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, &ins, 1);
 638                 btrfs_end_transaction(trans, root);
 639
 640                 if (ret) {
 641                         int i;
 642                         for (i = 0; i < async_extent->nr_pages; i++) {
 643                                 WARN_ON(async_extent->pages[i]->mapping);
 644                                 page_cache_release(async_extent->pages[i]);
 645                         }
 646                         kfree(async_extent->pages);
 647                         async_extent->nr_pages = 0;
 648                         async_extent->pages = NULL;
 649                         unlock_extent(io_tree, async_extent->start,
 650                                       async_extent->start +
 651                                       async_extent->ram_size - 1, GFP_NOFS);
 652                         goto retry;
 653                 }
 654
 655                 /*
 656                  * here we're doing allocation and writeback of the
 657                  * compressed pages
 658                  */
 659                 btrfs_drop_extent_cache(inode, async_extent->start,
 660                                         async_extent->start +
 661                                         async_extent->ram_size - 1, 0);
 662
 663                 em = alloc_extent_map();
 664                 BUG_ON(!em);
 665                 em->start = async_extent->start;
 666                 em->len = async_extent->ram_size;
 667                 em->orig_start = em->start;
 668
 669                 em->block_start = ins.objectid;
 670                 em->block_len = ins.offset;
 671                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 672                 em->compress_type = async_extent->compress_type;
 673                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 674                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 675
 676                 while (1) {
 677                         write_lock(&em_tree->lock);
 678                         ret = add_extent_mapping(em_tree, em);
 679                         write_unlock(&em_tree->lock);
 680                         if (ret != -EEXIST) {
 681                                 free_extent_map(em);
 682                                 break;
 683                         }
 684                         btrfs_drop_extent_cache(inode, async_extent->start,
 685                                                 async_extent->start +
 686                                                 async_extent->ram_size - 1, 0);
 687                 }
 688
 689                 ret = btrfs_add_ordered_extent_compress(inode,
 690                                                 async_extent->start,
 691                                                 ins.objectid,
 692                                                 async_extent->ram_size,
 693                                                 ins.offset,
 694                                                 BTRFS_ORDERED_COMPRESSED,
 695                                                 async_extent->compress_type);
 696                 BUG_ON(ret);
 697
 698                 /*
 699                  * clear dirty, set writeback and unlock the pages.
 700                  */
 701                 extent_clear_unlock_delalloc(inode,
 702                                 &BTRFS_I(inode)->io_tree,
 703                                 async_extent->start,
 704                                 async_extent->start +
 705                                 async_extent->ram_size - 1,
 706                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 707                                 EXTENT_CLEAR_UNLOCK |
 708                                 EXTENT_CLEAR_DELALLOC |
 709                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 710
 711                 ret = btrfs_submit_compressed_write(inode,
 712                                     async_extent->start,
 713                                     async_extent->ram_size,
 714                                     ins.objectid,
 715                                     ins.offset, async_extent->pages,
 716                                     async_extent->nr_pages);
 717
 718                 BUG_ON(ret);
 719                 alloc_hint = ins.objectid + ins.offset;
 720                 kfree(async_extent);
 721                 cond_resched();
 722         }
 723
 724         return 0;
 725 }
 726
 727 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 728                                       u64 num_bytes)
 729 {
 730         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 731         struct extent_map *em;
 732         u64 alloc_hint = 0;
 733
 734         read_lock(&em_tree->lock);
 735         em = search_extent_mapping(em_tree, start, num_bytes);
 736         if (em) {
 737                 /*
 738                  * if block start isn't an actual block number then find the
 739                  * first block in this inode and use that as a hint.  If that
 740                  * block is also bogus then just don't worry about it.
 741                  */
 742                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 743                         free_extent_map(em);
 744                         em = search_extent_mapping(em_tree, 0, 0);
 745                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 746                                 alloc_hint = em->block_start;
 747                         if (em)
 748                                 free_extent_map(em);
 749                 } else {
 750                         alloc_hint = em->block_start;
 751                         free_extent_map(em);
 752                 }
 753         }
 754         read_unlock(&em_tree->lock);
 755
 756         return alloc_hint;
 757 }
 758
 759 /*
 760  * when extent_io.c finds a delayed allocation range in the file,
 761  * the call backs end up in this code.  The basic idea is to
 762  * allocate extents on disk for the range, and create ordered data structs
 763  * in ram to track those extents.
 764  *
 765  * locked_page is the page that writepage had locked already.  We use
 766  * it to make sure we don't do extra locks or unlocks.
 767  *
 768  * *page_started is set to one if we unlock locked_page and do everything
 769  * required to start IO on it.  It may be clean and already done with
 770  * IO when we return.
 771  */
 772 static noinline int cow_file_range(struct inode *inode,
 773                                    struct page *locked_page,
 774                                    u64 start, u64 end, int *page_started,
 775                                    unsigned long *nr_written,
 776                                    int unlock)
 777 {
 778         struct btrfs_root *root = BTRFS_I(inode)->root;
 779         struct btrfs_trans_handle *trans;
 780         u64 alloc_hint = 0;
 781         u64 num_bytes;
 782         unsigned long ram_size;
 783         u64 disk_num_bytes;
 784         u64 cur_alloc_size;
 785         u64 blocksize = root->sectorsize;
 786         struct btrfs_key ins;
 787         struct extent_map *em;
 788         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 789         int ret = 0;
 790
 791         BUG_ON(btrfs_is_free_space_inode(root, inode));
 792         trans = btrfs_join_transaction(root);
 793         BUG_ON(IS_ERR(trans));
 794         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 795
 796         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 797         num_bytes = max(blocksize,  num_bytes);
 798         disk_num_bytes = num_bytes;
 799         ret = 0;
 800
 801         /* if this is a small write inside eof, kick off defrag */
 802         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 803                 btrfs_add_inode_defrag(trans, inode);
 804
 805         if (start == 0) {
 806                 /* lets try to make an inline extent */
 807                 ret = cow_file_range_inline(trans, root, inode,
 808                                             start, end, 0, 0, NULL);
 809                 if (ret == 0) {
 810                         extent_clear_unlock_delalloc(inode,
 811                                      &BTRFS_I(inode)->io_tree,
 812                                      start, end, NULL,
 813                                      EXTENT_CLEAR_UNLOCK_PAGE |
 814                                      EXTENT_CLEAR_UNLOCK |
 815                                      EXTENT_CLEAR_DELALLOC |
 816                                      EXTENT_CLEAR_DIRTY |
 817                                      EXTENT_SET_WRITEBACK |
 818                                      EXTENT_END_WRITEBACK);
 819
 820                         *nr_written = *nr_written +
 821                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 822                         *page_started = 1;
 823                         ret = 0;
 824                         goto out;
 825                 }
 826         }
 827
 828         BUG_ON(disk_num_bytes >
 829                btrfs_super_total_bytes(root->fs_info->super_copy));
 830
 831         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 832         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 833
 834         while (disk_num_bytes > 0) {
 835                 unsigned long op;
 836
 837                 cur_alloc_size = disk_num_bytes;
 838                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 839                                            root->sectorsize, 0, alloc_hint,
 840                                            &ins, 1);
 841                 BUG_ON(ret);
 842
 843                 em = alloc_extent_map();
 844                 BUG_ON(!em);
 845                 em->start = start;
 846                 em->orig_start = em->start;
 847                 ram_size = ins.offset;
 848                 em->len = ins.offset;
 849
 850                 em->block_start = ins.objectid;
 851                 em->block_len = ins.offset;
 852                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 853                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 854
 855                 while (1) {
 856                         write_lock(&em_tree->lock);
 857                         ret = add_extent_mapping(em_tree, em);
 858                         write_unlock(&em_tree->lock);
 859                         if (ret != -EEXIST) {
 860                                 free_extent_map(em);
 861                                 break;
 862                         }
 863                         btrfs_drop_extent_cache(inode, start,
 864                                                 start + ram_size - 1, 0);
 865                 }
 866
 867                 cur_alloc_size = ins.offset;
 868                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 869                                                ram_size, cur_alloc_size, 0);
 870                 BUG_ON(ret);
 871
 872                 if (root->root_key.objectid ==
 873                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 874                         ret = btrfs_reloc_clone_csums(inode, start,
 875                                                       cur_alloc_size);
 876                         BUG_ON(ret);
 877                 }
 878
 879                 if (disk_num_bytes < cur_alloc_size)
 880                         break;
 881
 882                 /* we're not doing compressed IO, don't unlock the first
 883                  * page (which the caller expects to stay locked), don't
 884                  * clear any dirty bits and don't set any writeback bits
 885                  *
 886                  * Do set the Private2 bit so we know this page was properly
 887                  * setup for writepage
 888                  */
 889                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 890                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 891                         EXTENT_SET_PRIVATE2;
 892
 893                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 894                                              start, start + ram_size - 1,
 895                                              locked_page, op);
 896                 disk_num_bytes -= cur_alloc_size;
 897                 num_bytes -= cur_alloc_size;
 898                 alloc_hint = ins.objectid + ins.offset;
 899                 start += cur_alloc_size;
 900         }
 901 out:
 902         ret = 0;
 903         btrfs_end_transaction(trans, root);
 904
 905         return ret;
 906 }
 907
 908 /*
 909  * work queue call back to started compression on a file and pages
 910  */
 911 static noinline void async_cow_start(struct btrfs_work *work)
 912 {
 913         struct async_cow *async_cow;
 914         int num_added = 0;
 915         async_cow = container_of(work, struct async_cow, work);
 916
 917         compress_file_range(async_cow->inode, async_cow->locked_page,
 918                             async_cow->start, async_cow->end, async_cow,
 919                             &num_added);
 920         if (num_added == 0)
 921                 async_cow->inode = NULL;
 922 }
 923
 924 /*
 925  * work queue call back to submit previously compressed pages
 926  */
 927 static noinline void async_cow_submit(struct btrfs_work *work)
 928 {
 929         struct async_cow *async_cow;
 930         struct btrfs_root *root;
 931         unsigned long nr_pages;
 932
 933         async_cow = container_of(work, struct async_cow, work);
 934
 935         root = async_cow->root;
 936         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 937                 PAGE_CACHE_SHIFT;
 938
 939         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 940
 941         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 942             5 * 1042 * 1024 &&
 943             waitqueue_active(&root->fs_info->async_submit_wait))
 944                 wake_up(&root->fs_info->async_submit_wait);
 945
 946         if (async_cow->inode)
 947                 submit_compressed_extents(async_cow->inode, async_cow);
 948 }
 949
 950 static noinline void async_cow_free(struct btrfs_work *work)
 951 {
 952         struct async_cow *async_cow;
 953         async_cow = container_of(work, struct async_cow, work);
 954         kfree(async_cow);
 955 }
 956
 957 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 958                                 u64 start, u64 end, int *page_started,
 959                                 unsigned long *nr_written)
 960 {
 961         struct async_cow *async_cow;
 962         struct btrfs_root *root = BTRFS_I(inode)->root;
 963         unsigned long nr_pages;
 964         u64 cur_end;
 965         int limit = 10 * 1024 * 1042;
 966
 967         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 968                          1, 0, NULL, GFP_NOFS);
 969         while (start < end) {
 970                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 971                 BUG_ON(!async_cow);
 972                 async_cow->inode = inode;
 973                 async_cow->root = root;
 974                 async_cow->locked_page = locked_page;
 975                 async_cow->start = start;
 976
 977                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 978                         cur_end = end;
 979                 else
 980                         cur_end = min(end, start + 512 * 1024 - 1);
 981
 982                 async_cow->end = cur_end;
 983                 INIT_LIST_HEAD(&async_cow->extents);
 984
 985                 async_cow->work.func = async_cow_start;
 986                 async_cow->work.ordered_func = async_cow_submit;
 987                 async_cow->work.ordered_free = async_cow_free;
 988                 async_cow->work.flags = 0;
 989
 990                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 991                         PAGE_CACHE_SHIFT;
 992                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 993
 994                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 995                                    &async_cow->work);
 996
 997                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 998                         wait_event(root->fs_info->async_submit_wait,
 999                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1000                             limit));
1001                 }
1002
1003                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1004                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1005                         wait_event(root->fs_info->async_submit_wait,
1006                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1007                            0));
1008                 }
1009
1010                 *nr_written += nr_pages;
1011                 start = cur_end + 1;
1012         }
1013         *page_started = 1;
1014         return 0;
1015 }
1016
1017 static noinline int csum_exist_in_range(struct btrfs_root *root,
1018                                         u64 bytenr, u64 num_bytes)
1019 {
1020         int ret;
1021         struct btrfs_ordered_sum *sums;
1022         LIST_HEAD(list);
1023
1024         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1025                                        bytenr + num_bytes - 1, &list, 0);
1026         if (ret == 0 && list_empty(&list))
1027                 return 0;
1028
1029         while (!list_empty(&list)) {
1030                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1031                 list_del(&sums->list);
1032                 kfree(sums);
1033         }
1034         return 1;
1035 }
1036
1037 /*
1038  * when nowcow writeback call back.  This checks for snapshots or COW copies
1039  * of the extents that exist in the file, and COWs the file as required.
1040  *
1041  * If no cow copies or snapshots exist, we write directly to the existing
1042  * blocks on disk
1043  */
1044 static noinline int run_delalloc_nocow(struct inode *inode,
1045                                        struct page *locked_page,
1046                               u64 start, u64 end, int *page_started, int force,
1047                               unsigned long *nr_written)
1048 {
1049         struct btrfs_root *root = BTRFS_I(inode)->root;
1050         struct btrfs_trans_handle *trans;
1051         struct extent_buffer *leaf;
1052         struct btrfs_path *path;
1053         struct btrfs_file_extent_item *fi;
1054         struct btrfs_key found_key;
1055         u64 cow_start;
1056         u64 cur_offset;
1057         u64 extent_end;
1058         u64 extent_offset;
1059         u64 disk_bytenr;
1060         u64 num_bytes;
1061         int extent_type;
1062         int ret;
1063         int type;
1064         int nocow;
1065         int check_prev = 1;
1066         bool nolock;
1067         u64 ino = btrfs_ino(inode);
1068
1069         path = btrfs_alloc_path();
1070         if (!path)
1071                 return -ENOMEM;
1072
1073         nolock = btrfs_is_free_space_inode(root, inode);
1074
1075         if (nolock)
1076                 trans = btrfs_join_transaction_nolock(root);
1077         else
1078                 trans = btrfs_join_transaction(root);
1079
1080         BUG_ON(IS_ERR(trans));
1081         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1082
1083         cow_start = (u64)-1;
1084         cur_offset = start;
1085         while (1) {
1086                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1087                                                cur_offset, 0);
1088                 BUG_ON(ret < 0);
1089                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1090                         leaf = path->nodes[0];
1091                         btrfs_item_key_to_cpu(leaf, &found_key,
1092                                               path->slots[0] - 1);
1093                         if (found_key.objectid == ino &&
1094                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1095                                 path->slots[0]--;
1096                 }
1097                 check_prev = 0;
1098 next_slot:
1099                 leaf = path->nodes[0];
1100                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1101                         ret = btrfs_next_leaf(root, path);
1102                         if (ret < 0)
1103                                 BUG_ON(1);
1104                         if (ret > 0)
1105                                 break;
1106                         leaf = path->nodes[0];
1107                 }
1108
1109                 nocow = 0;
1110                 disk_bytenr = 0;
1111                 num_bytes = 0;
1112                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1113
1114                 if (found_key.objectid > ino ||
1115                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1116                     found_key.offset > end)
1117                         break;
1118
1119                 if (found_key.offset > cur_offset) {
1120                         extent_end = found_key.offset;
1121                         extent_type = 0;
1122                         goto out_check;
1123                 }
1124
1125                 fi = btrfs_item_ptr(leaf, path->slots[0],
1126                                     struct btrfs_file_extent_item);
1127                 extent_type = btrfs_file_extent_type(leaf, fi);
1128
1129                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1130                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1131                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1132                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1133                         extent_end = found_key.offset +
1134                                 btrfs_file_extent_num_bytes(leaf, fi);
1135                         if (extent_end <= start) {
1136                                 path->slots[0]++;
1137                                 goto next_slot;
1138                         }
1139                         if (disk_bytenr == 0)
1140                                 goto out_check;
1141                         if (btrfs_file_extent_compression(leaf, fi) ||
1142                             btrfs_file_extent_encryption(leaf, fi) ||
1143                             btrfs_file_extent_other_encoding(leaf, fi))
1144                                 goto out_check;
1145                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1146                                 goto out_check;
1147                         if (btrfs_extent_readonly(root, disk_bytenr))
1148                                 goto out_check;
1149                         if (btrfs_cross_ref_exist(trans, root, ino,
1150                                                   found_key.offset -
1151                                                   extent_offset, disk_bytenr))
1152                                 goto out_check;
1153                         disk_bytenr += extent_offset;
1154                         disk_bytenr += cur_offset - found_key.offset;
1155                         num_bytes = min(end + 1, extent_end) - cur_offset;
1156                         /*
1157                          * force cow if csum exists in the range.
1158                          * this ensure that csum for a given extent are
1159                          * either valid or do not exist.
1160                          */
1161                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1162                                 goto out_check;
1163                         nocow = 1;
1164                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1165                         extent_end = found_key.offset +
1166                                 btrfs_file_extent_inline_len(leaf, fi);
1167                         extent_end = ALIGN(extent_end, root->sectorsize);
1168                 } else {
1169                         BUG_ON(1);
1170                 }
1171 out_check:
1172                 if (extent_end <= start) {
1173                         path->slots[0]++;
1174                         goto next_slot;
1175                 }
1176                 if (!nocow) {
1177                         if (cow_start == (u64)-1)
1178                                 cow_start = cur_offset;
1179                         cur_offset = extent_end;
1180                         if (cur_offset > end)
1181                                 break;
1182                         path->slots[0]++;
1183                         goto next_slot;
1184                 }
1185
1186                 btrfs_release_path(path);
1187                 if (cow_start != (u64)-1) {
1188                         ret = cow_file_range(inode, locked_page, cow_start,
1189                                         found_key.offset - 1, page_started,
1190                                         nr_written, 1);
1191                         BUG_ON(ret);
1192                         cow_start = (u64)-1;
1193                 }
1194
1195                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1196                         struct extent_map *em;
1197                         struct extent_map_tree *em_tree;
1198                         em_tree = &BTRFS_I(inode)->extent_tree;
1199                         em = alloc_extent_map();
1200                         BUG_ON(!em);
1201                         em->start = cur_offset;
1202                         em->orig_start = em->start;
1203                         em->len = num_bytes;
1204                         em->block_len = num_bytes;
1205                         em->block_start = disk_bytenr;
1206                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1207                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1208                         while (1) {
1209                                 write_lock(&em_tree->lock);
1210                                 ret = add_extent_mapping(em_tree, em);
1211                                 write_unlock(&em_tree->lock);
1212                                 if (ret != -EEXIST) {
1213                                         free_extent_map(em);
1214                                         break;
1215                                 }
1216                                 btrfs_drop_extent_cache(inode, em->start,
1217                                                 em->start + em->len - 1, 0);
1218                         }
1219                         type = BTRFS_ORDERED_PREALLOC;
1220                 } else {
1221                         type = BTRFS_ORDERED_NOCOW;
1222                 }
1223
1224                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1225                                                num_bytes, num_bytes, type);
1226                 BUG_ON(ret);
1227
1228                 if (root->root_key.objectid ==
1229                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1230                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1231                                                       num_bytes);
1232                         BUG_ON(ret);
1233                 }
1234
1235                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1236                                 cur_offset, cur_offset + num_bytes - 1,
1237                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1238                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1239                                 EXTENT_SET_PRIVATE2);
1240                 cur_offset = extent_end;
1241                 if (cur_offset > end)
1242                         break;
1243         }
1244         btrfs_release_path(path);
1245
1246         if (cur_offset <= end && cow_start == (u64)-1)
1247                 cow_start = cur_offset;
1248         if (cow_start != (u64)-1) {
1249                 ret = cow_file_range(inode, locked_page, cow_start, end,
1250                                      page_started, nr_written, 1);
1251                 BUG_ON(ret);
1252         }
1253
1254         if (nolock) {
1255                 ret = btrfs_end_transaction_nolock(trans, root);
1256                 BUG_ON(ret);
1257         } else {
1258                 ret = btrfs_end_transaction(trans, root);
1259                 BUG_ON(ret);
1260         }
1261         btrfs_free_path(path);
1262         return 0;
1263 }
1264
1265 /*
1266  * extent_io.c call back to do delayed allocation processing
1267  */
1268 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1269                               u64 start, u64 end, int *page_started,
1270                               unsigned long *nr_written)
1271 {
1272         int ret;
1273         struct btrfs_root *root = BTRFS_I(inode)->root;
1274
1275         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1276                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1277                                          page_started, 1, nr_written);
1278         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1279                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1280                                          page_started, 0, nr_written);
1281         else if (!btrfs_test_opt(root, COMPRESS) &&
1282                  !(BTRFS_I(inode)->force_compress) &&
1283                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1284                 ret = cow_file_range(inode, locked_page, start, end,
1285                                       page_started, nr_written, 1);
1286         else
1287                 ret = cow_file_range_async(inode, locked_page, start, end,
1288                                            page_started, nr_written);
1289         return ret;
1290 }
1291
1292 static void btrfs_split_extent_hook(struct inode *inode,
1293                                     struct extent_state *orig, u64 split)
1294 {
1295         /* not delalloc, ignore it */
1296         if (!(orig->state & EXTENT_DELALLOC))
1297                 return;
1298
1299         spin_lock(&BTRFS_I(inode)->lock);
1300         BTRFS_I(inode)->outstanding_extents++;
1301         spin_unlock(&BTRFS_I(inode)->lock);
1302 }
1303
1304 /*
1305  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1306  * extents so we can keep track of new extents that are just merged onto old
1307  * extents, such as when we are doing sequential writes, so we can properly
1308  * account for the metadata space we'll need.
1309  */
1310 static void btrfs_merge_extent_hook(struct inode *inode,
1311                                     struct extent_state *new,
1312                                     struct extent_state *other)
1313 {
1314         /* not delalloc, ignore it */
1315         if (!(other->state & EXTENT_DELALLOC))
1316                 return;
1317
1318         spin_lock(&BTRFS_I(inode)->lock);
1319         BTRFS_I(inode)->outstanding_extents--;
1320         spin_unlock(&BTRFS_I(inode)->lock);
1321 }
1322
1323 /*
1324  * extent_io.c set_bit_hook, used to track delayed allocation
1325  * bytes in this file, and to maintain the list of inodes that
1326  * have pending delalloc work to be done.
1327  */
1328 static void btrfs_set_bit_hook(struct inode *inode,
1329                                struct extent_state *state, int *bits)
1330 {
1331
1332         /*
1333          * set_bit and clear bit hooks normally require _irqsave/restore
1334          * but in this case, we are only testing for the DELALLOC
1335          * bit, which is only set or cleared with irqs on
1336          */
1337         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1338                 struct btrfs_root *root = BTRFS_I(inode)->root;
1339                 u64 len = state->end + 1 - state->start;
1340                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1341
1342                 if (*bits & EXTENT_FIRST_DELALLOC) {
1343                         *bits &= ~EXTENT_FIRST_DELALLOC;
1344                 } else {
1345                         spin_lock(&BTRFS_I(inode)->lock);
1346                         BTRFS_I(inode)->outstanding_extents++;
1347                         spin_unlock(&BTRFS_I(inode)->lock);
1348                 }
1349
1350                 spin_lock(&root->fs_info->delalloc_lock);
1351                 BTRFS_I(inode)->delalloc_bytes += len;
1352                 root->fs_info->delalloc_bytes += len;
1353                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1354                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1355                                       &root->fs_info->delalloc_inodes);
1356                 }
1357                 spin_unlock(&root->fs_info->delalloc_lock);
1358         }
1359 }
1360
1361 /*
1362  * extent_io.c clear_bit_hook, see set_bit_hook for why
1363  */
1364 static void btrfs_clear_bit_hook(struct inode *inode,
1365                                  struct extent_state *state, int *bits)
1366 {
1367         /*
1368          * set_bit and clear bit hooks normally require _irqsave/restore
1369          * but in this case, we are only testing for the DELALLOC
1370          * bit, which is only set or cleared with irqs on
1371          */
1372         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1373                 struct btrfs_root *root = BTRFS_I(inode)->root;
1374                 u64 len = state->end + 1 - state->start;
1375                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1376
1377                 if (*bits & EXTENT_FIRST_DELALLOC) {
1378                         *bits &= ~EXTENT_FIRST_DELALLOC;
1379                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1380                         spin_lock(&BTRFS_I(inode)->lock);
1381                         BTRFS_I(inode)->outstanding_extents--;
1382                         spin_unlock(&BTRFS_I(inode)->lock);
1383                 }
1384
1385                 if (*bits & EXTENT_DO_ACCOUNTING)
1386                         btrfs_delalloc_release_metadata(inode, len);
1387
1388                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1389                     && do_list)
1390                         btrfs_free_reserved_data_space(inode, len);
1391
1392                 spin_lock(&root->fs_info->delalloc_lock);
1393                 root->fs_info->delalloc_bytes -= len;
1394                 BTRFS_I(inode)->delalloc_bytes -= len;
1395
1396                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1397                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1398                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1399                 }
1400                 spin_unlock(&root->fs_info->delalloc_lock);
1401         }
1402 }
1403
1404 /*
1405  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1406  * we don't create bios that span stripes or chunks
1407  */
1408 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1409                          size_t size, struct bio *bio,
1410                          unsigned long bio_flags)
1411 {
1412         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1413         struct btrfs_mapping_tree *map_tree;
1414         u64 logical = (u64)bio->bi_sector << 9;
1415         u64 length = 0;
1416         u64 map_length;
1417         int ret;
1418
1419         if (bio_flags & EXTENT_BIO_COMPRESSED)
1420                 return 0;
1421
1422         length = bio->bi_size;
1423         map_tree = &root->fs_info->mapping_tree;
1424         map_length = length;
1425         ret = btrfs_map_block(map_tree, READ, logical,
1426                               &map_length, NULL, 0);
1427
1428         if (map_length < length + size)
1429                 return 1;
1430         return ret;
1431 }
1432
1433 /*
1434  * in order to insert checksums into the metadata in large chunks,
1435  * we wait until bio submission time.   All the pages in the bio are
1436  * checksummed and sums are attached onto the ordered extent record.
1437  *
1438  * At IO completion time the cums attached on the ordered extent record
1439  * are inserted into the btree
1440  */
1441 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1442                                     struct bio *bio, int mirror_num,
1443                                     unsigned long bio_flags,
1444                                     u64 bio_offset)
1445 {
1446         struct btrfs_root *root = BTRFS_I(inode)->root;
1447         int ret = 0;
1448
1449         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1450         BUG_ON(ret);
1451         return 0;
1452 }
1453
1454 /*
1455  * in order to insert checksums into the metadata in large chunks,
1456  * we wait until bio submission time.   All the pages in the bio are
1457  * checksummed and sums are attached onto the ordered extent record.
1458  *
1459  * At IO completion time the cums attached on the ordered extent record
1460  * are inserted into the btree
1461  */
1462 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1463                           int mirror_num, unsigned long bio_flags,
1464                           u64 bio_offset)
1465 {
1466         struct btrfs_root *root = BTRFS_I(inode)->root;
1467         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1468 }
1469
1470 /*
1471  * extent_io.c submission hook. This does the right thing for csum calculation
1472  * on write, or reading the csums from the tree before a read
1473  */
1474 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1475                           int mirror_num, unsigned long bio_flags,
1476                           u64 bio_offset)
1477 {
1478         struct btrfs_root *root = BTRFS_I(inode)->root;
1479         int ret = 0;
1480         int skip_sum;
1481
1482         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1483
1484         if (btrfs_is_free_space_inode(root, inode))
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1486         else
1487                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1488         BUG_ON(ret);
1489
1490         if (!(rw & REQ_WRITE)) {
1491                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1492                         return btrfs_submit_compressed_read(inode, bio,
1493                                                     mirror_num, bio_flags);
1494                 } else if (!skip_sum) {
1495                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1496                         if (ret)
1497                                 return ret;
1498                 }
1499                 goto mapit;
1500         } else if (!skip_sum) {
1501                 /* csum items have already been cloned */
1502                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1503                         goto mapit;
1504                 /* we're doing a write, do the async checksumming */
1505                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1506                                    inode, rw, bio, mirror_num,
1507                                    bio_flags, bio_offset,
1508                                    __btrfs_submit_bio_start,
1509                                    __btrfs_submit_bio_done);
1510         }
1511
1512 mapit:
1513         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1514 }
1515
1516 /*
1517  * given a list of ordered sums record them in the inode.  This happens
1518  * at IO completion time based on sums calculated at bio submission time.
1519  */
1520 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1521                              struct inode *inode, u64 file_offset,
1522                              struct list_head *list)
1523 {
1524         struct btrfs_ordered_sum *sum;
1525
1526         list_for_each_entry(sum, list, list) {
1527                 btrfs_csum_file_blocks(trans,
1528                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1529         }
1530         return 0;
1531 }
1532
1533 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1534                               struct extent_state **cached_state)
1535 {
1536         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1537                 WARN_ON(1);
1538         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1539                                    cached_state, GFP_NOFS);
1540 }
1541
1542 /* see btrfs_writepage_start_hook for details on why this is required */
1543 struct btrfs_writepage_fixup {
1544         struct page *page;
1545         struct btrfs_work work;
1546 };
1547
1548 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1549 {
1550         struct btrfs_writepage_fixup *fixup;
1551         struct btrfs_ordered_extent *ordered;
1552         struct extent_state *cached_state = NULL;
1553         struct page *page;
1554         struct inode *inode;
1555         u64 page_start;
1556         u64 page_end;
1557         int ret;
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                 btrfs_put_ordered_extent(ordered);
1586                 goto again;
1587         }
1588
1589         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1590         if (ret) {
1591                 mapping_set_error(page->mapping, ret);
1592                 end_extent_writepage(page, ret, page_start, page_end);
1593                 ClearPageChecked(page);
1594                 goto out;
1595          }
1596
1597         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1598         ClearPageChecked(page);
1599         set_page_dirty(page);
1600 out:
1601         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1602                              &cached_state, GFP_NOFS);
1603 out_page:
1604         unlock_page(page);
1605         page_cache_release(page);
1606         kfree(fixup);
1607 }
1608
1609 /*
1610  * There are a few paths in the higher layers of the kernel that directly
1611  * set the page dirty bit without asking the filesystem if it is a
1612  * good idea.  This causes problems because we want to make sure COW
1613  * properly happens and the data=ordered rules are followed.
1614  *
1615  * In our case any range that doesn't have the ORDERED bit set
1616  * hasn't been properly setup for IO.  We kick off an async process
1617  * to fix it up.  The async helper will wait for ordered extents, set
1618  * the delalloc bit and make it safe to write the page.
1619  */
1620 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1621 {
1622         struct inode *inode = page->mapping->host;
1623         struct btrfs_writepage_fixup *fixup;
1624         struct btrfs_root *root = BTRFS_I(inode)->root;
1625
1626         /* this page is properly in the ordered list */
1627         if (TestClearPagePrivate2(page))
1628                 return 0;
1629
1630         if (PageChecked(page))
1631                 return -EAGAIN;
1632
1633         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1634         if (!fixup)
1635                 return -EAGAIN;
1636
1637         SetPageChecked(page);
1638         page_cache_get(page);
1639         fixup->work.func = btrfs_writepage_fixup_worker;
1640         fixup->page = page;
1641         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1642         return -EBUSY;
1643 }
1644
1645 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1646                                        struct inode *inode, u64 file_pos,
1647                                        u64 disk_bytenr, u64 disk_num_bytes,
1648                                        u64 num_bytes, u64 ram_bytes,
1649                                        u8 compression, u8 encryption,
1650                                        u16 other_encoding, int extent_type)
1651 {
1652         struct btrfs_root *root = BTRFS_I(inode)->root;
1653         struct btrfs_file_extent_item *fi;
1654         struct btrfs_path *path;
1655         struct extent_buffer *leaf;
1656         struct btrfs_key ins;
1657         u64 hint;
1658         int ret;
1659
1660         path = btrfs_alloc_path();
1661         if (!path)
1662                 return -ENOMEM;
1663
1664         path->leave_spinning = 1;
1665
1666         /*
1667          * we may be replacing one extent in the tree with another.
1668          * The new extent is pinned in the extent map, and we don't want
1669          * to drop it from the cache until it is completely in the btree.
1670          *
1671          * So, tell btrfs_drop_extents to leave this extent in the cache.
1672          * the caller is expected to unpin it and allow it to be merged
1673          * with the others.
1674          */
1675         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1676                                  &hint, 0);
1677         BUG_ON(ret);
1678
1679         ins.objectid = btrfs_ino(inode);
1680         ins.offset = file_pos;
1681         ins.type = BTRFS_EXTENT_DATA_KEY;
1682         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1683         BUG_ON(ret);
1684         leaf = path->nodes[0];
1685         fi = btrfs_item_ptr(leaf, path->slots[0],
1686                             struct btrfs_file_extent_item);
1687         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1688         btrfs_set_file_extent_type(leaf, fi, extent_type);
1689         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1690         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1691         btrfs_set_file_extent_offset(leaf, fi, 0);
1692         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1693         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1694         btrfs_set_file_extent_compression(leaf, fi, compression);
1695         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1696         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1697
1698         btrfs_unlock_up_safe(path, 1);
1699         btrfs_set_lock_blocking(leaf);
1700
1701         btrfs_mark_buffer_dirty(leaf);
1702
1703         inode_add_bytes(inode, num_bytes);
1704
1705         ins.objectid = disk_bytenr;
1706         ins.offset = disk_num_bytes;
1707         ins.type = BTRFS_EXTENT_ITEM_KEY;
1708         ret = btrfs_alloc_reserved_file_extent(trans, root,
1709                                         root->root_key.objectid,
1710                                         btrfs_ino(inode), file_pos, &ins);
1711         BUG_ON(ret);
1712         btrfs_free_path(path);
1713
1714         return 0;
1715 }
1716
1717 /*
1718  * helper function for btrfs_finish_ordered_io, this
1719  * just reads in some of the csum leaves to prime them into ram
1720  * before we start the transaction.  It limits the amount of btree
1721  * reads required while inside the transaction.
1722  */
1723 /* as ordered data IO finishes, this gets called so we can finish
1724  * an ordered extent if the range of bytes in the file it covers are
1725  * fully written.
1726  */
1727 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1728 {
1729         struct btrfs_root *root = BTRFS_I(inode)->root;
1730         struct btrfs_trans_handle *trans = NULL;
1731         struct btrfs_ordered_extent *ordered_extent = NULL;
1732         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1733         struct extent_state *cached_state = NULL;
1734         int compress_type = 0;
1735         int ret;
1736         bool nolock;
1737
1738         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1739                                              end - start + 1);
1740         if (!ret)
1741                 return 0;
1742         BUG_ON(!ordered_extent);
1743
1744         nolock = btrfs_is_free_space_inode(root, inode);
1745
1746         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1747                 BUG_ON(!list_empty(&ordered_extent->list));
1748                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1749                 if (!ret) {
1750                         if (nolock)
1751                                 trans = btrfs_join_transaction_nolock(root);
1752                         else
1753                                 trans = btrfs_join_transaction(root);
1754                         BUG_ON(IS_ERR(trans));
1755                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1756                         ret = btrfs_update_inode_fallback(trans, root, inode);
1757                         BUG_ON(ret);
1758                 }
1759                 goto out;
1760         }
1761
1762         lock_extent_bits(io_tree, ordered_extent->file_offset,
1763                          ordered_extent->file_offset + ordered_extent->len - 1,
1764                          0, &cached_state, GFP_NOFS);
1765
1766         if (nolock)
1767                 trans = btrfs_join_transaction_nolock(root);
1768         else
1769                 trans = btrfs_join_transaction(root);
1770         BUG_ON(IS_ERR(trans));
1771         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1772
1773         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1774                 compress_type = ordered_extent->compress_type;
1775         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1776                 BUG_ON(compress_type);
1777                 ret = btrfs_mark_extent_written(trans, inode,
1778                                                 ordered_extent->file_offset,
1779                                                 ordered_extent->file_offset +
1780                                                 ordered_extent->len);
1781                 BUG_ON(ret);
1782         } else {
1783                 BUG_ON(root == root->fs_info->tree_root);
1784                 ret = insert_reserved_file_extent(trans, inode,
1785                                                 ordered_extent->file_offset,
1786                                                 ordered_extent->start,
1787                                                 ordered_extent->disk_len,
1788                                                 ordered_extent->len,
1789                                                 ordered_extent->len,
1790                                                 compress_type, 0, 0,
1791                                                 BTRFS_FILE_EXTENT_REG);
1792                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1793                                    ordered_extent->file_offset,
1794                                    ordered_extent->len);
1795                 BUG_ON(ret);
1796         }
1797         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1798                              ordered_extent->file_offset +
1799                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1800
1801         add_pending_csums(trans, inode, ordered_extent->file_offset,
1802                           &ordered_extent->list);
1803
1804         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1805         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1806                 ret = btrfs_update_inode_fallback(trans, root, inode);
1807                 BUG_ON(ret);
1808         }
1809         ret = 0;
1810 out:
1811         if (root != root->fs_info->tree_root)
1812                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1813         if (trans) {
1814                 if (nolock)
1815                         btrfs_end_transaction_nolock(trans, root);
1816                 else
1817                         btrfs_end_transaction(trans, root);
1818         }
1819
1820         /* once for us */
1821         btrfs_put_ordered_extent(ordered_extent);
1822         /* once for the tree */
1823         btrfs_put_ordered_extent(ordered_extent);
1824
1825         return 0;
1826 }
1827
1828 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1829                                 struct extent_state *state, int uptodate)
1830 {
1831         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1832
1833         ClearPagePrivate2(page);
1834         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1835 }
1836
1837 /*
1838  * when reads are done, we need to check csums to verify the data is correct
1839  * if there's a match, we allow the bio to finish.  If not, the code in
1840  * extent_io.c will try to find good copies for us.
1841  */
1842 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1843                                struct extent_state *state)
1844 {
1845         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1846         struct inode *inode = page->mapping->host;
1847         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1848         char *kaddr;
1849         u64 private = ~(u32)0;
1850         int ret;
1851         struct btrfs_root *root = BTRFS_I(inode)->root;
1852         u32 csum = ~(u32)0;
1853
1854         if (PageChecked(page)) {
1855                 ClearPageChecked(page);
1856                 goto good;
1857         }
1858
1859         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1860                 goto good;
1861
1862         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1863             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1864                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1865                                   GFP_NOFS);
1866                 return 0;
1867         }
1868
1869         if (state && state->start == start) {
1870                 private = state->private;
1871                 ret = 0;
1872         } else {
1873                 ret = get_state_private(io_tree, start, &private);
1874         }
1875         kaddr = kmap_atomic(page, KM_USER0);
1876         if (ret)
1877                 goto zeroit;
1878
1879         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1880         btrfs_csum_final(csum, (char *)&csum);
1881         if (csum != private)
1882                 goto zeroit;
1883
1884         kunmap_atomic(kaddr, KM_USER0);
1885 good:
1886         return 0;
1887
1888 zeroit:
1889         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1890                        "private %llu\n",
1891                        (unsigned long long)btrfs_ino(page->mapping->host),
1892                        (unsigned long long)start, csum,
1893                        (unsigned long long)private);
1894         memset(kaddr + offset, 1, end - start + 1);
1895         flush_dcache_page(page);
1896         kunmap_atomic(kaddr, KM_USER0);
1897         if (private == 0)
1898                 return 0;
1899         return -EIO;
1900 }
1901
1902 struct delayed_iput {
1903         struct list_head list;
1904         struct inode *inode;
1905 };
1906
1907 void btrfs_add_delayed_iput(struct inode *inode)
1908 {
1909         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1910         struct delayed_iput *delayed;
1911
1912         if (atomic_add_unless(&inode->i_count, -1, 1))
1913                 return;
1914
1915         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1916         delayed->inode = inode;
1917
1918         spin_lock(&fs_info->delayed_iput_lock);
1919         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1920         spin_unlock(&fs_info->delayed_iput_lock);
1921 }
1922
1923 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1924 {
1925         LIST_HEAD(list);
1926         struct btrfs_fs_info *fs_info = root->fs_info;
1927         struct delayed_iput *delayed;
1928         int empty;
1929
1930         spin_lock(&fs_info->delayed_iput_lock);
1931         empty = list_empty(&fs_info->delayed_iputs);
1932         spin_unlock(&fs_info->delayed_iput_lock);
1933         if (empty)
1934                 return;
1935
1936         down_read(&root->fs_info->cleanup_work_sem);
1937         spin_lock(&fs_info->delayed_iput_lock);
1938         list_splice_init(&fs_info->delayed_iputs, &list);
1939         spin_unlock(&fs_info->delayed_iput_lock);
1940
1941         while (!list_empty(&list)) {
1942                 delayed = list_entry(list.next, struct delayed_iput, list);
1943                 list_del(&delayed->list);
1944                 iput(delayed->inode);
1945                 kfree(delayed);
1946         }
1947         up_read(&root->fs_info->cleanup_work_sem);
1948 }
1949
1950 enum btrfs_orphan_cleanup_state {
1951         ORPHAN_CLEANUP_STARTED  = 1,
1952         ORPHAN_CLEANUP_DONE     = 2,
1953 };
1954
1955 /*
1956  * This is called in transaction commit time. If there are no orphan
1957  * files in the subvolume, it removes orphan item and frees block_rsv
1958  * structure.
1959  */
1960 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1961                               struct btrfs_root *root)
1962 {
1963         struct btrfs_block_rsv *block_rsv;
1964         int ret;
1965
1966         if (!list_empty(&root->orphan_list) ||
1967             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1968                 return;
1969
1970         spin_lock(&root->orphan_lock);
1971         if (!list_empty(&root->orphan_list)) {
1972                 spin_unlock(&root->orphan_lock);
1973                 return;
1974         }
1975
1976         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
1977                 spin_unlock(&root->orphan_lock);
1978                 return;
1979         }
1980
1981         block_rsv = root->orphan_block_rsv;
1982         root->orphan_block_rsv = NULL;
1983         spin_unlock(&root->orphan_lock);
1984
1985         if (root->orphan_item_inserted &&
1986             btrfs_root_refs(&root->root_item) > 0) {
1987                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1988                                             root->root_key.objectid);
1989                 BUG_ON(ret);
1990                 root->orphan_item_inserted = 0;
1991         }
1992
1993         if (block_rsv) {
1994                 WARN_ON(block_rsv->size > 0);
1995                 btrfs_free_block_rsv(root, block_rsv);
1996         }
1997 }
1998
1999 /*
2000  * This creates an orphan entry for the given inode in case something goes
2001  * wrong in the middle of an unlink/truncate.
2002  *
2003  * NOTE: caller of this function should reserve 5 units of metadata for
2004  *       this function.
2005  */
2006 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2007 {
2008         struct btrfs_root *root = BTRFS_I(inode)->root;
2009         struct btrfs_block_rsv *block_rsv = NULL;
2010         int reserve = 0;
2011         int insert = 0;
2012         int ret;
2013
2014         if (!root->orphan_block_rsv) {
2015                 block_rsv = btrfs_alloc_block_rsv(root);
2016                 if (!block_rsv)
2017                         return -ENOMEM;
2018         }
2019
2020         spin_lock(&root->orphan_lock);
2021         if (!root->orphan_block_rsv) {
2022                 root->orphan_block_rsv = block_rsv;
2023         } else if (block_rsv) {
2024                 btrfs_free_block_rsv(root, block_rsv);
2025                 block_rsv = NULL;
2026         }
2027
2028         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2029                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2030 #if 0
2031                 /*
2032                  * For proper ENOSPC handling, we should do orphan
2033                  * cleanup when mounting. But this introduces backward
2034                  * compatibility issue.
2035                  */
2036                 if (!xchg(&root->orphan_item_inserted, 1))
2037                         insert = 2;
2038                 else
2039                         insert = 1;
2040 #endif
2041                 insert = 1;
2042         }
2043
2044         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2045                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2046                 reserve = 1;
2047         }
2048         spin_unlock(&root->orphan_lock);
2049
2050         /* grab metadata reservation from transaction handle */
2051         if (reserve) {
2052                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2053                 BUG_ON(ret);
2054         }
2055
2056         /* insert an orphan item to track this unlinked/truncated file */
2057         if (insert >= 1) {
2058                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2059                 BUG_ON(ret && ret != -EEXIST);
2060         }
2061
2062         /* insert an orphan item to track subvolume contains orphan files */
2063         if (insert >= 2) {
2064                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2065                                                root->root_key.objectid);
2066                 BUG_ON(ret);
2067         }
2068         return 0;
2069 }
2070
2071 /*
2072  * We have done the truncate/delete so we can go ahead and remove the orphan
2073  * item for this particular inode.
2074  */
2075 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2076 {
2077         struct btrfs_root *root = BTRFS_I(inode)->root;
2078         int delete_item = 0;
2079         int release_rsv = 0;
2080         int ret = 0;
2081
2082         spin_lock(&root->orphan_lock);
2083         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2084                 list_del_init(&BTRFS_I(inode)->i_orphan);
2085                 delete_item = 1;
2086         }
2087
2088         if (BTRFS_I(inode)->orphan_meta_reserved) {
2089                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2090                 release_rsv = 1;
2091         }
2092         spin_unlock(&root->orphan_lock);
2093
2094         if (trans && delete_item) {
2095                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2096                 BUG_ON(ret);
2097         }
2098
2099         if (release_rsv)
2100                 btrfs_orphan_release_metadata(inode);
2101
2102         return 0;
2103 }
2104
2105 /*
2106  * this cleans up any orphans that may be left on the list from the last use
2107  * of this root.
2108  */
2109 int btrfs_orphan_cleanup(struct btrfs_root *root)
2110 {
2111         struct btrfs_path *path;
2112         struct extent_buffer *leaf;
2113         struct btrfs_key key, found_key;
2114         struct btrfs_trans_handle *trans;
2115         struct inode *inode;
2116         u64 last_objectid = 0;
2117         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2118
2119         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2120                 return 0;
2121
2122         path = btrfs_alloc_path();
2123         if (!path) {
2124                 ret = -ENOMEM;
2125                 goto out;
2126         }
2127         path->reada = -1;
2128
2129         key.objectid = BTRFS_ORPHAN_OBJECTID;
2130         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2131         key.offset = (u64)-1;
2132
2133         while (1) {
2134                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2135                 if (ret < 0)
2136                         goto out;
2137
2138                 /*
2139                  * if ret == 0 means we found what we were searching for, which
2140                  * is weird, but possible, so only screw with path if we didn't
2141                  * find the key and see if we have stuff that matches
2142                  */
2143                 if (ret > 0) {
2144                         ret = 0;
2145                         if (path->slots[0] == 0)
2146                                 break;
2147                         path->slots[0]--;
2148                 }
2149
2150                 /* pull out the item */
2151                 leaf = path->nodes[0];
2152                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2153
2154                 /* make sure the item matches what we want */
2155                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2156                         break;
2157                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2158                         break;
2159
2160                 /* release the path since we're done with it */
2161                 btrfs_release_path(path);
2162
2163                 /*
2164                  * this is where we are basically btrfs_lookup, without the
2165                  * crossing root thing.  we store the inode number in the
2166                  * offset of the orphan item.
2167                  */
2168
2169                 if (found_key.offset == last_objectid) {
2170                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2171                                "stopping orphan cleanup\n");
2172                         ret = -EINVAL;
2173                         goto out;
2174                 }
2175
2176                 last_objectid = found_key.offset;
2177
2178                 found_key.objectid = found_key.offset;
2179                 found_key.type = BTRFS_INODE_ITEM_KEY;
2180                 found_key.offset = 0;
2181                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2182                 ret = PTR_RET(inode);
2183                 if (ret && ret != -ESTALE)
2184                         goto out;
2185
2186                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2187                         struct btrfs_root *dead_root;
2188                         struct btrfs_fs_info *fs_info = root->fs_info;
2189                         int is_dead_root = 0;
2190
2191                         /*
2192                          * this is an orphan in the tree root. Currently these
2193                          * could come from 2 sources:
2194                          *  a) a snapshot deletion in progress
2195                          *  b) a free space cache inode
2196                          * We need to distinguish those two, as the snapshot
2197                          * orphan must not get deleted.
2198                          * find_dead_roots already ran before us, so if this
2199                          * is a snapshot deletion, we should find the root
2200                          * in the dead_roots list
2201                          */
2202                         spin_lock(&fs_info->trans_lock);
2203                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2204                                             root_list) {
2205                                 if (dead_root->root_key.objectid ==
2206                                     found_key.objectid) {
2207                                         is_dead_root = 1;
2208                                         break;
2209                                 }
2210                         }
2211                         spin_unlock(&fs_info->trans_lock);
2212                         if (is_dead_root) {
2213                                 /* prevent this orphan from being found again */
2214                                 key.offset = found_key.objectid - 1;
2215                                 continue;
2216                         }
2217                 }
2218                 /*
2219                  * Inode is already gone but the orphan item is still there,
2220                  * kill the orphan item.
2221                  */
2222                 if (ret == -ESTALE) {
2223                         trans = btrfs_start_transaction(root, 1);
2224                         if (IS_ERR(trans)) {
2225                                 ret = PTR_ERR(trans);
2226                                 goto out;
2227                         }
2228                         ret = btrfs_del_orphan_item(trans, root,
2229                                                     found_key.objectid);
2230                         BUG_ON(ret);
2231                         btrfs_end_transaction(trans, root);
2232                         continue;
2233                 }
2234
2235                 /*
2236                  * add this inode to the orphan list so btrfs_orphan_del does
2237                  * the proper thing when we hit it
2238                  */
2239                 spin_lock(&root->orphan_lock);
2240                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2241                 spin_unlock(&root->orphan_lock);
2242
2243                 /* if we have links, this was a truncate, lets do that */
2244                 if (inode->i_nlink) {
2245                         if (!S_ISREG(inode->i_mode)) {
2246                                 WARN_ON(1);
2247                                 iput(inode);
2248                                 continue;
2249                         }
2250                         nr_truncate++;
2251                         ret = btrfs_truncate(inode);
2252                 } else {
2253                         nr_unlink++;
2254                 }
2255
2256                 /* this will do delete_inode and everything for us */
2257                 iput(inode);
2258                 if (ret)
2259                         goto out;
2260         }
2261         /* release the path since we're done with it */
2262         btrfs_release_path(path);
2263
2264         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2265
2266         if (root->orphan_block_rsv)
2267                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2268                                         (u64)-1);
2269
2270         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2271                 trans = btrfs_join_transaction(root);
2272                 if (!IS_ERR(trans))
2273                         btrfs_end_transaction(trans, root);
2274         }
2275
2276         if (nr_unlink)
2277                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2278         if (nr_truncate)
2279                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2280
2281 out:
2282         if (ret)
2283                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2284         btrfs_free_path(path);
2285         return ret;
2286 }
2287
2288 /*
2289  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2290  * don't find any xattrs, we know there can't be any acls.
2291  *
2292  * slot is the slot the inode is in, objectid is the objectid of the inode
2293  */
2294 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2295                                           int slot, u64 objectid)
2296 {
2297         u32 nritems = btrfs_header_nritems(leaf);
2298         struct btrfs_key found_key;
2299         int scanned = 0;
2300
2301         slot++;
2302         while (slot < nritems) {
2303                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2304
2305                 /* we found a different objectid, there must not be acls */
2306                 if (found_key.objectid != objectid)
2307                         return 0;
2308
2309                 /* we found an xattr, assume we've got an acl */
2310                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2311                         return 1;
2312
2313                 /*
2314                  * we found a key greater than an xattr key, there can't
2315                  * be any acls later on
2316                  */
2317                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2318                         return 0;
2319
2320                 slot++;
2321                 scanned++;
2322
2323                 /*
2324                  * it goes inode, inode backrefs, xattrs, extents,
2325                  * so if there are a ton of hard links to an inode there can
2326                  * be a lot of backrefs.  Don't waste time searching too hard,
2327                  * this is just an optimization
2328                  */
2329                 if (scanned >= 8)
2330                         break;
2331         }
2332         /* we hit the end of the leaf before we found an xattr or
2333          * something larger than an xattr.  We have to assume the inode
2334          * has acls
2335          */
2336         return 1;
2337 }
2338
2339 /*
2340  * read an inode from the btree into the in-memory inode
2341  */
2342 static void btrfs_read_locked_inode(struct inode *inode)
2343 {
2344         struct btrfs_path *path;
2345         struct extent_buffer *leaf;
2346         struct btrfs_inode_item *inode_item;
2347         struct btrfs_timespec *tspec;
2348         struct btrfs_root *root = BTRFS_I(inode)->root;
2349         struct btrfs_key location;
2350         int maybe_acls;
2351         u32 rdev;
2352         int ret;
2353         bool filled = false;
2354
2355         ret = btrfs_fill_inode(inode, &rdev);
2356         if (!ret)
2357                 filled = true;
2358
2359         path = btrfs_alloc_path();
2360         if (!path)
2361                 goto make_bad;
2362
2363         path->leave_spinning = 1;
2364         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2365
2366         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2367         if (ret)
2368                 goto make_bad;
2369
2370         leaf = path->nodes[0];
2371
2372         if (filled)
2373                 goto cache_acl;
2374
2375         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2376                                     struct btrfs_inode_item);
2377         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2378         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2379         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2380         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2381         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2382
2383         tspec = btrfs_inode_atime(inode_item);
2384         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2385         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2386
2387         tspec = btrfs_inode_mtime(inode_item);
2388         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2389         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2390
2391         tspec = btrfs_inode_ctime(inode_item);
2392         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2393         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2394
2395         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2396         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2397         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2398         inode->i_generation = BTRFS_I(inode)->generation;
2399         inode->i_rdev = 0;
2400         rdev = btrfs_inode_rdev(leaf, inode_item);
2401
2402         BTRFS_I(inode)->index_cnt = (u64)-1;
2403         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2404 cache_acl:
2405         /*
2406          * try to precache a NULL acl entry for files that don't have
2407          * any xattrs or acls
2408          */
2409         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2410                                            btrfs_ino(inode));
2411         if (!maybe_acls)
2412                 cache_no_acl(inode);
2413
2414         btrfs_free_path(path);
2415
2416         switch (inode->i_mode & S_IFMT) {
2417         case S_IFREG:
2418                 inode->i_mapping->a_ops = &btrfs_aops;
2419                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2420                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2421                 inode->i_fop = &btrfs_file_operations;
2422                 inode->i_op = &btrfs_file_inode_operations;
2423                 break;
2424         case S_IFDIR:
2425                 inode->i_fop = &btrfs_dir_file_operations;
2426                 if (root == root->fs_info->tree_root)
2427                         inode->i_op = &btrfs_dir_ro_inode_operations;
2428                 else
2429                         inode->i_op = &btrfs_dir_inode_operations;
2430                 break;
2431         case S_IFLNK:
2432                 inode->i_op = &btrfs_symlink_inode_operations;
2433                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2434                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2435                 break;
2436         default:
2437                 inode->i_op = &btrfs_special_inode_operations;
2438                 init_special_inode(inode, inode->i_mode, rdev);
2439                 break;
2440         }
2441
2442         btrfs_update_iflags(inode);
2443         return;
2444
2445 make_bad:
2446         btrfs_free_path(path);
2447         make_bad_inode(inode);
2448 }
2449
2450 /*
2451  * given a leaf and an inode, copy the inode fields into the leaf
2452  */
2453 static void fill_inode_item(struct btrfs_trans_handle *trans,
2454                             struct extent_buffer *leaf,
2455                             struct btrfs_inode_item *item,
2456                             struct inode *inode)
2457 {
2458         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2459         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2460         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2461         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2462         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2463
2464         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2465                                inode->i_atime.tv_sec);
2466         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2467                                 inode->i_atime.tv_nsec);
2468
2469         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2470                                inode->i_mtime.tv_sec);
2471         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2472                                 inode->i_mtime.tv_nsec);
2473
2474         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2475                                inode->i_ctime.tv_sec);
2476         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2477                                 inode->i_ctime.tv_nsec);
2478
2479         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2480         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2481         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2482         btrfs_set_inode_transid(leaf, item, trans->transid);
2483         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2484         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2485         btrfs_set_inode_block_group(leaf, item, 0);
2486 }
2487
2488 /*
2489  * copy everything in the in-memory inode into the btree.
2490  */
2491 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2492                                 struct btrfs_root *root, struct inode *inode)
2493 {
2494         struct btrfs_inode_item *inode_item;
2495         struct btrfs_path *path;
2496         struct extent_buffer *leaf;
2497         int ret;
2498
2499         path = btrfs_alloc_path();
2500         if (!path)
2501                 return -ENOMEM;
2502
2503         path->leave_spinning = 1;
2504         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2505                                  1);
2506         if (ret) {
2507                 if (ret > 0)
2508                         ret = -ENOENT;
2509                 goto failed;
2510         }
2511
2512         btrfs_unlock_up_safe(path, 1);
2513         leaf = path->nodes[0];
2514         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2515                                     struct btrfs_inode_item);
2516
2517         fill_inode_item(trans, leaf, inode_item, inode);
2518         btrfs_mark_buffer_dirty(leaf);
2519         btrfs_set_inode_last_trans(trans, inode);
2520         ret = 0;
2521 failed:
2522         btrfs_free_path(path);
2523         return ret;
2524 }
2525
2526 /*
2527  * copy everything in the in-memory inode into the btree.
2528  */
2529 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2530                                 struct btrfs_root *root, struct inode *inode)
2531 {
2532         int ret;
2533
2534         /*
2535          * If the inode is a free space inode, we can deadlock during commit
2536          * if we put it into the delayed code.
2537          *
2538          * The data relocation inode should also be directly updated
2539          * without delay
2540          */
2541         if (!btrfs_is_free_space_inode(root, inode)
2542             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2543                 ret = btrfs_delayed_update_inode(trans, root, inode);
2544                 if (!ret)
2545                         btrfs_set_inode_last_trans(trans, inode);
2546                 return ret;
2547         }
2548
2549         return btrfs_update_inode_item(trans, root, inode);
2550 }
2551
2552 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2553                                 struct btrfs_root *root, struct inode *inode)
2554 {
2555         int ret;
2556
2557         ret = btrfs_update_inode(trans, root, inode);
2558         if (ret == -ENOSPC)
2559                 return btrfs_update_inode_item(trans, root, inode);
2560         return ret;
2561 }
2562
2563 /*
2564  * unlink helper that gets used here in inode.c and in the tree logging
2565  * recovery code.  It remove a link in a directory with a given name, and
2566  * also drops the back refs in the inode to the directory
2567  */
2568 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2569                                 struct btrfs_root *root,
2570                                 struct inode *dir, struct inode *inode,
2571                                 const char *name, int name_len)
2572 {
2573         struct btrfs_path *path;
2574         int ret = 0;
2575         struct extent_buffer *leaf;
2576         struct btrfs_dir_item *di;
2577         struct btrfs_key key;
2578         u64 index;
2579         u64 ino = btrfs_ino(inode);
2580         u64 dir_ino = btrfs_ino(dir);
2581
2582         path = btrfs_alloc_path();
2583         if (!path) {
2584                 ret = -ENOMEM;
2585                 goto out;
2586         }
2587
2588         path->leave_spinning = 1;
2589         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2590                                     name, name_len, -1);
2591         if (IS_ERR(di)) {
2592                 ret = PTR_ERR(di);
2593                 goto err;
2594         }
2595         if (!di) {
2596                 ret = -ENOENT;
2597                 goto err;
2598         }
2599         leaf = path->nodes[0];
2600         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2601         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2602         if (ret)
2603                 goto err;
2604         btrfs_release_path(path);
2605
2606         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2607                                   dir_ino, &index);
2608         if (ret) {
2609                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2610                        "inode %llu parent %llu\n", name_len, name,
2611                        (unsigned long long)ino, (unsigned long long)dir_ino);
2612                 goto err;
2613         }
2614
2615         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2616         if (ret)
2617                 goto err;
2618
2619         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2620                                          inode, dir_ino);
2621         BUG_ON(ret != 0 && ret != -ENOENT);
2622
2623         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2624                                            dir, index);
2625         if (ret == -ENOENT)
2626                 ret = 0;
2627 err:
2628         btrfs_free_path(path);
2629         if (ret)
2630                 goto out;
2631
2632         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2633         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2634         btrfs_update_inode(trans, root, dir);
2635 out:
2636         return ret;
2637 }
2638
2639 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2640                        struct btrfs_root *root,
2641                        struct inode *dir, struct inode *inode,
2642                        const char *name, int name_len)
2643 {
2644         int ret;
2645         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2646         if (!ret) {
2647                 btrfs_drop_nlink(inode);
2648                 ret = btrfs_update_inode(trans, root, inode);
2649         }
2650         return ret;
2651 }
2652
2653
2654 /* helper to check if there is any shared block in the path */
2655 static int check_path_shared(struct btrfs_root *root,
2656                              struct btrfs_path *path)
2657 {
2658         struct extent_buffer *eb;
2659         int level;
2660         u64 refs = 1;
2661
2662         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2663                 int ret;
2664
2665                 if (!path->nodes[level])
2666                         break;
2667                 eb = path->nodes[level];
2668                 if (!btrfs_block_can_be_shared(root, eb))
2669                         continue;
2670                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2671                                                &refs, NULL);
2672                 if (refs > 1)
2673                         return 1;
2674         }
2675         return 0;
2676 }
2677
2678 /*
2679  * helper to start transaction for unlink and rmdir.
2680  *
2681  * unlink and rmdir are special in btrfs, they do not always free space.
2682  * so in enospc case, we should make sure they will free space before
2683  * allowing them to use the global metadata reservation.
2684  */
2685 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2686                                                        struct dentry *dentry)
2687 {
2688         struct btrfs_trans_handle *trans;
2689         struct btrfs_root *root = BTRFS_I(dir)->root;
2690         struct btrfs_path *path;
2691         struct btrfs_inode_ref *ref;
2692         struct btrfs_dir_item *di;
2693         struct inode *inode = dentry->d_inode;
2694         u64 index;
2695         int check_link = 1;
2696         int err = -ENOSPC;
2697         int ret;
2698         u64 ino = btrfs_ino(inode);
2699         u64 dir_ino = btrfs_ino(dir);
2700
2701         /*
2702          * 1 for the possible orphan item
2703          * 1 for the dir item
2704          * 1 for the dir index
2705          * 1 for the inode ref
2706          * 1 for the inode ref in the tree log
2707          * 2 for the dir entries in the log
2708          * 1 for the inode
2709          */
2710         trans = btrfs_start_transaction(root, 8);
2711         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2712                 return trans;
2713
2714         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2715                 return ERR_PTR(-ENOSPC);
2716
2717         /* check if there is someone else holds reference */
2718         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2719                 return ERR_PTR(-ENOSPC);
2720
2721         if (atomic_read(&inode->i_count) > 2)
2722                 return ERR_PTR(-ENOSPC);
2723
2724         if (xchg(&root->fs_info->enospc_unlink, 1))
2725                 return ERR_PTR(-ENOSPC);
2726
2727         path = btrfs_alloc_path();
2728         if (!path) {
2729                 root->fs_info->enospc_unlink = 0;
2730                 return ERR_PTR(-ENOMEM);
2731         }
2732
2733         /* 1 for the orphan item */
2734         trans = btrfs_start_transaction(root, 1);
2735         if (IS_ERR(trans)) {
2736                 btrfs_free_path(path);
2737                 root->fs_info->enospc_unlink = 0;
2738                 return trans;
2739         }
2740
2741         path->skip_locking = 1;
2742         path->search_commit_root = 1;
2743
2744         ret = btrfs_lookup_inode(trans, root, path,
2745                                 &BTRFS_I(dir)->location, 0);
2746         if (ret < 0) {
2747                 err = ret;
2748                 goto out;
2749         }
2750         if (ret == 0) {
2751                 if (check_path_shared(root, path))
2752                         goto out;
2753         } else {
2754                 check_link = 0;
2755         }
2756         btrfs_release_path(path);
2757
2758         ret = btrfs_lookup_inode(trans, root, path,
2759                                 &BTRFS_I(inode)->location, 0);
2760         if (ret < 0) {
2761                 err = ret;
2762                 goto out;
2763         }
2764         if (ret == 0) {
2765                 if (check_path_shared(root, path))
2766                         goto out;
2767         } else {
2768                 check_link = 0;
2769         }
2770         btrfs_release_path(path);
2771
2772         if (ret == 0 && S_ISREG(inode->i_mode)) {
2773                 ret = btrfs_lookup_file_extent(trans, root, path,
2774                                                ino, (u64)-1, 0);
2775                 if (ret < 0) {
2776                         err = ret;
2777                         goto out;
2778                 }
2779                 BUG_ON(ret == 0);
2780                 if (check_path_shared(root, path))
2781                         goto out;
2782                 btrfs_release_path(path);
2783         }
2784
2785         if (!check_link) {
2786                 err = 0;
2787                 goto out;
2788         }
2789
2790         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2791                                 dentry->d_name.name, dentry->d_name.len, 0);
2792         if (IS_ERR(di)) {
2793                 err = PTR_ERR(di);
2794                 goto out;
2795         }
2796         if (di) {
2797                 if (check_path_shared(root, path))
2798                         goto out;
2799         } else {
2800                 err = 0;
2801                 goto out;
2802         }
2803         btrfs_release_path(path);
2804
2805         ref = btrfs_lookup_inode_ref(trans, root, path,
2806                                 dentry->d_name.name, dentry->d_name.len,
2807                                 ino, dir_ino, 0);
2808         if (IS_ERR(ref)) {
2809                 err = PTR_ERR(ref);
2810                 goto out;
2811         }
2812         BUG_ON(!ref);
2813         if (check_path_shared(root, path))
2814                 goto out;
2815         index = btrfs_inode_ref_index(path->nodes[0], ref);
2816         btrfs_release_path(path);
2817
2818         /*
2819          * This is a commit root search, if we can lookup inode item and other
2820          * relative items in the commit root, it means the transaction of
2821          * dir/file creation has been committed, and the dir index item that we
2822          * delay to insert has also been inserted into the commit root. So
2823          * we needn't worry about the delayed insertion of the dir index item
2824          * here.
2825          */
2826         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2827                                 dentry->d_name.name, dentry->d_name.len, 0);
2828         if (IS_ERR(di)) {
2829                 err = PTR_ERR(di);
2830                 goto out;
2831         }
2832         BUG_ON(ret == -ENOENT);
2833         if (check_path_shared(root, path))
2834                 goto out;
2835
2836         err = 0;
2837 out:
2838         btrfs_free_path(path);
2839         /* Migrate the orphan reservation over */
2840         if (!err)
2841                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2842                                 &root->fs_info->global_block_rsv,
2843                                 trans->bytes_reserved);
2844
2845         if (err) {
2846                 btrfs_end_transaction(trans, root);
2847                 root->fs_info->enospc_unlink = 0;
2848                 return ERR_PTR(err);
2849         }
2850
2851         trans->block_rsv = &root->fs_info->global_block_rsv;
2852         return trans;
2853 }
2854
2855 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2856                                struct btrfs_root *root)
2857 {
2858         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2859                 btrfs_block_rsv_release(root, trans->block_rsv,
2860                                         trans->bytes_reserved);
2861                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2862                 BUG_ON(!root->fs_info->enospc_unlink);
2863                 root->fs_info->enospc_unlink = 0;
2864         }
2865         btrfs_end_transaction(trans, root);
2866 }
2867
2868 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2869 {
2870         struct btrfs_root *root = BTRFS_I(dir)->root;
2871         struct btrfs_trans_handle *trans;
2872         struct inode *inode = dentry->d_inode;
2873         int ret;
2874         unsigned long nr = 0;
2875
2876         trans = __unlink_start_trans(dir, dentry);
2877         if (IS_ERR(trans))
2878                 return PTR_ERR(trans);
2879
2880         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2881
2882         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2883                                  dentry->d_name.name, dentry->d_name.len);
2884         if (ret)
2885                 goto out;
2886
2887         if (inode->i_nlink == 0) {
2888                 ret = btrfs_orphan_add(trans, inode);
2889                 if (ret)
2890                         goto out;
2891         }
2892
2893 out:
2894         nr = trans->blocks_used;
2895         __unlink_end_trans(trans, root);
2896         btrfs_btree_balance_dirty(root, nr);
2897         return ret;
2898 }
2899
2900 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2901                         struct btrfs_root *root,
2902                         struct inode *dir, u64 objectid,
2903                         const char *name, int name_len)
2904 {
2905         struct btrfs_path *path;
2906         struct extent_buffer *leaf;
2907         struct btrfs_dir_item *di;
2908         struct btrfs_key key;
2909         u64 index;
2910         int ret;
2911         u64 dir_ino = btrfs_ino(dir);
2912
2913         path = btrfs_alloc_path();
2914         if (!path)
2915                 return -ENOMEM;
2916
2917         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2918                                    name, name_len, -1);
2919         BUG_ON(IS_ERR_OR_NULL(di));
2920
2921         leaf = path->nodes[0];
2922         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2923         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2924         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2925         BUG_ON(ret);
2926         btrfs_release_path(path);
2927
2928         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2929                                  objectid, root->root_key.objectid,
2930                                  dir_ino, &index, name, name_len);
2931         if (ret < 0) {
2932                 BUG_ON(ret != -ENOENT);
2933                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2934                                                  name, name_len);
2935                 BUG_ON(IS_ERR_OR_NULL(di));
2936
2937                 leaf = path->nodes[0];
2938                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2939                 btrfs_release_path(path);
2940                 index = key.offset;
2941         }
2942         btrfs_release_path(path);
2943
2944         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2945         BUG_ON(ret);
2946
2947         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2948         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2949         ret = btrfs_update_inode(trans, root, dir);
2950         BUG_ON(ret);
2951
2952         btrfs_free_path(path);
2953         return 0;
2954 }
2955
2956 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2957 {
2958         struct inode *inode = dentry->d_inode;
2959         int err = 0;
2960         struct btrfs_root *root = BTRFS_I(dir)->root;
2961         struct btrfs_trans_handle *trans;
2962         unsigned long nr = 0;
2963
2964         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2965             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2966                 return -ENOTEMPTY;
2967
2968         trans = __unlink_start_trans(dir, dentry);
2969         if (IS_ERR(trans))
2970                 return PTR_ERR(trans);
2971
2972         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2973                 err = btrfs_unlink_subvol(trans, root, dir,
2974                                           BTRFS_I(inode)->location.objectid,
2975                                           dentry->d_name.name,
2976                                           dentry->d_name.len);
2977                 goto out;
2978         }
2979
2980         err = btrfs_orphan_add(trans, inode);
2981         if (err)
2982                 goto out;
2983
2984         /* now the directory is empty */
2985         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2986                                  dentry->d_name.name, dentry->d_name.len);
2987         if (!err)
2988                 btrfs_i_size_write(inode, 0);
2989 out:
2990         nr = trans->blocks_used;
2991         __unlink_end_trans(trans, root);
2992         btrfs_btree_balance_dirty(root, nr);
2993
2994         return err;
2995 }
2996
2997 /*
2998  * this can truncate away extent items, csum items and directory items.
2999  * It starts at a high offset and removes keys until it can't find
3000  * any higher than new_size
3001  *
3002  * csum items that cross the new i_size are truncated to the new size
3003  * as well.
3004  *
3005  * min_type is the minimum key type to truncate down to.  If set to 0, this
3006  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3007  */
3008 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3009                                struct btrfs_root *root,
3010                                struct inode *inode,
3011                                u64 new_size, u32 min_type)
3012 {
3013         struct btrfs_path *path;
3014         struct extent_buffer *leaf;
3015         struct btrfs_file_extent_item *fi;
3016         struct btrfs_key key;
3017         struct btrfs_key found_key;
3018         u64 extent_start = 0;
3019         u64 extent_num_bytes = 0;
3020         u64 extent_offset = 0;
3021         u64 item_end = 0;
3022         u64 mask = root->sectorsize - 1;
3023         u32 found_type = (u8)-1;
3024         int found_extent;
3025         int del_item;
3026         int pending_del_nr = 0;
3027         int pending_del_slot = 0;
3028         int extent_type = -1;
3029         int ret;
3030         int err = 0;
3031         u64 ino = btrfs_ino(inode);
3032
3033         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3034
3035         path = btrfs_alloc_path();
3036         if (!path)
3037                 return -ENOMEM;
3038         path->reada = -1;
3039
3040         if (root->ref_cows || root == root->fs_info->tree_root)
3041                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3042
3043         /*
3044          * This function is also used to drop the items in the log tree before
3045          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3046          * it is used to drop the loged items. So we shouldn't kill the delayed
3047          * items.
3048          */
3049         if (min_type == 0 && root == BTRFS_I(inode)->root)
3050                 btrfs_kill_delayed_inode_items(inode);
3051
3052         key.objectid = ino;
3053         key.offset = (u64)-1;
3054         key.type = (u8)-1;
3055
3056 search_again:
3057         path->leave_spinning = 1;
3058         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3059         if (ret < 0) {
3060                 err = ret;
3061                 goto out;
3062         }
3063
3064         if (ret > 0) {
3065                 /* there are no items in the tree for us to truncate, we're
3066                  * done
3067                  */
3068                 if (path->slots[0] == 0)
3069                         goto out;
3070                 path->slots[0]--;
3071         }
3072
3073         while (1) {
3074                 fi = NULL;
3075                 leaf = path->nodes[0];
3076                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3077                 found_type = btrfs_key_type(&found_key);
3078
3079                 if (found_key.objectid != ino)
3080                         break;
3081
3082                 if (found_type < min_type)
3083                         break;
3084
3085                 item_end = found_key.offset;
3086                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3087                         fi = btrfs_item_ptr(leaf, path->slots[0],
3088                                             struct btrfs_file_extent_item);
3089                         extent_type = btrfs_file_extent_type(leaf, fi);
3090                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3091                                 item_end +=
3092                                     btrfs_file_extent_num_bytes(leaf, fi);
3093                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3094                                 item_end += btrfs_file_extent_inline_len(leaf,
3095                                                                          fi);
3096                         }
3097                         item_end--;
3098                 }
3099                 if (found_type > min_type) {
3100                         del_item = 1;
3101                 } else {
3102                         if (item_end < new_size)
3103                                 break;
3104                         if (found_key.offset >= new_size)
3105                                 del_item = 1;
3106                         else
3107                                 del_item = 0;
3108                 }
3109                 found_extent = 0;
3110                 /* FIXME, shrink the extent if the ref count is only 1 */
3111                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3112                         goto delete;
3113
3114                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3115                         u64 num_dec;
3116                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3117                         if (!del_item) {
3118                                 u64 orig_num_bytes =
3119                                         btrfs_file_extent_num_bytes(leaf, fi);
3120                                 extent_num_bytes = new_size -
3121                                         found_key.offset + root->sectorsize - 1;
3122                                 extent_num_bytes = extent_num_bytes &
3123                                         ~((u64)root->sectorsize - 1);
3124                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3125                                                          extent_num_bytes);
3126                                 num_dec = (orig_num_bytes -
3127                                            extent_num_bytes);
3128                                 if (root->ref_cows && extent_start != 0)
3129                                         inode_sub_bytes(inode, num_dec);
3130                                 btrfs_mark_buffer_dirty(leaf);
3131                         } else {
3132                                 extent_num_bytes =
3133                                         btrfs_file_extent_disk_num_bytes(leaf,
3134                                                                          fi);
3135                                 extent_offset = found_key.offset -
3136                                         btrfs_file_extent_offset(leaf, fi);
3137
3138                                 /* FIXME blocksize != 4096 */
3139                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3140                                 if (extent_start != 0) {
3141                                         found_extent = 1;
3142                                         if (root->ref_cows)
3143                                                 inode_sub_bytes(inode, num_dec);
3144                                 }
3145                         }
3146                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3147                         /*
3148                          * we can't truncate inline items that have had
3149                          * special encodings
3150                          */
3151                         if (!del_item &&
3152                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3153                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3154                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3155                                 u32 size = new_size - found_key.offset;
3156
3157                                 if (root->ref_cows) {
3158                                         inode_sub_bytes(inode, item_end + 1 -
3159                                                         new_size);
3160                                 }
3161                                 size =
3162                                     btrfs_file_extent_calc_inline_size(size);
3163                                 ret = btrfs_truncate_item(trans, root, path,
3164                                                           size, 1);
3165                         } else if (root->ref_cows) {
3166                                 inode_sub_bytes(inode, item_end + 1 -
3167                                                 found_key.offset);
3168                         }
3169                 }
3170 delete:
3171                 if (del_item) {
3172                         if (!pending_del_nr) {
3173                                 /* no pending yet, add ourselves */
3174                                 pending_del_slot = path->slots[0];
3175                                 pending_del_nr = 1;
3176                         } else if (pending_del_nr &&
3177                                    path->slots[0] + 1 == pending_del_slot) {
3178                                 /* hop on the pending chunk */
3179                                 pending_del_nr++;
3180                                 pending_del_slot = path->slots[0];
3181                         } else {
3182                                 BUG();
3183                         }
3184                 } else {
3185                         break;
3186                 }
3187                 if (found_extent && (root->ref_cows ||
3188                                      root == root->fs_info->tree_root)) {
3189                         btrfs_set_path_blocking(path);
3190                         ret = btrfs_free_extent(trans, root, extent_start,
3191                                                 extent_num_bytes, 0,
3192                                                 btrfs_header_owner(leaf),
3193                                                 ino, extent_offset, 0);
3194                         BUG_ON(ret);
3195                 }
3196
3197                 if (found_type == BTRFS_INODE_ITEM_KEY)
3198                         break;
3199
3200                 if (path->slots[0] == 0 ||
3201                     path->slots[0] != pending_del_slot) {
3202                         if (root->ref_cows &&
3203                             BTRFS_I(inode)->location.objectid !=
3204                                                 BTRFS_FREE_INO_OBJECTID) {
3205                                 err = -EAGAIN;
3206                                 goto out;
3207                         }
3208                         if (pending_del_nr) {
3209                                 ret = btrfs_del_items(trans, root, path,
3210                                                 pending_del_slot,
3211                                                 pending_del_nr);
3212                                 BUG_ON(ret);
3213                                 pending_del_nr = 0;
3214                         }
3215                         btrfs_release_path(path);
3216                         goto search_again;
3217                 } else {
3218                         path->slots[0]--;
3219                 }
3220         }
3221 out:
3222         if (pending_del_nr) {
3223                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3224                                       pending_del_nr);
3225                 BUG_ON(ret);
3226         }
3227         btrfs_free_path(path);
3228         return err;
3229 }
3230
3231 /*
3232  * taken from block_truncate_page, but does cow as it zeros out
3233  * any bytes left in the last page in the file.
3234  */
3235 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3236 {
3237         struct inode *inode = mapping->host;
3238         struct btrfs_root *root = BTRFS_I(inode)->root;
3239         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3240         struct btrfs_ordered_extent *ordered;
3241         struct extent_state *cached_state = NULL;
3242         char *kaddr;
3243         u32 blocksize = root->sectorsize;
3244         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3245         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3246         struct page *page;
3247         gfp_t mask = btrfs_alloc_write_mask(mapping);
3248         int ret = 0;
3249         u64 page_start;
3250         u64 page_end;
3251
3252         if ((offset & (blocksize - 1)) == 0)
3253                 goto out;
3254         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3255         if (ret)
3256                 goto out;
3257
3258         ret = -ENOMEM;
3259 again:
3260         page = find_or_create_page(mapping, index, mask);
3261         if (!page) {
3262                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3263                 goto out;
3264         }
3265
3266         page_start = page_offset(page);
3267         page_end = page_start + PAGE_CACHE_SIZE - 1;
3268
3269         if (!PageUptodate(page)) {
3270                 ret = btrfs_readpage(NULL, page);
3271                 lock_page(page);
3272                 if (page->mapping != mapping) {
3273                         unlock_page(page);
3274                         page_cache_release(page);
3275                         goto again;
3276                 }
3277                 if (!PageUptodate(page)) {
3278                         ret = -EIO;
3279                         goto out_unlock;
3280                 }
3281         }
3282         wait_on_page_writeback(page);
3283
3284         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3285                          GFP_NOFS);
3286         set_page_extent_mapped(page);
3287
3288         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3289         if (ordered) {
3290                 unlock_extent_cached(io_tree, page_start, page_end,
3291                                      &cached_state, GFP_NOFS);
3292                 unlock_page(page);
3293                 page_cache_release(page);
3294                 btrfs_start_ordered_extent(inode, ordered, 1);
3295                 btrfs_put_ordered_extent(ordered);
3296                 goto again;
3297         }
3298
3299         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3300                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3301                           0, 0, &cached_state, GFP_NOFS);
3302
3303         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3304                                         &cached_state);
3305         if (ret) {
3306                 unlock_extent_cached(io_tree, page_start, page_end,
3307                                      &cached_state, GFP_NOFS);
3308                 goto out_unlock;
3309         }
3310
3311         ret = 0;
3312         if (offset != PAGE_CACHE_SIZE) {
3313                 kaddr = kmap(page);
3314                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3315                 flush_dcache_page(page);
3316                 kunmap(page);
3317         }
3318         ClearPageChecked(page);
3319         set_page_dirty(page);
3320         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3321                              GFP_NOFS);
3322
3323 out_unlock:
3324         if (ret)
3325                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3326         unlock_page(page);
3327         page_cache_release(page);
3328 out:
3329         return ret;
3330 }
3331
3332 /*
3333  * This function puts in dummy file extents for the area we're creating a hole
3334  * for.  So if we are truncating this file to a larger size we need to insert
3335  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3336  * the range between oldsize and size
3337  */
3338 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3339 {
3340         struct btrfs_trans_handle *trans;
3341         struct btrfs_root *root = BTRFS_I(inode)->root;
3342         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3343         struct extent_map *em = NULL;
3344         struct extent_state *cached_state = NULL;
3345         u64 mask = root->sectorsize - 1;
3346         u64 hole_start = (oldsize + mask) & ~mask;
3347         u64 block_end = (size + mask) & ~mask;
3348         u64 last_byte;
3349         u64 cur_offset;
3350         u64 hole_size;
3351         int err = 0;
3352
3353         if (size <= hole_start)
3354                 return 0;
3355
3356         while (1) {
3357                 struct btrfs_ordered_extent *ordered;
3358                 btrfs_wait_ordered_range(inode, hole_start,
3359                                          block_end - hole_start);
3360                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3361                                  &cached_state, GFP_NOFS);
3362                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3363                 if (!ordered)
3364                         break;
3365                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3366                                      &cached_state, GFP_NOFS);
3367                 btrfs_put_ordered_extent(ordered);
3368         }
3369
3370         cur_offset = hole_start;
3371         while (1) {
3372                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3373                                 block_end - cur_offset, 0);
3374                 BUG_ON(IS_ERR_OR_NULL(em));
3375                 last_byte = min(extent_map_end(em), block_end);
3376                 last_byte = (last_byte + mask) & ~mask;
3377                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3378                         u64 hint_byte = 0;
3379                         hole_size = last_byte - cur_offset;
3380
3381                         trans = btrfs_start_transaction(root, 3);
3382                         if (IS_ERR(trans)) {
3383                                 err = PTR_ERR(trans);
3384                                 break;
3385                         }
3386
3387                         err = btrfs_drop_extents(trans, inode, cur_offset,
3388                                                  cur_offset + hole_size,
3389                                                  &hint_byte, 1);
3390                         if (err) {
3391                                 btrfs_update_inode(trans, root, inode);
3392                                 btrfs_end_transaction(trans, root);
3393                                 break;
3394                         }
3395
3396                         err = btrfs_insert_file_extent(trans, root,
3397                                         btrfs_ino(inode), cur_offset, 0,
3398                                         0, hole_size, 0, hole_size,
3399                                         0, 0, 0);
3400                         if (err) {
3401                                 btrfs_update_inode(trans, root, inode);
3402                                 btrfs_end_transaction(trans, root);
3403                                 break;
3404                         }
3405
3406                         btrfs_drop_extent_cache(inode, hole_start,
3407                                         last_byte - 1, 0);
3408
3409                         btrfs_update_inode(trans, root, inode);
3410                         btrfs_end_transaction(trans, root);
3411                 }
3412                 free_extent_map(em);
3413                 em = NULL;
3414                 cur_offset = last_byte;
3415                 if (cur_offset >= block_end)
3416                         break;
3417         }
3418
3419         free_extent_map(em);
3420         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3421                              GFP_NOFS);
3422         return err;
3423 }
3424
3425 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3426 {
3427         struct btrfs_root *root = BTRFS_I(inode)->root;
3428         struct btrfs_trans_handle *trans;
3429         loff_t oldsize = i_size_read(inode);
3430         int ret;
3431
3432         if (newsize == oldsize)
3433                 return 0;
3434
3435         if (newsize > oldsize) {
3436                 truncate_pagecache(inode, oldsize, newsize);
3437                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3438                 if (ret)
3439                         return ret;
3440
3441                 trans = btrfs_start_transaction(root, 1);
3442                 if (IS_ERR(trans))
3443                         return PTR_ERR(trans);
3444
3445                 i_size_write(inode, newsize);
3446                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3447                 ret = btrfs_update_inode(trans, root, inode);
3448                 btrfs_end_transaction(trans, root);
3449         } else {
3450
3451                 /*
3452                  * We're truncating a file that used to have good data down to
3453                  * zero. Make sure it gets into the ordered flush list so that
3454                  * any new writes get down to disk quickly.
3455                  */
3456                 if (newsize == 0)
3457                         BTRFS_I(inode)->ordered_data_close = 1;
3458
3459                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3460                 truncate_setsize(inode, newsize);
3461                 ret = btrfs_truncate(inode);
3462         }
3463
3464         return ret;
3465 }
3466
3467 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3468 {
3469         struct inode *inode = dentry->d_inode;
3470         struct btrfs_root *root = BTRFS_I(inode)->root;
3471         int err;
3472
3473         if (btrfs_root_readonly(root))
3474                 return -EROFS;
3475
3476         err = inode_change_ok(inode, attr);
3477         if (err)
3478                 return err;
3479
3480         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3481                 err = btrfs_setsize(inode, attr->ia_size);
3482                 if (err)
3483                         return err;
3484         }
3485
3486         if (attr->ia_valid) {
3487                 setattr_copy(inode, attr);
3488                 err = btrfs_dirty_inode(inode);
3489
3490                 if (!err && attr->ia_valid & ATTR_MODE)
3491                         err = btrfs_acl_chmod(inode);
3492         }
3493
3494         return err;
3495 }
3496
3497 void btrfs_evict_inode(struct inode *inode)
3498 {
3499         struct btrfs_trans_handle *trans;
3500         struct btrfs_root *root = BTRFS_I(inode)->root;
3501         struct btrfs_block_rsv *rsv, *global_rsv;
3502         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3503         unsigned long nr;
3504         int ret;
3505
3506         trace_btrfs_inode_evict(inode);
3507
3508         truncate_inode_pages(&inode->i_data, 0);
3509         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3510                                btrfs_is_free_space_inode(root, inode)))
3511                 goto no_delete;
3512
3513         if (is_bad_inode(inode)) {
3514                 btrfs_orphan_del(NULL, inode);
3515                 goto no_delete;
3516         }
3517         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3518         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3519
3520         if (root->fs_info->log_root_recovering) {
3521                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3522                 goto no_delete;
3523         }
3524
3525         if (inode->i_nlink > 0) {
3526                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3527                 goto no_delete;
3528         }
3529
3530         rsv = btrfs_alloc_block_rsv(root);
3531         if (!rsv) {
3532                 btrfs_orphan_del(NULL, inode);
3533                 goto no_delete;
3534         }
3535         rsv->size = min_size;
3536         global_rsv = &root->fs_info->global_block_rsv;
3537
3538         btrfs_i_size_write(inode, 0);
3539
3540         /*
3541          * This is a bit simpler than btrfs_truncate since
3542          *
3543          * 1) We've already reserved our space for our orphan item in the
3544          *    unlink.
3545          * 2) We're going to delete the inode item, so we don't need to update
3546          *    it at all.
3547          *
3548          * So we just need to reserve some slack space in case we add bytes when
3549          * doing the truncate.
3550          */
3551         while (1) {
3552                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3553
3554                 /*
3555                  * Try and steal from the global reserve since we will
3556                  * likely not use this space anyway, we want to try as
3557                  * hard as possible to get this to work.
3558                  */
3559                 if (ret)
3560                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3561
3562                 if (ret) {
3563                         printk(KERN_WARNING "Could not get space for a "
3564                                "delete, will truncate on mount %d\n", ret);
3565                         btrfs_orphan_del(NULL, inode);
3566                         btrfs_free_block_rsv(root, rsv);
3567                         goto no_delete;
3568                 }
3569
3570                 trans = btrfs_start_transaction(root, 0);
3571                 if (IS_ERR(trans)) {
3572                         btrfs_orphan_del(NULL, inode);
3573                         btrfs_free_block_rsv(root, rsv);
3574                         goto no_delete;
3575                 }
3576
3577                 trans->block_rsv = rsv;
3578
3579                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3580                 if (ret != -EAGAIN)
3581                         break;
3582
3583                 nr = trans->blocks_used;
3584                 btrfs_end_transaction(trans, root);
3585                 trans = NULL;
3586                 btrfs_btree_balance_dirty(root, nr);
3587         }
3588
3589         btrfs_free_block_rsv(root, rsv);
3590
3591         if (ret == 0) {
3592                 trans->block_rsv = root->orphan_block_rsv;
3593                 ret = btrfs_orphan_del(trans, inode);
3594                 BUG_ON(ret);
3595         }
3596
3597         trans->block_rsv = &root->fs_info->trans_block_rsv;
3598         if (!(root == root->fs_info->tree_root ||
3599               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3600                 btrfs_return_ino(root, btrfs_ino(inode));
3601
3602         nr = trans->blocks_used;
3603         btrfs_end_transaction(trans, root);
3604         btrfs_btree_balance_dirty(root, nr);
3605 no_delete:
3606         end_writeback(inode);
3607         return;
3608 }
3609
3610 /*
3611  * this returns the key found in the dir entry in the location pointer.
3612  * If no dir entries were found, location->objectid is 0.
3613  */
3614 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3615                                struct btrfs_key *location)
3616 {
3617         const char *name = dentry->d_name.name;
3618         int namelen = dentry->d_name.len;
3619         struct btrfs_dir_item *di;
3620         struct btrfs_path *path;
3621         struct btrfs_root *root = BTRFS_I(dir)->root;
3622         int ret = 0;
3623
3624         path = btrfs_alloc_path();
3625         if (!path)
3626                 return -ENOMEM;
3627
3628         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3629                                     namelen, 0);
3630         if (IS_ERR(di))
3631                 ret = PTR_ERR(di);
3632
3633         if (IS_ERR_OR_NULL(di))
3634                 goto out_err;
3635
3636         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3637 out:
3638         btrfs_free_path(path);
3639         return ret;
3640 out_err:
3641         location->objectid = 0;
3642         goto out;
3643 }
3644
3645 /*
3646  * when we hit a tree root in a directory, the btrfs part of the inode
3647  * needs to be changed to reflect the root directory of the tree root.  This
3648  * is kind of like crossing a mount point.
3649  */
3650 static int fixup_tree_root_location(struct btrfs_root *root,
3651                                     struct inode *dir,
3652                                     struct dentry *dentry,
3653                                     struct btrfs_key *location,
3654                                     struct btrfs_root **sub_root)
3655 {
3656         struct btrfs_path *path;
3657         struct btrfs_root *new_root;
3658         struct btrfs_root_ref *ref;
3659         struct extent_buffer *leaf;
3660         int ret;
3661         int err = 0;
3662
3663         path = btrfs_alloc_path();
3664         if (!path) {
3665                 err = -ENOMEM;
3666                 goto out;
3667         }
3668
3669         err = -ENOENT;
3670         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3671                                   BTRFS_I(dir)->root->root_key.objectid,
3672                                   location->objectid);
3673         if (ret) {
3674                 if (ret < 0)
3675                         err = ret;
3676                 goto out;
3677         }
3678
3679         leaf = path->nodes[0];
3680         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3681         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3682             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3683                 goto out;
3684
3685         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3686                                    (unsigned long)(ref + 1),
3687                                    dentry->d_name.len);
3688         if (ret)
3689                 goto out;
3690
3691         btrfs_release_path(path);
3692
3693         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3694         if (IS_ERR(new_root)) {
3695                 err = PTR_ERR(new_root);
3696                 goto out;
3697         }
3698
3699         if (btrfs_root_refs(&new_root->root_item) == 0) {
3700                 err = -ENOENT;
3701                 goto out;
3702         }
3703
3704         *sub_root = new_root;
3705         location->objectid = btrfs_root_dirid(&new_root->root_item);
3706         location->type = BTRFS_INODE_ITEM_KEY;
3707         location->offset = 0;
3708         err = 0;
3709 out:
3710         btrfs_free_path(path);
3711         return err;
3712 }
3713
3714 static void inode_tree_add(struct inode *inode)
3715 {
3716         struct btrfs_root *root = BTRFS_I(inode)->root;
3717         struct btrfs_inode *entry;
3718         struct rb_node **p;
3719         struct rb_node *parent;
3720         u64 ino = btrfs_ino(inode);
3721 again:
3722         p = &root->inode_tree.rb_node;
3723         parent = NULL;
3724
3725         if (inode_unhashed(inode))
3726                 return;
3727
3728         spin_lock(&root->inode_lock);
3729         while (*p) {
3730                 parent = *p;
3731                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3732
3733                 if (ino < btrfs_ino(&entry->vfs_inode))
3734                         p = &parent->rb_left;
3735                 else if (ino > btrfs_ino(&entry->vfs_inode))
3736                         p = &parent->rb_right;
3737                 else {
3738                         WARN_ON(!(entry->vfs_inode.i_state &
3739                                   (I_WILL_FREE | I_FREEING)));
3740                         rb_erase(parent, &root->inode_tree);
3741                         RB_CLEAR_NODE(parent);
3742                         spin_unlock(&root->inode_lock);
3743                         goto again;
3744                 }
3745         }
3746         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3747         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3748         spin_unlock(&root->inode_lock);
3749 }
3750
3751 static void inode_tree_del(struct inode *inode)
3752 {
3753         struct btrfs_root *root = BTRFS_I(inode)->root;
3754         int empty = 0;
3755
3756         spin_lock(&root->inode_lock);
3757         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3758                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3759                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3760                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3761         }
3762         spin_unlock(&root->inode_lock);
3763
3764         /*
3765          * Free space cache has inodes in the tree root, but the tree root has a
3766          * root_refs of 0, so this could end up dropping the tree root as a
3767          * snapshot, so we need the extra !root->fs_info->tree_root check to
3768          * make sure we don't drop it.
3769          */
3770         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3771             root != root->fs_info->tree_root) {
3772                 synchronize_srcu(&root->fs_info->subvol_srcu);
3773                 spin_lock(&root->inode_lock);
3774                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3775                 spin_unlock(&root->inode_lock);
3776                 if (empty)
3777                         btrfs_add_dead_root(root);
3778         }
3779 }
3780
3781 int btrfs_invalidate_inodes(struct btrfs_root *root)
3782 {
3783         struct rb_node *node;
3784         struct rb_node *prev;
3785         struct btrfs_inode *entry;
3786         struct inode *inode;
3787         u64 objectid = 0;
3788
3789         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3790
3791         spin_lock(&root->inode_lock);
3792 again:
3793         node = root->inode_tree.rb_node;
3794         prev = NULL;
3795         while (node) {
3796                 prev = node;
3797                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3798
3799                 if (objectid < btrfs_ino(&entry->vfs_inode))
3800                         node = node->rb_left;
3801                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3802                         node = node->rb_right;
3803                 else
3804                         break;
3805         }
3806         if (!node) {
3807                 while (prev) {
3808                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3809                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3810                                 node = prev;
3811                                 break;
3812                         }
3813                         prev = rb_next(prev);
3814                 }
3815         }
3816         while (node) {
3817                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3818                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3819                 inode = igrab(&entry->vfs_inode);
3820                 if (inode) {
3821                         spin_unlock(&root->inode_lock);
3822                         if (atomic_read(&inode->i_count) > 1)
3823                                 d_prune_aliases(inode);
3824                         /*
3825                          * btrfs_drop_inode will have it removed from
3826                          * the inode cache when its usage count
3827                          * hits zero.
3828                          */
3829                         iput(inode);
3830                         cond_resched();
3831                         spin_lock(&root->inode_lock);
3832                         goto again;
3833                 }
3834
3835                 if (cond_resched_lock(&root->inode_lock))
3836                         goto again;
3837
3838                 node = rb_next(node);
3839         }
3840         spin_unlock(&root->inode_lock);
3841         return 0;
3842 }
3843
3844 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3845 {
3846         struct btrfs_iget_args *args = p;
3847         inode->i_ino = args->ino;
3848         BTRFS_I(inode)->root = args->root;
3849         btrfs_set_inode_space_info(args->root, inode);
3850         return 0;
3851 }
3852
3853 static int btrfs_find_actor(struct inode *inode, void *opaque)
3854 {
3855         struct btrfs_iget_args *args = opaque;
3856         return args->ino == btrfs_ino(inode) &&
3857                 args->root == BTRFS_I(inode)->root;
3858 }
3859
3860 static struct inode *btrfs_iget_locked(struct super_block *s,
3861                                        u64 objectid,
3862                                        struct btrfs_root *root)
3863 {
3864         struct inode *inode;
3865         struct btrfs_iget_args args;
3866         args.ino = objectid;
3867         args.root = root;
3868
3869         inode = iget5_locked(s, objectid, btrfs_find_actor,
3870                              btrfs_init_locked_inode,
3871                              (void *)&args);
3872         return inode;
3873 }
3874
3875 /* Get an inode object given its location and corresponding root.
3876  * Returns in *is_new if the inode was read from disk
3877  */
3878 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3879                          struct btrfs_root *root, int *new)
3880 {
3881         struct inode *inode;
3882
3883         inode = btrfs_iget_locked(s, location->objectid, root);
3884         if (!inode)
3885                 return ERR_PTR(-ENOMEM);
3886
3887         if (inode->i_state & I_NEW) {
3888                 BTRFS_I(inode)->root = root;
3889                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3890                 btrfs_read_locked_inode(inode);
3891                 if (!is_bad_inode(inode)) {
3892                         inode_tree_add(inode);
3893                         unlock_new_inode(inode);
3894                         if (new)
3895                                 *new = 1;
3896                 } else {
3897                         unlock_new_inode(inode);
3898                         iput(inode);
3899                         inode = ERR_PTR(-ESTALE);
3900                 }
3901         }
3902
3903         return inode;
3904 }
3905
3906 static struct inode *new_simple_dir(struct super_block *s,
3907                                     struct btrfs_key *key,
3908                                     struct btrfs_root *root)
3909 {
3910         struct inode *inode = new_inode(s);
3911
3912         if (!inode)
3913                 return ERR_PTR(-ENOMEM);
3914
3915         BTRFS_I(inode)->root = root;
3916         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3917         BTRFS_I(inode)->dummy_inode = 1;
3918
3919         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3920         inode->i_op = &simple_dir_inode_operations;
3921         inode->i_fop = &simple_dir_operations;
3922         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3923         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3924
3925         return inode;
3926 }
3927
3928 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3929 {
3930         struct inode *inode;
3931         struct btrfs_root *root = BTRFS_I(dir)->root;
3932         struct btrfs_root *sub_root = root;
3933         struct btrfs_key location;
3934         int index;
3935         int ret = 0;
3936
3937         if (dentry->d_name.len > BTRFS_NAME_LEN)
3938                 return ERR_PTR(-ENAMETOOLONG);
3939
3940         if (unlikely(d_need_lookup(dentry))) {
3941                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3942                 kfree(dentry->d_fsdata);
3943                 dentry->d_fsdata = NULL;
3944                 /* This thing is hashed, drop it for now */
3945                 d_drop(dentry);
3946         } else {
3947                 ret = btrfs_inode_by_name(dir, dentry, &location);
3948         }
3949
3950         if (ret < 0)
3951                 return ERR_PTR(ret);
3952
3953         if (location.objectid == 0)
3954                 return NULL;
3955
3956         if (location.type == BTRFS_INODE_ITEM_KEY) {
3957                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3958                 return inode;
3959         }
3960
3961         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3962
3963         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3964         ret = fixup_tree_root_location(root, dir, dentry,
3965                                        &location, &sub_root);
3966         if (ret < 0) {
3967                 if (ret != -ENOENT)
3968                         inode = ERR_PTR(ret);
3969                 else
3970                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3971         } else {
3972                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3973         }
3974         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3975
3976         if (!IS_ERR(inode) && root != sub_root) {
3977                 down_read(&root->fs_info->cleanup_work_sem);
3978                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3979                         ret = btrfs_orphan_cleanup(sub_root);
3980                 up_read(&root->fs_info->cleanup_work_sem);
3981                 if (ret)
3982                         inode = ERR_PTR(ret);
3983         }
3984
3985         return inode;
3986 }
3987
3988 static int btrfs_dentry_delete(const struct dentry *dentry)
3989 {
3990         struct btrfs_root *root;
3991
3992         if (!dentry->d_inode && !IS_ROOT(dentry))
3993                 dentry = dentry->d_parent;
3994
3995         if (dentry->d_inode) {
3996                 root = BTRFS_I(dentry->d_inode)->root;
3997                 if (btrfs_root_refs(&root->root_item) == 0)
3998                         return 1;
3999         }
4000         return 0;
4001 }
4002
4003 static void btrfs_dentry_release(struct dentry *dentry)
4004 {
4005         if (dentry->d_fsdata)
4006                 kfree(dentry->d_fsdata);
4007 }
4008
4009 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4010                                    struct nameidata *nd)
4011 {
4012         struct dentry *ret;
4013
4014         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4015         if (unlikely(d_need_lookup(dentry))) {
4016                 spin_lock(&dentry->d_lock);
4017                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4018                 spin_unlock(&dentry->d_lock);
4019         }
4020         return ret;
4021 }
4022
4023 unsigned char btrfs_filetype_table[] = {
4024         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4025 };
4026
4027 static int btrfs_real_readdir(struct file *filp, void *dirent,
4028                               filldir_t filldir)
4029 {
4030         struct inode *inode = filp->f_dentry->d_inode;
4031         struct btrfs_root *root = BTRFS_I(inode)->root;
4032         struct btrfs_item *item;
4033         struct btrfs_dir_item *di;
4034         struct btrfs_key key;
4035         struct btrfs_key found_key;
4036         struct btrfs_path *path;
4037         struct list_head ins_list;
4038         struct list_head del_list;
4039         struct qstr q;
4040         int ret;
4041         struct extent_buffer *leaf;
4042         int slot;
4043         unsigned char d_type;
4044         int over = 0;
4045         u32 di_cur;
4046         u32 di_total;
4047         u32 di_len;
4048         int key_type = BTRFS_DIR_INDEX_KEY;
4049         char tmp_name[32];
4050         char *name_ptr;
4051         int name_len;
4052         int is_curr = 0;        /* filp->f_pos points to the current index? */
4053
4054         /* FIXME, use a real flag for deciding about the key type */
4055         if (root->fs_info->tree_root == root)
4056                 key_type = BTRFS_DIR_ITEM_KEY;
4057
4058         /* special case for "." */
4059         if (filp->f_pos == 0) {
4060                 over = filldir(dirent, ".", 1,
4061                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4062                 if (over)
4063                         return 0;
4064                 filp->f_pos = 1;
4065         }
4066         /* special case for .., just use the back ref */
4067         if (filp->f_pos == 1) {
4068                 u64 pino = parent_ino(filp->f_path.dentry);
4069                 over = filldir(dirent, "..", 2,
4070                                filp->f_pos, pino, DT_DIR);
4071                 if (over)
4072                         return 0;
4073                 filp->f_pos = 2;
4074         }
4075         path = btrfs_alloc_path();
4076         if (!path)
4077                 return -ENOMEM;
4078
4079         path->reada = 1;
4080
4081         if (key_type == BTRFS_DIR_INDEX_KEY) {
4082                 INIT_LIST_HEAD(&ins_list);
4083                 INIT_LIST_HEAD(&del_list);
4084                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4085         }
4086
4087         btrfs_set_key_type(&key, key_type);
4088         key.offset = filp->f_pos;
4089         key.objectid = btrfs_ino(inode);
4090
4091         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4092         if (ret < 0)
4093                 goto err;
4094
4095         while (1) {
4096                 leaf = path->nodes[0];
4097                 slot = path->slots[0];
4098                 if (slot >= btrfs_header_nritems(leaf)) {
4099                         ret = btrfs_next_leaf(root, path);
4100                         if (ret < 0)
4101                                 goto err;
4102                         else if (ret > 0)
4103                                 break;
4104                         continue;
4105                 }
4106
4107                 item = btrfs_item_nr(leaf, slot);
4108                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4109
4110                 if (found_key.objectid != key.objectid)
4111                         break;
4112                 if (btrfs_key_type(&found_key) != key_type)
4113                         break;
4114                 if (found_key.offset < filp->f_pos)
4115                         goto next;
4116                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4117                     btrfs_should_delete_dir_index(&del_list,
4118                                                   found_key.offset))
4119                         goto next;
4120
4121                 filp->f_pos = found_key.offset;
4122                 is_curr = 1;
4123
4124                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4125                 di_cur = 0;
4126                 di_total = btrfs_item_size(leaf, item);
4127
4128                 while (di_cur < di_total) {
4129                         struct btrfs_key location;
4130                         struct dentry *tmp;
4131
4132                         if (verify_dir_item(root, leaf, di))
4133                                 break;
4134
4135                         name_len = btrfs_dir_name_len(leaf, di);
4136                         if (name_len <= sizeof(tmp_name)) {
4137                                 name_ptr = tmp_name;
4138                         } else {
4139                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4140                                 if (!name_ptr) {
4141                                         ret = -ENOMEM;
4142                                         goto err;
4143                                 }
4144                         }
4145                         read_extent_buffer(leaf, name_ptr,
4146                                            (unsigned long)(di + 1), name_len);
4147
4148                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4149                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4150
4151                         q.name = name_ptr;
4152                         q.len = name_len;
4153                         q.hash = full_name_hash(q.name, q.len);
4154                         tmp = d_lookup(filp->f_dentry, &q);
4155                         if (!tmp) {
4156                                 struct btrfs_key *newkey;
4157
4158                                 newkey = kzalloc(sizeof(struct btrfs_key),
4159                                                  GFP_NOFS);
4160                                 if (!newkey)
4161                                         goto no_dentry;
4162                                 tmp = d_alloc(filp->f_dentry, &q);
4163                                 if (!tmp) {
4164                                         kfree(newkey);
4165                                         dput(tmp);
4166                                         goto no_dentry;
4167                                 }
4168                                 memcpy(newkey, &location,
4169                                        sizeof(struct btrfs_key));
4170                                 tmp->d_fsdata = newkey;
4171                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4172                                 d_rehash(tmp);
4173                                 dput(tmp);
4174                         } else {
4175                                 dput(tmp);
4176                         }
4177 no_dentry:
4178                         /* is this a reference to our own snapshot? If so
4179                          * skip it
4180                          */
4181                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4182                             location.objectid == root->root_key.objectid) {
4183                                 over = 0;
4184                                 goto skip;
4185                         }
4186                         over = filldir(dirent, name_ptr, name_len,
4187                                        found_key.offset, location.objectid,
4188                                        d_type);
4189
4190 skip:
4191                         if (name_ptr != tmp_name)
4192                                 kfree(name_ptr);
4193
4194                         if (over)
4195                                 goto nopos;
4196                         di_len = btrfs_dir_name_len(leaf, di) +
4197                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4198                         di_cur += di_len;
4199                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4200                 }
4201 next:
4202                 path->slots[0]++;
4203         }
4204
4205         if (key_type == BTRFS_DIR_INDEX_KEY) {
4206                 if (is_curr)
4207                         filp->f_pos++;
4208                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4209                                                       &ins_list);
4210                 if (ret)
4211                         goto nopos;
4212         }
4213
4214         /* Reached end of directory/root. Bump pos past the last item. */
4215         if (key_type == BTRFS_DIR_INDEX_KEY)
4216                 /*
4217                  * 32-bit glibc will use getdents64, but then strtol -
4218                  * so the last number we can serve is this.
4219                  */
4220                 filp->f_pos = 0x7fffffff;
4221         else
4222                 filp->f_pos++;
4223 nopos:
4224         ret = 0;
4225 err:
4226         if (key_type == BTRFS_DIR_INDEX_KEY)
4227                 btrfs_put_delayed_items(&ins_list, &del_list);
4228         btrfs_free_path(path);
4229         return ret;
4230 }
4231
4232 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4233 {
4234         struct btrfs_root *root = BTRFS_I(inode)->root;
4235         struct btrfs_trans_handle *trans;
4236         int ret = 0;
4237         bool nolock = false;
4238
4239         if (BTRFS_I(inode)->dummy_inode)
4240                 return 0;
4241
4242         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4243                 nolock = true;
4244
4245         if (wbc->sync_mode == WB_SYNC_ALL) {
4246                 if (nolock)
4247                         trans = btrfs_join_transaction_nolock(root);
4248                 else
4249                         trans = btrfs_join_transaction(root);
4250                 if (IS_ERR(trans))
4251                         return PTR_ERR(trans);
4252                 if (nolock)
4253                         ret = btrfs_end_transaction_nolock(trans, root);
4254                 else
4255                         ret = btrfs_commit_transaction(trans, root);
4256         }
4257         return ret;
4258 }
4259
4260 /*
4261  * This is somewhat expensive, updating the tree every time the
4262  * inode changes.  But, it is most likely to find the inode in cache.
4263  * FIXME, needs more benchmarking...there are no reasons other than performance
4264  * to keep or drop this code.
4265  */
4266 int btrfs_dirty_inode(struct inode *inode)
4267 {
4268         struct btrfs_root *root = BTRFS_I(inode)->root;
4269         struct btrfs_trans_handle *trans;
4270         int ret;
4271
4272         if (BTRFS_I(inode)->dummy_inode)
4273                 return 0;
4274
4275         trans = btrfs_join_transaction(root);
4276         if (IS_ERR(trans))
4277                 return PTR_ERR(trans);
4278
4279         ret = btrfs_update_inode(trans, root, inode);
4280         if (ret && ret == -ENOSPC) {
4281                 /* whoops, lets try again with the full transaction */
4282                 btrfs_end_transaction(trans, root);
4283                 trans = btrfs_start_transaction(root, 1);
4284                 if (IS_ERR(trans))
4285                         return PTR_ERR(trans);
4286
4287                 ret = btrfs_update_inode(trans, root, inode);
4288         }
4289         btrfs_end_transaction(trans, root);
4290         if (BTRFS_I(inode)->delayed_node)
4291                 btrfs_balance_delayed_items(root);
4292
4293         return ret;
4294 }
4295
4296 /*
4297  * This is a copy of file_update_time.  We need this so we can return error on
4298  * ENOSPC for updating the inode in the case of file write and mmap writes.
4299  */
4300 int btrfs_update_time(struct file *file)
4301 {
4302         struct inode *inode = file->f_path.dentry->d_inode;
4303         struct timespec now;
4304         int ret;
4305         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4306
4307         /* First try to exhaust all avenues to not sync */
4308         if (IS_NOCMTIME(inode))
4309                 return 0;
4310
4311         now = current_fs_time(inode->i_sb);
4312         if (!timespec_equal(&inode->i_mtime, &now))
4313                 sync_it = S_MTIME;
4314
4315         if (!timespec_equal(&inode->i_ctime, &now))
4316                 sync_it |= S_CTIME;
4317
4318         if (IS_I_VERSION(inode))
4319                 sync_it |= S_VERSION;
4320
4321         if (!sync_it)
4322                 return 0;
4323
4324         /* Finally allowed to write? Takes lock. */
4325         if (mnt_want_write_file(file))
4326                 return 0;
4327
4328         /* Only change inode inside the lock region */
4329         if (sync_it & S_VERSION)
4330                 inode_inc_iversion(inode);
4331         if (sync_it & S_CTIME)
4332                 inode->i_ctime = now;
4333         if (sync_it & S_MTIME)
4334                 inode->i_mtime = now;
4335         ret = btrfs_dirty_inode(inode);
4336         if (!ret)
4337                 mark_inode_dirty_sync(inode);
4338         mnt_drop_write(file->f_path.mnt);
4339         return ret;
4340 }
4341
4342 /*
4343  * find the highest existing sequence number in a directory
4344  * and then set the in-memory index_cnt variable to reflect
4345  * free sequence numbers
4346  */
4347 static int btrfs_set_inode_index_count(struct inode *inode)
4348 {
4349         struct btrfs_root *root = BTRFS_I(inode)->root;
4350         struct btrfs_key key, found_key;
4351         struct btrfs_path *path;
4352         struct extent_buffer *leaf;
4353         int ret;
4354
4355         key.objectid = btrfs_ino(inode);
4356         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4357         key.offset = (u64)-1;
4358
4359         path = btrfs_alloc_path();
4360         if (!path)
4361                 return -ENOMEM;
4362
4363         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4364         if (ret < 0)
4365                 goto out;
4366         /* FIXME: we should be able to handle this */
4367         if (ret == 0)
4368                 goto out;
4369         ret = 0;
4370
4371         /*
4372          * MAGIC NUMBER EXPLANATION:
4373          * since we search a directory based on f_pos we have to start at 2
4374          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4375          * else has to start at 2
4376          */
4377         if (path->slots[0] == 0) {
4378                 BTRFS_I(inode)->index_cnt = 2;
4379                 goto out;
4380         }
4381
4382         path->slots[0]--;
4383
4384         leaf = path->nodes[0];
4385         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4386
4387         if (found_key.objectid != btrfs_ino(inode) ||
4388             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4389                 BTRFS_I(inode)->index_cnt = 2;
4390                 goto out;
4391         }
4392
4393         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4394 out:
4395         btrfs_free_path(path);
4396         return ret;
4397 }
4398
4399 /*
4400  * helper to find a free sequence number in a given directory.  This current
4401  * code is very simple, later versions will do smarter things in the btree
4402  */
4403 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4404 {
4405         int ret = 0;
4406
4407         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4408                 ret = btrfs_inode_delayed_dir_index_count(dir);
4409                 if (ret) {
4410                         ret = btrfs_set_inode_index_count(dir);
4411                         if (ret)
4412                                 return ret;
4413                 }
4414         }
4415
4416         *index = BTRFS_I(dir)->index_cnt;
4417         BTRFS_I(dir)->index_cnt++;
4418
4419         return ret;
4420 }
4421
4422 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4423                                      struct btrfs_root *root,
4424                                      struct inode *dir,
4425                                      const char *name, int name_len,
4426                                      u64 ref_objectid, u64 objectid,
4427                                      umode_t mode, u64 *index)
4428 {
4429         struct inode *inode;
4430         struct btrfs_inode_item *inode_item;
4431         struct btrfs_key *location;
4432         struct btrfs_path *path;
4433         struct btrfs_inode_ref *ref;
4434         struct btrfs_key key[2];
4435         u32 sizes[2];
4436         unsigned long ptr;
4437         int ret;
4438         int owner;
4439
4440         path = btrfs_alloc_path();
4441         if (!path)
4442                 return ERR_PTR(-ENOMEM);
4443
4444         inode = new_inode(root->fs_info->sb);
4445         if (!inode) {
4446                 btrfs_free_path(path);
4447                 return ERR_PTR(-ENOMEM);
4448         }
4449
4450         /*
4451          * we have to initialize this early, so we can reclaim the inode
4452          * number if we fail afterwards in this function.
4453          */
4454         inode->i_ino = objectid;
4455
4456         if (dir) {
4457                 trace_btrfs_inode_request(dir);
4458
4459                 ret = btrfs_set_inode_index(dir, index);
4460                 if (ret) {
4461                         btrfs_free_path(path);
4462                         iput(inode);
4463                         return ERR_PTR(ret);
4464                 }
4465         }
4466         /*
4467          * index_cnt is ignored for everything but a dir,
4468          * btrfs_get_inode_index_count has an explanation for the magic
4469          * number
4470          */
4471         BTRFS_I(inode)->index_cnt = 2;
4472         BTRFS_I(inode)->root = root;
4473         BTRFS_I(inode)->generation = trans->transid;
4474         inode->i_generation = BTRFS_I(inode)->generation;
4475         btrfs_set_inode_space_info(root, inode);
4476
4477         if (S_ISDIR(mode))
4478                 owner = 0;
4479         else
4480                 owner = 1;
4481
4482         key[0].objectid = objectid;
4483         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4484         key[0].offset = 0;
4485
4486         key[1].objectid = objectid;
4487         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4488         key[1].offset = ref_objectid;
4489
4490         sizes[0] = sizeof(struct btrfs_inode_item);
4491         sizes[1] = name_len + sizeof(*ref);
4492
4493         path->leave_spinning = 1;
4494         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4495         if (ret != 0)
4496                 goto fail;
4497
4498         inode_init_owner(inode, dir, mode);
4499         inode_set_bytes(inode, 0);
4500         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4501         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4502                                   struct btrfs_inode_item);
4503         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4504
4505         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4506                              struct btrfs_inode_ref);
4507         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4508         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4509         ptr = (unsigned long)(ref + 1);
4510         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4511
4512         btrfs_mark_buffer_dirty(path->nodes[0]);
4513         btrfs_free_path(path);
4514
4515         location = &BTRFS_I(inode)->location;
4516         location->objectid = objectid;
4517         location->offset = 0;
4518         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4519
4520         btrfs_inherit_iflags(inode, dir);
4521
4522         if (S_ISREG(mode)) {
4523                 if (btrfs_test_opt(root, NODATASUM))
4524                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4525                 if (btrfs_test_opt(root, NODATACOW) ||
4526                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4527                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4528         }
4529
4530         insert_inode_hash(inode);
4531         inode_tree_add(inode);
4532
4533         trace_btrfs_inode_new(inode);
4534         btrfs_set_inode_last_trans(trans, inode);
4535
4536         return inode;
4537 fail:
4538         if (dir)
4539                 BTRFS_I(dir)->index_cnt--;
4540         btrfs_free_path(path);
4541         iput(inode);
4542         return ERR_PTR(ret);
4543 }
4544
4545 static inline u8 btrfs_inode_type(struct inode *inode)
4546 {
4547         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4548 }
4549
4550 /*
4551  * utility function to add 'inode' into 'parent_inode' with
4552  * a give name and a given sequence number.
4553  * if 'add_backref' is true, also insert a backref from the
4554  * inode to the parent directory.
4555  */
4556 int btrfs_add_link(struct btrfs_trans_handle *trans,
4557                    struct inode *parent_inode, struct inode *inode,
4558                    const char *name, int name_len, int add_backref, u64 index)
4559 {
4560         int ret = 0;
4561         struct btrfs_key key;
4562         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4563         u64 ino = btrfs_ino(inode);
4564         u64 parent_ino = btrfs_ino(parent_inode);
4565
4566         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4567                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4568         } else {
4569                 key.objectid = ino;
4570                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4571                 key.offset = 0;
4572         }
4573
4574         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4575                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4576                                          key.objectid, root->root_key.objectid,
4577                                          parent_ino, index, name, name_len);
4578         } else if (add_backref) {
4579                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4580                                              parent_ino, index);
4581         }
4582
4583         if (ret == 0) {
4584                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4585                                             parent_inode, &key,
4586                                             btrfs_inode_type(inode), index);
4587                 if (ret)
4588                         goto fail_dir_item;
4589
4590                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4591                                    name_len * 2);
4592                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4593                 ret = btrfs_update_inode(trans, root, parent_inode);
4594         }
4595         return ret;
4596
4597 fail_dir_item:
4598         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4599                 u64 local_index;
4600                 int err;
4601                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4602                                  key.objectid, root->root_key.objectid,
4603                                  parent_ino, &local_index, name, name_len);
4604
4605         } else if (add_backref) {
4606                 u64 local_index;
4607                 int err;
4608
4609                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4610                                           ino, parent_ino, &local_index);
4611         }
4612         return ret;
4613 }
4614
4615 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4616                             struct inode *dir, struct dentry *dentry,
4617                             struct inode *inode, int backref, u64 index)
4618 {
4619         int err = btrfs_add_link(trans, dir, inode,
4620                                  dentry->d_name.name, dentry->d_name.len,
4621                                  backref, index);
4622         if (err > 0)
4623                 err = -EEXIST;
4624         return err;
4625 }
4626
4627 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4628                         umode_t mode, dev_t rdev)
4629 {
4630         struct btrfs_trans_handle *trans;
4631         struct btrfs_root *root = BTRFS_I(dir)->root;
4632         struct inode *inode = NULL;
4633         int err;
4634         int drop_inode = 0;
4635         u64 objectid;
4636         unsigned long nr = 0;
4637         u64 index = 0;
4638
4639         if (!new_valid_dev(rdev))
4640                 return -EINVAL;
4641
4642         /*
4643          * 2 for inode item and ref
4644          * 2 for dir items
4645          * 1 for xattr if selinux is on
4646          */
4647         trans = btrfs_start_transaction(root, 5);
4648         if (IS_ERR(trans))
4649                 return PTR_ERR(trans);
4650
4651         err = btrfs_find_free_ino(root, &objectid);
4652         if (err)
4653                 goto out_unlock;
4654
4655         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4656                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4657                                 mode, &index);
4658         if (IS_ERR(inode)) {
4659                 err = PTR_ERR(inode);
4660                 goto out_unlock;
4661         }
4662
4663         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4664         if (err) {
4665                 drop_inode = 1;
4666                 goto out_unlock;
4667         }
4668
4669         /*
4670         * If the active LSM wants to access the inode during
4671         * d_instantiate it needs these. Smack checks to see
4672         * if the filesystem supports xattrs by looking at the
4673         * ops vector.
4674         */
4675
4676         inode->i_op = &btrfs_special_inode_operations;
4677         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4678         if (err)
4679                 drop_inode = 1;
4680         else {
4681                 init_special_inode(inode, inode->i_mode, rdev);
4682                 btrfs_update_inode(trans, root, inode);
4683                 d_instantiate(dentry, inode);
4684         }
4685 out_unlock:
4686         nr = trans->blocks_used;
4687         btrfs_end_transaction(trans, root);
4688         btrfs_btree_balance_dirty(root, nr);
4689         if (drop_inode) {
4690                 inode_dec_link_count(inode);
4691                 iput(inode);
4692         }
4693         return err;
4694 }
4695
4696 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4697                         umode_t mode, struct nameidata *nd)
4698 {
4699         struct btrfs_trans_handle *trans;
4700         struct btrfs_root *root = BTRFS_I(dir)->root;
4701         struct inode *inode = NULL;
4702         int drop_inode = 0;
4703         int err;
4704         unsigned long nr = 0;
4705         u64 objectid;
4706         u64 index = 0;
4707
4708         /*
4709          * 2 for inode item and ref
4710          * 2 for dir items
4711          * 1 for xattr if selinux is on
4712          */
4713         trans = btrfs_start_transaction(root, 5);
4714         if (IS_ERR(trans))
4715                 return PTR_ERR(trans);
4716
4717         err = btrfs_find_free_ino(root, &objectid);
4718         if (err)
4719                 goto out_unlock;
4720
4721         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4722                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4723                                 mode, &index);
4724         if (IS_ERR(inode)) {
4725                 err = PTR_ERR(inode);
4726                 goto out_unlock;
4727         }
4728
4729         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4730         if (err) {
4731                 drop_inode = 1;
4732                 goto out_unlock;
4733         }
4734
4735         /*
4736         * If the active LSM wants to access the inode during
4737         * d_instantiate it needs these. Smack checks to see
4738         * if the filesystem supports xattrs by looking at the
4739         * ops vector.
4740         */
4741         inode->i_fop = &btrfs_file_operations;
4742         inode->i_op = &btrfs_file_inode_operations;
4743
4744         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4745         if (err)
4746                 drop_inode = 1;
4747         else {
4748                 inode->i_mapping->a_ops = &btrfs_aops;
4749                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4750                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4751                 d_instantiate(dentry, inode);
4752         }
4753 out_unlock:
4754         nr = trans->blocks_used;
4755         btrfs_end_transaction(trans, root);
4756         if (drop_inode) {
4757                 inode_dec_link_count(inode);
4758                 iput(inode);
4759         }
4760         btrfs_btree_balance_dirty(root, nr);
4761         return err;
4762 }
4763
4764 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4765                       struct dentry *dentry)
4766 {
4767         struct btrfs_trans_handle *trans;
4768         struct btrfs_root *root = BTRFS_I(dir)->root;
4769         struct inode *inode = old_dentry->d_inode;
4770         u64 index;
4771         unsigned long nr = 0;
4772         int err;
4773         int drop_inode = 0;
4774
4775         /* do not allow sys_link's with other subvols of the same device */
4776         if (root->objectid != BTRFS_I(inode)->root->objectid)
4777                 return -EXDEV;
4778
4779         if (inode->i_nlink == ~0U)
4780                 return -EMLINK;
4781
4782         err = btrfs_set_inode_index(dir, &index);
4783         if (err)
4784                 goto fail;
4785
4786         /*
4787          * 2 items for inode and inode ref
4788          * 2 items for dir items
4789          * 1 item for parent inode
4790          */
4791         trans = btrfs_start_transaction(root, 5);
4792         if (IS_ERR(trans)) {
4793                 err = PTR_ERR(trans);
4794                 goto fail;
4795         }
4796
4797         btrfs_inc_nlink(inode);
4798         inode->i_ctime = CURRENT_TIME;
4799         ihold(inode);
4800
4801         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4802
4803         if (err) {
4804                 drop_inode = 1;
4805         } else {
4806                 struct dentry *parent = dentry->d_parent;
4807                 err = btrfs_update_inode(trans, root, inode);
4808                 BUG_ON(err);
4809                 d_instantiate(dentry, inode);
4810                 btrfs_log_new_name(trans, inode, NULL, parent);
4811         }
4812
4813         nr = trans->blocks_used;
4814         btrfs_end_transaction(trans, root);
4815 fail:
4816         if (drop_inode) {
4817                 inode_dec_link_count(inode);
4818                 iput(inode);
4819         }
4820         btrfs_btree_balance_dirty(root, nr);
4821         return err;
4822 }
4823
4824 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4825 {
4826         struct inode *inode = NULL;
4827         struct btrfs_trans_handle *trans;
4828         struct btrfs_root *root = BTRFS_I(dir)->root;
4829         int err = 0;
4830         int drop_on_err = 0;
4831         u64 objectid = 0;
4832         u64 index = 0;
4833         unsigned long nr = 1;
4834
4835         /*
4836          * 2 items for inode and ref
4837          * 2 items for dir items
4838          * 1 for xattr if selinux is on
4839          */
4840         trans = btrfs_start_transaction(root, 5);
4841         if (IS_ERR(trans))
4842                 return PTR_ERR(trans);
4843
4844         err = btrfs_find_free_ino(root, &objectid);
4845         if (err)
4846                 goto out_fail;
4847
4848         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4849                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4850                                 S_IFDIR | mode, &index);
4851         if (IS_ERR(inode)) {
4852                 err = PTR_ERR(inode);
4853                 goto out_fail;
4854         }
4855
4856         drop_on_err = 1;
4857
4858         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4859         if (err)
4860                 goto out_fail;
4861
4862         inode->i_op = &btrfs_dir_inode_operations;
4863         inode->i_fop = &btrfs_dir_file_operations;
4864
4865         btrfs_i_size_write(inode, 0);
4866         err = btrfs_update_inode(trans, root, inode);
4867         if (err)
4868                 goto out_fail;
4869
4870         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4871                              dentry->d_name.len, 0, index);
4872         if (err)
4873                 goto out_fail;
4874
4875         d_instantiate(dentry, inode);
4876         drop_on_err = 0;
4877
4878 out_fail:
4879         nr = trans->blocks_used;
4880         btrfs_end_transaction(trans, root);
4881         if (drop_on_err)
4882                 iput(inode);
4883         btrfs_btree_balance_dirty(root, nr);
4884         return err;
4885 }
4886
4887 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4888  * and an extent that you want to insert, deal with overlap and insert
4889  * the new extent into the tree.
4890  */
4891 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4892                                 struct extent_map *existing,
4893                                 struct extent_map *em,
4894                                 u64 map_start, u64 map_len)
4895 {
4896         u64 start_diff;
4897
4898         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4899         start_diff = map_start - em->start;
4900         em->start = map_start;
4901         em->len = map_len;
4902         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4903             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4904                 em->block_start += start_diff;
4905                 em->block_len -= start_diff;
4906         }
4907         return add_extent_mapping(em_tree, em);
4908 }
4909
4910 static noinline int uncompress_inline(struct btrfs_path *path,
4911                                       struct inode *inode, struct page *page,
4912                                       size_t pg_offset, u64 extent_offset,
4913                                       struct btrfs_file_extent_item *item)
4914 {
4915         int ret;
4916         struct extent_buffer *leaf = path->nodes[0];
4917         char *tmp;
4918         size_t max_size;
4919         unsigned long inline_size;
4920         unsigned long ptr;
4921         int compress_type;
4922
4923         WARN_ON(pg_offset != 0);
4924         compress_type = btrfs_file_extent_compression(leaf, item);
4925         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4926         inline_size = btrfs_file_extent_inline_item_len(leaf,
4927                                         btrfs_item_nr(leaf, path->slots[0]));
4928         tmp = kmalloc(inline_size, GFP_NOFS);
4929         if (!tmp)
4930                 return -ENOMEM;
4931         ptr = btrfs_file_extent_inline_start(item);
4932
4933         read_extent_buffer(leaf, tmp, ptr, inline_size);
4934
4935         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4936         ret = btrfs_decompress(compress_type, tmp, page,
4937                                extent_offset, inline_size, max_size);
4938         if (ret) {
4939                 char *kaddr = kmap_atomic(page, KM_USER0);
4940                 unsigned long copy_size = min_t(u64,
4941                                   PAGE_CACHE_SIZE - pg_offset,
4942                                   max_size - extent_offset);
4943                 memset(kaddr + pg_offset, 0, copy_size);
4944                 kunmap_atomic(kaddr, KM_USER0);
4945         }
4946         kfree(tmp);
4947         return 0;
4948 }
4949
4950 /*
4951  * a bit scary, this does extent mapping from logical file offset to the disk.
4952  * the ugly parts come from merging extents from the disk with the in-ram
4953  * representation.  This gets more complex because of the data=ordered code,
4954  * where the in-ram extents might be locked pending data=ordered completion.
4955  *
4956  * This also copies inline extents directly into the page.
4957  */
4958
4959 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4960                                     size_t pg_offset, u64 start, u64 len,
4961                                     int create)
4962 {
4963         int ret;
4964         int err = 0;
4965         u64 bytenr;
4966         u64 extent_start = 0;
4967         u64 extent_end = 0;
4968         u64 objectid = btrfs_ino(inode);
4969         u32 found_type;
4970         struct btrfs_path *path = NULL;
4971         struct btrfs_root *root = BTRFS_I(inode)->root;
4972         struct btrfs_file_extent_item *item;
4973         struct extent_buffer *leaf;
4974         struct btrfs_key found_key;
4975         struct extent_map *em = NULL;
4976         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4977         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4978         struct btrfs_trans_handle *trans = NULL;
4979         int compress_type;
4980
4981 again:
4982         read_lock(&em_tree->lock);
4983         em = lookup_extent_mapping(em_tree, start, len);
4984         if (em)
4985                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4986         read_unlock(&em_tree->lock);
4987
4988         if (em) {
4989                 if (em->start > start || em->start + em->len <= start)
4990                         free_extent_map(em);
4991                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4992                         free_extent_map(em);
4993                 else
4994                         goto out;
4995         }
4996         em = alloc_extent_map();
4997         if (!em) {
4998                 err = -ENOMEM;
4999                 goto out;
5000         }
5001         em->bdev = root->fs_info->fs_devices->latest_bdev;
5002         em->start = EXTENT_MAP_HOLE;
5003         em->orig_start = EXTENT_MAP_HOLE;
5004         em->len = (u64)-1;
5005         em->block_len = (u64)-1;
5006
5007         if (!path) {
5008                 path = btrfs_alloc_path();
5009                 if (!path) {
5010                         err = -ENOMEM;
5011                         goto out;
5012                 }
5013                 /*
5014                  * Chances are we'll be called again, so go ahead and do
5015                  * readahead
5016                  */
5017                 path->reada = 1;
5018         }
5019
5020         ret = btrfs_lookup_file_extent(trans, root, path,
5021                                        objectid, start, trans != NULL);
5022         if (ret < 0) {
5023                 err = ret;
5024                 goto out;
5025         }
5026
5027         if (ret != 0) {
5028                 if (path->slots[0] == 0)
5029                         goto not_found;
5030                 path->slots[0]--;
5031         }
5032
5033         leaf = path->nodes[0];
5034         item = btrfs_item_ptr(leaf, path->slots[0],
5035                               struct btrfs_file_extent_item);
5036         /* are we inside the extent that was found? */
5037         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5038         found_type = btrfs_key_type(&found_key);
5039         if (found_key.objectid != objectid ||
5040             found_type != BTRFS_EXTENT_DATA_KEY) {
5041                 goto not_found;
5042         }
5043
5044         found_type = btrfs_file_extent_type(leaf, item);
5045         extent_start = found_key.offset;
5046         compress_type = btrfs_file_extent_compression(leaf, item);
5047         if (found_type == BTRFS_FILE_EXTENT_REG ||
5048             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5049                 extent_end = extent_start +
5050                        btrfs_file_extent_num_bytes(leaf, item);
5051         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5052                 size_t size;
5053                 size = btrfs_file_extent_inline_len(leaf, item);
5054                 extent_end = (extent_start + size + root->sectorsize - 1) &
5055                         ~((u64)root->sectorsize - 1);
5056         }
5057
5058         if (start >= extent_end) {
5059                 path->slots[0]++;
5060                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5061                         ret = btrfs_next_leaf(root, path);
5062                         if (ret < 0) {
5063                                 err = ret;
5064                                 goto out;
5065                         }
5066                         if (ret > 0)
5067                                 goto not_found;
5068                         leaf = path->nodes[0];
5069                 }
5070                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5071                 if (found_key.objectid != objectid ||
5072                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5073                         goto not_found;
5074                 if (start + len <= found_key.offset)
5075                         goto not_found;
5076                 em->start = start;
5077                 em->len = found_key.offset - start;
5078                 goto not_found_em;
5079         }
5080
5081         if (found_type == BTRFS_FILE_EXTENT_REG ||
5082             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5083                 em->start = extent_start;
5084                 em->len = extent_end - extent_start;
5085                 em->orig_start = extent_start -
5086                                  btrfs_file_extent_offset(leaf, item);
5087                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5088                 if (bytenr == 0) {
5089                         em->block_start = EXTENT_MAP_HOLE;
5090                         goto insert;
5091                 }
5092                 if (compress_type != BTRFS_COMPRESS_NONE) {
5093                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5094                         em->compress_type = compress_type;
5095                         em->block_start = bytenr;
5096                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5097                                                                          item);
5098                 } else {
5099                         bytenr += btrfs_file_extent_offset(leaf, item);
5100                         em->block_start = bytenr;
5101                         em->block_len = em->len;
5102                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5103                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5104                 }
5105                 goto insert;
5106         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5107                 unsigned long ptr;
5108                 char *map;
5109                 size_t size;
5110                 size_t extent_offset;
5111                 size_t copy_size;
5112
5113                 em->block_start = EXTENT_MAP_INLINE;
5114                 if (!page || create) {
5115                         em->start = extent_start;
5116                         em->len = extent_end - extent_start;
5117                         goto out;
5118                 }
5119
5120                 size = btrfs_file_extent_inline_len(leaf, item);
5121                 extent_offset = page_offset(page) + pg_offset - extent_start;
5122                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5123                                 size - extent_offset);
5124                 em->start = extent_start + extent_offset;
5125                 em->len = (copy_size + root->sectorsize - 1) &
5126                         ~((u64)root->sectorsize - 1);
5127                 em->orig_start = EXTENT_MAP_INLINE;
5128                 if (compress_type) {
5129                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5130                         em->compress_type = compress_type;
5131                 }
5132                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5133                 if (create == 0 && !PageUptodate(page)) {
5134                         if (btrfs_file_extent_compression(leaf, item) !=
5135                             BTRFS_COMPRESS_NONE) {
5136                                 ret = uncompress_inline(path, inode, page,
5137                                                         pg_offset,
5138                                                         extent_offset, item);
5139                                 BUG_ON(ret);
5140                         } else {
5141                                 map = kmap(page);
5142                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5143                                                    copy_size);
5144                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5145                                         memset(map + pg_offset + copy_size, 0,
5146                                                PAGE_CACHE_SIZE - pg_offset -
5147                                                copy_size);
5148                                 }
5149                                 kunmap(page);
5150                         }
5151                         flush_dcache_page(page);
5152                 } else if (create && PageUptodate(page)) {
5153                         BUG();
5154                         if (!trans) {
5155                                 kunmap(page);
5156                                 free_extent_map(em);
5157                                 em = NULL;
5158
5159                                 btrfs_release_path(path);
5160                                 trans = btrfs_join_transaction(root);
5161
5162                                 if (IS_ERR(trans))
5163                                         return ERR_CAST(trans);
5164                                 goto again;
5165                         }
5166                         map = kmap(page);
5167                         write_extent_buffer(leaf, map + pg_offset, ptr,
5168                                             copy_size);
5169                         kunmap(page);
5170                         btrfs_mark_buffer_dirty(leaf);
5171                 }
5172                 set_extent_uptodate(io_tree, em->start,
5173                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5174                 goto insert;
5175         } else {
5176                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5177                 WARN_ON(1);
5178         }
5179 not_found:
5180         em->start = start;
5181         em->len = len;
5182 not_found_em:
5183         em->block_start = EXTENT_MAP_HOLE;
5184         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5185 insert:
5186         btrfs_release_path(path);
5187         if (em->start > start || extent_map_end(em) <= start) {
5188                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5189                        "[%llu %llu]\n", (unsigned long long)em->start,
5190                        (unsigned long long)em->len,
5191                        (unsigned long long)start,
5192                        (unsigned long long)len);
5193                 err = -EIO;
5194                 goto out;
5195         }
5196
5197         err = 0;
5198         write_lock(&em_tree->lock);
5199         ret = add_extent_mapping(em_tree, em);
5200         /* it is possible that someone inserted the extent into the tree
5201          * while we had the lock dropped.  It is also possible that
5202          * an overlapping map exists in the tree
5203          */
5204         if (ret == -EEXIST) {
5205                 struct extent_map *existing;
5206
5207                 ret = 0;
5208
5209                 existing = lookup_extent_mapping(em_tree, start, len);
5210                 if (existing && (existing->start > start ||
5211                     existing->start + existing->len <= start)) {
5212                         free_extent_map(existing);
5213                         existing = NULL;
5214                 }
5215                 if (!existing) {
5216                         existing = lookup_extent_mapping(em_tree, em->start,
5217                                                          em->len);
5218                         if (existing) {
5219                                 err = merge_extent_mapping(em_tree, existing,
5220                                                            em, start,
5221                                                            root->sectorsize);
5222                                 free_extent_map(existing);
5223                                 if (err) {
5224                                         free_extent_map(em);
5225                                         em = NULL;
5226                                 }
5227                         } else {
5228                                 err = -EIO;
5229                                 free_extent_map(em);
5230                                 em = NULL;
5231                         }
5232                 } else {
5233                         free_extent_map(em);
5234                         em = existing;
5235                         err = 0;
5236                 }
5237         }
5238         write_unlock(&em_tree->lock);
5239 out:
5240
5241         trace_btrfs_get_extent(root, em);
5242
5243         if (path)
5244                 btrfs_free_path(path);
5245         if (trans) {
5246                 ret = btrfs_end_transaction(trans, root);
5247                 if (!err)
5248                         err = ret;
5249         }
5250         if (err) {
5251                 free_extent_map(em);
5252                 return ERR_PTR(err);
5253         }
5254         return em;
5255 }
5256
5257 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5258                                            size_t pg_offset, u64 start, u64 len,
5259                                            int create)
5260 {
5261         struct extent_map *em;
5262         struct extent_map *hole_em = NULL;
5263         u64 range_start = start;
5264         u64 end;
5265         u64 found;
5266         u64 found_end;
5267         int err = 0;
5268
5269         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5270         if (IS_ERR(em))
5271                 return em;
5272         if (em) {
5273                 /*
5274                  * if our em maps to a hole, there might
5275                  * actually be delalloc bytes behind it
5276                  */
5277                 if (em->block_start != EXTENT_MAP_HOLE)
5278                         return em;
5279                 else
5280                         hole_em = em;
5281         }
5282
5283         /* check to see if we've wrapped (len == -1 or similar) */
5284         end = start + len;
5285         if (end < start)
5286                 end = (u64)-1;
5287         else
5288                 end -= 1;
5289
5290         em = NULL;
5291
5292         /* ok, we didn't find anything, lets look for delalloc */
5293         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5294                                  end, len, EXTENT_DELALLOC, 1);
5295         found_end = range_start + found;
5296         if (found_end < range_start)
5297                 found_end = (u64)-1;
5298
5299         /*
5300          * we didn't find anything useful, return
5301          * the original results from get_extent()
5302          */
5303         if (range_start > end || found_end <= start) {
5304                 em = hole_em;
5305                 hole_em = NULL;
5306                 goto out;
5307         }
5308
5309         /* adjust the range_start to make sure it doesn't
5310          * go backwards from the start they passed in
5311          */
5312         range_start = max(start,range_start);
5313         found = found_end - range_start;
5314
5315         if (found > 0) {
5316                 u64 hole_start = start;
5317                 u64 hole_len = len;
5318
5319                 em = alloc_extent_map();
5320                 if (!em) {
5321                         err = -ENOMEM;
5322                         goto out;
5323                 }
5324                 /*
5325                  * when btrfs_get_extent can't find anything it
5326                  * returns one huge hole
5327                  *
5328                  * make sure what it found really fits our range, and
5329                  * adjust to make sure it is based on the start from
5330                  * the caller
5331                  */
5332                 if (hole_em) {
5333                         u64 calc_end = extent_map_end(hole_em);
5334
5335                         if (calc_end <= start || (hole_em->start > end)) {
5336                                 free_extent_map(hole_em);
5337                                 hole_em = NULL;
5338                         } else {
5339                                 hole_start = max(hole_em->start, start);
5340                                 hole_len = calc_end - hole_start;
5341                         }
5342                 }
5343                 em->bdev = NULL;
5344                 if (hole_em && range_start > hole_start) {
5345                         /* our hole starts before our delalloc, so we
5346                          * have to return just the parts of the hole
5347                          * that go until  the delalloc starts
5348                          */
5349                         em->len = min(hole_len,
5350                                       range_start - hole_start);
5351                         em->start = hole_start;
5352                         em->orig_start = hole_start;
5353                         /*
5354                          * don't adjust block start at all,
5355                          * it is fixed at EXTENT_MAP_HOLE
5356                          */
5357                         em->block_start = hole_em->block_start;
5358                         em->block_len = hole_len;
5359                 } else {
5360                         em->start = range_start;
5361                         em->len = found;
5362                         em->orig_start = range_start;
5363                         em->block_start = EXTENT_MAP_DELALLOC;
5364                         em->block_len = found;
5365                 }
5366         } else if (hole_em) {
5367                 return hole_em;
5368         }
5369 out:
5370
5371         free_extent_map(hole_em);
5372         if (err) {
5373                 free_extent_map(em);
5374                 return ERR_PTR(err);
5375         }
5376         return em;
5377 }
5378
5379 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5380                                                   struct extent_map *em,
5381                                                   u64 start, u64 len)
5382 {
5383         struct btrfs_root *root = BTRFS_I(inode)->root;
5384         struct btrfs_trans_handle *trans;
5385         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5386         struct btrfs_key ins;
5387         u64 alloc_hint;
5388         int ret;
5389         bool insert = false;
5390
5391         /*
5392          * Ok if the extent map we looked up is a hole and is for the exact
5393          * range we want, there is no reason to allocate a new one, however if
5394          * it is not right then we need to free this one and drop the cache for
5395          * our range.
5396          */
5397         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5398             em->len != len) {
5399                 free_extent_map(em);
5400                 em = NULL;
5401                 insert = true;
5402                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5403         }
5404
5405         trans = btrfs_join_transaction(root);
5406         if (IS_ERR(trans))
5407                 return ERR_CAST(trans);
5408
5409         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5410                 btrfs_add_inode_defrag(trans, inode);
5411
5412         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5413
5414         alloc_hint = get_extent_allocation_hint(inode, start, len);
5415         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5416                                    alloc_hint, &ins, 1);
5417         if (ret) {
5418                 em = ERR_PTR(ret);
5419                 goto out;
5420         }
5421
5422         if (!em) {
5423                 em = alloc_extent_map();
5424                 if (!em) {
5425                         em = ERR_PTR(-ENOMEM);
5426                         goto out;
5427                 }
5428         }
5429
5430         em->start = start;
5431         em->orig_start = em->start;
5432         em->len = ins.offset;
5433
5434         em->block_start = ins.objectid;
5435         em->block_len = ins.offset;
5436         em->bdev = root->fs_info->fs_devices->latest_bdev;
5437
5438         /*
5439          * We need to do this because if we're using the original em we searched
5440          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5441          */
5442         em->flags = 0;
5443         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5444
5445         while (insert) {
5446                 write_lock(&em_tree->lock);
5447                 ret = add_extent_mapping(em_tree, em);
5448                 write_unlock(&em_tree->lock);
5449                 if (ret != -EEXIST)
5450                         break;
5451                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5452         }
5453
5454         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5455                                            ins.offset, ins.offset, 0);
5456         if (ret) {
5457                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5458                 em = ERR_PTR(ret);
5459         }
5460 out:
5461         btrfs_end_transaction(trans, root);
5462         return em;
5463 }
5464
5465 /*
5466  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5467  * block must be cow'd
5468  */
5469 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5470                                       struct inode *inode, u64 offset, u64 len)
5471 {
5472         struct btrfs_path *path;
5473         int ret;
5474         struct extent_buffer *leaf;
5475         struct btrfs_root *root = BTRFS_I(inode)->root;
5476         struct btrfs_file_extent_item *fi;
5477         struct btrfs_key key;
5478         u64 disk_bytenr;
5479         u64 backref_offset;
5480         u64 extent_end;
5481         u64 num_bytes;
5482         int slot;
5483         int found_type;
5484
5485         path = btrfs_alloc_path();
5486         if (!path)
5487                 return -ENOMEM;
5488
5489         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5490                                        offset, 0);
5491         if (ret < 0)
5492                 goto out;
5493
5494         slot = path->slots[0];
5495         if (ret == 1) {
5496                 if (slot == 0) {
5497                         /* can't find the item, must cow */
5498                         ret = 0;
5499                         goto out;
5500                 }
5501                 slot--;
5502         }
5503         ret = 0;
5504         leaf = path->nodes[0];
5505         btrfs_item_key_to_cpu(leaf, &key, slot);
5506         if (key.objectid != btrfs_ino(inode) ||
5507             key.type != BTRFS_EXTENT_DATA_KEY) {
5508                 /* not our file or wrong item type, must cow */
5509                 goto out;
5510         }
5511
5512         if (key.offset > offset) {
5513                 /* Wrong offset, must cow */
5514                 goto out;
5515         }
5516
5517         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5518         found_type = btrfs_file_extent_type(leaf, fi);
5519         if (found_type != BTRFS_FILE_EXTENT_REG &&
5520             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5521                 /* not a regular extent, must cow */
5522                 goto out;
5523         }
5524         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5525         backref_offset = btrfs_file_extent_offset(leaf, fi);
5526
5527         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5528         if (extent_end < offset + len) {
5529                 /* extent doesn't include our full range, must cow */
5530                 goto out;
5531         }
5532
5533         if (btrfs_extent_readonly(root, disk_bytenr))
5534                 goto out;
5535
5536         /*
5537          * look for other files referencing this extent, if we
5538          * find any we must cow
5539          */
5540         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5541                                   key.offset - backref_offset, disk_bytenr))
5542                 goto out;
5543
5544         /*
5545          * adjust disk_bytenr and num_bytes to cover just the bytes
5546          * in this extent we are about to write.  If there
5547          * are any csums in that range we have to cow in order
5548          * to keep the csums correct
5549          */
5550         disk_bytenr += backref_offset;
5551         disk_bytenr += offset - key.offset;
5552         num_bytes = min(offset + len, extent_end) - offset;
5553         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5554                                 goto out;
5555         /*
5556          * all of the above have passed, it is safe to overwrite this extent
5557          * without cow
5558          */
5559         ret = 1;
5560 out:
5561         btrfs_free_path(path);
5562         return ret;
5563 }
5564
5565 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5566                                    struct buffer_head *bh_result, int create)
5567 {
5568         struct extent_map *em;
5569         struct btrfs_root *root = BTRFS_I(inode)->root;
5570         u64 start = iblock << inode->i_blkbits;
5571         u64 len = bh_result->b_size;
5572         struct btrfs_trans_handle *trans;
5573
5574         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5575         if (IS_ERR(em))
5576                 return PTR_ERR(em);
5577
5578         /*
5579          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5580          * io.  INLINE is special, and we could probably kludge it in here, but
5581          * it's still buffered so for safety lets just fall back to the generic
5582          * buffered path.
5583          *
5584          * For COMPRESSED we _have_ to read the entire extent in so we can
5585          * decompress it, so there will be buffering required no matter what we
5586          * do, so go ahead and fallback to buffered.
5587          *
5588          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5589          * to buffered IO.  Don't blame me, this is the price we pay for using
5590          * the generic code.
5591          */
5592         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5593             em->block_start == EXTENT_MAP_INLINE) {
5594                 free_extent_map(em);
5595                 return -ENOTBLK;
5596         }
5597
5598         /* Just a good old fashioned hole, return */
5599         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5600                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5601                 free_extent_map(em);
5602                 /* DIO will do one hole at a time, so just unlock a sector */
5603                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5604                               start + root->sectorsize - 1, GFP_NOFS);
5605                 return 0;
5606         }
5607
5608         /*
5609          * We don't allocate a new extent in the following cases
5610          *
5611          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5612          * existing extent.
5613          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5614          * just use the extent.
5615          *
5616          */
5617         if (!create) {
5618                 len = em->len - (start - em->start);
5619                 goto map;
5620         }
5621
5622         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5623             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5624              em->block_start != EXTENT_MAP_HOLE)) {
5625                 int type;
5626                 int ret;
5627                 u64 block_start;
5628
5629                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5630                         type = BTRFS_ORDERED_PREALLOC;
5631                 else
5632                         type = BTRFS_ORDERED_NOCOW;
5633                 len = min(len, em->len - (start - em->start));
5634                 block_start = em->block_start + (start - em->start);
5635
5636                 /*
5637                  * we're not going to log anything, but we do need
5638                  * to make sure the current transaction stays open
5639                  * while we look for nocow cross refs
5640                  */
5641                 trans = btrfs_join_transaction(root);
5642                 if (IS_ERR(trans))
5643                         goto must_cow;
5644
5645                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5646                         ret = btrfs_add_ordered_extent_dio(inode, start,
5647                                            block_start, len, len, type);
5648                         btrfs_end_transaction(trans, root);
5649                         if (ret) {
5650                                 free_extent_map(em);
5651                                 return ret;
5652                         }
5653                         goto unlock;
5654                 }
5655                 btrfs_end_transaction(trans, root);
5656         }
5657 must_cow:
5658         /*
5659          * this will cow the extent, reset the len in case we changed
5660          * it above
5661          */
5662         len = bh_result->b_size;
5663         em = btrfs_new_extent_direct(inode, em, start, len);
5664         if (IS_ERR(em))
5665                 return PTR_ERR(em);
5666         len = min(len, em->len - (start - em->start));
5667 unlock:
5668         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5669                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5670                           0, NULL, GFP_NOFS);
5671 map:
5672         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5673                 inode->i_blkbits;
5674         bh_result->b_size = len;
5675         bh_result->b_bdev = em->bdev;
5676         set_buffer_mapped(bh_result);
5677         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5678                 set_buffer_new(bh_result);
5679
5680         free_extent_map(em);
5681
5682         return 0;
5683 }
5684
5685 struct btrfs_dio_private {
5686         struct inode *inode;
5687         u64 logical_offset;
5688         u64 disk_bytenr;
5689         u64 bytes;
5690         u32 *csums;
5691         void *private;
5692
5693         /* number of bios pending for this dio */
5694         atomic_t pending_bios;
5695
5696         /* IO errors */
5697         int errors;
5698
5699         struct bio *orig_bio;
5700 };
5701
5702 static void btrfs_endio_direct_read(struct bio *bio, int err)
5703 {
5704         struct btrfs_dio_private *dip = bio->bi_private;
5705         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5706         struct bio_vec *bvec = bio->bi_io_vec;
5707         struct inode *inode = dip->inode;
5708         struct btrfs_root *root = BTRFS_I(inode)->root;
5709         u64 start;
5710         u32 *private = dip->csums;
5711
5712         start = dip->logical_offset;
5713         do {
5714                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5715                         struct page *page = bvec->bv_page;
5716                         char *kaddr;
5717                         u32 csum = ~(u32)0;
5718                         unsigned long flags;
5719
5720                         local_irq_save(flags);
5721                         kaddr = kmap_atomic(page, KM_IRQ0);
5722                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5723                                                csum, bvec->bv_len);
5724                         btrfs_csum_final(csum, (char *)&csum);
5725                         kunmap_atomic(kaddr, KM_IRQ0);
5726                         local_irq_restore(flags);
5727
5728                         flush_dcache_page(bvec->bv_page);
5729                         if (csum != *private) {
5730                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5731                                       " %llu csum %u private %u\n",
5732                                       (unsigned long long)btrfs_ino(inode),
5733                                       (unsigned long long)start,
5734                                       csum, *private);
5735                                 err = -EIO;
5736                         }
5737                 }
5738
5739                 start += bvec->bv_len;
5740                 private++;
5741                 bvec++;
5742         } while (bvec <= bvec_end);
5743
5744         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5745                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5746         bio->bi_private = dip->private;
5747
5748         kfree(dip->csums);
5749         kfree(dip);
5750
5751         /* If we had a csum failure make sure to clear the uptodate flag */
5752         if (err)
5753                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5754         dio_end_io(bio, err);
5755 }
5756
5757 static void btrfs_endio_direct_write(struct bio *bio, int err)
5758 {
5759         struct btrfs_dio_private *dip = bio->bi_private;
5760         struct inode *inode = dip->inode;
5761         struct btrfs_root *root = BTRFS_I(inode)->root;
5762         struct btrfs_trans_handle *trans;
5763         struct btrfs_ordered_extent *ordered = NULL;
5764         struct extent_state *cached_state = NULL;
5765         u64 ordered_offset = dip->logical_offset;
5766         u64 ordered_bytes = dip->bytes;
5767         int ret;
5768
5769         if (err)
5770                 goto out_done;
5771 again:
5772         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5773                                                    &ordered_offset,
5774                                                    ordered_bytes);
5775         if (!ret)
5776                 goto out_test;
5777
5778         BUG_ON(!ordered);
5779
5780         trans = btrfs_join_transaction(root);
5781         if (IS_ERR(trans)) {
5782                 err = -ENOMEM;
5783                 goto out;
5784         }
5785         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5786
5787         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5788                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5789                 if (!ret)
5790                         err = btrfs_update_inode_fallback(trans, root, inode);
5791                 goto out;
5792         }
5793
5794         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5795                          ordered->file_offset + ordered->len - 1, 0,
5796                          &cached_state, GFP_NOFS);
5797
5798         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5799                 ret = btrfs_mark_extent_written(trans, inode,
5800                                                 ordered->file_offset,
5801                                                 ordered->file_offset +
5802                                                 ordered->len);
5803                 if (ret) {
5804                         err = ret;
5805                         goto out_unlock;
5806                 }
5807         } else {
5808                 ret = insert_reserved_file_extent(trans, inode,
5809                                                   ordered->file_offset,
5810                                                   ordered->start,
5811                                                   ordered->disk_len,
5812                                                   ordered->len,
5813                                                   ordered->len,
5814                                                   0, 0, 0,
5815                                                   BTRFS_FILE_EXTENT_REG);
5816                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5817                                    ordered->file_offset, ordered->len);
5818                 if (ret) {
5819                         err = ret;
5820                         WARN_ON(1);
5821                         goto out_unlock;
5822                 }
5823         }
5824
5825         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5826         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5827         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5828                 btrfs_update_inode_fallback(trans, root, inode);
5829         ret = 0;
5830 out_unlock:
5831         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5832                              ordered->file_offset + ordered->len - 1,
5833                              &cached_state, GFP_NOFS);
5834 out:
5835         btrfs_delalloc_release_metadata(inode, ordered->len);
5836         btrfs_end_transaction(trans, root);
5837         ordered_offset = ordered->file_offset + ordered->len;
5838         btrfs_put_ordered_extent(ordered);
5839         btrfs_put_ordered_extent(ordered);
5840
5841 out_test:
5842         /*
5843          * our bio might span multiple ordered extents.  If we haven't
5844          * completed the accounting for the whole dio, go back and try again
5845          */
5846         if (ordered_offset < dip->logical_offset + dip->bytes) {
5847                 ordered_bytes = dip->logical_offset + dip->bytes -
5848                         ordered_offset;
5849                 goto again;
5850         }
5851 out_done:
5852         bio->bi_private = dip->private;
5853
5854         kfree(dip->csums);
5855         kfree(dip);
5856
5857         /* If we had an error make sure to clear the uptodate flag */
5858         if (err)
5859                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5860         dio_end_io(bio, err);
5861 }
5862
5863 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5864                                     struct bio *bio, int mirror_num,
5865                                     unsigned long bio_flags, u64 offset)
5866 {
5867         int ret;
5868         struct btrfs_root *root = BTRFS_I(inode)->root;
5869         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5870         BUG_ON(ret);
5871         return 0;
5872 }
5873
5874 static void btrfs_end_dio_bio(struct bio *bio, int err)
5875 {
5876         struct btrfs_dio_private *dip = bio->bi_private;
5877
5878         if (err) {
5879                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5880                       "sector %#Lx len %u err no %d\n",
5881                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5882                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5883                 dip->errors = 1;
5884
5885                 /*
5886                  * before atomic variable goto zero, we must make sure
5887                  * dip->errors is perceived to be set.
5888                  */
5889                 smp_mb__before_atomic_dec();
5890         }
5891
5892         /* if there are more bios still pending for this dio, just exit */
5893         if (!atomic_dec_and_test(&dip->pending_bios))
5894                 goto out;
5895
5896         if (dip->errors)
5897                 bio_io_error(dip->orig_bio);
5898         else {
5899                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5900                 bio_endio(dip->orig_bio, 0);
5901         }
5902 out:
5903         bio_put(bio);
5904 }
5905
5906 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5907                                        u64 first_sector, gfp_t gfp_flags)
5908 {
5909         int nr_vecs = bio_get_nr_vecs(bdev);
5910         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5911 }
5912
5913 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5914                                          int rw, u64 file_offset, int skip_sum,
5915                                          u32 *csums, int async_submit)
5916 {
5917         int write = rw & REQ_WRITE;
5918         struct btrfs_root *root = BTRFS_I(inode)->root;
5919         int ret;
5920
5921         bio_get(bio);
5922         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5923         if (ret)
5924                 goto err;
5925
5926         if (skip_sum)
5927                 goto map;
5928
5929         if (write && async_submit) {
5930                 ret = btrfs_wq_submit_bio(root->fs_info,
5931                                    inode, rw, bio, 0, 0,
5932                                    file_offset,
5933                                    __btrfs_submit_bio_start_direct_io,
5934                                    __btrfs_submit_bio_done);
5935                 goto err;
5936         } else if (write) {
5937                 /*
5938                  * If we aren't doing async submit, calculate the csum of the
5939                  * bio now.
5940                  */
5941                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5942                 if (ret)
5943                         goto err;
5944         } else if (!skip_sum) {
5945                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5946                                           file_offset, csums);
5947                 if (ret)
5948                         goto err;
5949         }
5950
5951 map:
5952         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5953 err:
5954         bio_put(bio);
5955         return ret;
5956 }
5957
5958 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5959                                     int skip_sum)
5960 {
5961         struct inode *inode = dip->inode;
5962         struct btrfs_root *root = BTRFS_I(inode)->root;
5963         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5964         struct bio *bio;
5965         struct bio *orig_bio = dip->orig_bio;
5966         struct bio_vec *bvec = orig_bio->bi_io_vec;
5967         u64 start_sector = orig_bio->bi_sector;
5968         u64 file_offset = dip->logical_offset;
5969         u64 submit_len = 0;
5970         u64 map_length;
5971         int nr_pages = 0;
5972         u32 *csums = dip->csums;
5973         int ret = 0;
5974         int async_submit = 0;
5975         int write = rw & REQ_WRITE;
5976
5977         map_length = orig_bio->bi_size;
5978         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5979                               &map_length, NULL, 0);
5980         if (ret) {
5981                 bio_put(orig_bio);
5982                 return -EIO;
5983         }
5984
5985         if (map_length >= orig_bio->bi_size) {
5986                 bio = orig_bio;
5987                 goto submit;
5988         }
5989
5990         async_submit = 1;
5991         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5992         if (!bio)
5993                 return -ENOMEM;
5994         bio->bi_private = dip;
5995         bio->bi_end_io = btrfs_end_dio_bio;
5996         atomic_inc(&dip->pending_bios);
5997
5998         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5999                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6000                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6001                                  bvec->bv_offset) < bvec->bv_len)) {
6002                         /*
6003                          * inc the count before we submit the bio so
6004                          * we know the end IO handler won't happen before
6005                          * we inc the count. Otherwise, the dip might get freed
6006                          * before we're done setting it up
6007                          */
6008                         atomic_inc(&dip->pending_bios);
6009                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6010                                                      file_offset, skip_sum,
6011                                                      csums, async_submit);
6012                         if (ret) {
6013                                 bio_put(bio);
6014                                 atomic_dec(&dip->pending_bios);
6015                                 goto out_err;
6016                         }
6017
6018                         /* Write's use the ordered csums */
6019                         if (!write && !skip_sum)
6020                                 csums = csums + nr_pages;
6021                         start_sector += submit_len >> 9;
6022                         file_offset += submit_len;
6023
6024                         submit_len = 0;
6025                         nr_pages = 0;
6026
6027                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6028                                                   start_sector, GFP_NOFS);
6029                         if (!bio)
6030                                 goto out_err;
6031                         bio->bi_private = dip;
6032                         bio->bi_end_io = btrfs_end_dio_bio;
6033
6034                         map_length = orig_bio->bi_size;
6035                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6036                                               &map_length, NULL, 0);
6037                         if (ret) {
6038                                 bio_put(bio);
6039                                 goto out_err;
6040                         }
6041                 } else {
6042                         submit_len += bvec->bv_len;
6043                         nr_pages ++;
6044                         bvec++;
6045                 }
6046         }
6047
6048 submit:
6049         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6050                                      csums, async_submit);
6051         if (!ret)
6052                 return 0;
6053
6054         bio_put(bio);
6055 out_err:
6056         dip->errors = 1;
6057         /*
6058          * before atomic variable goto zero, we must
6059          * make sure dip->errors is perceived to be set.
6060          */
6061         smp_mb__before_atomic_dec();
6062         if (atomic_dec_and_test(&dip->pending_bios))
6063                 bio_io_error(dip->orig_bio);
6064
6065         /* bio_end_io() will handle error, so we needn't return it */
6066         return 0;
6067 }
6068
6069 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6070                                 loff_t file_offset)
6071 {
6072         struct btrfs_root *root = BTRFS_I(inode)->root;
6073         struct btrfs_dio_private *dip;
6074         struct bio_vec *bvec = bio->bi_io_vec;
6075         int skip_sum;
6076         int write = rw & REQ_WRITE;
6077         int ret = 0;
6078
6079         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6080
6081         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6082         if (!dip) {
6083                 ret = -ENOMEM;
6084                 goto free_ordered;
6085         }
6086         dip->csums = NULL;
6087
6088         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6089         if (!write && !skip_sum) {
6090                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6091                 if (!dip->csums) {
6092                         kfree(dip);
6093                         ret = -ENOMEM;
6094                         goto free_ordered;
6095                 }
6096         }
6097
6098         dip->private = bio->bi_private;
6099         dip->inode = inode;
6100         dip->logical_offset = file_offset;
6101
6102         dip->bytes = 0;
6103         do {
6104                 dip->bytes += bvec->bv_len;
6105                 bvec++;
6106         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6107
6108         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6109         bio->bi_private = dip;
6110         dip->errors = 0;
6111         dip->orig_bio = bio;
6112         atomic_set(&dip->pending_bios, 0);
6113
6114         if (write)
6115                 bio->bi_end_io = btrfs_endio_direct_write;
6116         else
6117                 bio->bi_end_io = btrfs_endio_direct_read;
6118
6119         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6120         if (!ret)
6121                 return;
6122 free_ordered:
6123         /*
6124          * If this is a write, we need to clean up the reserved space and kill
6125          * the ordered extent.
6126          */
6127         if (write) {
6128                 struct btrfs_ordered_extent *ordered;
6129                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6130                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6131                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6132                         btrfs_free_reserved_extent(root, ordered->start,
6133                                                    ordered->disk_len);
6134                 btrfs_put_ordered_extent(ordered);
6135                 btrfs_put_ordered_extent(ordered);
6136         }
6137         bio_endio(bio, ret);
6138 }
6139
6140 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6141                         const struct iovec *iov, loff_t offset,
6142                         unsigned long nr_segs)
6143 {
6144         int seg;
6145         int i;
6146         size_t size;
6147         unsigned long addr;
6148         unsigned blocksize_mask = root->sectorsize - 1;
6149         ssize_t retval = -EINVAL;
6150         loff_t end = offset;
6151
6152         if (offset & blocksize_mask)
6153                 goto out;
6154
6155         /* Check the memory alignment.  Blocks cannot straddle pages */
6156         for (seg = 0; seg < nr_segs; seg++) {
6157                 addr = (unsigned long)iov[seg].iov_base;
6158                 size = iov[seg].iov_len;
6159                 end += size;
6160                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6161                         goto out;
6162
6163                 /* If this is a write we don't need to check anymore */
6164                 if (rw & WRITE)
6165                         continue;
6166
6167                 /*
6168                  * Check to make sure we don't have duplicate iov_base's in this
6169                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6170                  * when reading back.
6171                  */
6172                 for (i = seg + 1; i < nr_segs; i++) {
6173                         if (iov[seg].iov_base == iov[i].iov_base)
6174                                 goto out;
6175                 }
6176         }
6177         retval = 0;
6178 out:
6179         return retval;
6180 }
6181 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6182                         const struct iovec *iov, loff_t offset,
6183                         unsigned long nr_segs)
6184 {
6185         struct file *file = iocb->ki_filp;
6186         struct inode *inode = file->f_mapping->host;
6187         struct btrfs_ordered_extent *ordered;
6188         struct extent_state *cached_state = NULL;
6189         u64 lockstart, lockend;
6190         ssize_t ret;
6191         int writing = rw & WRITE;
6192         int write_bits = 0;
6193         size_t count = iov_length(iov, nr_segs);
6194
6195         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6196                             offset, nr_segs)) {
6197                 return 0;
6198         }
6199
6200         lockstart = offset;
6201         lockend = offset + count - 1;
6202
6203         if (writing) {
6204                 ret = btrfs_delalloc_reserve_space(inode, count);
6205                 if (ret)
6206                         goto out;
6207         }
6208
6209         while (1) {
6210                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6211                                  0, &cached_state, GFP_NOFS);
6212                 /*
6213                  * We're concerned with the entire range that we're going to be
6214                  * doing DIO to, so we need to make sure theres no ordered
6215                  * extents in this range.
6216                  */
6217                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6218                                                      lockend - lockstart + 1);
6219                 if (!ordered)
6220                         break;
6221                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6222                                      &cached_state, GFP_NOFS);
6223                 btrfs_start_ordered_extent(inode, ordered, 1);
6224                 btrfs_put_ordered_extent(ordered);
6225                 cond_resched();
6226         }
6227
6228         /*
6229          * we don't use btrfs_set_extent_delalloc because we don't want
6230          * the dirty or uptodate bits
6231          */
6232         if (writing) {
6233                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6234                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6235                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6236                                      GFP_NOFS);
6237                 if (ret) {
6238                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6239                                          lockend, EXTENT_LOCKED | write_bits,
6240                                          1, 0, &cached_state, GFP_NOFS);
6241                         goto out;
6242                 }
6243         }
6244
6245         free_extent_state(cached_state);
6246         cached_state = NULL;
6247
6248         ret = __blockdev_direct_IO(rw, iocb, inode,
6249                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6250                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6251                    btrfs_submit_direct, 0);
6252
6253         if (ret < 0 && ret != -EIOCBQUEUED) {
6254                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6255                               offset + iov_length(iov, nr_segs) - 1,
6256                               EXTENT_LOCKED | write_bits, 1, 0,
6257                               &cached_state, GFP_NOFS);
6258         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6259                 /*
6260                  * We're falling back to buffered, unlock the section we didn't
6261                  * do IO on.
6262                  */
6263                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6264                               offset + iov_length(iov, nr_segs) - 1,
6265                               EXTENT_LOCKED | write_bits, 1, 0,
6266                               &cached_state, GFP_NOFS);
6267         }
6268 out:
6269         free_extent_state(cached_state);
6270         return ret;
6271 }
6272
6273 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6274                 __u64 start, __u64 len)
6275 {
6276         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6277 }
6278
6279 int btrfs_readpage(struct file *file, struct page *page)
6280 {
6281         struct extent_io_tree *tree;
6282         tree = &BTRFS_I(page->mapping->host)->io_tree;
6283         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6284 }
6285
6286 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6287 {
6288         struct extent_io_tree *tree;
6289
6290
6291         if (current->flags & PF_MEMALLOC) {
6292                 redirty_page_for_writepage(wbc, page);
6293                 unlock_page(page);
6294                 return 0;
6295         }
6296         tree = &BTRFS_I(page->mapping->host)->io_tree;
6297         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6298 }
6299
6300 int btrfs_writepages(struct address_space *mapping,
6301                      struct writeback_control *wbc)
6302 {
6303         struct extent_io_tree *tree;
6304
6305         tree = &BTRFS_I(mapping->host)->io_tree;
6306         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6307 }
6308
6309 static int
6310 btrfs_readpages(struct file *file, struct address_space *mapping,
6311                 struct list_head *pages, unsigned nr_pages)
6312 {
6313         struct extent_io_tree *tree;
6314         tree = &BTRFS_I(mapping->host)->io_tree;
6315         return extent_readpages(tree, mapping, pages, nr_pages,
6316                                 btrfs_get_extent);
6317 }
6318 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6319 {
6320         struct extent_io_tree *tree;
6321         struct extent_map_tree *map;
6322         int ret;
6323
6324         tree = &BTRFS_I(page->mapping->host)->io_tree;
6325         map = &BTRFS_I(page->mapping->host)->extent_tree;
6326         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6327         if (ret == 1) {
6328                 ClearPagePrivate(page);
6329                 set_page_private(page, 0);
6330                 page_cache_release(page);
6331         }
6332         return ret;
6333 }
6334
6335 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6336 {
6337         if (PageWriteback(page) || PageDirty(page))
6338                 return 0;
6339         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6340 }
6341
6342 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6343 {
6344         struct extent_io_tree *tree;
6345         struct btrfs_ordered_extent *ordered;
6346         struct extent_state *cached_state = NULL;
6347         u64 page_start = page_offset(page);
6348         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6349
6350
6351         /*
6352          * we have the page locked, so new writeback can't start,
6353          * and the dirty bit won't be cleared while we are here.
6354          *
6355          * Wait for IO on this page so that we can safely clear
6356          * the PagePrivate2 bit and do ordered accounting
6357          */
6358         wait_on_page_writeback(page);
6359
6360         tree = &BTRFS_I(page->mapping->host)->io_tree;
6361         if (offset) {
6362                 btrfs_releasepage(page, GFP_NOFS);
6363                 return;
6364         }
6365         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6366                          GFP_NOFS);
6367         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6368                                            page_offset(page));
6369         if (ordered) {
6370                 /*
6371                  * IO on this page will never be started, so we need
6372                  * to account for any ordered extents now
6373                  */
6374                 clear_extent_bit(tree, page_start, page_end,
6375                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6376                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6377                                  &cached_state, GFP_NOFS);
6378                 /*
6379                  * whoever cleared the private bit is responsible
6380                  * for the finish_ordered_io
6381                  */
6382                 if (TestClearPagePrivate2(page)) {
6383                         btrfs_finish_ordered_io(page->mapping->host,
6384                                                 page_start, page_end);
6385                 }
6386                 btrfs_put_ordered_extent(ordered);
6387                 cached_state = NULL;
6388                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6389                                  GFP_NOFS);
6390         }
6391         clear_extent_bit(tree, page_start, page_end,
6392                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6393                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6394         __btrfs_releasepage(page, GFP_NOFS);
6395
6396         ClearPageChecked(page);
6397         if (PagePrivate(page)) {
6398                 ClearPagePrivate(page);
6399                 set_page_private(page, 0);
6400                 page_cache_release(page);
6401         }
6402 }
6403
6404 /*
6405  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6406  * called from a page fault handler when a page is first dirtied. Hence we must
6407  * be careful to check for EOF conditions here. We set the page up correctly
6408  * for a written page which means we get ENOSPC checking when writing into
6409  * holes and correct delalloc and unwritten extent mapping on filesystems that
6410  * support these features.
6411  *
6412  * We are not allowed to take the i_mutex here so we have to play games to
6413  * protect against truncate races as the page could now be beyond EOF.  Because
6414  * vmtruncate() writes the inode size before removing pages, once we have the
6415  * page lock we can determine safely if the page is beyond EOF. If it is not
6416  * beyond EOF, then the page is guaranteed safe against truncation until we
6417  * unlock the page.
6418  */
6419 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6420 {
6421         struct page *page = vmf->page;
6422         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6423         struct btrfs_root *root = BTRFS_I(inode)->root;
6424         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6425         struct btrfs_ordered_extent *ordered;
6426         struct extent_state *cached_state = NULL;
6427         char *kaddr;
6428         unsigned long zero_start;
6429         loff_t size;
6430         int ret;
6431         int reserved = 0;
6432         u64 page_start;
6433         u64 page_end;
6434
6435         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6436         if (!ret) {
6437                 ret = btrfs_update_time(vma->vm_file);
6438                 reserved = 1;
6439         }
6440         if (ret) {
6441                 if (ret == -ENOMEM)
6442                         ret = VM_FAULT_OOM;
6443                 else /* -ENOSPC, -EIO, etc */
6444                         ret = VM_FAULT_SIGBUS;
6445                 if (reserved)
6446                         goto out;
6447                 goto out_noreserve;
6448         }
6449
6450         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6451 again:
6452         lock_page(page);
6453         size = i_size_read(inode);
6454         page_start = page_offset(page);
6455         page_end = page_start + PAGE_CACHE_SIZE - 1;
6456
6457         if ((page->mapping != inode->i_mapping) ||
6458             (page_start >= size)) {
6459                 /* page got truncated out from underneath us */
6460                 goto out_unlock;
6461         }
6462         wait_on_page_writeback(page);
6463
6464         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6465                          GFP_NOFS);
6466         set_page_extent_mapped(page);
6467
6468         /*
6469          * we can't set the delalloc bits if there are pending ordered
6470          * extents.  Drop our locks and wait for them to finish
6471          */
6472         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6473         if (ordered) {
6474                 unlock_extent_cached(io_tree, page_start, page_end,
6475                                      &cached_state, GFP_NOFS);
6476                 unlock_page(page);
6477                 btrfs_start_ordered_extent(inode, ordered, 1);
6478                 btrfs_put_ordered_extent(ordered);
6479                 goto again;
6480         }
6481
6482         /*
6483          * XXX - page_mkwrite gets called every time the page is dirtied, even
6484          * if it was already dirty, so for space accounting reasons we need to
6485          * clear any delalloc bits for the range we are fixing to save.  There
6486          * is probably a better way to do this, but for now keep consistent with
6487          * prepare_pages in the normal write path.
6488          */
6489         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6490                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6491                           0, 0, &cached_state, GFP_NOFS);
6492
6493         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6494                                         &cached_state);
6495         if (ret) {
6496                 unlock_extent_cached(io_tree, page_start, page_end,
6497                                      &cached_state, GFP_NOFS);
6498                 ret = VM_FAULT_SIGBUS;
6499                 goto out_unlock;
6500         }
6501         ret = 0;
6502
6503         /* page is wholly or partially inside EOF */
6504         if (page_start + PAGE_CACHE_SIZE > size)
6505                 zero_start = size & ~PAGE_CACHE_MASK;
6506         else
6507                 zero_start = PAGE_CACHE_SIZE;
6508
6509         if (zero_start != PAGE_CACHE_SIZE) {
6510                 kaddr = kmap(page);
6511                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6512                 flush_dcache_page(page);
6513                 kunmap(page);
6514         }
6515         ClearPageChecked(page);
6516         set_page_dirty(page);
6517         SetPageUptodate(page);
6518
6519         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6520         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6521
6522         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6523
6524 out_unlock:
6525         if (!ret)
6526                 return VM_FAULT_LOCKED;
6527         unlock_page(page);
6528 out:
6529         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6530 out_noreserve:
6531         return ret;
6532 }
6533
6534 static int btrfs_truncate(struct inode *inode)
6535 {
6536         struct btrfs_root *root = BTRFS_I(inode)->root;
6537         struct btrfs_block_rsv *rsv;
6538         int ret;
6539         int err = 0;
6540         struct btrfs_trans_handle *trans;
6541         unsigned long nr;
6542         u64 mask = root->sectorsize - 1;
6543         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6544
6545         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6546         if (ret)
6547                 return ret;
6548
6549         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6550         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6551
6552         /*
6553          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6554          * 3 things going on here
6555          *
6556          * 1) We need to reserve space for our orphan item and the space to
6557          * delete our orphan item.  Lord knows we don't want to have a dangling
6558          * orphan item because we didn't reserve space to remove it.
6559          *
6560          * 2) We need to reserve space to update our inode.
6561          *
6562          * 3) We need to have something to cache all the space that is going to
6563          * be free'd up by the truncate operation, but also have some slack
6564          * space reserved in case it uses space during the truncate (thank you
6565          * very much snapshotting).
6566          *
6567          * And we need these to all be seperate.  The fact is we can use alot of
6568          * space doing the truncate, and we have no earthly idea how much space
6569          * we will use, so we need the truncate reservation to be seperate so it
6570          * doesn't end up using space reserved for updating the inode or
6571          * removing the orphan item.  We also need to be able to stop the
6572          * transaction and start a new one, which means we need to be able to
6573          * update the inode several times, and we have no idea of knowing how
6574          * many times that will be, so we can't just reserve 1 item for the
6575          * entirety of the opration, so that has to be done seperately as well.
6576          * Then there is the orphan item, which does indeed need to be held on
6577          * to for the whole operation, and we need nobody to touch this reserved
6578          * space except the orphan code.
6579          *
6580          * So that leaves us with
6581          *
6582          * 1) root->orphan_block_rsv - for the orphan deletion.
6583          * 2) rsv - for the truncate reservation, which we will steal from the
6584          * transaction reservation.
6585          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6586          * updating the inode.
6587          */
6588         rsv = btrfs_alloc_block_rsv(root);
6589         if (!rsv)
6590                 return -ENOMEM;
6591         rsv->size = min_size;
6592
6593         /*
6594          * 1 for the truncate slack space
6595          * 1 for the orphan item we're going to add
6596          * 1 for the orphan item deletion
6597          * 1 for updating the inode.
6598          */
6599         trans = btrfs_start_transaction(root, 4);
6600         if (IS_ERR(trans)) {
6601                 err = PTR_ERR(trans);
6602                 goto out;
6603         }
6604
6605         /* Migrate the slack space for the truncate to our reserve */
6606         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6607                                       min_size);
6608         BUG_ON(ret);
6609
6610         ret = btrfs_orphan_add(trans, inode);
6611         if (ret) {
6612                 btrfs_end_transaction(trans, root);
6613                 goto out;
6614         }
6615
6616         /*
6617          * setattr is responsible for setting the ordered_data_close flag,
6618          * but that is only tested during the last file release.  That
6619          * could happen well after the next commit, leaving a great big
6620          * window where new writes may get lost if someone chooses to write
6621          * to this file after truncating to zero
6622          *
6623          * The inode doesn't have any dirty data here, and so if we commit
6624          * this is a noop.  If someone immediately starts writing to the inode
6625          * it is very likely we'll catch some of their writes in this
6626          * transaction, and the commit will find this file on the ordered
6627          * data list with good things to send down.
6628          *
6629          * This is a best effort solution, there is still a window where
6630          * using truncate to replace the contents of the file will
6631          * end up with a zero length file after a crash.
6632          */
6633         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6634                 btrfs_add_ordered_operation(trans, root, inode);
6635
6636         while (1) {
6637                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6638                 if (ret) {
6639                         /*
6640                          * This can only happen with the original transaction we
6641                          * started above, every other time we shouldn't have a
6642                          * transaction started yet.
6643                          */
6644                         if (ret == -EAGAIN)
6645                                 goto end_trans;
6646                         err = ret;
6647                         break;
6648                 }
6649
6650                 if (!trans) {
6651                         /* Just need the 1 for updating the inode */
6652                         trans = btrfs_start_transaction(root, 1);
6653                         if (IS_ERR(trans)) {
6654                                 ret = err = PTR_ERR(trans);
6655                                 trans = NULL;
6656                                 break;
6657                         }
6658                 }
6659
6660                 trans->block_rsv = rsv;
6661
6662                 ret = btrfs_truncate_inode_items(trans, root, inode,
6663                                                  inode->i_size,
6664                                                  BTRFS_EXTENT_DATA_KEY);
6665                 if (ret != -EAGAIN) {
6666                         err = ret;
6667                         break;
6668                 }
6669
6670                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6671                 ret = btrfs_update_inode(trans, root, inode);
6672                 if (ret) {
6673                         err = ret;
6674                         break;
6675                 }
6676 end_trans:
6677                 nr = trans->blocks_used;
6678                 btrfs_end_transaction(trans, root);
6679                 trans = NULL;
6680                 btrfs_btree_balance_dirty(root, nr);
6681         }
6682
6683         if (ret == 0 && inode->i_nlink > 0) {
6684                 trans->block_rsv = root->orphan_block_rsv;
6685                 ret = btrfs_orphan_del(trans, inode);
6686                 if (ret)
6687                         err = ret;
6688         } else if (ret && inode->i_nlink > 0) {
6689                 /*
6690                  * Failed to do the truncate, remove us from the in memory
6691                  * orphan list.
6692                  */
6693                 ret = btrfs_orphan_del(NULL, inode);
6694         }
6695
6696         if (trans) {
6697                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6698                 ret = btrfs_update_inode(trans, root, inode);
6699                 if (ret && !err)
6700                         err = ret;
6701
6702                 nr = trans->blocks_used;
6703                 ret = btrfs_end_transaction(trans, root);
6704                 btrfs_btree_balance_dirty(root, nr);
6705         }
6706
6707 out:
6708         btrfs_free_block_rsv(root, rsv);
6709
6710         if (ret && !err)
6711                 err = ret;
6712
6713         return err;
6714 }
6715
6716 /*
6717  * create a new subvolume directory/inode (helper for the ioctl).
6718  */
6719 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6720                              struct btrfs_root *new_root, u64 new_dirid)
6721 {
6722         struct inode *inode;
6723         int err;
6724         u64 index = 0;
6725
6726         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6727                                 new_dirid, new_dirid,
6728                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6729                                 &index);
6730         if (IS_ERR(inode))
6731                 return PTR_ERR(inode);
6732         inode->i_op = &btrfs_dir_inode_operations;
6733         inode->i_fop = &btrfs_dir_file_operations;
6734
6735         set_nlink(inode, 1);
6736         btrfs_i_size_write(inode, 0);
6737
6738         err = btrfs_update_inode(trans, new_root, inode);
6739         BUG_ON(err);
6740
6741         iput(inode);
6742         return 0;
6743 }
6744
6745 struct inode *btrfs_alloc_inode(struct super_block *sb)
6746 {
6747         struct btrfs_inode *ei;
6748         struct inode *inode;
6749
6750         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6751         if (!ei)
6752                 return NULL;
6753
6754         ei->root = NULL;
6755         ei->space_info = NULL;
6756         ei->generation = 0;
6757         ei->sequence = 0;
6758         ei->last_trans = 0;
6759         ei->last_sub_trans = 0;
6760         ei->logged_trans = 0;
6761         ei->delalloc_bytes = 0;
6762         ei->disk_i_size = 0;
6763         ei->flags = 0;
6764         ei->csum_bytes = 0;
6765         ei->index_cnt = (u64)-1;
6766         ei->last_unlink_trans = 0;
6767
6768         spin_lock_init(&ei->lock);
6769         ei->outstanding_extents = 0;
6770         ei->reserved_extents = 0;
6771
6772         ei->ordered_data_close = 0;
6773         ei->orphan_meta_reserved = 0;
6774         ei->dummy_inode = 0;
6775         ei->in_defrag = 0;
6776         ei->delalloc_meta_reserved = 0;
6777         ei->force_compress = BTRFS_COMPRESS_NONE;
6778
6779         ei->delayed_node = NULL;
6780
6781         inode = &ei->vfs_inode;
6782         extent_map_tree_init(&ei->extent_tree);
6783         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6784         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6785         mutex_init(&ei->log_mutex);
6786         mutex_init(&ei->delalloc_mutex);
6787         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6788         INIT_LIST_HEAD(&ei->i_orphan);
6789         INIT_LIST_HEAD(&ei->delalloc_inodes);
6790         INIT_LIST_HEAD(&ei->ordered_operations);
6791         RB_CLEAR_NODE(&ei->rb_node);
6792
6793         return inode;
6794 }
6795
6796 static void btrfs_i_callback(struct rcu_head *head)
6797 {
6798         struct inode *inode = container_of(head, struct inode, i_rcu);
6799         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6800 }
6801
6802 void btrfs_destroy_inode(struct inode *inode)
6803 {
6804         struct btrfs_ordered_extent *ordered;
6805         struct btrfs_root *root = BTRFS_I(inode)->root;
6806
6807         WARN_ON(!list_empty(&inode->i_dentry));
6808         WARN_ON(inode->i_data.nrpages);
6809         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6810         WARN_ON(BTRFS_I(inode)->reserved_extents);
6811         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6812         WARN_ON(BTRFS_I(inode)->csum_bytes);
6813
6814         /*
6815          * This can happen where we create an inode, but somebody else also
6816          * created the same inode and we need to destroy the one we already
6817          * created.
6818          */
6819         if (!root)
6820                 goto free;
6821
6822         /*
6823          * Make sure we're properly removed from the ordered operation
6824          * lists.
6825          */
6826         smp_mb();
6827         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6828                 spin_lock(&root->fs_info->ordered_extent_lock);
6829                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6830                 spin_unlock(&root->fs_info->ordered_extent_lock);
6831         }
6832
6833         spin_lock(&root->orphan_lock);
6834         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6835                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6836                        (unsigned long long)btrfs_ino(inode));
6837                 list_del_init(&BTRFS_I(inode)->i_orphan);
6838         }
6839         spin_unlock(&root->orphan_lock);
6840
6841         while (1) {
6842                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6843                 if (!ordered)
6844                         break;
6845                 else {
6846                         printk(KERN_ERR "btrfs found ordered "
6847                                "extent %llu %llu on inode cleanup\n",
6848                                (unsigned long long)ordered->file_offset,
6849                                (unsigned long long)ordered->len);
6850                         btrfs_remove_ordered_extent(inode, ordered);
6851                         btrfs_put_ordered_extent(ordered);
6852                         btrfs_put_ordered_extent(ordered);
6853                 }
6854         }
6855         inode_tree_del(inode);
6856         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6857 free:
6858         btrfs_remove_delayed_node(inode);
6859         call_rcu(&inode->i_rcu, btrfs_i_callback);
6860 }
6861
6862 int btrfs_drop_inode(struct inode *inode)
6863 {
6864         struct btrfs_root *root = BTRFS_I(inode)->root;
6865
6866         if (btrfs_root_refs(&root->root_item) == 0 &&
6867             !btrfs_is_free_space_inode(root, inode))
6868                 return 1;
6869         else
6870                 return generic_drop_inode(inode);
6871 }
6872
6873 static void init_once(void *foo)
6874 {
6875         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6876
6877         inode_init_once(&ei->vfs_inode);
6878 }
6879
6880 void btrfs_destroy_cachep(void)
6881 {
6882         if (btrfs_inode_cachep)
6883                 kmem_cache_destroy(btrfs_inode_cachep);
6884         if (btrfs_trans_handle_cachep)
6885                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6886         if (btrfs_transaction_cachep)
6887                 kmem_cache_destroy(btrfs_transaction_cachep);
6888         if (btrfs_path_cachep)
6889                 kmem_cache_destroy(btrfs_path_cachep);
6890         if (btrfs_free_space_cachep)
6891                 kmem_cache_destroy(btrfs_free_space_cachep);
6892 }
6893
6894 int btrfs_init_cachep(void)
6895 {
6896         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6897                         sizeof(struct btrfs_inode), 0,
6898                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6899         if (!btrfs_inode_cachep)
6900                 goto fail;
6901
6902         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6903                         sizeof(struct btrfs_trans_handle), 0,
6904                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6905         if (!btrfs_trans_handle_cachep)
6906                 goto fail;
6907
6908         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6909                         sizeof(struct btrfs_transaction), 0,
6910                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6911         if (!btrfs_transaction_cachep)
6912                 goto fail;
6913
6914         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6915                         sizeof(struct btrfs_path), 0,
6916                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6917         if (!btrfs_path_cachep)
6918                 goto fail;
6919
6920         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6921                         sizeof(struct btrfs_free_space), 0,
6922                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6923         if (!btrfs_free_space_cachep)
6924                 goto fail;
6925
6926         return 0;
6927 fail:
6928         btrfs_destroy_cachep();
6929         return -ENOMEM;
6930 }
6931
6932 static int btrfs_getattr(struct vfsmount *mnt,
6933                          struct dentry *dentry, struct kstat *stat)
6934 {
6935         struct inode *inode = dentry->d_inode;
6936         u32 blocksize = inode->i_sb->s_blocksize;
6937
6938         generic_fillattr(inode, stat);
6939         stat->dev = BTRFS_I(inode)->root->anon_dev;
6940         stat->blksize = PAGE_CACHE_SIZE;
6941         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6942                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6943         return 0;
6944 }
6945
6946 /*
6947  * If a file is moved, it will inherit the cow and compression flags of the new
6948  * directory.
6949  */
6950 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6951 {
6952         struct btrfs_inode *b_dir = BTRFS_I(dir);
6953         struct btrfs_inode *b_inode = BTRFS_I(inode);
6954
6955         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6956                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6957         else
6958                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6959
6960         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6961                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6962         else
6963                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6964 }
6965
6966 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6967                            struct inode *new_dir, struct dentry *new_dentry)
6968 {
6969         struct btrfs_trans_handle *trans;
6970         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6971         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6972         struct inode *new_inode = new_dentry->d_inode;
6973         struct inode *old_inode = old_dentry->d_inode;
6974         struct timespec ctime = CURRENT_TIME;
6975         u64 index = 0;
6976         u64 root_objectid;
6977         int ret;
6978         u64 old_ino = btrfs_ino(old_inode);
6979
6980         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6981                 return -EPERM;
6982
6983         /* we only allow rename subvolume link between subvolumes */
6984         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6985                 return -EXDEV;
6986
6987         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6988             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6989                 return -ENOTEMPTY;
6990
6991         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6992             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6993                 return -ENOTEMPTY;
6994         /*
6995          * we're using rename to replace one file with another.
6996          * and the replacement file is large.  Start IO on it now so
6997          * we don't add too much work to the end of the transaction
6998          */
6999         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7000             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7001                 filemap_flush(old_inode->i_mapping);
7002
7003         /* close the racy window with snapshot create/destroy ioctl */
7004         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7005                 down_read(&root->fs_info->subvol_sem);
7006         /*
7007          * We want to reserve the absolute worst case amount of items.  So if
7008          * both inodes are subvols and we need to unlink them then that would
7009          * require 4 item modifications, but if they are both normal inodes it
7010          * would require 5 item modifications, so we'll assume their normal
7011          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7012          * should cover the worst case number of items we'll modify.
7013          */
7014         trans = btrfs_start_transaction(root, 20);
7015         if (IS_ERR(trans)) {
7016                 ret = PTR_ERR(trans);
7017                 goto out_notrans;
7018         }
7019
7020         if (dest != root)
7021                 btrfs_record_root_in_trans(trans, dest);
7022
7023         ret = btrfs_set_inode_index(new_dir, &index);
7024         if (ret)
7025                 goto out_fail;
7026
7027         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7028                 /* force full log commit if subvolume involved. */
7029                 root->fs_info->last_trans_log_full_commit = trans->transid;
7030         } else {
7031                 ret = btrfs_insert_inode_ref(trans, dest,
7032                                              new_dentry->d_name.name,
7033                                              new_dentry->d_name.len,
7034                                              old_ino,
7035                                              btrfs_ino(new_dir), index);
7036                 if (ret)
7037                         goto out_fail;
7038                 /*
7039                  * this is an ugly little race, but the rename is required
7040                  * to make sure that if we crash, the inode is either at the
7041                  * old name or the new one.  pinning the log transaction lets
7042                  * us make sure we don't allow a log commit to come in after
7043                  * we unlink the name but before we add the new name back in.
7044                  */
7045                 btrfs_pin_log_trans(root);
7046         }
7047         /*
7048          * make sure the inode gets flushed if it is replacing
7049          * something.
7050          */
7051         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7052                 btrfs_add_ordered_operation(trans, root, old_inode);
7053
7054         old_dir->i_ctime = old_dir->i_mtime = ctime;
7055         new_dir->i_ctime = new_dir->i_mtime = ctime;
7056         old_inode->i_ctime = ctime;
7057
7058         if (old_dentry->d_parent != new_dentry->d_parent)
7059                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7060
7061         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7062                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7063                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7064                                         old_dentry->d_name.name,
7065                                         old_dentry->d_name.len);
7066         } else {
7067                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7068                                         old_dentry->d_inode,
7069                                         old_dentry->d_name.name,
7070                                         old_dentry->d_name.len);
7071                 if (!ret)
7072                         ret = btrfs_update_inode(trans, root, old_inode);
7073         }
7074         BUG_ON(ret);
7075
7076         if (new_inode) {
7077                 new_inode->i_ctime = CURRENT_TIME;
7078                 if (unlikely(btrfs_ino(new_inode) ==
7079                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7080                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7081                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7082                                                 root_objectid,
7083                                                 new_dentry->d_name.name,
7084                                                 new_dentry->d_name.len);
7085                         BUG_ON(new_inode->i_nlink == 0);
7086                 } else {
7087                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7088                                                  new_dentry->d_inode,
7089                                                  new_dentry->d_name.name,
7090                                                  new_dentry->d_name.len);
7091                 }
7092                 BUG_ON(ret);
7093                 if (new_inode->i_nlink == 0) {
7094                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7095                         BUG_ON(ret);
7096                 }
7097         }
7098
7099         fixup_inode_flags(new_dir, old_inode);
7100
7101         ret = btrfs_add_link(trans, new_dir, old_inode,
7102                              new_dentry->d_name.name,
7103                              new_dentry->d_name.len, 0, index);
7104         BUG_ON(ret);
7105
7106         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7107                 struct dentry *parent = new_dentry->d_parent;
7108                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7109                 btrfs_end_log_trans(root);
7110         }
7111 out_fail:
7112         btrfs_end_transaction(trans, root);
7113 out_notrans:
7114         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7115                 up_read(&root->fs_info->subvol_sem);
7116
7117         return ret;
7118 }
7119
7120 /*
7121  * some fairly slow code that needs optimization. This walks the list
7122  * of all the inodes with pending delalloc and forces them to disk.
7123  */
7124 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7125 {
7126         struct list_head *head = &root->fs_info->delalloc_inodes;
7127         struct btrfs_inode *binode;
7128         struct inode *inode;
7129
7130         if (root->fs_info->sb->s_flags & MS_RDONLY)
7131                 return -EROFS;
7132
7133         spin_lock(&root->fs_info->delalloc_lock);
7134         while (!list_empty(head)) {
7135                 binode = list_entry(head->next, struct btrfs_inode,
7136                                     delalloc_inodes);
7137                 inode = igrab(&binode->vfs_inode);
7138                 if (!inode)
7139                         list_del_init(&binode->delalloc_inodes);
7140                 spin_unlock(&root->fs_info->delalloc_lock);
7141                 if (inode) {
7142                         filemap_flush(inode->i_mapping);
7143                         if (delay_iput)
7144                                 btrfs_add_delayed_iput(inode);
7145                         else
7146                                 iput(inode);
7147                 }
7148                 cond_resched();
7149                 spin_lock(&root->fs_info->delalloc_lock);
7150         }
7151         spin_unlock(&root->fs_info->delalloc_lock);
7152
7153         /* the filemap_flush will queue IO into the worker threads, but
7154          * we have to make sure the IO is actually started and that
7155          * ordered extents get created before we return
7156          */
7157         atomic_inc(&root->fs_info->async_submit_draining);
7158         while (atomic_read(&root->fs_info->nr_async_submits) ||
7159               atomic_read(&root->fs_info->async_delalloc_pages)) {
7160                 wait_event(root->fs_info->async_submit_wait,
7161                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7162                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7163         }
7164         atomic_dec(&root->fs_info->async_submit_draining);
7165         return 0;
7166 }
7167
7168 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7169                          const char *symname)
7170 {
7171         struct btrfs_trans_handle *trans;
7172         struct btrfs_root *root = BTRFS_I(dir)->root;
7173         struct btrfs_path *path;
7174         struct btrfs_key key;
7175         struct inode *inode = NULL;
7176         int err;
7177         int drop_inode = 0;
7178         u64 objectid;
7179         u64 index = 0 ;
7180         int name_len;
7181         int datasize;
7182         unsigned long ptr;
7183         struct btrfs_file_extent_item *ei;
7184         struct extent_buffer *leaf;
7185         unsigned long nr = 0;
7186
7187         name_len = strlen(symname) + 1;
7188         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7189                 return -ENAMETOOLONG;
7190
7191         /*
7192          * 2 items for inode item and ref
7193          * 2 items for dir items
7194          * 1 item for xattr if selinux is on
7195          */
7196         trans = btrfs_start_transaction(root, 5);
7197         if (IS_ERR(trans))
7198                 return PTR_ERR(trans);
7199
7200         err = btrfs_find_free_ino(root, &objectid);
7201         if (err)
7202                 goto out_unlock;
7203
7204         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7205                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7206                                 S_IFLNK|S_IRWXUGO, &index);
7207         if (IS_ERR(inode)) {
7208                 err = PTR_ERR(inode);
7209                 goto out_unlock;
7210         }
7211
7212         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7213         if (err) {
7214                 drop_inode = 1;
7215                 goto out_unlock;
7216         }
7217
7218         /*
7219         * If the active LSM wants to access the inode during
7220         * d_instantiate it needs these. Smack checks to see
7221         * if the filesystem supports xattrs by looking at the
7222         * ops vector.
7223         */
7224         inode->i_fop = &btrfs_file_operations;
7225         inode->i_op = &btrfs_file_inode_operations;
7226
7227         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7228         if (err)
7229                 drop_inode = 1;
7230         else {
7231                 inode->i_mapping->a_ops = &btrfs_aops;
7232                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7233                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7234         }
7235         if (drop_inode)
7236                 goto out_unlock;
7237
7238         path = btrfs_alloc_path();
7239         if (!path) {
7240                 err = -ENOMEM;
7241                 drop_inode = 1;
7242                 goto out_unlock;
7243         }
7244         key.objectid = btrfs_ino(inode);
7245         key.offset = 0;
7246         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7247         datasize = btrfs_file_extent_calc_inline_size(name_len);
7248         err = btrfs_insert_empty_item(trans, root, path, &key,
7249                                       datasize);
7250         if (err) {
7251                 drop_inode = 1;
7252                 btrfs_free_path(path);
7253                 goto out_unlock;
7254         }
7255         leaf = path->nodes[0];
7256         ei = btrfs_item_ptr(leaf, path->slots[0],
7257                             struct btrfs_file_extent_item);
7258         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7259         btrfs_set_file_extent_type(leaf, ei,
7260                                    BTRFS_FILE_EXTENT_INLINE);
7261         btrfs_set_file_extent_encryption(leaf, ei, 0);
7262         btrfs_set_file_extent_compression(leaf, ei, 0);
7263         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7264         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7265
7266         ptr = btrfs_file_extent_inline_start(ei);
7267         write_extent_buffer(leaf, symname, ptr, name_len);
7268         btrfs_mark_buffer_dirty(leaf);
7269         btrfs_free_path(path);
7270
7271         inode->i_op = &btrfs_symlink_inode_operations;
7272         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7273         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7274         inode_set_bytes(inode, name_len);
7275         btrfs_i_size_write(inode, name_len - 1);
7276         err = btrfs_update_inode(trans, root, inode);
7277         if (err)
7278                 drop_inode = 1;
7279
7280 out_unlock:
7281         if (!err)
7282                 d_instantiate(dentry, inode);
7283         nr = trans->blocks_used;
7284         btrfs_end_transaction(trans, root);
7285         if (drop_inode) {
7286                 inode_dec_link_count(inode);
7287                 iput(inode);
7288         }
7289         btrfs_btree_balance_dirty(root, nr);
7290         return err;
7291 }
7292
7293 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7294                                        u64 start, u64 num_bytes, u64 min_size,
7295                                        loff_t actual_len, u64 *alloc_hint,
7296                                        struct btrfs_trans_handle *trans)
7297 {
7298         struct btrfs_root *root = BTRFS_I(inode)->root;
7299         struct btrfs_key ins;
7300         u64 cur_offset = start;
7301         u64 i_size;
7302         int ret = 0;
7303         bool own_trans = true;
7304
7305         if (trans)
7306                 own_trans = false;
7307         while (num_bytes > 0) {
7308                 if (own_trans) {
7309                         trans = btrfs_start_transaction(root, 3);
7310                         if (IS_ERR(trans)) {
7311                                 ret = PTR_ERR(trans);
7312                                 break;
7313                         }
7314                 }
7315
7316                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7317                                            0, *alloc_hint, &ins, 1);
7318                 if (ret) {
7319                         if (own_trans)
7320                                 btrfs_end_transaction(trans, root);
7321                         break;
7322                 }
7323
7324                 ret = insert_reserved_file_extent(trans, inode,
7325                                                   cur_offset, ins.objectid,
7326                                                   ins.offset, ins.offset,
7327                                                   ins.offset, 0, 0, 0,
7328                                                   BTRFS_FILE_EXTENT_PREALLOC);
7329                 BUG_ON(ret);
7330                 btrfs_drop_extent_cache(inode, cur_offset,
7331                                         cur_offset + ins.offset -1, 0);
7332
7333                 num_bytes -= ins.offset;
7334                 cur_offset += ins.offset;
7335                 *alloc_hint = ins.objectid + ins.offset;
7336
7337                 inode->i_ctime = CURRENT_TIME;
7338                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7339                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7340                     (actual_len > inode->i_size) &&
7341                     (cur_offset > inode->i_size)) {
7342                         if (cur_offset > actual_len)
7343                                 i_size = actual_len;
7344                         else
7345                                 i_size = cur_offset;
7346                         i_size_write(inode, i_size);
7347                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7348                 }
7349
7350                 ret = btrfs_update_inode(trans, root, inode);
7351                 BUG_ON(ret);
7352
7353                 if (own_trans)
7354                         btrfs_end_transaction(trans, root);
7355         }
7356         return ret;
7357 }
7358
7359 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7360                               u64 start, u64 num_bytes, u64 min_size,
7361                               loff_t actual_len, u64 *alloc_hint)
7362 {
7363         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7364                                            min_size, actual_len, alloc_hint,
7365                                            NULL);
7366 }
7367
7368 int btrfs_prealloc_file_range_trans(struct inode *inode,
7369                                     struct btrfs_trans_handle *trans, int mode,
7370                                     u64 start, u64 num_bytes, u64 min_size,
7371                                     loff_t actual_len, u64 *alloc_hint)
7372 {
7373         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7374                                            min_size, actual_len, alloc_hint, trans);
7375 }
7376
7377 static int btrfs_set_page_dirty(struct page *page)
7378 {
7379         return __set_page_dirty_nobuffers(page);
7380 }
7381
7382 static int btrfs_permission(struct inode *inode, int mask)
7383 {
7384         struct btrfs_root *root = BTRFS_I(inode)->root;
7385         umode_t mode = inode->i_mode;
7386
7387         if (mask & MAY_WRITE &&
7388             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7389                 if (btrfs_root_readonly(root))
7390                         return -EROFS;
7391                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7392                         return -EACCES;
7393         }
7394         return generic_permission(inode, mask);
7395 }
7396
7397 static const struct inode_operations btrfs_dir_inode_operations = {
7398         .getattr        = btrfs_getattr,
7399         .lookup         = btrfs_lookup,
7400         .create         = btrfs_create,
7401         .unlink         = btrfs_unlink,
7402         .link           = btrfs_link,
7403         .mkdir          = btrfs_mkdir,
7404         .rmdir          = btrfs_rmdir,
7405         .rename         = btrfs_rename,
7406         .symlink        = btrfs_symlink,
7407         .setattr        = btrfs_setattr,
7408         .mknod          = btrfs_mknod,
7409         .setxattr       = btrfs_setxattr,
7410         .getxattr       = btrfs_getxattr,
7411         .listxattr      = btrfs_listxattr,
7412         .removexattr    = btrfs_removexattr,
7413         .permission     = btrfs_permission,
7414         .get_acl        = btrfs_get_acl,
7415 };
7416 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7417         .lookup         = btrfs_lookup,
7418         .permission     = btrfs_permission,
7419         .get_acl        = btrfs_get_acl,
7420 };
7421
7422 static const struct file_operations btrfs_dir_file_operations = {
7423         .llseek         = generic_file_llseek,
7424         .read           = generic_read_dir,
7425         .readdir        = btrfs_real_readdir,
7426         .unlocked_ioctl = btrfs_ioctl,
7427 #ifdef CONFIG_COMPAT
7428         .compat_ioctl   = btrfs_ioctl,
7429 #endif
7430         .release        = btrfs_release_file,
7431         .fsync          = btrfs_sync_file,
7432 };
7433
7434 static struct extent_io_ops btrfs_extent_io_ops = {
7435         .fill_delalloc = run_delalloc_range,
7436         .submit_bio_hook = btrfs_submit_bio_hook,
7437         .merge_bio_hook = btrfs_merge_bio_hook,
7438         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7439         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7440         .writepage_start_hook = btrfs_writepage_start_hook,
7441         .set_bit_hook = btrfs_set_bit_hook,
7442         .clear_bit_hook = btrfs_clear_bit_hook,
7443         .merge_extent_hook = btrfs_merge_extent_hook,
7444         .split_extent_hook = btrfs_split_extent_hook,
7445 };
7446
7447 /*
7448  * btrfs doesn't support the bmap operation because swapfiles
7449  * use bmap to make a mapping of extents in the file.  They assume
7450  * these extents won't change over the life of the file and they
7451  * use the bmap result to do IO directly to the drive.
7452  *
7453  * the btrfs bmap call would return logical addresses that aren't
7454  * suitable for IO and they also will change frequently as COW
7455  * operations happen.  So, swapfile + btrfs == corruption.
7456  *
7457  * For now we're avoiding this by dropping bmap.
7458  */
7459 static const struct address_space_operations btrfs_aops = {
7460         .readpage       = btrfs_readpage,
7461         .writepage      = btrfs_writepage,
7462         .writepages     = btrfs_writepages,
7463         .readpages      = btrfs_readpages,
7464         .direct_IO      = btrfs_direct_IO,
7465         .invalidatepage = btrfs_invalidatepage,
7466         .releasepage    = btrfs_releasepage,
7467         .set_page_dirty = btrfs_set_page_dirty,
7468         .error_remove_page = generic_error_remove_page,
7469 };
7470
7471 static const struct address_space_operations btrfs_symlink_aops = {
7472         .readpage       = btrfs_readpage,
7473         .writepage      = btrfs_writepage,
7474         .invalidatepage = btrfs_invalidatepage,
7475         .releasepage    = btrfs_releasepage,
7476 };
7477
7478 static const struct inode_operations btrfs_file_inode_operations = {
7479         .getattr        = btrfs_getattr,
7480         .setattr        = btrfs_setattr,
7481         .setxattr       = btrfs_setxattr,
7482         .getxattr       = btrfs_getxattr,
7483         .listxattr      = btrfs_listxattr,
7484         .removexattr    = btrfs_removexattr,
7485         .permission     = btrfs_permission,
7486         .fiemap         = btrfs_fiemap,
7487         .get_acl        = btrfs_get_acl,
7488 };
7489 static const struct inode_operations btrfs_special_inode_operations = {
7490         .getattr        = btrfs_getattr,
7491         .setattr        = btrfs_setattr,
7492         .permission     = btrfs_permission,
7493         .setxattr       = btrfs_setxattr,
7494         .getxattr       = btrfs_getxattr,
7495         .listxattr      = btrfs_listxattr,
7496         .removexattr    = btrfs_removexattr,
7497         .get_acl        = btrfs_get_acl,
7498 };
7499 static const struct inode_operations btrfs_symlink_inode_operations = {
7500         .readlink       = generic_readlink,
7501         .follow_link    = page_follow_link_light,
7502         .put_link       = page_put_link,
7503         .getattr        = btrfs_getattr,
7504         .setattr        = btrfs_setattr,
7505         .permission     = btrfs_permission,
7506         .setxattr       = btrfs_setxattr,
7507         .getxattr       = btrfs_getxattr,
7508         .listxattr      = btrfs_listxattr,
7509         .removexattr    = btrfs_removexattr,
7510         .get_acl        = btrfs_get_acl,
7511 };
7512
7513 const struct dentry_operations btrfs_dentry_operations = {
7514         .d_delete       = btrfs_dentry_delete,
7515         .d_release      = btrfs_dentry_release,
7516 };