn't really the fastest way to fix things, but this is a
 365          * very uncommon corner.
 366          */
 367         if (actual_end <= start)
 368                 goto cleanup_and_bail_uncompressed;
 369
 370         total_compressed = actual_end - start;
 371
 372         /* we want to make sure that amount of ram required to uncompress
 373          * an extent is reasonable, so we limit the total size in ram
 374          * of a compressed extent to 128k.  This is a crucial number
 375          * because it also controls how easily we can spread reads across
 376          * cpus for decompression.
 377          *
 378          * We also want to make sure the amount of IO required to do
 379          * a random read is reasonably small, so we limit the size of
 380          * a compressed extent to 128k.
 381          */
 382         total_compressed = min(total_compressed, max_uncompressed);
 383         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 384         num_bytes = max(blocksize,  num_bytes);
 385         total_in = 0;
 386         ret = 0;
 387
 388         /*
 389          * we do compression for mount -o compress and when the
 390          * inode has not been flagged as nocompress.  This flag can
 391          * change at any time if we discover bad compression ratios.
 392          */
 393         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 394             (btrfs_test_opt(root, COMPRESS) ||
 395              (BTRFS_I(inode)->force_compress) ||
 396              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 397                 WARN_ON(pages);
 398                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 399                 if (!pages) {
 400                         /* just bail out to the uncompressed code */
 401                         goto cont;
 402                 }
 403
 404                 if (BTRFS_I(inode)->force_compress)
 405                         compress_type = BTRFS_I(inode)->force_compress;
 406
 407                 ret = btrfs_compress_pages(compress_type,
 408                                            inode->i_mapping, start,
 409                                            total_compressed, pages,
 410                                            nr_pages, &nr_pages_ret,
 411                                            &total_in,
 412                                            &total_compressed,
 413                                            max_compressed);
 414
 415                 if (!ret) {
 416                         unsigned long offset = total_compressed &
 417                                 (PAGE_CACHE_SIZE - 1);
 418                         struct page *page = pages[nr_pages_ret - 1];
 419                         char *kaddr;
 420
 421                         /* zero the tail end of the last page, we might be
 422                          * sending it down to disk
 423                          */
 424                         if (offset) {
 425                                 kaddr = kmap_atomic(page, KM_USER0);
 426                                 memset(kaddr + offset, 0,
 427                                        PAGE_CACHE_SIZE - offset);
 428                                 kunmap_atomic(kaddr, KM_USER0);
 429                         }
 430                         will_compress = 1;
 431                 }
 432         }
 433 cont:
 434         if (start == 0) {
 435                 trans = btrfs_join_transaction(root);
 436                 BUG_ON(IS_ERR(trans));
 437                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 438
 439                 /* lets try to make an inline extent */
 440                 if (ret || total_in < (actual_end - start)) {
 441                         /* we didn't compress the entire range, try
 442                          * to make an uncompressed inline extent.
 443                          */
 444                         ret = cow_file_range_inline(trans, root, inode,
 445                                                     start, end, 0, 0, NULL);
 446                 } else {
 447                         /* try making a compressed inline extent */
 448                         ret = cow_file_range_inline(trans, root, inode,
 449                                                     start, end,
 450                                                     total_compressed,
 451                                                     compress_type, pages);
 452                 }
 453                 if (ret == 0) {
 454                         /*
 455                          * inline extent creation worked, we don't need
 456                          * to create any more async work items.  Unlock
 457                          * and free up our temp pages.
 458                          */
 459                         extent_clear_unlock_delalloc(inode,
 460                              &BTRFS_I(inode)->io_tree,
 461                              start, end, NULL,
 462                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 463                              EXTENT_CLEAR_DELALLOC |
 464                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 465
 466                         btrfs_end_transaction(trans, root);
 467                         goto free_pages_out;
 468                 }
 469                 btrfs_end_transaction(trans, root);
 470         }
 471
 472         if (will_compress) {
 473                 /*
 474                  * we aren't doing an inline extent round the compressed size
 475                  * up to a block size boundary so the allocator does sane
 476                  * things
 477                  */
 478                 total_compressed = (total_compressed + blocksize - 1) &
 479                         ~(blocksize - 1);
 480
 481                 /*
 482                  * one last check to make sure the compression is really a
 483                  * win, compare the page count read with the blocks on disk
 484                  */
 485                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 486                         ~(PAGE_CACHE_SIZE - 1);
 487                 if (total_compressed >= total_in) {
 488                         will_compress = 0;
 489                 } else {
 490                         num_bytes = total_in;
 491                 }
 492         }
 493         if (!will_compress && pages) {
 494                 /*
 495                  * the compression code ran but failed to make things smaller,
 496                  * free any pages it allocated and our page pointer array
 497                  */
 498                 for (i = 0; i < nr_pages_ret; i++) {
 499                         WARN_ON(pages[i]->mapping);
 500                         page_cache_release(pages[i]);
 501                 }
 502                 kfree(pages);
 503                 pages = NULL;
 504                 total_compressed = 0;
 505                 nr_pages_ret = 0;
 506
 507                 /* flag the file so we don't compress in the future */
 508                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 509                     !(BTRFS_I(inode)->force_compress)) {
 510                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 511                 }
 512         }
 513         if (will_compress) {
 514                 *num_added += 1;
 515
 516                 /* the async work queues will take care of doing actual
 517                  * allocation on disk for these compressed pages,
 518                  * and will submit them to the elevator.
 519                  */
 520                 add_async_extent(async_cow, start, num_bytes,
 521                                  total_compressed, pages, nr_pages_ret,
 522                                  compress_type);
 523
 524                 if (start + num_bytes < end) {
 525                         start += num_bytes;
 526                         pages = NULL;
 527                         cond_resched();
 528                         goto again;
 529                 }
 530         } else {
 531 cleanup_and_bail_uncompressed:
 532                 /*
 533                  * No compression, but we still need to write the pages in
 534                  * the file we've been given so far.  redirty the locked
 535                  * page if it corresponds to our extent and set things up
 536                  * for the async work queue to run cow_file_range to do
 537                  * the normal delalloc dance
 538                  */
 539                 if (page_offset(locked_page) >= start &&
 540                     page_offset(locked_page) <= end) {
 541                         __set_page_dirty_nobuffers(locked_page);
 542                         /* unlocked later on in the async handlers */
 543                 }
 544                 add_async_extent(async_cow, start, end - start + 1,
 545                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 546                 *num_added += 1;
 547         }
 548
 549 out:
 550         return 0;
 551
 552 free_pages_out:
 553         for (i = 0; i < nr_pages_ret; i++) {
 554                 WARN_ON(pages[i]->mapping);
 555                 page_cache_release(pages[i]);
 556         }
 557         kfree(pages);
 558
 559         goto out;
 560 }
 561
 562 /*
 563  * phase two of compressed writeback.  This is the ordered portion
 564  * of the code, which only gets called in the order the work was
 565  * queued.  We walk all the async extents created by compress_file_range
 566  * and send them down to the disk.
 567  */
 568 static noinline int submit_compressed_extents(struct inode *inode,
 569                                               struct async_cow *async_cow)
 570 {
 571         struct async_extent *async_extent;
 572         u64 alloc_hint = 0;
 573         struct btrfs_trans_handle *trans;
 574         struct btrfs_key ins;
 575         struct extent_map *em;
 576         struct btrfs_root *root = BTRFS_I(inode)->root;
 577         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 578         struct extent_io_tree *io_tree;
 579         int ret = 0;
 580
 581         if (list_empty(&async_cow->extents))
 582                 return 0;
 583
 584
 585         while (!list_empty(&async_cow->extents)) {
 586                 async_extent = list_entry(async_cow->extents.next,
 587                                           struct async_extent, list);
 588                 list_del(&async_extent->list);
 589
 590                 io_tree = &BTRFS_I(inode)->io_tree;
 591
 592 retry:
 593                 /* did the compression code fall back to uncompressed IO? */
 594                 if (!async_extent->pages) {
 595                         int page_started = 0;
 596                         unsigned long nr_written = 0;
 597
 598                         lock_extent(io_tree, async_extent->start,
 599                                          async_extent->start +
 600                                          async_extent->ram_size - 1, GFP_NOFS);
 601
 602                         /* allocate blocks */
 603                         ret = cow_file_range(inode, async_cow->locked_page,
 604                                              async_extent->start,
 605                                              async_extent->start +
 606                                              async_extent->ram_size - 1,
 607                                              &page_started, &nr_written, 0);
 608
 609                         /*
 610                          * if page_started, cow_file_range inserted an
 611                          * inline extent and took care of all the unlocking
 612                          * and IO for us.  Otherwise, we need to submit
 613                          * all those pages down to the drive.
 614                          */
 615                         if (!page_started && !ret)
 616                                 extent_write_locked_range(io_tree,
 617                                                   inode, async_extent->start,
 618                                                   async_extent->start +
 619                                                   async_extent->ram_size - 1,
 620                                                   btrfs_get_extent,
 621                                                   WB_SYNC_ALL);
 622                         kfree(async_extent);
 623                         cond_resched();
 624                         continue;
 625                 }
 626
 627                 lock_extent(io_tree, async_extent->start,
 628                             async_extent->start + async_extent->ram_size - 1,
 629                             GFP_NOFS);
 630
 631                 trans = btrfs_join_transaction(root);
 632                 BUG_ON(IS_ERR(trans));
 633                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 634                 ret = btrfs_reserve_extent(trans, root,
 635                                            async_extent->compressed_size,
 636                                            async_extent->compressed_size,
 637                                            0, alloc_hint,
 638                                            (u64)-1, &ins, 1);
 639                 btrfs_end_transaction(trans, root);
 640
 641                 if (ret) {
 642                         int i;
 643                         for (i = 0; i < async_extent->nr_pages; i++) {
 644                                 WARN_ON(async_extent->pages[i]->mapping);
 645                                 page_cache_release(async_extent->pages[i]);
 646                         }
 647                         kfree(async_extent->pages);
 648                         async_extent->nr_pages = 0;
 649                         async_extent->pages = NULL;
 650                         unlock_extent(io_tree, async_extent->start,
 651                                       async_extent->start +
 652                                       async_extent->ram_size - 1, GFP_NOFS);
 653                         goto retry;
 654                 }
 655
 656                 /*
 657                  * here we're doing allocation and writeback of the
 658                  * compressed pages
 659                  */
 660                 btrfs_drop_extent_cache(inode, async_extent->start,
 661                                         async_extent->start +
 662                                         async_extent->ram_size - 1, 0);
 663
 664                 em = alloc_extent_map();
 665                 BUG_ON(!em);
 666                 em->start = async_extent->start;
 667                 em->len = async_extent->ram_size;
 668                 em->orig_start = em->start;
 669
 670                 em->block_start = ins.objectid;
 671                 em->block_len = ins.offset;
 672                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 673                 em->compress_type = async_extent->compress_type;
 674                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 675                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 676
 677                 while (1) {
 678                         write_lock(&em_tree->lock);
 679                         ret = add_extent_mapping(em_tree, em);
 680                         write_unlock(&em_tree->lock);
 681                         if (ret != -EEXIST) {
 682                                 free_extent_map(em);
 683                                 break;
 684                         }
 685                         btrfs_drop_extent_cache(inode, async_extent->start,
 686                                                 async_extent->start +
 687                                                 async_extent->ram_size - 1, 0);
 688                 }
 689
 690                 ret = btrfs_add_ordered_extent_compress(inode,
 691                                                 async_extent->start,
 692                                                 ins.objectid,
 693                                                 async_extent->ram_size,
 694                                                 ins.offset,
 695                                                 BTRFS_ORDERED_COMPRESSED,
 696                                                 async_extent->compress_type);
 697                 BUG_ON(ret);
 698
 699                 /*
 700                  * clear dirty, set writeback and unlock the pages.
 701                  */
 702                 extent_clear_unlock_delalloc(inode,
 703                                 &BTRFS_I(inode)->io_tree,
 704                                 async_extent->start,
 705                                 async_extent->start +
 706                                 async_extent->ram_size - 1,
 707                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 708                                 EXTENT_CLEAR_UNLOCK |
 709                                 EXTENT_CLEAR_DELALLOC |
 710                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 711
 712                 ret = btrfs_submit_compressed_write(inode,
 713                                     async_extent->start,
 714                                     async_extent->ram_size,
 715                                     ins.objectid,
 716                                     ins.offset, async_extent->pages,
 717                                     async_extent->nr_pages);
 718
 719                 BUG_ON(ret);
 720                 alloc_hint = ins.objectid + ins.offset;
 721                 kfree(async_extent);
 722                 cond_resched();
 723         }
 724
 725         return 0;
 726 }
 727
 728 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 729                                       u64 num_bytes)
 730 {
 731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 732         struct extent_map *em;
 733         u64 alloc_hint = 0;
 734
 735         read_lock(&em_tree->lock);
 736         em = search_extent_mapping(em_tree, start, num_bytes);
 737         if (em) {
 738                 /*
 739                  * if block start isn't an actual block number then find the
 740                  * first block in this inode and use that as a hint.  If that
 741                  * block is also bogus then just don't worry about it.
 742                  */
 743                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 744                         free_extent_map(em);
 745                         em = search_extent_mapping(em_tree, 0, 0);
 746                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 747                                 alloc_hint = em->block_start;
 748                         if (em)
 749                                 free_extent_map(em);
 750                 } else {
 751                         alloc_hint = em->block_start;
 752                         free_extent_map(em);
 753                 }
 754         }
 755         read_unlock(&em_tree->lock);
 756
 757         return alloc_hint;
 758 }
 759
 760 /*
 761  * when extent_io.c finds a delayed allocation range in the file,
 762  * the call backs end up in this code.  The basic idea is to
 763  * allocate extents on disk for the range, and create ordered data structs
 764  * in ram to track those extents.
 765  *
 766  * locked_page is the page that writepage had locked already.  We use
 767  * it to make sure we don't do extra locks or unlocks.
 768  *
 769  * *page_started is set to one if we unlock locked_page and do everything
 770  * required to start IO on it.  It may be clean and already done with
 771  * IO when we return.
 772  */
 773 static noinline int cow_file_range(struct inode *inode,
 774                                    struct page *locked_page,
 775                                    u64 start, u64 end, int *page_started,
 776                                    unsigned long *nr_written,
 777                                    int unlock)
 778 {
 779         struct btrfs_root *root = BTRFS_I(inode)->root;
 780         struct btrfs_trans_handle *trans;
 781         u64 alloc_hint = 0;
 782         u64 num_bytes;
 783         unsigned long ram_size;
 784         u64 disk_num_bytes;
 785         u64 cur_alloc_size;
 786         u64 blocksize = root->sectorsize;
 787         struct btrfs_key ins;
 788         struct extent_map *em;
 789         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 790         int ret = 0;
 791
 792         BUG_ON(btrfs_is_free_space_inode(root, inode));
 793         trans = btrfs_join_transaction(root);
 794         BUG_ON(IS_ERR(trans));
 795         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 796
 797         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 798         num_bytes = max(blocksize,  num_bytes);
 799         disk_num_bytes = num_bytes;
 800         ret = 0;
 801
 802         /* if this is a small write inside eof, kick off defrag */
 803         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 804                 btrfs_add_inode_defrag(trans, inode);
 805
 806         if (start == 0) {
 807                 /* lets try to make an inline extent */
 808                 ret = cow_file_range_inline(trans, root, inode,
 809                                             start, end, 0, 0, NULL);
 810                 if (ret == 0) {
 811                         extent_clear_unlock_delalloc(inode,
 812                                      &BTRFS_I(inode)->io_tree,
 813                                      start, end, NULL,
 814                                      EXTENT_CLEAR_UNLOCK_PAGE |
 815                                      EXTENT_CLEAR_UNLOCK |
 816                                      EXTENT_CLEAR_DELALLOC |
 817                                      EXTENT_CLEAR_DIRTY |
 818                                      EXTENT_SET_WRITEBACK |
 819                                      EXTENT_END_WRITEBACK);
 820
 821                         *nr_written = *nr_written +
 822                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 823                         *page_started = 1;
 824                         ret = 0;
 825                         goto out;
 826                 }
 827         }
 828
 829         BUG_ON(disk_num_bytes >
 830                btrfs_super_total_bytes(root->fs_info->super_copy));
 831
 832         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 833         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 834
 835         while (disk_num_bytes > 0) {
 836                 unsigned long op;
 837
 838                 cur_alloc_size = disk_num_bytes;
 839                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 840                                            root->sectorsize, 0, alloc_hint,
 841                                            (u64)-1, &ins, 1);
 842                 BUG_ON(ret);
 843
 844                 em = alloc_extent_map();
 845                 BUG_ON(!em);
 846                 em->start = start;
 847                 em->orig_start = em->start;
 848                 ram_size = ins.offset;
 849                 em->len = ins.offset;
 850
 851                 em->block_start = ins.objectid;
 852                 em->block_len = ins.offset;
 853                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 854                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 855
 856                 while (1) {
 857                         write_lock(&em_tree->lock);
 858                         ret = add_extent_mapping(em_tree, em);
 859                         write_unlock(&em_tree->lock);
 860                         if (ret != -EEXIST) {
 861                                 free_extent_map(em);
 862                                 break;
 863                         }
 864                         btrfs_drop_extent_cache(inode, start,
 865                                                 start + ram_size - 1, 0);
 866                 }
 867
 868                 cur_alloc_size = ins.offset;
 869                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 870                                                ram_size, cur_alloc_size, 0);
 871                 BUG_ON(ret);
 872
 873                 if (root->root_key.objectid ==
 874                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 875                         ret = btrfs_reloc_clone_csums(inode, start,
 876                                                       cur_alloc_size);
 877                         BUG_ON(ret);
 878                 }
 879
 880                 if (disk_num_bytes < cur_alloc_size)
 881                         break;
 882
 883                 /* we're not doing compressed IO, don't unlock the first
 884                  * page (which the caller expects to stay locked), don't
 885                  * clear any dirty bits and don't set any writeback bits
 886                  *
 887                  * Do set the Private2 bit so we know this page was properly
 888                  * setup for writepage
 889                  */
 890                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 891                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 892                         EXTENT_SET_PRIVATE2;
 893
 894                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 895                                              start, start + ram_size - 1,
 896                                              locked_page, op);
 897                 disk_num_bytes -= cur_alloc_size;
 898                 num_bytes -= cur_alloc_size;
 899                 alloc_hint = ins.objectid + ins.offset;
 900                 start += cur_alloc_size;
 901         }
 902 out:
 903         ret = 0;
 904         btrfs_end_transaction(trans, root);
 905
 906         return ret;
 907 }
 908
 909 /*
 910  * work queue call back to started compression on a file and pages
 911  */
 912 static noinline void async_cow_start(struct btrfs_work *work)
 913 {
 914         struct async_cow *async_cow;
 915         int num_added = 0;
 916         async_cow = container_of(work, struct async_cow, work);
 917
 918         compress_file_range(async_cow->inode, async_cow->locked_page,
 919                             async_cow->start, async_cow->end, async_cow,
 920                             &num_added);
 921         if (num_added == 0)
 922                 async_cow->inode = NULL;
 923 }
 924
 925 /*
 926  * work queue call back to submit previously compressed pages
 927  */
 928 static noinline void async_cow_submit(struct btrfs_work *work)
 929 {
 930         struct async_cow *async_cow;
 931         struct btrfs_root *root;
 932         unsigned long nr_pages;
 933
 934         async_cow = container_of(work, struct async_cow, work);
 935
 936         root = async_cow->root;
 937         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 938                 PAGE_CACHE_SHIFT;
 939
 940         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 941
 942         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 943             5 * 1042 * 1024 &&
 944             waitqueue_active(&root->fs_info->async_submit_wait))
 945                 wake_up(&root->fs_info->async_submit_wait);
 946
 947         if (async_cow->inode)
 948                 submit_compressed_extents(async_cow->inode, async_cow);
 949 }
 950
 951 static noinline void async_cow_free(struct btrfs_work *work)
 952 {
 953         struct async_cow *async_cow;
 954         async_cow = container_of(work, struct async_cow, work);
 955         kfree(async_cow);
 956 }
 957
 958 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 959                                 u64 start, u64 end, int *page_started,
 960                                 unsigned long *nr_written)
 961 {
 962         struct async_cow *async_cow;
 963         struct btrfs_root *root = BTRFS_I(inode)->root;
 964         unsigned long nr_pages;
 965         u64 cur_end;
 966         int limit = 10 * 1024 * 1042;
 967
 968         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 969                          1, 0, NULL, GFP_NOFS);
 970         while (start < end) {
 971                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 972                 BUG_ON(!async_cow);
 973                 async_cow->inode = inode;
 974                 async_cow->root = root;
 975                 async_cow->locked_page = locked_page;
 976                 async_cow->start = start;
 977
 978                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 979                         cur_end = end;
 980                 else
 981                         cur_end = min(end, start + 512 * 1024 - 1);
 982
 983                 async_cow->end = cur_end;
 984                 INIT_LIST_HEAD(&async_cow->extents);
 985
 986                 async_cow->work.func = async_cow_start;
 987                 async_cow->work.ordered_func = async_cow_submit;
 988                 async_cow->work.ordered_free = async_cow_free;
 989                 async_cow->work.flags = 0;
 990
 991                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 992                         PAGE_CACHE_SHIFT;
 993                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 994
 995                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 996                                    &async_cow->work);
 997
 998                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1001                             limit));
1002                 }
1003
1004                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1005                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1006                         wait_event(root->fs_info->async_submit_wait,
1007                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008                            0));
1009                 }
1010
1011                 *nr_written += nr_pages;
1012                 start = cur_end + 1;
1013         }
1014         *page_started = 1;
1015         return 0;
1016 }
1017
1018 static noinline int csum_exist_in_range(struct btrfs_root *root,
1019                                         u64 bytenr, u64 num_bytes)
1020 {
1021         int ret;
1022         struct btrfs_ordered_sum *sums;
1023         LIST_HEAD(list);
1024
1025         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1026                                        bytenr + num_bytes - 1, &list, 0);
1027         if (ret == 0 && list_empty(&list))
1028                 return 0;
1029
1030         while (!list_empty(&list)) {
1031                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1032                 list_del(&sums->list);
1033                 kfree(sums);
1034         }
1035         return 1;
1036 }
1037
1038 /*
1039  * when nowcow writeback call back.  This checks for snapshots or COW copies
1040  * of the extents that exist in the file, and COWs the file as required.
1041  *
1042  * If no cow copies or snapshots exist, we write directly to the existing
1043  * blocks on disk
1044  */
1045 static noinline int run_delalloc_nocow(struct inode *inode,
1046                                        struct page *locked_page,
1047                               u64 start, u64 end, int *page_started, int force,
1048                               unsigned long *nr_written)
1049 {
1050         struct btrfs_root *root = BTRFS_I(inode)->root;
1051         struct btrfs_trans_handle *trans;
1052         struct extent_buffer *leaf;
1053         struct btrfs_path *path;
1054         struct btrfs_file_extent_item *fi;
1055         struct btrfs_key found_key;
1056         u64 cow_start;
1057         u64 cur_offset;
1058         u64 extent_end;
1059         u64 extent_offset;
1060         u64 disk_bytenr;
1061         u64 num_bytes;
1062         int extent_type;
1063         int ret;
1064         int type;
1065         int nocow;
1066         int check_prev = 1;
1067         bool nolock;
1068         u64 ino = btrfs_ino(inode);
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         nolock = btrfs_is_free_space_inode(root, inode);
1075
1076         if (nolock)
1077                 trans = btrfs_join_transaction_nolock(root);
1078         else
1079                 trans = btrfs_join_transaction(root);
1080
1081         BUG_ON(IS_ERR(trans));
1082         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083
1084         cow_start = (u64)-1;
1085         cur_offset = start;
1086         while (1) {
1087                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1088                                                cur_offset, 0);
1089                 BUG_ON(ret < 0);
1090                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1091                         leaf = path->nodes[0];
1092                         btrfs_item_key_to_cpu(leaf, &found_key,
1093                                               path->slots[0] - 1);
1094                         if (found_key.objectid == ino &&
1095                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1096                                 path->slots[0]--;
1097                 }
1098                 check_prev = 0;
1099 next_slot:
1100                 leaf = path->nodes[0];
1101                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1102                         ret = btrfs_next_leaf(root, path);
1103                         if (ret < 0)
1104                                 BUG_ON(1);
1105                         if (ret > 0)
1106                                 break;
1107                         leaf = path->nodes[0];
1108                 }
1109
1110                 nocow = 0;
1111                 disk_bytenr = 0;
1112                 num_bytes = 0;
1113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114
1115                 if (found_key.objectid > ino ||
1116                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1117                     found_key.offset > end)
1118                         break;
1119
1120                 if (found_key.offset > cur_offset) {
1121                         extent_end = found_key.offset;
1122                         extent_type = 0;
1123                         goto out_check;
1124                 }
1125
1126                 fi = btrfs_item_ptr(leaf, path->slots[0],
1127                                     struct btrfs_file_extent_item);
1128                 extent_type = btrfs_file_extent_type(leaf, fi);
1129
1130                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1131                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1132                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1133                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1134                         extent_end = found_key.offset +
1135                                 btrfs_file_extent_num_bytes(leaf, fi);
1136                         if (extent_end <= start) {
1137                                 path->slots[0]++;
1138                                 goto next_slot;
1139                         }
1140                         if (disk_bytenr == 0)
1141                                 goto out_check;
1142                         if (btrfs_file_extent_compression(leaf, fi) ||
1143                             btrfs_file_extent_encryption(leaf, fi) ||
1144                             btrfs_file_extent_other_encoding(leaf, fi))
1145                                 goto out_check;
1146                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147                                 goto out_check;
1148                         if (btrfs_extent_readonly(root, disk_bytenr))
1149                                 goto out_check;
1150                         if (btrfs_cross_ref_exist(trans, root, ino,
1151                                                   found_key.offset -
1152                                                   extent_offset, disk_bytenr))
1153                                 goto out_check;
1154                         disk_bytenr += extent_offset;
1155                         disk_bytenr += cur_offset - found_key.offset;
1156                         num_bytes = min(end + 1, extent_end) - cur_offset;
1157                         /*
1158                          * force cow if csum exists in the range.
1159                          * this ensure that csum for a given extent are
1160                          * either valid or do not exist.
1161                          */
1162                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1163                                 goto out_check;
1164                         nocow = 1;
1165                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1166                         extent_end = found_key.offset +
1167                                 btrfs_file_extent_inline_len(leaf, fi);
1168                         extent_end = ALIGN(extent_end, root->sectorsize);
1169                 } else {
1170                         BUG_ON(1);
1171                 }
1172 out_check:
1173                 if (extent_end <= start) {
1174                         path->slots[0]++;
1175                         goto next_slot;
1176                 }
1177                 if (!nocow) {
1178                         if (cow_start == (u64)-1)
1179                                 cow_start = cur_offset;
1180                         cur_offset = extent_end;
1181                         if (cur_offset > end)
1182                                 break;
1183                         path->slots[0]++;
1184                         goto next_slot;
1185                 }
1186
1187                 btrfs_release_path(path);
1188                 if (cow_start != (u64)-1) {
1189                         ret = cow_file_range(inode, locked_page, cow_start,
1190                                         found_key.offset - 1, page_started,
1191                                         nr_written, 1);
1192                         BUG_ON(ret);
1193                         cow_start = (u64)-1;
1194                 }
1195
1196                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1197                         struct extent_map *em;
1198                         struct extent_map_tree *em_tree;
1199                         em_tree = &BTRFS_I(inode)->extent_tree;
1200                         em = alloc_extent_map();
1201                         BUG_ON(!em);
1202                         em->start = cur_offset;
1203                         em->orig_start = em->start;
1204                         em->len = num_bytes;
1205                         em->block_len = num_bytes;
1206                         em->block_start = disk_bytenr;
1207                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1208                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209                         while (1) {
1210                                 write_lock(&em_tree->lock);
1211                                 ret = add_extent_mapping(em_tree, em);
1212                                 write_unlock(&em_tree->lock);
1213                                 if (ret != -EEXIST) {
1214                                         free_extent_map(em);
1215                                         break;
1216                                 }
1217                                 btrfs_drop_extent_cache(inode, em->start,
1218                                                 em->start + em->len - 1, 0);
1219                         }
1220                         type = BTRFS_ORDERED_PREALLOC;
1221                 } else {
1222                         type = BTRFS_ORDERED_NOCOW;
1223                 }
1224
1225                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1226                                                num_bytes, num_bytes, type);
1227                 BUG_ON(ret);
1228
1229                 if (root->root_key.objectid ==
1230                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1231                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1232                                                       num_bytes);
1233                         BUG_ON(ret);
1234                 }
1235
1236                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1237                                 cur_offset, cur_offset + num_bytes - 1,
1238                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1239                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1240                                 EXTENT_SET_PRIVATE2);
1241                 cur_offset = extent_end;
1242                 if (cur_offset > end)
1243                         break;
1244         }
1245         btrfs_release_path(path);
1246
1247         if (cur_offset <= end && cow_start == (u64)-1)
1248                 cow_start = cur_offset;
1249         if (cow_start != (u64)-1) {
1250                 ret = cow_file_range(inode, locked_page, cow_start, end,
1251                                      page_started, nr_written, 1);
1252                 BUG_ON(ret);
1253         }
1254
1255         if (nolock) {
1256                 ret = btrfs_end_transaction_nolock(trans, root);
1257                 BUG_ON(ret);
1258         } else {
1259                 ret = btrfs_end_transaction(trans, root);
1260                 BUG_ON(ret);
1261         }
1262         btrfs_free_path(path);
1263         return 0;
1264 }
1265
1266 /*
1267  * extent_io.c call back to do delayed allocation processing
1268  */
1269 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1270                               u64 start, u64 end, int *page_started,
1271                               unsigned long *nr_written)
1272 {
1273         int ret;
1274         struct btrfs_root *root = BTRFS_I(inode)->root;
1275
1276         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 1, nr_written);
1279         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1280                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1281                                          page_started, 0, nr_written);
1282         else if (!btrfs_test_opt(root, COMPRESS) &&
1283                  !(BTRFS_I(inode)->force_compress) &&
1284                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1285                 ret = cow_file_range(inode, locked_page, start, end,
1286                                       page_started, nr_written, 1);
1287         else
1288                 ret = cow_file_range_async(inode, locked_page, start, end,
1289                                            page_started, nr_written);
1290         return ret;
1291 }
1292
1293 static void btrfs_split_extent_hook(struct inode *inode,
1294                                     struct extent_state *orig, u64 split)
1295 {
1296         /* not delalloc, ignore it */
1297         if (!(orig->state & EXTENT_DELALLOC))
1298                 return;
1299
1300         spin_lock(&BTRFS_I(inode)->lock);
1301         BTRFS_I(inode)->outstanding_extents++;
1302         spin_unlock(&BTRFS_I(inode)->lock);
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static void btrfs_merge_extent_hook(struct inode *inode,
1312                                     struct extent_state *new,
1313                                     struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return;
1318
1319         spin_lock(&BTRFS_I(inode)->lock);
1320         BTRFS_I(inode)->outstanding_extents--;
1321         spin_unlock(&BTRFS_I(inode)->lock);
1322 }
1323
1324 /*
1325  * extent_io.c set_bit_hook, used to track delayed allocation
1326  * bytes in this file, and to maintain the list of inodes that
1327  * have pending delalloc work to be done.
1328  */
1329 static void btrfs_set_bit_hook(struct inode *inode,
1330                                struct extent_state *state, int *bits)
1331 {
1332
1333         /*
1334          * set_bit and clear bit hooks normally require _irqsave/restore
1335          * but in this case, we are only testing for the DELALLOC
1336          * bit, which is only set or cleared with irqs on
1337          */
1338         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1339                 struct btrfs_root *root = BTRFS_I(inode)->root;
1340                 u64 len = state->end + 1 - state->start;
1341                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342
1343                 if (*bits & EXTENT_FIRST_DELALLOC) {
1344                         *bits &= ~EXTENT_FIRST_DELALLOC;
1345                 } else {
1346                         spin_lock(&BTRFS_I(inode)->lock);
1347                         BTRFS_I(inode)->outstanding_extents++;
1348                         spin_unlock(&BTRFS_I(inode)->lock);
1349                 }
1350
1351                 spin_lock(&root->fs_info->delalloc_lock);
1352                 BTRFS_I(inode)->delalloc_bytes += len;
1353                 root->fs_info->delalloc_bytes += len;
1354                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1355                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1356                                       &root->fs_info->delalloc_inodes);
1357                 }
1358                 spin_unlock(&root->fs_info->delalloc_lock);
1359         }
1360 }
1361
1362 /*
1363  * extent_io.c clear_bit_hook, see set_bit_hook for why
1364  */
1365 static void btrfs_clear_bit_hook(struct inode *inode,
1366                                  struct extent_state *state, int *bits)
1367 {
1368         /*
1369          * set_bit and clear bit hooks normally require _irqsave/restore
1370          * but in this case, we are only testing for the DELALLOC
1371          * bit, which is only set or cleared with irqs on
1372          */
1373         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1374                 struct btrfs_root *root = BTRFS_I(inode)->root;
1375                 u64 len = state->end + 1 - state->start;
1376                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377
1378                 if (*bits & EXTENT_FIRST_DELALLOC) {
1379                         *bits &= ~EXTENT_FIRST_DELALLOC;
1380                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1381                         spin_lock(&BTRFS_I(inode)->lock);
1382                         BTRFS_I(inode)->outstanding_extents--;
1383                         spin_unlock(&BTRFS_I(inode)->lock);
1384                 }
1385
1386                 if (*bits & EXTENT_DO_ACCOUNTING)
1387                         btrfs_delalloc_release_metadata(inode, len);
1388
1389                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390                     && do_list)
1391                         btrfs_free_reserved_data_space(inode, len);
1392
1393                 spin_lock(&root->fs_info->delalloc_lock);
1394                 root->fs_info->delalloc_bytes -= len;
1395                 BTRFS_I(inode)->delalloc_bytes -= len;
1396
1397                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1398                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1399                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400                 }
1401                 spin_unlock(&root->fs_info->delalloc_lock);
1402         }
1403 }
1404
1405 /*
1406  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407  * we don't create bios that span stripes or chunks
1408  */
1409 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1410                          size_t size, struct bio *bio,
1411                          unsigned long bio_flags)
1412 {
1413         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1414         struct btrfs_mapping_tree *map_tree;
1415         u64 logical = (u64)bio->bi_sector << 9;
1416         u64 length = 0;
1417         u64 map_length;
1418         int ret;
1419
1420         if (bio_flags & EXTENT_BIO_COMPRESSED)
1421                 return 0;
1422
1423         length = bio->bi_size;
1424         map_tree = &root->fs_info->mapping_tree;
1425         map_length = length;
1426         ret = btrfs_map_block(map_tree, READ, logical,
1427                               &map_length, NULL, 0);
1428
1429         if (map_length < length + size)
1430                 return 1;
1431         return ret;
1432 }
1433
1434 /*
1435  * in order to insert checksums into the metadata in large chunks,
1436  * we wait until bio submission time.   All the pages in the bio are
1437  * checksummed and sums are attached onto the ordered extent record.
1438  *
1439  * At IO completion time the cums attached on the ordered extent record
1440  * are inserted into the btree
1441  */
1442 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443                                     struct bio *bio, int mirror_num,
1444                                     unsigned long bio_flags,
1445                                     u64 bio_offset)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         int ret = 0;
1449
1450         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1451         BUG_ON(ret);
1452         return 0;
1453 }
1454
1455 /*
1456  * in order to insert checksums into the metadata in large chunks,
1457  * we wait until bio submission time.   All the pages in the bio are
1458  * checksummed and sums are attached onto the ordered extent record.
1459  *
1460  * At IO completion time the cums attached on the ordered extent record
1461  * are inserted into the btree
1462  */
1463 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1464                           int mirror_num, unsigned long bio_flags,
1465                           u64 bio_offset)
1466 {
1467         struct btrfs_root *root = BTRFS_I(inode)->root;
1468         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1469 }
1470
1471 /*
1472  * extent_io.c submission hook. This does the right thing for csum calculation
1473  * on write, or reading the csums from the tree before a read
1474  */
1475 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1476                           int mirror_num, unsigned long bio_flags,
1477                           u64 bio_offset)
1478 {
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480         int ret = 0;
1481         int skip_sum;
1482
1483         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484
1485         if (btrfs_is_free_space_inode(root, inode))
1486                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487         else
1488                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1489         BUG_ON(ret);
1490
1491         if (!(rw & REQ_WRITE)) {
1492                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1493                         return btrfs_submit_compressed_read(inode, bio,
1494                                                     mirror_num, bio_flags);
1495                 } else if (!skip_sum) {
1496                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1497                         if (ret)
1498                                 return ret;
1499                 }
1500                 goto mapit;
1501         } else if (!skip_sum) {
1502                 /* csum items have already been cloned */
1503                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504                         goto mapit;
1505                 /* we're doing a write, do the async checksumming */
1506                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1507                                    inode, rw, bio, mirror_num,
1508                                    bio_flags, bio_offset,
1509                                    __btrfs_submit_bio_start,
1510                                    __btrfs_submit_bio_done);
1511         }
1512
1513 mapit:
1514         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1515 }
1516
1517 /*
1518  * given a list of ordered sums record them in the inode.  This happens
1519  * at IO completion time based on sums calculated at bio submission time.
1520  */
1521 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1522                              struct inode *inode, u64 file_offset,
1523                              struct list_head *list)
1524 {
1525         struct btrfs_ordered_sum *sum;
1526
1527         list_for_each_entry(sum, list, list) {
1528                 btrfs_csum_file_blocks(trans,
1529                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1530         }
1531         return 0;
1532 }
1533
1534 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1535                               struct extent_state **cached_state)
1536 {
1537         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538                 WARN_ON(1);
1539         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1540                                    cached_state, GFP_NOFS);
1541 }
1542
1543 /* see btrfs_writepage_start_hook for details on why this is required */
1544 struct btrfs_writepage_fixup {
1545         struct page *page;
1546         struct btrfs_work work;
1547 };
1548
1549 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 {
1551         struct btrfs_writepage_fixup *fixup;
1552         struct btrfs_ordered_extent *ordered;
1553         struct extent_state *cached_state = NULL;
1554         struct page *page;
1555         struct inode *inode;
1556         u64 page_start;
1557         u64 page_end;
1558
1559         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560         page = fixup->page;
1561 again:
1562         lock_page(page);
1563         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564                 ClearPageChecked(page);
1565                 goto out_page;
1566         }
1567
1568         inode = page->mapping->host;
1569         page_start = page_offset(page);
1570         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571
1572         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573                          &cached_state, GFP_NOFS);
1574
1575         /* already ordered? We're done */
1576         if (PagePrivate2(page))
1577                 goto out;
1578
1579         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580         if (ordered) {
1581                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582                                      page_end, &cached_state, GFP_NOFS);
1583                 unlock_page(page);
1584                 btrfs_start_ordered_extent(inode, ordered, 1);
1585                 goto again;
1586         }
1587
1588         BUG();
1589         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1590         ClearPageChecked(page);
1591 out:
1592         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1593                              &cached_state, GFP_NOFS);
1594 out_page:
1595         unlock_page(page);
1596         page_cache_release(page);
1597         kfree(fixup);
1598 }
1599
1600 /*
1601  * There are a few paths in the higher layers of the kernel that directly
1602  * set the page dirty bit without asking the filesystem if it is a
1603  * good idea.  This causes problems because we want to make sure COW
1604  * properly happens and the data=ordered rules are followed.
1605  *
1606  * In our case any range that doesn't have the ORDERED bit set
1607  * hasn't been properly setup for IO.  We kick off an async process
1608  * to fix it up.  The async helper will wait for ordered extents, set
1609  * the delalloc bit and make it safe to write the page.
1610  */
1611 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1612 {
1613         struct inode *inode = page->mapping->host;
1614         struct btrfs_writepage_fixup *fixup;
1615         struct btrfs_root *root = BTRFS_I(inode)->root;
1616
1617         /* this page is properly in the ordered list */
1618         if (TestClearPagePrivate2(page))
1619                 return 0;
1620
1621         if (PageChecked(page))
1622                 return -EAGAIN;
1623
1624         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1625         if (!fixup)
1626                 return -EAGAIN;
1627
1628         SetPageChecked(page);
1629         page_cache_get(page);
1630         fixup->work.func = btrfs_writepage_fixup_worker;
1631         fixup->page = page;
1632         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1633         return -EAGAIN;
1634 }
1635
1636 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1637                                        struct inode *inode, u64 file_pos,
1638                                        u64 disk_bytenr, u64 disk_num_bytes,
1639                                        u64 num_bytes, u64 ram_bytes,
1640                                        u8 compression, u8 encryption,
1641                                        u16 other_encoding, int extent_type)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         struct btrfs_file_extent_item *fi;
1645         struct btrfs_path *path;
1646         struct extent_buffer *leaf;
1647         struct btrfs_key ins;
1648         u64 hint;
1649         int ret;
1650
1651         path = btrfs_alloc_path();
1652         if (!path)
1653                 return -ENOMEM;
1654
1655         path->leave_spinning = 1;
1656
1657         /*
1658          * we may be replacing one extent in the tree with another.
1659          * The new extent is pinned in the extent map, and we don't want
1660          * to drop it from the cache until it is completely in the btree.
1661          *
1662          * So, tell btrfs_drop_extents to leave this extent in the cache.
1663          * the caller is expected to unpin it and allow it to be merged
1664          * with the others.
1665          */
1666         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1667                                  &hint, 0);
1668         BUG_ON(ret);
1669
1670         ins.objectid = btrfs_ino(inode);
1671         ins.offset = file_pos;
1672         ins.type = BTRFS_EXTENT_DATA_KEY;
1673         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1674         BUG_ON(ret);
1675         leaf = path->nodes[0];
1676         fi = btrfs_item_ptr(leaf, path->slots[0],
1677                             struct btrfs_file_extent_item);
1678         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1679         btrfs_set_file_extent_type(leaf, fi, extent_type);
1680         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1681         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1682         btrfs_set_file_extent_offset(leaf, fi, 0);
1683         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1684         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1685         btrfs_set_file_extent_compression(leaf, fi, compression);
1686         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1687         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1688
1689         btrfs_unlock_up_safe(path, 1);
1690         btrfs_set_lock_blocking(leaf);
1691
1692         btrfs_mark_buffer_dirty(leaf);
1693
1694         inode_add_bytes(inode, num_bytes);
1695
1696         ins.objectid = disk_bytenr;
1697         ins.offset = disk_num_bytes;
1698         ins.type = BTRFS_EXTENT_ITEM_KEY;
1699         ret = btrfs_alloc_reserved_file_extent(trans, root,
1700                                         root->root_key.objectid,
1701                                         btrfs_ino(inode), file_pos, &ins);
1702         BUG_ON(ret);
1703         btrfs_free_path(path);
1704
1705         return 0;
1706 }
1707
1708 /*
1709  * helper function for btrfs_finish_ordered_io, this
1710  * just reads in some of the csum leaves to prime them into ram
1711  * before we start the transaction.  It limits the amount of btree
1712  * reads required while inside the transaction.
1713  */
1714 /* as ordered data IO finishes, this gets called so we can finish
1715  * an ordered extent if the range of bytes in the file it covers are
1716  * fully written.
1717  */
1718 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1719 {
1720         struct btrfs_root *root = BTRFS_I(inode)->root;
1721         struct btrfs_trans_handle *trans = NULL;
1722         struct btrfs_ordered_extent *ordered_extent = NULL;
1723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1724         struct extent_state *cached_state = NULL;
1725         int compress_type = 0;
1726         int ret;
1727         bool nolock;
1728
1729         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1730                                              end - start + 1);
1731         if (!ret)
1732                 return 0;
1733         BUG_ON(!ordered_extent);
1734
1735         nolock = btrfs_is_free_space_inode(root, inode);
1736
1737         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1738                 BUG_ON(!list_empty(&ordered_extent->list));
1739                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1740                 if (!ret) {
1741                         if (nolock)
1742                                 trans = btrfs_join_transaction_nolock(root);
1743                         else
1744                                 trans = btrfs_join_transaction(root);
1745                         BUG_ON(IS_ERR(trans));
1746                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1747                         ret = btrfs_update_inode_fallback(trans, root, inode);
1748                         BUG_ON(ret);
1749                 }
1750                 goto out;
1751         }
1752
1753         lock_extent_bits(io_tree, ordered_extent->file_offset,
1754                          ordered_extent->file_offset + ordered_extent->len - 1,
1755                          0, &cached_state, GFP_NOFS);
1756
1757         if (nolock)
1758                 trans = btrfs_join_transaction_nolock(root);
1759         else
1760                 trans = btrfs_join_transaction(root);
1761         BUG_ON(IS_ERR(trans));
1762         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1763
1764         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1765                 compress_type = ordered_extent->compress_type;
1766         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1767                 BUG_ON(compress_type);
1768                 ret = btrfs_mark_extent_written(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->file_offset +
1771                                                 ordered_extent->len);
1772                 BUG_ON(ret);
1773         } else {
1774                 BUG_ON(root == root->fs_info->tree_root);
1775                 ret = insert_reserved_file_extent(trans, inode,
1776                                                 ordered_extent->file_offset,
1777                                                 ordered_extent->start,
1778                                                 ordered_extent->disk_len,
1779                                                 ordered_extent->len,
1780                                                 ordered_extent->len,
1781                                                 compress_type, 0, 0,
1782                                                 BTRFS_FILE_EXTENT_REG);
1783                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1784                                    ordered_extent->file_offset,
1785                                    ordered_extent->len);
1786                 BUG_ON(ret);
1787         }
1788         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1789                              ordered_extent->file_offset +
1790                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1791
1792         add_pending_csums(trans, inode, ordered_extent->file_offset,
1793                           &ordered_extent->list);
1794
1795         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1796         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1797                 ret = btrfs_update_inode_fallback(trans, root, inode);
1798                 BUG_ON(ret);
1799         }
1800         ret = 0;
1801 out:
1802         if (root != root->fs_info->tree_root)
1803                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1804         if (trans) {
1805                 if (nolock)
1806                         btrfs_end_transaction_nolock(trans, root);
1807                 else
1808                         btrfs_end_transaction(trans, root);
1809         }
1810
1811         /* once for us */
1812         btrfs_put_ordered_extent(ordered_extent);
1813         /* once for the tree */
1814         btrfs_put_ordered_extent(ordered_extent);
1815
1816         return 0;
1817 }
1818
1819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1820                                 struct extent_state *state, int uptodate)
1821 {
1822         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1823
1824         ClearPagePrivate2(page);
1825         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1826 }
1827
1828 /*
1829  * when reads are done, we need to check csums to verify the data is correct
1830  * if there's a match, we allow the bio to finish.  If not, the code in
1831  * extent_io.c will try to find good copies for us.
1832  */
1833 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1834                                struct extent_state *state)
1835 {
1836         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1837         struct inode *inode = page->mapping->host;
1838         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1839         char *kaddr;
1840         u64 private = ~(u32)0;
1841         int ret;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843         u32 csum = ~(u32)0;
1844
1845         if (PageChecked(page)) {
1846                 ClearPageChecked(page);
1847                 goto good;
1848         }
1849
1850         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1851                 goto good;
1852
1853         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1854             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1855                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1856                                   GFP_NOFS);
1857                 return 0;
1858         }
1859
1860         if (state && state->start == start) {
1861                 private = state->private;
1862                 ret = 0;
1863         } else {
1864                 ret = get_state_private(io_tree, start, &private);
1865         }
1866         kaddr = kmap_atomic(page, KM_USER0);
1867         if (ret)
1868                 goto zeroit;
1869
1870         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1871         btrfs_csum_final(csum, (char *)&csum);
1872         if (csum != private)
1873                 goto zeroit;
1874
1875         kunmap_atomic(kaddr, KM_USER0);
1876 good:
1877         return 0;
1878
1879 zeroit:
1880         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1881                        "private %llu\n",
1882                        (unsigned long long)btrfs_ino(page->mapping->host),
1883                        (unsigned long long)start, csum,
1884                        (unsigned long long)private);
1885         memset(kaddr + offset, 1, end - start + 1);
1886         flush_dcache_page(page);
1887         kunmap_atomic(kaddr, KM_USER0);
1888         if (private == 0)
1889                 return 0;
1890         return -EIO;
1891 }
1892
1893 struct delayed_iput {
1894         struct list_head list;
1895         struct inode *inode;
1896 };
1897
1898 void btrfs_add_delayed_iput(struct inode *inode)
1899 {
1900         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1901         struct delayed_iput *delayed;
1902
1903         if (atomic_add_unless(&inode->i_count, -1, 1))
1904                 return;
1905
1906         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1907         delayed->inode = inode;
1908
1909         spin_lock(&fs_info->delayed_iput_lock);
1910         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1911         spin_unlock(&fs_info->delayed_iput_lock);
1912 }
1913
1914 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1915 {
1916         LIST_HEAD(list);
1917         struct btrfs_fs_info *fs_info = root->fs_info;
1918         struct delayed_iput *delayed;
1919         int empty;
1920
1921         spin_lock(&fs_info->delayed_iput_lock);
1922         empty = list_empty(&fs_info->delayed_iputs);
1923         spin_unlock(&fs_info->delayed_iput_lock);
1924         if (empty)
1925                 return;
1926
1927         down_read(&root->fs_info->cleanup_work_sem);
1928         spin_lock(&fs_info->delayed_iput_lock);
1929         list_splice_init(&fs_info->delayed_iputs, &list);
1930         spin_unlock(&fs_info->delayed_iput_lock);
1931
1932         while (!list_empty(&list)) {
1933                 delayed = list_entry(list.next, struct delayed_iput, list);
1934                 list_del(&delayed->list);
1935                 iput(delayed->inode);
1936                 kfree(delayed);
1937         }
1938         up_read(&root->fs_info->cleanup_work_sem);
1939 }
1940
1941 enum btrfs_orphan_cleanup_state {
1942         ORPHAN_CLEANUP_STARTED  = 1,
1943         ORPHAN_CLEANUP_DONE     = 2,
1944 };
1945
1946 /*
1947  * This is called in transaction commmit time. If there are no orphan
1948  * files in the subvolume, it removes orphan item and frees block_rsv
1949  * structure.
1950  */
1951 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1952                               struct btrfs_root *root)
1953 {
1954         struct btrfs_block_rsv *block_rsv;
1955         int ret;
1956
1957         if (!list_empty(&root->orphan_list) ||
1958             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1959                 return;
1960
1961         spin_lock(&root->orphan_lock);
1962         if (!list_empty(&root->orphan_list)) {
1963                 spin_unlock(&root->orphan_lock);
1964                 return;
1965         }
1966
1967         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1968                 spin_unlock(&root->orphan_lock);
1969                 return;
1970         }
1971
1972         block_rsv = root->orphan_block_rsv;
1973         root->orphan_block_rsv = NULL;
1974         spin_unlock(&root->orphan_lock);
1975
1976         if (root->orphan_item_inserted &&
1977             btrfs_root_refs(&root->root_item) > 0) {
1978                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1979                                             root->root_key.objectid);
1980                 BUG_ON(ret);
1981                 root->orphan_item_inserted = 0;
1982         }
1983
1984         if (block_rsv) {
1985                 WARN_ON(block_rsv->size > 0);
1986                 btrfs_free_block_rsv(root, block_rsv);
1987         }
1988 }
1989
1990 /*
1991  * This creates an orphan entry for the given inode in case something goes
1992  * wrong in the middle of an unlink/truncate.
1993  *
1994  * NOTE: caller of this function should reserve 5 units of metadata for
1995  *       this function.
1996  */
1997 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1998 {
1999         struct btrfs_root *root = BTRFS_I(inode)->root;
2000         struct btrfs_block_rsv *block_rsv = NULL;
2001         int reserve = 0;
2002         int insert = 0;
2003         int ret;
2004
2005         if (!root->orphan_block_rsv) {
2006                 block_rsv = btrfs_alloc_block_rsv(root);
2007                 if (!block_rsv)
2008                         return -ENOMEM;
2009         }
2010
2011         spin_lock(&root->orphan_lock);
2012         if (!root->orphan_block_rsv) {
2013                 root->orphan_block_rsv = block_rsv;
2014         } else if (block_rsv) {
2015                 btrfs_free_block_rsv(root, block_rsv);
2016                 block_rsv = NULL;
2017         }
2018
2019         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2020                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2021 #if 0
2022                 /*
2023                  * For proper ENOSPC handling, we should do orphan
2024                  * cleanup when mounting. But this introduces backward
2025                  * compatibility issue.
2026                  */
2027                 if (!xchg(&root->orphan_item_inserted, 1))
2028                         insert = 2;
2029                 else
2030                         insert = 1;
2031 #endif
2032                 insert = 1;
2033         }
2034
2035         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2036                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2037                 reserve = 1;
2038         }
2039         spin_unlock(&root->orphan_lock);
2040
2041         /* grab metadata reservation from transaction handle */
2042         if (reserve) {
2043                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2044                 BUG_ON(ret);
2045         }
2046
2047         /* insert an orphan item to track this unlinked/truncated file */
2048         if (insert >= 1) {
2049                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2050                 BUG_ON(ret && ret != -EEXIST);
2051         }
2052
2053         /* insert an orphan item to track subvolume contains orphan files */
2054         if (insert >= 2) {
2055                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2056                                                root->root_key.objectid);
2057                 BUG_ON(ret);
2058         }
2059         return 0;
2060 }
2061
2062 /*
2063  * We have done the truncate/delete so we can go ahead and remove the orphan
2064  * item for this particular inode.
2065  */
2066 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2067 {
2068         struct btrfs_root *root = BTRFS_I(inode)->root;
2069         int delete_item = 0;
2070         int release_rsv = 0;
2071         int ret = 0;
2072
2073         spin_lock(&root->orphan_lock);
2074         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2075                 list_del_init(&BTRFS_I(inode)->i_orphan);
2076                 delete_item = 1;
2077         }
2078
2079         if (BTRFS_I(inode)->orphan_meta_reserved) {
2080                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2081                 release_rsv = 1;
2082         }
2083         spin_unlock(&root->orphan_lock);
2084
2085         if (trans && delete_item) {
2086                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2087                 BUG_ON(ret);
2088         }
2089
2090         if (release_rsv)
2091                 btrfs_orphan_release_metadata(inode);
2092
2093         return 0;
2094 }
2095
2096 /*
2097  * this cleans up any orphans that may be left on the list from the last use
2098  * of this root.
2099  */
2100 int btrfs_orphan_cleanup(struct btrfs_root *root)
2101 {
2102         struct btrfs_path *path;
2103         struct extent_buffer *leaf;
2104         struct btrfs_key key, found_key;
2105         struct btrfs_trans_handle *trans;
2106         struct inode *inode;
2107         u64 last_objectid = 0;
2108         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2109
2110         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2111                 return 0;
2112
2113         path = btrfs_alloc_path();
2114         if (!path) {
2115                 ret = -ENOMEM;
2116                 goto out;
2117         }
2118         path->reada = -1;
2119
2120         key.objectid = BTRFS_ORPHAN_OBJECTID;
2121         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2122         key.offset = (u64)-1;
2123
2124         while (1) {
2125                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2126                 if (ret < 0)
2127                         goto out;
2128
2129                 /*
2130                  * if ret == 0 means we found what we were searching for, which
2131                  * is weird, but possible, so only screw with path if we didn't
2132                  * find the key and see if we have stuff that matches
2133                  */
2134                 if (ret > 0) {
2135                         ret = 0;
2136                         if (path->slots[0] == 0)
2137                                 break;
2138                         path->slots[0]--;
2139                 }
2140
2141                 /* pull out the item */
2142                 leaf = path->nodes[0];
2143                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2144
2145                 /* make sure the item matches what we want */
2146                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2147                         break;
2148                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2149                         break;
2150
2151                 /* release the path since we're done with it */
2152                 btrfs_release_path(path);
2153
2154                 /*
2155                  * this is where we are basically btrfs_lookup, without the
2156                  * crossing root thing.  we store the inode number in the
2157                  * offset of the orphan item.
2158                  */
2159
2160                 if (found_key.offset == last_objectid) {
2161                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2162                                "stopping orphan cleanup\n");
2163                         ret = -EINVAL;
2164                         goto out;
2165                 }
2166
2167                 last_objectid = found_key.offset;
2168
2169                 found_key.objectid = found_key.offset;
2170                 found_key.type = BTRFS_INODE_ITEM_KEY;
2171                 found_key.offset = 0;
2172                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2173                 ret = PTR_RET(inode);
2174                 if (ret && ret != -ESTALE)
2175                         goto out;
2176
2177                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2178                         struct btrfs_root *dead_root;
2179                         struct btrfs_fs_info *fs_info = root->fs_info;
2180                         int is_dead_root = 0;
2181
2182                         /*
2183                          * this is an orphan in the tree root. Currently these
2184                          * could come from 2 sources:
2185                          *  a) a snapshot deletion in progress
2186                          *  b) a free space cache inode
2187                          * We need to distinguish those two, as the snapshot
2188                          * orphan must not get deleted.
2189                          * find_dead_roots already ran before us, so if this
2190                          * is a snapshot deletion, we should find the root
2191                          * in the dead_roots list
2192                          */
2193                         spin_lock(&fs_info->trans_lock);
2194                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2195                                             root_list) {
2196                                 if (dead_root->root_key.objectid ==
2197                                     found_key.objectid) {
2198                                         is_dead_root = 1;
2199                                         break;
2200                                 }
2201                         }
2202                         spin_unlock(&fs_info->trans_lock);
2203                         if (is_dead_root) {
2204                                 /* prevent this orphan from being found again */
2205                                 key.offset = found_key.objectid - 1;
2206                                 continue;
2207                         }
2208                 }
2209                 /*
2210                  * Inode is already gone but the orphan item is still there,
2211                  * kill the orphan item.
2212                  */
2213                 if (ret == -ESTALE) {
2214                         trans = btrfs_start_transaction(root, 1);
2215                         if (IS_ERR(trans)) {
2216                                 ret = PTR_ERR(trans);
2217                                 goto out;
2218                         }
2219                         ret = btrfs_del_orphan_item(trans, root,
2220                                                     found_key.objectid);
2221                         BUG_ON(ret);
2222                         btrfs_end_transaction(trans, root);
2223                         continue;
2224                 }
2225
2226                 /*
2227                  * add this inode to the orphan list so btrfs_orphan_del does
2228                  * the proper thing when we hit it
2229                  */
2230                 spin_lock(&root->orphan_lock);
2231                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2232                 spin_unlock(&root->orphan_lock);
2233
2234                 /* if we have links, this was a truncate, lets do that */
2235                 if (inode->i_nlink) {
2236                         if (!S_ISREG(inode->i_mode)) {
2237                                 WARN_ON(1);
2238                                 iput(inode);
2239                                 continue;
2240                         }
2241                         nr_truncate++;
2242                         /*
2243                          * Need to hold the imutex for reservation purposes, not
2244                          * a huge deal here but I have a WARN_ON in
2245                          * btrfs_delalloc_reserve_space to catch offenders.
2246                          */
2247                         mutex_lock(&inode->i_mutex);
2248                         ret = btrfs_truncate(inode);
2249                         mutex_unlock(&inode->i_mutex);
2250                 } else {
2251                         nr_unlink++;
2252                 }
2253
2254                 /* this will do delete_inode and everything for us */
2255                 iput(inode);
2256                 if (ret)
2257                         goto out;
2258         }
2259         /* release the path since we're done with it */
2260         btrfs_release_path(path);
2261
2262         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2263
2264         if (root->orphan_block_rsv)
2265                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2266                                         (u64)-1);
2267
2268         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2269                 trans = btrfs_join_transaction(root);
2270                 if (!IS_ERR(trans))
2271                         btrfs_end_transaction(trans, root);
2272         }
2273
2274         if (nr_unlink)
2275                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2276         if (nr_truncate)
2277                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2278
2279 out:
2280         if (ret)
2281                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2282         btrfs_free_path(path);
2283         return ret;
2284 }
2285
2286 /*
2287  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2288  * don't find any xattrs, we know there can't be any acls.
2289  *
2290  * slot is the slot the inode is in, objectid is the objectid of the inode
2291  */
2292 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2293                                           int slot, u64 objectid)
2294 {
2295         u32 nritems = btrfs_header_nritems(leaf);
2296         struct btrfs_key found_key;
2297         int scanned = 0;
2298
2299         slot++;
2300         while (slot < nritems) {
2301                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2302
2303                 /* we found a different objectid, there must not be acls */
2304                 if (found_key.objectid != objectid)
2305                         return 0;
2306
2307                 /* we found an xattr, assume we've got an acl */
2308                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2309                         return 1;
2310
2311                 /*
2312                  * we found a key greater than an xattr key, there can't
2313                  * be any acls later on
2314                  */
2315                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2316                         return 0;
2317
2318                 slot++;
2319                 scanned++;
2320
2321                 /*
2322                  * it goes inode, inode backrefs, xattrs, extents,
2323                  * so if there are a ton of hard links to an inode there can
2324                  * be a lot of backrefs.  Don't waste time searching too hard,
2325                  * this is just an optimization
2326                  */
2327                 if (scanned >= 8)
2328                         break;
2329         }
2330         /* we hit the end of the leaf before we found an xattr or
2331          * something larger than an xattr.  We have to assume the inode
2332          * has acls
2333          */
2334         return 1;
2335 }
2336
2337 /*
2338  * read an inode from the btree into the in-memory inode
2339  */
2340 static void btrfs_read_locked_inode(struct inode *inode)
2341 {
2342         struct btrfs_path *path;
2343         struct extent_buffer *leaf;
2344         struct btrfs_inode_item *inode_item;
2345         struct btrfs_timespec *tspec;
2346         struct btrfs_root *root = BTRFS_I(inode)->root;
2347         struct btrfs_key location;
2348         int maybe_acls;
2349         u32 rdev;
2350         int ret;
2351         bool filled = false;
2352
2353         ret = btrfs_fill_inode(inode, &rdev);
2354         if (!ret)
2355                 filled = true;
2356
2357         path = btrfs_alloc_path();
2358         if (!path)
2359                 goto make_bad;
2360
2361         path->leave_spinning = 1;
2362         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2363
2364         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2365         if (ret)
2366                 goto make_bad;
2367
2368         leaf = path->nodes[0];
2369
2370         if (filled)
2371                 goto cache_acl;
2372
2373         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2374                                     struct btrfs_inode_item);
2375         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2376         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2377         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2378         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2379         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2380
2381         tspec = btrfs_inode_atime(inode_item);
2382         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2383         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2384
2385         tspec = btrfs_inode_mtime(inode_item);
2386         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2387         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2388
2389         tspec = btrfs_inode_ctime(inode_item);
2390         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2391         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2392
2393         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2394         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2395         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2396         inode->i_generation = BTRFS_I(inode)->generation;
2397         inode->i_rdev = 0;
2398         rdev = btrfs_inode_rdev(leaf, inode_item);
2399
2400         BTRFS_I(inode)->index_cnt = (u64)-1;
2401         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2402 cache_acl:
2403         /*
2404          * try to precache a NULL acl entry for files that don't have
2405          * any xattrs or acls
2406          */
2407         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2408                                            btrfs_ino(inode));
2409         if (!maybe_acls)
2410                 cache_no_acl(inode);
2411
2412         btrfs_free_path(path);
2413
2414         switch (inode->i_mode & S_IFMT) {
2415         case S_IFREG:
2416                 inode->i_mapping->a_ops = &btrfs_aops;
2417                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2418                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2419                 inode->i_fop = &btrfs_file_operations;
2420                 inode->i_op = &btrfs_file_inode_operations;
2421                 break;
2422         case S_IFDIR:
2423                 inode->i_fop = &btrfs_dir_file_operations;
2424                 if (root == root->fs_info->tree_root)
2425                         inode->i_op = &btrfs_dir_ro_inode_operations;
2426                 else
2427                         inode->i_op = &btrfs_dir_inode_operations;
2428                 break;
2429         case S_IFLNK:
2430                 inode->i_op = &btrfs_symlink_inode_operations;
2431                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2432                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2433                 break;
2434         default:
2435                 inode->i_op = &btrfs_special_inode_operations;
2436                 init_special_inode(inode, inode->i_mode, rdev);
2437                 break;
2438         }
2439
2440         btrfs_update_iflags(inode);
2441         return;
2442
2443 make_bad:
2444         btrfs_free_path(path);
2445         make_bad_inode(inode);
2446 }
2447
2448 /*
2449  * given a leaf and an inode, copy the inode fields into the leaf
2450  */
2451 static void fill_inode_item(struct btrfs_trans_handle *trans,
2452                             struct extent_buffer *leaf,
2453                             struct btrfs_inode_item *item,
2454                             struct inode *inode)
2455 {
2456         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2457         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2458         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2459         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2460         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2461
2462         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2463                                inode->i_atime.tv_sec);
2464         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2465                                 inode->i_atime.tv_nsec);
2466
2467         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2468                                inode->i_mtime.tv_sec);
2469         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2470                                 inode->i_mtime.tv_nsec);
2471
2472         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2473                                inode->i_ctime.tv_sec);
2474         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2475                                 inode->i_ctime.tv_nsec);
2476
2477         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2478         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2479         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2480         btrfs_set_inode_transid(leaf, item, trans->transid);
2481         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2482         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2483         btrfs_set_inode_block_group(leaf, item, 0);
2484 }
2485
2486 /*
2487  * copy everything in the in-memory inode into the btree.
2488  */
2489 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2490                                 struct btrfs_root *root, struct inode *inode)
2491 {
2492         struct btrfs_inode_item *inode_item;
2493         struct btrfs_path *path;
2494         struct extent_buffer *leaf;
2495         int ret;
2496
2497         path = btrfs_alloc_path();
2498         if (!path)
2499                 return -ENOMEM;
2500
2501         path->leave_spinning = 1;
2502         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2503                                  1);
2504         if (ret) {
2505                 if (ret > 0)
2506                         ret = -ENOENT;
2507                 goto failed;
2508         }
2509
2510         btrfs_unlock_up_safe(path, 1);
2511         leaf = path->nodes[0];
2512         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2513                                     struct btrfs_inode_item);
2514
2515         fill_inode_item(trans, leaf, inode_item, inode);
2516         btrfs_mark_buffer_dirty(leaf);
2517         btrfs_set_inode_last_trans(trans, inode);
2518         ret = 0;
2519 failed:
2520         btrfs_free_path(path);
2521         return ret;
2522 }
2523
2524 /*
2525  * copy everything in the in-memory inode into the btree.
2526  */
2527 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2528                                 struct btrfs_root *root, struct inode *inode)
2529 {
2530         int ret;
2531
2532         /*
2533          * If the inode is a free space inode, we can deadlock during commit
2534          * if we put it into the delayed code.
2535          *
2536          * The data relocation inode should also be directly updated
2537          * without delay
2538          */
2539         if (!btrfs_is_free_space_inode(root, inode)
2540             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2541                 ret = btrfs_delayed_update_inode(trans, root, inode);
2542                 if (!ret)
2543                         btrfs_set_inode_last_trans(trans, inode);
2544                 return ret;
2545         }
2546
2547         return btrfs_update_inode_item(trans, root, inode);
2548 }
2549
2550 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2551                                 struct btrfs_root *root, struct inode *inode)
2552 {
2553         int ret;
2554
2555         ret = btrfs_update_inode(trans, root, inode);
2556         if (ret == -ENOSPC)
2557                 return btrfs_update_inode_item(trans, root, inode);
2558         return ret;
2559 }
2560
2561 /*
2562  * unlink helper that gets used here in inode.c and in the tree logging
2563  * recovery code.  It remove a link in a directory with a given name, and
2564  * also drops the back refs in the inode to the directory
2565  */
2566 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2567                                 struct btrfs_root *root,
2568                                 struct inode *dir, struct inode *inode,
2569                                 const char *name, int name_len)
2570 {
2571         struct btrfs_path *path;
2572         int ret = 0;
2573         struct extent_buffer *leaf;
2574         struct btrfs_dir_item *di;
2575         struct btrfs_key key;
2576         u64 index;
2577         u64 ino = btrfs_ino(inode);
2578         u64 dir_ino = btrfs_ino(dir);
2579
2580         path = btrfs_alloc_path();
2581         if (!path) {
2582                 ret = -ENOMEM;
2583                 goto out;
2584         }
2585
2586         path->leave_spinning = 1;
2587         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2588                                     name, name_len, -1);
2589         if (IS_ERR(di)) {
2590                 ret = PTR_ERR(di);
2591                 goto err;
2592         }
2593         if (!di) {
2594                 ret = -ENOENT;
2595                 goto err;
2596         }
2597         leaf = path->nodes[0];
2598         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2599         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2600         if (ret)
2601                 goto err;
2602         btrfs_release_path(path);
2603
2604         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2605                                   dir_ino, &index);
2606         if (ret) {
2607                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2608                        "inode %llu parent %llu\n", name_len, name,
2609                        (unsigned long long)ino, (unsigned long long)dir_ino);
2610                 goto err;
2611         }
2612
2613         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2614         if (ret)
2615                 goto err;
2616
2617         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2618                                          inode, dir_ino);
2619         BUG_ON(ret != 0 && ret != -ENOENT);
2620
2621         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2622                                            dir, index);
2623         if (ret == -ENOENT)
2624                 ret = 0;
2625 err:
2626         btrfs_free_path(path);
2627         if (ret)
2628                 goto out;
2629
2630         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2631         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2632         btrfs_update_inode(trans, root, dir);
2633 out:
2634         return ret;
2635 }
2636
2637 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2638                        struct btrfs_root *root,
2639                        struct inode *dir, struct inode *inode,
2640                        const char *name, int name_len)
2641 {
2642         int ret;
2643         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2644         if (!ret) {
2645                 btrfs_drop_nlink(inode);
2646                 ret = btrfs_update_inode(trans, root, inode);
2647         }
2648         return ret;
2649 }
2650
2651
2652 /* helper to check if there is any shared block in the path */
2653 static int check_path_shared(struct btrfs_root *root,
2654                              struct btrfs_path *path)
2655 {
2656         struct extent_buffer *eb;
2657         int level;
2658         u64 refs = 1;
2659
2660         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2661                 int ret;
2662
2663                 if (!path->nodes[level])
2664                         break;
2665                 eb = path->nodes[level];
2666                 if (!btrfs_block_can_be_shared(root, eb))
2667                         continue;
2668                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2669                                                &refs, NULL);
2670                 if (refs > 1)
2671                         return 1;
2672         }
2673         return 0;
2674 }
2675
2676 /*
2677  * helper to start transaction for unlink and rmdir.
2678  *
2679  * unlink and rmdir are special in btrfs, they do not always free space.
2680  * so in enospc case, we should make sure they will free space before
2681  * allowing them to use the global metadata reservation.
2682  */
2683 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2684                                                        struct dentry *dentry)
2685 {
2686         struct btrfs_trans_handle *trans;
2687         struct btrfs_root *root = BTRFS_I(dir)->root;
2688         struct btrfs_path *path;
2689         struct btrfs_inode_ref *ref;
2690         struct btrfs_dir_item *di;
2691         struct inode *inode = dentry->d_inode;
2692         u64 index;
2693         int check_link = 1;
2694         int err = -ENOSPC;
2695         int ret;
2696         u64 ino = btrfs_ino(inode);
2697         u64 dir_ino = btrfs_ino(dir);
2698
2699         /*
2700          * 1 for the possible orphan item
2701          * 1 for the dir item
2702          * 1 for the dir index
2703          * 1 for the inode ref
2704          * 1 for the inode ref in the tree log
2705          * 2 for the dir entries in the log
2706          * 1 for the inode
2707          */
2708         trans = btrfs_start_transaction(root, 8);
2709         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2710                 return trans;
2711
2712         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2713                 return ERR_PTR(-ENOSPC);
2714
2715         /* check if there is someone else holds reference */
2716         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2717                 return ERR_PTR(-ENOSPC);
2718
2719         if (atomic_read(&inode->i_count) > 2)
2720                 return ERR_PTR(-ENOSPC);
2721
2722         if (xchg(&root->fs_info->enospc_unlink, 1))
2723                 return ERR_PTR(-ENOSPC);
2724
2725         path = btrfs_alloc_path();
2726         if (!path) {
2727                 root->fs_info->enospc_unlink = 0;
2728                 return ERR_PTR(-ENOMEM);
2729         }
2730
2731         /* 1 for the orphan item */
2732         trans = btrfs_start_transaction(root, 1);
2733         if (IS_ERR(trans)) {
2734                 btrfs_free_path(path);
2735                 root->fs_info->enospc_unlink = 0;
2736                 return trans;
2737         }
2738
2739         path->skip_locking = 1;
2740         path->search_commit_root = 1;
2741
2742         ret = btrfs_lookup_inode(trans, root, path,
2743                                 &BTRFS_I(dir)->location, 0);
2744         if (ret < 0) {
2745                 err = ret;
2746                 goto out;
2747         }
2748         if (ret == 0) {
2749                 if (check_path_shared(root, path))
2750                         goto out;
2751         } else {
2752                 check_link = 0;
2753         }
2754         btrfs_release_path(path);
2755
2756         ret = btrfs_lookup_inode(trans, root, path,
2757                                 &BTRFS_I(inode)->location, 0);
2758         if (ret < 0) {
2759                 err = ret;
2760                 goto out;
2761         }
2762         if (ret == 0) {
2763                 if (check_path_shared(root, path))
2764                         goto out;
2765         } else {
2766                 check_link = 0;
2767         }
2768         btrfs_release_path(path);
2769
2770         if (ret == 0 && S_ISREG(inode->i_mode)) {
2771                 ret = btrfs_lookup_file_extent(trans, root, path,
2772                                                ino, (u64)-1, 0);
2773                 if (ret < 0) {
2774                         err = ret;
2775                         goto out;
2776                 }
2777                 BUG_ON(ret == 0);
2778                 if (check_path_shared(root, path))
2779                         goto out;
2780                 btrfs_release_path(path);
2781         }
2782
2783         if (!check_link) {
2784                 err = 0;
2785                 goto out;
2786         }
2787
2788         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2789                                 dentry->d_name.name, dentry->d_name.len, 0);
2790         if (IS_ERR(di)) {
2791                 err = PTR_ERR(di);
2792                 goto out;
2793         }
2794         if (di) {
2795                 if (check_path_shared(root, path))
2796                         goto out;
2797         } else {
2798                 err = 0;
2799                 goto out;
2800         }
2801         btrfs_release_path(path);
2802
2803         ref = btrfs_lookup_inode_ref(trans, root, path,
2804                                 dentry->d_name.name, dentry->d_name.len,
2805                                 ino, dir_ino, 0);
2806         if (IS_ERR(ref)) {
2807                 err = PTR_ERR(ref);
2808                 goto out;
2809         }
2810         BUG_ON(!ref);
2811         if (check_path_shared(root, path))
2812                 goto out;
2813         index = btrfs_inode_ref_index(path->nodes[0], ref);
2814         btrfs_release_path(path);
2815
2816         /*
2817          * This is a commit root search, if we can lookup inode item and other
2818          * relative items in the commit root, it means the transaction of
2819          * dir/file creation has been committed, and the dir index item that we
2820          * delay to insert has also been inserted into the commit root. So
2821          * we needn't worry about the delayed insertion of the dir index item
2822          * here.
2823          */
2824         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2825                                 dentry->d_name.name, dentry->d_name.len, 0);
2826         if (IS_ERR(di)) {
2827                 err = PTR_ERR(di);
2828                 goto out;
2829         }
2830         BUG_ON(ret == -ENOENT);
2831         if (check_path_shared(root, path))
2832                 goto out;
2833
2834         err = 0;
2835 out:
2836         btrfs_free_path(path);
2837         /* Migrate the orphan reservation over */
2838         if (!err)
2839                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2840                                 &root->fs_info->global_block_rsv,
2841                                 trans->bytes_reserved);
2842
2843         if (err) {
2844                 btrfs_end_transaction(trans, root);
2845                 root->fs_info->enospc_unlink = 0;
2846                 return ERR_PTR(err);
2847         }
2848
2849         trans->block_rsv = &root->fs_info->global_block_rsv;
2850         return trans;
2851 }
2852
2853 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2854                                struct btrfs_root *root)
2855 {
2856         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2857                 btrfs_block_rsv_release(root, trans->block_rsv,
2858                                         trans->bytes_reserved);
2859                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2860                 BUG_ON(!root->fs_info->enospc_unlink);
2861                 root->fs_info->enospc_unlink = 0;
2862         }
2863         btrfs_end_transaction(trans, root);
2864 }
2865
2866 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2867 {
2868         struct btrfs_root *root = BTRFS_I(dir)->root;
2869         struct btrfs_trans_handle *trans;
2870         struct inode *inode = dentry->d_inode;
2871         int ret;
2872         unsigned long nr = 0;
2873
2874         trans = __unlink_start_trans(dir, dentry);
2875         if (IS_ERR(trans))
2876                 return PTR_ERR(trans);
2877
2878         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2879
2880         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2881                                  dentry->d_name.name, dentry->d_name.len);
2882         if (ret)
2883                 goto out;
2884
2885         if (inode->i_nlink == 0) {
2886                 ret = btrfs_orphan_add(trans, inode);
2887                 if (ret)
2888                         goto out;
2889         }
2890
2891 out:
2892         nr = trans->blocks_used;
2893         __unlink_end_trans(trans, root);
2894         btrfs_btree_balance_dirty(root, nr);
2895         return ret;
2896 }
2897
2898 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2899                         struct btrfs_root *root,
2900                         struct inode *dir, u64 objectid,
2901                         const char *name, int name_len)
2902 {
2903         struct btrfs_path *path;
2904         struct extent_buffer *leaf;
2905         struct btrfs_dir_item *di;
2906         struct btrfs_key key;
2907         u64 index;
2908         int ret;
2909         u64 dir_ino = btrfs_ino(dir);
2910
2911         path = btrfs_alloc_path();
2912         if (!path)
2913                 return -ENOMEM;
2914
2915         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2916                                    name, name_len, -1);
2917         BUG_ON(IS_ERR_OR_NULL(di));
2918
2919         leaf = path->nodes[0];
2920         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2921         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2922         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2923         BUG_ON(ret);
2924         btrfs_release_path(path);
2925
2926         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2927                                  objectid, root->root_key.objectid,
2928                                  dir_ino, &index, name, name_len);
2929         if (ret < 0) {
2930                 BUG_ON(ret != -ENOENT);
2931                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2932                                                  name, name_len);
2933                 BUG_ON(IS_ERR_OR_NULL(di));
2934
2935                 leaf = path->nodes[0];
2936                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2937                 btrfs_release_path(path);
2938                 index = key.offset;
2939         }
2940         btrfs_release_path(path);
2941
2942         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2943         BUG_ON(ret);
2944
2945         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2946         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2947         ret = btrfs_update_inode(trans, root, dir);
2948         BUG_ON(ret);
2949
2950         btrfs_free_path(path);
2951         return 0;
2952 }
2953
2954 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2955 {
2956         struct inode *inode = dentry->d_inode;
2957         int err = 0;
2958         struct btrfs_root *root = BTRFS_I(dir)->root;
2959         struct btrfs_trans_handle *trans;
2960         unsigned long nr = 0;
2961
2962         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2963             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2964                 return -ENOTEMPTY;
2965
2966         trans = __unlink_start_trans(dir, dentry);
2967         if (IS_ERR(trans))
2968                 return PTR_ERR(trans);
2969
2970         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2971                 err = btrfs_unlink_subvol(trans, root, dir,
2972                                           BTRFS_I(inode)->location.objectid,
2973                                           dentry->d_name.name,
2974                                           dentry->d_name.len);
2975                 goto out;
2976         }
2977
2978         err = btrfs_orphan_add(trans, inode);
2979         if (err)
2980                 goto out;
2981
2982         /* now the directory is empty */
2983         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2984                                  dentry->d_name.name, dentry->d_name.len);
2985         if (!err)
2986                 btrfs_i_size_write(inode, 0);
2987 out:
2988         nr = trans->blocks_used;
2989         __unlink_end_trans(trans, root);
2990         btrfs_btree_balance_dirty(root, nr);
2991
2992         return err;
2993 }
2994
2995 /*
2996  * this can truncate away extent items, csum items and directory items.
2997  * It starts at a high offset and removes keys until it can't find
2998  * any higher than new_size
2999  *
3000  * csum items that cross the new i_size are truncated to the new size
3001  * as well.
3002  *
3003  * min_type is the minimum key type to truncate down to.  If set to 0, this
3004  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3005  */
3006 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3007                                struct btrfs_root *root,
3008                                struct inode *inode,
3009                                u64 new_size, u32 min_type)
3010 {
3011         struct btrfs_path *path;
3012         struct extent_buffer *leaf;
3013         struct btrfs_file_extent_item *fi;
3014         struct btrfs_key key;
3015         struct btrfs_key found_key;
3016         u64 extent_start = 0;
3017         u64 extent_num_bytes = 0;
3018         u64 extent_offset = 0;
3019         u64 item_end = 0;
3020         u64 mask = root->sectorsize - 1;
3021         u32 found_type = (u8)-1;
3022         int found_extent;
3023         int del_item;
3024         int pending_del_nr = 0;
3025         int pending_del_slot = 0;
3026         int extent_type = -1;
3027         int ret;
3028         int err = 0;
3029         u64 ino = btrfs_ino(inode);
3030
3031         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3032
3033         path = btrfs_alloc_path();
3034         if (!path)
3035                 return -ENOMEM;
3036         path->reada = -1;
3037
3038         if (root->ref_cows || root == root->fs_info->tree_root)
3039                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3040
3041         /*
3042          * This function is also used to drop the items in the log tree before
3043          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3044          * it is used to drop the loged items. So we shouldn't kill the delayed
3045          * items.
3046          */
3047         if (min_type == 0 && root == BTRFS_I(inode)->root)
3048                 btrfs_kill_delayed_inode_items(inode);
3049
3050         key.objectid = ino;
3051         key.offset = (u64)-1;
3052         key.type = (u8)-1;
3053
3054 search_again:
3055         path->leave_spinning = 1;
3056         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3057         if (ret < 0) {
3058                 err = ret;
3059                 goto out;
3060         }
3061
3062         if (ret > 0) {
3063                 /* there are no items in the tree for us to truncate, we're
3064                  * done
3065                  */
3066                 if (path->slots[0] == 0)
3067                         goto out;
3068                 path->slots[0]--;
3069         }
3070
3071         while (1) {
3072                 fi = NULL;
3073                 leaf = path->nodes[0];
3074                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3075                 found_type = btrfs_key_type(&found_key);
3076
3077                 if (found_key.objectid != ino)
3078                         break;
3079
3080                 if (found_type < min_type)
3081                         break;
3082
3083                 item_end = found_key.offset;
3084                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3085                         fi = btrfs_item_ptr(leaf, path->slots[0],
3086                                             struct btrfs_file_extent_item);
3087                         extent_type = btrfs_file_extent_type(leaf, fi);
3088                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3089                                 item_end +=
3090                                     btrfs_file_extent_num_bytes(leaf, fi);
3091                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3092                                 item_end += btrfs_file_extent_inline_len(leaf,
3093                                                                          fi);
3094                         }
3095                         item_end--;
3096                 }
3097                 if (found_type > min_type) {
3098                         del_item = 1;
3099                 } else {
3100                         if (item_end < new_size)
3101                                 break;
3102                         if (found_key.offset >= new_size)
3103                                 del_item = 1;
3104                         else
3105                                 del_item = 0;
3106                 }
3107                 found_extent = 0;
3108                 /* FIXME, shrink the extent if the ref count is only 1 */
3109                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3110                         goto delete;
3111
3112                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3113                         u64 num_dec;
3114                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3115                         if (!del_item) {
3116                                 u64 orig_num_bytes =
3117                                         btrfs_file_extent_num_bytes(leaf, fi);
3118                                 extent_num_bytes = new_size -
3119                                         found_key.offset + root->sectorsize - 1;
3120                                 extent_num_bytes = extent_num_bytes &
3121                                         ~((u64)root->sectorsize - 1);
3122                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3123                                                          extent_num_bytes);
3124                                 num_dec = (orig_num_bytes -
3125                                            extent_num_bytes);
3126                                 if (root->ref_cows && extent_start != 0)
3127                                         inode_sub_bytes(inode, num_dec);
3128                                 btrfs_mark_buffer_dirty(leaf);
3129                         } else {
3130                                 extent_num_bytes =
3131                                         btrfs_file_extent_disk_num_bytes(leaf,
3132                                                                          fi);
3133                                 extent_offset = found_key.offset -
3134                                         btrfs_file_extent_offset(leaf, fi);
3135
3136                                 /* FIXME blocksize != 4096 */
3137                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3138                                 if (extent_start != 0) {
3139                                         found_extent = 1;
3140                                         if (root->ref_cows)
3141                                                 inode_sub_bytes(inode, num_dec);
3142                                 }
3143                         }
3144                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3145                         /*
3146                          * we can't truncate inline items that have had
3147                          * special encodings
3148                          */
3149                         if (!del_item &&
3150                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3151                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3152                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3153                                 u32 size = new_size - found_key.offset;
3154
3155                                 if (root->ref_cows) {
3156                                         inode_sub_bytes(inode, item_end + 1 -
3157                                                         new_size);
3158                                 }
3159                                 size =
3160                                     btrfs_file_extent_calc_inline_size(size);
3161                                 ret = btrfs_truncate_item(trans, root, path,
3162                                                           size, 1);
3163                         } else if (root->ref_cows) {
3164                                 inode_sub_bytes(inode, item_end + 1 -
3165                                                 found_key.offset);
3166                         }
3167                 }
3168 delete:
3169                 if (del_item) {
3170                         if (!pending_del_nr) {
3171                                 /* no pending yet, add ourselves */
3172                                 pending_del_slot = path->slots[0];
3173                                 pending_del_nr = 1;
3174                         } else if (pending_del_nr &&
3175                                    path->slots[0] + 1 == pending_del_slot) {
3176                                 /* hop on the pending chunk */
3177                                 pending_del_nr++;
3178                                 pending_del_slot = path->slots[0];
3179                         } else {
3180                                 BUG();
3181                         }
3182                 } else {
3183                         break;
3184                 }
3185                 if (found_extent && (root->ref_cows ||
3186                                      root == root->fs_info->tree_root)) {
3187                         btrfs_set_path_blocking(path);
3188                         ret = btrfs_free_extent(trans, root, extent_start,
3189                                                 extent_num_bytes, 0,
3190                                                 btrfs_header_owner(leaf),
3191                                                 ino, extent_offset, 0);
3192                         BUG_ON(ret);
3193                 }
3194
3195                 if (found_type == BTRFS_INODE_ITEM_KEY)
3196                         break;
3197
3198                 if (path->slots[0] == 0 ||
3199                     path->slots[0] != pending_del_slot) {
3200                         if (root->ref_cows &&
3201                             BTRFS_I(inode)->location.objectid !=
3202                                                 BTRFS_FREE_INO_OBJECTID) {
3203                                 err = -EAGAIN;
3204                                 goto out;
3205                         }
3206                         if (pending_del_nr) {
3207                                 ret = btrfs_del_items(trans, root, path,
3208                                                 pending_del_slot,
3209                                                 pending_del_nr);
3210                                 BUG_ON(ret);
3211                                 pending_del_nr = 0;
3212                         }
3213                         btrfs_release_path(path);
3214                         goto search_again;
3215                 } else {
3216                         path->slots[0]--;
3217                 }
3218         }
3219 out:
3220         if (pending_del_nr) {
3221                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3222                                       pending_del_nr);
3223                 BUG_ON(ret);
3224         }
3225         btrfs_free_path(path);
3226         return err;
3227 }
3228
3229 /*
3230  * taken from block_truncate_page, but does cow as it zeros out
3231  * any bytes left in the last page in the file.
3232  */
3233 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3234 {
3235         struct inode *inode = mapping->host;
3236         struct btrfs_root *root = BTRFS_I(inode)->root;
3237         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3238         struct btrfs_ordered_extent *ordered;
3239         struct extent_state *cached_state = NULL;
3240         char *kaddr;
3241         u32 blocksize = root->sectorsize;
3242         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3243         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3244         struct page *page;
3245         gfp_t mask = btrfs_alloc_write_mask(mapping);
3246         int ret = 0;
3247         u64 page_start;
3248         u64 page_end;
3249
3250         if ((offset & (blocksize - 1)) == 0)
3251                 goto out;
3252         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3253         if (ret)
3254                 goto out;
3255
3256         ret = -ENOMEM;
3257 again:
3258         page = find_or_create_page(mapping, index, mask);
3259         if (!page) {
3260                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3261                 goto out;
3262         }
3263
3264         page_start = page_offset(page);
3265         page_end = page_start + PAGE_CACHE_SIZE - 1;
3266
3267         if (!PageUptodate(page)) {
3268                 ret = btrfs_readpage(NULL, page);
3269                 lock_page(page);
3270                 if (page->mapping != mapping) {
3271                         unlock_page(page);
3272                         page_cache_release(page);
3273                         goto again;
3274                 }
3275                 if (!PageUptodate(page)) {
3276                         ret = -EIO;
3277                         goto out_unlock;
3278                 }
3279         }
3280         wait_on_page_writeback(page);
3281
3282         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3283                          GFP_NOFS);
3284         set_page_extent_mapped(page);
3285
3286         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3287         if (ordered) {
3288                 unlock_extent_cached(io_tree, page_start, page_end,
3289                                      &cached_state, GFP_NOFS);
3290                 unlock_page(page);
3291                 page_cache_release(page);
3292                 btrfs_start_ordered_extent(inode, ordered, 1);
3293                 btrfs_put_ordered_extent(ordered);
3294                 goto again;
3295         }
3296
3297         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3298                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3299                           0, 0, &cached_state, GFP_NOFS);
3300
3301         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3302                                         &cached_state);
3303         if (ret) {
3304                 unlock_extent_cached(io_tree, page_start, page_end,
3305                                      &cached_state, GFP_NOFS);
3306                 goto out_unlock;
3307         }
3308
3309         ret = 0;
3310         if (offset != PAGE_CACHE_SIZE) {
3311                 kaddr = kmap(page);
3312                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3313                 flush_dcache_page(page);
3314                 kunmap(page);
3315         }
3316         ClearPageChecked(page);
3317         set_page_dirty(page);
3318         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3319                              GFP_NOFS);
3320
3321 out_unlock:
3322         if (ret)
3323                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3324         unlock_page(page);
3325         page_cache_release(page);
3326 out:
3327         return ret;
3328 }
3329
3330 /*
3331  * This function puts in dummy file extents for the area we're creating a hole
3332  * for.  So if we are truncating this file to a larger size we need to insert
3333  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3334  * the range between oldsize and size
3335  */
3336 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3337 {
3338         struct btrfs_trans_handle *trans;
3339         struct btrfs_root *root = BTRFS_I(inode)->root;
3340         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3341         struct extent_map *em = NULL;
3342         struct extent_state *cached_state = NULL;
3343         u64 mask = root->sectorsize - 1;
3344         u64 hole_start = (oldsize + mask) & ~mask;
3345         u64 block_end = (size + mask) & ~mask;
3346         u64 last_byte;
3347         u64 cur_offset;
3348         u64 hole_size;
3349         int err = 0;
3350
3351         if (size <= hole_start)
3352                 return 0;
3353
3354         while (1) {
3355                 struct btrfs_ordered_extent *ordered;
3356                 btrfs_wait_ordered_range(inode, hole_start,
3357                                          block_end - hole_start);
3358                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3359                                  &cached_state, GFP_NOFS);
3360                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3361                 if (!ordered)
3362                         break;
3363                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3364                                      &cached_state, GFP_NOFS);
3365                 btrfs_put_ordered_extent(ordered);
3366         }
3367
3368         cur_offset = hole_start;
3369         while (1) {
3370                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3371                                 block_end - cur_offset, 0);
3372                 BUG_ON(IS_ERR_OR_NULL(em));
3373                 last_byte = min(extent_map_end(em), block_end);
3374                 last_byte = (last_byte + mask) & ~mask;
3375                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3376                         u64 hint_byte = 0;
3377                         hole_size = last_byte - cur_offset;
3378
3379                         trans = btrfs_start_transaction(root, 3);
3380                         if (IS_ERR(trans)) {
3381                                 err = PTR_ERR(trans);
3382                                 break;
3383                         }
3384
3385                         err = btrfs_drop_extents(trans, inode, cur_offset,
3386                                                  cur_offset + hole_size,
3387                                                  &hint_byte, 1);
3388                         if (err) {
3389                                 btrfs_update_inode(trans, root, inode);
3390                                 btrfs_end_transaction(trans, root);
3391                                 break;
3392                         }
3393
3394                         err = btrfs_insert_file_extent(trans, root,
3395                                         btrfs_ino(inode), cur_offset, 0,
3396                                         0, hole_size, 0, hole_size,
3397                                         0, 0, 0);
3398                         if (err) {
3399                                 btrfs_update_inode(trans, root, inode);
3400                                 btrfs_end_transaction(trans, root);
3401                                 break;
3402                         }
3403
3404                         btrfs_drop_extent_cache(inode, hole_start,
3405                                         last_byte - 1, 0);
3406
3407                         btrfs_update_inode(trans, root, inode);
3408                         btrfs_end_transaction(trans, root);
3409                 }
3410                 free_extent_map(em);
3411                 em = NULL;
3412                 cur_offset = last_byte;
3413                 if (cur_offset >= block_end)
3414                         break;
3415         }
3416
3417         free_extent_map(em);
3418         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3419                              GFP_NOFS);
3420         return err;
3421 }
3422
3423 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3424 {
3425         struct btrfs_root *root = BTRFS_I(inode)->root;
3426         struct btrfs_trans_handle *trans;
3427         loff_t oldsize = i_size_read(inode);
3428         int ret;
3429
3430         if (newsize == oldsize)
3431                 return 0;
3432
3433         if (newsize > oldsize) {
3434                 truncate_pagecache(inode, oldsize, newsize);
3435                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3436                 if (ret)
3437                         return ret;
3438
3439                 trans = btrfs_start_transaction(root, 1);
3440                 if (IS_ERR(trans))
3441                         return PTR_ERR(trans);
3442
3443                 i_size_write(inode, newsize);
3444                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3445                 ret = btrfs_update_inode(trans, root, inode);
3446                 btrfs_end_transaction(trans, root);
3447         } else {
3448
3449                 /*
3450                  * We're truncating a file that used to have good data down to
3451                  * zero. Make sure it gets into the ordered flush list so that
3452                  * any new writes get down to disk quickly.
3453                  */
3454                 if (newsize == 0)
3455                         BTRFS_I(inode)->ordered_data_close = 1;
3456
3457                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3458                 truncate_setsize(inode, newsize);
3459                 ret = btrfs_truncate(inode);
3460         }
3461
3462         return ret;
3463 }
3464
3465 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3466 {
3467         struct inode *inode = dentry->d_inode;
3468         struct btrfs_root *root = BTRFS_I(inode)->root;
3469         int err;
3470
3471         if (btrfs_root_readonly(root))
3472                 return -EROFS;
3473
3474         err = inode_change_ok(inode, attr);
3475         if (err)
3476                 return err;
3477
3478         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3479                 err = btrfs_setsize(inode, attr->ia_size);
3480                 if (err)
3481                         return err;
3482         }
3483
3484         if (attr->ia_valid) {
3485                 setattr_copy(inode, attr);
3486                 err = btrfs_dirty_inode(inode);
3487
3488                 if (!err && attr->ia_valid & ATTR_MODE)
3489                         err = btrfs_acl_chmod(inode);
3490         }
3491
3492         return err;
3493 }
3494
3495 void btrfs_evict_inode(struct inode *inode)
3496 {
3497         struct btrfs_trans_handle *trans;
3498         struct btrfs_root *root = BTRFS_I(inode)->root;
3499         struct btrfs_block_rsv *rsv, *global_rsv;
3500         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3501         unsigned long nr;
3502         int ret;
3503
3504         trace_btrfs_inode_evict(inode);
3505
3506         truncate_inode_pages(&inode->i_data, 0);
3507         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3508                                btrfs_is_free_space_inode(root, inode)))
3509                 goto no_delete;
3510
3511         if (is_bad_inode(inode)) {
3512                 btrfs_orphan_del(NULL, inode);
3513                 goto no_delete;
3514         }
3515         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3516         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3517
3518         if (root->fs_info->log_root_recovering) {
3519                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3520                 goto no_delete;
3521         }
3522
3523         if (inode->i_nlink > 0) {
3524                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3525                 goto no_delete;
3526         }
3527
3528         rsv = btrfs_alloc_block_rsv(root);
3529         if (!rsv) {
3530                 btrfs_orphan_del(NULL, inode);
3531                 goto no_delete;
3532         }
3533         rsv->size = min_size;
3534         global_rsv = &root->fs_info->global_block_rsv;
3535
3536         btrfs_i_size_write(inode, 0);
3537
3538         /*
3539          * This is a bit simpler than btrfs_truncate since
3540          *
3541          * 1) We've already reserved our space for our orphan item in the
3542          *    unlink.
3543          * 2) We're going to delete the inode item, so we don't need to update
3544          *    it at all.
3545          *
3546          * So we just need to reserve some slack space in case we add bytes when
3547          * doing the truncate.
3548          */
3549         while (1) {
3550                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3551
3552                 /*
3553                  * Try and steal from the global reserve since we will
3554                  * likely not use this space anyway, we want to try as
3555                  * hard as possible to get this to work.
3556                  */
3557                 if (ret)
3558                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3559
3560                 if (ret) {
3561                         printk(KERN_WARNING "Could not get space for a "
3562                                "delete, will truncate on mount %d\n", ret);
3563                         btrfs_orphan_del(NULL, inode);
3564                         btrfs_free_block_rsv(root, rsv);
3565                         goto no_delete;
3566                 }
3567
3568                 trans = btrfs_start_transaction(root, 0);
3569                 if (IS_ERR(trans)) {
3570                         btrfs_orphan_del(NULL, inode);
3571                         btrfs_free_block_rsv(root, rsv);
3572                         goto no_delete;
3573                 }
3574
3575                 trans->block_rsv = rsv;
3576
3577                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3578                 if (ret != -EAGAIN)
3579                         break;
3580
3581                 nr = trans->blocks_used;
3582                 btrfs_end_transaction(trans, root);
3583                 trans = NULL;
3584                 btrfs_btree_balance_dirty(root, nr);
3585         }
3586
3587         btrfs_free_block_rsv(root, rsv);
3588
3589         if (ret == 0) {
3590                 trans->block_rsv = root->orphan_block_rsv;
3591                 ret = btrfs_orphan_del(trans, inode);
3592                 BUG_ON(ret);
3593         }
3594
3595         trans->block_rsv = &root->fs_info->trans_block_rsv;
3596         if (!(root == root->fs_info->tree_root ||
3597               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3598                 btrfs_return_ino(root, btrfs_ino(inode));
3599
3600         nr = trans->blocks_used;
3601         btrfs_end_transaction(trans, root);
3602         btrfs_btree_balance_dirty(root, nr);
3603 no_delete:
3604         end_writeback(inode);
3605         return;
3606 }
3607
3608 /*
3609  * this returns the key found in the dir entry in the location pointer.
3610  * If no dir entries were found, location->objectid is 0.
3611  */
3612 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3613                                struct btrfs_key *location)
3614 {
3615         const char *name = dentry->d_name.name;
3616         int namelen = dentry->d_name.len;
3617         struct btrfs_dir_item *di;
3618         struct btrfs_path *path;
3619         struct btrfs_root *root = BTRFS_I(dir)->root;
3620         int ret = 0;
3621
3622         path = btrfs_alloc_path();
3623         if (!path)
3624                 return -ENOMEM;
3625
3626         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3627                                     namelen, 0);
3628         if (IS_ERR(di))
3629                 ret = PTR_ERR(di);
3630
3631         if (IS_ERR_OR_NULL(di))
3632                 goto out_err;
3633
3634         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3635 out:
3636         btrfs_free_path(path);
3637         return ret;
3638 out_err:
3639         location->objectid = 0;
3640         goto out;
3641 }
3642
3643 /*
3644  * when we hit a tree root in a directory, the btrfs part of the inode
3645  * needs to be changed to reflect the root directory of the tree root.  This
3646  * is kind of like crossing a mount point.
3647  */
3648 static int fixup_tree_root_location(struct btrfs_root *root,
3649                                     struct inode *dir,
3650                                     struct dentry *dentry,
3651                                     struct btrfs_key *location,
3652                                     struct btrfs_root **sub_root)
3653 {
3654         struct btrfs_path *path;
3655         struct btrfs_root *new_root;
3656         struct btrfs_root_ref *ref;
3657         struct extent_buffer *leaf;
3658         int ret;
3659         int err = 0;
3660
3661         path = btrfs_alloc_path();
3662         if (!path) {
3663                 err = -ENOMEM;
3664                 goto out;
3665         }
3666
3667         err = -ENOENT;
3668         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3669                                   BTRFS_I(dir)->root->root_key.objectid,
3670                                   location->objectid);
3671         if (ret) {
3672                 if (ret < 0)
3673                         err = ret;
3674                 goto out;
3675         }
3676
3677         leaf = path->nodes[0];
3678         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3679         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3680             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3681                 goto out;
3682
3683         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3684                                    (unsigned long)(ref + 1),
3685                                    dentry->d_name.len);
3686         if (ret)
3687                 goto out;
3688
3689         btrfs_release_path(path);
3690
3691         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3692         if (IS_ERR(new_root)) {
3693                 err = PTR_ERR(new_root);
3694                 goto out;
3695         }
3696
3697         if (btrfs_root_refs(&new_root->root_item) == 0) {
3698                 err = -ENOENT;
3699                 goto out;
3700         }
3701
3702         *sub_root = new_root;
3703         location->objectid = btrfs_root_dirid(&new_root->root_item);
3704         location->type = BTRFS_INODE_ITEM_KEY;
3705         location->offset = 0;
3706         err = 0;
3707 out:
3708         btrfs_free_path(path);
3709         return err;
3710 }
3711
3712 static void inode_tree_add(struct inode *inode)
3713 {
3714         struct btrfs_root *root = BTRFS_I(inode)->root;
3715         struct btrfs_inode *entry;
3716         struct rb_node **p;
3717         struct rb_node *parent;
3718         u64 ino = btrfs_ino(inode);
3719 again:
3720         p = &root->inode_tree.rb_node;
3721         parent = NULL;
3722
3723         if (inode_unhashed(inode))
3724                 return;
3725
3726         spin_lock(&root->inode_lock);
3727         while (*p) {
3728                 parent = *p;
3729                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3730
3731                 if (ino < btrfs_ino(&entry->vfs_inode))
3732                         p = &parent->rb_left;
3733                 else if (ino > btrfs_ino(&entry->vfs_inode))
3734                         p = &parent->rb_right;
3735                 else {
3736                         WARN_ON(!(entry->vfs_inode.i_state &
3737                                   (I_WILL_FREE | I_FREEING)));
3738                         rb_erase(parent, &root->inode_tree);
3739                         RB_CLEAR_NODE(parent);
3740                         spin_unlock(&root->inode_lock);
3741                         goto again;
3742                 }
3743         }
3744         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3745         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3746         spin_unlock(&root->inode_lock);
3747 }
3748
3749 static void inode_tree_del(struct inode *inode)
3750 {
3751         struct btrfs_root *root = BTRFS_I(inode)->root;
3752         int empty = 0;
3753
3754         spin_lock(&root->inode_lock);
3755         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3756                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3757                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3758                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3759         }
3760         spin_unlock(&root->inode_lock);
3761
3762         /*
3763          * Free space cache has inodes in the tree root, but the tree root has a
3764          * root_refs of 0, so this could end up dropping the tree root as a
3765          * snapshot, so we need the extra !root->fs_info->tree_root check to
3766          * make sure we don't drop it.
3767          */
3768         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3769             root != root->fs_info->tree_root) {
3770                 synchronize_srcu(&root->fs_info->subvol_srcu);
3771                 spin_lock(&root->inode_lock);
3772                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3773                 spin_unlock(&root->inode_lock);
3774                 if (empty)
3775                         btrfs_add_dead_root(root);
3776         }
3777 }
3778
3779 int btrfs_invalidate_inodes(struct btrfs_root *root)
3780 {
3781         struct rb_node *node;
3782         struct rb_node *prev;
3783         struct btrfs_inode *entry;
3784         struct inode *inode;
3785         u64 objectid = 0;
3786
3787         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3788
3789         spin_lock(&root->inode_lock);
3790 again:
3791         node = root->inode_tree.rb_node;
3792         prev = NULL;
3793         while (node) {
3794                 prev = node;
3795                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3796
3797                 if (objectid < btrfs_ino(&entry->vfs_inode))
3798                         node = node->rb_left;
3799                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3800                         node = node->rb_right;
3801                 else
3802                         break;
3803         }
3804         if (!node) {
3805                 while (prev) {
3806                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3807                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3808                                 node = prev;
3809                                 break;
3810                         }
3811                         prev = rb_next(prev);
3812                 }
3813         }
3814         while (node) {
3815                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3816                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3817                 inode = igrab(&entry->vfs_inode);
3818                 if (inode) {
3819                         spin_unlock(&root->inode_lock);
3820                         if (atomic_read(&inode->i_count) > 1)
3821                                 d_prune_aliases(inode);
3822                         /*
3823                          * btrfs_drop_inode will have it removed from
3824                          * the inode cache when its usage count
3825                          * hits zero.
3826                          */
3827                         iput(inode);
3828                         cond_resched();
3829                         spin_lock(&root->inode_lock);
3830                         goto again;
3831                 }
3832
3833                 if (cond_resched_lock(&root->inode_lock))
3834                         goto again;
3835
3836                 node = rb_next(node);
3837         }
3838         spin_unlock(&root->inode_lock);
3839         return 0;
3840 }
3841
3842 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3843 {
3844         struct btrfs_iget_args *args = p;
3845         inode->i_ino = args->ino;
3846         BTRFS_I(inode)->root = args->root;
3847         btrfs_set_inode_space_info(args->root, inode);
3848         return 0;
3849 }
3850
3851 static int btrfs_find_actor(struct inode *inode, void *opaque)
3852 {
3853         struct btrfs_iget_args *args = opaque;
3854         return args->ino == btrfs_ino(inode) &&
3855                 args->root == BTRFS_I(inode)->root;
3856 }
3857
3858 static struct inode *btrfs_iget_locked(struct super_block *s,
3859                                        u64 objectid,
3860                                        struct btrfs_root *root)
3861 {
3862         struct inode *inode;
3863         struct btrfs_iget_args args;
3864         args.ino = objectid;
3865         args.root = root;
3866
3867         inode = iget5_locked(s, objectid, btrfs_find_actor,
3868                              btrfs_init_locked_inode,
3869                              (void *)&args);
3870         return inode;
3871 }
3872
3873 /* Get an inode object given its location and corresponding root.
3874  * Returns in *is_new if the inode was read from disk
3875  */
3876 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3877                          struct btrfs_root *root, int *new)
3878 {
3879         struct inode *inode;
3880
3881         inode = btrfs_iget_locked(s, location->objectid, root);
3882         if (!inode)
3883                 return ERR_PTR(-ENOMEM);
3884
3885         if (inode->i_state & I_NEW) {
3886                 BTRFS_I(inode)->root = root;
3887                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3888                 btrfs_read_locked_inode(inode);
3889                 if (!is_bad_inode(inode)) {
3890                         inode_tree_add(inode);
3891                         unlock_new_inode(inode);
3892                         if (new)
3893                                 *new = 1;
3894                 } else {
3895                         unlock_new_inode(inode);
3896                         iput(inode);
3897                         inode = ERR_PTR(-ESTALE);
3898                 }
3899         }
3900
3901         return inode;
3902 }
3903
3904 static struct inode *new_simple_dir(struct super_block *s,
3905                                     struct btrfs_key *key,
3906                                     struct btrfs_root *root)
3907 {
3908         struct inode *inode = new_inode(s);
3909
3910         if (!inode)
3911                 return ERR_PTR(-ENOMEM);
3912
3913         BTRFS_I(inode)->root = root;
3914         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3915         BTRFS_I(inode)->dummy_inode = 1;
3916
3917         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3918         inode->i_op = &simple_dir_inode_operations;
3919         inode->i_fop = &simple_dir_operations;
3920         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3921         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3922
3923         return inode;
3924 }
3925
3926 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3927 {
3928         struct inode *inode;
3929         struct btrfs_root *root = BTRFS_I(dir)->root;
3930         struct btrfs_root *sub_root = root;
3931         struct btrfs_key location;
3932         int index;
3933         int ret = 0;
3934
3935         if (dentry->d_name.len > BTRFS_NAME_LEN)
3936                 return ERR_PTR(-ENAMETOOLONG);
3937
3938         if (unlikely(d_need_lookup(dentry))) {
3939                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3940                 kfree(dentry->d_fsdata);
3941                 dentry->d_fsdata = NULL;
3942                 /* This thing is hashed, drop it for now */
3943                 d_drop(dentry);
3944         } else {
3945                 ret = btrfs_inode_by_name(dir, dentry, &location);
3946         }
3947
3948         if (ret < 0)
3949                 return ERR_PTR(ret);
3950
3951         if (location.objectid == 0)
3952                 return NULL;
3953
3954         if (location.type == BTRFS_INODE_ITEM_KEY) {
3955                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3956                 return inode;
3957         }
3958
3959         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3960
3961         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3962         ret = fixup_tree_root_location(root, dir, dentry,
3963                                        &location, &sub_root);
3964         if (ret < 0) {
3965                 if (ret != -ENOENT)
3966                         inode = ERR_PTR(ret);
3967                 else
3968                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3969         } else {
3970                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3971         }
3972         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3973
3974         if (!IS_ERR(inode) && root != sub_root) {
3975                 down_read(&root->fs_info->cleanup_work_sem);
3976                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3977                         ret = btrfs_orphan_cleanup(sub_root);
3978                 up_read(&root->fs_info->cleanup_work_sem);
3979                 if (ret)
3980                         inode = ERR_PTR(ret);
3981         }
3982
3983         return inode;
3984 }
3985
3986 static int btrfs_dentry_delete(const struct dentry *dentry)
3987 {
3988         struct btrfs_root *root;
3989
3990         if (!dentry->d_inode && !IS_ROOT(dentry))
3991                 dentry = dentry->d_parent;
3992
3993         if (dentry->d_inode) {
3994                 root = BTRFS_I(dentry->d_inode)->root;
3995                 if (btrfs_root_refs(&root->root_item) == 0)
3996                         return 1;
3997         }
3998         return 0;
3999 }
4000
4001 static void btrfs_dentry_release(struct dentry *dentry)
4002 {
4003         if (dentry->d_fsdata)
4004                 kfree(dentry->d_fsdata);
4005 }
4006
4007 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4008                                    struct nameidata *nd)
4009 {
4010         struct dentry *ret;
4011
4012         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4013         if (unlikely(d_need_lookup(dentry))) {
4014                 spin_lock(&dentry->d_lock);
4015                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4016                 spin_unlock(&dentry->d_lock);
4017         }
4018         return ret;
4019 }
4020
4021 unsigned char btrfs_filetype_table[] = {
4022         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4023 };
4024
4025 static int btrfs_real_readdir(struct file *filp, void *dirent,
4026                               filldir_t filldir)
4027 {
4028         struct inode *inode = filp->f_dentry->d_inode;
4029         struct btrfs_root *root = BTRFS_I(inode)->root;
4030         struct btrfs_item *item;
4031         struct btrfs_dir_item *di;
4032         struct btrfs_key key;
4033         struct btrfs_key found_key;
4034         struct btrfs_path *path;
4035         struct list_head ins_list;
4036         struct list_head del_list;
4037         struct qstr q;
4038         int ret;
4039         struct extent_buffer *leaf;
4040         int slot;
4041         unsigned char d_type;
4042         int over = 0;
4043         u32 di_cur;
4044         u32 di_total;
4045         u32 di_len;
4046         int key_type = BTRFS_DIR_INDEX_KEY;
4047         char tmp_name[32];
4048         char *name_ptr;
4049         int name_len;
4050         int is_curr = 0;        /* filp->f_pos points to the current index? */
4051
4052         /* FIXME, use a real flag for deciding about the key type */
4053         if (root->fs_info->tree_root == root)
4054                 key_type = BTRFS_DIR_ITEM_KEY;
4055
4056         /* special case for "." */
4057         if (filp->f_pos == 0) {
4058                 over = filldir(dirent, ".", 1,
4059                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4060                 if (over)
4061                         return 0;
4062                 filp->f_pos = 1;
4063         }
4064         /* special case for .., just use the back ref */
4065         if (filp->f_pos == 1) {
4066                 u64 pino = parent_ino(filp->f_path.dentry);
4067                 over = filldir(dirent, "..", 2,
4068                                filp->f_pos, pino, DT_DIR);
4069                 if (over)
4070                         return 0;
4071                 filp->f_pos = 2;
4072         }
4073         path = btrfs_alloc_path();
4074         if (!path)
4075                 return -ENOMEM;
4076
4077         path->reada = 1;
4078
4079         if (key_type == BTRFS_DIR_INDEX_KEY) {
4080                 INIT_LIST_HEAD(&ins_list);
4081                 INIT_LIST_HEAD(&del_list);
4082                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4083         }
4084
4085         btrfs_set_key_type(&key, key_type);
4086         key.offset = filp->f_pos;
4087         key.objectid = btrfs_ino(inode);
4088
4089         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4090         if (ret < 0)
4091                 goto err;
4092
4093         while (1) {
4094                 leaf = path->nodes[0];
4095                 slot = path->slots[0];
4096                 if (slot >= btrfs_header_nritems(leaf)) {
4097                         ret = btrfs_next_leaf(root, path);
4098                         if (ret < 0)
4099                                 goto err;
4100                         else if (ret > 0)
4101                                 break;
4102                         continue;
4103                 }
4104
4105                 item = btrfs_item_nr(leaf, slot);
4106                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4107
4108                 if (found_key.objectid != key.objectid)
4109                         break;
4110                 if (btrfs_key_type(&found_key) != key_type)
4111                         break;
4112                 if (found_key.offset < filp->f_pos)
4113                         goto next;
4114                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4115                     btrfs_should_delete_dir_index(&del_list,
4116                                                   found_key.offset))
4117                         goto next;
4118
4119                 filp->f_pos = found_key.offset;
4120                 is_curr = 1;
4121
4122                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4123                 di_cur = 0;
4124                 di_total = btrfs_item_size(leaf, item);
4125
4126                 while (di_cur < di_total) {
4127                         struct btrfs_key location;
4128                         struct dentry *tmp;
4129
4130                         if (verify_dir_item(root, leaf, di))
4131                                 break;
4132
4133                         name_len = btrfs_dir_name_len(leaf, di);
4134                         if (name_len <= sizeof(tmp_name)) {
4135                                 name_ptr = tmp_name;
4136                         } else {
4137                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4138                                 if (!name_ptr) {
4139                                         ret = -ENOMEM;
4140                                         goto err;
4141                                 }
4142                         }
4143                         read_extent_buffer(leaf, name_ptr,
4144                                            (unsigned long)(di + 1), name_len);
4145
4146                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4147                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4148
4149                         q.name = name_ptr;
4150                         q.len = name_len;
4151                         q.hash = full_name_hash(q.name, q.len);
4152                         tmp = d_lookup(filp->f_dentry, &q);
4153                         if (!tmp) {
4154                                 struct btrfs_key *newkey;
4155
4156                                 newkey = kzalloc(sizeof(struct btrfs_key),
4157                                                  GFP_NOFS);
4158                                 if (!newkey)
4159                                         goto no_dentry;
4160                                 tmp = d_alloc(filp->f_dentry, &q);
4161                                 if (!tmp) {
4162                                         kfree(newkey);
4163                                         dput(tmp);
4164                                         goto no_dentry;
4165                                 }
4166                                 memcpy(newkey, &location,
4167                                        sizeof(struct btrfs_key));
4168                                 tmp->d_fsdata = newkey;
4169                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4170                                 d_rehash(tmp);
4171                                 dput(tmp);
4172                         } else {
4173                                 dput(tmp);
4174                         }
4175 no_dentry:
4176                         /* is this a reference to our own snapshot? If so
4177                          * skip it
4178                          */
4179                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4180                             location.objectid == root->root_key.objectid) {
4181                                 over = 0;
4182                                 goto skip;
4183                         }
4184                         over = filldir(dirent, name_ptr, name_len,
4185                                        found_key.offset, location.objectid,
4186                                        d_type);
4187
4188 skip:
4189                         if (name_ptr != tmp_name)
4190                                 kfree(name_ptr);
4191
4192                         if (over)
4193                                 goto nopos;
4194                         di_len = btrfs_dir_name_len(leaf, di) +
4195                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4196                         di_cur += di_len;
4197                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4198                 }
4199 next:
4200                 path->slots[0]++;
4201         }
4202
4203         if (key_type == BTRFS_DIR_INDEX_KEY) {
4204                 if (is_curr)
4205                         filp->f_pos++;
4206                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4207                                                       &ins_list);
4208                 if (ret)
4209                         goto nopos;
4210         }
4211
4212         /* Reached end of directory/root. Bump pos past the last item. */
4213         if (key_type == BTRFS_DIR_INDEX_KEY)
4214                 /*
4215                  * 32-bit glibc will use getdents64, but then strtol -
4216                  * so the last number we can serve is this.
4217                  */
4218                 filp->f_pos = 0x7fffffff;
4219         else
4220                 filp->f_pos++;
4221 nopos:
4222         ret = 0;
4223 err:
4224         if (key_type == BTRFS_DIR_INDEX_KEY)
4225                 btrfs_put_delayed_items(&ins_list, &del_list);
4226         btrfs_free_path(path);
4227         return ret;
4228 }
4229
4230 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4231 {
4232         struct btrfs_root *root = BTRFS_I(inode)->root;
4233         struct btrfs_trans_handle *trans;
4234         int ret = 0;
4235         bool nolock = false;
4236
4237         if (BTRFS_I(inode)->dummy_inode)
4238                 return 0;
4239
4240         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4241                 nolock = true;
4242
4243         if (wbc->sync_mode == WB_SYNC_ALL) {
4244                 if (nolock)
4245                         trans = btrfs_join_transaction_nolock(root);
4246                 else
4247                         trans = btrfs_join_transaction(root);
4248                 if (IS_ERR(trans))
4249                         return PTR_ERR(trans);
4250                 if (nolock)
4251                         ret = btrfs_end_transaction_nolock(trans, root);
4252                 else
4253                         ret = btrfs_commit_transaction(trans, root);
4254         }
4255         return ret;
4256 }
4257
4258 /*
4259  * This is somewhat expensive, updating the tree every time the
4260  * inode changes.  But, it is most likely to find the inode in cache.
4261  * FIXME, needs more benchmarking...there are no reasons other than performance
4262  * to keep or drop this code.
4263  */
4264 int btrfs_dirty_inode(struct inode *inode)
4265 {
4266         struct btrfs_root *root = BTRFS_I(inode)->root;
4267         struct btrfs_trans_handle *trans;
4268         int ret;
4269
4270         if (BTRFS_I(inode)->dummy_inode)
4271                 return 0;
4272
4273         trans = btrfs_join_transaction(root);
4274         if (IS_ERR(trans))
4275                 return PTR_ERR(trans);
4276
4277         ret = btrfs_update_inode(trans, root, inode);
4278         if (ret && ret == -ENOSPC) {
4279                 /* whoops, lets try again with the full transaction */
4280                 btrfs_end_transaction(trans, root);
4281                 trans = btrfs_start_transaction(root, 1);
4282                 if (IS_ERR(trans))
4283                         return PTR_ERR(trans);
4284
4285                 ret = btrfs_update_inode(trans, root, inode);
4286         }
4287         btrfs_end_transaction(trans, root);
4288         if (BTRFS_I(inode)->delayed_node)
4289                 btrfs_balance_delayed_items(root);
4290
4291         return ret;
4292 }
4293
4294 /*
4295  * This is a copy of file_update_time.  We need this so we can return error on
4296  * ENOSPC for updating the inode in the case of file write and mmap writes.
4297  */
4298 int btrfs_update_time(struct file *file)
4299 {
4300         struct inode *inode = file->f_path.dentry->d_inode;
4301         struct timespec now;
4302         int ret;
4303         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4304
4305         /* First try to exhaust all avenues to not sync */
4306         if (IS_NOCMTIME(inode))
4307                 return 0;
4308
4309         now = current_fs_time(inode->i_sb);
4310         if (!timespec_equal(&inode->i_mtime, &now))
4311                 sync_it = S_MTIME;
4312
4313         if (!timespec_equal(&inode->i_ctime, &now))
4314                 sync_it |= S_CTIME;
4315
4316         if (IS_I_VERSION(inode))
4317                 sync_it |= S_VERSION;
4318
4319         if (!sync_it)
4320                 return 0;
4321
4322         /* Finally allowed to write? Takes lock. */
4323         if (mnt_want_write_file(file))
4324                 return 0;
4325
4326         /* Only change inode inside the lock region */
4327         if (sync_it & S_VERSION)
4328                 inode_inc_iversion(inode);
4329         if (sync_it & S_CTIME)
4330                 inode->i_ctime = now;
4331         if (sync_it & S_MTIME)
4332                 inode->i_mtime = now;
4333         ret = btrfs_dirty_inode(inode);
4334         if (!ret)
4335                 mark_inode_dirty_sync(inode);
4336         mnt_drop_write(file->f_path.mnt);
4337         return ret;
4338 }
4339
4340 /*
4341  * find the highest existing sequence number in a directory
4342  * and then set the in-memory index_cnt variable to reflect
4343  * free sequence numbers
4344  */
4345 static int btrfs_set_inode_index_count(struct inode *inode)
4346 {
4347         struct btrfs_root *root = BTRFS_I(inode)->root;
4348         struct btrfs_key key, found_key;
4349         struct btrfs_path *path;
4350         struct extent_buffer *leaf;
4351         int ret;
4352
4353         key.objectid = btrfs_ino(inode);
4354         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4355         key.offset = (u64)-1;
4356
4357         path = btrfs_alloc_path();
4358         if (!path)
4359                 return -ENOMEM;
4360
4361         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4362         if (ret < 0)
4363                 goto out;
4364         /* FIXME: we should be able to handle this */
4365         if (ret == 0)
4366                 goto out;
4367         ret = 0;
4368
4369         /*
4370          * MAGIC NUMBER EXPLANATION:
4371          * since we search a directory based on f_pos we have to start at 2
4372          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4373          * else has to start at 2
4374          */
4375         if (path->slots[0] == 0) {
4376                 BTRFS_I(inode)->index_cnt = 2;
4377                 goto out;
4378         }
4379
4380         path->slots[0]--;
4381
4382         leaf = path->nodes[0];
4383         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4384
4385         if (found_key.objectid != btrfs_ino(inode) ||
4386             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4387                 BTRFS_I(inode)->index_cnt = 2;
4388                 goto out;
4389         }
4390
4391         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4392 out:
4393         btrfs_free_path(path);
4394         return ret;
4395 }
4396
4397 /*
4398  * helper to find a free sequence number in a given directory.  This current
4399  * code is very simple, later versions will do smarter things in the btree
4400  */
4401 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4402 {
4403         int ret = 0;
4404
4405         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4406                 ret = btrfs_inode_delayed_dir_index_count(dir);
4407                 if (ret) {
4408                         ret = btrfs_set_inode_index_count(dir);
4409                         if (ret)
4410                                 return ret;
4411                 }
4412         }
4413
4414         *index = BTRFS_I(dir)->index_cnt;
4415         BTRFS_I(dir)->index_cnt++;
4416
4417         return ret;
4418 }
4419
4420 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4421                                      struct btrfs_root *root,
4422                                      struct inode *dir,
4423                                      const char *name, int name_len,
4424                                      u64 ref_objectid, u64 objectid, int mode,
4425                                      u64 *index)
4426 {
4427         struct inode *inode;
4428         struct btrfs_inode_item *inode_item;
4429         struct btrfs_key *location;
4430         struct btrfs_path *path;
4431         struct btrfs_inode_ref *ref;
4432         struct btrfs_key key[2];
4433         u32 sizes[2];
4434         unsigned long ptr;
4435         int ret;
4436         int owner;
4437
4438         path = btrfs_alloc_path();
4439         if (!path)
4440                 return ERR_PTR(-ENOMEM);
4441
4442         inode = new_inode(root->fs_info->sb);
4443         if (!inode) {
4444                 btrfs_free_path(path);
4445                 return ERR_PTR(-ENOMEM);
4446         }
4447
4448         /*
4449          * we have to initialize this early, so we can reclaim the inode
4450          * number if we fail afterwards in this function.
4451          */
4452         inode->i_ino = objectid;
4453
4454         if (dir) {
4455                 trace_btrfs_inode_request(dir);
4456
4457                 ret = btrfs_set_inode_index(dir, index);
4458                 if (ret) {
4459                         btrfs_free_path(path);
4460                         iput(inode);
4461                         return ERR_PTR(ret);
4462                 }
4463         }
4464         /*
4465          * index_cnt is ignored for everything but a dir,
4466          * btrfs_get_inode_index_count has an explanation for the magic
4467          * number
4468          */
4469         BTRFS_I(inode)->index_cnt = 2;
4470         BTRFS_I(inode)->root = root;
4471         BTRFS_I(inode)->generation = trans->transid;
4472         inode->i_generation = BTRFS_I(inode)->generation;
4473         btrfs_set_inode_space_info(root, inode);
4474
4475         if (S_ISDIR(mode))
4476                 owner = 0;
4477         else
4478                 owner = 1;
4479
4480         key[0].objectid = objectid;
4481         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4482         key[0].offset = 0;
4483
4484         key[1].objectid = objectid;
4485         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4486         key[1].offset = ref_objectid;
4487
4488         sizes[0] = sizeof(struct btrfs_inode_item);
4489         sizes[1] = name_len + sizeof(*ref);
4490
4491         path->leave_spinning = 1;
4492         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4493         if (ret != 0)
4494                 goto fail;
4495
4496         inode_init_owner(inode, dir, mode);
4497         inode_set_bytes(inode, 0);
4498         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4499         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4500                                   struct btrfs_inode_item);
4501         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4502
4503         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4504                              struct btrfs_inode_ref);
4505         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4506         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4507         ptr = (unsigned long)(ref + 1);
4508         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4509
4510         btrfs_mark_buffer_dirty(path->nodes[0]);
4511         btrfs_free_path(path);
4512
4513         location = &BTRFS_I(inode)->location;
4514         location->objectid = objectid;
4515         location->offset = 0;
4516         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4517
4518         btrfs_inherit_iflags(inode, dir);
4519
4520         if (S_ISREG(mode)) {
4521                 if (btrfs_test_opt(root, NODATASUM))
4522                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4523                 if (btrfs_test_opt(root, NODATACOW) ||
4524                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4525                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4526         }
4527
4528         insert_inode_hash(inode);
4529         inode_tree_add(inode);
4530
4531         trace_btrfs_inode_new(inode);
4532         btrfs_set_inode_last_trans(trans, inode);
4533
4534         return inode;
4535 fail:
4536         if (dir)
4537                 BTRFS_I(dir)->index_cnt--;
4538         btrfs_free_path(path);
4539         iput(inode);
4540         return ERR_PTR(ret);
4541 }
4542
4543 static inline u8 btrfs_inode_type(struct inode *inode)
4544 {
4545         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4546 }
4547
4548 /*
4549  * utility function to add 'inode' into 'parent_inode' with
4550  * a give name and a given sequence number.
4551  * if 'add_backref' is true, also insert a backref from the
4552  * inode to the parent directory.
4553  */
4554 int btrfs_add_link(struct btrfs_trans_handle *trans,
4555                    struct inode *parent_inode, struct inode *inode,
4556                    const char *name, int name_len, int add_backref, u64 index)
4557 {
4558         int ret = 0;
4559         struct btrfs_key key;
4560         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4561         u64 ino = btrfs_ino(inode);
4562         u64 parent_ino = btrfs_ino(parent_inode);
4563
4564         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4565                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4566         } else {
4567                 key.objectid = ino;
4568                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4569                 key.offset = 0;
4570         }
4571
4572         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4573                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4574                                          key.objectid, root->root_key.objectid,
4575                                          parent_ino, index, name, name_len);
4576         } else if (add_backref) {
4577                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4578                                              parent_ino, index);
4579         }
4580
4581         if (ret == 0) {
4582                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4583                                             parent_inode, &key,
4584                                             btrfs_inode_type(inode), index);
4585                 BUG_ON(ret);
4586
4587                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4588                                    name_len * 2);
4589                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4590                 ret = btrfs_update_inode(trans, root, parent_inode);
4591         }
4592         return ret;
4593 }
4594
4595 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4596                             struct inode *dir, struct dentry *dentry,
4597                             struct inode *inode, int backref, u64 index)
4598 {
4599         int err = btrfs_add_link(trans, dir, inode,
4600                                  dentry->d_name.name, dentry->d_name.len,
4601                                  backref, index);
4602         if (err > 0)
4603                 err = -EEXIST;
4604         return err;
4605 }
4606
4607 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4608                         int mode, dev_t rdev)
4609 {
4610         struct btrfs_trans_handle *trans;
4611         struct btrfs_root *root = BTRFS_I(dir)->root;
4612         struct inode *inode = NULL;
4613         int err;
4614         int drop_inode = 0;
4615         u64 objectid;
4616         unsigned long nr = 0;
4617         u64 index = 0;
4618
4619         if (!new_valid_dev(rdev))
4620                 return -EINVAL;
4621
4622         /*
4623          * 2 for inode item and ref
4624          * 2 for dir items
4625          * 1 for xattr if selinux is on
4626          */
4627         trans = btrfs_start_transaction(root, 5);
4628         if (IS_ERR(trans))
4629                 return PTR_ERR(trans);
4630
4631         err = btrfs_find_free_ino(root, &objectid);
4632         if (err)
4633                 goto out_unlock;
4634
4635         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4636                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4637                                 mode, &index);
4638         if (IS_ERR(inode)) {
4639                 err = PTR_ERR(inode);
4640                 goto out_unlock;
4641         }
4642
4643         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4644         if (err) {
4645                 drop_inode = 1;
4646                 goto out_unlock;
4647         }
4648
4649         /*
4650         * If the active LSM wants to access the inode during
4651         * d_instantiate it needs these. Smack checks to see
4652         * if the filesystem supports xattrs by looking at the
4653         * ops vector.
4654         */
4655
4656         inode->i_op = &btrfs_special_inode_operations;
4657         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4658         if (err)
4659                 drop_inode = 1;
4660         else {
4661                 init_special_inode(inode, inode->i_mode, rdev);
4662                 btrfs_update_inode(trans, root, inode);
4663                 d_instantiate(dentry, inode);
4664         }
4665 out_unlock:
4666         nr = trans->blocks_used;
4667         btrfs_end_transaction(trans, root);
4668         btrfs_btree_balance_dirty(root, nr);
4669         if (drop_inode) {
4670                 inode_dec_link_count(inode);
4671                 iput(inode);
4672         }
4673         return err;
4674 }
4675
4676 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4677                         int mode, struct nameidata *nd)
4678 {
4679         struct btrfs_trans_handle *trans;
4680         struct btrfs_root *root = BTRFS_I(dir)->root;
4681         struct inode *inode = NULL;
4682         int drop_inode = 0;
4683         int err;
4684         unsigned long nr = 0;
4685         u64 objectid;
4686         u64 index = 0;
4687
4688         /*
4689          * 2 for inode item and ref
4690          * 2 for dir items
4691          * 1 for xattr if selinux is on
4692          */
4693         trans = btrfs_start_transaction(root, 5);
4694         if (IS_ERR(trans))
4695                 return PTR_ERR(trans);
4696
4697         err = btrfs_find_free_ino(root, &objectid);
4698         if (err)
4699                 goto out_unlock;
4700
4701         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4702                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4703                                 mode, &index);
4704         if (IS_ERR(inode)) {
4705                 err = PTR_ERR(inode);
4706                 goto out_unlock;
4707         }
4708
4709         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4710         if (err) {
4711                 drop_inode = 1;
4712                 goto out_unlock;
4713         }
4714
4715         /*
4716         * If the active LSM wants to access the inode during
4717         * d_instantiate it needs these. Smack checks to see
4718         * if the filesystem supports xattrs by looking at the
4719         * ops vector.
4720         */
4721         inode->i_fop = &btrfs_file_operations;
4722         inode->i_op = &btrfs_file_inode_operations;
4723
4724         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4725         if (err)
4726                 drop_inode = 1;
4727         else {
4728                 inode->i_mapping->a_ops = &btrfs_aops;
4729                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4730                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4731                 d_instantiate(dentry, inode);
4732         }
4733 out_unlock:
4734         nr = trans->blocks_used;
4735         btrfs_end_transaction(trans, root);
4736         if (drop_inode) {
4737                 inode_dec_link_count(inode);
4738                 iput(inode);
4739         }
4740         btrfs_btree_balance_dirty(root, nr);
4741         return err;
4742 }
4743
4744 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4745                       struct dentry *dentry)
4746 {
4747         struct btrfs_trans_handle *trans;
4748         struct btrfs_root *root = BTRFS_I(dir)->root;
4749         struct inode *inode = old_dentry->d_inode;
4750         u64 index;
4751         unsigned long nr = 0;
4752         int err;
4753         int drop_inode = 0;
4754
4755         /* do not allow sys_link's with other subvols of the same device */
4756         if (root->objectid != BTRFS_I(inode)->root->objectid)
4757                 return -EXDEV;
4758
4759         if (inode->i_nlink == ~0U)
4760                 return -EMLINK;
4761
4762         err = btrfs_set_inode_index(dir, &index);
4763         if (err)
4764                 goto fail;
4765
4766         /*
4767          * 2 items for inode and inode ref
4768          * 2 items for dir items
4769          * 1 item for parent inode
4770          */
4771         trans = btrfs_start_transaction(root, 5);
4772         if (IS_ERR(trans)) {
4773                 err = PTR_ERR(trans);
4774                 goto fail;
4775         }
4776
4777         btrfs_inc_nlink(inode);
4778         inode->i_ctime = CURRENT_TIME;
4779         ihold(inode);
4780
4781         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4782
4783         if (err) {
4784                 drop_inode = 1;
4785         } else {
4786                 struct dentry *parent = dentry->d_parent;
4787                 err = btrfs_update_inode(trans, root, inode);
4788                 BUG_ON(err);
4789                 d_instantiate(dentry, inode);
4790                 btrfs_log_new_name(trans, inode, NULL, parent);
4791         }
4792
4793         nr = trans->blocks_used;
4794         btrfs_end_transaction(trans, root);
4795 fail:
4796         if (drop_inode) {
4797                 inode_dec_link_count(inode);
4798                 iput(inode);
4799         }
4800         btrfs_btree_balance_dirty(root, nr);
4801         return err;
4802 }
4803
4804 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4805 {
4806         struct inode *inode = NULL;
4807         struct btrfs_trans_handle *trans;
4808         struct btrfs_root *root = BTRFS_I(dir)->root;
4809         int err = 0;
4810         int drop_on_err = 0;
4811         u64 objectid = 0;
4812         u64 index = 0;
4813         unsigned long nr = 1;
4814
4815         /*
4816          * 2 items for inode and ref
4817          * 2 items for dir items
4818          * 1 for xattr if selinux is on
4819          */
4820         trans = btrfs_start_transaction(root, 5);
4821         if (IS_ERR(trans))
4822                 return PTR_ERR(trans);
4823
4824         err = btrfs_find_free_ino(root, &objectid);
4825         if (err)
4826                 goto out_fail;
4827
4828         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4829                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4830                                 S_IFDIR | mode, &index);
4831         if (IS_ERR(inode)) {
4832                 err = PTR_ERR(inode);
4833                 goto out_fail;
4834         }
4835
4836         drop_on_err = 1;
4837
4838         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4839         if (err)
4840                 goto out_fail;
4841
4842         inode->i_op = &btrfs_dir_inode_operations;
4843         inode->i_fop = &btrfs_dir_file_operations;
4844
4845         btrfs_i_size_write(inode, 0);
4846         err = btrfs_update_inode(trans, root, inode);
4847         if (err)
4848                 goto out_fail;
4849
4850         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4851                              dentry->d_name.len, 0, index);
4852         if (err)
4853                 goto out_fail;
4854
4855         d_instantiate(dentry, inode);
4856         drop_on_err = 0;
4857
4858 out_fail:
4859         nr = trans->blocks_used;
4860         btrfs_end_transaction(trans, root);
4861         if (drop_on_err)
4862                 iput(inode);
4863         btrfs_btree_balance_dirty(root, nr);
4864         return err;
4865 }
4866
4867 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4868  * and an extent that you want to insert, deal with overlap and insert
4869  * the new extent into the tree.
4870  */
4871 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4872                                 struct extent_map *existing,
4873                                 struct extent_map *em,
4874                                 u64 map_start, u64 map_len)
4875 {
4876         u64 start_diff;
4877
4878         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4879         start_diff = map_start - em->start;
4880         em->start = map_start;
4881         em->len = map_len;
4882         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4883             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4884                 em->block_start += start_diff;
4885                 em->block_len -= start_diff;
4886         }
4887         return add_extent_mapping(em_tree, em);
4888 }
4889
4890 static noinline int uncompress_inline(struct btrfs_path *path,
4891                                       struct inode *inode, struct page *page,
4892                                       size_t pg_offset, u64 extent_offset,
4893                                       struct btrfs_file_extent_item *item)
4894 {
4895         int ret;
4896         struct extent_buffer *leaf = path->nodes[0];
4897         char *tmp;
4898         size_t max_size;
4899         unsigned long inline_size;
4900         unsigned long ptr;
4901         int compress_type;
4902
4903         WARN_ON(pg_offset != 0);
4904         compress_type = btrfs_file_extent_compression(leaf, item);
4905         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4906         inline_size = btrfs_file_extent_inline_item_len(leaf,
4907                                         btrfs_item_nr(leaf, path->slots[0]));
4908         tmp = kmalloc(inline_size, GFP_NOFS);
4909         if (!tmp)
4910                 return -ENOMEM;
4911         ptr = btrfs_file_extent_inline_start(item);
4912
4913         read_extent_buffer(leaf, tmp, ptr, inline_size);
4914
4915         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4916         ret = btrfs_decompress(compress_type, tmp, page,
4917                                extent_offset, inline_size, max_size);
4918         if (ret) {
4919                 char *kaddr = kmap_atomic(page, KM_USER0);
4920                 unsigned long copy_size = min_t(u64,
4921                                   PAGE_CACHE_SIZE - pg_offset,
4922                                   max_size - extent_offset);
4923                 memset(kaddr + pg_offset, 0, copy_size);
4924                 kunmap_atomic(kaddr, KM_USER0);
4925         }
4926         kfree(tmp);
4927         return 0;
4928 }
4929
4930 /*
4931  * a bit scary, this does extent mapping from logical file offset to the disk.
4932  * the ugly parts come from merging extents from the disk with the in-ram
4933  * representation.  This gets more complex because of the data=ordered code,
4934  * where the in-ram extents might be locked pending data=ordered completion.
4935  *
4936  * This also copies inline extents directly into the page.
4937  */
4938
4939 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4940                                     size_t pg_offset, u64 start, u64 len,
4941                                     int create)
4942 {
4943         int ret;
4944         int err = 0;
4945         u64 bytenr;
4946         u64 extent_start = 0;
4947         u64 extent_end = 0;
4948         u64 objectid = btrfs_ino(inode);
4949         u32 found_type;
4950         struct btrfs_path *path = NULL;
4951         struct btrfs_root *root = BTRFS_I(inode)->root;
4952         struct btrfs_file_extent_item *item;
4953         struct extent_buffer *leaf;
4954         struct btrfs_key found_key;
4955         struct extent_map *em = NULL;
4956         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4957         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4958         struct btrfs_trans_handle *trans = NULL;
4959         int compress_type;
4960
4961 again:
4962         read_lock(&em_tree->lock);
4963         em = lookup_extent_mapping(em_tree, start, len);
4964         if (em)
4965                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4966         read_unlock(&em_tree->lock);
4967
4968         if (em) {
4969                 if (em->start > start || em->start + em->len <= start)
4970                         free_extent_map(em);
4971                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4972                         free_extent_map(em);
4973                 else
4974                         goto out;
4975         }
4976         em = alloc_extent_map();
4977         if (!em) {
4978                 err = -ENOMEM;
4979                 goto out;
4980         }
4981         em->bdev = root->fs_info->fs_devices->latest_bdev;
4982         em->start = EXTENT_MAP_HOLE;
4983         em->orig_start = EXTENT_MAP_HOLE;
4984         em->len = (u64)-1;
4985         em->block_len = (u64)-1;
4986
4987         if (!path) {
4988                 path = btrfs_alloc_path();
4989                 if (!path) {
4990                         err = -ENOMEM;
4991                         goto out;
4992                 }
4993                 /*
4994                  * Chances are we'll be called again, so go ahead and do
4995                  * readahead
4996                  */
4997                 path->reada = 1;
4998         }
4999
5000         ret = btrfs_lookup_file_extent(trans, root, path,
5001                                        objectid, start, trans != NULL);
5002         if (ret < 0) {
5003                 err = ret;
5004                 goto out;
5005         }
5006
5007         if (ret != 0) {
5008                 if (path->slots[0] == 0)
5009                         goto not_found;
5010                 path->slots[0]--;
5011         }
5012
5013         leaf = path->nodes[0];
5014         item = btrfs_item_ptr(leaf, path->slots[0],
5015                               struct btrfs_file_extent_item);
5016         /* are we inside the extent that was found? */
5017         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5018         found_type = btrfs_key_type(&found_key);
5019         if (found_key.objectid != objectid ||
5020             found_type != BTRFS_EXTENT_DATA_KEY) {
5021                 goto not_found;
5022         }
5023
5024         found_type = btrfs_file_extent_type(leaf, item);
5025         extent_start = found_key.offset;
5026         compress_type = btrfs_file_extent_compression(leaf, item);
5027         if (found_type == BTRFS_FILE_EXTENT_REG ||
5028             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5029                 extent_end = extent_start +
5030                        btrfs_file_extent_num_bytes(leaf, item);
5031         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5032                 size_t size;
5033                 size = btrfs_file_extent_inline_len(leaf, item);
5034                 extent_end = (extent_start + size + root->sectorsize - 1) &
5035                         ~((u64)root->sectorsize - 1);
5036         }
5037
5038         if (start >= extent_end) {
5039                 path->slots[0]++;
5040                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5041                         ret = btrfs_next_leaf(root, path);
5042                         if (ret < 0) {
5043                                 err = ret;
5044                                 goto out;
5045                         }
5046                         if (ret > 0)
5047                                 goto not_found;
5048                         leaf = path->nodes[0];
5049                 }
5050                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5051                 if (found_key.objectid != objectid ||
5052                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5053                         goto not_found;
5054                 if (start + len <= found_key.offset)
5055                         goto not_found;
5056                 em->start = start;
5057                 em->len = found_key.offset - start;
5058                 goto not_found_em;
5059         }
5060
5061         if (found_type == BTRFS_FILE_EXTENT_REG ||
5062             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5063                 em->start = extent_start;
5064                 em->len = extent_end - extent_start;
5065                 em->orig_start = extent_start -
5066                                  btrfs_file_extent_offset(leaf, item);
5067                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5068                 if (bytenr == 0) {
5069                         em->block_start = EXTENT_MAP_HOLE;
5070                         goto insert;
5071                 }
5072                 if (compress_type != BTRFS_COMPRESS_NONE) {
5073                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5074                         em->compress_type = compress_type;
5075                         em->block_start = bytenr;
5076                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5077                                                                          item);
5078                 } else {
5079                         bytenr += btrfs_file_extent_offset(leaf, item);
5080                         em->block_start = bytenr;
5081                         em->block_len = em->len;
5082                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5083                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5084                 }
5085                 goto insert;
5086         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5087                 unsigned long ptr;
5088                 char *map;
5089                 size_t size;
5090                 size_t extent_offset;
5091                 size_t copy_size;
5092
5093                 em->block_start = EXTENT_MAP_INLINE;
5094                 if (!page || create) {
5095                         em->start = extent_start;
5096                         em->len = extent_end - extent_start;
5097                         goto out;
5098                 }
5099
5100                 size = btrfs_file_extent_inline_len(leaf, item);
5101                 extent_offset = page_offset(page) + pg_offset - extent_start;
5102                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5103                                 size - extent_offset);
5104                 em->start = extent_start + extent_offset;
5105                 em->len = (copy_size + root->sectorsize - 1) &
5106                         ~((u64)root->sectorsize - 1);
5107                 em->orig_start = EXTENT_MAP_INLINE;
5108                 if (compress_type) {
5109                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5110                         em->compress_type = compress_type;
5111                 }
5112                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5113                 if (create == 0 && !PageUptodate(page)) {
5114                         if (btrfs_file_extent_compression(leaf, item) !=
5115                             BTRFS_COMPRESS_NONE) {
5116                                 ret = uncompress_inline(path, inode, page,
5117                                                         pg_offset,
5118                                                         extent_offset, item);
5119                                 BUG_ON(ret);
5120                         } else {
5121                                 map = kmap(page);
5122                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5123                                                    copy_size);
5124                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5125                                         memset(map + pg_offset + copy_size, 0,
5126                                                PAGE_CACHE_SIZE - pg_offset -
5127                                                copy_size);
5128                                 }
5129                                 kunmap(page);
5130                         }
5131                         flush_dcache_page(page);
5132                 } else if (create && PageUptodate(page)) {
5133                         BUG();
5134                         if (!trans) {
5135                                 kunmap(page);
5136                                 free_extent_map(em);
5137                                 em = NULL;
5138
5139                                 btrfs_release_path(path);
5140                                 trans = btrfs_join_transaction(root);
5141
5142                                 if (IS_ERR(trans))
5143                                         return ERR_CAST(trans);
5144                                 goto again;
5145                         }
5146                         map = kmap(page);
5147                         write_extent_buffer(leaf, map + pg_offset, ptr,
5148                                             copy_size);
5149                         kunmap(page);
5150                         btrfs_mark_buffer_dirty(leaf);
5151                 }
5152                 set_extent_uptodate(io_tree, em->start,
5153                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5154                 goto insert;
5155         } else {
5156                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5157                 WARN_ON(1);
5158         }
5159 not_found:
5160         em->start = start;
5161         em->len = len;
5162 not_found_em:
5163         em->block_start = EXTENT_MAP_HOLE;
5164         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5165 insert:
5166         btrfs_release_path(path);
5167         if (em->start > start || extent_map_end(em) <= start) {
5168                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5169                        "[%llu %llu]\n", (unsigned long long)em->start,
5170                        (unsigned long long)em->len,
5171                        (unsigned long long)start,
5172                        (unsigned long long)len);
5173                 err = -EIO;
5174                 goto out;
5175         }
5176
5177         err = 0;
5178         write_lock(&em_tree->lock);
5179         ret = add_extent_mapping(em_tree, em);
5180         /* it is possible that someone inserted the extent into the tree
5181          * while we had the lock dropped.  It is also possible that
5182          * an overlapping map exists in the tree
5183          */
5184         if (ret == -EEXIST) {
5185                 struct extent_map *existing;
5186
5187                 ret = 0;
5188
5189                 existing = lookup_extent_mapping(em_tree, start, len);
5190                 if (existing && (existing->start > start ||
5191                     existing->start + existing->len <= start)) {
5192                         free_extent_map(existing);
5193                         existing = NULL;
5194                 }
5195                 if (!existing) {
5196                         existing = lookup_extent_mapping(em_tree, em->start,
5197                                                          em->len);
5198                         if (existing) {
5199                                 err = merge_extent_mapping(em_tree, existing,
5200                                                            em, start,
5201                                                            root->sectorsize);
5202                                 free_extent_map(existing);
5203                                 if (err) {
5204                                         free_extent_map(em);
5205                                         em = NULL;
5206                                 }
5207                         } else {
5208                                 err = -EIO;
5209                                 free_extent_map(em);
5210                                 em = NULL;
5211                         }
5212                 } else {
5213                         free_extent_map(em);
5214                         em = existing;
5215                         err = 0;
5216                 }
5217         }
5218         write_unlock(&em_tree->lock);
5219 out:
5220
5221         trace_btrfs_get_extent(root, em);
5222
5223         if (path)
5224                 btrfs_free_path(path);
5225         if (trans) {
5226                 ret = btrfs_end_transaction(trans, root);
5227                 if (!err)
5228                         err = ret;
5229         }
5230         if (err) {
5231                 free_extent_map(em);
5232                 return ERR_PTR(err);
5233         }
5234         return em;
5235 }
5236
5237 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5238                                            size_t pg_offset, u64 start, u64 len,
5239                                            int create)
5240 {
5241         struct extent_map *em;
5242         struct extent_map *hole_em = NULL;
5243         u64 range_start = start;
5244         u64 end;
5245         u64 found;
5246         u64 found_end;
5247         int err = 0;
5248
5249         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5250         if (IS_ERR(em))
5251                 return em;
5252         if (em) {
5253                 /*
5254                  * if our em maps to a hole, there might
5255                  * actually be delalloc bytes behind it
5256                  */
5257                 if (em->block_start != EXTENT_MAP_HOLE)
5258                         return em;
5259                 else
5260                         hole_em = em;
5261         }
5262
5263         /* check to see if we've wrapped (len == -1 or similar) */
5264         end = start + len;
5265         if (end < start)
5266                 end = (u64)-1;
5267         else
5268                 end -= 1;
5269
5270         em = NULL;
5271
5272         /* ok, we didn't find anything, lets look for delalloc */
5273         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5274                                  end, len, EXTENT_DELALLOC, 1);
5275         found_end = range_start + found;
5276         if (found_end < range_start)
5277                 found_end = (u64)-1;
5278
5279         /*
5280          * we didn't find anything useful, return
5281          * the original results from get_extent()
5282          */
5283         if (range_start > end || found_end <= start) {
5284                 em = hole_em;
5285                 hole_em = NULL;
5286                 goto out;
5287         }
5288
5289         /* adjust the range_start to make sure it doesn't
5290          * go backwards from the start they passed in
5291          */
5292         range_start = max(start,range_start);
5293         found = found_end - range_start;
5294
5295         if (found > 0) {
5296                 u64 hole_start = start;
5297                 u64 hole_len = len;
5298
5299                 em = alloc_extent_map();
5300                 if (!em) {
5301                         err = -ENOMEM;
5302                         goto out;
5303                 }
5304                 /*
5305                  * when btrfs_get_extent can't find anything it
5306                  * returns one huge hole
5307                  *
5308                  * make sure what it found really fits our range, and
5309                  * adjust to make sure it is based on the start from
5310                  * the caller
5311                  */
5312                 if (hole_em) {
5313                         u64 calc_end = extent_map_end(hole_em);
5314
5315                         if (calc_end <= start || (hole_em->start > end)) {
5316                                 free_extent_map(hole_em);
5317                                 hole_em = NULL;
5318                         } else {
5319                                 hole_start = max(hole_em->start, start);
5320                                 hole_len = calc_end - hole_start;
5321                         }
5322                 }
5323                 em->bdev = NULL;
5324                 if (hole_em && range_start > hole_start) {
5325                         /* our hole starts before our delalloc, so we
5326                          * have to return just the parts of the hole
5327                          * that go until  the delalloc starts
5328                          */
5329                         em->len = min(hole_len,
5330                                       range_start - hole_start);
5331                         em->start = hole_start;
5332                         em->orig_start = hole_start;
5333                         /*
5334                          * don't adjust block start at all,
5335                          * it is fixed at EXTENT_MAP_HOLE
5336                          */
5337                         em->block_start = hole_em->block_start;
5338                         em->block_len = hole_len;
5339                 } else {
5340                         em->start = range_start;
5341                         em->len = found;
5342                         em->orig_start = range_start;
5343                         em->block_start = EXTENT_MAP_DELALLOC;
5344                         em->block_len = found;
5345                 }
5346         } else if (hole_em) {
5347                 return hole_em;
5348         }
5349 out:
5350
5351         free_extent_map(hole_em);
5352         if (err) {
5353                 free_extent_map(em);
5354                 return ERR_PTR(err);
5355         }
5356         return em;
5357 }
5358
5359 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5360                                                   struct extent_map *em,
5361                                                   u64 start, u64 len)
5362 {
5363         struct btrfs_root *root = BTRFS_I(inode)->root;
5364         struct btrfs_trans_handle *trans;
5365         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5366         struct btrfs_key ins;
5367         u64 alloc_hint;
5368         int ret;
5369         bool insert = false;
5370
5371         /*
5372          * Ok if the extent map we looked up is a hole and is for the exact
5373          * range we want, there is no reason to allocate a new one, however if
5374          * it is not right then we need to free this one and drop the cache for
5375          * our range.
5376          */
5377         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5378             em->len != len) {
5379                 free_extent_map(em);
5380                 em = NULL;
5381                 insert = true;
5382                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5383         }
5384
5385         trans = btrfs_join_transaction(root);
5386         if (IS_ERR(trans))
5387                 return ERR_CAST(trans);
5388
5389         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5390                 btrfs_add_inode_defrag(trans, inode);
5391
5392         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5393
5394         alloc_hint = get_extent_allocation_hint(inode, start, len);
5395         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5396                                    alloc_hint, (u64)-1, &ins, 1);
5397         if (ret) {
5398                 em = ERR_PTR(ret);
5399                 goto out;
5400         }
5401
5402         if (!em) {
5403                 em = alloc_extent_map();
5404                 if (!em) {
5405                         em = ERR_PTR(-ENOMEM);
5406                         goto out;
5407                 }
5408         }
5409
5410         em->start = start;
5411         em->orig_start = em->start;
5412         em->len = ins.offset;
5413
5414         em->block_start = ins.objectid;
5415         em->block_len = ins.offset;
5416         em->bdev = root->fs_info->fs_devices->latest_bdev;
5417
5418         /*
5419          * We need to do this because if we're using the original em we searched
5420          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5421          */
5422         em->flags = 0;
5423         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5424
5425         while (insert) {
5426                 write_lock(&em_tree->lock);
5427                 ret = add_extent_mapping(em_tree, em);
5428                 write_unlock(&em_tree->lock);
5429                 if (ret != -EEXIST)
5430                         break;
5431                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5432         }
5433
5434         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5435                                            ins.offset, ins.offset, 0);
5436         if (ret) {
5437                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5438                 em = ERR_PTR(ret);
5439         }
5440 out:
5441         btrfs_end_transaction(trans, root);
5442         return em;
5443 }
5444
5445 /*
5446  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5447  * block must be cow'd
5448  */
5449 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5450                                       struct inode *inode, u64 offset, u64 len)
5451 {
5452         struct btrfs_path *path;
5453         int ret;
5454         struct extent_buffer *leaf;
5455         struct btrfs_root *root = BTRFS_I(inode)->root;
5456         struct btrfs_file_extent_item *fi;
5457         struct btrfs_key key;
5458         u64 disk_bytenr;
5459         u64 backref_offset;
5460         u64 extent_end;
5461         u64 num_bytes;
5462         int slot;
5463         int found_type;
5464
5465         path = btrfs_alloc_path();
5466         if (!path)
5467                 return -ENOMEM;
5468
5469         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5470                                        offset, 0);
5471         if (ret < 0)
5472                 goto out;
5473
5474         slot = path->slots[0];
5475         if (ret == 1) {
5476                 if (slot == 0) {
5477                         /* can't find the item, must cow */
5478                         ret = 0;
5479                         goto out;
5480                 }
5481                 slot--;
5482         }
5483         ret = 0;
5484         leaf = path->nodes[0];
5485         btrfs_item_key_to_cpu(leaf, &key, slot);
5486         if (key.objectid != btrfs_ino(inode) ||
5487             key.type != BTRFS_EXTENT_DATA_KEY) {
5488                 /* not our file or wrong item type, must cow */
5489                 goto out;
5490         }
5491
5492         if (key.offset > offset) {
5493                 /* Wrong offset, must cow */
5494                 goto out;
5495         }
5496
5497         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5498         found_type = btrfs_file_extent_type(leaf, fi);
5499         if (found_type != BTRFS_FILE_EXTENT_REG &&
5500             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5501                 /* not a regular extent, must cow */
5502                 goto out;
5503         }
5504         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5505         backref_offset = btrfs_file_extent_offset(leaf, fi);
5506
5507         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5508         if (extent_end < offset + len) {
5509                 /* extent doesn't include our full range, must cow */
5510                 goto out;
5511         }
5512
5513         if (btrfs_extent_readonly(root, disk_bytenr))
5514                 goto out;
5515
5516         /*
5517          * look for other files referencing this extent, if we
5518          * find any we must cow
5519          */
5520         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5521                                   key.offset - backref_offset, disk_bytenr))
5522                 goto out;
5523
5524         /*
5525          * adjust disk_bytenr and num_bytes to cover just the bytes
5526          * in this extent we are about to write.  If there
5527          * are any csums in that range we have to cow in order
5528          * to keep the csums correct
5529          */
5530         disk_bytenr += backref_offset;
5531         disk_bytenr += offset - key.offset;
5532         num_bytes = min(offset + len, extent_end) - offset;
5533         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5534                                 goto out;
5535         /*
5536          * all of the above have passed, it is safe to overwrite this extent
5537          * without cow
5538          */
5539         ret = 1;
5540 out:
5541         btrfs_free_path(path);
5542         return ret;
5543 }
5544
5545 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5546                                    struct buffer_head *bh_result, int create)
5547 {
5548         struct extent_map *em;
5549         struct btrfs_root *root = BTRFS_I(inode)->root;
5550         u64 start = iblock << inode->i_blkbits;
5551         u64 len = bh_result->b_size;
5552         struct btrfs_trans_handle *trans;
5553
5554         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5555         if (IS_ERR(em))
5556                 return PTR_ERR(em);
5557
5558         /*
5559          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5560          * io.  INLINE is special, and we could probably kludge it in here, but
5561          * it's still buffered so for safety lets just fall back to the generic
5562          * buffered path.
5563          *
5564          * For COMPRESSED we _have_ to read the entire extent in so we can
5565          * decompress it, so there will be buffering required no matter what we
5566          * do, so go ahead and fallback to buffered.
5567          *
5568          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5569          * to buffered IO.  Don't blame me, this is the price we pay for using
5570          * the generic code.
5571          */
5572         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5573             em->block_start == EXTENT_MAP_INLINE) {
5574                 free_extent_map(em);
5575                 return -ENOTBLK;
5576         }
5577
5578         /* Just a good old fashioned hole, return */
5579         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5580                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5581                 free_extent_map(em);
5582                 /* DIO will do one hole at a time, so just unlock a sector */
5583                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5584                               start + root->sectorsize - 1, GFP_NOFS);
5585                 return 0;
5586         }
5587
5588         /*
5589          * We don't allocate a new extent in the following cases
5590          *
5591          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5592          * existing extent.
5593          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5594          * just use the extent.
5595          *
5596          */
5597         if (!create) {
5598                 len = em->len - (start - em->start);
5599                 goto map;
5600         }
5601
5602         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5603             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5604              em->block_start != EXTENT_MAP_HOLE)) {
5605                 int type;
5606                 int ret;
5607                 u64 block_start;
5608
5609                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5610                         type = BTRFS_ORDERED_PREALLOC;
5611                 else
5612                         type = BTRFS_ORDERED_NOCOW;
5613                 len = min(len, em->len - (start - em->start));
5614                 block_start = em->block_start + (start - em->start);
5615
5616                 /*
5617                  * we're not going to log anything, but we do need
5618                  * to make sure the current transaction stays open
5619                  * while we look for nocow cross refs
5620                  */
5621                 trans = btrfs_join_transaction(root);
5622                 if (IS_ERR(trans))
5623                         goto must_cow;
5624
5625                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5626                         ret = btrfs_add_ordered_extent_dio(inode, start,
5627                                            block_start, len, len, type);
5628                         btrfs_end_transaction(trans, root);
5629                         if (ret) {
5630                                 free_extent_map(em);
5631                                 return ret;
5632                         }
5633                         goto unlock;
5634                 }
5635                 btrfs_end_transaction(trans, root);
5636         }
5637 must_cow:
5638         /*
5639          * this will cow the extent, reset the len in case we changed
5640          * it above
5641          */
5642         len = bh_result->b_size;
5643         em = btrfs_new_extent_direct(inode, em, start, len);
5644         if (IS_ERR(em))
5645                 return PTR_ERR(em);
5646         len = min(len, em->len - (start - em->start));
5647 unlock:
5648         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5649                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5650                           0, NULL, GFP_NOFS);
5651 map:
5652         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5653                 inode->i_blkbits;
5654         bh_result->b_size = len;
5655         bh_result->b_bdev = em->bdev;
5656         set_buffer_mapped(bh_result);
5657         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5658                 set_buffer_new(bh_result);
5659
5660         free_extent_map(em);
5661
5662         return 0;
5663 }
5664
5665 struct btrfs_dio_private {
5666         struct inode *inode;
5667         u64 logical_offset;
5668         u64 disk_bytenr;
5669         u64 bytes;
5670         u32 *csums;
5671         void *private;
5672
5673         /* number of bios pending for this dio */
5674         atomic_t pending_bios;
5675
5676         /* IO errors */
5677         int errors;
5678
5679         struct bio *orig_bio;
5680 };
5681
5682 static void btrfs_endio_direct_read(struct bio *bio, int err)
5683 {
5684         struct btrfs_dio_private *dip = bio->bi_private;
5685         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5686         struct bio_vec *bvec = bio->bi_io_vec;
5687         struct inode *inode = dip->inode;
5688         struct btrfs_root *root = BTRFS_I(inode)->root;
5689         u64 start;
5690         u32 *private = dip->csums;
5691
5692         start = dip->logical_offset;
5693         do {
5694                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5695                         struct page *page = bvec->bv_page;
5696                         char *kaddr;
5697                         u32 csum = ~(u32)0;
5698                         unsigned long flags;
5699
5700                         local_irq_save(flags);
5701                         kaddr = kmap_atomic(page, KM_IRQ0);
5702                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5703                                                csum, bvec->bv_len);
5704                         btrfs_csum_final(csum, (char *)&csum);
5705                         kunmap_atomic(kaddr, KM_IRQ0);
5706                         local_irq_restore(flags);
5707
5708                         flush_dcache_page(bvec->bv_page);
5709                         if (csum != *private) {
5710                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5711                                       " %llu csum %u private %u\n",
5712                                       (unsigned long long)btrfs_ino(inode),
5713                                       (unsigned long long)start,
5714                                       csum, *private);
5715                                 err = -EIO;
5716                         }
5717                 }
5718
5719                 start += bvec->bv_len;
5720                 private++;
5721                 bvec++;
5722         } while (bvec <= bvec_end);
5723
5724         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5725                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5726         bio->bi_private = dip->private;
5727
5728         kfree(dip->csums);
5729         kfree(dip);
5730
5731         /* If we had a csum failure make sure to clear the uptodate flag */
5732         if (err)
5733                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5734         dio_end_io(bio, err);
5735 }
5736
5737 static void btrfs_endio_direct_write(struct bio *bio, int err)
5738 {
5739         struct btrfs_dio_private *dip = bio->bi_private;
5740         struct inode *inode = dip->inode;
5741         struct btrfs_root *root = BTRFS_I(inode)->root;
5742         struct btrfs_trans_handle *trans;
5743         struct btrfs_ordered_extent *ordered = NULL;
5744         struct extent_state *cached_state = NULL;
5745         u64 ordered_offset = dip->logical_offset;
5746         u64 ordered_bytes = dip->bytes;
5747         int ret;
5748
5749         if (err)
5750                 goto out_done;
5751 again:
5752         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5753                                                    &ordered_offset,
5754                                                    ordered_bytes);
5755         if (!ret)
5756                 goto out_test;
5757
5758         BUG_ON(!ordered);
5759
5760         trans = btrfs_join_transaction(root);
5761         if (IS_ERR(trans)) {
5762                 err = -ENOMEM;
5763                 goto out;
5764         }
5765         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5766
5767         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5768                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5769                 if (!ret)
5770                         err = btrfs_update_inode_fallback(trans, root, inode);
5771                 goto out;
5772         }
5773
5774         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5775                          ordered->file_offset + ordered->len - 1, 0,
5776                          &cached_state, GFP_NOFS);
5777
5778         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5779                 ret = btrfs_mark_extent_written(trans, inode,
5780                                                 ordered->file_offset,
5781                                                 ordered->file_offset +
5782                                                 ordered->len);
5783                 if (ret) {
5784                         err = ret;
5785                         goto out_unlock;
5786                 }
5787         } else {
5788                 ret = insert_reserved_file_extent(trans, inode,
5789                                                   ordered->file_offset,
5790                                                   ordered->start,
5791                                                   ordered->disk_len,
5792                                                   ordered->len,
5793                                                   ordered->len,
5794                                                   0, 0, 0,
5795                                                   BTRFS_FILE_EXTENT_REG);
5796                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5797                                    ordered->file_offset, ordered->len);
5798                 if (ret) {
5799                         err = ret;
5800                         WARN_ON(1);
5801                         goto out_unlock;
5802                 }
5803         }
5804
5805         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5806         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5807         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5808                 btrfs_update_inode_fallback(trans, root, inode);
5809         ret = 0;
5810 out_unlock:
5811         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5812                              ordered->file_offset + ordered->len - 1,
5813                              &cached_state, GFP_NOFS);
5814 out:
5815         btrfs_delalloc_release_metadata(inode, ordered->len);
5816         btrfs_end_transaction(trans, root);
5817         ordered_offset = ordered->file_offset + ordered->len;
5818         btrfs_put_ordered_extent(ordered);
5819         btrfs_put_ordered_extent(ordered);
5820
5821 out_test:
5822         /*
5823          * our bio might span multiple ordered extents.  If we haven't
5824          * completed the accounting for the whole dio, go back and try again
5825          */
5826         if (ordered_offset < dip->logical_offset + dip->bytes) {
5827                 ordered_bytes = dip->logical_offset + dip->bytes -
5828                         ordered_offset;
5829                 goto again;
5830         }
5831 out_done:
5832         bio->bi_private = dip->private;
5833
5834         kfree(dip->csums);
5835         kfree(dip);
5836
5837         /* If we had an error make sure to clear the uptodate flag */
5838         if (err)
5839                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5840         dio_end_io(bio, err);
5841 }
5842
5843 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5844                                     struct bio *bio, int mirror_num,
5845                                     unsigned long bio_flags, u64 offset)
5846 {
5847         int ret;
5848         struct btrfs_root *root = BTRFS_I(inode)->root;
5849         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5850         BUG_ON(ret);
5851         return 0;
5852 }
5853
5854 static void btrfs_end_dio_bio(struct bio *bio, int err)
5855 {
5856         struct btrfs_dio_private *dip = bio->bi_private;
5857
5858         if (err) {
5859                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5860                       "sector %#Lx len %u err no %d\n",
5861                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5862                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5863                 dip->errors = 1;
5864
5865                 /*
5866                  * before atomic variable goto zero, we must make sure
5867                  * dip->errors is perceived to be set.
5868                  */
5869                 smp_mb__before_atomic_dec();
5870         }
5871
5872         /* if there are more bios still pending for this dio, just exit */
5873         if (!atomic_dec_and_test(&dip->pending_bios))
5874                 goto out;
5875
5876         if (dip->errors)
5877                 bio_io_error(dip->orig_bio);
5878         else {
5879                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5880                 bio_endio(dip->orig_bio, 0);
5881         }
5882 out:
5883         bio_put(bio);
5884 }
5885
5886 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5887                                        u64 first_sector, gfp_t gfp_flags)
5888 {
5889         int nr_vecs = bio_get_nr_vecs(bdev);
5890         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5891 }
5892
5893 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5894                                          int rw, u64 file_offset, int skip_sum,
5895                                          u32 *csums, int async_submit)
5896 {
5897         int write = rw & REQ_WRITE;
5898         struct btrfs_root *root = BTRFS_I(inode)->root;
5899         int ret;
5900
5901         bio_get(bio);
5902         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5903         if (ret)
5904                 goto err;
5905
5906         if (skip_sum)
5907                 goto map;
5908
5909         if (write && async_submit) {
5910                 ret = btrfs_wq_submit_bio(root->fs_info,
5911                                    inode, rw, bio, 0, 0,
5912                                    file_offset,
5913                                    __btrfs_submit_bio_start_direct_io,
5914                                    __btrfs_submit_bio_done);
5915                 goto err;
5916         } else if (write) {
5917                 /*
5918                  * If we aren't doing async submit, calculate the csum of the
5919                  * bio now.
5920                  */
5921                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5922                 if (ret)
5923                         goto err;
5924         } else if (!skip_sum) {
5925                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5926                                           file_offset, csums);
5927                 if (ret)
5928                         goto err;
5929         }
5930
5931 map:
5932         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5933 err:
5934         bio_put(bio);
5935         return ret;
5936 }
5937
5938 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5939                                     int skip_sum)
5940 {
5941         struct inode *inode = dip->inode;
5942         struct btrfs_root *root = BTRFS_I(inode)->root;
5943         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5944         struct bio *bio;
5945         struct bio *orig_bio = dip->orig_bio;
5946         struct bio_vec *bvec = orig_bio->bi_io_vec;
5947         u64 start_sector = orig_bio->bi_sector;
5948         u64 file_offset = dip->logical_offset;
5949         u64 submit_len = 0;
5950         u64 map_length;
5951         int nr_pages = 0;
5952         u32 *csums = dip->csums;
5953         int ret = 0;
5954         int async_submit = 0;
5955         int write = rw & REQ_WRITE;
5956
5957         map_length = orig_bio->bi_size;
5958         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5959                               &map_length, NULL, 0);
5960         if (ret) {
5961                 bio_put(orig_bio);
5962                 return -EIO;
5963         }
5964
5965         if (map_length >= orig_bio->bi_size) {
5966                 bio = orig_bio;
5967                 goto submit;
5968         }
5969
5970         async_submit = 1;
5971         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5972         if (!bio)
5973                 return -ENOMEM;
5974         bio->bi_private = dip;
5975         bio->bi_end_io = btrfs_end_dio_bio;
5976         atomic_inc(&dip->pending_bios);
5977
5978         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5979                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5980                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5981                                  bvec->bv_offset) < bvec->bv_len)) {
5982                         /*
5983                          * inc the count before we submit the bio so
5984                          * we know the end IO handler won't happen before
5985                          * we inc the count. Otherwise, the dip might get freed
5986                          * before we're done setting it up
5987                          */
5988                         atomic_inc(&dip->pending_bios);
5989                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5990                                                      file_offset, skip_sum,
5991                                                      csums, async_submit);
5992                         if (ret) {
5993                                 bio_put(bio);
5994                                 atomic_dec(&dip->pending_bios);
5995                                 goto out_err;
5996                         }
5997
5998                         /* Write's use the ordered csums */
5999                         if (!write && !skip_sum)
6000                                 csums = csums + nr_pages;
6001                         start_sector += submit_len >> 9;
6002                         file_offset += submit_len;
6003
6004                         submit_len = 0;
6005                         nr_pages = 0;
6006
6007                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6008                                                   start_sector, GFP_NOFS);
6009                         if (!bio)
6010                                 goto out_err;
6011                         bio->bi_private = dip;
6012                         bio->bi_end_io = btrfs_end_dio_bio;
6013
6014                         map_length = orig_bio->bi_size;
6015                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6016                                               &map_length, NULL, 0);
6017                         if (ret) {
6018                                 bio_put(bio);
6019                                 goto out_err;
6020                         }
6021                 } else {
6022                         submit_len += bvec->bv_len;
6023                         nr_pages ++;
6024                         bvec++;
6025                 }
6026         }
6027
6028 submit:
6029         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6030                                      csums, async_submit);
6031         if (!ret)
6032                 return 0;
6033
6034         bio_put(bio);
6035 out_err:
6036         dip->errors = 1;
6037         /*
6038          * before atomic variable goto zero, we must
6039          * make sure dip->errors is perceived to be set.
6040          */
6041         smp_mb__before_atomic_dec();
6042         if (atomic_dec_and_test(&dip->pending_bios))
6043                 bio_io_error(dip->orig_bio);
6044
6045         /* bio_end_io() will handle error, so we needn't return it */
6046         return 0;
6047 }
6048
6049 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6050                                 loff_t file_offset)
6051 {
6052         struct btrfs_root *root = BTRFS_I(inode)->root;
6053         struct btrfs_dio_private *dip;
6054         struct bio_vec *bvec = bio->bi_io_vec;
6055         int skip_sum;
6056         int write = rw & REQ_WRITE;
6057         int ret = 0;
6058
6059         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6060
6061         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6062         if (!dip) {
6063                 ret = -ENOMEM;
6064                 goto free_ordered;
6065         }
6066         dip->csums = NULL;
6067
6068         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6069         if (!write && !skip_sum) {
6070                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6071                 if (!dip->csums) {
6072                         kfree(dip);
6073                         ret = -ENOMEM;
6074                         goto free_ordered;
6075                 }
6076         }
6077
6078         dip->private = bio->bi_private;
6079         dip->inode = inode;
6080         dip->logical_offset = file_offset;
6081
6082         dip->bytes = 0;
6083         do {
6084                 dip->bytes += bvec->bv_len;
6085                 bvec++;
6086         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6087
6088         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6089         bio->bi_private = dip;
6090         dip->errors = 0;
6091         dip->orig_bio = bio;
6092         atomic_set(&dip->pending_bios, 0);
6093
6094         if (write)
6095                 bio->bi_end_io = btrfs_endio_direct_write;
6096         else
6097                 bio->bi_end_io = btrfs_endio_direct_read;
6098
6099         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6100         if (!ret)
6101                 return;
6102 free_ordered:
6103         /*
6104          * If this is a write, we need to clean up the reserved space and kill
6105          * the ordered extent.
6106          */
6107         if (write) {
6108                 struct btrfs_ordered_extent *ordered;
6109                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6110                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6111                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6112                         btrfs_free_reserved_extent(root, ordered->start,
6113                                                    ordered->disk_len);
6114                 btrfs_put_ordered_extent(ordered);
6115                 btrfs_put_ordered_extent(ordered);
6116         }
6117         bio_endio(bio, ret);
6118 }
6119
6120 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6121                         const struct iovec *iov, loff_t offset,
6122                         unsigned long nr_segs)
6123 {
6124         int seg;
6125         int i;
6126         size_t size;
6127         unsigned long addr;
6128         unsigned blocksize_mask = root->sectorsize - 1;
6129         ssize_t retval = -EINVAL;
6130         loff_t end = offset;
6131
6132         if (offset & blocksize_mask)
6133                 goto out;
6134
6135         /* Check the memory alignment.  Blocks cannot straddle pages */
6136         for (seg = 0; seg < nr_segs; seg++) {
6137                 addr = (unsigned long)iov[seg].iov_base;
6138                 size = iov[seg].iov_len;
6139                 end += size;
6140                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6141                         goto out;
6142
6143                 /* If this is a write we don't need to check anymore */
6144                 if (rw & WRITE)
6145                         continue;
6146
6147                 /*
6148                  * Check to make sure we don't have duplicate iov_base's in this
6149                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6150                  * when reading back.
6151                  */
6152                 for (i = seg + 1; i < nr_segs; i++) {
6153                         if (iov[seg].iov_base == iov[i].iov_base)
6154                                 goto out;
6155                 }
6156         }
6157         retval = 0;
6158 out:
6159         return retval;
6160 }
6161 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6162                         const struct iovec *iov, loff_t offset,
6163                         unsigned long nr_segs)
6164 {
6165         struct file *file = iocb->ki_filp;
6166         struct inode *inode = file->f_mapping->host;
6167         struct btrfs_ordered_extent *ordered;
6168         struct extent_state *cached_state = NULL;
6169         u64 lockstart, lockend;
6170         ssize_t ret;
6171         int writing = rw & WRITE;
6172         int write_bits = 0;
6173         size_t count = iov_length(iov, nr_segs);
6174
6175         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6176                             offset, nr_segs)) {
6177                 return 0;
6178         }
6179
6180         lockstart = offset;
6181         lockend = offset + count - 1;
6182
6183         if (writing) {
6184                 ret = btrfs_delalloc_reserve_space(inode, count);
6185                 if (ret)
6186                         goto out;
6187         }
6188
6189         while (1) {
6190                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6191                                  0, &cached_state, GFP_NOFS);
6192                 /*
6193                  * We're concerned with the entire range that we're going to be
6194                  * doing DIO to, so we need to make sure theres no ordered
6195                  * extents in this range.
6196                  */
6197                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6198                                                      lockend - lockstart + 1);
6199                 if (!ordered)
6200                         break;
6201                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6202                                      &cached_state, GFP_NOFS);
6203                 btrfs_start_ordered_extent(inode, ordered, 1);
6204                 btrfs_put_ordered_extent(ordered);
6205                 cond_resched();
6206         }
6207
6208         /*
6209          * we don't use btrfs_set_extent_delalloc because we don't want
6210          * the dirty or uptodate bits
6211          */
6212         if (writing) {
6213                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6214                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6215                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6216                                      GFP_NOFS);
6217                 if (ret) {
6218                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6219                                          lockend, EXTENT_LOCKED | write_bits,
6220                                          1, 0, &cached_state, GFP_NOFS);
6221                         goto out;
6222                 }
6223         }
6224
6225         free_extent_state(cached_state);
6226         cached_state = NULL;
6227
6228         ret = __blockdev_direct_IO(rw, iocb, inode,
6229                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6230                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6231                    btrfs_submit_direct, 0);
6232
6233         if (ret < 0 && ret != -EIOCBQUEUED) {
6234                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6235                               offset + iov_length(iov, nr_segs) - 1,
6236                               EXTENT_LOCKED | write_bits, 1, 0,
6237                               &cached_state, GFP_NOFS);
6238         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6239                 /*
6240                  * We're falling back to buffered, unlock the section we didn't
6241                  * do IO on.
6242                  */
6243                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6244                               offset + iov_length(iov, nr_segs) - 1,
6245                               EXTENT_LOCKED | write_bits, 1, 0,
6246                               &cached_state, GFP_NOFS);
6247         }
6248 out:
6249         free_extent_state(cached_state);
6250         return ret;
6251 }
6252
6253 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6254                 __u64 start, __u64 len)
6255 {
6256         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6257 }
6258
6259 int btrfs_readpage(struct file *file, struct page *page)
6260 {
6261         struct extent_io_tree *tree;
6262         tree = &BTRFS_I(page->mapping->host)->io_tree;
6263         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6264 }
6265
6266 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6267 {
6268         struct extent_io_tree *tree;
6269
6270
6271         if (current->flags & PF_MEMALLOC) {
6272                 redirty_page_for_writepage(wbc, page);
6273                 unlock_page(page);
6274                 return 0;
6275         }
6276         tree = &BTRFS_I(page->mapping->host)->io_tree;
6277         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6278 }
6279
6280 int btrfs_writepages(struct address_space *mapping,
6281                      struct writeback_control *wbc)
6282 {
6283         struct extent_io_tree *tree;
6284
6285         tree = &BTRFS_I(mapping->host)->io_tree;
6286         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6287 }
6288
6289 static int
6290 btrfs_readpages(struct file *file, struct address_space *mapping,
6291                 struct list_head *pages, unsigned nr_pages)
6292 {
6293         struct extent_io_tree *tree;
6294         tree = &BTRFS_I(mapping->host)->io_tree;
6295         return extent_readpages(tree, mapping, pages, nr_pages,
6296                                 btrfs_get_extent);
6297 }
6298 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6299 {
6300         struct extent_io_tree *tree;
6301         struct extent_map_tree *map;
6302         int ret;
6303
6304         tree = &BTRFS_I(page->mapping->host)->io_tree;
6305         map = &BTRFS_I(page->mapping->host)->extent_tree;
6306         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6307         if (ret == 1) {
6308                 ClearPagePrivate(page);
6309                 set_page_private(page, 0);
6310                 page_cache_release(page);
6311         }
6312         return ret;
6313 }
6314
6315 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6316 {
6317         if (PageWriteback(page) || PageDirty(page))
6318                 return 0;
6319         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6320 }
6321
6322 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6323 {
6324         struct extent_io_tree *tree;
6325         struct btrfs_ordered_extent *ordered;
6326         struct extent_state *cached_state = NULL;
6327         u64 page_start = page_offset(page);
6328         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6329
6330
6331         /*
6332          * we have the page locked, so new writeback can't start,
6333          * and the dirty bit won't be cleared while we are here.
6334          *
6335          * Wait for IO on this page so that we can safely clear
6336          * the PagePrivate2 bit and do ordered accounting
6337          */
6338         wait_on_page_writeback(page);
6339
6340         tree = &BTRFS_I(page->mapping->host)->io_tree;
6341         if (offset) {
6342                 btrfs_releasepage(page, GFP_NOFS);
6343                 return;
6344         }
6345         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6346                          GFP_NOFS);
6347         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6348                                            page_offset(page));
6349         if (ordered) {
6350                 /*
6351                  * IO on this page will never be started, so we need
6352                  * to account for any ordered extents now
6353                  */
6354                 clear_extent_bit(tree, page_start, page_end,
6355                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6356                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6357                                  &cached_state, GFP_NOFS);
6358                 /*
6359                  * whoever cleared the private bit is responsible
6360                  * for the finish_ordered_io
6361                  */
6362                 if (TestClearPagePrivate2(page)) {
6363                         btrfs_finish_ordered_io(page->mapping->host,
6364                                                 page_start, page_end);
6365                 }
6366                 btrfs_put_ordered_extent(ordered);
6367                 cached_state = NULL;
6368                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6369                                  GFP_NOFS);
6370         }
6371         clear_extent_bit(tree, page_start, page_end,
6372                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6373                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6374         __btrfs_releasepage(page, GFP_NOFS);
6375
6376         ClearPageChecked(page);
6377         if (PagePrivate(page)) {
6378                 ClearPagePrivate(page);
6379                 set_page_private(page, 0);
6380                 page_cache_release(page);
6381         }
6382 }
6383
6384 /*
6385  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6386  * called from a page fault handler when a page is first dirtied. Hence we must
6387  * be careful to check for EOF conditions here. We set the page up correctly
6388  * for a written page which means we get ENOSPC checking when writing into
6389  * holes and correct delalloc and unwritten extent mapping on filesystems that
6390  * support these features.
6391  *
6392  * We are not allowed to take the i_mutex here so we have to play games to
6393  * protect against truncate races as the page could now be beyond EOF.  Because
6394  * vmtruncate() writes the inode size before removing pages, once we have the
6395  * page lock we can determine safely if the page is beyond EOF. If it is not
6396  * beyond EOF, then the page is guaranteed safe against truncation until we
6397  * unlock the page.
6398  */
6399 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6400 {
6401         struct page *page = vmf->page;
6402         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6403         struct btrfs_root *root = BTRFS_I(inode)->root;
6404         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6405         struct btrfs_ordered_extent *ordered;
6406         struct extent_state *cached_state = NULL;
6407         char *kaddr;
6408         unsigned long zero_start;
6409         loff_t size;
6410         int ret;
6411         u64 page_start;
6412         u64 page_end;
6413
6414         /* Need this to keep space reservations serialized */
6415         mutex_lock(&inode->i_mutex);
6416         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6417         mutex_unlock(&inode->i_mutex);
6418         if (!ret)
6419                 ret = btrfs_update_time(vma->vm_file);
6420         if (ret) {
6421                 if (ret == -ENOMEM)
6422                         ret = VM_FAULT_OOM;
6423                 else /* -ENOSPC, -EIO, etc */
6424                         ret = VM_FAULT_SIGBUS;
6425                 goto out;
6426         }
6427
6428         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6429 again:
6430         lock_page(page);
6431         size = i_size_read(inode);
6432         page_start = page_offset(page);
6433         page_end = page_start + PAGE_CACHE_SIZE - 1;
6434
6435         if ((page->mapping != inode->i_mapping) ||
6436             (page_start >= size)) {
6437                 /* page got truncated out from underneath us */
6438                 goto out_unlock;
6439         }
6440         wait_on_page_writeback(page);
6441
6442         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6443                          GFP_NOFS);
6444         set_page_extent_mapped(page);
6445
6446         /*
6447          * we can't set the delalloc bits if there are pending ordered
6448          * extents.  Drop our locks and wait for them to finish
6449          */
6450         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6451         if (ordered) {
6452                 unlock_extent_cached(io_tree, page_start, page_end,
6453                                      &cached_state, GFP_NOFS);
6454                 unlock_page(page);
6455                 btrfs_start_ordered_extent(inode, ordered, 1);
6456                 btrfs_put_ordered_extent(ordered);
6457                 goto again;
6458         }
6459
6460         /*
6461          * XXX - page_mkwrite gets called every time the page is dirtied, even
6462          * if it was already dirty, so for space accounting reasons we need to
6463          * clear any delalloc bits for the range we are fixing to save.  There
6464          * is probably a better way to do this, but for now keep consistent with
6465          * prepare_pages in the normal write path.
6466          */
6467         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6468                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6469                           0, 0, &cached_state, GFP_NOFS);
6470
6471         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6472                                         &cached_state);
6473         if (ret) {
6474                 unlock_extent_cached(io_tree, page_start, page_end,
6475                                      &cached_state, GFP_NOFS);
6476                 ret = VM_FAULT_SIGBUS;
6477                 goto out_unlock;
6478         }
6479         ret = 0;
6480
6481         /* page is wholly or partially inside EOF */
6482         if (page_start + PAGE_CACHE_SIZE > size)
6483                 zero_start = size & ~PAGE_CACHE_MASK;
6484         else
6485                 zero_start = PAGE_CACHE_SIZE;
6486
6487         if (zero_start != PAGE_CACHE_SIZE) {
6488                 kaddr = kmap(page);
6489                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6490                 flush_dcache_page(page);
6491                 kunmap(page);
6492         }
6493         ClearPageChecked(page);
6494         set_page_dirty(page);
6495         SetPageUptodate(page);
6496
6497         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6498         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6499
6500         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6501
6502 out_unlock:
6503         if (!ret)
6504                 return VM_FAULT_LOCKED;
6505         unlock_page(page);
6506 out:
6507         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6508         return ret;
6509 }
6510
6511 static int btrfs_truncate(struct inode *inode)
6512 {
6513         struct btrfs_root *root = BTRFS_I(inode)->root;
6514         struct btrfs_block_rsv *rsv;
6515         int ret;
6516         int err = 0;
6517         struct btrfs_trans_handle *trans;
6518         unsigned long nr;
6519         u64 mask = root->sectorsize - 1;
6520         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6521
6522         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6523         if (ret)
6524                 return ret;
6525
6526         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6527         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6528
6529         /*
6530          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6531          * 3 things going on here
6532          *
6533          * 1) We need to reserve space for our orphan item and the space to
6534          * delete our orphan item.  Lord knows we don't want to have a dangling
6535          * orphan item because we didn't reserve space to remove it.
6536          *
6537          * 2) We need to reserve space to update our inode.
6538          *
6539          * 3) We need to have something to cache all the space that is going to
6540          * be free'd up by the truncate operation, but also have some slack
6541          * space reserved in case it uses space during the truncate (thank you
6542          * very much snapshotting).
6543          *
6544          * And we need these to all be seperate.  The fact is we can use alot of
6545          * space doing the truncate, and we have no earthly idea how much space
6546          * we will use, so we need the truncate reservation to be seperate so it
6547          * doesn't end up using space reserved for updating the inode or
6548          * removing the orphan item.  We also need to be able to stop the
6549          * transaction and start a new one, which means we need to be able to
6550          * update the inode several times, and we have no idea of knowing how
6551          * many times that will be, so we can't just reserve 1 item for the
6552          * entirety of the opration, so that has to be done seperately as well.
6553          * Then there is the orphan item, which does indeed need to be held on
6554          * to for the whole operation, and we need nobody to touch this reserved
6555          * space except the orphan code.
6556          *
6557          * So that leaves us with
6558          *
6559          * 1) root->orphan_block_rsv - for the orphan deletion.
6560          * 2) rsv - for the truncate reservation, which we will steal from the
6561          * transaction reservation.
6562          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6563          * updating the inode.
6564          */
6565         rsv = btrfs_alloc_block_rsv(root);
6566         if (!rsv)
6567                 return -ENOMEM;
6568         rsv->size = min_size;
6569
6570         /*
6571          * 1 for the truncate slack space
6572          * 1 for the orphan item we're going to add
6573          * 1 for the orphan item deletion
6574          * 1 for updating the inode.
6575          */
6576         trans = btrfs_start_transaction(root, 4);
6577         if (IS_ERR(trans)) {
6578                 err = PTR_ERR(trans);
6579                 goto out;
6580         }
6581
6582         /* Migrate the slack space for the truncate to our reserve */
6583         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6584                                       min_size);
6585         BUG_ON(ret);
6586
6587         ret = btrfs_orphan_add(trans, inode);
6588         if (ret) {
6589                 btrfs_end_transaction(trans, root);
6590                 goto out;
6591         }
6592
6593         /*
6594          * setattr is responsible for setting the ordered_data_close flag,
6595          * but that is only tested during the last file release.  That
6596          * could happen well after the next commit, leaving a great big
6597          * window where new writes may get lost if someone chooses to write
6598          * to this file after truncating to zero
6599          *
6600          * The inode doesn't have any dirty data here, and so if we commit
6601          * this is a noop.  If someone immediately starts writing to the inode
6602          * it is very likely we'll catch some of their writes in this
6603          * transaction, and the commit will find this file on the ordered
6604          * data list with good things to send down.
6605          *
6606          * This is a best effort solution, there is still a window where
6607          * using truncate to replace the contents of the file will
6608          * end up with a zero length file after a crash.
6609          */
6610         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6611                 btrfs_add_ordered_operation(trans, root, inode);
6612
6613         while (1) {
6614                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6615                 if (ret) {
6616                         /*
6617                          * This can only happen with the original transaction we
6618                          * started above, every other time we shouldn't have a
6619                          * transaction started yet.
6620                          */
6621                         if (ret == -EAGAIN)
6622                                 goto end_trans;
6623                         err = ret;
6624                         break;
6625                 }
6626
6627                 if (!trans) {
6628                         /* Just need the 1 for updating the inode */
6629                         trans = btrfs_start_transaction(root, 1);
6630                         if (IS_ERR(trans)) {
6631                                 ret = err = PTR_ERR(trans);
6632                                 trans = NULL;
6633                                 break;
6634                         }
6635                 }
6636
6637                 trans->block_rsv = rsv;
6638
6639                 ret = btrfs_truncate_inode_items(trans, root, inode,
6640                                                  inode->i_size,
6641                                                  BTRFS_EXTENT_DATA_KEY);
6642                 if (ret != -EAGAIN) {
6643                         err = ret;
6644                         break;
6645                 }
6646
6647                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6648                 ret = btrfs_update_inode(trans, root, inode);
6649                 if (ret) {
6650                         err = ret;
6651                         break;
6652                 }
6653 end_trans:
6654                 nr = trans->blocks_used;
6655                 btrfs_end_transaction(trans, root);
6656                 trans = NULL;
6657                 btrfs_btree_balance_dirty(root, nr);
6658         }
6659
6660         if (ret == 0 && inode->i_nlink > 0) {
6661                 trans->block_rsv = root->orphan_block_rsv;
6662                 ret = btrfs_orphan_del(trans, inode);
6663                 if (ret)
6664                         err = ret;
6665         } else if (ret && inode->i_nlink > 0) {
6666                 /*
6667                  * Failed to do the truncate, remove us from the in memory
6668                  * orphan list.
6669                  */
6670                 ret = btrfs_orphan_del(NULL, inode);
6671         }
6672
6673         if (trans) {
6674                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6675                 ret = btrfs_update_inode(trans, root, inode);
6676                 if (ret && !err)
6677                         err = ret;
6678
6679                 nr = trans->blocks_used;
6680                 ret = btrfs_end_transaction(trans, root);
6681                 btrfs_btree_balance_dirty(root, nr);
6682         }
6683
6684 out:
6685         btrfs_free_block_rsv(root, rsv);
6686
6687         if (ret && !err)
6688                 err = ret;
6689
6690         return err;
6691 }
6692
6693 /*
6694  * create a new subvolume directory/inode (helper for the ioctl).
6695  */
6696 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6697                              struct btrfs_root *new_root, u64 new_dirid)
6698 {
6699         struct inode *inode;
6700         int err;
6701         u64 index = 0;
6702
6703         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6704                                 new_dirid, S_IFDIR | 0700, &index);
6705         if (IS_ERR(inode))
6706                 return PTR_ERR(inode);
6707         inode->i_op = &btrfs_dir_inode_operations;
6708         inode->i_fop = &btrfs_dir_file_operations;
6709
6710         set_nlink(inode, 1);
6711         btrfs_i_size_write(inode, 0);
6712
6713         err = btrfs_update_inode(trans, new_root, inode);
6714         BUG_ON(err);
6715
6716         iput(inode);
6717         return 0;
6718 }
6719
6720 struct inode *btrfs_alloc_inode(struct super_block *sb)
6721 {
6722         struct btrfs_inode *ei;
6723         struct inode *inode;
6724
6725         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6726         if (!ei)
6727                 return NULL;
6728
6729         ei->root = NULL;
6730         ei->space_info = NULL;
6731         ei->generation = 0;
6732         ei->sequence = 0;
6733         ei->last_trans = 0;
6734         ei->last_sub_trans = 0;
6735         ei->logged_trans = 0;
6736         ei->delalloc_bytes = 0;
6737         ei->disk_i_size = 0;
6738         ei->flags = 0;
6739         ei->csum_bytes = 0;
6740         ei->index_cnt = (u64)-1;
6741         ei->last_unlink_trans = 0;
6742
6743         spin_lock_init(&ei->lock);
6744         ei->outstanding_extents = 0;
6745         ei->reserved_extents = 0;
6746
6747         ei->ordered_data_close = 0;
6748         ei->orphan_meta_reserved = 0;
6749         ei->dummy_inode = 0;
6750         ei->in_defrag = 0;
6751         ei->delalloc_meta_reserved = 0;
6752         ei->force_compress = BTRFS_COMPRESS_NONE;
6753
6754         ei->delayed_node = NULL;
6755
6756         inode = &ei->vfs_inode;
6757         extent_map_tree_init(&ei->extent_tree);
6758         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6759         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6760         mutex_init(&ei->log_mutex);
6761         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6762         INIT_LIST_HEAD(&ei->i_orphan);
6763         INIT_LIST_HEAD(&ei->delalloc_inodes);
6764         INIT_LIST_HEAD(&ei->ordered_operations);
6765         RB_CLEAR_NODE(&ei->rb_node);
6766
6767         return inode;
6768 }
6769
6770 static void btrfs_i_callback(struct rcu_head *head)
6771 {
6772         struct inode *inode = container_of(head, struct inode, i_rcu);
6773         INIT_LIST_HEAD(&inode->i_dentry);
6774         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6775 }
6776
6777 void btrfs_destroy_inode(struct inode *inode)
6778 {
6779         struct btrfs_ordered_extent *ordered;
6780         struct btrfs_root *root = BTRFS_I(inode)->root;
6781
6782         WARN_ON(!list_empty(&inode->i_dentry));
6783         WARN_ON(inode->i_data.nrpages);
6784         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6785         WARN_ON(BTRFS_I(inode)->reserved_extents);
6786         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6787         WARN_ON(BTRFS_I(inode)->csum_bytes);
6788
6789         /*
6790          * This can happen where we create an inode, but somebody else also
6791          * created the same inode and we need to destroy the one we already
6792          * created.
6793          */
6794         if (!root)
6795                 goto free;
6796
6797         /*
6798          * Make sure we're properly removed from the ordered operation
6799          * lists.
6800          */
6801         smp_mb();
6802         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6803                 spin_lock(&root->fs_info->ordered_extent_lock);
6804                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6805                 spin_unlock(&root->fs_info->ordered_extent_lock);
6806         }
6807
6808         spin_lock(&root->orphan_lock);
6809         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6810                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6811                        (unsigned long long)btrfs_ino(inode));
6812                 list_del_init(&BTRFS_I(inode)->i_orphan);
6813         }
6814         spin_unlock(&root->orphan_lock);
6815
6816         while (1) {
6817                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6818                 if (!ordered)
6819                         break;
6820                 else {
6821                         printk(KERN_ERR "btrfs found ordered "
6822                                "extent %llu %llu on inode cleanup\n",
6823                                (unsigned long long)ordered->file_offset,
6824                                (unsigned long long)ordered->len);
6825                         btrfs_remove_ordered_extent(inode, ordered);
6826                         btrfs_put_ordered_extent(ordered);
6827                         btrfs_put_ordered_extent(ordered);
6828                 }
6829         }
6830         inode_tree_del(inode);
6831         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6832 free:
6833         btrfs_remove_delayed_node(inode);
6834         call_rcu(&inode->i_rcu, btrfs_i_callback);
6835 }
6836
6837 int btrfs_drop_inode(struct inode *inode)
6838 {
6839         struct btrfs_root *root = BTRFS_I(inode)->root;
6840
6841         if (btrfs_root_refs(&root->root_item) == 0 &&
6842             !btrfs_is_free_space_inode(root, inode))
6843                 return 1;
6844         else
6845                 return generic_drop_inode(inode);
6846 }
6847
6848 static void init_once(void *foo)
6849 {
6850         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6851
6852         inode_init_once(&ei->vfs_inode);
6853 }
6854
6855 void btrfs_destroy_cachep(void)
6856 {
6857         if (btrfs_inode_cachep)
6858                 kmem_cache_destroy(btrfs_inode_cachep);
6859         if (btrfs_trans_handle_cachep)
6860                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6861         if (btrfs_transaction_cachep)
6862                 kmem_cache_destroy(btrfs_transaction_cachep);
6863         if (btrfs_path_cachep)
6864                 kmem_cache_destroy(btrfs_path_cachep);
6865         if (btrfs_free_space_cachep)
6866                 kmem_cache_destroy(btrfs_free_space_cachep);
6867 }
6868
6869 int btrfs_init_cachep(void)
6870 {
6871         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6872                         sizeof(struct btrfs_inode), 0,
6873                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6874         if (!btrfs_inode_cachep)
6875                 goto fail;
6876
6877         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6878                         sizeof(struct btrfs_trans_handle), 0,
6879                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6880         if (!btrfs_trans_handle_cachep)
6881                 goto fail;
6882
6883         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6884                         sizeof(struct btrfs_transaction), 0,
6885                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6886         if (!btrfs_transaction_cachep)
6887                 goto fail;
6888
6889         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6890                         sizeof(struct btrfs_path), 0,
6891                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6892         if (!btrfs_path_cachep)
6893                 goto fail;
6894
6895         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6896                         sizeof(struct btrfs_free_space), 0,
6897                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6898         if (!btrfs_free_space_cachep)
6899                 goto fail;
6900
6901         return 0;
6902 fail:
6903         btrfs_destroy_cachep();
6904         return -ENOMEM;
6905 }
6906
6907 static int btrfs_getattr(struct vfsmount *mnt,
6908                          struct dentry *dentry, struct kstat *stat)
6909 {
6910         struct inode *inode = dentry->d_inode;
6911         u32 blocksize = inode->i_sb->s_blocksize;
6912
6913         generic_fillattr(inode, stat);
6914         stat->dev = BTRFS_I(inode)->root->anon_dev;
6915         stat->blksize = PAGE_CACHE_SIZE;
6916         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6917                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6918         return 0;
6919 }
6920
6921 /*
6922  * If a file is moved, it will inherit the cow and compression flags of the new
6923  * directory.
6924  */
6925 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6926 {
6927         struct btrfs_inode *b_dir = BTRFS_I(dir);
6928         struct btrfs_inode *b_inode = BTRFS_I(inode);
6929
6930         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6931                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6932         else
6933                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6934
6935         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6936                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6937         else
6938                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6939 }
6940
6941 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6942                            struct inode *new_dir, struct dentry *new_dentry)
6943 {
6944         struct btrfs_trans_handle *trans;
6945         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6946         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6947         struct inode *new_inode = new_dentry->d_inode;
6948         struct inode *old_inode = old_dentry->d_inode;
6949         struct timespec ctime = CURRENT_TIME;
6950         u64 index = 0;
6951         u64 root_objectid;
6952         int ret;
6953         u64 old_ino = btrfs_ino(old_inode);
6954
6955         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6956                 return -EPERM;
6957
6958         /* we only allow rename subvolume link between subvolumes */
6959         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6960                 return -EXDEV;
6961
6962         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6963             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6964                 return -ENOTEMPTY;
6965
6966         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6967             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6968                 return -ENOTEMPTY;
6969         /*
6970          * we're using rename to replace one file with another.
6971          * and the replacement file is large.  Start IO on it now so
6972          * we don't add too much work to the end of the transaction
6973          */
6974         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6975             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6976                 filemap_flush(old_inode->i_mapping);
6977
6978         /* close the racy window with snapshot create/destroy ioctl */
6979         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6980                 down_read(&root->fs_info->subvol_sem);
6981         /*
6982          * We want to reserve the absolute worst case amount of items.  So if
6983          * both inodes are subvols and we need to unlink them then that would
6984          * require 4 item modifications, but if they are both normal inodes it
6985          * would require 5 item modifications, so we'll assume their normal
6986          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6987          * should cover the worst case number of items we'll modify.
6988          */
6989         trans = btrfs_start_transaction(root, 20);
6990         if (IS_ERR(trans)) {
6991                 ret = PTR_ERR(trans);
6992                 goto out_notrans;
6993         }
6994
6995         if (dest != root)
6996                 btrfs_record_root_in_trans(trans, dest);
6997
6998         ret = btrfs_set_inode_index(new_dir, &index);
6999         if (ret)
7000                 goto out_fail;
7001
7002         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7003                 /* force full log commit if subvolume involved. */
7004                 root->fs_info->last_trans_log_full_commit = trans->transid;
7005         } else {
7006                 ret = btrfs_insert_inode_ref(trans, dest,
7007                                              new_dentry->d_name.name,
7008                                              new_dentry->d_name.len,
7009                                              old_ino,
7010                                              btrfs_ino(new_dir), index);
7011                 if (ret)
7012                         goto out_fail;
7013                 /*
7014                  * this is an ugly little race, but the rename is required
7015                  * to make sure that if we crash, the inode is either at the
7016                  * old name or the new one.  pinning the log transaction lets
7017                  * us make sure we don't allow a log commit to come in after
7018                  * we unlink the name but before we add the new name back in.
7019                  */
7020                 btrfs_pin_log_trans(root);
7021         }
7022         /*
7023          * make sure the inode gets flushed if it is replacing
7024          * something.
7025          */
7026         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7027                 btrfs_add_ordered_operation(trans, root, old_inode);
7028
7029         old_dir->i_ctime = old_dir->i_mtime = ctime;
7030         new_dir->i_ctime = new_dir->i_mtime = ctime;
7031         old_inode->i_ctime = ctime;
7032
7033         if (old_dentry->d_parent != new_dentry->d_parent)
7034                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7035
7036         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7037                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7038                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7039                                         old_dentry->d_name.name,
7040                                         old_dentry->d_name.len);
7041         } else {
7042                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7043                                         old_dentry->d_inode,
7044                                         old_dentry->d_name.name,
7045                                         old_dentry->d_name.len);
7046                 if (!ret)
7047                         ret = btrfs_update_inode(trans, root, old_inode);
7048         }
7049         BUG_ON(ret);
7050
7051         if (new_inode) {
7052                 new_inode->i_ctime = CURRENT_TIME;
7053                 if (unlikely(btrfs_ino(new_inode) ==
7054                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7055                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7056                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7057                                                 root_objectid,
7058                                                 new_dentry->d_name.name,
7059                                                 new_dentry->d_name.len);
7060                         BUG_ON(new_inode->i_nlink == 0);
7061                 } else {
7062                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7063                                                  new_dentry->d_inode,
7064                                                  new_dentry->d_name.name,
7065                                                  new_dentry->d_name.len);
7066                 }
7067                 BUG_ON(ret);
7068                 if (new_inode->i_nlink == 0) {
7069                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7070                         BUG_ON(ret);
7071                 }
7072         }
7073
7074         fixup_inode_flags(new_dir, old_inode);
7075
7076         ret = btrfs_add_link(trans, new_dir, old_inode,
7077                              new_dentry->d_name.name,
7078                              new_dentry->d_name.len, 0, index);
7079         BUG_ON(ret);
7080
7081         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7082                 struct dentry *parent = new_dentry->d_parent;
7083                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7084                 btrfs_end_log_trans(root);
7085         }
7086 out_fail:
7087         btrfs_end_transaction(trans, root);
7088 out_notrans:
7089         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7090                 up_read(&root->fs_info->subvol_sem);
7091
7092         return ret;
7093 }
7094
7095 /*
7096  * some fairly slow code that needs optimization. This walks the list
7097  * of all the inodes with pending delalloc and forces them to disk.
7098  */
7099 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7100 {
7101         struct list_head *head = &root->fs_info->delalloc_inodes;
7102         struct btrfs_inode *binode;
7103         struct inode *inode;
7104
7105         if (root->fs_info->sb->s_flags & MS_RDONLY)
7106                 return -EROFS;
7107
7108         spin_lock(&root->fs_info->delalloc_lock);
7109         while (!list_empty(head)) {
7110                 binode = list_entry(head->next, struct btrfs_inode,
7111                                     delalloc_inodes);
7112                 inode = igrab(&binode->vfs_inode);
7113                 if (!inode)
7114                         list_del_init(&binode->delalloc_inodes);
7115                 spin_unlock(&root->fs_info->delalloc_lock);
7116                 if (inode) {
7117                         filemap_flush(inode->i_mapping);
7118                         if (delay_iput)
7119                                 btrfs_add_delayed_iput(inode);
7120                         else
7121                                 iput(inode);
7122                 }
7123                 cond_resched();
7124                 spin_lock(&root->fs_info->delalloc_lock);
7125         }
7126         spin_unlock(&root->fs_info->delalloc_lock);
7127
7128         /* the filemap_flush will queue IO into the worker threads, but
7129          * we have to make sure the IO is actually started and that
7130          * ordered extents get created before we return
7131          */
7132         atomic_inc(&root->fs_info->async_submit_draining);
7133         while (atomic_read(&root->fs_info->nr_async_submits) ||
7134               atomic_read(&root->fs_info->async_delalloc_pages)) {
7135                 wait_event(root->fs_info->async_submit_wait,
7136                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7137                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7138         }
7139         atomic_dec(&root->fs_info->async_submit_draining);
7140         return 0;
7141 }
7142
7143 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7144                          const char *symname)
7145 {
7146         struct btrfs_trans_handle *trans;
7147         struct btrfs_root *root = BTRFS_I(dir)->root;
7148         struct btrfs_path *path;
7149         struct btrfs_key key;
7150         struct inode *inode = NULL;
7151         int err;
7152         int drop_inode = 0;
7153         u64 objectid;
7154         u64 index = 0 ;
7155         int name_len;
7156         int datasize;
7157         unsigned long ptr;
7158         struct btrfs_file_extent_item *ei;
7159         struct extent_buffer *leaf;
7160         unsigned long nr = 0;
7161
7162         name_len = strlen(symname) + 1;
7163         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7164                 return -ENAMETOOLONG;
7165
7166         /*
7167          * 2 items for inode item and ref
7168          * 2 items for dir items
7169          * 1 item for xattr if selinux is on
7170          */
7171         trans = btrfs_start_transaction(root, 5);
7172         if (IS_ERR(trans))
7173                 return PTR_ERR(trans);
7174
7175         err = btrfs_find_free_ino(root, &objectid);
7176         if (err)
7177                 goto out_unlock;
7178
7179         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7180                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7181                                 S_IFLNK|S_IRWXUGO, &index);
7182         if (IS_ERR(inode)) {
7183                 err = PTR_ERR(inode);
7184                 goto out_unlock;
7185         }
7186
7187         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7188         if (err) {
7189                 drop_inode = 1;
7190                 goto out_unlock;
7191         }
7192
7193         /*
7194         * If the active LSM wants to access the inode during
7195         * d_instantiate it needs these. Smack checks to see
7196         * if the filesystem supports xattrs by looking at the
7197         * ops vector.
7198         */
7199         inode->i_fop = &btrfs_file_operations;
7200         inode->i_op = &btrfs_file_inode_operations;
7201
7202         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7203         if (err)
7204                 drop_inode = 1;
7205         else {
7206                 inode->i_mapping->a_ops = &btrfs_aops;
7207                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7208                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7209         }
7210         if (drop_inode)
7211                 goto out_unlock;
7212
7213         path = btrfs_alloc_path();
7214         if (!path) {
7215                 err = -ENOMEM;
7216                 drop_inode = 1;
7217                 goto out_unlock;
7218         }
7219         key.objectid = btrfs_ino(inode);
7220         key.offset = 0;
7221         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7222         datasize = btrfs_file_extent_calc_inline_size(name_len);
7223         err = btrfs_insert_empty_item(trans, root, path, &key,
7224                                       datasize);
7225         if (err) {
7226                 drop_inode = 1;
7227                 btrfs_free_path(path);
7228                 goto out_unlock;
7229         }
7230         leaf = path->nodes[0];
7231         ei = btrfs_item_ptr(leaf, path->slots[0],
7232                             struct btrfs_file_extent_item);
7233         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7234         btrfs_set_file_extent_type(leaf, ei,
7235                                    BTRFS_FILE_EXTENT_INLINE);
7236         btrfs_set_file_extent_encryption(leaf, ei, 0);
7237         btrfs_set_file_extent_compression(leaf, ei, 0);
7238         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7239         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7240
7241         ptr = btrfs_file_extent_inline_start(ei);
7242         write_extent_buffer(leaf, symname, ptr, name_len);
7243         btrfs_mark_buffer_dirty(leaf);
7244         btrfs_free_path(path);
7245
7246         inode->i_op = &btrfs_symlink_inode_operations;
7247         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7248         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7249         inode_set_bytes(inode, name_len);
7250         btrfs_i_size_write(inode, name_len - 1);
7251         err = btrfs_update_inode(trans, root, inode);
7252         if (err)
7253                 drop_inode = 1;
7254
7255 out_unlock:
7256         if (!err)
7257                 d_instantiate(dentry, inode);
7258         nr = trans->blocks_used;
7259         btrfs_end_transaction(trans, root);
7260         if (drop_inode) {
7261                 inode_dec_link_count(inode);
7262                 iput(inode);
7263         }
7264         btrfs_btree_balance_dirty(root, nr);
7265         return err;
7266 }
7267
7268 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7269                                        u64 start, u64 num_bytes, u64 min_size,
7270                                        loff_t actual_len, u64 *alloc_hint,
7271                                        struct btrfs_trans_handle *trans)
7272 {
7273         struct btrfs_root *root = BTRFS_I(inode)->root;
7274         struct btrfs_key ins;
7275         u64 cur_offset = start;
7276         u64 i_size;
7277         int ret = 0;
7278         bool own_trans = true;
7279
7280         if (trans)
7281                 own_trans = false;
7282         while (num_bytes > 0) {
7283                 if (own_trans) {
7284                         trans = btrfs_start_transaction(root, 3);
7285                         if (IS_ERR(trans)) {
7286                                 ret = PTR_ERR(trans);
7287                                 break;
7288                         }
7289                 }
7290
7291                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7292                                            0, *alloc_hint, (u64)-1, &ins, 1);
7293                 if (ret) {
7294                         if (own_trans)
7295                                 btrfs_end_transaction(trans, root);
7296                         break;
7297                 }
7298
7299                 ret = insert_reserved_file_extent(trans, inode,
7300                                                   cur_offset, ins.objectid,
7301                                                   ins.offset, ins.offset,
7302                                                   ins.offset, 0, 0, 0,
7303                                                   BTRFS_FILE_EXTENT_PREALLOC);
7304                 BUG_ON(ret);
7305                 btrfs_drop_extent_cache(inode, cur_offset,
7306                                         cur_offset + ins.offset -1, 0);
7307
7308                 num_bytes -= ins.offset;
7309                 cur_offset += ins.offset;
7310                 *alloc_hint = ins.objectid + ins.offset;
7311
7312                 inode->i_ctime = CURRENT_TIME;
7313                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7314                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7315                     (actual_len > inode->i_size) &&
7316                     (cur_offset > inode->i_size)) {
7317                         if (cur_offset > actual_len)
7318                                 i_size = actual_len;
7319                         else
7320                                 i_size = cur_offset;
7321                         i_size_write(inode, i_size);
7322                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7323                 }
7324
7325                 ret = btrfs_update_inode(trans, root, inode);
7326                 BUG_ON(ret);
7327
7328                 if (own_trans)
7329                         btrfs_end_transaction(trans, root);
7330         }
7331         return ret;
7332 }
7333
7334 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7335                               u64 start, u64 num_bytes, u64 min_size,
7336                               loff_t actual_len, u64 *alloc_hint)
7337 {
7338         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7339                                            min_size, actual_len, alloc_hint,
7340                                            NULL);
7341 }
7342
7343 int btrfs_prealloc_file_range_trans(struct inode *inode,
7344                                     struct btrfs_trans_handle *trans, int mode,
7345                                     u64 start, u64 num_bytes, u64 min_size,
7346                                     loff_t actual_len, u64 *alloc_hint)
7347 {
7348         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7349                                            min_size, actual_len, alloc_hint, trans);
7350 }
7351
7352 static int btrfs_set_page_dirty(struct page *page)
7353 {
7354         return __set_page_dirty_nobuffers(page);
7355 }
7356
7357 static int btrfs_permission(struct inode *inode, int mask)
7358 {
7359         struct btrfs_root *root = BTRFS_I(inode)->root;
7360         umode_t mode = inode->i_mode;
7361
7362         if (mask & MAY_WRITE &&
7363             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7364                 if (btrfs_root_readonly(root))
7365                         return -EROFS;
7366                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7367                         return -EACCES;
7368         }
7369         return generic_permission(inode, mask);
7370 }
7371
7372 static const struct inode_operations btrfs_dir_inode_operations = {
7373         .getattr        = btrfs_getattr,
7374         .lookup         = btrfs_lookup,
7375         .create         = btrfs_create,
7376         .unlink         = btrfs_unlink,
7377         .link           = btrfs_link,
7378         .mkdir          = btrfs_mkdir,
7379         .rmdir          = btrfs_rmdir,
7380         .rename         = btrfs_rename,
7381         .symlink        = btrfs_symlink,
7382         .setattr        = btrfs_setattr,
7383         .mknod          = btrfs_mknod,
7384         .setxattr       = btrfs_setxattr,
7385         .getxattr       = btrfs_getxattr,
7386         .listxattr      = btrfs_listxattr,
7387         .removexattr    = btrfs_removexattr,
7388         .permission     = btrfs_permission,
7389         .get_acl        = btrfs_get_acl,
7390 };
7391 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7392         .lookup         = btrfs_lookup,
7393         .permission     = btrfs_permission,
7394         .get_acl        = btrfs_get_acl,
7395 };
7396
7397 static const struct file_operations btrfs_dir_file_operations = {
7398         .llseek         = generic_file_llseek,
7399         .read           = generic_read_dir,
7400         .readdir        = btrfs_real_readdir,
7401         .unlocked_ioctl = btrfs_ioctl,
7402 #ifdef CONFIG_COMPAT
7403         .compat_ioctl   = btrfs_ioctl,
7404 #endif
7405         .release        = btrfs_release_file,
7406         .fsync          = btrfs_sync_file,
7407 };
7408
7409 static struct extent_io_ops btrfs_extent_io_ops = {
7410         .fill_delalloc = run_delalloc_range,
7411         .submit_bio_hook = btrfs_submit_bio_hook,
7412         .merge_bio_hook = btrfs_merge_bio_hook,
7413         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7414         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7415         .writepage_start_hook = btrfs_writepage_start_hook,
7416         .set_bit_hook = btrfs_set_bit_hook,
7417         .clear_bit_hook = btrfs_clear_bit_hook,
7418         .merge_extent_hook = btrfs_merge_extent_hook,
7419         .split_extent_hook = btrfs_split_extent_hook,
7420 };
7421
7422 /*
7423  * btrfs doesn't support the bmap operation because swapfiles
7424  * use bmap to make a mapping of extents in the file.  They assume
7425  * these extents won't change over the life of the file and they
7426  * use the bmap result to do IO directly to the drive.
7427  *
7428  * the btrfs bmap call would return logical addresses that aren't
7429  * suitable for IO and they also will change frequently as COW
7430  * operations happen.  So, swapfile + btrfs == corruption.
7431  *
7432  * For now we're avoiding this by dropping bmap.
7433  */
7434 static const struct address_space_operations btrfs_aops = {
7435         .readpage       = btrfs_readpage,
7436         .writepage      = btrfs_writepage,
7437         .writepages     = btrfs_writepages,
7438         .readpages      = btrfs_readpages,
7439         .direct_IO      = btrfs_direct_IO,
7440         .invalidatepage = btrfs_invalidatepage,
7441         .releasepage    = btrfs_releasepage,
7442         .set_page_dirty = btrfs_set_page_dirty,
7443         .error_remove_page = generic_error_remove_page,
7444 };
7445
7446 static const struct address_space_operations btrfs_symlink_aops = {
7447         .readpage       = btrfs_readpage,
7448         .writepage      = btrfs_writepage,
7449         .invalidatepage = btrfs_invalidatepage,
7450         .releasepage    = btrfs_releasepage,
7451 };
7452
7453 static const struct inode_operations btrfs_file_inode_operations = {
7454         .getattr        = btrfs_getattr,
7455         .setattr        = btrfs_setattr,
7456         .setxattr       = btrfs_setxattr,
7457         .getxattr       = btrfs_getxattr,
7458         .listxattr      = btrfs_listxattr,
7459         .removexattr    = btrfs_removexattr,
7460         .permission     = btrfs_permission,
7461         .fiemap         = btrfs_fiemap,
7462         .get_acl        = btrfs_get_acl,
7463 };
7464 static const struct inode_operations btrfs_special_inode_operations = {
7465         .getattr        = btrfs_getattr,
7466         .setattr        = btrfs_setattr,
7467         .permission     = btrfs_permission,
7468         .setxattr       = btrfs_setxattr,
7469         .getxattr       = btrfs_getxattr,
7470         .listxattr      = btrfs_listxattr,
7471         .removexattr    = btrfs_removexattr,
7472         .get_acl        = btrfs_get_acl,
7473 };
7474 static const struct inode_operations btrfs_symlink_inode_operations = {
7475         .readlink       = generic_readlink,
7476         .follow_link    = page_follow_link_light,
7477         .put_link       = page_put_link,
7478         .getattr        = btrfs_getattr,
7479         .setattr        = btrfs_setattr,
7480         .permission     = btrfs_permission,
7481         .setxattr       = btrfs_setxattr,
7482         .getxattr       = btrfs_getxattr,
7483         .listxattr      = btrfs_listxattr,
7484         .removexattr    = btrfs_removexattr,
7485         .get_acl        = btrfs_get_acl,
7486 };
7487
7488 const struct dentry_operations btrfs_dentry_operations = {
7489         .d_delete       = btrfs_dentry_delete,
7490         .d_release      = btrfs_dentry_release,
7491 };