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 "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "volumes.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 if (!pages) {
 397                         /* just bail out to the uncompressed code */
 398                         goto cont;
 399                 }
 400
 401                 if (BTRFS_I(inode)->force_compress)
 402                         compress_type = BTRFS_I(inode)->force_compress;
 403
 404                 ret = btrfs_compress_pages(compress_type,
 405                                            inode->i_mapping, start,
 406                                            total_compressed, pages,
 407                                            nr_pages, &nr_pages_ret,
 408                                            &total_in,
 409                                            &total_compressed,
 410                                            max_compressed);
 411
 412                 if (!ret) {
 413                         unsigned long offset = total_compressed &
 414                                 (PAGE_CACHE_SIZE - 1);
 415                         struct page *page = pages[nr_pages_ret - 1];
 416                         char *kaddr;
 417
 418                         /* zero the tail end of the last page, we might be
 419                          * sending it down to disk
 420                          */
 421                         if (offset) {
 422                                 kaddr = kmap_atomic(page, KM_USER0);
 423                                 memset(kaddr + offset, 0,
 424                                        PAGE_CACHE_SIZE - offset);
 425                                 kunmap_atomic(kaddr, KM_USER0);
 426                         }
 427                         will_compress = 1;
 428                 }
 429         }
 430 cont:
 431         if (start == 0) {
 432                 trans = btrfs_join_transaction(root);
 433                 BUG_ON(IS_ERR(trans));
 434                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 435
 436                 /* lets try to make an inline extent */
 437                 if (ret || total_in < (actual_end - start)) {
 438                         /* we didn't compress the entire range, try
 439                          * to make an uncompressed inline extent.
 440                          */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end, 0, 0, NULL);
 443                 } else {
 444                         /* try making a compressed inline extent */
 445                         ret = cow_file_range_inline(trans, root, inode,
 446                                                     start, end,
 447                                                     total_compressed,
 448                                                     compress_type, pages);
 449                 }
 450                 if (ret == 0) {
 451                         /*
 452                          * inline extent creation worked, we don't need
 453                          * to create any more async work items.  Unlock
 454                          * and free up our temp pages.
 455                          */
 456                         extent_clear_unlock_delalloc(inode,
 457                              &BTRFS_I(inode)->io_tree,
 458                              start, end, NULL,
 459                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 460                              EXTENT_CLEAR_DELALLOC |
 461                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 462
 463                         btrfs_end_transaction(trans, root);
 464                         goto free_pages_out;
 465                 }
 466                 btrfs_end_transaction(trans, root);
 467         }
 468
 469         if (will_compress) {
 470                 /*
 471                  * we aren't doing an inline extent round the compressed size
 472                  * up to a block size boundary so the allocator does sane
 473                  * things
 474                  */
 475                 total_compressed = (total_compressed + blocksize - 1) &
 476                         ~(blocksize - 1);
 477
 478                 /*
 479                  * one last check to make sure the compression is really a
 480                  * win, compare the page count read with the blocks on disk
 481                  */
 482                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 483                         ~(PAGE_CACHE_SIZE - 1);
 484                 if (total_compressed >= total_in) {
 485                         will_compress = 0;
 486                 } else {
 487                         num_bytes = total_in;
 488                 }
 489         }
 490         if (!will_compress && pages) {
 491                 /*
 492                  * the compression code ran but failed to make things smaller,
 493                  * free any pages it allocated and our page pointer array
 494                  */
 495                 for (i = 0; i < nr_pages_ret; i++) {
 496                         WARN_ON(pages[i]->mapping);
 497                         page_cache_release(pages[i]);
 498                 }
 499                 kfree(pages);
 500                 pages = NULL;
 501                 total_compressed = 0;
 502                 nr_pages_ret = 0;
 503
 504                 /* flag the file so we don't compress in the future */
 505                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 506                     !(BTRFS_I(inode)->force_compress)) {
 507                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 508                 }
 509         }
 510         if (will_compress) {
 511                 *num_added += 1;
 512
 513                 /* the async work queues will take care of doing actual
 514                  * allocation on disk for these compressed pages,
 515                  * and will submit them to the elevator.
 516                  */
 517                 add_async_extent(async_cow, start, num_bytes,
 518                                  total_compressed, pages, nr_pages_ret,
 519                                  compress_type);
 520
 521                 if (start + num_bytes < end) {
 522                         start += num_bytes;
 523                         pages = NULL;
 524                         cond_resched();
 525                         goto again;
 526                 }
 527         } else {
 528 cleanup_and_bail_uncompressed:
 529                 /*
 530                  * No compression, but we still need to write the pages in
 531                  * the file we've been given so far.  redirty the locked
 532                  * page if it corresponds to our extent and set things up
 533                  * for the async work queue to run cow_file_range to do
 534                  * the normal delalloc dance
 535                  */
 536                 if (page_offset(locked_page) >= start &&
 537                     page_offset(locked_page) <= end) {
 538                         __set_page_dirty_nobuffers(locked_page);
 539                         /* unlocked later on in the async handlers */
 540                 }
 541                 add_async_extent(async_cow, start, end - start + 1,
 542                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 543                 *num_added += 1;
 544         }
 545
 546 out:
 547         return 0;
 548
 549 free_pages_out:
 550         for (i = 0; i < nr_pages_ret; i++) {
 551                 WARN_ON(pages[i]->mapping);
 552                 page_cache_release(pages[i]);
 553         }
 554         kfree(pages);
 555
 556         goto out;
 557 }
 558
 559 /*
 560  * phase two of compressed writeback.  This is the ordered portion
 561  * of the code, which only gets called in the order the work was
 562  * queued.  We walk all the async extents created by compress_file_range
 563  * and send them down to the disk.
 564  */
 565 static noinline int submit_compressed_extents(struct inode *inode,
 566                                               struct async_cow *async_cow)
 567 {
 568         struct async_extent *async_extent;
 569         u64 alloc_hint = 0;
 570         struct btrfs_trans_handle *trans;
 571         struct btrfs_key ins;
 572         struct extent_map *em;
 573         struct btrfs_root *root = BTRFS_I(inode)->root;
 574         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 575         struct extent_io_tree *io_tree;
 576         int ret = 0;
 577
 578         if (list_empty(&async_cow->extents))
 579                 return 0;
 580
 581
 582         while (!list_empty(&async_cow->extents)) {
 583                 async_extent = list_entry(async_cow->extents.next,
 584                                           struct async_extent, list);
 585                 list_del(&async_extent->list);
 586
 587                 io_tree = &BTRFS_I(inode)->io_tree;
 588
 589 retry:
 590                 /* did the compression code fall back to uncompressed IO? */
 591                 if (!async_extent->pages) {
 592                         int page_started = 0;
 593                         unsigned long nr_written = 0;
 594
 595                         lock_extent(io_tree, async_extent->start,
 596                                          async_extent->start +
 597                                          async_extent->ram_size - 1, GFP_NOFS);
 598
 599                         /* allocate blocks */
 600                         ret = cow_file_range(inode, async_cow->locked_page,
 601                                              async_extent->start,
 602                                              async_extent->start +
 603                                              async_extent->ram_size - 1,
 604                                              &page_started, &nr_written, 0);
 605
 606                         /*
 607                          * if page_started, cow_file_range inserted an
 608                          * inline extent and took care of all the unlocking
 609                          * and IO for us.  Otherwise, we need to submit
 610                          * all those pages down to the drive.
 611                          */
 612                         if (!page_started && !ret)
 613                                 extent_write_locked_range(io_tree,
 614                                                   inode, async_extent->start,
 615                                                   async_extent->start +
 616                                                   async_extent->ram_size - 1,
 617                                                   btrfs_get_extent,
 618                                                   WB_SYNC_ALL);
 619                         kfree(async_extent);
 620                         cond_resched();
 621                         continue;
 622                 }
 623
 624                 lock_extent(io_tree, async_extent->start,
 625                             async_extent->start + async_extent->ram_size - 1,
 626                             GFP_NOFS);
 627
 628                 trans = btrfs_join_transaction(root);
 629                 BUG_ON(IS_ERR(trans));
 630                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 631                 ret = btrfs_reserve_extent(trans, root,
 632                                            async_extent->compressed_size,
 633                                            async_extent->compressed_size,
 634                                            0, alloc_hint,
 635                                            (u64)-1, &ins, 1);
 636                 btrfs_end_transaction(trans, root);
 637
 638                 if (ret) {
 639                         int i;
 640                         for (i = 0; i < async_extent->nr_pages; i++) {
 641                                 WARN_ON(async_extent->pages[i]->mapping);
 642                                 page_cache_release(async_extent->pages[i]);
 643                         }
 644                         kfree(async_extent->pages);
 645                         async_extent->nr_pages = 0;
 646                         async_extent->pages = NULL;
 647                         unlock_extent(io_tree, async_extent->start,
 648                                       async_extent->start +
 649                                       async_extent->ram_size - 1, GFP_NOFS);
 650                         goto retry;
 651                 }
 652
 653                 /*
 654                  * here we're doing allocation and writeback of the
 655                  * compressed pages
 656                  */
 657                 btrfs_drop_extent_cache(inode, async_extent->start,
 658                                         async_extent->start +
 659                                         async_extent->ram_size - 1, 0);
 660
 661                 em = alloc_extent_map();
 662                 BUG_ON(!em);
 663                 em->start = async_extent->start;
 664                 em->len = async_extent->ram_size;
 665                 em->orig_start = em->start;
 666
 667                 em->block_start = ins.objectid;
 668                 em->block_len = ins.offset;
 669                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 670                 em->compress_type = async_extent->compress_type;
 671                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 672                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 673
 674                 while (1) {
 675                         write_lock(&em_tree->lock);
 676                         ret = add_extent_mapping(em_tree, em);
 677                         write_unlock(&em_tree->lock);
 678                         if (ret != -EEXIST) {
 679                                 free_extent_map(em);
 680                                 break;
 681                         }
 682                         btrfs_drop_extent_cache(inode, async_extent->start,
 683                                                 async_extent->start +
 684                                                 async_extent->ram_size - 1, 0);
 685                 }
 686
 687                 ret = btrfs_add_ordered_extent_compress(inode,
 688                                                 async_extent->start,
 689                                                 ins.objectid,
 690                                                 async_extent->ram_size,
 691                                                 ins.offset,
 692                                                 BTRFS_ORDERED_COMPRESSED,
 693                                                 async_extent->compress_type);
 694                 BUG_ON(ret);
 695
 696                 /*
 697                  * clear dirty, set writeback and unlock the pages.
 698                  */
 699                 extent_clear_unlock_delalloc(inode,
 700                                 &BTRFS_I(inode)->io_tree,
 701                                 async_extent->start,
 702                                 async_extent->start +
 703                                 async_extent->ram_size - 1,
 704                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 705                                 EXTENT_CLEAR_UNLOCK |
 706                                 EXTENT_CLEAR_DELALLOC |
 707                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 708
 709                 ret = btrfs_submit_compressed_write(inode,
 710                                     async_extent->start,
 711                                     async_extent->ram_size,
 712                                     ins.objectid,
 713                                     ins.offset, async_extent->pages,
 714                                     async_extent->nr_pages);
 715
 716                 BUG_ON(ret);
 717                 alloc_hint = ins.objectid + ins.offset;
 718                 kfree(async_extent);
 719                 cond_resched();
 720         }
 721
 722         return 0;
 723 }
 724
 725 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 726                                       u64 num_bytes)
 727 {
 728         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 729         struct extent_map *em;
 730         u64 alloc_hint = 0;
 731
 732         read_lock(&em_tree->lock);
 733         em = search_extent_mapping(em_tree, start, num_bytes);
 734         if (em) {
 735                 /*
 736                  * if block start isn't an actual block number then find the
 737                  * first block in this inode and use that as a hint.  If that
 738                  * block is also bogus then just don't worry about it.
 739                  */
 740                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 741                         free_extent_map(em);
 742                         em = search_extent_mapping(em_tree, 0, 0);
 743                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 744                                 alloc_hint = em->block_start;
 745                         if (em)
 746                                 free_extent_map(em);
 747                 } else {
 748                         alloc_hint = em->block_start;
 749                         free_extent_map(em);
 750                 }
 751         }
 752         read_unlock(&em_tree->lock);
 753
 754         return alloc_hint;
 755 }
 756
 757 /*
 758  * when extent_io.c finds a delayed allocation range in the file,
 759  * the call backs end up in this code.  The basic idea is to
 760  * allocate extents on disk for the range, and create ordered data structs
 761  * in ram to track those extents.
 762  *
 763  * locked_page is the page that writepage had locked already.  We use
 764  * it to make sure we don't do extra locks or unlocks.
 765  *
 766  * *page_started is set to one if we unlock locked_page and do everything
 767  * required to start IO on it.  It may be clean and already done with
 768  * IO when we return.
 769  */
 770 static noinline int cow_file_range(struct inode *inode,
 771                                    struct page *locked_page,
 772                                    u64 start, u64 end, int *page_started,
 773                                    unsigned long *nr_written,
 774                                    int unlock)
 775 {
 776         struct btrfs_root *root = BTRFS_I(inode)->root;
 777         struct btrfs_trans_handle *trans;
 778         u64 alloc_hint = 0;
 779         u64 num_bytes;
 780         unsigned long ram_size;
 781         u64 disk_num_bytes;
 782         u64 cur_alloc_size;
 783         u64 blocksize = root->sectorsize;
 784         struct btrfs_key ins;
 785         struct extent_map *em;
 786         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 787         int ret = 0;
 788
 789         BUG_ON(btrfs_is_free_space_inode(root, inode));
 790         trans = btrfs_join_transaction(root);
 791         BUG_ON(IS_ERR(trans));
 792         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 793
 794         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 795         num_bytes = max(blocksize,  num_bytes);
 796         disk_num_bytes = num_bytes;
 797         ret = 0;
 798
 799         /* if this is a small write inside eof, kick off defrag */
 800         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 801                 btrfs_add_inode_defrag(trans, inode);
 802
 803         if (start == 0) {
 804                 /* lets try to make an inline extent */
 805                 ret = cow_file_range_inline(trans, root, inode,
 806                                             start, end, 0, 0, NULL);
 807                 if (ret == 0) {
 808                         extent_clear_unlock_delalloc(inode,
 809                                      &BTRFS_I(inode)->io_tree,
 810                                      start, end, NULL,
 811                                      EXTENT_CLEAR_UNLOCK_PAGE |
 812                                      EXTENT_CLEAR_UNLOCK |
 813                                      EXTENT_CLEAR_DELALLOC |
 814                                      EXTENT_CLEAR_DIRTY |
 815                                      EXTENT_SET_WRITEBACK |
 816                                      EXTENT_END_WRITEBACK);
 817
 818                         *nr_written = *nr_written +
 819                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 820                         *page_started = 1;
 821                         ret = 0;
 822                         goto out;
 823                 }
 824         }
 825
 826         BUG_ON(disk_num_bytes >
 827                btrfs_super_total_bytes(root->fs_info->super_copy));
 828
 829         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 830         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 831
 832         while (disk_num_bytes > 0) {
 833                 unsigned long op;
 834
 835                 cur_alloc_size = disk_num_bytes;
 836                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 837                                            root->sectorsize, 0, alloc_hint,
 838                                            (u64)-1, &ins, 1);
 839                 BUG_ON(ret);
 840
 841                 em = alloc_extent_map();
 842                 BUG_ON(!em);
 843                 em->start = start;
 844                 em->orig_start = em->start;
 845                 ram_size = ins.offset;
 846                 em->len = ins.offset;
 847
 848                 em->block_start = ins.objectid;
 849                 em->block_len = ins.offset;
 850                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 851                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 852
 853                 while (1) {
 854                         write_lock(&em_tree->lock);
 855                         ret = add_extent_mapping(em_tree, em);
 856                         write_unlock(&em_tree->lock);
 857                         if (ret != -EEXIST) {
 858                                 free_extent_map(em);
 859                                 break;
 860                         }
 861                         btrfs_drop_extent_cache(inode, start,
 862                                                 start + ram_size - 1, 0);
 863                 }
 864
 865                 cur_alloc_size = ins.offset;
 866                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 867                                                ram_size, cur_alloc_size, 0);
 868                 BUG_ON(ret);
 869
 870                 if (root->root_key.objectid ==
 871                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 872                         ret = btrfs_reloc_clone_csums(inode, start,
 873                                                       cur_alloc_size);
 874                         BUG_ON(ret);
 875                 }
 876
 877                 if (disk_num_bytes < cur_alloc_size)
 878                         break;
 879
 880                 /* we're not doing compressed IO, don't unlock the first
 881                  * page (which the caller expects to stay locked), don't
 882                  * clear any dirty bits and don't set any writeback bits
 883                  *
 884                  * Do set the Private2 bit so we know this page was properly
 885                  * setup for writepage
 886                  */
 887                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 888                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 889                         EXTENT_SET_PRIVATE2;
 890
 891                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 892                                              start, start + ram_size - 1,
 893                                              locked_page, op);
 894                 disk_num_bytes -= cur_alloc_size;
 895                 num_bytes -= cur_alloc_size;
 896                 alloc_hint = ins.objectid + ins.offset;
 897                 start += cur_alloc_size;
 898         }
 899 out:
 900         ret = 0;
 901         btrfs_end_transaction(trans, root);
 902
 903         return ret;
 904 }
 905
 906 /*
 907  * work queue call back to started compression on a file and pages
 908  */
 909 static noinline void async_cow_start(struct btrfs_work *work)
 910 {
 911         struct async_cow *async_cow;
 912         int num_added = 0;
 913         async_cow = container_of(work, struct async_cow, work);
 914
 915         compress_file_range(async_cow->inode, async_cow->locked_page,
 916                             async_cow->start, async_cow->end, async_cow,
 917                             &num_added);
 918         if (num_added == 0)
 919                 async_cow->inode = NULL;
 920 }
 921
 922 /*
 923  * work queue call back to submit previously compressed pages
 924  */
 925 static noinline void async_cow_submit(struct btrfs_work *work)
 926 {
 927         struct async_cow *async_cow;
 928         struct btrfs_root *root;
 929         unsigned long nr_pages;
 930
 931         async_cow = container_of(work, struct async_cow, work);
 932
 933         root = async_cow->root;
 934         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 935                 PAGE_CACHE_SHIFT;
 936
 937         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 938
 939         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 940             5 * 1042 * 1024 &&
 941             waitqueue_active(&root->fs_info->async_submit_wait))
 942                 wake_up(&root->fs_info->async_submit_wait);
 943
 944         if (async_cow->inode)
 945                 submit_compressed_extents(async_cow->inode, async_cow);
 946 }
 947
 948 static noinline void async_cow_free(struct btrfs_work *work)
 949 {
 950         struct async_cow *async_cow;
 951         async_cow = container_of(work, struct async_cow, work);
 952         kfree(async_cow);
 953 }
 954
 955 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 956                                 u64 start, u64 end, int *page_started,
 957                                 unsigned long *nr_written)
 958 {
 959         struct async_cow *async_cow;
 960         struct btrfs_root *root = BTRFS_I(inode)->root;
 961         unsigned long nr_pages;
 962         u64 cur_end;
 963         int limit = 10 * 1024 * 1042;
 964
 965         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 966                          1, 0, NULL, GFP_NOFS);
 967         while (start < end) {
 968                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 969                 BUG_ON(!async_cow);
 970                 async_cow->inode = inode;
 971                 async_cow->root = root;
 972                 async_cow->locked_page = locked_page;
 973                 async_cow->start = start;
 974
 975                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 976                         cur_end = end;
 977                 else
 978                         cur_end = min(end, start + 512 * 1024 - 1);
 979
 980                 async_cow->end = cur_end;
 981                 INIT_LIST_HEAD(&async_cow->extents);
 982
 983                 async_cow->work.func = async_cow_start;
 984                 async_cow->work.ordered_func = async_cow_submit;
 985                 async_cow->work.ordered_free = async_cow_free;
 986                 async_cow->work.flags = 0;
 987
 988                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 989                         PAGE_CACHE_SHIFT;
 990                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 991
 992                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 993                                    &async_cow->work);
 994
 995                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 996                         wait_event(root->fs_info->async_submit_wait,
 997                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 998                             limit));
 999                 }
1000
1001                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1002                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1003                         wait_event(root->fs_info->async_submit_wait,
1004                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1005                            0));
1006                 }
1007
1008                 *nr_written += nr_pages;
1009                 start = cur_end + 1;
1010         }
1011         *page_started = 1;
1012         return 0;
1013 }
1014
1015 static noinline int csum_exist_in_range(struct btrfs_root *root,
1016                                         u64 bytenr, u64 num_bytes)
1017 {
1018         int ret;
1019         struct btrfs_ordered_sum *sums;
1020         LIST_HEAD(list);
1021
1022         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1023                                        bytenr + num_bytes - 1, &list, 0);
1024         if (ret == 0 && list_empty(&list))
1025                 return 0;
1026
1027         while (!list_empty(&list)) {
1028                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1029                 list_del(&sums->list);
1030                 kfree(sums);
1031         }
1032         return 1;
1033 }
1034
1035 /*
1036  * when nowcow writeback call back.  This checks for snapshots or COW copies
1037  * of the extents that exist in the file, and COWs the file as required.
1038  *
1039  * If no cow copies or snapshots exist, we write directly to the existing
1040  * blocks on disk
1041  */
1042 static noinline int run_delalloc_nocow(struct inode *inode,
1043                                        struct page *locked_page,
1044                               u64 start, u64 end, int *page_started, int force,
1045                               unsigned long *nr_written)
1046 {
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         struct btrfs_trans_handle *trans;
1049         struct extent_buffer *leaf;
1050         struct btrfs_path *path;
1051         struct btrfs_file_extent_item *fi;
1052         struct btrfs_key found_key;
1053         u64 cow_start;
1054         u64 cur_offset;
1055         u64 extent_end;
1056         u64 extent_offset;
1057         u64 disk_bytenr;
1058         u64 num_bytes;
1059         int extent_type;
1060         int ret;
1061         int type;
1062         int nocow;
1063         int check_prev = 1;
1064         bool nolock;
1065         u64 ino = btrfs_ino(inode);
1066
1067         path = btrfs_alloc_path();
1068         if (!path)
1069                 return -ENOMEM;
1070
1071         nolock = btrfs_is_free_space_inode(root, inode);
1072
1073         if (nolock)
1074                 trans = btrfs_join_transaction_nolock(root);
1075         else
1076                 trans = btrfs_join_transaction(root);
1077
1078         BUG_ON(IS_ERR(trans));
1079         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1080
1081         cow_start = (u64)-1;
1082         cur_offset = start;
1083         while (1) {
1084                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1085                                                cur_offset, 0);
1086                 BUG_ON(ret < 0);
1087                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1088                         leaf = path->nodes[0];
1089                         btrfs_item_key_to_cpu(leaf, &found_key,
1090                                               path->slots[0] - 1);
1091                         if (found_key.objectid == ino &&
1092                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1093                                 path->slots[0]--;
1094                 }
1095                 check_prev = 0;
1096 next_slot:
1097                 leaf = path->nodes[0];
1098                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1099                         ret = btrfs_next_leaf(root, path);
1100                         if (ret < 0)
1101                                 BUG_ON(1);
1102                         if (ret > 0)
1103                                 break;
1104                         leaf = path->nodes[0];
1105                 }
1106
1107                 nocow = 0;
1108                 disk_bytenr = 0;
1109                 num_bytes = 0;
1110                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1111
1112                 if (found_key.objectid > ino ||
1113                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1114                     found_key.offset > end)
1115                         break;
1116
1117                 if (found_key.offset > cur_offset) {
1118                         extent_end = found_key.offset;
1119                         extent_type = 0;
1120                         goto out_check;
1121                 }
1122
1123                 fi = btrfs_item_ptr(leaf, path->slots[0],
1124                                     struct btrfs_file_extent_item);
1125                 extent_type = btrfs_file_extent_type(leaf, fi);
1126
1127                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1128                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1129                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1130                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1131                         extent_end = found_key.offset +
1132                                 btrfs_file_extent_num_bytes(leaf, fi);
1133                         if (extent_end <= start) {
1134                                 path->slots[0]++;
1135                                 goto next_slot;
1136                         }
1137                         if (disk_bytenr == 0)
1138                                 goto out_check;
1139                         if (btrfs_file_extent_compression(leaf, fi) ||
1140                             btrfs_file_extent_encryption(leaf, fi) ||
1141                             btrfs_file_extent_other_encoding(leaf, fi))
1142                                 goto out_check;
1143                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1144                                 goto out_check;
1145                         if (btrfs_extent_readonly(root, disk_bytenr))
1146                                 goto out_check;
1147                         if (btrfs_cross_ref_exist(trans, root, ino,
1148                                                   found_key.offset -
1149                                                   extent_offset, disk_bytenr))
1150                                 goto out_check;
1151                         disk_bytenr += extent_offset;
1152                         disk_bytenr += cur_offset - found_key.offset;
1153                         num_bytes = min(end + 1, extent_end) - cur_offset;
1154                         /*
1155                          * force cow if csum exists in the range.
1156                          * this ensure that csum for a given extent are
1157                          * either valid or do not exist.
1158                          */
1159                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1160                                 goto out_check;
1161                         nocow = 1;
1162                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1163                         extent_end = found_key.offset +
1164                                 btrfs_file_extent_inline_len(leaf, fi);
1165                         extent_end = ALIGN(extent_end, root->sectorsize);
1166                 } else {
1167                         BUG_ON(1);
1168                 }
1169 out_check:
1170                 if (extent_end <= start) {
1171                         path->slots[0]++;
1172                         goto next_slot;
1173                 }
1174                 if (!nocow) {
1175                         if (cow_start == (u64)-1)
1176                                 cow_start = cur_offset;
1177                         cur_offset = extent_end;
1178                         if (cur_offset > end)
1179                                 break;
1180                         path->slots[0]++;
1181                         goto next_slot;
1182                 }
1183
1184                 btrfs_release_path(path);
1185                 if (cow_start != (u64)-1) {
1186                         ret = cow_file_range(inode, locked_page, cow_start,
1187                                         found_key.offset - 1, page_started,
1188                                         nr_written, 1);
1189                         BUG_ON(ret);
1190                         cow_start = (u64)-1;
1191                 }
1192
1193                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1194                         struct extent_map *em;
1195                         struct extent_map_tree *em_tree;
1196                         em_tree = &BTRFS_I(inode)->extent_tree;
1197                         em = alloc_extent_map();
1198                         BUG_ON(!em);
1199                         em->start = cur_offset;
1200                         em->orig_start = em->start;
1201                         em->len = num_bytes;
1202                         em->block_len = num_bytes;
1203                         em->block_start = disk_bytenr;
1204                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1205                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1206                         while (1) {
1207                                 write_lock(&em_tree->lock);
1208                                 ret = add_extent_mapping(em_tree, em);
1209                                 write_unlock(&em_tree->lock);
1210                                 if (ret != -EEXIST) {
1211                                         free_extent_map(em);
1212                                         break;
1213                                 }
1214                                 btrfs_drop_extent_cache(inode, em->start,
1215                                                 em->start + em->len - 1, 0);
1216                         }
1217                         type = BTRFS_ORDERED_PREALLOC;
1218                 } else {
1219                         type = BTRFS_ORDERED_NOCOW;
1220                 }
1221
1222                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1223                                                num_bytes, num_bytes, type);
1224                 BUG_ON(ret);
1225
1226                 if (root->root_key.objectid ==
1227                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1228                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1229                                                       num_bytes);
1230                         BUG_ON(ret);
1231                 }
1232
1233                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1234                                 cur_offset, cur_offset + num_bytes - 1,
1235                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1236                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1237                                 EXTENT_SET_PRIVATE2);
1238                 cur_offset = extent_end;
1239                 if (cur_offset > end)
1240                         break;
1241         }
1242         btrfs_release_path(path);
1243
1244         if (cur_offset <= end && cow_start == (u64)-1)
1245                 cow_start = cur_offset;
1246         if (cow_start != (u64)-1) {
1247                 ret = cow_file_range(inode, locked_page, cow_start, end,
1248                                      page_started, nr_written, 1);
1249                 BUG_ON(ret);
1250         }
1251
1252         if (nolock) {
1253                 ret = btrfs_end_transaction_nolock(trans, root);
1254                 BUG_ON(ret);
1255         } else {
1256                 ret = btrfs_end_transaction(trans, root);
1257                 BUG_ON(ret);
1258         }
1259         btrfs_free_path(path);
1260         return 0;
1261 }
1262
1263 /*
1264  * extent_io.c call back to do delayed allocation processing
1265  */
1266 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1267                               u64 start, u64 end, int *page_started,
1268                               unsigned long *nr_written)
1269 {
1270         int ret;
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272
1273         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1274                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1275                                          page_started, 1, nr_written);
1276         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 0, nr_written);
1279         else if (!btrfs_test_opt(root, COMPRESS) &&
1280                  !(BTRFS_I(inode)->force_compress) &&
1281                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1282                 ret = cow_file_range(inode, locked_page, start, end,
1283                                       page_started, nr_written, 1);
1284         else
1285                 ret = cow_file_range_async(inode, locked_page, start, end,
1286                                            page_started, nr_written);
1287         return ret;
1288 }
1289
1290 static void btrfs_split_extent_hook(struct inode *inode,
1291                                     struct extent_state *orig, u64 split)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(orig->state & EXTENT_DELALLOC))
1295                 return;
1296
1297         spin_lock(&BTRFS_I(inode)->lock);
1298         BTRFS_I(inode)->outstanding_extents++;
1299         spin_unlock(&BTRFS_I(inode)->lock);
1300 }
1301
1302 /*
1303  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1304  * extents so we can keep track of new extents that are just merged onto old
1305  * extents, such as when we are doing sequential writes, so we can properly
1306  * account for the metadata space we'll need.
1307  */
1308 static void btrfs_merge_extent_hook(struct inode *inode,
1309                                     struct extent_state *new,
1310                                     struct extent_state *other)
1311 {
1312         /* not delalloc, ignore it */
1313         if (!(other->state & EXTENT_DELALLOC))
1314                 return;
1315
1316         spin_lock(&BTRFS_I(inode)->lock);
1317         BTRFS_I(inode)->outstanding_extents--;
1318         spin_unlock(&BTRFS_I(inode)->lock);
1319 }
1320
1321 /*
1322  * extent_io.c set_bit_hook, used to track delayed allocation
1323  * bytes in this file, and to maintain the list of inodes that
1324  * have pending delalloc work to be done.
1325  */
1326 static void btrfs_set_bit_hook(struct inode *inode,
1327                                struct extent_state *state, int *bits)
1328 {
1329
1330         /*
1331          * set_bit and clear bit hooks normally require _irqsave/restore
1332          * but in this case, we are only testing for the DELALLOC
1333          * bit, which is only set or cleared with irqs on
1334          */
1335         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1336                 struct btrfs_root *root = BTRFS_I(inode)->root;
1337                 u64 len = state->end + 1 - state->start;
1338                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1339
1340                 if (*bits & EXTENT_FIRST_DELALLOC) {
1341                         *bits &= ~EXTENT_FIRST_DELALLOC;
1342                 } else {
1343                         spin_lock(&BTRFS_I(inode)->lock);
1344                         BTRFS_I(inode)->outstanding_extents++;
1345                         spin_unlock(&BTRFS_I(inode)->lock);
1346                 }
1347
1348                 spin_lock(&root->fs_info->delalloc_lock);
1349                 BTRFS_I(inode)->delalloc_bytes += len;
1350                 root->fs_info->delalloc_bytes += len;
1351                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1352                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1353                                       &root->fs_info->delalloc_inodes);
1354                 }
1355                 spin_unlock(&root->fs_info->delalloc_lock);
1356         }
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static void btrfs_clear_bit_hook(struct inode *inode,
1363                                  struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC) {
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1378                         spin_lock(&BTRFS_I(inode)->lock);
1379                         BTRFS_I(inode)->outstanding_extents--;
1380                         spin_unlock(&BTRFS_I(inode)->lock);
1381                 }
1382
1383                 if (*bits & EXTENT_DO_ACCOUNTING)
1384                         btrfs_delalloc_release_metadata(inode, len);
1385
1386                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1387                     && do_list)
1388                         btrfs_free_reserved_data_space(inode, len);
1389
1390                 spin_lock(&root->fs_info->delalloc_lock);
1391                 root->fs_info->delalloc_bytes -= len;
1392                 BTRFS_I(inode)->delalloc_bytes -= len;
1393
1394                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1395                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1396                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1397                 }
1398                 spin_unlock(&root->fs_info->delalloc_lock);
1399         }
1400 }
1401
1402 /*
1403  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1404  * we don't create bios that span stripes or chunks
1405  */
1406 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1407                          size_t size, struct bio *bio,
1408                          unsigned long bio_flags)
1409 {
1410         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1411         struct btrfs_mapping_tree *map_tree;
1412         u64 logical = (u64)bio->bi_sector << 9;
1413         u64 length = 0;
1414         u64 map_length;
1415         int ret;
1416
1417         if (bio_flags & EXTENT_BIO_COMPRESSED)
1418                 return 0;
1419
1420         length = bio->bi_size;
1421         map_tree = &root->fs_info->mapping_tree;
1422         map_length = length;
1423         ret = btrfs_map_block(map_tree, READ, logical,
1424                               &map_length, NULL, 0);
1425
1426         if (map_length < length + size)
1427                 return 1;
1428         return ret;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1440                                     struct bio *bio, int mirror_num,
1441                                     unsigned long bio_flags,
1442                                     u64 bio_offset)
1443 {
1444         struct btrfs_root *root = BTRFS_I(inode)->root;
1445         int ret = 0;
1446
1447         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1448         BUG_ON(ret);
1449         return 0;
1450 }
1451
1452 /*
1453  * in order to insert checksums into the metadata in large chunks,
1454  * we wait until bio submission time.   All the pages in the bio are
1455  * checksummed and sums are attached onto the ordered extent record.
1456  *
1457  * At IO completion time the cums attached on the ordered extent record
1458  * are inserted into the btree
1459  */
1460 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1461                           int mirror_num, unsigned long bio_flags,
1462                           u64 bio_offset)
1463 {
1464         struct btrfs_root *root = BTRFS_I(inode)->root;
1465         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1466 }
1467
1468 /*
1469  * extent_io.c submission hook. This does the right thing for csum calculation
1470  * on write, or reading the csums from the tree before a read
1471  */
1472 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1473                           int mirror_num, unsigned long bio_flags,
1474                           u64 bio_offset)
1475 {
1476         struct btrfs_root *root = BTRFS_I(inode)->root;
1477         int ret = 0;
1478         int skip_sum;
1479
1480         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1481
1482         if (btrfs_is_free_space_inode(root, inode))
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1484         else
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1486         BUG_ON(ret);
1487
1488         if (!(rw & REQ_WRITE)) {
1489                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1490                         return btrfs_submit_compressed_read(inode, bio,
1491                                                     mirror_num, bio_flags);
1492                 } else if (!skip_sum) {
1493                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1494                         if (ret)
1495                                 return ret;
1496                 }
1497                 goto mapit;
1498         } else if (!skip_sum) {
1499                 /* csum items have already been cloned */
1500                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1501                         goto mapit;
1502                 /* we're doing a write, do the async checksumming */
1503                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1504                                    inode, rw, bio, mirror_num,
1505                                    bio_flags, bio_offset,
1506                                    __btrfs_submit_bio_start,
1507                                    __btrfs_submit_bio_done);
1508         }
1509
1510 mapit:
1511         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1512 }
1513
1514 /*
1515  * given a list of ordered sums record them in the inode.  This happens
1516  * at IO completion time based on sums calculated at bio submission time.
1517  */
1518 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1519                              struct inode *inode, u64 file_offset,
1520                              struct list_head *list)
1521 {
1522         struct btrfs_ordered_sum *sum;
1523
1524         list_for_each_entry(sum, list, list) {
1525                 btrfs_csum_file_blocks(trans,
1526                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1527         }
1528         return 0;
1529 }
1530
1531 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1532                               struct extent_state **cached_state)
1533 {
1534         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1535                 WARN_ON(1);
1536         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1537                                    cached_state, GFP_NOFS);
1538 }
1539
1540 /* see btrfs_writepage_start_hook for details on why this is required */
1541 struct btrfs_writepage_fixup {
1542         struct page *page;
1543         struct btrfs_work work;
1544 };
1545
1546 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1547 {
1548         struct btrfs_writepage_fixup *fixup;
1549         struct btrfs_ordered_extent *ordered;
1550         struct extent_state *cached_state = NULL;
1551         struct page *page;
1552         struct inode *inode;
1553         u64 page_start;
1554         u64 page_end;
1555
1556         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1557         page = fixup->page;
1558 again:
1559         lock_page(page);
1560         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1561                 ClearPageChecked(page);
1562                 goto out_page;
1563         }
1564
1565         inode = page->mapping->host;
1566         page_start = page_offset(page);
1567         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1568
1569         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1570                          &cached_state, GFP_NOFS);
1571
1572         /* already ordered? We're done */
1573         if (PagePrivate2(page))
1574                 goto out;
1575
1576         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1577         if (ordered) {
1578                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1579                                      page_end, &cached_state, GFP_NOFS);
1580                 unlock_page(page);
1581                 btrfs_start_ordered_extent(inode, ordered, 1);
1582                 goto again;
1583         }
1584
1585         BUG();
1586         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1587         ClearPageChecked(page);
1588 out:
1589         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1590                              &cached_state, GFP_NOFS);
1591 out_page:
1592         unlock_page(page);
1593         page_cache_release(page);
1594         kfree(fixup);
1595 }
1596
1597 /*
1598  * There are a few paths in the higher layers of the kernel that directly
1599  * set the page dirty bit without asking the filesystem if it is a
1600  * good idea.  This causes problems because we want to make sure COW
1601  * properly happens and the data=ordered rules are followed.
1602  *
1603  * In our case any range that doesn't have the ORDERED bit set
1604  * hasn't been properly setup for IO.  We kick off an async process
1605  * to fix it up.  The async helper will wait for ordered extents, set
1606  * the delalloc bit and make it safe to write the page.
1607  */
1608 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1609 {
1610         struct inode *inode = page->mapping->host;
1611         struct btrfs_writepage_fixup *fixup;
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613
1614         /* this page is properly in the ordered list */
1615         if (TestClearPagePrivate2(page))
1616                 return 0;
1617
1618         if (PageChecked(page))
1619                 return -EAGAIN;
1620
1621         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1622         if (!fixup)
1623                 return -EAGAIN;
1624
1625         SetPageChecked(page);
1626         page_cache_get(page);
1627         fixup->work.func = btrfs_writepage_fixup_worker;
1628         fixup->page = page;
1629         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1630         return -EAGAIN;
1631 }
1632
1633 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1634                                        struct inode *inode, u64 file_pos,
1635                                        u64 disk_bytenr, u64 disk_num_bytes,
1636                                        u64 num_bytes, u64 ram_bytes,
1637                                        u8 compression, u8 encryption,
1638                                        u16 other_encoding, int extent_type)
1639 {
1640         struct btrfs_root *root = BTRFS_I(inode)->root;
1641         struct btrfs_file_extent_item *fi;
1642         struct btrfs_path *path;
1643         struct extent_buffer *leaf;
1644         struct btrfs_key ins;
1645         u64 hint;
1646         int ret;
1647
1648         path = btrfs_alloc_path();
1649         if (!path)
1650                 return -ENOMEM;
1651
1652         path->leave_spinning = 1;
1653
1654         /*
1655          * we may be replacing one extent in the tree with another.
1656          * The new extent is pinned in the extent map, and we don't want
1657          * to drop it from the cache until it is completely in the btree.
1658          *
1659          * So, tell btrfs_drop_extents to leave this extent in the cache.
1660          * the caller is expected to unpin it and allow it to be merged
1661          * with the others.
1662          */
1663         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1664                                  &hint, 0);
1665         BUG_ON(ret);
1666
1667         ins.objectid = btrfs_ino(inode);
1668         ins.offset = file_pos;
1669         ins.type = BTRFS_EXTENT_DATA_KEY;
1670         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1671         BUG_ON(ret);
1672         leaf = path->nodes[0];
1673         fi = btrfs_item_ptr(leaf, path->slots[0],
1674                             struct btrfs_file_extent_item);
1675         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1676         btrfs_set_file_extent_type(leaf, fi, extent_type);
1677         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1678         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1679         btrfs_set_file_extent_offset(leaf, fi, 0);
1680         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1681         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1682         btrfs_set_file_extent_compression(leaf, fi, compression);
1683         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1684         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1685
1686         btrfs_unlock_up_safe(path, 1);
1687         btrfs_set_lock_blocking(leaf);
1688
1689         btrfs_mark_buffer_dirty(leaf);
1690
1691         inode_add_bytes(inode, num_bytes);
1692
1693         ins.objectid = disk_bytenr;
1694         ins.offset = disk_num_bytes;
1695         ins.type = BTRFS_EXTENT_ITEM_KEY;
1696         ret = btrfs_alloc_reserved_file_extent(trans, root,
1697                                         root->root_key.objectid,
1698                                         btrfs_ino(inode), file_pos, &ins);
1699         BUG_ON(ret);
1700         btrfs_free_path(path);
1701
1702         return 0;
1703 }
1704
1705 /*
1706  * helper function for btrfs_finish_ordered_io, this
1707  * just reads in some of the csum leaves to prime them into ram
1708  * before we start the transaction.  It limits the amount of btree
1709  * reads required while inside the transaction.
1710  */
1711 /* as ordered data IO finishes, this gets called so we can finish
1712  * an ordered extent if the range of bytes in the file it covers are
1713  * fully written.
1714  */
1715 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1716 {
1717         struct btrfs_root *root = BTRFS_I(inode)->root;
1718         struct btrfs_trans_handle *trans = NULL;
1719         struct btrfs_ordered_extent *ordered_extent = NULL;
1720         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1721         struct extent_state *cached_state = NULL;
1722         int compress_type = 0;
1723         int ret;
1724         bool nolock;
1725
1726         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1727                                              end - start + 1);
1728         if (!ret)
1729                 return 0;
1730         BUG_ON(!ordered_extent);
1731
1732         nolock = btrfs_is_free_space_inode(root, inode);
1733
1734         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1735                 BUG_ON(!list_empty(&ordered_extent->list));
1736                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1737                 if (!ret) {
1738                         if (nolock)
1739                                 trans = btrfs_join_transaction_nolock(root);
1740                         else
1741                                 trans = btrfs_join_transaction(root);
1742                         BUG_ON(IS_ERR(trans));
1743                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1744                         ret = btrfs_update_inode(trans, root, inode);
1745                         BUG_ON(ret);
1746                 }
1747                 goto out;
1748         }
1749
1750         lock_extent_bits(io_tree, ordered_extent->file_offset,
1751                          ordered_extent->file_offset + ordered_extent->len - 1,
1752                          0, &cached_state, GFP_NOFS);
1753
1754         if (nolock)
1755                 trans = btrfs_join_transaction_nolock(root);
1756         else
1757                 trans = btrfs_join_transaction(root);
1758         BUG_ON(IS_ERR(trans));
1759         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1760
1761         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1762                 compress_type = ordered_extent->compress_type;
1763         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1764                 BUG_ON(compress_type);
1765                 ret = btrfs_mark_extent_written(trans, inode,
1766                                                 ordered_extent->file_offset,
1767                                                 ordered_extent->file_offset +
1768                                                 ordered_extent->len);
1769                 BUG_ON(ret);
1770         } else {
1771                 BUG_ON(root == root->fs_info->tree_root);
1772                 ret = insert_reserved_file_extent(trans, inode,
1773                                                 ordered_extent->file_offset,
1774                                                 ordered_extent->start,
1775                                                 ordered_extent->disk_len,
1776                                                 ordered_extent->len,
1777                                                 ordered_extent->len,
1778                                                 compress_type, 0, 0,
1779                                                 BTRFS_FILE_EXTENT_REG);
1780                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1781                                    ordered_extent->file_offset,
1782                                    ordered_extent->len);
1783                 BUG_ON(ret);
1784         }
1785         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1786                              ordered_extent->file_offset +
1787                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1788
1789         add_pending_csums(trans, inode, ordered_extent->file_offset,
1790                           &ordered_extent->list);
1791
1792         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1793         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1794                 ret = btrfs_update_inode(trans, root, inode);
1795                 BUG_ON(ret);
1796         }
1797         ret = 0;
1798 out:
1799         if (root != root->fs_info->tree_root)
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801         if (trans) {
1802                 if (nolock)
1803                         btrfs_end_transaction_nolock(trans, root);
1804                 else
1805                         btrfs_end_transaction(trans, root);
1806         }
1807
1808         /* once for us */
1809         btrfs_put_ordered_extent(ordered_extent);
1810         /* once for the tree */
1811         btrfs_put_ordered_extent(ordered_extent);
1812
1813         return 0;
1814 }
1815
1816 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1817                                 struct extent_state *state, int uptodate)
1818 {
1819         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1820
1821         ClearPagePrivate2(page);
1822         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1823 }
1824
1825 /*
1826  * when reads are done, we need to check csums to verify the data is correct
1827  * if there's a match, we allow the bio to finish.  If not, the code in
1828  * extent_io.c will try to find good copies for us.
1829  */
1830 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1831                                struct extent_state *state)
1832 {
1833         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1834         struct inode *inode = page->mapping->host;
1835         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1836         char *kaddr;
1837         u64 private = ~(u32)0;
1838         int ret;
1839         struct btrfs_root *root = BTRFS_I(inode)->root;
1840         u32 csum = ~(u32)0;
1841
1842         if (PageChecked(page)) {
1843                 ClearPageChecked(page);
1844                 goto good;
1845         }
1846
1847         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1848                 goto good;
1849
1850         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1851             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1852                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1853                                   GFP_NOFS);
1854                 return 0;
1855         }
1856
1857         if (state && state->start == start) {
1858                 private = state->private;
1859                 ret = 0;
1860         } else {
1861                 ret = get_state_private(io_tree, start, &private);
1862         }
1863         kaddr = kmap_atomic(page, KM_USER0);
1864         if (ret)
1865                 goto zeroit;
1866
1867         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1868         btrfs_csum_final(csum, (char *)&csum);
1869         if (csum != private)
1870                 goto zeroit;
1871
1872         kunmap_atomic(kaddr, KM_USER0);
1873 good:
1874         return 0;
1875
1876 zeroit:
1877         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1878                        "private %llu\n",
1879                        (unsigned long long)btrfs_ino(page->mapping->host),
1880                        (unsigned long long)start, csum,
1881                        (unsigned long long)private);
1882         memset(kaddr + offset, 1, end - start + 1);
1883         flush_dcache_page(page);
1884         kunmap_atomic(kaddr, KM_USER0);
1885         if (private == 0)
1886                 return 0;
1887         return -EIO;
1888 }
1889
1890 struct delayed_iput {
1891         struct list_head list;
1892         struct inode *inode;
1893 };
1894
1895 void btrfs_add_delayed_iput(struct inode *inode)
1896 {
1897         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1898         struct delayed_iput *delayed;
1899
1900         if (atomic_add_unless(&inode->i_count, -1, 1))
1901                 return;
1902
1903         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1904         delayed->inode = inode;
1905
1906         spin_lock(&fs_info->delayed_iput_lock);
1907         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1908         spin_unlock(&fs_info->delayed_iput_lock);
1909 }
1910
1911 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1912 {
1913         LIST_HEAD(list);
1914         struct btrfs_fs_info *fs_info = root->fs_info;
1915         struct delayed_iput *delayed;
1916         int empty;
1917
1918         spin_lock(&fs_info->delayed_iput_lock);
1919         empty = list_empty(&fs_info->delayed_iputs);
1920         spin_unlock(&fs_info->delayed_iput_lock);
1921         if (empty)
1922                 return;
1923
1924         down_read(&root->fs_info->cleanup_work_sem);
1925         spin_lock(&fs_info->delayed_iput_lock);
1926         list_splice_init(&fs_info->delayed_iputs, &list);
1927         spin_unlock(&fs_info->delayed_iput_lock);
1928
1929         while (!list_empty(&list)) {
1930                 delayed = list_entry(list.next, struct delayed_iput, list);
1931                 list_del(&delayed->list);
1932                 iput(delayed->inode);
1933                 kfree(delayed);
1934         }
1935         up_read(&root->fs_info->cleanup_work_sem);
1936 }
1937
1938 enum btrfs_orphan_cleanup_state {
1939         ORPHAN_CLEANUP_STARTED  = 1,
1940         ORPHAN_CLEANUP_DONE     = 2,
1941 };
1942
1943 /*
1944  * This is called in transaction commmit time. If there are no orphan
1945  * files in the subvolume, it removes orphan item and frees block_rsv
1946  * structure.
1947  */
1948 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1949                               struct btrfs_root *root)
1950 {
1951         int ret;
1952
1953         if (!list_empty(&root->orphan_list) ||
1954             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1955                 return;
1956
1957         if (root->orphan_item_inserted &&
1958             btrfs_root_refs(&root->root_item) > 0) {
1959                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1960                                             root->root_key.objectid);
1961                 BUG_ON(ret);
1962                 root->orphan_item_inserted = 0;
1963         }
1964
1965         if (root->orphan_block_rsv) {
1966                 WARN_ON(root->orphan_block_rsv->size > 0);
1967                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1968                 root->orphan_block_rsv = NULL;
1969         }
1970 }
1971
1972 /*
1973  * This creates an orphan entry for the given inode in case something goes
1974  * wrong in the middle of an unlink/truncate.
1975  *
1976  * NOTE: caller of this function should reserve 5 units of metadata for
1977  *       this function.
1978  */
1979 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1980 {
1981         struct btrfs_root *root = BTRFS_I(inode)->root;
1982         struct btrfs_block_rsv *block_rsv = NULL;
1983         int reserve = 0;
1984         int insert = 0;
1985         int ret;
1986
1987         if (!root->orphan_block_rsv) {
1988                 block_rsv = btrfs_alloc_block_rsv(root);
1989                 if (!block_rsv)
1990                         return -ENOMEM;
1991         }
1992
1993         spin_lock(&root->orphan_lock);
1994         if (!root->orphan_block_rsv) {
1995                 root->orphan_block_rsv = block_rsv;
1996         } else if (block_rsv) {
1997                 btrfs_free_block_rsv(root, block_rsv);
1998                 block_rsv = NULL;
1999         }
2000
2001         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2002                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2003 #if 0
2004                 /*
2005                  * For proper ENOSPC handling, we should do orphan
2006                  * cleanup when mounting. But this introduces backward
2007                  * compatibility issue.
2008                  */
2009                 if (!xchg(&root->orphan_item_inserted, 1))
2010                         insert = 2;
2011                 else
2012                         insert = 1;
2013 #endif
2014                 insert = 1;
2015         }
2016
2017         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2018                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2019                 reserve = 1;
2020         }
2021         spin_unlock(&root->orphan_lock);
2022
2023         /* grab metadata reservation from transaction handle */
2024         if (reserve) {
2025                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2026                 BUG_ON(ret);
2027         }
2028
2029         /* insert an orphan item to track this unlinked/truncated file */
2030         if (insert >= 1) {
2031                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2032                 BUG_ON(ret);
2033         }
2034
2035         /* insert an orphan item to track subvolume contains orphan files */
2036         if (insert >= 2) {
2037                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2038                                                root->root_key.objectid);
2039                 BUG_ON(ret);
2040         }
2041         return 0;
2042 }
2043
2044 /*
2045  * We have done the truncate/delete so we can go ahead and remove the orphan
2046  * item for this particular inode.
2047  */
2048 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2049 {
2050         struct btrfs_root *root = BTRFS_I(inode)->root;
2051         int delete_item = 0;
2052         int release_rsv = 0;
2053         int ret = 0;
2054
2055         spin_lock(&root->orphan_lock);
2056         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2057                 list_del_init(&BTRFS_I(inode)->i_orphan);
2058                 delete_item = 1;
2059         }
2060
2061         if (BTRFS_I(inode)->orphan_meta_reserved) {
2062                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2063                 release_rsv = 1;
2064         }
2065         spin_unlock(&root->orphan_lock);
2066
2067         if (trans && delete_item) {
2068                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2069                 BUG_ON(ret);
2070         }
2071
2072         if (release_rsv)
2073                 btrfs_orphan_release_metadata(inode);
2074
2075         return 0;
2076 }
2077
2078 /*
2079  * this cleans up any orphans that may be left on the list from the last use
2080  * of this root.
2081  */
2082 int btrfs_orphan_cleanup(struct btrfs_root *root)
2083 {
2084         struct btrfs_path *path;
2085         struct extent_buffer *leaf;
2086         struct btrfs_key key, found_key;
2087         struct btrfs_trans_handle *trans;
2088         struct inode *inode;
2089         u64 last_objectid = 0;
2090         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2091
2092         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2093                 return 0;
2094
2095         path = btrfs_alloc_path();
2096         if (!path) {
2097                 ret = -ENOMEM;
2098                 goto out;
2099         }
2100         path->reada = -1;
2101
2102         key.objectid = BTRFS_ORPHAN_OBJECTID;
2103         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2104         key.offset = (u64)-1;
2105
2106         while (1) {
2107                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2108                 if (ret < 0)
2109                         goto out;
2110
2111                 /*
2112                  * if ret == 0 means we found what we were searching for, which
2113                  * is weird, but possible, so only screw with path if we didn't
2114                  * find the key and see if we have stuff that matches
2115                  */
2116                 if (ret > 0) {
2117                         ret = 0;
2118                         if (path->slots[0] == 0)
2119                                 break;
2120                         path->slots[0]--;
2121                 }
2122
2123                 /* pull out the item */
2124                 leaf = path->nodes[0];
2125                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2126
2127                 /* make sure the item matches what we want */
2128                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2129                         break;
2130                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2131                         break;
2132
2133                 /* release the path since we're done with it */
2134                 btrfs_release_path(path);
2135
2136                 /*
2137                  * this is where we are basically btrfs_lookup, without the
2138                  * crossing root thing.  we store the inode number in the
2139                  * offset of the orphan item.
2140                  */
2141
2142                 if (found_key.offset == last_objectid) {
2143                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2144                                "stopping orphan cleanup\n");
2145                         ret = -EINVAL;
2146                         goto out;
2147                 }
2148
2149                 last_objectid = found_key.offset;
2150
2151                 found_key.objectid = found_key.offset;
2152                 found_key.type = BTRFS_INODE_ITEM_KEY;
2153                 found_key.offset = 0;
2154                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2155                 ret = PTR_RET(inode);
2156                 if (ret && ret != -ESTALE)
2157                         goto out;
2158
2159                 /*
2160                  * Inode is already gone but the orphan item is still there,
2161                  * kill the orphan item.
2162                  */
2163                 if (ret == -ESTALE) {
2164                         trans = btrfs_start_transaction(root, 1);
2165                         if (IS_ERR(trans)) {
2166                                 ret = PTR_ERR(trans);
2167                                 goto out;
2168                         }
2169                         ret = btrfs_del_orphan_item(trans, root,
2170                                                     found_key.objectid);
2171                         BUG_ON(ret);
2172                         btrfs_end_transaction(trans, root);
2173                         continue;
2174                 }
2175
2176                 /*
2177                  * add this inode to the orphan list so btrfs_orphan_del does
2178                  * the proper thing when we hit it
2179                  */
2180                 spin_lock(&root->orphan_lock);
2181                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2182                 spin_unlock(&root->orphan_lock);
2183
2184                 /* if we have links, this was a truncate, lets do that */
2185                 if (inode->i_nlink) {
2186                         if (!S_ISREG(inode->i_mode)) {
2187                                 WARN_ON(1);
2188                                 iput(inode);
2189                                 continue;
2190                         }
2191                         nr_truncate++;
2192                         ret = btrfs_truncate(inode);
2193                 } else {
2194                         nr_unlink++;
2195                 }
2196
2197                 /* this will do delete_inode and everything for us */
2198                 iput(inode);
2199                 if (ret)
2200                         goto out;
2201         }
2202         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2203
2204         if (root->orphan_block_rsv)
2205                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2206                                         (u64)-1);
2207
2208         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2209                 trans = btrfs_join_transaction(root);
2210                 if (!IS_ERR(trans))
2211                         btrfs_end_transaction(trans, root);
2212         }
2213
2214         if (nr_unlink)
2215                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2216         if (nr_truncate)
2217                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2218
2219 out:
2220         if (ret)
2221                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2222         btrfs_free_path(path);
2223         return ret;
2224 }
2225
2226 /*
2227  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2228  * don't find any xattrs, we know there can't be any acls.
2229  *
2230  * slot is the slot the inode is in, objectid is the objectid of the inode
2231  */
2232 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2233                                           int slot, u64 objectid)
2234 {
2235         u32 nritems = btrfs_header_nritems(leaf);
2236         struct btrfs_key found_key;
2237         int scanned = 0;
2238
2239         slot++;
2240         while (slot < nritems) {
2241                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2242
2243                 /* we found a different objectid, there must not be acls */
2244                 if (found_key.objectid != objectid)
2245                         return 0;
2246
2247                 /* we found an xattr, assume we've got an acl */
2248                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2249                         return 1;
2250
2251                 /*
2252                  * we found a key greater than an xattr key, there can't
2253                  * be any acls later on
2254                  */
2255                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2256                         return 0;
2257
2258                 slot++;
2259                 scanned++;
2260
2261                 /*
2262                  * it goes inode, inode backrefs, xattrs, extents,
2263                  * so if there are a ton of hard links to an inode there can
2264                  * be a lot of backrefs.  Don't waste time searching too hard,
2265                  * this is just an optimization
2266                  */
2267                 if (scanned >= 8)
2268                         break;
2269         }
2270         /* we hit the end of the leaf before we found an xattr or
2271          * something larger than an xattr.  We have to assume the inode
2272          * has acls
2273          */
2274         return 1;
2275 }
2276
2277 /*
2278  * read an inode from the btree into the in-memory inode
2279  */
2280 static void btrfs_read_locked_inode(struct inode *inode)
2281 {
2282         struct btrfs_path *path;
2283         struct extent_buffer *leaf;
2284         struct btrfs_inode_item *inode_item;
2285         struct btrfs_timespec *tspec;
2286         struct btrfs_root *root = BTRFS_I(inode)->root;
2287         struct btrfs_key location;
2288         int maybe_acls;
2289         u32 rdev;
2290         int ret;
2291         bool filled = false;
2292
2293         ret = btrfs_fill_inode(inode, &rdev);
2294         if (!ret)
2295                 filled = true;
2296
2297         path = btrfs_alloc_path();
2298         if (!path)
2299                 goto make_bad;
2300
2301         path->leave_spinning = 1;
2302         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2303
2304         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2305         if (ret)
2306                 goto make_bad;
2307
2308         leaf = path->nodes[0];
2309
2310         if (filled)
2311                 goto cache_acl;
2312
2313         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2314                                     struct btrfs_inode_item);
2315         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2316         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2317         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2318         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2319         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2320
2321         tspec = btrfs_inode_atime(inode_item);
2322         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2323         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2324
2325         tspec = btrfs_inode_mtime(inode_item);
2326         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2327         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2328
2329         tspec = btrfs_inode_ctime(inode_item);
2330         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2331         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2332
2333         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2334         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2335         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2336         inode->i_generation = BTRFS_I(inode)->generation;
2337         inode->i_rdev = 0;
2338         rdev = btrfs_inode_rdev(leaf, inode_item);
2339
2340         BTRFS_I(inode)->index_cnt = (u64)-1;
2341         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2342 cache_acl:
2343         /*
2344          * try to precache a NULL acl entry for files that don't have
2345          * any xattrs or acls
2346          */
2347         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2348                                            btrfs_ino(inode));
2349         if (!maybe_acls)
2350                 cache_no_acl(inode);
2351
2352         btrfs_free_path(path);
2353
2354         switch (inode->i_mode & S_IFMT) {
2355         case S_IFREG:
2356                 inode->i_mapping->a_ops = &btrfs_aops;
2357                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2358                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2359                 inode->i_fop = &btrfs_file_operations;
2360                 inode->i_op = &btrfs_file_inode_operations;
2361                 break;
2362         case S_IFDIR:
2363                 inode->i_fop = &btrfs_dir_file_operations;
2364                 if (root == root->fs_info->tree_root)
2365                         inode->i_op = &btrfs_dir_ro_inode_operations;
2366                 else
2367                         inode->i_op = &btrfs_dir_inode_operations;
2368                 break;
2369         case S_IFLNK:
2370                 inode->i_op = &btrfs_symlink_inode_operations;
2371                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2372                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2373                 break;
2374         default:
2375                 inode->i_op = &btrfs_special_inode_operations;
2376                 init_special_inode(inode, inode->i_mode, rdev);
2377                 break;
2378         }
2379
2380         btrfs_update_iflags(inode);
2381         return;
2382
2383 make_bad:
2384         btrfs_free_path(path);
2385         make_bad_inode(inode);
2386 }
2387
2388 /*
2389  * given a leaf and an inode, copy the inode fields into the leaf
2390  */
2391 static void fill_inode_item(struct btrfs_trans_handle *trans,
2392                             struct extent_buffer *leaf,
2393                             struct btrfs_inode_item *item,
2394                             struct inode *inode)
2395 {
2396         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2397         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2398         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2399         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2400         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2401
2402         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2403                                inode->i_atime.tv_sec);
2404         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2405                                 inode->i_atime.tv_nsec);
2406
2407         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2408                                inode->i_mtime.tv_sec);
2409         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2410                                 inode->i_mtime.tv_nsec);
2411
2412         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2413                                inode->i_ctime.tv_sec);
2414         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2415                                 inode->i_ctime.tv_nsec);
2416
2417         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2418         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2419         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2420         btrfs_set_inode_transid(leaf, item, trans->transid);
2421         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2422         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2423         btrfs_set_inode_block_group(leaf, item, 0);
2424 }
2425
2426 /*
2427  * copy everything in the in-memory inode into the btree.
2428  */
2429 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2430                                 struct btrfs_root *root, struct inode *inode)
2431 {
2432         struct btrfs_inode_item *inode_item;
2433         struct btrfs_path *path;
2434         struct extent_buffer *leaf;
2435         int ret;
2436
2437         /*
2438          * If the inode is a free space inode, we can deadlock during commit
2439          * if we put it into the delayed code.
2440          *
2441          * The data relocation inode should also be directly updated
2442          * without delay
2443          */
2444         if (!btrfs_is_free_space_inode(root, inode)
2445             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2446                 ret = btrfs_delayed_update_inode(trans, root, inode);
2447                 if (!ret)
2448                         btrfs_set_inode_last_trans(trans, inode);
2449                 return ret;
2450         }
2451
2452         path = btrfs_alloc_path();
2453         if (!path)
2454                 return -ENOMEM;
2455
2456         path->leave_spinning = 1;
2457         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2458                                  1);
2459         if (ret) {
2460                 if (ret > 0)
2461                         ret = -ENOENT;
2462                 goto failed;
2463         }
2464
2465         btrfs_unlock_up_safe(path, 1);
2466         leaf = path->nodes[0];
2467         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2468                                     struct btrfs_inode_item);
2469
2470         fill_inode_item(trans, leaf, inode_item, inode);
2471         btrfs_mark_buffer_dirty(leaf);
2472         btrfs_set_inode_last_trans(trans, inode);
2473         ret = 0;
2474 failed:
2475         btrfs_free_path(path);
2476         return ret;
2477 }
2478
2479 /*
2480  * unlink helper that gets used here in inode.c and in the tree logging
2481  * recovery code.  It remove a link in a directory with a given name, and
2482  * also drops the back refs in the inode to the directory
2483  */
2484 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2485                                 struct btrfs_root *root,
2486                                 struct inode *dir, struct inode *inode,
2487                                 const char *name, int name_len)
2488 {
2489         struct btrfs_path *path;
2490         int ret = 0;
2491         struct extent_buffer *leaf;
2492         struct btrfs_dir_item *di;
2493         struct btrfs_key key;
2494         u64 index;
2495         u64 ino = btrfs_ino(inode);
2496         u64 dir_ino = btrfs_ino(dir);
2497
2498         path = btrfs_alloc_path();
2499         if (!path) {
2500                 ret = -ENOMEM;
2501                 goto out;
2502         }
2503
2504         path->leave_spinning = 1;
2505         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2506                                     name, name_len, -1);
2507         if (IS_ERR(di)) {
2508                 ret = PTR_ERR(di);
2509                 goto err;
2510         }
2511         if (!di) {
2512                 ret = -ENOENT;
2513                 goto err;
2514         }
2515         leaf = path->nodes[0];
2516         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2517         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2518         if (ret)
2519                 goto err;
2520         btrfs_release_path(path);
2521
2522         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2523                                   dir_ino, &index);
2524         if (ret) {
2525                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2526                        "inode %llu parent %llu\n", name_len, name,
2527                        (unsigned long long)ino, (unsigned long long)dir_ino);
2528                 goto err;
2529         }
2530
2531         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2532         if (ret)
2533                 goto err;
2534
2535         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2536                                          inode, dir_ino);
2537         BUG_ON(ret != 0 && ret != -ENOENT);
2538
2539         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2540                                            dir, index);
2541         if (ret == -ENOENT)
2542                 ret = 0;
2543 err:
2544         btrfs_free_path(path);
2545         if (ret)
2546                 goto out;
2547
2548         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2549         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2550         btrfs_update_inode(trans, root, dir);
2551 out:
2552         return ret;
2553 }
2554
2555 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2556                        struct btrfs_root *root,
2557                        struct inode *dir, struct inode *inode,
2558                        const char *name, int name_len)
2559 {
2560         int ret;
2561         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2562         if (!ret) {
2563                 btrfs_drop_nlink(inode);
2564                 ret = btrfs_update_inode(trans, root, inode);
2565         }
2566         return ret;
2567 }
2568
2569
2570 /* helper to check if there is any shared block in the path */
2571 static int check_path_shared(struct btrfs_root *root,
2572                              struct btrfs_path *path)
2573 {
2574         struct extent_buffer *eb;
2575         int level;
2576         u64 refs = 1;
2577
2578         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2579                 int ret;
2580
2581                 if (!path->nodes[level])
2582                         break;
2583                 eb = path->nodes[level];
2584                 if (!btrfs_block_can_be_shared(root, eb))
2585                         continue;
2586                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2587                                                &refs, NULL);
2588                 if (refs > 1)
2589                         return 1;
2590         }
2591         return 0;
2592 }
2593
2594 /*
2595  * helper to start transaction for unlink and rmdir.
2596  *
2597  * unlink and rmdir are special in btrfs, they do not always free space.
2598  * so in enospc case, we should make sure they will free space before
2599  * allowing them to use the global metadata reservation.
2600  */
2601 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2602                                                        struct dentry *dentry)
2603 {
2604         struct btrfs_trans_handle *trans;
2605         struct btrfs_root *root = BTRFS_I(dir)->root;
2606         struct btrfs_path *path;
2607         struct btrfs_inode_ref *ref;
2608         struct btrfs_dir_item *di;
2609         struct inode *inode = dentry->d_inode;
2610         u64 index;
2611         int check_link = 1;
2612         int err = -ENOSPC;
2613         int ret;
2614         u64 ino = btrfs_ino(inode);
2615         u64 dir_ino = btrfs_ino(dir);
2616
2617         /*
2618          * 1 for the possible orphan item
2619          * 1 for the dir item
2620          * 1 for the dir index
2621          * 1 for the inode ref
2622          * 1 for the inode ref in the tree log
2623          * 2 for the dir entries in the log
2624          * 1 for the inode
2625          */
2626         trans = btrfs_start_transaction(root, 8);
2627         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2628                 return trans;
2629
2630         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2631                 return ERR_PTR(-ENOSPC);
2632
2633         /* check if there is someone else holds reference */
2634         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2635                 return ERR_PTR(-ENOSPC);
2636
2637         if (atomic_read(&inode->i_count) > 2)
2638                 return ERR_PTR(-ENOSPC);
2639
2640         if (xchg(&root->fs_info->enospc_unlink, 1))
2641                 return ERR_PTR(-ENOSPC);
2642
2643         path = btrfs_alloc_path();
2644         if (!path) {
2645                 root->fs_info->enospc_unlink = 0;
2646                 return ERR_PTR(-ENOMEM);
2647         }
2648
2649         /* 1 for the orphan item */
2650         trans = btrfs_start_transaction(root, 1);
2651         if (IS_ERR(trans)) {
2652                 btrfs_free_path(path);
2653                 root->fs_info->enospc_unlink = 0;
2654                 return trans;
2655         }
2656
2657         path->skip_locking = 1;
2658         path->search_commit_root = 1;
2659
2660         ret = btrfs_lookup_inode(trans, root, path,
2661                                 &BTRFS_I(dir)->location, 0);
2662         if (ret < 0) {
2663                 err = ret;
2664                 goto out;
2665         }
2666         if (ret == 0) {
2667                 if (check_path_shared(root, path))
2668                         goto out;
2669         } else {
2670                 check_link = 0;
2671         }
2672         btrfs_release_path(path);
2673
2674         ret = btrfs_lookup_inode(trans, root, path,
2675                                 &BTRFS_I(inode)->location, 0);
2676         if (ret < 0) {
2677                 err = ret;
2678                 goto out;
2679         }
2680         if (ret == 0) {
2681                 if (check_path_shared(root, path))
2682                         goto out;
2683         } else {
2684                 check_link = 0;
2685         }
2686         btrfs_release_path(path);
2687
2688         if (ret == 0 && S_ISREG(inode->i_mode)) {
2689                 ret = btrfs_lookup_file_extent(trans, root, path,
2690                                                ino, (u64)-1, 0);
2691                 if (ret < 0) {
2692                         err = ret;
2693                         goto out;
2694                 }
2695                 BUG_ON(ret == 0);
2696                 if (check_path_shared(root, path))
2697                         goto out;
2698                 btrfs_release_path(path);
2699         }
2700
2701         if (!check_link) {
2702                 err = 0;
2703                 goto out;
2704         }
2705
2706         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2707                                 dentry->d_name.name, dentry->d_name.len, 0);
2708         if (IS_ERR(di)) {
2709                 err = PTR_ERR(di);
2710                 goto out;
2711         }
2712         if (di) {
2713                 if (check_path_shared(root, path))
2714                         goto out;
2715         } else {
2716                 err = 0;
2717                 goto out;
2718         }
2719         btrfs_release_path(path);
2720
2721         ref = btrfs_lookup_inode_ref(trans, root, path,
2722                                 dentry->d_name.name, dentry->d_name.len,
2723                                 ino, dir_ino, 0);
2724         if (IS_ERR(ref)) {
2725                 err = PTR_ERR(ref);
2726                 goto out;
2727         }
2728         BUG_ON(!ref);
2729         if (check_path_shared(root, path))
2730                 goto out;
2731         index = btrfs_inode_ref_index(path->nodes[0], ref);
2732         btrfs_release_path(path);
2733
2734         /*
2735          * This is a commit root search, if we can lookup inode item and other
2736          * relative items in the commit root, it means the transaction of
2737          * dir/file creation has been committed, and the dir index item that we
2738          * delay to insert has also been inserted into the commit root. So
2739          * we needn't worry about the delayed insertion of the dir index item
2740          * here.
2741          */
2742         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2743                                 dentry->d_name.name, dentry->d_name.len, 0);
2744         if (IS_ERR(di)) {
2745                 err = PTR_ERR(di);
2746                 goto out;
2747         }
2748         BUG_ON(ret == -ENOENT);
2749         if (check_path_shared(root, path))
2750                 goto out;
2751
2752         err = 0;
2753 out:
2754         btrfs_free_path(path);
2755         /* Migrate the orphan reservation over */
2756         if (!err)
2757                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2758                                 &root->fs_info->global_block_rsv,
2759                                 trans->bytes_reserved);
2760
2761         if (err) {
2762                 btrfs_end_transaction(trans, root);
2763                 root->fs_info->enospc_unlink = 0;
2764                 return ERR_PTR(err);
2765         }
2766
2767         trans->block_rsv = &root->fs_info->global_block_rsv;
2768         return trans;
2769 }
2770
2771 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2772                                struct btrfs_root *root)
2773 {
2774         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2775                 btrfs_block_rsv_release(root, trans->block_rsv,
2776                                         trans->bytes_reserved);
2777                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2778                 BUG_ON(!root->fs_info->enospc_unlink);
2779                 root->fs_info->enospc_unlink = 0;
2780         }
2781         btrfs_end_transaction_throttle(trans, root);
2782 }
2783
2784 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2785 {
2786         struct btrfs_root *root = BTRFS_I(dir)->root;
2787         struct btrfs_trans_handle *trans;
2788         struct inode *inode = dentry->d_inode;
2789         int ret;
2790         unsigned long nr = 0;
2791
2792         trans = __unlink_start_trans(dir, dentry);
2793         if (IS_ERR(trans))
2794                 return PTR_ERR(trans);
2795
2796         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2797
2798         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2799                                  dentry->d_name.name, dentry->d_name.len);
2800         if (ret)
2801                 goto out;
2802
2803         if (inode->i_nlink == 0) {
2804                 ret = btrfs_orphan_add(trans, inode);
2805                 if (ret)
2806                         goto out;
2807         }
2808
2809 out:
2810         nr = trans->blocks_used;
2811         __unlink_end_trans(trans, root);
2812         btrfs_btree_balance_dirty(root, nr);
2813         return ret;
2814 }
2815
2816 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2817                         struct btrfs_root *root,
2818                         struct inode *dir, u64 objectid,
2819                         const char *name, int name_len)
2820 {
2821         struct btrfs_path *path;
2822         struct extent_buffer *leaf;
2823         struct btrfs_dir_item *di;
2824         struct btrfs_key key;
2825         u64 index;
2826         int ret;
2827         u64 dir_ino = btrfs_ino(dir);
2828
2829         path = btrfs_alloc_path();
2830         if (!path)
2831                 return -ENOMEM;
2832
2833         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2834                                    name, name_len, -1);
2835         BUG_ON(IS_ERR_OR_NULL(di));
2836
2837         leaf = path->nodes[0];
2838         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2839         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2840         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2841         BUG_ON(ret);
2842         btrfs_release_path(path);
2843
2844         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2845                                  objectid, root->root_key.objectid,
2846                                  dir_ino, &index, name, name_len);
2847         if (ret < 0) {
2848                 BUG_ON(ret != -ENOENT);
2849                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2850                                                  name, name_len);
2851                 BUG_ON(IS_ERR_OR_NULL(di));
2852
2853                 leaf = path->nodes[0];
2854                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2855                 btrfs_release_path(path);
2856                 index = key.offset;
2857         }
2858         btrfs_release_path(path);
2859
2860         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2861         BUG_ON(ret);
2862
2863         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2864         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2865         ret = btrfs_update_inode(trans, root, dir);
2866         BUG_ON(ret);
2867
2868         btrfs_free_path(path);
2869         return 0;
2870 }
2871
2872 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2873 {
2874         struct inode *inode = dentry->d_inode;
2875         int err = 0;
2876         struct btrfs_root *root = BTRFS_I(dir)->root;
2877         struct btrfs_trans_handle *trans;
2878         unsigned long nr = 0;
2879
2880         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2881             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2882                 return -ENOTEMPTY;
2883
2884         trans = __unlink_start_trans(dir, dentry);
2885         if (IS_ERR(trans))
2886                 return PTR_ERR(trans);
2887
2888         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2889                 err = btrfs_unlink_subvol(trans, root, dir,
2890                                           BTRFS_I(inode)->location.objectid,
2891                                           dentry->d_name.name,
2892                                           dentry->d_name.len);
2893                 goto out;
2894         }
2895
2896         err = btrfs_orphan_add(trans, inode);
2897         if (err)
2898                 goto out;
2899
2900         /* now the directory is empty */
2901         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2902                                  dentry->d_name.name, dentry->d_name.len);
2903         if (!err)
2904                 btrfs_i_size_write(inode, 0);
2905 out:
2906         nr = trans->blocks_used;
2907         __unlink_end_trans(trans, root);
2908         btrfs_btree_balance_dirty(root, nr);
2909
2910         return err;
2911 }
2912
2913 /*
2914  * this can truncate away extent items, csum items and directory items.
2915  * It starts at a high offset and removes keys until it can't find
2916  * any higher than new_size
2917  *
2918  * csum items that cross the new i_size are truncated to the new size
2919  * as well.
2920  *
2921  * min_type is the minimum key type to truncate down to.  If set to 0, this
2922  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2923  */
2924 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2925                                struct btrfs_root *root,
2926                                struct inode *inode,
2927                                u64 new_size, u32 min_type)
2928 {
2929         struct btrfs_path *path;
2930         struct extent_buffer *leaf;
2931         struct btrfs_file_extent_item *fi;
2932         struct btrfs_key key;
2933         struct btrfs_key found_key;
2934         u64 extent_start = 0;
2935         u64 extent_num_bytes = 0;
2936         u64 extent_offset = 0;
2937         u64 item_end = 0;
2938         u64 mask = root->sectorsize - 1;
2939         u32 found_type = (u8)-1;
2940         int found_extent;
2941         int del_item;
2942         int pending_del_nr = 0;
2943         int pending_del_slot = 0;
2944         int extent_type = -1;
2945         int encoding;
2946         int ret;
2947         int err = 0;
2948         u64 ino = btrfs_ino(inode);
2949
2950         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
2951
2952         path = btrfs_alloc_path();
2953         if (!path)
2954                 return -ENOMEM;
2955         path->reada = -1;
2956
2957         if (root->ref_cows || root == root->fs_info->tree_root)
2958                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
2959
2960         /*
2961          * This function is also used to drop the items in the log tree before
2962          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
2963          * it is used to drop the loged items. So we shouldn't kill the delayed
2964          * items.
2965          */
2966         if (min_type == 0 && root == BTRFS_I(inode)->root)
2967                 btrfs_kill_delayed_inode_items(inode);
2968
2969         key.objectid = ino;
2970         key.offset = (u64)-1;
2971         key.type = (u8)-1;
2972
2973 search_again:
2974         path->leave_spinning = 1;
2975         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2976         if (ret < 0) {
2977                 err = ret;
2978                 goto out;
2979         }
2980
2981         if (ret > 0) {
2982                 /* there are no items in the tree for us to truncate, we're
2983                  * done
2984                  */
2985                 if (path->slots[0] == 0)
2986                         goto out;
2987                 path->slots[0]--;
2988         }
2989
2990         while (1) {
2991                 fi = NULL;
2992                 leaf = path->nodes[0];
2993                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2994                 found_type = btrfs_key_type(&found_key);
2995                 encoding = 0;
2996
2997                 if (found_key.objectid != ino)
2998                         break;
2999
3000                 if (found_type < min_type)
3001                         break;
3002
3003                 item_end = found_key.offset;
3004                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3005                         fi = btrfs_item_ptr(leaf, path->slots[0],
3006                                             struct btrfs_file_extent_item);
3007                         extent_type = btrfs_file_extent_type(leaf, fi);
3008                         encoding = btrfs_file_extent_compression(leaf, fi);
3009                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3010                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3011
3012                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3013                                 item_end +=
3014                                     btrfs_file_extent_num_bytes(leaf, fi);
3015                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3016                                 item_end += btrfs_file_extent_inline_len(leaf,
3017                                                                          fi);
3018                         }
3019                         item_end--;
3020                 }
3021                 if (found_type > min_type) {
3022                         del_item = 1;
3023                 } else {
3024                         if (item_end < new_size)
3025                                 break;
3026                         if (found_key.offset >= new_size)
3027                                 del_item = 1;
3028                         else
3029                                 del_item = 0;
3030                 }
3031                 found_extent = 0;
3032                 /* FIXME, shrink the extent if the ref count is only 1 */
3033                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3034                         goto delete;
3035
3036                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3037                         u64 num_dec;
3038                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3039                         if (!del_item && !encoding) {
3040                                 u64 orig_num_bytes =
3041                                         btrfs_file_extent_num_bytes(leaf, fi);
3042                                 extent_num_bytes = new_size -
3043                                         found_key.offset + root->sectorsize - 1;
3044                                 extent_num_bytes = extent_num_bytes &
3045                                         ~((u64)root->sectorsize - 1);
3046                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3047                                                          extent_num_bytes);
3048                                 num_dec = (orig_num_bytes -
3049                                            extent_num_bytes);
3050                                 if (root->ref_cows && extent_start != 0)
3051                                         inode_sub_bytes(inode, num_dec);
3052                                 btrfs_mark_buffer_dirty(leaf);
3053                         } else {
3054                                 extent_num_bytes =
3055                                         btrfs_file_extent_disk_num_bytes(leaf,
3056                                                                          fi);
3057                                 extent_offset = found_key.offset -
3058                                         btrfs_file_extent_offset(leaf, fi);
3059
3060                                 /* FIXME blocksize != 4096 */
3061                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3062                                 if (extent_start != 0) {
3063                                         found_extent = 1;
3064                                         if (root->ref_cows)
3065                                                 inode_sub_bytes(inode, num_dec);
3066                                 }
3067                         }
3068                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3069                         /*
3070                          * we can't truncate inline items that have had
3071                          * special encodings
3072                          */
3073                         if (!del_item &&
3074                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3075                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3076                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3077                                 u32 size = new_size - found_key.offset;
3078
3079                                 if (root->ref_cows) {
3080                                         inode_sub_bytes(inode, item_end + 1 -
3081                                                         new_size);
3082                                 }
3083                                 size =
3084                                     btrfs_file_extent_calc_inline_size(size);
3085                                 ret = btrfs_truncate_item(trans, root, path,
3086                                                           size, 1);
3087                         } else if (root->ref_cows) {
3088                                 inode_sub_bytes(inode, item_end + 1 -
3089                                                 found_key.offset);
3090                         }
3091                 }
3092 delete:
3093                 if (del_item) {
3094                         if (!pending_del_nr) {
3095                                 /* no pending yet, add ourselves */
3096                                 pending_del_slot = path->slots[0];
3097                                 pending_del_nr = 1;
3098                         } else if (pending_del_nr &&
3099                                    path->slots[0] + 1 == pending_del_slot) {
3100                                 /* hop on the pending chunk */
3101                                 pending_del_nr++;
3102                                 pending_del_slot = path->slots[0];
3103                         } else {
3104                                 BUG();
3105                         }
3106                 } else {
3107                         break;
3108                 }
3109                 if (found_extent && (root->ref_cows ||
3110                                      root == root->fs_info->tree_root)) {
3111                         btrfs_set_path_blocking(path);
3112                         ret = btrfs_free_extent(trans, root, extent_start,
3113                                                 extent_num_bytes, 0,
3114                                                 btrfs_header_owner(leaf),
3115                                                 ino, extent_offset);
3116                         BUG_ON(ret);
3117                 }
3118
3119                 if (found_type == BTRFS_INODE_ITEM_KEY)
3120                         break;
3121
3122                 if (path->slots[0] == 0 ||
3123                     path->slots[0] != pending_del_slot) {
3124                         if (root->ref_cows &&
3125                             BTRFS_I(inode)->location.objectid !=
3126                                                 BTRFS_FREE_INO_OBJECTID) {
3127                                 err = -EAGAIN;
3128                                 goto out;
3129                         }
3130                         if (pending_del_nr) {
3131                                 ret = btrfs_del_items(trans, root, path,
3132                                                 pending_del_slot,
3133                                                 pending_del_nr);
3134                                 BUG_ON(ret);
3135                                 pending_del_nr = 0;
3136                         }
3137                         btrfs_release_path(path);
3138                         goto search_again;
3139                 } else {
3140                         path->slots[0]--;
3141                 }
3142         }
3143 out:
3144         if (pending_del_nr) {
3145                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3146                                       pending_del_nr);
3147                 BUG_ON(ret);
3148         }
3149         btrfs_free_path(path);
3150         return err;
3151 }
3152
3153 /*
3154  * taken from block_truncate_page, but does cow as it zeros out
3155  * any bytes left in the last page in the file.
3156  */
3157 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3158 {
3159         struct inode *inode = mapping->host;
3160         struct btrfs_root *root = BTRFS_I(inode)->root;
3161         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3162         struct btrfs_ordered_extent *ordered;
3163         struct extent_state *cached_state = NULL;
3164         char *kaddr;
3165         u32 blocksize = root->sectorsize;
3166         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3167         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3168         struct page *page;
3169         gfp_t mask = btrfs_alloc_write_mask(mapping);
3170         int ret = 0;
3171         u64 page_start;
3172         u64 page_end;
3173
3174         if ((offset & (blocksize - 1)) == 0)
3175                 goto out;
3176         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3177         if (ret)
3178                 goto out;
3179
3180         ret = -ENOMEM;
3181 again:
3182         page = find_or_create_page(mapping, index, mask);
3183         if (!page) {
3184                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3185                 goto out;
3186         }
3187
3188         page_start = page_offset(page);
3189         page_end = page_start + PAGE_CACHE_SIZE - 1;
3190
3191         if (!PageUptodate(page)) {
3192                 ret = btrfs_readpage(NULL, page);
3193                 lock_page(page);
3194                 if (page->mapping != mapping) {
3195                         unlock_page(page);
3196                         page_cache_release(page);
3197                         goto again;
3198                 }
3199                 if (!PageUptodate(page)) {
3200                         ret = -EIO;
3201                         goto out_unlock;
3202                 }
3203         }
3204         wait_on_page_writeback(page);
3205
3206         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3207                          GFP_NOFS);
3208         set_page_extent_mapped(page);
3209
3210         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3211         if (ordered) {
3212                 unlock_extent_cached(io_tree, page_start, page_end,
3213                                      &cached_state, GFP_NOFS);
3214                 unlock_page(page);
3215                 page_cache_release(page);
3216                 btrfs_start_ordered_extent(inode, ordered, 1);
3217                 btrfs_put_ordered_extent(ordered);
3218                 goto again;
3219         }
3220
3221         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3222                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3223                           0, 0, &cached_state, GFP_NOFS);
3224
3225         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3226                                         &cached_state);
3227         if (ret) {
3228                 unlock_extent_cached(io_tree, page_start, page_end,
3229                                      &cached_state, GFP_NOFS);
3230                 goto out_unlock;
3231         }
3232
3233         ret = 0;
3234         if (offset != PAGE_CACHE_SIZE) {
3235                 kaddr = kmap(page);
3236                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3237                 flush_dcache_page(page);
3238                 kunmap(page);
3239         }
3240         ClearPageChecked(page);
3241         set_page_dirty(page);
3242         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3243                              GFP_NOFS);
3244
3245 out_unlock:
3246         if (ret)
3247                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3248         unlock_page(page);
3249         page_cache_release(page);
3250 out:
3251         return ret;
3252 }
3253
3254 /*
3255  * This function puts in dummy file extents for the area we're creating a hole
3256  * for.  So if we are truncating this file to a larger size we need to insert
3257  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3258  * the range between oldsize and size
3259  */
3260 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3261 {
3262         struct btrfs_trans_handle *trans;
3263         struct btrfs_root *root = BTRFS_I(inode)->root;
3264         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3265         struct extent_map *em = NULL;
3266         struct extent_state *cached_state = NULL;
3267         u64 mask = root->sectorsize - 1;
3268         u64 hole_start = (oldsize + mask) & ~mask;
3269         u64 block_end = (size + mask) & ~mask;
3270         u64 last_byte;
3271         u64 cur_offset;
3272         u64 hole_size;
3273         int err = 0;
3274
3275         if (size <= hole_start)
3276                 return 0;
3277
3278         while (1) {
3279                 struct btrfs_ordered_extent *ordered;
3280                 btrfs_wait_ordered_range(inode, hole_start,
3281                                          block_end - hole_start);
3282                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3283                                  &cached_state, GFP_NOFS);
3284                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3285                 if (!ordered)
3286                         break;
3287                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3288                                      &cached_state, GFP_NOFS);
3289                 btrfs_put_ordered_extent(ordered);
3290         }
3291
3292         cur_offset = hole_start;
3293         while (1) {
3294                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3295                                 block_end - cur_offset, 0);
3296                 BUG_ON(IS_ERR_OR_NULL(em));
3297                 last_byte = min(extent_map_end(em), block_end);
3298                 last_byte = (last_byte + mask) & ~mask;
3299                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3300                         u64 hint_byte = 0;
3301                         hole_size = last_byte - cur_offset;
3302
3303                         trans = btrfs_start_transaction(root, 2);
3304                         if (IS_ERR(trans)) {
3305                                 err = PTR_ERR(trans);
3306                                 break;
3307                         }
3308
3309                         err = btrfs_drop_extents(trans, inode, cur_offset,
3310                                                  cur_offset + hole_size,
3311                                                  &hint_byte, 1);
3312                         if (err) {
3313                                 btrfs_end_transaction(trans, root);
3314                                 break;
3315                         }
3316
3317                         err = btrfs_insert_file_extent(trans, root,
3318                                         btrfs_ino(inode), cur_offset, 0,
3319                                         0, hole_size, 0, hole_size,
3320                                         0, 0, 0);
3321                         if (err) {
3322                                 btrfs_end_transaction(trans, root);
3323                                 break;
3324                         }
3325
3326                         btrfs_drop_extent_cache(inode, hole_start,
3327                                         last_byte - 1, 0);
3328
3329                         btrfs_end_transaction(trans, root);
3330                 }
3331                 free_extent_map(em);
3332                 em = NULL;
3333                 cur_offset = last_byte;
3334                 if (cur_offset >= block_end)
3335                         break;
3336         }
3337
3338         free_extent_map(em);
3339         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3340                              GFP_NOFS);
3341         return err;
3342 }
3343
3344 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3345 {
3346         loff_t oldsize = i_size_read(inode);
3347         int ret;
3348
3349         if (newsize == oldsize)
3350                 return 0;
3351
3352         if (newsize > oldsize) {
3353                 i_size_write(inode, newsize);
3354                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3355                 truncate_pagecache(inode, oldsize, newsize);
3356                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3357                 if (ret) {
3358                         btrfs_setsize(inode, oldsize);
3359                         return ret;
3360                 }
3361
3362                 mark_inode_dirty(inode);
3363         } else {
3364
3365                 /*
3366                  * We're truncating a file that used to have good data down to
3367                  * zero. Make sure it gets into the ordered flush list so that
3368                  * any new writes get down to disk quickly.
3369                  */
3370                 if (newsize == 0)
3371                         BTRFS_I(inode)->ordered_data_close = 1;
3372
3373                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3374                 truncate_setsize(inode, newsize);
3375                 ret = btrfs_truncate(inode);
3376         }
3377
3378         return ret;
3379 }
3380
3381 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3382 {
3383         struct inode *inode = dentry->d_inode;
3384         struct btrfs_root *root = BTRFS_I(inode)->root;
3385         int err;
3386
3387         if (btrfs_root_readonly(root))
3388                 return -EROFS;
3389
3390         err = inode_change_ok(inode, attr);
3391         if (err)
3392                 return err;
3393
3394         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3395                 err = btrfs_setsize(inode, attr->ia_size);
3396                 if (err)
3397                         return err;
3398         }
3399
3400         if (attr->ia_valid) {
3401                 setattr_copy(inode, attr);
3402                 mark_inode_dirty(inode);
3403
3404                 if (attr->ia_valid & ATTR_MODE)
3405                         err = btrfs_acl_chmod(inode);
3406         }
3407
3408         return err;
3409 }
3410
3411 void btrfs_evict_inode(struct inode *inode)
3412 {
3413         struct btrfs_trans_handle *trans;
3414         struct btrfs_root *root = BTRFS_I(inode)->root;
3415         struct btrfs_block_rsv *rsv, *global_rsv;
3416         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3417         unsigned long nr;
3418         int ret;
3419
3420         trace_btrfs_inode_evict(inode);
3421
3422         truncate_inode_pages(&inode->i_data, 0);
3423         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3424                                btrfs_is_free_space_inode(root, inode)))
3425                 goto no_delete;
3426
3427         if (is_bad_inode(inode)) {
3428                 btrfs_orphan_del(NULL, inode);
3429                 goto no_delete;
3430         }
3431         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3432         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3433
3434         if (root->fs_info->log_root_recovering) {
3435                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3436                 goto no_delete;
3437         }
3438
3439         if (inode->i_nlink > 0) {
3440                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3441                 goto no_delete;
3442         }
3443
3444         rsv = btrfs_alloc_block_rsv(root);
3445         if (!rsv) {
3446                 btrfs_orphan_del(NULL, inode);
3447                 goto no_delete;
3448         }
3449         rsv->size = min_size;
3450         global_rsv = &root->fs_info->global_block_rsv;
3451
3452         btrfs_i_size_write(inode, 0);
3453
3454         /*
3455          * This is a bit simpler than btrfs_truncate since
3456          *
3457          * 1) We've already reserved our space for our orphan item in the
3458          *    unlink.
3459          * 2) We're going to delete the inode item, so we don't need to update
3460          *    it at all.
3461          *
3462          * So we just need to reserve some slack space in case we add bytes when
3463          * doing the truncate.
3464          */
3465         while (1) {
3466                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3467
3468                 /*
3469                  * Try and steal from the global reserve since we will
3470                  * likely not use this space anyway, we want to try as
3471                  * hard as possible to get this to work.
3472                  */
3473                 if (ret)
3474                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3475
3476                 if (ret) {
3477                         printk(KERN_WARNING "Could not get space for a "
3478                                "delete, will truncate on mount %d\n", ret);
3479                         btrfs_orphan_del(NULL, inode);
3480                         btrfs_free_block_rsv(root, rsv);
3481                         goto no_delete;
3482                 }
3483
3484                 trans = btrfs_start_transaction(root, 0);
3485                 if (IS_ERR(trans)) {
3486                         btrfs_orphan_del(NULL, inode);
3487                         btrfs_free_block_rsv(root, rsv);
3488                         goto no_delete;
3489                 }
3490
3491                 trans->block_rsv = rsv;
3492
3493                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3494                 if (ret != -EAGAIN)
3495                         break;
3496
3497                 nr = trans->blocks_used;
3498                 btrfs_end_transaction(trans, root);
3499                 trans = NULL;
3500                 btrfs_btree_balance_dirty(root, nr);
3501         }
3502
3503         btrfs_free_block_rsv(root, rsv);
3504
3505         if (ret == 0) {
3506                 trans->block_rsv = root->orphan_block_rsv;
3507                 ret = btrfs_orphan_del(trans, inode);
3508                 BUG_ON(ret);
3509         }
3510
3511         trans->block_rsv = &root->fs_info->trans_block_rsv;
3512         if (!(root == root->fs_info->tree_root ||
3513               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3514                 btrfs_return_ino(root, btrfs_ino(inode));
3515
3516         nr = trans->blocks_used;
3517         btrfs_end_transaction(trans, root);
3518         btrfs_btree_balance_dirty(root, nr);
3519 no_delete:
3520         end_writeback(inode);
3521         return;
3522 }
3523
3524 /*
3525  * this returns the key found in the dir entry in the location pointer.
3526  * If no dir entries were found, location->objectid is 0.
3527  */
3528 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3529                                struct btrfs_key *location)
3530 {
3531         const char *name = dentry->d_name.name;
3532         int namelen = dentry->d_name.len;
3533         struct btrfs_dir_item *di;
3534         struct btrfs_path *path;
3535         struct btrfs_root *root = BTRFS_I(dir)->root;
3536         int ret = 0;
3537
3538         path = btrfs_alloc_path();
3539         if (!path)
3540                 return -ENOMEM;
3541
3542         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3543                                     namelen, 0);
3544         if (IS_ERR(di))
3545                 ret = PTR_ERR(di);
3546
3547         if (IS_ERR_OR_NULL(di))
3548                 goto out_err;
3549
3550         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3551 out:
3552         btrfs_free_path(path);
3553         return ret;
3554 out_err:
3555         location->objectid = 0;
3556         goto out;
3557 }
3558
3559 /*
3560  * when we hit a tree root in a directory, the btrfs part of the inode
3561  * needs to be changed to reflect the root directory of the tree root.  This
3562  * is kind of like crossing a mount point.
3563  */
3564 static int fixup_tree_root_location(struct btrfs_root *root,
3565                                     struct inode *dir,
3566                                     struct dentry *dentry,
3567                                     struct btrfs_key *location,
3568                                     struct btrfs_root **sub_root)
3569 {
3570         struct btrfs_path *path;
3571         struct btrfs_root *new_root;
3572         struct btrfs_root_ref *ref;
3573         struct extent_buffer *leaf;
3574         int ret;
3575         int err = 0;
3576
3577         path = btrfs_alloc_path();
3578         if (!path) {
3579                 err = -ENOMEM;
3580                 goto out;
3581         }
3582
3583         err = -ENOENT;
3584         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3585                                   BTRFS_I(dir)->root->root_key.objectid,
3586                                   location->objectid);
3587         if (ret) {
3588                 if (ret < 0)
3589                         err = ret;
3590                 goto out;
3591         }
3592
3593         leaf = path->nodes[0];
3594         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3595         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3596             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3597                 goto out;
3598
3599         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3600                                    (unsigned long)(ref + 1),
3601                                    dentry->d_name.len);
3602         if (ret)
3603                 goto out;
3604
3605         btrfs_release_path(path);
3606
3607         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3608         if (IS_ERR(new_root)) {
3609                 err = PTR_ERR(new_root);
3610                 goto out;
3611         }
3612
3613         if (btrfs_root_refs(&new_root->root_item) == 0) {
3614                 err = -ENOENT;
3615                 goto out;
3616         }
3617
3618         *sub_root = new_root;
3619         location->objectid = btrfs_root_dirid(&new_root->root_item);
3620         location->type = BTRFS_INODE_ITEM_KEY;
3621         location->offset = 0;
3622         err = 0;
3623 out:
3624         btrfs_free_path(path);
3625         return err;
3626 }
3627
3628 static void inode_tree_add(struct inode *inode)
3629 {
3630         struct btrfs_root *root = BTRFS_I(inode)->root;
3631         struct btrfs_inode *entry;
3632         struct rb_node **p;
3633         struct rb_node *parent;
3634         u64 ino = btrfs_ino(inode);
3635 again:
3636         p = &root->inode_tree.rb_node;
3637         parent = NULL;
3638
3639         if (inode_unhashed(inode))
3640                 return;
3641
3642         spin_lock(&root->inode_lock);
3643         while (*p) {
3644                 parent = *p;
3645                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3646
3647                 if (ino < btrfs_ino(&entry->vfs_inode))
3648                         p = &parent->rb_left;
3649                 else if (ino > btrfs_ino(&entry->vfs_inode))
3650                         p = &parent->rb_right;
3651                 else {
3652                         WARN_ON(!(entry->vfs_inode.i_state &
3653                                   (I_WILL_FREE | I_FREEING)));
3654                         rb_erase(parent, &root->inode_tree);
3655                         RB_CLEAR_NODE(parent);
3656                         spin_unlock(&root->inode_lock);
3657                         goto again;
3658                 }
3659         }
3660         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3661         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3662         spin_unlock(&root->inode_lock);
3663 }
3664
3665 static void inode_tree_del(struct inode *inode)
3666 {
3667         struct btrfs_root *root = BTRFS_I(inode)->root;
3668         int empty = 0;
3669
3670         spin_lock(&root->inode_lock);
3671         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3672                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3673                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3674                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3675         }
3676         spin_unlock(&root->inode_lock);
3677
3678         /*
3679          * Free space cache has inodes in the tree root, but the tree root has a
3680          * root_refs of 0, so this could end up dropping the tree root as a
3681          * snapshot, so we need the extra !root->fs_info->tree_root check to
3682          * make sure we don't drop it.
3683          */
3684         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3685             root != root->fs_info->tree_root) {
3686                 synchronize_srcu(&root->fs_info->subvol_srcu);
3687                 spin_lock(&root->inode_lock);
3688                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3689                 spin_unlock(&root->inode_lock);
3690                 if (empty)
3691                         btrfs_add_dead_root(root);
3692         }
3693 }
3694
3695 int btrfs_invalidate_inodes(struct btrfs_root *root)
3696 {
3697         struct rb_node *node;
3698         struct rb_node *prev;
3699         struct btrfs_inode *entry;
3700         struct inode *inode;
3701         u64 objectid = 0;
3702
3703         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3704
3705         spin_lock(&root->inode_lock);
3706 again:
3707         node = root->inode_tree.rb_node;
3708         prev = NULL;
3709         while (node) {
3710                 prev = node;
3711                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3712
3713                 if (objectid < btrfs_ino(&entry->vfs_inode))
3714                         node = node->rb_left;
3715                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3716                         node = node->rb_right;
3717                 else
3718                         break;
3719         }
3720         if (!node) {
3721                 while (prev) {
3722                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3723                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3724                                 node = prev;
3725                                 break;
3726                         }
3727                         prev = rb_next(prev);
3728                 }
3729         }
3730         while (node) {
3731                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3732                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3733                 inode = igrab(&entry->vfs_inode);
3734                 if (inode) {
3735                         spin_unlock(&root->inode_lock);
3736                         if (atomic_read(&inode->i_count) > 1)
3737                                 d_prune_aliases(inode);
3738                         /*
3739                          * btrfs_drop_inode will have it removed from
3740                          * the inode cache when its usage count
3741                          * hits zero.
3742                          */
3743                         iput(inode);
3744                         cond_resched();
3745                         spin_lock(&root->inode_lock);
3746                         goto again;
3747                 }
3748
3749                 if (cond_resched_lock(&root->inode_lock))
3750                         goto again;
3751
3752                 node = rb_next(node);
3753         }
3754         spin_unlock(&root->inode_lock);
3755         return 0;
3756 }
3757
3758 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3759 {
3760         struct btrfs_iget_args *args = p;
3761         inode->i_ino = args->ino;
3762         BTRFS_I(inode)->root = args->root;
3763         btrfs_set_inode_space_info(args->root, inode);
3764         return 0;
3765 }
3766
3767 static int btrfs_find_actor(struct inode *inode, void *opaque)
3768 {
3769         struct btrfs_iget_args *args = opaque;
3770         return args->ino == btrfs_ino(inode) &&
3771                 args->root == BTRFS_I(inode)->root;
3772 }
3773
3774 static struct inode *btrfs_iget_locked(struct super_block *s,
3775                                        u64 objectid,
3776                                        struct btrfs_root *root)
3777 {
3778         struct inode *inode;
3779         struct btrfs_iget_args args;
3780         args.ino = objectid;
3781         args.root = root;
3782
3783         inode = iget5_locked(s, objectid, btrfs_find_actor,
3784                              btrfs_init_locked_inode,
3785                              (void *)&args);
3786         return inode;
3787 }
3788
3789 /* Get an inode object given its location and corresponding root.
3790  * Returns in *is_new if the inode was read from disk
3791  */
3792 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3793                          struct btrfs_root *root, int *new)
3794 {
3795         struct inode *inode;
3796
3797         inode = btrfs_iget_locked(s, location->objectid, root);
3798         if (!inode)
3799                 return ERR_PTR(-ENOMEM);
3800
3801         if (inode->i_state & I_NEW) {
3802                 BTRFS_I(inode)->root = root;
3803                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3804                 btrfs_read_locked_inode(inode);
3805                 if (!is_bad_inode(inode)) {
3806                         inode_tree_add(inode);
3807                         unlock_new_inode(inode);
3808                         if (new)
3809                                 *new = 1;
3810                 } else {
3811                         unlock_new_inode(inode);
3812                         iput(inode);
3813                         inode = ERR_PTR(-ESTALE);
3814                 }
3815         }
3816
3817         return inode;
3818 }
3819
3820 static struct inode *new_simple_dir(struct super_block *s,
3821                                     struct btrfs_key *key,
3822                                     struct btrfs_root *root)
3823 {
3824         struct inode *inode = new_inode(s);
3825
3826         if (!inode)
3827                 return ERR_PTR(-ENOMEM);
3828
3829         BTRFS_I(inode)->root = root;
3830         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3831         BTRFS_I(inode)->dummy_inode = 1;
3832
3833         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3834         inode->i_op = &simple_dir_inode_operations;
3835         inode->i_fop = &simple_dir_operations;
3836         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3837         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3838
3839         return inode;
3840 }
3841
3842 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3843 {
3844         struct inode *inode;
3845         struct btrfs_root *root = BTRFS_I(dir)->root;
3846         struct btrfs_root *sub_root = root;
3847         struct btrfs_key location;
3848         int index;
3849         int ret = 0;
3850
3851         if (dentry->d_name.len > BTRFS_NAME_LEN)
3852                 return ERR_PTR(-ENAMETOOLONG);
3853
3854         if (unlikely(d_need_lookup(dentry))) {
3855                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3856                 kfree(dentry->d_fsdata);
3857                 dentry->d_fsdata = NULL;
3858                 /* This thing is hashed, drop it for now */
3859                 d_drop(dentry);
3860         } else {
3861                 ret = btrfs_inode_by_name(dir, dentry, &location);
3862         }
3863
3864         if (ret < 0)
3865                 return ERR_PTR(ret);
3866
3867         if (location.objectid == 0)
3868                 return NULL;
3869
3870         if (location.type == BTRFS_INODE_ITEM_KEY) {
3871                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3872                 return inode;
3873         }
3874
3875         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3876
3877         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3878         ret = fixup_tree_root_location(root, dir, dentry,
3879                                        &location, &sub_root);
3880         if (ret < 0) {
3881                 if (ret != -ENOENT)
3882                         inode = ERR_PTR(ret);
3883                 else
3884                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3885         } else {
3886                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3887         }
3888         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3889
3890         if (!IS_ERR(inode) && root != sub_root) {
3891                 down_read(&root->fs_info->cleanup_work_sem);
3892                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3893                         ret = btrfs_orphan_cleanup(sub_root);
3894                 up_read(&root->fs_info->cleanup_work_sem);
3895                 if (ret)
3896                         inode = ERR_PTR(ret);
3897         }
3898
3899         return inode;
3900 }
3901
3902 static int btrfs_dentry_delete(const struct dentry *dentry)
3903 {
3904         struct btrfs_root *root;
3905
3906         if (!dentry->d_inode && !IS_ROOT(dentry))
3907                 dentry = dentry->d_parent;
3908
3909         if (dentry->d_inode) {
3910                 root = BTRFS_I(dentry->d_inode)->root;
3911                 if (btrfs_root_refs(&root->root_item) == 0)
3912                         return 1;
3913         }
3914         return 0;
3915 }
3916
3917 static void btrfs_dentry_release(struct dentry *dentry)
3918 {
3919         if (dentry->d_fsdata)
3920                 kfree(dentry->d_fsdata);
3921 }
3922
3923 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3924                                    struct nameidata *nd)
3925 {
3926         struct dentry *ret;
3927
3928         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3929         if (unlikely(d_need_lookup(dentry))) {
3930                 spin_lock(&dentry->d_lock);
3931                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
3932                 spin_unlock(&dentry->d_lock);
3933         }
3934         return ret;
3935 }
3936
3937 unsigned char btrfs_filetype_table[] = {
3938         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
3939 };
3940
3941 static int btrfs_real_readdir(struct file *filp, void *dirent,
3942                               filldir_t filldir)
3943 {
3944         struct inode *inode = filp->f_dentry->d_inode;
3945         struct btrfs_root *root = BTRFS_I(inode)->root;
3946         struct btrfs_item *item;
3947         struct btrfs_dir_item *di;
3948         struct btrfs_key key;
3949         struct btrfs_key found_key;
3950         struct btrfs_path *path;
3951         struct list_head ins_list;
3952         struct list_head del_list;
3953         struct qstr q;
3954         int ret;
3955         struct extent_buffer *leaf;
3956         int slot;
3957         unsigned char d_type;
3958         int over = 0;
3959         u32 di_cur;
3960         u32 di_total;
3961         u32 di_len;
3962         int key_type = BTRFS_DIR_INDEX_KEY;
3963         char tmp_name[32];
3964         char *name_ptr;
3965         int name_len;
3966         int is_curr = 0;        /* filp->f_pos points to the current index? */
3967
3968         /* FIXME, use a real flag for deciding about the key type */
3969         if (root->fs_info->tree_root == root)
3970                 key_type = BTRFS_DIR_ITEM_KEY;
3971
3972         /* special case for "." */
3973         if (filp->f_pos == 0) {
3974                 over = filldir(dirent, ".", 1,
3975                                filp->f_pos, btrfs_ino(inode), DT_DIR);
3976                 if (over)
3977                         return 0;
3978                 filp->f_pos = 1;
3979         }
3980         /* special case for .., just use the back ref */
3981         if (filp->f_pos == 1) {
3982                 u64 pino = parent_ino(filp->f_path.dentry);
3983                 over = filldir(dirent, "..", 2,
3984                                filp->f_pos, pino, DT_DIR);
3985                 if (over)
3986                         return 0;
3987                 filp->f_pos = 2;
3988         }
3989         path = btrfs_alloc_path();
3990         if (!path)
3991                 return -ENOMEM;
3992
3993         path->reada = 1;
3994
3995         if (key_type == BTRFS_DIR_INDEX_KEY) {
3996                 INIT_LIST_HEAD(&ins_list);
3997                 INIT_LIST_HEAD(&del_list);
3998                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
3999         }
4000
4001         btrfs_set_key_type(&key, key_type);
4002         key.offset = filp->f_pos;
4003         key.objectid = btrfs_ino(inode);
4004
4005         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4006         if (ret < 0)
4007                 goto err;
4008
4009         while (1) {
4010                 leaf = path->nodes[0];
4011                 slot = path->slots[0];
4012                 if (slot >= btrfs_header_nritems(leaf)) {
4013                         ret = btrfs_next_leaf(root, path);
4014                         if (ret < 0)
4015                                 goto err;
4016                         else if (ret > 0)
4017                                 break;
4018                         continue;
4019                 }
4020
4021                 item = btrfs_item_nr(leaf, slot);
4022                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4023
4024                 if (found_key.objectid != key.objectid)
4025                         break;
4026                 if (btrfs_key_type(&found_key) != key_type)
4027                         break;
4028                 if (found_key.offset < filp->f_pos)
4029                         goto next;
4030                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4031                     btrfs_should_delete_dir_index(&del_list,
4032                                                   found_key.offset))
4033                         goto next;
4034
4035                 filp->f_pos = found_key.offset;
4036                 is_curr = 1;
4037
4038                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4039                 di_cur = 0;
4040                 di_total = btrfs_item_size(leaf, item);
4041
4042                 while (di_cur < di_total) {
4043                         struct btrfs_key location;
4044                         struct dentry *tmp;
4045
4046                         if (verify_dir_item(root, leaf, di))
4047                                 break;
4048
4049                         name_len = btrfs_dir_name_len(leaf, di);
4050                         if (name_len <= sizeof(tmp_name)) {
4051                                 name_ptr = tmp_name;
4052                         } else {
4053                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4054                                 if (!name_ptr) {
4055                                         ret = -ENOMEM;
4056                                         goto err;
4057                                 }
4058                         }
4059                         read_extent_buffer(leaf, name_ptr,
4060                                            (unsigned long)(di + 1), name_len);
4061
4062                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4063                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4064
4065                         q.name = name_ptr;
4066                         q.len = name_len;
4067                         q.hash = full_name_hash(q.name, q.len);
4068                         tmp = d_lookup(filp->f_dentry, &q);
4069                         if (!tmp) {
4070                                 struct btrfs_key *newkey;
4071
4072                                 newkey = kzalloc(sizeof(struct btrfs_key),
4073                                                  GFP_NOFS);
4074                                 if (!newkey)
4075                                         goto no_dentry;
4076                                 tmp = d_alloc(filp->f_dentry, &q);
4077                                 if (!tmp) {
4078                                         kfree(newkey);
4079                                         dput(tmp);
4080                                         goto no_dentry;
4081                                 }
4082                                 memcpy(newkey, &location,
4083                                        sizeof(struct btrfs_key));
4084                                 tmp->d_fsdata = newkey;
4085                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4086                                 d_rehash(tmp);
4087                                 dput(tmp);
4088                         } else {
4089                                 dput(tmp);
4090                         }
4091 no_dentry:
4092                         /* is this a reference to our own snapshot? If so
4093                          * skip it
4094                          */
4095                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4096                             location.objectid == root->root_key.objectid) {
4097                                 over = 0;
4098                                 goto skip;
4099                         }
4100                         over = filldir(dirent, name_ptr, name_len,
4101                                        found_key.offset, location.objectid,
4102                                        d_type);
4103
4104 skip:
4105                         if (name_ptr != tmp_name)
4106                                 kfree(name_ptr);
4107
4108                         if (over)
4109                                 goto nopos;
4110                         di_len = btrfs_dir_name_len(leaf, di) +
4111                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4112                         di_cur += di_len;
4113                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4114                 }
4115 next:
4116                 path->slots[0]++;
4117         }
4118
4119         if (key_type == BTRFS_DIR_INDEX_KEY) {
4120                 if (is_curr)
4121                         filp->f_pos++;
4122                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4123                                                       &ins_list);
4124                 if (ret)
4125                         goto nopos;
4126         }
4127
4128         /* Reached end of directory/root. Bump pos past the last item. */
4129         if (key_type == BTRFS_DIR_INDEX_KEY)
4130                 /*
4131                  * 32-bit glibc will use getdents64, but then strtol -
4132                  * so the last number we can serve is this.
4133                  */
4134                 filp->f_pos = 0x7fffffff;
4135         else
4136                 filp->f_pos++;
4137 nopos:
4138         ret = 0;
4139 err:
4140         if (key_type == BTRFS_DIR_INDEX_KEY)
4141                 btrfs_put_delayed_items(&ins_list, &del_list);
4142         btrfs_free_path(path);
4143         return ret;
4144 }
4145
4146 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4147 {
4148         struct btrfs_root *root = BTRFS_I(inode)->root;
4149         struct btrfs_trans_handle *trans;
4150         int ret = 0;
4151         bool nolock = false;
4152
4153         if (BTRFS_I(inode)->dummy_inode)
4154                 return 0;
4155
4156         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4157                 nolock = true;
4158
4159         if (wbc->sync_mode == WB_SYNC_ALL) {
4160                 if (nolock)
4161                         trans = btrfs_join_transaction_nolock(root);
4162                 else
4163                         trans = btrfs_join_transaction(root);
4164                 if (IS_ERR(trans))
4165                         return PTR_ERR(trans);
4166                 if (nolock)
4167                         ret = btrfs_end_transaction_nolock(trans, root);
4168                 else
4169                         ret = btrfs_commit_transaction(trans, root);
4170         }
4171         return ret;
4172 }
4173
4174 /*
4175  * This is somewhat expensive, updating the tree every time the
4176  * inode changes.  But, it is most likely to find the inode in cache.
4177  * FIXME, needs more benchmarking...there are no reasons other than performance
4178  * to keep or drop this code.
4179  */
4180 void btrfs_dirty_inode(struct inode *inode, int flags)
4181 {
4182         struct btrfs_root *root = BTRFS_I(inode)->root;
4183         struct btrfs_trans_handle *trans;
4184         int ret;
4185
4186         if (BTRFS_I(inode)->dummy_inode)
4187                 return;
4188
4189         trans = btrfs_join_transaction(root);
4190         BUG_ON(IS_ERR(trans));
4191
4192         ret = btrfs_update_inode(trans, root, inode);
4193         if (ret && ret == -ENOSPC) {
4194                 /* whoops, lets try again with the full transaction */
4195                 btrfs_end_transaction(trans, root);
4196                 trans = btrfs_start_transaction(root, 1);
4197                 if (IS_ERR(trans)) {
4198                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4199                                        "dirty  inode %llu error %ld\n",
4200                                        (unsigned long long)btrfs_ino(inode),
4201                                        PTR_ERR(trans));
4202                         return;
4203                 }
4204
4205                 ret = btrfs_update_inode(trans, root, inode);
4206                 if (ret) {
4207                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4208                                        "dirty  inode %llu error %d\n",
4209                                        (unsigned long long)btrfs_ino(inode),
4210                                        ret);
4211                 }
4212         }
4213         btrfs_end_transaction(trans, root);
4214         if (BTRFS_I(inode)->delayed_node)
4215                 btrfs_balance_delayed_items(root);
4216 }
4217
4218 /*
4219  * find the highest existing sequence number in a directory
4220  * and then set the in-memory index_cnt variable to reflect
4221  * free sequence numbers
4222  */
4223 static int btrfs_set_inode_index_count(struct inode *inode)
4224 {
4225         struct btrfs_root *root = BTRFS_I(inode)->root;
4226         struct btrfs_key key, found_key;
4227         struct btrfs_path *path;
4228         struct extent_buffer *leaf;
4229         int ret;
4230
4231         key.objectid = btrfs_ino(inode);
4232         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4233         key.offset = (u64)-1;
4234
4235         path = btrfs_alloc_path();
4236         if (!path)
4237                 return -ENOMEM;
4238
4239         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4240         if (ret < 0)
4241                 goto out;
4242         /* FIXME: we should be able to handle this */
4243         if (ret == 0)
4244                 goto out;
4245         ret = 0;
4246
4247         /*
4248          * MAGIC NUMBER EXPLANATION:
4249          * since we search a directory based on f_pos we have to start at 2
4250          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4251          * else has to start at 2
4252          */
4253         if (path->slots[0] == 0) {
4254                 BTRFS_I(inode)->index_cnt = 2;
4255                 goto out;
4256         }
4257
4258         path->slots[0]--;
4259
4260         leaf = path->nodes[0];
4261         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4262
4263         if (found_key.objectid != btrfs_ino(inode) ||
4264             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4265                 BTRFS_I(inode)->index_cnt = 2;
4266                 goto out;
4267         }
4268
4269         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4270 out:
4271         btrfs_free_path(path);
4272         return ret;
4273 }
4274
4275 /*
4276  * helper to find a free sequence number in a given directory.  This current
4277  * code is very simple, later versions will do smarter things in the btree
4278  */
4279 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4280 {
4281         int ret = 0;
4282
4283         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4284                 ret = btrfs_inode_delayed_dir_index_count(dir);
4285                 if (ret) {
4286                         ret = btrfs_set_inode_index_count(dir);
4287                         if (ret)
4288                                 return ret;
4289                 }
4290         }
4291
4292         *index = BTRFS_I(dir)->index_cnt;
4293         BTRFS_I(dir)->index_cnt++;
4294
4295         return ret;
4296 }
4297
4298 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4299                                      struct btrfs_root *root,
4300                                      struct inode *dir,
4301                                      const char *name, int name_len,
4302                                      u64 ref_objectid, u64 objectid, int mode,
4303                                      u64 *index)
4304 {
4305         struct inode *inode;
4306         struct btrfs_inode_item *inode_item;
4307         struct btrfs_key *location;
4308         struct btrfs_path *path;
4309         struct btrfs_inode_ref *ref;
4310         struct btrfs_key key[2];
4311         u32 sizes[2];
4312         unsigned long ptr;
4313         int ret;
4314         int owner;
4315
4316         path = btrfs_alloc_path();
4317         if (!path)
4318                 return ERR_PTR(-ENOMEM);
4319
4320         inode = new_inode(root->fs_info->sb);
4321         if (!inode) {
4322                 btrfs_free_path(path);
4323                 return ERR_PTR(-ENOMEM);
4324         }
4325
4326         /*
4327          * we have to initialize this early, so we can reclaim the inode
4328          * number if we fail afterwards in this function.
4329          */
4330         inode->i_ino = objectid;
4331
4332         if (dir) {
4333                 trace_btrfs_inode_request(dir);
4334
4335                 ret = btrfs_set_inode_index(dir, index);
4336                 if (ret) {
4337                         btrfs_free_path(path);
4338                         iput(inode);
4339                         return ERR_PTR(ret);
4340                 }
4341         }
4342         /*
4343          * index_cnt is ignored for everything but a dir,
4344          * btrfs_get_inode_index_count has an explanation for the magic
4345          * number
4346          */
4347         BTRFS_I(inode)->index_cnt = 2;
4348         BTRFS_I(inode)->root = root;
4349         BTRFS_I(inode)->generation = trans->transid;
4350         inode->i_generation = BTRFS_I(inode)->generation;
4351         btrfs_set_inode_space_info(root, inode);
4352
4353         if (S_ISDIR(mode))
4354                 owner = 0;
4355         else
4356                 owner = 1;
4357
4358         key[0].objectid = objectid;
4359         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4360         key[0].offset = 0;
4361
4362         key[1].objectid = objectid;
4363         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4364         key[1].offset = ref_objectid;
4365
4366         sizes[0] = sizeof(struct btrfs_inode_item);
4367         sizes[1] = name_len + sizeof(*ref);
4368
4369         path->leave_spinning = 1;
4370         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4371         if (ret != 0)
4372                 goto fail;
4373
4374         inode_init_owner(inode, dir, mode);
4375         inode_set_bytes(inode, 0);
4376         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4377         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4378                                   struct btrfs_inode_item);
4379         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4380
4381         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4382                              struct btrfs_inode_ref);
4383         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4384         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4385         ptr = (unsigned long)(ref + 1);
4386         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4387
4388         btrfs_mark_buffer_dirty(path->nodes[0]);
4389         btrfs_free_path(path);
4390
4391         location = &BTRFS_I(inode)->location;
4392         location->objectid = objectid;
4393         location->offset = 0;
4394         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4395
4396         btrfs_inherit_iflags(inode, dir);
4397
4398         if (S_ISREG(mode)) {
4399                 if (btrfs_test_opt(root, NODATASUM))
4400                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4401                 if (btrfs_test_opt(root, NODATACOW) ||
4402                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4403                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4404         }
4405
4406         insert_inode_hash(inode);
4407         inode_tree_add(inode);
4408
4409         trace_btrfs_inode_new(inode);
4410         btrfs_set_inode_last_trans(trans, inode);
4411
4412         return inode;
4413 fail:
4414         if (dir)
4415                 BTRFS_I(dir)->index_cnt--;
4416         btrfs_free_path(path);
4417         iput(inode);
4418         return ERR_PTR(ret);
4419 }
4420
4421 static inline u8 btrfs_inode_type(struct inode *inode)
4422 {
4423         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4424 }
4425
4426 /*
4427  * utility function to add 'inode' into 'parent_inode' with
4428  * a give name and a given sequence number.
4429  * if 'add_backref' is true, also insert a backref from the
4430  * inode to the parent directory.
4431  */
4432 int btrfs_add_link(struct btrfs_trans_handle *trans,
4433                    struct inode *parent_inode, struct inode *inode,
4434                    const char *name, int name_len, int add_backref, u64 index)
4435 {
4436         int ret = 0;
4437         struct btrfs_key key;
4438         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4439         u64 ino = btrfs_ino(inode);
4440         u64 parent_ino = btrfs_ino(parent_inode);
4441
4442         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4443                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4444         } else {
4445                 key.objectid = ino;
4446                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4447                 key.offset = 0;
4448         }
4449
4450         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4451                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4452                                          key.objectid, root->root_key.objectid,
4453                                          parent_ino, index, name, name_len);
4454         } else if (add_backref) {
4455                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4456                                              parent_ino, index);
4457         }
4458
4459         if (ret == 0) {
4460                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4461                                             parent_inode, &key,
4462                                             btrfs_inode_type(inode), index);
4463                 BUG_ON(ret);
4464
4465                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4466                                    name_len * 2);
4467                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4468                 ret = btrfs_update_inode(trans, root, parent_inode);
4469         }
4470         return ret;
4471 }
4472
4473 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4474                             struct inode *dir, struct dentry *dentry,
4475                             struct inode *inode, int backref, u64 index)
4476 {
4477         int err = btrfs_add_link(trans, dir, inode,
4478                                  dentry->d_name.name, dentry->d_name.len,
4479                                  backref, index);
4480         if (!err) {
4481                 d_instantiate(dentry, inode);
4482                 return 0;
4483         }
4484         if (err > 0)
4485                 err = -EEXIST;
4486         return err;
4487 }
4488
4489 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4490                         int mode, dev_t rdev)
4491 {
4492         struct btrfs_trans_handle *trans;
4493         struct btrfs_root *root = BTRFS_I(dir)->root;
4494         struct inode *inode = NULL;
4495         int err;
4496         int drop_inode = 0;
4497         u64 objectid;
4498         unsigned long nr = 0;
4499         u64 index = 0;
4500
4501         if (!new_valid_dev(rdev))
4502                 return -EINVAL;
4503
4504         /*
4505          * 2 for inode item and ref
4506          * 2 for dir items
4507          * 1 for xattr if selinux is on
4508          */
4509         trans = btrfs_start_transaction(root, 5);
4510         if (IS_ERR(trans))
4511                 return PTR_ERR(trans);
4512
4513         err = btrfs_find_free_ino(root, &objectid);
4514         if (err)
4515                 goto out_unlock;
4516
4517         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4518                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4519                                 mode, &index);
4520         if (IS_ERR(inode)) {
4521                 err = PTR_ERR(inode);
4522                 goto out_unlock;
4523         }
4524
4525         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4526         if (err) {
4527                 drop_inode = 1;
4528                 goto out_unlock;
4529         }
4530
4531         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4532         if (err)
4533                 drop_inode = 1;
4534         else {
4535                 inode->i_op = &btrfs_special_inode_operations;
4536                 init_special_inode(inode, inode->i_mode, rdev);
4537                 btrfs_update_inode(trans, root, inode);
4538         }
4539 out_unlock:
4540         nr = trans->blocks_used;
4541         btrfs_end_transaction_throttle(trans, root);
4542         btrfs_btree_balance_dirty(root, nr);
4543         if (drop_inode) {
4544                 inode_dec_link_count(inode);
4545                 iput(inode);
4546         }
4547         return err;
4548 }
4549
4550 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4551                         int mode, struct nameidata *nd)
4552 {
4553         struct btrfs_trans_handle *trans;
4554         struct btrfs_root *root = BTRFS_I(dir)->root;
4555         struct inode *inode = NULL;
4556         int drop_inode = 0;
4557         int err;
4558         unsigned long nr = 0;
4559         u64 objectid;
4560         u64 index = 0;
4561
4562         /*
4563          * 2 for inode item and ref
4564          * 2 for dir items
4565          * 1 for xattr if selinux is on
4566          */
4567         trans = btrfs_start_transaction(root, 5);
4568         if (IS_ERR(trans))
4569                 return PTR_ERR(trans);
4570
4571         err = btrfs_find_free_ino(root, &objectid);
4572         if (err)
4573                 goto out_unlock;
4574
4575         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4576                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4577                                 mode, &index);
4578         if (IS_ERR(inode)) {
4579                 err = PTR_ERR(inode);
4580                 goto out_unlock;
4581         }
4582
4583         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4584         if (err) {
4585                 drop_inode = 1;
4586                 goto out_unlock;
4587         }
4588
4589         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4590         if (err)
4591                 drop_inode = 1;
4592         else {
4593                 inode->i_mapping->a_ops = &btrfs_aops;
4594                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4595                 inode->i_fop = &btrfs_file_operations;
4596                 inode->i_op = &btrfs_file_inode_operations;
4597                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4598         }
4599 out_unlock:
4600         nr = trans->blocks_used;
4601         btrfs_end_transaction_throttle(trans, root);
4602         if (drop_inode) {
4603                 inode_dec_link_count(inode);
4604                 iput(inode);
4605         }
4606         btrfs_btree_balance_dirty(root, nr);
4607         return err;
4608 }
4609
4610 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4611                       struct dentry *dentry)
4612 {
4613         struct btrfs_trans_handle *trans;
4614         struct btrfs_root *root = BTRFS_I(dir)->root;
4615         struct inode *inode = old_dentry->d_inode;
4616         u64 index;
4617         unsigned long nr = 0;
4618         int err;
4619         int drop_inode = 0;
4620
4621         /* do not allow sys_link's with other subvols of the same device */
4622         if (root->objectid != BTRFS_I(inode)->root->objectid)
4623                 return -EXDEV;
4624
4625         if (inode->i_nlink == ~0U)
4626                 return -EMLINK;
4627
4628         err = btrfs_set_inode_index(dir, &index);
4629         if (err)
4630                 goto fail;
4631
4632         /*
4633          * 2 items for inode and inode ref
4634          * 2 items for dir items
4635          * 1 item for parent inode
4636          */
4637         trans = btrfs_start_transaction(root, 5);
4638         if (IS_ERR(trans)) {
4639                 err = PTR_ERR(trans);
4640                 goto fail;
4641         }
4642
4643         btrfs_inc_nlink(inode);
4644         inode->i_ctime = CURRENT_TIME;
4645         ihold(inode);
4646
4647         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4648
4649         if (err) {
4650                 drop_inode = 1;
4651         } else {
4652                 struct dentry *parent = dentry->d_parent;
4653                 err = btrfs_update_inode(trans, root, inode);
4654                 BUG_ON(err);
4655                 btrfs_log_new_name(trans, inode, NULL, parent);
4656         }
4657
4658         nr = trans->blocks_used;
4659         btrfs_end_transaction_throttle(trans, root);
4660 fail:
4661         if (drop_inode) {
4662                 inode_dec_link_count(inode);
4663                 iput(inode);
4664         }
4665         btrfs_btree_balance_dirty(root, nr);
4666         return err;
4667 }
4668
4669 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4670 {
4671         struct inode *inode = NULL;
4672         struct btrfs_trans_handle *trans;
4673         struct btrfs_root *root = BTRFS_I(dir)->root;
4674         int err = 0;
4675         int drop_on_err = 0;
4676         u64 objectid = 0;
4677         u64 index = 0;
4678         unsigned long nr = 1;
4679
4680         /*
4681          * 2 items for inode and ref
4682          * 2 items for dir items
4683          * 1 for xattr if selinux is on
4684          */
4685         trans = btrfs_start_transaction(root, 5);
4686         if (IS_ERR(trans))
4687                 return PTR_ERR(trans);
4688
4689         err = btrfs_find_free_ino(root, &objectid);
4690         if (err)
4691                 goto out_fail;
4692
4693         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4694                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4695                                 S_IFDIR | mode, &index);
4696         if (IS_ERR(inode)) {
4697                 err = PTR_ERR(inode);
4698                 goto out_fail;
4699         }
4700
4701         drop_on_err = 1;
4702
4703         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4704         if (err)
4705                 goto out_fail;
4706
4707         inode->i_op = &btrfs_dir_inode_operations;
4708         inode->i_fop = &btrfs_dir_file_operations;
4709
4710         btrfs_i_size_write(inode, 0);
4711         err = btrfs_update_inode(trans, root, inode);
4712         if (err)
4713                 goto out_fail;
4714
4715         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4716                              dentry->d_name.len, 0, index);
4717         if (err)
4718                 goto out_fail;
4719
4720         d_instantiate(dentry, inode);
4721         drop_on_err = 0;
4722
4723 out_fail:
4724         nr = trans->blocks_used;
4725         btrfs_end_transaction_throttle(trans, root);
4726         if (drop_on_err)
4727                 iput(inode);
4728         btrfs_btree_balance_dirty(root, nr);
4729         return err;
4730 }
4731
4732 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4733  * and an extent that you want to insert, deal with overlap and insert
4734  * the new extent into the tree.
4735  */
4736 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4737                                 struct extent_map *existing,
4738                                 struct extent_map *em,
4739                                 u64 map_start, u64 map_len)
4740 {
4741         u64 start_diff;
4742
4743         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4744         start_diff = map_start - em->start;
4745         em->start = map_start;
4746         em->len = map_len;
4747         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4748             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4749                 em->block_start += start_diff;
4750                 em->block_len -= start_diff;
4751         }
4752         return add_extent_mapping(em_tree, em);
4753 }
4754
4755 static noinline int uncompress_inline(struct btrfs_path *path,
4756                                       struct inode *inode, struct page *page,
4757                                       size_t pg_offset, u64 extent_offset,
4758                                       struct btrfs_file_extent_item *item)
4759 {
4760         int ret;
4761         struct extent_buffer *leaf = path->nodes[0];
4762         char *tmp;
4763         size_t max_size;
4764         unsigned long inline_size;
4765         unsigned long ptr;
4766         int compress_type;
4767
4768         WARN_ON(pg_offset != 0);
4769         compress_type = btrfs_file_extent_compression(leaf, item);
4770         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4771         inline_size = btrfs_file_extent_inline_item_len(leaf,
4772                                         btrfs_item_nr(leaf, path->slots[0]));
4773         tmp = kmalloc(inline_size, GFP_NOFS);
4774         if (!tmp)
4775                 return -ENOMEM;
4776         ptr = btrfs_file_extent_inline_start(item);
4777
4778         read_extent_buffer(leaf, tmp, ptr, inline_size);
4779
4780         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4781         ret = btrfs_decompress(compress_type, tmp, page,
4782                                extent_offset, inline_size, max_size);
4783         if (ret) {
4784                 char *kaddr = kmap_atomic(page, KM_USER0);
4785                 unsigned long copy_size = min_t(u64,
4786                                   PAGE_CACHE_SIZE - pg_offset,
4787                                   max_size - extent_offset);
4788                 memset(kaddr + pg_offset, 0, copy_size);
4789                 kunmap_atomic(kaddr, KM_USER0);
4790         }
4791         kfree(tmp);
4792         return 0;
4793 }
4794
4795 /*
4796  * a bit scary, this does extent mapping from logical file offset to the disk.
4797  * the ugly parts come from merging extents from the disk with the in-ram
4798  * representation.  This gets more complex because of the data=ordered code,
4799  * where the in-ram extents might be locked pending data=ordered completion.
4800  *
4801  * This also copies inline extents directly into the page.
4802  */
4803
4804 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4805                                     size_t pg_offset, u64 start, u64 len,
4806                                     int create)
4807 {
4808         int ret;
4809         int err = 0;
4810         u64 bytenr;
4811         u64 extent_start = 0;
4812         u64 extent_end = 0;
4813         u64 objectid = btrfs_ino(inode);
4814         u32 found_type;
4815         struct btrfs_path *path = NULL;
4816         struct btrfs_root *root = BTRFS_I(inode)->root;
4817         struct btrfs_file_extent_item *item;
4818         struct extent_buffer *leaf;
4819         struct btrfs_key found_key;
4820         struct extent_map *em = NULL;
4821         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4822         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4823         struct btrfs_trans_handle *trans = NULL;
4824         int compress_type;
4825
4826 again:
4827         read_lock(&em_tree->lock);
4828         em = lookup_extent_mapping(em_tree, start, len);
4829         if (em)
4830                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4831         read_unlock(&em_tree->lock);
4832
4833         if (em) {
4834                 if (em->start > start || em->start + em->len <= start)
4835                         free_extent_map(em);
4836                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4837                         free_extent_map(em);
4838                 else
4839                         goto out;
4840         }
4841         em = alloc_extent_map();
4842         if (!em) {
4843                 err = -ENOMEM;
4844                 goto out;
4845         }
4846         em->bdev = root->fs_info->fs_devices->latest_bdev;
4847         em->start = EXTENT_MAP_HOLE;
4848         em->orig_start = EXTENT_MAP_HOLE;
4849         em->len = (u64)-1;
4850         em->block_len = (u64)-1;
4851
4852         if (!path) {
4853                 path = btrfs_alloc_path();
4854                 if (!path) {
4855                         err = -ENOMEM;
4856                         goto out;
4857                 }
4858                 /*
4859                  * Chances are we'll be called again, so go ahead and do
4860                  * readahead
4861                  */
4862                 path->reada = 1;
4863         }
4864
4865         ret = btrfs_lookup_file_extent(trans, root, path,
4866                                        objectid, start, trans != NULL);
4867         if (ret < 0) {
4868                 err = ret;
4869                 goto out;
4870         }
4871
4872         if (ret != 0) {
4873                 if (path->slots[0] == 0)
4874                         goto not_found;
4875                 path->slots[0]--;
4876         }
4877
4878         leaf = path->nodes[0];
4879         item = btrfs_item_ptr(leaf, path->slots[0],
4880                               struct btrfs_file_extent_item);
4881         /* are we inside the extent that was found? */
4882         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4883         found_type = btrfs_key_type(&found_key);
4884         if (found_key.objectid != objectid ||
4885             found_type != BTRFS_EXTENT_DATA_KEY) {
4886                 goto not_found;
4887         }
4888
4889         found_type = btrfs_file_extent_type(leaf, item);
4890         extent_start = found_key.offset;
4891         compress_type = btrfs_file_extent_compression(leaf, item);
4892         if (found_type == BTRFS_FILE_EXTENT_REG ||
4893             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4894                 extent_end = extent_start +
4895                        btrfs_file_extent_num_bytes(leaf, item);
4896         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4897                 size_t size;
4898                 size = btrfs_file_extent_inline_len(leaf, item);
4899                 extent_end = (extent_start + size + root->sectorsize - 1) &
4900                         ~((u64)root->sectorsize - 1);
4901         }
4902
4903         if (start >= extent_end) {
4904                 path->slots[0]++;
4905                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
4906                         ret = btrfs_next_leaf(root, path);
4907                         if (ret < 0) {
4908                                 err = ret;
4909                                 goto out;
4910                         }
4911                         if (ret > 0)
4912                                 goto not_found;
4913                         leaf = path->nodes[0];
4914                 }
4915                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4916                 if (found_key.objectid != objectid ||
4917                     found_key.type != BTRFS_EXTENT_DATA_KEY)
4918                         goto not_found;
4919                 if (start + len <= found_key.offset)
4920                         goto not_found;
4921                 em->start = start;
4922                 em->len = found_key.offset - start;
4923                 goto not_found_em;
4924         }
4925
4926         if (found_type == BTRFS_FILE_EXTENT_REG ||
4927             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
4928                 em->start = extent_start;
4929                 em->len = extent_end - extent_start;
4930                 em->orig_start = extent_start -
4931                                  btrfs_file_extent_offset(leaf, item);
4932                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
4933                 if (bytenr == 0) {
4934                         em->block_start = EXTENT_MAP_HOLE;
4935                         goto insert;
4936                 }
4937                 if (compress_type != BTRFS_COMPRESS_NONE) {
4938                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4939                         em->compress_type = compress_type;
4940                         em->block_start = bytenr;
4941                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
4942                                                                          item);
4943                 } else {
4944                         bytenr += btrfs_file_extent_offset(leaf, item);
4945                         em->block_start = bytenr;
4946                         em->block_len = em->len;
4947                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
4948                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
4949                 }
4950                 goto insert;
4951         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
4952                 unsigned long ptr;
4953                 char *map;
4954                 size_t size;
4955                 size_t extent_offset;
4956                 size_t copy_size;
4957
4958                 em->block_start = EXTENT_MAP_INLINE;
4959                 if (!page || create) {
4960                         em->start = extent_start;
4961                         em->len = extent_end - extent_start;
4962                         goto out;
4963                 }
4964
4965                 size = btrfs_file_extent_inline_len(leaf, item);
4966                 extent_offset = page_offset(page) + pg_offset - extent_start;
4967                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
4968                                 size - extent_offset);
4969                 em->start = extent_start + extent_offset;
4970                 em->len = (copy_size + root->sectorsize - 1) &
4971                         ~((u64)root->sectorsize - 1);
4972                 em->orig_start = EXTENT_MAP_INLINE;
4973                 if (compress_type) {
4974                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
4975                         em->compress_type = compress_type;
4976                 }
4977                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
4978                 if (create == 0 && !PageUptodate(page)) {
4979                         if (btrfs_file_extent_compression(leaf, item) !=
4980                             BTRFS_COMPRESS_NONE) {
4981                                 ret = uncompress_inline(path, inode, page,
4982                                                         pg_offset,
4983                                                         extent_offset, item);
4984                                 BUG_ON(ret);
4985                         } else {
4986                                 map = kmap(page);
4987                                 read_extent_buffer(leaf, map + pg_offset, ptr,
4988                                                    copy_size);
4989                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
4990                                         memset(map + pg_offset + copy_size, 0,
4991                                                PAGE_CACHE_SIZE - pg_offset -
4992                                                copy_size);
4993                                 }
4994                                 kunmap(page);
4995                         }
4996                         flush_dcache_page(page);
4997                 } else if (create && PageUptodate(page)) {
4998                         WARN_ON(1);
4999                         if (!trans) {
5000                                 kunmap(page);
5001                                 free_extent_map(em);
5002                                 em = NULL;
5003
5004                                 btrfs_release_path(path);
5005                                 trans = btrfs_join_transaction(root);
5006
5007                                 if (IS_ERR(trans))
5008                                         return ERR_CAST(trans);
5009                                 goto again;
5010                         }
5011                         map = kmap(page);
5012                         write_extent_buffer(leaf, map + pg_offset, ptr,
5013                                             copy_size);
5014                         kunmap(page);
5015                         btrfs_mark_buffer_dirty(leaf);
5016                 }
5017                 set_extent_uptodate(io_tree, em->start,
5018                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5019                 goto insert;
5020         } else {
5021                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5022                 WARN_ON(1);
5023         }
5024 not_found:
5025         em->start = start;
5026         em->len = len;
5027 not_found_em:
5028         em->block_start = EXTENT_MAP_HOLE;
5029         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5030 insert:
5031         btrfs_release_path(path);
5032         if (em->start > start || extent_map_end(em) <= start) {
5033                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5034                        "[%llu %llu]\n", (unsigned long long)em->start,
5035                        (unsigned long long)em->len,
5036                        (unsigned long long)start,
5037                        (unsigned long long)len);
5038                 err = -EIO;
5039                 goto out;
5040         }
5041
5042         err = 0;
5043         write_lock(&em_tree->lock);
5044         ret = add_extent_mapping(em_tree, em);
5045         /* it is possible that someone inserted the extent into the tree
5046          * while we had the lock dropped.  It is also possible that
5047          * an overlapping map exists in the tree
5048          */
5049         if (ret == -EEXIST) {
5050                 struct extent_map *existing;
5051
5052                 ret = 0;
5053
5054                 existing = lookup_extent_mapping(em_tree, start, len);
5055                 if (existing && (existing->start > start ||
5056                     existing->start + existing->len <= start)) {
5057                         free_extent_map(existing);
5058                         existing = NULL;
5059                 }
5060                 if (!existing) {
5061                         existing = lookup_extent_mapping(em_tree, em->start,
5062                                                          em->len);
5063                         if (existing) {
5064                                 err = merge_extent_mapping(em_tree, existing,
5065                                                            em, start,
5066                                                            root->sectorsize);
5067                                 free_extent_map(existing);
5068                                 if (err) {
5069                                         free_extent_map(em);
5070                                         em = NULL;
5071                                 }
5072                         } else {
5073                                 err = -EIO;
5074                                 free_extent_map(em);
5075                                 em = NULL;
5076                         }
5077                 } else {
5078                         free_extent_map(em);
5079                         em = existing;
5080                         err = 0;
5081                 }
5082         }
5083         write_unlock(&em_tree->lock);
5084 out:
5085
5086         trace_btrfs_get_extent(root, em);
5087
5088         if (path)
5089                 btrfs_free_path(path);
5090         if (trans) {
5091                 ret = btrfs_end_transaction(trans, root);
5092                 if (!err)
5093                         err = ret;
5094         }
5095         if (err) {
5096                 free_extent_map(em);
5097                 return ERR_PTR(err);
5098         }
5099         return em;
5100 }
5101
5102 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5103                                            size_t pg_offset, u64 start, u64 len,
5104                                            int create)
5105 {
5106         struct extent_map *em;
5107         struct extent_map *hole_em = NULL;
5108         u64 range_start = start;
5109         u64 end;
5110         u64 found;
5111         u64 found_end;
5112         int err = 0;
5113
5114         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5115         if (IS_ERR(em))
5116                 return em;
5117         if (em) {
5118                 /*
5119                  * if our em maps to a hole, there might
5120                  * actually be delalloc bytes behind it
5121                  */
5122                 if (em->block_start != EXTENT_MAP_HOLE)
5123                         return em;
5124                 else
5125                         hole_em = em;
5126         }
5127
5128         /* check to see if we've wrapped (len == -1 or similar) */
5129         end = start + len;
5130         if (end < start)
5131                 end = (u64)-1;
5132         else
5133                 end -= 1;
5134
5135         em = NULL;
5136
5137         /* ok, we didn't find anything, lets look for delalloc */
5138         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5139                                  end, len, EXTENT_DELALLOC, 1);
5140         found_end = range_start + found;
5141         if (found_end < range_start)
5142                 found_end = (u64)-1;
5143
5144         /*
5145          * we didn't find anything useful, return
5146          * the original results from get_extent()
5147          */
5148         if (range_start > end || found_end <= start) {
5149                 em = hole_em;
5150                 hole_em = NULL;
5151                 goto out;
5152         }
5153
5154         /* adjust the range_start to make sure it doesn't
5155          * go backwards from the start they passed in
5156          */
5157         range_start = max(start,range_start);
5158         found = found_end - range_start;
5159
5160         if (found > 0) {
5161                 u64 hole_start = start;
5162                 u64 hole_len = len;
5163
5164                 em = alloc_extent_map();
5165                 if (!em) {
5166                         err = -ENOMEM;
5167                         goto out;
5168                 }
5169                 /*
5170                  * when btrfs_get_extent can't find anything it
5171                  * returns one huge hole
5172                  *
5173                  * make sure what it found really fits our range, and
5174                  * adjust to make sure it is based on the start from
5175                  * the caller
5176                  */
5177                 if (hole_em) {
5178                         u64 calc_end = extent_map_end(hole_em);
5179
5180                         if (calc_end <= start || (hole_em->start > end)) {
5181                                 free_extent_map(hole_em);
5182                                 hole_em = NULL;
5183                         } else {
5184                                 hole_start = max(hole_em->start, start);
5185                                 hole_len = calc_end - hole_start;
5186                         }
5187                 }
5188                 em->bdev = NULL;
5189                 if (hole_em && range_start > hole_start) {
5190                         /* our hole starts before our delalloc, so we
5191                          * have to return just the parts of the hole
5192                          * that go until  the delalloc starts
5193                          */
5194                         em->len = min(hole_len,
5195                                       range_start - hole_start);
5196                         em->start = hole_start;
5197                         em->orig_start = hole_start;
5198                         /*
5199                          * don't adjust block start at all,
5200                          * it is fixed at EXTENT_MAP_HOLE
5201                          */
5202                         em->block_start = hole_em->block_start;
5203                         em->block_len = hole_len;
5204                 } else {
5205                         em->start = range_start;
5206                         em->len = found;
5207                         em->orig_start = range_start;
5208                         em->block_start = EXTENT_MAP_DELALLOC;
5209                         em->block_len = found;
5210                 }
5211         } else if (hole_em) {
5212                 return hole_em;
5213         }
5214 out:
5215
5216         free_extent_map(hole_em);
5217         if (err) {
5218                 free_extent_map(em);
5219                 return ERR_PTR(err);
5220         }
5221         return em;
5222 }
5223
5224 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5225                                                   struct extent_map *em,
5226                                                   u64 start, u64 len)
5227 {
5228         struct btrfs_root *root = BTRFS_I(inode)->root;
5229         struct btrfs_trans_handle *trans;
5230         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5231         struct btrfs_key ins;
5232         u64 alloc_hint;
5233         int ret;
5234         bool insert = false;
5235
5236         /*
5237          * Ok if the extent map we looked up is a hole and is for the exact
5238          * range we want, there is no reason to allocate a new one, however if
5239          * it is not right then we need to free this one and drop the cache for
5240          * our range.
5241          */
5242         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5243             em->len != len) {
5244                 free_extent_map(em);
5245                 em = NULL;
5246                 insert = true;
5247                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5248         }
5249
5250         trans = btrfs_join_transaction(root);
5251         if (IS_ERR(trans))
5252                 return ERR_CAST(trans);
5253
5254         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5255                 btrfs_add_inode_defrag(trans, inode);
5256
5257         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5258
5259         alloc_hint = get_extent_allocation_hint(inode, start, len);
5260         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5261                                    alloc_hint, (u64)-1, &ins, 1);
5262         if (ret) {
5263                 em = ERR_PTR(ret);
5264                 goto out;
5265         }
5266
5267         if (!em) {
5268                 em = alloc_extent_map();
5269                 if (!em) {
5270                         em = ERR_PTR(-ENOMEM);
5271                         goto out;
5272                 }
5273         }
5274
5275         em->start = start;
5276         em->orig_start = em->start;
5277         em->len = ins.offset;
5278
5279         em->block_start = ins.objectid;
5280         em->block_len = ins.offset;
5281         em->bdev = root->fs_info->fs_devices->latest_bdev;
5282
5283         /*
5284          * We need to do this because if we're using the original em we searched
5285          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5286          */
5287         em->flags = 0;
5288         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5289
5290         while (insert) {
5291                 write_lock(&em_tree->lock);
5292                 ret = add_extent_mapping(em_tree, em);
5293                 write_unlock(&em_tree->lock);
5294                 if (ret != -EEXIST)
5295                         break;
5296                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5297         }
5298
5299         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5300                                            ins.offset, ins.offset, 0);
5301         if (ret) {
5302                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5303                 em = ERR_PTR(ret);
5304         }
5305 out:
5306         btrfs_end_transaction(trans, root);
5307         return em;
5308 }
5309
5310 /*
5311  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5312  * block must be cow'd
5313  */
5314 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5315                                       struct inode *inode, u64 offset, u64 len)
5316 {
5317         struct btrfs_path *path;
5318         int ret;
5319         struct extent_buffer *leaf;
5320         struct btrfs_root *root = BTRFS_I(inode)->root;
5321         struct btrfs_file_extent_item *fi;
5322         struct btrfs_key key;
5323         u64 disk_bytenr;
5324         u64 backref_offset;
5325         u64 extent_end;
5326         u64 num_bytes;
5327         int slot;
5328         int found_type;
5329
5330         path = btrfs_alloc_path();
5331         if (!path)
5332                 return -ENOMEM;
5333
5334         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5335                                        offset, 0);
5336         if (ret < 0)
5337                 goto out;
5338
5339         slot = path->slots[0];
5340         if (ret == 1) {
5341                 if (slot == 0) {
5342                         /* can't find the item, must cow */
5343                         ret = 0;
5344                         goto out;
5345                 }
5346                 slot--;
5347         }
5348         ret = 0;
5349         leaf = path->nodes[0];
5350         btrfs_item_key_to_cpu(leaf, &key, slot);
5351         if (key.objectid != btrfs_ino(inode) ||
5352             key.type != BTRFS_EXTENT_DATA_KEY) {
5353                 /* not our file or wrong item type, must cow */
5354                 goto out;
5355         }
5356
5357         if (key.offset > offset) {
5358                 /* Wrong offset, must cow */
5359                 goto out;
5360         }
5361
5362         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5363         found_type = btrfs_file_extent_type(leaf, fi);
5364         if (found_type != BTRFS_FILE_EXTENT_REG &&
5365             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5366                 /* not a regular extent, must cow */
5367                 goto out;
5368         }
5369         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5370         backref_offset = btrfs_file_extent_offset(leaf, fi);
5371
5372         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5373         if (extent_end < offset + len) {
5374                 /* extent doesn't include our full range, must cow */
5375                 goto out;
5376         }
5377
5378         if (btrfs_extent_readonly(root, disk_bytenr))
5379                 goto out;
5380
5381         /*
5382          * look for other files referencing this extent, if we
5383          * find any we must cow
5384          */
5385         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5386                                   key.offset - backref_offset, disk_bytenr))
5387                 goto out;
5388
5389         /*
5390          * adjust disk_bytenr and num_bytes to cover just the bytes
5391          * in this extent we are about to write.  If there
5392          * are any csums in that range we have to cow in order
5393          * to keep the csums correct
5394          */
5395         disk_bytenr += backref_offset;
5396         disk_bytenr += offset - key.offset;
5397         num_bytes = min(offset + len, extent_end) - offset;
5398         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5399                                 goto out;
5400         /*
5401          * all of the above have passed, it is safe to overwrite this extent
5402          * without cow
5403          */
5404         ret = 1;
5405 out:
5406         btrfs_free_path(path);
5407         return ret;
5408 }
5409
5410 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5411                                    struct buffer_head *bh_result, int create)
5412 {
5413         struct extent_map *em;
5414         struct btrfs_root *root = BTRFS_I(inode)->root;
5415         u64 start = iblock << inode->i_blkbits;
5416         u64 len = bh_result->b_size;
5417         struct btrfs_trans_handle *trans;
5418
5419         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5420         if (IS_ERR(em))
5421                 return PTR_ERR(em);
5422
5423         /*
5424          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5425          * io.  INLINE is special, and we could probably kludge it in here, but
5426          * it's still buffered so for safety lets just fall back to the generic
5427          * buffered path.
5428          *
5429          * For COMPRESSED we _have_ to read the entire extent in so we can
5430          * decompress it, so there will be buffering required no matter what we
5431          * do, so go ahead and fallback to buffered.
5432          *
5433          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5434          * to buffered IO.  Don't blame me, this is the price we pay for using
5435          * the generic code.
5436          */
5437         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5438             em->block_start == EXTENT_MAP_INLINE) {
5439                 free_extent_map(em);
5440                 return -ENOTBLK;
5441         }
5442
5443         /* Just a good old fashioned hole, return */
5444         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5445                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5446                 free_extent_map(em);
5447                 /* DIO will do one hole at a time, so just unlock a sector */
5448                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5449                               start + root->sectorsize - 1, GFP_NOFS);
5450                 return 0;
5451         }
5452
5453         /*
5454          * We don't allocate a new extent in the following cases
5455          *
5456          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5457          * existing extent.
5458          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5459          * just use the extent.
5460          *
5461          */
5462         if (!create) {
5463                 len = em->len - (start - em->start);
5464                 goto map;
5465         }
5466
5467         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5468             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5469              em->block_start != EXTENT_MAP_HOLE)) {
5470                 int type;
5471                 int ret;
5472                 u64 block_start;
5473
5474                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5475                         type = BTRFS_ORDERED_PREALLOC;
5476                 else
5477                         type = BTRFS_ORDERED_NOCOW;
5478                 len = min(len, em->len - (start - em->start));
5479                 block_start = em->block_start + (start - em->start);
5480
5481                 /*
5482                  * we're not going to log anything, but we do need
5483                  * to make sure the current transaction stays open
5484                  * while we look for nocow cross refs
5485                  */
5486                 trans = btrfs_join_transaction(root);
5487                 if (IS_ERR(trans))
5488                         goto must_cow;
5489
5490                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5491                         ret = btrfs_add_ordered_extent_dio(inode, start,
5492                                            block_start, len, len, type);
5493                         btrfs_end_transaction(trans, root);
5494                         if (ret) {
5495                                 free_extent_map(em);
5496                                 return ret;
5497                         }
5498                         goto unlock;
5499                 }
5500                 btrfs_end_transaction(trans, root);
5501         }
5502 must_cow:
5503         /*
5504          * this will cow the extent, reset the len in case we changed
5505          * it above
5506          */
5507         len = bh_result->b_size;
5508         em = btrfs_new_extent_direct(inode, em, start, len);
5509         if (IS_ERR(em))
5510                 return PTR_ERR(em);
5511         len = min(len, em->len - (start - em->start));
5512 unlock:
5513         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5514                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5515                           0, NULL, GFP_NOFS);
5516 map:
5517         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5518                 inode->i_blkbits;
5519         bh_result->b_size = len;
5520         bh_result->b_bdev = em->bdev;
5521         set_buffer_mapped(bh_result);
5522         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5523                 set_buffer_new(bh_result);
5524
5525         free_extent_map(em);
5526
5527         return 0;
5528 }
5529
5530 struct btrfs_dio_private {
5531         struct inode *inode;
5532         u64 logical_offset;
5533         u64 disk_bytenr;
5534         u64 bytes;
5535         u32 *csums;
5536         void *private;
5537
5538         /* number of bios pending for this dio */
5539         atomic_t pending_bios;
5540
5541         /* IO errors */
5542         int errors;
5543
5544         struct bio *orig_bio;
5545 };
5546
5547 static void btrfs_endio_direct_read(struct bio *bio, int err)
5548 {
5549         struct btrfs_dio_private *dip = bio->bi_private;
5550         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5551         struct bio_vec *bvec = bio->bi_io_vec;
5552         struct inode *inode = dip->inode;
5553         struct btrfs_root *root = BTRFS_I(inode)->root;
5554         u64 start;
5555         u32 *private = dip->csums;
5556
5557         start = dip->logical_offset;
5558         do {
5559                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5560                         struct page *page = bvec->bv_page;
5561                         char *kaddr;
5562                         u32 csum = ~(u32)0;
5563                         unsigned long flags;
5564
5565                         local_irq_save(flags);
5566                         kaddr = kmap_atomic(page, KM_IRQ0);
5567                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5568                                                csum, bvec->bv_len);
5569                         btrfs_csum_final(csum, (char *)&csum);
5570                         kunmap_atomic(kaddr, KM_IRQ0);
5571                         local_irq_restore(flags);
5572
5573                         flush_dcache_page(bvec->bv_page);
5574                         if (csum != *private) {
5575                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5576                                       " %llu csum %u private %u\n",
5577                                       (unsigned long long)btrfs_ino(inode),
5578                                       (unsigned long long)start,
5579                                       csum, *private);
5580                                 err = -EIO;
5581                         }
5582                 }
5583
5584                 start += bvec->bv_len;
5585                 private++;
5586                 bvec++;
5587         } while (bvec <= bvec_end);
5588
5589         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5590                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5591         bio->bi_private = dip->private;
5592
5593         kfree(dip->csums);
5594         kfree(dip);
5595
5596         /* If we had a csum failure make sure to clear the uptodate flag */
5597         if (err)
5598                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5599         dio_end_io(bio, err);
5600 }
5601
5602 static void btrfs_endio_direct_write(struct bio *bio, int err)
5603 {
5604         struct btrfs_dio_private *dip = bio->bi_private;
5605         struct inode *inode = dip->inode;
5606         struct btrfs_root *root = BTRFS_I(inode)->root;
5607         struct btrfs_trans_handle *trans;
5608         struct btrfs_ordered_extent *ordered = NULL;
5609         struct extent_state *cached_state = NULL;
5610         u64 ordered_offset = dip->logical_offset;
5611         u64 ordered_bytes = dip->bytes;
5612         int ret;
5613
5614         if (err)
5615                 goto out_done;
5616 again:
5617         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5618                                                    &ordered_offset,
5619                                                    ordered_bytes);
5620         if (!ret)
5621                 goto out_test;
5622
5623         BUG_ON(!ordered);
5624
5625         trans = btrfs_join_transaction(root);
5626         if (IS_ERR(trans)) {
5627                 err = -ENOMEM;
5628                 goto out;
5629         }
5630         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5631
5632         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5633                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5634                 if (!ret)
5635                         err = btrfs_update_inode(trans, root, inode);
5636                 goto out;
5637         }
5638
5639         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5640                          ordered->file_offset + ordered->len - 1, 0,
5641                          &cached_state, GFP_NOFS);
5642
5643         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5644                 ret = btrfs_mark_extent_written(trans, inode,
5645                                                 ordered->file_offset,
5646                                                 ordered->file_offset +
5647                                                 ordered->len);
5648                 if (ret) {
5649                         err = ret;
5650                         goto out_unlock;
5651                 }
5652         } else {
5653                 ret = insert_reserved_file_extent(trans, inode,
5654                                                   ordered->file_offset,
5655                                                   ordered->start,
5656                                                   ordered->disk_len,
5657                                                   ordered->len,
5658                                                   ordered->len,
5659                                                   0, 0, 0,
5660                                                   BTRFS_FILE_EXTENT_REG);
5661                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5662                                    ordered->file_offset, ordered->len);
5663                 if (ret) {
5664                         err = ret;
5665                         WARN_ON(1);
5666                         goto out_unlock;
5667                 }
5668         }
5669
5670         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5671         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5672         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5673                 btrfs_update_inode(trans, root, inode);
5674         ret = 0;
5675 out_unlock:
5676         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5677                              ordered->file_offset + ordered->len - 1,
5678                              &cached_state, GFP_NOFS);
5679 out:
5680         btrfs_delalloc_release_metadata(inode, ordered->len);
5681         btrfs_end_transaction(trans, root);
5682         ordered_offset = ordered->file_offset + ordered->len;
5683         btrfs_put_ordered_extent(ordered);
5684         btrfs_put_ordered_extent(ordered);
5685
5686 out_test:
5687         /*
5688          * our bio might span multiple ordered extents.  If we haven't
5689          * completed the accounting for the whole dio, go back and try again
5690          */
5691         if (ordered_offset < dip->logical_offset + dip->bytes) {
5692                 ordered_bytes = dip->logical_offset + dip->bytes -
5693                         ordered_offset;
5694                 goto again;
5695         }
5696 out_done:
5697         bio->bi_private = dip->private;
5698
5699         kfree(dip->csums);
5700         kfree(dip);
5701
5702         /* If we had an error make sure to clear the uptodate flag */
5703         if (err)
5704                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5705         dio_end_io(bio, err);
5706 }
5707
5708 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5709                                     struct bio *bio, int mirror_num,
5710                                     unsigned long bio_flags, u64 offset)
5711 {
5712         int ret;
5713         struct btrfs_root *root = BTRFS_I(inode)->root;
5714         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5715         BUG_ON(ret);
5716         return 0;
5717 }
5718
5719 static void btrfs_end_dio_bio(struct bio *bio, int err)
5720 {
5721         struct btrfs_dio_private *dip = bio->bi_private;
5722
5723         if (err) {
5724                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5725                       "sector %#Lx len %u err no %d\n",
5726                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5727                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5728                 dip->errors = 1;
5729
5730                 /*
5731                  * before atomic variable goto zero, we must make sure
5732                  * dip->errors is perceived to be set.
5733                  */
5734                 smp_mb__before_atomic_dec();
5735         }
5736
5737         /* if there are more bios still pending for this dio, just exit */
5738         if (!atomic_dec_and_test(&dip->pending_bios))
5739                 goto out;
5740
5741         if (dip->errors)
5742                 bio_io_error(dip->orig_bio);
5743         else {
5744                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5745                 bio_endio(dip->orig_bio, 0);
5746         }
5747 out:
5748         bio_put(bio);
5749 }
5750
5751 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5752                                        u64 first_sector, gfp_t gfp_flags)
5753 {
5754         int nr_vecs = bio_get_nr_vecs(bdev);
5755         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5756 }
5757
5758 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5759                                          int rw, u64 file_offset, int skip_sum,
5760                                          u32 *csums, int async_submit)
5761 {
5762         int write = rw & REQ_WRITE;
5763         struct btrfs_root *root = BTRFS_I(inode)->root;
5764         int ret;
5765
5766         bio_get(bio);
5767         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5768         if (ret)
5769                 goto err;
5770
5771         if (skip_sum)
5772                 goto map;
5773
5774         if (write && async_submit) {
5775                 ret = btrfs_wq_submit_bio(root->fs_info,
5776                                    inode, rw, bio, 0, 0,
5777                                    file_offset,
5778                                    __btrfs_submit_bio_start_direct_io,
5779                                    __btrfs_submit_bio_done);
5780                 goto err;
5781         } else if (write) {
5782                 /*
5783                  * If we aren't doing async submit, calculate the csum of the
5784                  * bio now.
5785                  */
5786                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5787                 if (ret)
5788                         goto err;
5789         } else if (!skip_sum) {
5790                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5791                                           file_offset, csums);
5792                 if (ret)
5793                         goto err;
5794         }
5795
5796 map:
5797         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5798 err:
5799         bio_put(bio);
5800         return ret;
5801 }
5802
5803 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5804                                     int skip_sum)
5805 {
5806         struct inode *inode = dip->inode;
5807         struct btrfs_root *root = BTRFS_I(inode)->root;
5808         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5809         struct bio *bio;
5810         struct bio *orig_bio = dip->orig_bio;
5811         struct bio_vec *bvec = orig_bio->bi_io_vec;
5812         u64 start_sector = orig_bio->bi_sector;
5813         u64 file_offset = dip->logical_offset;
5814         u64 submit_len = 0;
5815         u64 map_length;
5816         int nr_pages = 0;
5817         u32 *csums = dip->csums;
5818         int ret = 0;
5819         int async_submit = 0;
5820         int write = rw & REQ_WRITE;
5821
5822         map_length = orig_bio->bi_size;
5823         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5824                               &map_length, NULL, 0);
5825         if (ret) {
5826                 bio_put(orig_bio);
5827                 return -EIO;
5828         }
5829
5830         if (map_length >= orig_bio->bi_size) {
5831                 bio = orig_bio;
5832                 goto submit;
5833         }
5834
5835         async_submit = 1;
5836         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5837         if (!bio)
5838                 return -ENOMEM;
5839         bio->bi_private = dip;
5840         bio->bi_end_io = btrfs_end_dio_bio;
5841         atomic_inc(&dip->pending_bios);
5842
5843         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5844                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5845                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5846                                  bvec->bv_offset) < bvec->bv_len)) {
5847                         /*
5848                          * inc the count before we submit the bio so
5849                          * we know the end IO handler won't happen before
5850                          * we inc the count. Otherwise, the dip might get freed
5851                          * before we're done setting it up
5852                          */
5853                         atomic_inc(&dip->pending_bios);
5854                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5855                                                      file_offset, skip_sum,
5856                                                      csums, async_submit);
5857                         if (ret) {
5858                                 bio_put(bio);
5859                                 atomic_dec(&dip->pending_bios);
5860                                 goto out_err;
5861                         }
5862
5863                         /* Write's use the ordered csums */
5864                         if (!write && !skip_sum)
5865                                 csums = csums + nr_pages;
5866                         start_sector += submit_len >> 9;
5867                         file_offset += submit_len;
5868
5869                         submit_len = 0;
5870                         nr_pages = 0;
5871
5872                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5873                                                   start_sector, GFP_NOFS);
5874                         if (!bio)
5875                                 goto out_err;
5876                         bio->bi_private = dip;
5877                         bio->bi_end_io = btrfs_end_dio_bio;
5878
5879                         map_length = orig_bio->bi_size;
5880                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5881                                               &map_length, NULL, 0);
5882                         if (ret) {
5883                                 bio_put(bio);
5884                                 goto out_err;
5885                         }
5886                 } else {
5887                         submit_len += bvec->bv_len;
5888                         nr_pages ++;
5889                         bvec++;
5890                 }
5891         }
5892
5893 submit:
5894         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5895                                      csums, async_submit);
5896         if (!ret)
5897                 return 0;
5898
5899         bio_put(bio);
5900 out_err:
5901         dip->errors = 1;
5902         /*
5903          * before atomic variable goto zero, we must
5904          * make sure dip->errors is perceived to be set.
5905          */
5906         smp_mb__before_atomic_dec();
5907         if (atomic_dec_and_test(&dip->pending_bios))
5908                 bio_io_error(dip->orig_bio);
5909
5910         /* bio_end_io() will handle error, so we needn't return it */
5911         return 0;
5912 }
5913
5914 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5915                                 loff_t file_offset)
5916 {
5917         struct btrfs_root *root = BTRFS_I(inode)->root;
5918         struct btrfs_dio_private *dip;
5919         struct bio_vec *bvec = bio->bi_io_vec;
5920         int skip_sum;
5921         int write = rw & REQ_WRITE;
5922         int ret = 0;
5923
5924         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5925
5926         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5927         if (!dip) {
5928                 ret = -ENOMEM;
5929                 goto free_ordered;
5930         }
5931         dip->csums = NULL;
5932
5933         /* Write's use the ordered csum stuff, so we don't need dip->csums */
5934         if (!write && !skip_sum) {
5935                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5936                 if (!dip->csums) {
5937                         kfree(dip);
5938                         ret = -ENOMEM;
5939                         goto free_ordered;
5940                 }
5941         }
5942
5943         dip->private = bio->bi_private;
5944         dip->inode = inode;
5945         dip->logical_offset = file_offset;
5946
5947         dip->bytes = 0;
5948         do {
5949                 dip->bytes += bvec->bv_len;
5950                 bvec++;
5951         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5952
5953         dip->disk_bytenr = (u64)bio->bi_sector << 9;
5954         bio->bi_private = dip;
5955         dip->errors = 0;
5956         dip->orig_bio = bio;
5957         atomic_set(&dip->pending_bios, 0);
5958
5959         if (write)
5960                 bio->bi_end_io = btrfs_endio_direct_write;
5961         else
5962                 bio->bi_end_io = btrfs_endio_direct_read;
5963
5964         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5965         if (!ret)
5966                 return;
5967 free_ordered:
5968         /*
5969          * If this is a write, we need to clean up the reserved space and kill
5970          * the ordered extent.
5971          */
5972         if (write) {
5973                 struct btrfs_ordered_extent *ordered;
5974                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
5975                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
5976                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
5977                         btrfs_free_reserved_extent(root, ordered->start,
5978                                                    ordered->disk_len);
5979                 btrfs_put_ordered_extent(ordered);
5980                 btrfs_put_ordered_extent(ordered);
5981         }
5982         bio_endio(bio, ret);
5983 }
5984
5985 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
5986                         const struct iovec *iov, loff_t offset,
5987                         unsigned long nr_segs)
5988 {
5989         int seg;
5990         int i;
5991         size_t size;
5992         unsigned long addr;
5993         unsigned blocksize_mask = root->sectorsize - 1;
5994         ssize_t retval = -EINVAL;
5995         loff_t end = offset;
5996
5997         if (offset & blocksize_mask)
5998                 goto out;
5999
6000         /* Check the memory alignment.  Blocks cannot straddle pages */
6001         for (seg = 0; seg < nr_segs; seg++) {
6002                 addr = (unsigned long)iov[seg].iov_base;
6003                 size = iov[seg].iov_len;
6004                 end += size;
6005                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6006                         goto out;
6007
6008                 /* If this is a write we don't need to check anymore */
6009                 if (rw & WRITE)
6010                         continue;
6011
6012                 /*
6013                  * Check to make sure we don't have duplicate iov_base's in this
6014                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6015                  * when reading back.
6016                  */
6017                 for (i = seg + 1; i < nr_segs; i++) {
6018                         if (iov[seg].iov_base == iov[i].iov_base)
6019                                 goto out;
6020                 }
6021         }
6022         retval = 0;
6023 out:
6024         return retval;
6025 }
6026 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6027                         const struct iovec *iov, loff_t offset,
6028                         unsigned long nr_segs)
6029 {
6030         struct file *file = iocb->ki_filp;
6031         struct inode *inode = file->f_mapping->host;
6032         struct btrfs_ordered_extent *ordered;
6033         struct extent_state *cached_state = NULL;
6034         u64 lockstart, lockend;
6035         ssize_t ret;
6036         int writing = rw & WRITE;
6037         int write_bits = 0;
6038         size_t count = iov_length(iov, nr_segs);
6039
6040         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6041                             offset, nr_segs)) {
6042                 return 0;
6043         }
6044
6045         lockstart = offset;
6046         lockend = offset + count - 1;
6047
6048         if (writing) {
6049                 ret = btrfs_delalloc_reserve_space(inode, count);
6050                 if (ret)
6051                         goto out;
6052         }
6053
6054         while (1) {
6055                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6056                                  0, &cached_state, GFP_NOFS);
6057                 /*
6058                  * We're concerned with the entire range that we're going to be
6059                  * doing DIO to, so we need to make sure theres no ordered
6060                  * extents in this range.
6061                  */
6062                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6063                                                      lockend - lockstart + 1);
6064                 if (!ordered)
6065                         break;
6066                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6067                                      &cached_state, GFP_NOFS);
6068                 btrfs_start_ordered_extent(inode, ordered, 1);
6069                 btrfs_put_ordered_extent(ordered);
6070                 cond_resched();
6071         }
6072
6073         /*
6074          * we don't use btrfs_set_extent_delalloc because we don't want
6075          * the dirty or uptodate bits
6076          */
6077         if (writing) {
6078                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6079                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6080                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6081                                      GFP_NOFS);
6082                 if (ret) {
6083                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6084                                          lockend, EXTENT_LOCKED | write_bits,
6085                                          1, 0, &cached_state, GFP_NOFS);
6086                         goto out;
6087                 }
6088         }
6089
6090         free_extent_state(cached_state);
6091         cached_state = NULL;
6092
6093         ret = __blockdev_direct_IO(rw, iocb, inode,
6094                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6095                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6096                    btrfs_submit_direct, 0);
6097
6098         if (ret < 0 && ret != -EIOCBQUEUED) {
6099                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6100                               offset + iov_length(iov, nr_segs) - 1,
6101                               EXTENT_LOCKED | write_bits, 1, 0,
6102                               &cached_state, GFP_NOFS);
6103         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6104                 /*
6105                  * We're falling back to buffered, unlock the section we didn't
6106                  * do IO on.
6107                  */
6108                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6109                               offset + iov_length(iov, nr_segs) - 1,
6110                               EXTENT_LOCKED | write_bits, 1, 0,
6111                               &cached_state, GFP_NOFS);
6112         }
6113 out:
6114         free_extent_state(cached_state);
6115         return ret;
6116 }
6117
6118 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6119                 __u64 start, __u64 len)
6120 {
6121         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6122 }
6123
6124 int btrfs_readpage(struct file *file, struct page *page)
6125 {
6126         struct extent_io_tree *tree;
6127         tree = &BTRFS_I(page->mapping->host)->io_tree;
6128         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6129 }
6130
6131 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6132 {
6133         struct extent_io_tree *tree;
6134
6135
6136         if (current->flags & PF_MEMALLOC) {
6137                 redirty_page_for_writepage(wbc, page);
6138                 unlock_page(page);
6139                 return 0;
6140         }
6141         tree = &BTRFS_I(page->mapping->host)->io_tree;
6142         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6143 }
6144
6145 int btrfs_writepages(struct address_space *mapping,
6146                      struct writeback_control *wbc)
6147 {
6148         struct extent_io_tree *tree;
6149
6150         tree = &BTRFS_I(mapping->host)->io_tree;
6151         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6152 }
6153
6154 static int
6155 btrfs_readpages(struct file *file, struct address_space *mapping,
6156                 struct list_head *pages, unsigned nr_pages)
6157 {
6158         struct extent_io_tree *tree;
6159         tree = &BTRFS_I(mapping->host)->io_tree;
6160         return extent_readpages(tree, mapping, pages, nr_pages,
6161                                 btrfs_get_extent);
6162 }
6163 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6164 {
6165         struct extent_io_tree *tree;
6166         struct extent_map_tree *map;
6167         int ret;
6168
6169         tree = &BTRFS_I(page->mapping->host)->io_tree;
6170         map = &BTRFS_I(page->mapping->host)->extent_tree;
6171         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6172         if (ret == 1) {
6173                 ClearPagePrivate(page);
6174                 set_page_private(page, 0);
6175                 page_cache_release(page);
6176         }
6177         return ret;
6178 }
6179
6180 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6181 {
6182         if (PageWriteback(page) || PageDirty(page))
6183                 return 0;
6184         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6185 }
6186
6187 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6188 {
6189         struct extent_io_tree *tree;
6190         struct btrfs_ordered_extent *ordered;
6191         struct extent_state *cached_state = NULL;
6192         u64 page_start = page_offset(page);
6193         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6194
6195
6196         /*
6197          * we have the page locked, so new writeback can't start,
6198          * and the dirty bit won't be cleared while we are here.
6199          *
6200          * Wait for IO on this page so that we can safely clear
6201          * the PagePrivate2 bit and do ordered accounting
6202          */
6203         wait_on_page_writeback(page);
6204
6205         tree = &BTRFS_I(page->mapping->host)->io_tree;
6206         if (offset) {
6207                 btrfs_releasepage(page, GFP_NOFS);
6208                 return;
6209         }
6210         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6211                          GFP_NOFS);
6212         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6213                                            page_offset(page));
6214         if (ordered) {
6215                 /*
6216                  * IO on this page will never be started, so we need
6217                  * to account for any ordered extents now
6218                  */
6219                 clear_extent_bit(tree, page_start, page_end,
6220                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6221                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6222                                  &cached_state, GFP_NOFS);
6223                 /*
6224                  * whoever cleared the private bit is responsible
6225                  * for the finish_ordered_io
6226                  */
6227                 if (TestClearPagePrivate2(page)) {
6228                         btrfs_finish_ordered_io(page->mapping->host,
6229                                                 page_start, page_end);
6230                 }
6231                 btrfs_put_ordered_extent(ordered);
6232                 cached_state = NULL;
6233                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6234                                  GFP_NOFS);
6235         }
6236         clear_extent_bit(tree, page_start, page_end,
6237                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6238                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6239         __btrfs_releasepage(page, GFP_NOFS);
6240
6241         ClearPageChecked(page);
6242         if (PagePrivate(page)) {
6243                 ClearPagePrivate(page);
6244                 set_page_private(page, 0);
6245                 page_cache_release(page);
6246         }
6247 }
6248
6249 /*
6250  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6251  * called from a page fault handler when a page is first dirtied. Hence we must
6252  * be careful to check for EOF conditions here. We set the page up correctly
6253  * for a written page which means we get ENOSPC checking when writing into
6254  * holes and correct delalloc and unwritten extent mapping on filesystems that
6255  * support these features.
6256  *
6257  * We are not allowed to take the i_mutex here so we have to play games to
6258  * protect against truncate races as the page could now be beyond EOF.  Because
6259  * vmtruncate() writes the inode size before removing pages, once we have the
6260  * page lock we can determine safely if the page is beyond EOF. If it is not
6261  * beyond EOF, then the page is guaranteed safe against truncation until we
6262  * unlock the page.
6263  */
6264 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6265 {
6266         struct page *page = vmf->page;
6267         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6268         struct btrfs_root *root = BTRFS_I(inode)->root;
6269         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6270         struct btrfs_ordered_extent *ordered;
6271         struct extent_state *cached_state = NULL;
6272         char *kaddr;
6273         unsigned long zero_start;
6274         loff_t size;
6275         int ret;
6276         u64 page_start;
6277         u64 page_end;
6278
6279         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6280         if (ret) {
6281                 if (ret == -ENOMEM)
6282                         ret = VM_FAULT_OOM;
6283                 else /* -ENOSPC, -EIO, etc */
6284                         ret = VM_FAULT_SIGBUS;
6285                 goto out;
6286         }
6287
6288         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6289 again:
6290         lock_page(page);
6291         size = i_size_read(inode);
6292         page_start = page_offset(page);
6293         page_end = page_start + PAGE_CACHE_SIZE - 1;
6294
6295         if ((page->mapping != inode->i_mapping) ||
6296             (page_start >= size)) {
6297                 /* page got truncated out from underneath us */
6298                 goto out_unlock;
6299         }
6300         wait_on_page_writeback(page);
6301
6302         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6303                          GFP_NOFS);
6304         set_page_extent_mapped(page);
6305
6306         /*
6307          * we can't set the delalloc bits if there are pending ordered
6308          * extents.  Drop our locks and wait for them to finish
6309          */
6310         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6311         if (ordered) {
6312                 unlock_extent_cached(io_tree, page_start, page_end,
6313                                      &cached_state, GFP_NOFS);
6314                 unlock_page(page);
6315                 btrfs_start_ordered_extent(inode, ordered, 1);
6316                 btrfs_put_ordered_extent(ordered);
6317                 goto again;
6318         }
6319
6320         /*
6321          * XXX - page_mkwrite gets called every time the page is dirtied, even
6322          * if it was already dirty, so for space accounting reasons we need to
6323          * clear any delalloc bits for the range we are fixing to save.  There
6324          * is probably a better way to do this, but for now keep consistent with
6325          * prepare_pages in the normal write path.
6326          */
6327         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6328                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6329                           0, 0, &cached_state, GFP_NOFS);
6330
6331         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6332                                         &cached_state);
6333         if (ret) {
6334                 unlock_extent_cached(io_tree, page_start, page_end,
6335                                      &cached_state, GFP_NOFS);
6336                 ret = VM_FAULT_SIGBUS;
6337                 goto out_unlock;
6338         }
6339         ret = 0;
6340
6341         /* page is wholly or partially inside EOF */
6342         if (page_start + PAGE_CACHE_SIZE > size)
6343                 zero_start = size & ~PAGE_CACHE_MASK;
6344         else
6345                 zero_start = PAGE_CACHE_SIZE;
6346
6347         if (zero_start != PAGE_CACHE_SIZE) {
6348                 kaddr = kmap(page);
6349                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6350                 flush_dcache_page(page);
6351                 kunmap(page);
6352         }
6353         ClearPageChecked(page);
6354         set_page_dirty(page);
6355         SetPageUptodate(page);
6356
6357         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6358         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6359
6360         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6361
6362 out_unlock:
6363         if (!ret)
6364                 return VM_FAULT_LOCKED;
6365         unlock_page(page);
6366         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6367 out:
6368         return ret;
6369 }
6370
6371 static int btrfs_truncate(struct inode *inode)
6372 {
6373         struct btrfs_root *root = BTRFS_I(inode)->root;
6374         struct btrfs_block_rsv *rsv;
6375         int ret;
6376         int err = 0;
6377         struct btrfs_trans_handle *trans;
6378         unsigned long nr;
6379         u64 mask = root->sectorsize - 1;
6380         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6381
6382         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6383         if (ret)
6384                 return ret;
6385
6386         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6387         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6388
6389         /*
6390          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6391          * 3 things going on here
6392          *
6393          * 1) We need to reserve space for our orphan item and the space to
6394          * delete our orphan item.  Lord knows we don't want to have a dangling
6395          * orphan item because we didn't reserve space to remove it.
6396          *
6397          * 2) We need to reserve space to update our inode.
6398          *
6399          * 3) We need to have something to cache all the space that is going to
6400          * be free'd up by the truncate operation, but also have some slack
6401          * space reserved in case it uses space during the truncate (thank you
6402          * very much snapshotting).
6403          *
6404          * And we need these to all be seperate.  The fact is we can use alot of
6405          * space doing the truncate, and we have no earthly idea how much space
6406          * we will use, so we need the truncate reservation to be seperate so it
6407          * doesn't end up using space reserved for updating the inode or
6408          * removing the orphan item.  We also need to be able to stop the
6409          * transaction and start a new one, which means we need to be able to
6410          * update the inode several times, and we have no idea of knowing how
6411          * many times that will be, so we can't just reserve 1 item for the
6412          * entirety of the opration, so that has to be done seperately as well.
6413          * Then there is the orphan item, which does indeed need to be held on
6414          * to for the whole operation, and we need nobody to touch this reserved
6415          * space except the orphan code.
6416          *
6417          * So that leaves us with
6418          *
6419          * 1) root->orphan_block_rsv - for the orphan deletion.
6420          * 2) rsv - for the truncate reservation, which we will steal from the
6421          * transaction reservation.
6422          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6423          * updating the inode.
6424          */
6425         rsv = btrfs_alloc_block_rsv(root);
6426         if (!rsv)
6427                 return -ENOMEM;
6428         rsv->size = min_size;
6429
6430         /*
6431          * 1 for the truncate slack space
6432          * 1 for the orphan item we're going to add
6433          * 1 for the orphan item deletion
6434          * 1 for updating the inode.
6435          */
6436         trans = btrfs_start_transaction(root, 4);
6437         if (IS_ERR(trans)) {
6438                 err = PTR_ERR(trans);
6439                 goto out;
6440         }
6441
6442         /* Migrate the slack space for the truncate to our reserve */
6443         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6444                                       min_size);
6445         BUG_ON(ret);
6446
6447         ret = btrfs_orphan_add(trans, inode);
6448         if (ret) {
6449                 btrfs_end_transaction(trans, root);
6450                 goto out;
6451         }
6452
6453         /*
6454          * setattr is responsible for setting the ordered_data_close flag,
6455          * but that is only tested during the last file release.  That
6456          * could happen well after the next commit, leaving a great big
6457          * window where new writes may get lost if someone chooses to write
6458          * to this file after truncating to zero
6459          *
6460          * The inode doesn't have any dirty data here, and so if we commit
6461          * this is a noop.  If someone immediately starts writing to the inode
6462          * it is very likely we'll catch some of their writes in this
6463          * transaction, and the commit will find this file on the ordered
6464          * data list with good things to send down.
6465          *
6466          * This is a best effort solution, there is still a window where
6467          * using truncate to replace the contents of the file will
6468          * end up with a zero length file after a crash.
6469          */
6470         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6471                 btrfs_add_ordered_operation(trans, root, inode);
6472
6473         while (1) {
6474                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6475                 if (ret) {
6476                         /*
6477                          * This can only happen with the original transaction we
6478                          * started above, every other time we shouldn't have a
6479                          * transaction started yet.
6480                          */
6481                         if (ret == -EAGAIN)
6482                                 goto end_trans;
6483                         err = ret;
6484                         break;
6485                 }
6486
6487                 if (!trans) {
6488                         /* Just need the 1 for updating the inode */
6489                         trans = btrfs_start_transaction(root, 1);
6490                         if (IS_ERR(trans)) {
6491                                 err = PTR_ERR(trans);
6492                                 goto out;
6493                         }
6494                 }
6495
6496                 trans->block_rsv = rsv;
6497
6498                 ret = btrfs_truncate_inode_items(trans, root, inode,
6499                                                  inode->i_size,
6500                                                  BTRFS_EXTENT_DATA_KEY);
6501                 if (ret != -EAGAIN) {
6502                         err = ret;
6503                         break;
6504                 }
6505
6506                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6507                 ret = btrfs_update_inode(trans, root, inode);
6508                 if (ret) {
6509                         err = ret;
6510                         break;
6511                 }
6512 end_trans:
6513                 nr = trans->blocks_used;
6514                 btrfs_end_transaction(trans, root);
6515                 trans = NULL;
6516                 btrfs_btree_balance_dirty(root, nr);
6517         }
6518
6519         if (ret == 0 && inode->i_nlink > 0) {
6520                 trans->block_rsv = root->orphan_block_rsv;
6521                 ret = btrfs_orphan_del(trans, inode);
6522                 if (ret)
6523                         err = ret;
6524         } else if (ret && inode->i_nlink > 0) {
6525                 /*
6526                  * Failed to do the truncate, remove us from the in memory
6527                  * orphan list.
6528                  */
6529                 ret = btrfs_orphan_del(NULL, inode);
6530         }
6531
6532         trans->block_rsv = &root->fs_info->trans_block_rsv;
6533         ret = btrfs_update_inode(trans, root, inode);
6534         if (ret && !err)
6535                 err = ret;
6536
6537         nr = trans->blocks_used;
6538         ret = btrfs_end_transaction_throttle(trans, root);
6539         btrfs_btree_balance_dirty(root, nr);
6540
6541 out:
6542         btrfs_free_block_rsv(root, rsv);
6543
6544         if (ret && !err)
6545                 err = ret;
6546
6547         return err;
6548 }
6549
6550 /*
6551  * create a new subvolume directory/inode (helper for the ioctl).
6552  */
6553 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6554                              struct btrfs_root *new_root, u64 new_dirid)
6555 {
6556         struct inode *inode;
6557         int err;
6558         u64 index = 0;
6559
6560         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6561                                 new_dirid, S_IFDIR | 0700, &index);
6562         if (IS_ERR(inode))
6563                 return PTR_ERR(inode);
6564         inode->i_op = &btrfs_dir_inode_operations;
6565         inode->i_fop = &btrfs_dir_file_operations;
6566
6567         set_nlink(inode, 1);
6568         btrfs_i_size_write(inode, 0);
6569
6570         err = btrfs_update_inode(trans, new_root, inode);
6571         BUG_ON(err);
6572
6573         iput(inode);
6574         return 0;
6575 }
6576
6577 struct inode *btrfs_alloc_inode(struct super_block *sb)
6578 {
6579         struct btrfs_inode *ei;
6580         struct inode *inode;
6581
6582         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6583         if (!ei)
6584                 return NULL;
6585
6586         ei->root = NULL;
6587         ei->space_info = NULL;
6588         ei->generation = 0;
6589         ei->sequence = 0;
6590         ei->last_trans = 0;
6591         ei->last_sub_trans = 0;
6592         ei->logged_trans = 0;
6593         ei->delalloc_bytes = 0;
6594         ei->disk_i_size = 0;
6595         ei->flags = 0;
6596         ei->csum_bytes = 0;
6597         ei->index_cnt = (u64)-1;
6598         ei->last_unlink_trans = 0;
6599
6600         spin_lock_init(&ei->lock);
6601         ei->outstanding_extents = 0;
6602         ei->reserved_extents = 0;
6603
6604         ei->ordered_data_close = 0;
6605         ei->orphan_meta_reserved = 0;
6606         ei->dummy_inode = 0;
6607         ei->in_defrag = 0;
6608         ei->force_compress = BTRFS_COMPRESS_NONE;
6609
6610         ei->delayed_node = NULL;
6611
6612         inode = &ei->vfs_inode;
6613         extent_map_tree_init(&ei->extent_tree);
6614         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6615         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6616         mutex_init(&ei->log_mutex);
6617         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6618         INIT_LIST_HEAD(&ei->i_orphan);
6619         INIT_LIST_HEAD(&ei->delalloc_inodes);
6620         INIT_LIST_HEAD(&ei->ordered_operations);
6621         RB_CLEAR_NODE(&ei->rb_node);
6622
6623         return inode;
6624 }
6625
6626 static void btrfs_i_callback(struct rcu_head *head)
6627 {
6628         struct inode *inode = container_of(head, struct inode, i_rcu);
6629         INIT_LIST_HEAD(&inode->i_dentry);
6630         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6631 }
6632
6633 void btrfs_destroy_inode(struct inode *inode)
6634 {
6635         struct btrfs_ordered_extent *ordered;
6636         struct btrfs_root *root = BTRFS_I(inode)->root;
6637
6638         WARN_ON(!list_empty(&inode->i_dentry));
6639         WARN_ON(inode->i_data.nrpages);
6640         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6641         WARN_ON(BTRFS_I(inode)->reserved_extents);
6642         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6643         WARN_ON(BTRFS_I(inode)->csum_bytes);
6644
6645         /*
6646          * This can happen where we create an inode, but somebody else also
6647          * created the same inode and we need to destroy the one we already
6648          * created.
6649          */
6650         if (!root)
6651                 goto free;
6652
6653         /*
6654          * Make sure we're properly removed from the ordered operation
6655          * lists.
6656          */
6657         smp_mb();
6658         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6659                 spin_lock(&root->fs_info->ordered_extent_lock);
6660                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6661                 spin_unlock(&root->fs_info->ordered_extent_lock);
6662         }
6663
6664         spin_lock(&root->orphan_lock);
6665         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6666                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6667                        (unsigned long long)btrfs_ino(inode));
6668                 list_del_init(&BTRFS_I(inode)->i_orphan);
6669         }
6670         spin_unlock(&root->orphan_lock);
6671
6672         while (1) {
6673                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6674                 if (!ordered)
6675                         break;
6676                 else {
6677                         printk(KERN_ERR "btrfs found ordered "
6678                                "extent %llu %llu on inode cleanup\n",
6679                                (unsigned long long)ordered->file_offset,
6680                                (unsigned long long)ordered->len);
6681                         btrfs_remove_ordered_extent(inode, ordered);
6682                         btrfs_put_ordered_extent(ordered);
6683                         btrfs_put_ordered_extent(ordered);
6684                 }
6685         }
6686         inode_tree_del(inode);
6687         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6688 free:
6689         btrfs_remove_delayed_node(inode);
6690         call_rcu(&inode->i_rcu, btrfs_i_callback);
6691 }
6692
6693 int btrfs_drop_inode(struct inode *inode)
6694 {
6695         struct btrfs_root *root = BTRFS_I(inode)->root;
6696
6697         if (btrfs_root_refs(&root->root_item) == 0 &&
6698             !btrfs_is_free_space_inode(root, inode))
6699                 return 1;
6700         else
6701                 return generic_drop_inode(inode);
6702 }
6703
6704 static void init_once(void *foo)
6705 {
6706         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6707
6708         inode_init_once(&ei->vfs_inode);
6709 }
6710
6711 void btrfs_destroy_cachep(void)
6712 {
6713         if (btrfs_inode_cachep)
6714                 kmem_cache_destroy(btrfs_inode_cachep);
6715         if (btrfs_trans_handle_cachep)
6716                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6717         if (btrfs_transaction_cachep)
6718                 kmem_cache_destroy(btrfs_transaction_cachep);
6719         if (btrfs_path_cachep)
6720                 kmem_cache_destroy(btrfs_path_cachep);
6721         if (btrfs_free_space_cachep)
6722                 kmem_cache_destroy(btrfs_free_space_cachep);
6723 }
6724
6725 int btrfs_init_cachep(void)
6726 {
6727         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6728                         sizeof(struct btrfs_inode), 0,
6729                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6730         if (!btrfs_inode_cachep)
6731                 goto fail;
6732
6733         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6734                         sizeof(struct btrfs_trans_handle), 0,
6735                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6736         if (!btrfs_trans_handle_cachep)
6737                 goto fail;
6738
6739         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6740                         sizeof(struct btrfs_transaction), 0,
6741                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6742         if (!btrfs_transaction_cachep)
6743                 goto fail;
6744
6745         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6746                         sizeof(struct btrfs_path), 0,
6747                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6748         if (!btrfs_path_cachep)
6749                 goto fail;
6750
6751         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6752                         sizeof(struct btrfs_free_space), 0,
6753                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6754         if (!btrfs_free_space_cachep)
6755                 goto fail;
6756
6757         return 0;
6758 fail:
6759         btrfs_destroy_cachep();
6760         return -ENOMEM;
6761 }
6762
6763 static int btrfs_getattr(struct vfsmount *mnt,
6764                          struct dentry *dentry, struct kstat *stat)
6765 {
6766         struct inode *inode = dentry->d_inode;
6767         generic_fillattr(inode, stat);
6768         stat->dev = BTRFS_I(inode)->root->anon_dev;
6769         stat->blksize = PAGE_CACHE_SIZE;
6770         stat->blocks = (inode_get_bytes(inode) +
6771                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6772         return 0;
6773 }
6774
6775 /*
6776  * If a file is moved, it will inherit the cow and compression flags of the new
6777  * directory.
6778  */
6779 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6780 {
6781         struct btrfs_inode *b_dir = BTRFS_I(dir);
6782         struct btrfs_inode *b_inode = BTRFS_I(inode);
6783
6784         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6785                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6786         else
6787                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6788
6789         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6790                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6791         else
6792                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6793 }
6794
6795 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6796                            struct inode *new_dir, struct dentry *new_dentry)
6797 {
6798         struct btrfs_trans_handle *trans;
6799         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6800         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6801         struct inode *new_inode = new_dentry->d_inode;
6802         struct inode *old_inode = old_dentry->d_inode;
6803         struct timespec ctime = CURRENT_TIME;
6804         u64 index = 0;
6805         u64 root_objectid;
6806         int ret;
6807         u64 old_ino = btrfs_ino(old_inode);
6808
6809         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6810                 return -EPERM;
6811
6812         /* we only allow rename subvolume link between subvolumes */
6813         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6814                 return -EXDEV;
6815
6816         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6817             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6818                 return -ENOTEMPTY;
6819
6820         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6821             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6822                 return -ENOTEMPTY;
6823         /*
6824          * we're using rename to replace one file with another.
6825          * and the replacement file is large.  Start IO on it now so
6826          * we don't add too much work to the end of the transaction
6827          */
6828         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6829             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6830                 filemap_flush(old_inode->i_mapping);
6831
6832         /* close the racy window with snapshot create/destroy ioctl */
6833         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6834                 down_read(&root->fs_info->subvol_sem);
6835         /*
6836          * We want to reserve the absolute worst case amount of items.  So if
6837          * both inodes are subvols and we need to unlink them then that would
6838          * require 4 item modifications, but if they are both normal inodes it
6839          * would require 5 item modifications, so we'll assume their normal
6840          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6841          * should cover the worst case number of items we'll modify.
6842          */
6843         trans = btrfs_start_transaction(root, 20);
6844         if (IS_ERR(trans)) {
6845                 ret = PTR_ERR(trans);
6846                 goto out_notrans;
6847         }
6848
6849         if (dest != root)
6850                 btrfs_record_root_in_trans(trans, dest);
6851
6852         ret = btrfs_set_inode_index(new_dir, &index);
6853         if (ret)
6854                 goto out_fail;
6855
6856         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6857                 /* force full log commit if subvolume involved. */
6858                 root->fs_info->last_trans_log_full_commit = trans->transid;
6859         } else {
6860                 ret = btrfs_insert_inode_ref(trans, dest,
6861                                              new_dentry->d_name.name,
6862                                              new_dentry->d_name.len,
6863                                              old_ino,
6864                                              btrfs_ino(new_dir), index);
6865                 if (ret)
6866                         goto out_fail;
6867                 /*
6868                  * this is an ugly little race, but the rename is required
6869                  * to make sure that if we crash, the inode is either at the
6870                  * old name or the new one.  pinning the log transaction lets
6871                  * us make sure we don't allow a log commit to come in after
6872                  * we unlink the name but before we add the new name back in.
6873                  */
6874                 btrfs_pin_log_trans(root);
6875         }
6876         /*
6877          * make sure the inode gets flushed if it is replacing
6878          * something.
6879          */
6880         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6881                 btrfs_add_ordered_operation(trans, root, old_inode);
6882
6883         old_dir->i_ctime = old_dir->i_mtime = ctime;
6884         new_dir->i_ctime = new_dir->i_mtime = ctime;
6885         old_inode->i_ctime = ctime;
6886
6887         if (old_dentry->d_parent != new_dentry->d_parent)
6888                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6889
6890         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6891                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6892                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6893                                         old_dentry->d_name.name,
6894                                         old_dentry->d_name.len);
6895         } else {
6896                 ret = __btrfs_unlink_inode(trans, root, old_dir,
6897                                         old_dentry->d_inode,
6898                                         old_dentry->d_name.name,
6899                                         old_dentry->d_name.len);
6900                 if (!ret)
6901                         ret = btrfs_update_inode(trans, root, old_inode);
6902         }
6903         BUG_ON(ret);
6904
6905         if (new_inode) {
6906                 new_inode->i_ctime = CURRENT_TIME;
6907                 if (unlikely(btrfs_ino(new_inode) ==
6908                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6909                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6910                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6911                                                 root_objectid,
6912                                                 new_dentry->d_name.name,
6913                                                 new_dentry->d_name.len);
6914                         BUG_ON(new_inode->i_nlink == 0);
6915                 } else {
6916                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6917                                                  new_dentry->d_inode,
6918                                                  new_dentry->d_name.name,
6919                                                  new_dentry->d_name.len);
6920                 }
6921                 BUG_ON(ret);
6922                 if (new_inode->i_nlink == 0) {
6923                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6924                         BUG_ON(ret);
6925                 }
6926         }
6927
6928         fixup_inode_flags(new_dir, old_inode);
6929
6930         ret = btrfs_add_link(trans, new_dir, old_inode,
6931                              new_dentry->d_name.name,
6932                              new_dentry->d_name.len, 0, index);
6933         BUG_ON(ret);
6934
6935         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
6936                 struct dentry *parent = new_dentry->d_parent;
6937                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6938                 btrfs_end_log_trans(root);
6939         }
6940 out_fail:
6941         btrfs_end_transaction_throttle(trans, root);
6942 out_notrans:
6943         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6944                 up_read(&root->fs_info->subvol_sem);
6945
6946         return ret;
6947 }
6948
6949 /*
6950  * some fairly slow code that needs optimization. This walks the list
6951  * of all the inodes with pending delalloc and forces them to disk.
6952  */
6953 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6954 {
6955         struct list_head *head = &root->fs_info->delalloc_inodes;
6956         struct btrfs_inode *binode;
6957         struct inode *inode;
6958
6959         if (root->fs_info->sb->s_flags & MS_RDONLY)
6960                 return -EROFS;
6961
6962         spin_lock(&root->fs_info->delalloc_lock);
6963         while (!list_empty(head)) {
6964                 binode = list_entry(head->next, struct btrfs_inode,
6965                                     delalloc_inodes);
6966                 inode = igrab(&binode->vfs_inode);
6967                 if (!inode)
6968                         list_del_init(&binode->delalloc_inodes);
6969                 spin_unlock(&root->fs_info->delalloc_lock);
6970                 if (inode) {
6971                         filemap_flush(inode->i_mapping);
6972                         if (delay_iput)
6973                                 btrfs_add_delayed_iput(inode);
6974                         else
6975                                 iput(inode);
6976                 }
6977                 cond_resched();
6978                 spin_lock(&root->fs_info->delalloc_lock);
6979         }
6980         spin_unlock(&root->fs_info->delalloc_lock);
6981
6982         /* the filemap_flush will queue IO into the worker threads, but
6983          * we have to make sure the IO is actually started and that
6984          * ordered extents get created before we return
6985          */
6986         atomic_inc(&root->fs_info->async_submit_draining);
6987         while (atomic_read(&root->fs_info->nr_async_submits) ||
6988               atomic_read(&root->fs_info->async_delalloc_pages)) {
6989                 wait_event(root->fs_info->async_submit_wait,
6990                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
6991                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
6992         }
6993         atomic_dec(&root->fs_info->async_submit_draining);
6994         return 0;
6995 }
6996
6997 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
6998                          const char *symname)
6999 {
7000         struct btrfs_trans_handle *trans;
7001         struct btrfs_root *root = BTRFS_I(dir)->root;
7002         struct btrfs_path *path;
7003         struct btrfs_key key;
7004         struct inode *inode = NULL;
7005         int err;
7006         int drop_inode = 0;
7007         u64 objectid;
7008         u64 index = 0 ;
7009         int name_len;
7010         int datasize;
7011         unsigned long ptr;
7012         struct btrfs_file_extent_item *ei;
7013         struct extent_buffer *leaf;
7014         unsigned long nr = 0;
7015
7016         name_len = strlen(symname) + 1;
7017         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7018                 return -ENAMETOOLONG;
7019
7020         /*
7021          * 2 items for inode item and ref
7022          * 2 items for dir items
7023          * 1 item for xattr if selinux is on
7024          */
7025         trans = btrfs_start_transaction(root, 5);
7026         if (IS_ERR(trans))
7027                 return PTR_ERR(trans);
7028
7029         err = btrfs_find_free_ino(root, &objectid);
7030         if (err)
7031                 goto out_unlock;
7032
7033         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7034                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7035                                 S_IFLNK|S_IRWXUGO, &index);
7036         if (IS_ERR(inode)) {
7037                 err = PTR_ERR(inode);
7038                 goto out_unlock;
7039         }
7040
7041         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7042         if (err) {
7043                 drop_inode = 1;
7044                 goto out_unlock;
7045         }
7046
7047         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7048         if (err)
7049                 drop_inode = 1;
7050         else {
7051                 inode->i_mapping->a_ops = &btrfs_aops;
7052                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7053                 inode->i_fop = &btrfs_file_operations;
7054                 inode->i_op = &btrfs_file_inode_operations;
7055                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7056         }
7057         if (drop_inode)
7058                 goto out_unlock;
7059
7060         path = btrfs_alloc_path();
7061         if (!path) {
7062                 err = -ENOMEM;
7063                 drop_inode = 1;
7064                 goto out_unlock;
7065         }
7066         key.objectid = btrfs_ino(inode);
7067         key.offset = 0;
7068         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7069         datasize = btrfs_file_extent_calc_inline_size(name_len);
7070         err = btrfs_insert_empty_item(trans, root, path, &key,
7071                                       datasize);
7072         if (err) {
7073                 drop_inode = 1;
7074                 btrfs_free_path(path);
7075                 goto out_unlock;
7076         }
7077         leaf = path->nodes[0];
7078         ei = btrfs_item_ptr(leaf, path->slots[0],
7079                             struct btrfs_file_extent_item);
7080         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7081         btrfs_set_file_extent_type(leaf, ei,
7082                                    BTRFS_FILE_EXTENT_INLINE);
7083         btrfs_set_file_extent_encryption(leaf, ei, 0);
7084         btrfs_set_file_extent_compression(leaf, ei, 0);
7085         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7086         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7087
7088         ptr = btrfs_file_extent_inline_start(ei);
7089         write_extent_buffer(leaf, symname, ptr, name_len);
7090         btrfs_mark_buffer_dirty(leaf);
7091         btrfs_free_path(path);
7092
7093         inode->i_op = &btrfs_symlink_inode_operations;
7094         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7095         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7096         inode_set_bytes(inode, name_len);
7097         btrfs_i_size_write(inode, name_len - 1);
7098         err = btrfs_update_inode(trans, root, inode);
7099         if (err)
7100                 drop_inode = 1;
7101
7102 out_unlock:
7103         nr = trans->blocks_used;
7104         btrfs_end_transaction_throttle(trans, root);
7105         if (drop_inode) {
7106                 inode_dec_link_count(inode);
7107                 iput(inode);
7108         }
7109         btrfs_btree_balance_dirty(root, nr);
7110         return err;
7111 }
7112
7113 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7114                                        u64 start, u64 num_bytes, u64 min_size,
7115                                        loff_t actual_len, u64 *alloc_hint,
7116                                        struct btrfs_trans_handle *trans)
7117 {
7118         struct btrfs_root *root = BTRFS_I(inode)->root;
7119         struct btrfs_key ins;
7120         u64 cur_offset = start;
7121         u64 i_size;
7122         int ret = 0;
7123         bool own_trans = true;
7124
7125         if (trans)
7126                 own_trans = false;
7127         while (num_bytes > 0) {
7128                 if (own_trans) {
7129                         trans = btrfs_start_transaction(root, 3);
7130                         if (IS_ERR(trans)) {
7131                                 ret = PTR_ERR(trans);
7132                                 break;
7133                         }
7134                 }
7135
7136                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7137                                            0, *alloc_hint, (u64)-1, &ins, 1);
7138                 if (ret) {
7139                         if (own_trans)
7140                                 btrfs_end_transaction(trans, root);
7141                         break;
7142                 }
7143
7144                 ret = insert_reserved_file_extent(trans, inode,
7145                                                   cur_offset, ins.objectid,
7146                                                   ins.offset, ins.offset,
7147                                                   ins.offset, 0, 0, 0,
7148                                                   BTRFS_FILE_EXTENT_PREALLOC);
7149                 BUG_ON(ret);
7150                 btrfs_drop_extent_cache(inode, cur_offset,
7151                                         cur_offset + ins.offset -1, 0);
7152
7153                 num_bytes -= ins.offset;
7154                 cur_offset += ins.offset;
7155                 *alloc_hint = ins.objectid + ins.offset;
7156
7157                 inode->i_ctime = CURRENT_TIME;
7158                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7159                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7160                     (actual_len > inode->i_size) &&
7161                     (cur_offset > inode->i_size)) {
7162                         if (cur_offset > actual_len)
7163                                 i_size = actual_len;
7164                         else
7165                                 i_size = cur_offset;
7166                         i_size_write(inode, i_size);
7167                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7168                 }
7169
7170                 ret = btrfs_update_inode(trans, root, inode);
7171                 BUG_ON(ret);
7172
7173                 if (own_trans)
7174                         btrfs_end_transaction(trans, root);
7175         }
7176         return ret;
7177 }
7178
7179 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7180                               u64 start, u64 num_bytes, u64 min_size,
7181                               loff_t actual_len, u64 *alloc_hint)
7182 {
7183         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7184                                            min_size, actual_len, alloc_hint,
7185                                            NULL);
7186 }
7187
7188 int btrfs_prealloc_file_range_trans(struct inode *inode,
7189                                     struct btrfs_trans_handle *trans, int mode,
7190                                     u64 start, u64 num_bytes, u64 min_size,
7191                                     loff_t actual_len, u64 *alloc_hint)
7192 {
7193         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7194                                            min_size, actual_len, alloc_hint, trans);
7195 }
7196
7197 static int btrfs_set_page_dirty(struct page *page)
7198 {
7199         return __set_page_dirty_nobuffers(page);
7200 }
7201
7202 static int btrfs_permission(struct inode *inode, int mask)
7203 {
7204         struct btrfs_root *root = BTRFS_I(inode)->root;
7205         umode_t mode = inode->i_mode;
7206
7207         if (mask & MAY_WRITE &&
7208             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7209                 if (btrfs_root_readonly(root))
7210                         return -EROFS;
7211                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7212                         return -EACCES;
7213         }
7214         return generic_permission(inode, mask);
7215 }
7216
7217 static const struct inode_operations btrfs_dir_inode_operations = {
7218         .getattr        = btrfs_getattr,
7219         .lookup         = btrfs_lookup,
7220         .create         = btrfs_create,
7221         .unlink         = btrfs_unlink,
7222         .link           = btrfs_link,
7223         .mkdir          = btrfs_mkdir,
7224         .rmdir          = btrfs_rmdir,
7225         .rename         = btrfs_rename,
7226         .symlink        = btrfs_symlink,
7227         .setattr        = btrfs_setattr,
7228         .mknod          = btrfs_mknod,
7229         .setxattr       = btrfs_setxattr,
7230         .getxattr       = btrfs_getxattr,
7231         .listxattr      = btrfs_listxattr,
7232         .removexattr    = btrfs_removexattr,
7233         .permission     = btrfs_permission,
7234         .get_acl        = btrfs_get_acl,
7235 };
7236 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7237         .lookup         = btrfs_lookup,
7238         .permission     = btrfs_permission,
7239         .get_acl        = btrfs_get_acl,
7240 };
7241
7242 static const struct file_operations btrfs_dir_file_operations = {
7243         .llseek         = generic_file_llseek,
7244         .read           = generic_read_dir,
7245         .readdir        = btrfs_real_readdir,
7246         .unlocked_ioctl = btrfs_ioctl,
7247 #ifdef CONFIG_COMPAT
7248         .compat_ioctl   = btrfs_ioctl,
7249 #endif
7250         .release        = btrfs_release_file,
7251         .fsync          = btrfs_sync_file,
7252 };
7253
7254 static struct extent_io_ops btrfs_extent_io_ops = {
7255         .fill_delalloc = run_delalloc_range,
7256         .submit_bio_hook = btrfs_submit_bio_hook,
7257         .merge_bio_hook = btrfs_merge_bio_hook,
7258         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7259         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7260         .writepage_start_hook = btrfs_writepage_start_hook,
7261         .set_bit_hook = btrfs_set_bit_hook,
7262         .clear_bit_hook = btrfs_clear_bit_hook,
7263         .merge_extent_hook = btrfs_merge_extent_hook,
7264         .split_extent_hook = btrfs_split_extent_hook,
7265 };
7266
7267 /*
7268  * btrfs doesn't support the bmap operation because swapfiles
7269  * use bmap to make a mapping of extents in the file.  They assume
7270  * these extents won't change over the life of the file and they
7271  * use the bmap result to do IO directly to the drive.
7272  *
7273  * the btrfs bmap call would return logical addresses that aren't
7274  * suitable for IO and they also will change frequently as COW
7275  * operations happen.  So, swapfile + btrfs == corruption.
7276  *
7277  * For now we're avoiding this by dropping bmap.
7278  */
7279 static const struct address_space_operations btrfs_aops = {
7280         .readpage       = btrfs_readpage,
7281         .writepage      = btrfs_writepage,
7282         .writepages     = btrfs_writepages,
7283         .readpages      = btrfs_readpages,
7284         .direct_IO      = btrfs_direct_IO,
7285         .invalidatepage = btrfs_invalidatepage,
7286         .releasepage    = btrfs_releasepage,
7287         .set_page_dirty = btrfs_set_page_dirty,
7288         .error_remove_page = generic_error_remove_page,
7289 };
7290
7291 static const struct address_space_operations btrfs_symlink_aops = {
7292         .readpage       = btrfs_readpage,
7293         .writepage      = btrfs_writepage,
7294         .invalidatepage = btrfs_invalidatepage,
7295         .releasepage    = btrfs_releasepage,
7296 };
7297
7298 static const struct inode_operations btrfs_file_inode_operations = {
7299         .getattr        = btrfs_getattr,
7300         .setattr        = btrfs_setattr,
7301         .setxattr       = btrfs_setxattr,
7302         .getxattr       = btrfs_getxattr,
7303         .listxattr      = btrfs_listxattr,
7304         .removexattr    = btrfs_removexattr,
7305         .permission     = btrfs_permission,
7306         .fiemap         = btrfs_fiemap,
7307         .get_acl        = btrfs_get_acl,
7308 };
7309 static const struct inode_operations btrfs_special_inode_operations = {
7310         .getattr        = btrfs_getattr,
7311         .setattr        = btrfs_setattr,
7312         .permission     = btrfs_permission,
7313         .setxattr       = btrfs_setxattr,
7314         .getxattr       = btrfs_getxattr,
7315         .listxattr      = btrfs_listxattr,
7316         .removexattr    = btrfs_removexattr,
7317         .get_acl        = btrfs_get_acl,
7318 };
7319 static const struct inode_operations btrfs_symlink_inode_operations = {
7320         .readlink       = generic_readlink,
7321         .follow_link    = page_follow_link_light,
7322         .put_link       = page_put_link,
7323         .getattr        = btrfs_getattr,
7324         .permission     = btrfs_permission,
7325         .setxattr       = btrfs_setxattr,
7326         .getxattr       = btrfs_getxattr,
7327         .listxattr      = btrfs_listxattr,
7328         .removexattr    = btrfs_removexattr,
7329         .get_acl        = btrfs_get_acl,
7330 };
7331
7332 const struct dentry_operations btrfs_dentry_operations = {
7333         .d_delete       = btrfs_dentry_delete,
7334         .d_release      = btrfs_dentry_release,
7335 };