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 "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.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         btrfs_set_trans_block_group(trans, inode);
 142
 143         key.objectid = btrfs_ino(inode);
 144         key.offset = start;
 145         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 146         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 147
 148         inode_add_bytes(inode, size);
 149         ret = btrfs_insert_empty_item(trans, root, path, &key,
 150                                       datasize);
 151         BUG_ON(ret);
 152         if (ret) {
 153                 err = ret;
 154                 goto fail;
 155         }
 156         leaf = path->nodes[0];
 157         ei = btrfs_item_ptr(leaf, path->slots[0],
 158                             struct btrfs_file_extent_item);
 159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 161         btrfs_set_file_extent_encryption(leaf, ei, 0);
 162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 164         ptr = btrfs_file_extent_inline_start(ei);
 165
 166         if (compress_type != BTRFS_COMPRESS_NONE) {
 167                 struct page *cpage;
 168                 int i = 0;
 169                 while (compressed_size > 0) {
 170                         cpage = compressed_pages[i];
 171                         cur_size = min_t(unsigned long, compressed_size,
 172                                        PAGE_CACHE_SIZE);
 173
 174                         kaddr = kmap_atomic(cpage, KM_USER0);
 175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 176                         kunmap_atomic(kaddr, KM_USER0);
 177
 178                         i++;
 179                         ptr += cur_size;
 180                         compressed_size -= cur_size;
 181                 }
 182                 btrfs_set_file_extent_compression(leaf, ei,
 183                                                   compress_type);
 184         } else {
 185                 page = find_get_page(inode->i_mapping,
 186                                      start >> PAGE_CACHE_SHIFT);
 187                 btrfs_set_file_extent_compression(leaf, ei, 0);
 188                 kaddr = kmap_atomic(page, KM_USER0);
 189                 offset = start & (PAGE_CACHE_SIZE - 1);
 190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 191                 kunmap_atomic(kaddr, KM_USER0);
 192                 page_cache_release(page);
 193         }
 194         btrfs_mark_buffer_dirty(leaf);
 195         btrfs_free_path(path);
 196
 197         /*
 198          * we're an inline extent, so nobody can
 199          * extend the file past i_size without locking
 200          * a page we already have locked.
 201          *
 202          * We must do any isize and inode updates
 203          * before we unlock the pages.  Otherwise we
 204          * could end up racing with unlink.
 205          */
 206         BTRFS_I(inode)->disk_i_size = inode->i_size;
 207         btrfs_update_inode(trans, root, inode);
 208
 209         return 0;
 210 fail:
 211         btrfs_free_path(path);
 212         return err;
 213 }
 214
 215
 216 /*
 217  * conditionally insert an inline extent into the file.  This
 218  * does the checks required to make sure the data is small enough
 219  * to fit as an inline extent.
 220  */
 221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 222                                  struct btrfs_root *root,
 223                                  struct inode *inode, u64 start, u64 end,
 224                                  size_t compressed_size, int compress_type,
 225                                  struct page **compressed_pages)
 226 {
 227         u64 isize = i_size_read(inode);
 228         u64 actual_end = min(end + 1, isize);
 229         u64 inline_len = actual_end - start;
 230         u64 aligned_end = (end + root->sectorsize - 1) &
 231                         ~((u64)root->sectorsize - 1);
 232         u64 hint_byte;
 233         u64 data_len = inline_len;
 234         int ret;
 235
 236         if (compressed_size)
 237                 data_len = compressed_size;
 238
 239         if (start > 0 ||
 240             actual_end >= PAGE_CACHE_SIZE ||
 241             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 242             (!compressed_size &&
 243             (actual_end & (root->sectorsize - 1)) == 0) ||
 244             end + 1 < isize ||
 245             data_len > root->fs_info->max_inline) {
 246                 return 1;
 247         }
 248
 249         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 250                                  &hint_byte, 1);
 251         BUG_ON(ret);
 252
 253         if (isize > actual_end)
 254                 inline_len = min_t(u64, isize, actual_end);
 255         ret = insert_inline_extent(trans, root, inode, start,
 256                                    inline_len, compressed_size,
 257                                    compress_type, compressed_pages);
 258         BUG_ON(ret);
 259         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 260         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 261         return 0;
 262 }
 263
 264 struct async_extent {
 265         u64 start;
 266         u64 ram_size;
 267         u64 compressed_size;
 268         struct page **pages;
 269         unsigned long nr_pages;
 270         int compress_type;
 271         struct list_head list;
 272 };
 273
 274 struct async_cow {
 275         struct inode *inode;
 276         struct btrfs_root *root;
 277         struct page *locked_page;
 278         u64 start;
 279         u64 end;
 280         struct list_head extents;
 281         struct btrfs_work work;
 282 };
 283
 284 static noinline int add_async_extent(struct async_cow *cow,
 285                                      u64 start, u64 ram_size,
 286                                      u64 compressed_size,
 287                                      struct page **pages,
 288                                      unsigned long nr_pages,
 289                                      int compress_type)
 290 {
 291         struct async_extent *async_extent;
 292
 293         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 294         BUG_ON(!async_extent);
 295         async_extent->start = start;
 296         async_extent->ram_size = ram_size;
 297         async_extent->compressed_size = compressed_size;
 298         async_extent->pages = pages;
 299         async_extent->nr_pages = nr_pages;
 300         async_extent->compress_type = compress_type;
 301         list_add_tail(&async_extent->list, &cow->extents);
 302         return 0;
 303 }
 304
 305 /*
 306  * we create compressed extents in two phases.  The first
 307  * phase compresses a range of pages that have already been
 308  * locked (both pages and state bits are locked).
 309  *
 310  * This is done inside an ordered work queue, and the compression
 311  * is spread across many cpus.  The actual IO submission is step
 312  * two, and the ordered work queue takes care of making sure that
 313  * happens in the same order things were put onto the queue by
 314  * writepages and friends.
 315  *
 316  * If this code finds it can't get good compression, it puts an
 317  * entry onto the work queue to write the uncompressed bytes.  This
 318  * makes sure that both compressed inodes and uncompressed inodes
 319  * are written in the same order that pdflush sent them down.
 320  */
 321 static noinline int compress_file_range(struct inode *inode,
 322                                         struct page *locked_page,
 323                                         u64 start, u64 end,
 324                                         struct async_cow *async_cow,
 325                                         int *num_added)
 326 {
 327         struct btrfs_root *root = BTRFS_I(inode)->root;
 328         struct btrfs_trans_handle *trans;
 329         u64 num_bytes;
 330         u64 blocksize = root->sectorsize;
 331         u64 actual_end;
 332         u64 isize = i_size_read(inode);
 333         int ret = 0;
 334         struct page **pages = NULL;
 335         unsigned long nr_pages;
 336         unsigned long nr_pages_ret = 0;
 337         unsigned long total_compressed = 0;
 338         unsigned long total_in = 0;
 339         unsigned long max_compressed = 128 * 1024;
 340         unsigned long max_uncompressed = 128 * 1024;
 341         int i;
 342         int will_compress;
 343         int compress_type = root->fs_info->compress_type;
 344
 345         actual_end = min_t(u64, isize, end + 1);
 346 again:
 347         will_compress = 0;
 348         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 349         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 350
 351         /*
 352          * we don't want to send crud past the end of i_size through
 353          * compression, that's just a waste of CPU time.  So, if the
 354          * end of the file is before the start of our current
 355          * requested range of bytes, we bail out to the uncompressed
 356          * cleanup code that can deal with all of this.
 357          *
 358          * It isn't really the fastest way to fix things, but this is a
 359          * very uncommon corner.
 360          */
 361         if (actual_end <= start)
 362                 goto cleanup_and_bail_uncompressed;
 363
 364         total_compressed = actual_end - start;
 365
 366         /* we want to make sure that amount of ram required to uncompress
 367          * an extent is reasonable, so we limit the total size in ram
 368          * of a compressed extent to 128k.  This is a crucial number
 369          * because it also controls how easily we can spread reads across
 370          * cpus for decompression.
 371          *
 372          * We also want to make sure the amount of IO required to do
 373          * a random read is reasonably small, so we limit the size of
 374          * a compressed extent to 128k.
 375          */
 376         total_compressed = min(total_compressed, max_uncompressed);
 377         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 378         num_bytes = max(blocksize,  num_bytes);
 379         total_in = 0;
 380         ret = 0;
 381
 382         /*
 383          * we do compression for mount -o compress and when the
 384          * inode has not been flagged as nocompress.  This flag can
 385          * change at any time if we discover bad compression ratios.
 386          */
 387         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 388             (btrfs_test_opt(root, COMPRESS) ||
 389              (BTRFS_I(inode)->force_compress) ||
 390              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 391                 WARN_ON(pages);
 392                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 393                 BUG_ON(!pages);
 394
 395                 if (BTRFS_I(inode)->force_compress)
 396                         compress_type = BTRFS_I(inode)->force_compress;
 397
 398                 ret = btrfs_compress_pages(compress_type,
 399                                            inode->i_mapping, start,
 400                                            total_compressed, pages,
 401                                            nr_pages, &nr_pages_ret,
 402                                            &total_in,
 403                                            &total_compressed,
 404                                            max_compressed);
 405
 406                 if (!ret) {
 407                         unsigned long offset = total_compressed &
 408                                 (PAGE_CACHE_SIZE - 1);
 409                         struct page *page = pages[nr_pages_ret - 1];
 410                         char *kaddr;
 411
 412                         /* zero the tail end of the last page, we might be
 413                          * sending it down to disk
 414                          */
 415                         if (offset) {
 416                                 kaddr = kmap_atomic(page, KM_USER0);
 417                                 memset(kaddr + offset, 0,
 418                                        PAGE_CACHE_SIZE - offset);
 419                                 kunmap_atomic(kaddr, KM_USER0);
 420                         }
 421                         will_compress = 1;
 422                 }
 423         }
 424         if (start == 0) {
 425                 trans = btrfs_join_transaction(root, 1);
 426                 BUG_ON(IS_ERR(trans));
 427                 btrfs_set_trans_block_group(trans, inode);
 428                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 429
 430                 /* lets try to make an inline extent */
 431                 if (ret || total_in < (actual_end - start)) {
 432                         /* we didn't compress the entire range, try
 433                          * to make an uncompressed inline extent.
 434                          */
 435                         ret = cow_file_range_inline(trans, root, inode,
 436                                                     start, end, 0, 0, NULL);
 437                 } else {
 438                         /* try making a compressed inline extent */
 439                         ret = cow_file_range_inline(trans, root, inode,
 440                                                     start, end,
 441                                                     total_compressed,
 442                                                     compress_type, pages);
 443                 }
 444                 if (ret == 0) {
 445                         /*
 446                          * inline extent creation worked, we don't need
 447                          * to create any more async work items.  Unlock
 448                          * and free up our temp pages.
 449                          */
 450                         extent_clear_unlock_delalloc(inode,
 451                              &BTRFS_I(inode)->io_tree,
 452                              start, end, NULL,
 453                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 454                              EXTENT_CLEAR_DELALLOC |
 455                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 456
 457                         btrfs_end_transaction(trans, root);
 458                         goto free_pages_out;
 459                 }
 460                 btrfs_end_transaction(trans, root);
 461         }
 462
 463         if (will_compress) {
 464                 /*
 465                  * we aren't doing an inline extent round the compressed size
 466                  * up to a block size boundary so the allocator does sane
 467                  * things
 468                  */
 469                 total_compressed = (total_compressed + blocksize - 1) &
 470                         ~(blocksize - 1);
 471
 472                 /*
 473                  * one last check to make sure the compression is really a
 474                  * win, compare the page count read with the blocks on disk
 475                  */
 476                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 477                         ~(PAGE_CACHE_SIZE - 1);
 478                 if (total_compressed >= total_in) {
 479                         will_compress = 0;
 480                 } else {
 481                         num_bytes = total_in;
 482                 }
 483         }
 484         if (!will_compress && pages) {
 485                 /*
 486                  * the compression code ran but failed to make things smaller,
 487                  * free any pages it allocated and our page pointer array
 488                  */
 489                 for (i = 0; i < nr_pages_ret; i++) {
 490                         WARN_ON(pages[i]->mapping);
 491                         page_cache_release(pages[i]);
 492                 }
 493                 kfree(pages);
 494                 pages = NULL;
 495                 total_compressed = 0;
 496                 nr_pages_ret = 0;
 497
 498                 /* flag the file so we don't compress in the future */
 499                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 500                     !(BTRFS_I(inode)->force_compress)) {
 501                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 502                 }
 503         }
 504         if (will_compress) {
 505                 *num_added += 1;
 506
 507                 /* the async work queues will take care of doing actual
 508                  * allocation on disk for these compressed pages,
 509                  * and will submit them to the elevator.
 510                  */
 511                 add_async_extent(async_cow, start, num_bytes,
 512                                  total_compressed, pages, nr_pages_ret,
 513                                  compress_type);
 514
 515                 if (start + num_bytes < end) {
 516                         start += num_bytes;
 517                         pages = NULL;
 518                         cond_resched();
 519                         goto again;
 520                 }
 521         } else {
 522 cleanup_and_bail_uncompressed:
 523                 /*
 524                  * No compression, but we still need to write the pages in
 525                  * the file we've been given so far.  redirty the locked
 526                  * page if it corresponds to our extent and set things up
 527                  * for the async work queue to run cow_file_range to do
 528                  * the normal delalloc dance
 529                  */
 530                 if (page_offset(locked_page) >= start &&
 531                     page_offset(locked_page) <= end) {
 532                         __set_page_dirty_nobuffers(locked_page);
 533                         /* unlocked later on in the async handlers */
 534                 }
 535                 add_async_extent(async_cow, start, end - start + 1,
 536                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 537                 *num_added += 1;
 538         }
 539
 540 out:
 541         return 0;
 542
 543 free_pages_out:
 544         for (i = 0; i < nr_pages_ret; i++) {
 545                 WARN_ON(pages[i]->mapping);
 546                 page_cache_release(pages[i]);
 547         }
 548         kfree(pages);
 549
 550         goto out;
 551 }
 552
 553 /*
 554  * phase two of compressed writeback.  This is the ordered portion
 555  * of the code, which only gets called in the order the work was
 556  * queued.  We walk all the async extents created by compress_file_range
 557  * and send them down to the disk.
 558  */
 559 static noinline int submit_compressed_extents(struct inode *inode,
 560                                               struct async_cow *async_cow)
 561 {
 562         struct async_extent *async_extent;
 563         u64 alloc_hint = 0;
 564         struct btrfs_trans_handle *trans;
 565         struct btrfs_key ins;
 566         struct extent_map *em;
 567         struct btrfs_root *root = BTRFS_I(inode)->root;
 568         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 569         struct extent_io_tree *io_tree;
 570         int ret = 0;
 571
 572         if (list_empty(&async_cow->extents))
 573                 return 0;
 574
 575
 576         while (!list_empty(&async_cow->extents)) {
 577                 async_extent = list_entry(async_cow->extents.next,
 578                                           struct async_extent, list);
 579                 list_del(&async_extent->list);
 580
 581                 io_tree = &BTRFS_I(inode)->io_tree;
 582
 583 retry:
 584                 /* did the compression code fall back to uncompressed IO? */
 585                 if (!async_extent->pages) {
 586                         int page_started = 0;
 587                         unsigned long nr_written = 0;
 588
 589                         lock_extent(io_tree, async_extent->start,
 590                                          async_extent->start +
 591                                          async_extent->ram_size - 1, GFP_NOFS);
 592
 593                         /* allocate blocks */
 594                         ret = cow_file_range(inode, async_cow->locked_page,
 595                                              async_extent->start,
 596                                              async_extent->start +
 597                                              async_extent->ram_size - 1,
 598                                              &page_started, &nr_written, 0);
 599
 600                         /*
 601                          * if page_started, cow_file_range inserted an
 602                          * inline extent and took care of all the unlocking
 603                          * and IO for us.  Otherwise, we need to submit
 604                          * all those pages down to the drive.
 605                          */
 606                         if (!page_started && !ret)
 607                                 extent_write_locked_range(io_tree,
 608                                                   inode, async_extent->start,
 609                                                   async_extent->start +
 610                                                   async_extent->ram_size - 1,
 611                                                   btrfs_get_extent,
 612                                                   WB_SYNC_ALL);
 613                         kfree(async_extent);
 614                         cond_resched();
 615                         continue;
 616                 }
 617
 618                 lock_extent(io_tree, async_extent->start,
 619                             async_extent->start + async_extent->ram_size - 1,
 620                             GFP_NOFS);
 621
 622                 trans = btrfs_join_transaction(root, 1);
 623                 BUG_ON(IS_ERR(trans));
 624                 ret = btrfs_reserve_extent(trans, root,
 625                                            async_extent->compressed_size,
 626                                            async_extent->compressed_size,
 627                                            0, alloc_hint,
 628                                            (u64)-1, &ins, 1);
 629                 btrfs_end_transaction(trans, root);
 630
 631                 if (ret) {
 632                         int i;
 633                         for (i = 0; i < async_extent->nr_pages; i++) {
 634                                 WARN_ON(async_extent->pages[i]->mapping);
 635                                 page_cache_release(async_extent->pages[i]);
 636                         }
 637                         kfree(async_extent->pages);
 638                         async_extent->nr_pages = 0;
 639                         async_extent->pages = NULL;
 640                         unlock_extent(io_tree, async_extent->start,
 641                                       async_extent->start +
 642                                       async_extent->ram_size - 1, GFP_NOFS);
 643                         goto retry;
 644                 }
 645
 646                 /*
 647                  * here we're doing allocation and writeback of the
 648                  * compressed pages
 649                  */
 650                 btrfs_drop_extent_cache(inode, async_extent->start,
 651                                         async_extent->start +
 652                                         async_extent->ram_size - 1, 0);
 653
 654                 em = alloc_extent_map();
 655                 BUG_ON(!em);
 656                 em->start = async_extent->start;
 657                 em->len = async_extent->ram_size;
 658                 em->orig_start = em->start;
 659
 660                 em->block_start = ins.objectid;
 661                 em->block_len = ins.offset;
 662                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 663                 em->compress_type = async_extent->compress_type;
 664                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 665                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 666
 667                 while (1) {
 668                         write_lock(&em_tree->lock);
 669                         ret = add_extent_mapping(em_tree, em);
 670                         write_unlock(&em_tree->lock);
 671                         if (ret != -EEXIST) {
 672                                 free_extent_map(em);
 673                                 break;
 674                         }
 675                         btrfs_drop_extent_cache(inode, async_extent->start,
 676                                                 async_extent->start +
 677                                                 async_extent->ram_size - 1, 0);
 678                 }
 679
 680                 ret = btrfs_add_ordered_extent_compress(inode,
 681                                                 async_extent->start,
 682                                                 ins.objectid,
 683                                                 async_extent->ram_size,
 684                                                 ins.offset,
 685                                                 BTRFS_ORDERED_COMPRESSED,
 686                                                 async_extent->compress_type);
 687                 BUG_ON(ret);
 688
 689                 /*
 690                  * clear dirty, set writeback and unlock the pages.
 691                  */
 692                 extent_clear_unlock_delalloc(inode,
 693                                 &BTRFS_I(inode)->io_tree,
 694                                 async_extent->start,
 695                                 async_extent->start +
 696                                 async_extent->ram_size - 1,
 697                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 698                                 EXTENT_CLEAR_UNLOCK |
 699                                 EXTENT_CLEAR_DELALLOC |
 700                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 701
 702                 ret = btrfs_submit_compressed_write(inode,
 703                                     async_extent->start,
 704                                     async_extent->ram_size,
 705                                     ins.objectid,
 706                                     ins.offset, async_extent->pages,
 707                                     async_extent->nr_pages);
 708
 709                 BUG_ON(ret);
 710                 alloc_hint = ins.objectid + ins.offset;
 711                 kfree(async_extent);
 712                 cond_resched();
 713         }
 714
 715         return 0;
 716 }
 717
 718 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 719                                       u64 num_bytes)
 720 {
 721         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 722         struct extent_map *em;
 723         u64 alloc_hint = 0;
 724
 725         read_lock(&em_tree->lock);
 726         em = search_extent_mapping(em_tree, start, num_bytes);
 727         if (em) {
 728                 /*
 729                  * if block start isn't an actual block number then find the
 730                  * first block in this inode and use that as a hint.  If that
 731                  * block is also bogus then just don't worry about it.
 732                  */
 733                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 734                         free_extent_map(em);
 735                         em = search_extent_mapping(em_tree, 0, 0);
 736                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 737                                 alloc_hint = em->block_start;
 738                         if (em)
 739                                 free_extent_map(em);
 740                 } else {
 741                         alloc_hint = em->block_start;
 742                         free_extent_map(em);
 743                 }
 744         }
 745         read_unlock(&em_tree->lock);
 746
 747         return alloc_hint;
 748 }
 749
 750 static inline bool is_free_space_inode(struct btrfs_root *root,
 751                                        struct inode *inode)
 752 {
 753         if (root == root->fs_info->tree_root ||
 754             BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
 755                 return true;
 756         return false;
 757 }
 758
 759 /*
 760  * when extent_io.c finds a delayed allocation range in the file,
 761  * the call backs end up in this code.  The basic idea is to
 762  * allocate extents on disk for the range, and create ordered data structs
 763  * in ram to track those extents.
 764  *
 765  * locked_page is the page that writepage had locked already.  We use
 766  * it to make sure we don't do extra locks or unlocks.
 767  *
 768  * *page_started is set to one if we unlock locked_page and do everything
 769  * required to start IO on it.  It may be clean and already done with
 770  * IO when we return.
 771  */
 772 static noinline int cow_file_range(struct inode *inode,
 773                                    struct page *locked_page,
 774                                    u64 start, u64 end, int *page_started,
 775                                    unsigned long *nr_written,
 776                                    int unlock)
 777 {
 778         struct btrfs_root *root = BTRFS_I(inode)->root;
 779         struct btrfs_trans_handle *trans;
 780         u64 alloc_hint = 0;
 781         u64 num_bytes;
 782         unsigned long ram_size;
 783         u64 disk_num_bytes;
 784         u64 cur_alloc_size;
 785         u64 blocksize = root->sectorsize;
 786         struct btrfs_key ins;
 787         struct extent_map *em;
 788         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 789         int ret = 0;
 790
 791         BUG_ON(is_free_space_inode(root, inode));
 792         trans = btrfs_join_transaction(root, 1);
 793         BUG_ON(IS_ERR(trans));
 794         btrfs_set_trans_block_group(trans, inode);
 795         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 796
 797         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 798         num_bytes = max(blocksize,  num_bytes);
 799         disk_num_bytes = num_bytes;
 800         ret = 0;
 801
 802         if (start == 0) {
 803                 /* lets try to make an inline extent */
 804                 ret = cow_file_range_inline(trans, root, inode,
 805                                             start, end, 0, 0, NULL);
 806                 if (ret == 0) {
 807                         extent_clear_unlock_delalloc(inode,
 808                                      &BTRFS_I(inode)->io_tree,
 809                                      start, end, NULL,
 810                                      EXTENT_CLEAR_UNLOCK_PAGE |
 811                                      EXTENT_CLEAR_UNLOCK |
 812                                      EXTENT_CLEAR_DELALLOC |
 813                                      EXTENT_CLEAR_DIRTY |
 814                                      EXTENT_SET_WRITEBACK |
 815                                      EXTENT_END_WRITEBACK);
 816
 817                         *nr_written = *nr_written +
 818                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 819                         *page_started = 1;
 820                         ret = 0;
 821                         goto out;
 822                 }
 823         }
 824
 825         BUG_ON(disk_num_bytes >
 826                btrfs_super_total_bytes(&root->fs_info->super_copy));
 827
 828         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 829         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 830
 831         while (disk_num_bytes > 0) {
 832                 unsigned long op;
 833
 834                 cur_alloc_size = disk_num_bytes;
 835                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 836                                            root->sectorsize, 0, alloc_hint,
 837                                            (u64)-1, &ins, 1);
 838                 BUG_ON(ret);
 839
 840                 em = alloc_extent_map();
 841                 BUG_ON(!em);
 842                 em->start = start;
 843                 em->orig_start = em->start;
 844                 ram_size = ins.offset;
 845                 em->len = ins.offset;
 846
 847                 em->block_start = ins.objectid;
 848                 em->block_len = ins.offset;
 849                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 850                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 851
 852                 while (1) {
 853                         write_lock(&em_tree->lock);
 854                         ret = add_extent_mapping(em_tree, em);
 855                         write_unlock(&em_tree->lock);
 856                         if (ret != -EEXIST) {
 857                                 free_extent_map(em);
 858                                 break;
 859                         }
 860                         btrfs_drop_extent_cache(inode, start,
 861                                                 start + ram_size - 1, 0);
 862                 }
 863
 864                 cur_alloc_size = ins.offset;
 865                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 866                                                ram_size, cur_alloc_size, 0);
 867                 BUG_ON(ret);
 868
 869                 if (root->root_key.objectid ==
 870                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 871                         ret = btrfs_reloc_clone_csums(inode, start,
 872                                                       cur_alloc_size);
 873                         BUG_ON(ret);
 874                 }
 875
 876                 if (disk_num_bytes < cur_alloc_size)
 877                         break;
 878
 879                 /* we're not doing compressed IO, don't unlock the first
 880                  * page (which the caller expects to stay locked), don't
 881                  * clear any dirty bits and don't set any writeback bits
 882                  *
 883                  * Do set the Private2 bit so we know this page was properly
 884                  * setup for writepage
 885                  */
 886                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 887                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 888                         EXTENT_SET_PRIVATE2;
 889
 890                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 891                                              start, start + ram_size - 1,
 892                                              locked_page, op);
 893                 disk_num_bytes -= cur_alloc_size;
 894                 num_bytes -= cur_alloc_size;
 895                 alloc_hint = ins.objectid + ins.offset;
 896                 start += cur_alloc_size;
 897         }
 898 out:
 899         ret = 0;
 900         btrfs_end_transaction(trans, root);
 901
 902         return ret;
 903 }
 904
 905 /*
 906  * work queue call back to started compression on a file and pages
 907  */
 908 static noinline void async_cow_start(struct btrfs_work *work)
 909 {
 910         struct async_cow *async_cow;
 911         int num_added = 0;
 912         async_cow = container_of(work, struct async_cow, work);
 913
 914         compress_file_range(async_cow->inode, async_cow->locked_page,
 915                             async_cow->start, async_cow->end, async_cow,
 916                             &num_added);
 917         if (num_added == 0)
 918                 async_cow->inode = NULL;
 919 }
 920
 921 /*
 922  * work queue call back to submit previously compressed pages
 923  */
 924 static noinline void async_cow_submit(struct btrfs_work *work)
 925 {
 926         struct async_cow *async_cow;
 927         struct btrfs_root *root;
 928         unsigned long nr_pages;
 929
 930         async_cow = container_of(work, struct async_cow, work);
 931
 932         root = async_cow->root;
 933         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 934                 PAGE_CACHE_SHIFT;
 935
 936         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 937
 938         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 939             5 * 1042 * 1024 &&
 940             waitqueue_active(&root->fs_info->async_submit_wait))
 941                 wake_up(&root->fs_info->async_submit_wait);
 942
 943         if (async_cow->inode)
 944                 submit_compressed_extents(async_cow->inode, async_cow);
 945 }
 946
 947 static noinline void async_cow_free(struct btrfs_work *work)
 948 {
 949         struct async_cow *async_cow;
 950         async_cow = container_of(work, struct async_cow, work);
 951         kfree(async_cow);
 952 }
 953
 954 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 955                                 u64 start, u64 end, int *page_started,
 956                                 unsigned long *nr_written)
 957 {
 958         struct async_cow *async_cow;
 959         struct btrfs_root *root = BTRFS_I(inode)->root;
 960         unsigned long nr_pages;
 961         u64 cur_end;
 962         int limit = 10 * 1024 * 1042;
 963
 964         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 965                          1, 0, NULL, GFP_NOFS);
 966         while (start < end) {
 967                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 968                 BUG_ON(!async_cow);
 969                 async_cow->inode = inode;
 970                 async_cow->root = root;
 971                 async_cow->locked_page = locked_page;
 972                 async_cow->start = start;
 973
 974                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 975                         cur_end = end;
 976                 else
 977                         cur_end = min(end, start + 512 * 1024 - 1);
 978
 979                 async_cow->end = cur_end;
 980                 INIT_LIST_HEAD(&async_cow->extents);
 981
 982                 async_cow->work.func = async_cow_start;
 983                 async_cow->work.ordered_func = async_cow_submit;
 984                 async_cow->work.ordered_free = async_cow_free;
 985                 async_cow->work.flags = 0;
 986
 987                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 988                         PAGE_CACHE_SHIFT;
 989                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 990
 991                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 992                                    &async_cow->work);
 993
 994                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 995                         wait_event(root->fs_info->async_submit_wait,
 996                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 997                             limit));
 998                 }
 999
1000                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1001                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1002                         wait_event(root->fs_info->async_submit_wait,
1003                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1004                            0));
1005                 }
1006
1007                 *nr_written += nr_pages;
1008                 start = cur_end + 1;
1009         }
1010         *page_started = 1;
1011         return 0;
1012 }
1013
1014 static noinline int csum_exist_in_range(struct btrfs_root *root,
1015                                         u64 bytenr, u64 num_bytes)
1016 {
1017         int ret;
1018         struct btrfs_ordered_sum *sums;
1019         LIST_HEAD(list);
1020
1021         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1022                                        bytenr + num_bytes - 1, &list, 0);
1023         if (ret == 0 && list_empty(&list))
1024                 return 0;
1025
1026         while (!list_empty(&list)) {
1027                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1028                 list_del(&sums->list);
1029                 kfree(sums);
1030         }
1031         return 1;
1032 }
1033
1034 /*
1035  * when nowcow writeback call back.  This checks for snapshots or COW copies
1036  * of the extents that exist in the file, and COWs the file as required.
1037  *
1038  * If no cow copies or snapshots exist, we write directly to the existing
1039  * blocks on disk
1040  */
1041 static noinline int run_delalloc_nocow(struct inode *inode,
1042                                        struct page *locked_page,
1043                               u64 start, u64 end, int *page_started, int force,
1044                               unsigned long *nr_written)
1045 {
1046         struct btrfs_root *root = BTRFS_I(inode)->root;
1047         struct btrfs_trans_handle *trans;
1048         struct extent_buffer *leaf;
1049         struct btrfs_path *path;
1050         struct btrfs_file_extent_item *fi;
1051         struct btrfs_key found_key;
1052         u64 cow_start;
1053         u64 cur_offset;
1054         u64 extent_end;
1055         u64 extent_offset;
1056         u64 disk_bytenr;
1057         u64 num_bytes;
1058         int extent_type;
1059         int ret;
1060         int type;
1061         int nocow;
1062         int check_prev = 1;
1063         bool nolock;
1064         u64 ino = btrfs_ino(inode);
1065
1066         path = btrfs_alloc_path();
1067         BUG_ON(!path);
1068
1069         nolock = is_free_space_inode(root, inode);
1070
1071         if (nolock)
1072                 trans = btrfs_join_transaction_nolock(root, 1);
1073         else
1074                 trans = btrfs_join_transaction(root, 1);
1075         BUG_ON(IS_ERR(trans));
1076
1077         cow_start = (u64)-1;
1078         cur_offset = start;
1079         while (1) {
1080                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1081                                                cur_offset, 0);
1082                 BUG_ON(ret < 0);
1083                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1084                         leaf = path->nodes[0];
1085                         btrfs_item_key_to_cpu(leaf, &found_key,
1086                                               path->slots[0] - 1);
1087                         if (found_key.objectid == ino &&
1088                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1089                                 path->slots[0]--;
1090                 }
1091                 check_prev = 0;
1092 next_slot:
1093                 leaf = path->nodes[0];
1094                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1095                         ret = btrfs_next_leaf(root, path);
1096                         if (ret < 0)
1097                                 BUG_ON(1);
1098                         if (ret > 0)
1099                                 break;
1100                         leaf = path->nodes[0];
1101                 }
1102
1103                 nocow = 0;
1104                 disk_bytenr = 0;
1105                 num_bytes = 0;
1106                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1107
1108                 if (found_key.objectid > ino ||
1109                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1110                     found_key.offset > end)
1111                         break;
1112
1113                 if (found_key.offset > cur_offset) {
1114                         extent_end = found_key.offset;
1115                         extent_type = 0;
1116                         goto out_check;
1117                 }
1118
1119                 fi = btrfs_item_ptr(leaf, path->slots[0],
1120                                     struct btrfs_file_extent_item);
1121                 extent_type = btrfs_file_extent_type(leaf, fi);
1122
1123                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1124                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1125                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1126                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1127                         extent_end = found_key.offset +
1128                                 btrfs_file_extent_num_bytes(leaf, fi);
1129                         if (extent_end <= start) {
1130                                 path->slots[0]++;
1131                                 goto next_slot;
1132                         }
1133                         if (disk_bytenr == 0)
1134                                 goto out_check;
1135                         if (btrfs_file_extent_compression(leaf, fi) ||
1136                             btrfs_file_extent_encryption(leaf, fi) ||
1137                             btrfs_file_extent_other_encoding(leaf, fi))
1138                                 goto out_check;
1139                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1140                                 goto out_check;
1141                         if (btrfs_extent_readonly(root, disk_bytenr))
1142                                 goto out_check;
1143                         if (btrfs_cross_ref_exist(trans, root, ino,
1144                                                   found_key.offset -
1145                                                   extent_offset, disk_bytenr))
1146                                 goto out_check;
1147                         disk_bytenr += extent_offset;
1148                         disk_bytenr += cur_offset - found_key.offset;
1149                         num_bytes = min(end + 1, extent_end) - cur_offset;
1150                         /*
1151                          * force cow if csum exists in the range.
1152                          * this ensure that csum for a given extent are
1153                          * either valid or do not exist.
1154                          */
1155                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1156                                 goto out_check;
1157                         nocow = 1;
1158                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1159                         extent_end = found_key.offset +
1160                                 btrfs_file_extent_inline_len(leaf, fi);
1161                         extent_end = ALIGN(extent_end, root->sectorsize);
1162                 } else {
1163                         BUG_ON(1);
1164                 }
1165 out_check:
1166                 if (extent_end <= start) {
1167                         path->slots[0]++;
1168                         goto next_slot;
1169                 }
1170                 if (!nocow) {
1171                         if (cow_start == (u64)-1)
1172                                 cow_start = cur_offset;
1173                         cur_offset = extent_end;
1174                         if (cur_offset > end)
1175                                 break;
1176                         path->slots[0]++;
1177                         goto next_slot;
1178                 }
1179
1180                 btrfs_release_path(path);
1181                 if (cow_start != (u64)-1) {
1182                         ret = cow_file_range(inode, locked_page, cow_start,
1183                                         found_key.offset - 1, page_started,
1184                                         nr_written, 1);
1185                         BUG_ON(ret);
1186                         cow_start = (u64)-1;
1187                 }
1188
1189                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1190                         struct extent_map *em;
1191                         struct extent_map_tree *em_tree;
1192                         em_tree = &BTRFS_I(inode)->extent_tree;
1193                         em = alloc_extent_map();
1194                         BUG_ON(!em);
1195                         em->start = cur_offset;
1196                         em->orig_start = em->start;
1197                         em->len = num_bytes;
1198                         em->block_len = num_bytes;
1199                         em->block_start = disk_bytenr;
1200                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1201                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1202                         while (1) {
1203                                 write_lock(&em_tree->lock);
1204                                 ret = add_extent_mapping(em_tree, em);
1205                                 write_unlock(&em_tree->lock);
1206                                 if (ret != -EEXIST) {
1207                                         free_extent_map(em);
1208                                         break;
1209                                 }
1210                                 btrfs_drop_extent_cache(inode, em->start,
1211                                                 em->start + em->len - 1, 0);
1212                         }
1213                         type = BTRFS_ORDERED_PREALLOC;
1214                 } else {
1215                         type = BTRFS_ORDERED_NOCOW;
1216                 }
1217
1218                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1219                                                num_bytes, num_bytes, type);
1220                 BUG_ON(ret);
1221
1222                 if (root->root_key.objectid ==
1223                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1224                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1225                                                       num_bytes);
1226                         BUG_ON(ret);
1227                 }
1228
1229                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1230                                 cur_offset, cur_offset + num_bytes - 1,
1231                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1232                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1233                                 EXTENT_SET_PRIVATE2);
1234                 cur_offset = extent_end;
1235                 if (cur_offset > end)
1236                         break;
1237         }
1238         btrfs_release_path(path);
1239
1240         if (cur_offset <= end && cow_start == (u64)-1)
1241                 cow_start = cur_offset;
1242         if (cow_start != (u64)-1) {
1243                 ret = cow_file_range(inode, locked_page, cow_start, end,
1244                                      page_started, nr_written, 1);
1245                 BUG_ON(ret);
1246         }
1247
1248         if (nolock) {
1249                 ret = btrfs_end_transaction_nolock(trans, root);
1250                 BUG_ON(ret);
1251         } else {
1252                 ret = btrfs_end_transaction(trans, root);
1253                 BUG_ON(ret);
1254         }
1255         btrfs_free_path(path);
1256         return 0;
1257 }
1258
1259 /*
1260  * extent_io.c call back to do delayed allocation processing
1261  */
1262 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1263                               u64 start, u64 end, int *page_started,
1264                               unsigned long *nr_written)
1265 {
1266         int ret;
1267         struct btrfs_root *root = BTRFS_I(inode)->root;
1268
1269         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1270                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1271                                          page_started, 1, nr_written);
1272         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1273                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1274                                          page_started, 0, nr_written);
1275         else if (!btrfs_test_opt(root, COMPRESS) &&
1276                  !(BTRFS_I(inode)->force_compress) &&
1277                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1278                 ret = cow_file_range(inode, locked_page, start, end,
1279                                       page_started, nr_written, 1);
1280         else
1281                 ret = cow_file_range_async(inode, locked_page, start, end,
1282                                            page_started, nr_written);
1283         return ret;
1284 }
1285
1286 static int btrfs_split_extent_hook(struct inode *inode,
1287                                    struct extent_state *orig, u64 split)
1288 {
1289         /* not delalloc, ignore it */
1290         if (!(orig->state & EXTENT_DELALLOC))
1291                 return 0;
1292
1293         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1294         return 0;
1295 }
1296
1297 /*
1298  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1299  * extents so we can keep track of new extents that are just merged onto old
1300  * extents, such as when we are doing sequential writes, so we can properly
1301  * account for the metadata space we'll need.
1302  */
1303 static int btrfs_merge_extent_hook(struct inode *inode,
1304                                    struct extent_state *new,
1305                                    struct extent_state *other)
1306 {
1307         /* not delalloc, ignore it */
1308         if (!(other->state & EXTENT_DELALLOC))
1309                 return 0;
1310
1311         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1312         return 0;
1313 }
1314
1315 /*
1316  * extent_io.c set_bit_hook, used to track delayed allocation
1317  * bytes in this file, and to maintain the list of inodes that
1318  * have pending delalloc work to be done.
1319  */
1320 static int btrfs_set_bit_hook(struct inode *inode,
1321                               struct extent_state *state, int *bits)
1322 {
1323
1324         /*
1325          * set_bit and clear bit hooks normally require _irqsave/restore
1326          * but in this case, we are only testeing for the DELALLOC
1327          * bit, which is only set or cleared with irqs on
1328          */
1329         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1330                 struct btrfs_root *root = BTRFS_I(inode)->root;
1331                 u64 len = state->end + 1 - state->start;
1332                 bool do_list = !is_free_space_inode(root, inode);
1333
1334                 if (*bits & EXTENT_FIRST_DELALLOC)
1335                         *bits &= ~EXTENT_FIRST_DELALLOC;
1336                 else
1337                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1338
1339                 spin_lock(&root->fs_info->delalloc_lock);
1340                 BTRFS_I(inode)->delalloc_bytes += len;
1341                 root->fs_info->delalloc_bytes += len;
1342                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1343                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1344                                       &root->fs_info->delalloc_inodes);
1345                 }
1346                 spin_unlock(&root->fs_info->delalloc_lock);
1347         }
1348         return 0;
1349 }
1350
1351 /*
1352  * extent_io.c clear_bit_hook, see set_bit_hook for why
1353  */
1354 static int btrfs_clear_bit_hook(struct inode *inode,
1355                                 struct extent_state *state, int *bits)
1356 {
1357         /*
1358          * set_bit and clear bit hooks normally require _irqsave/restore
1359          * but in this case, we are only testeing for the DELALLOC
1360          * bit, which is only set or cleared with irqs on
1361          */
1362         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1363                 struct btrfs_root *root = BTRFS_I(inode)->root;
1364                 u64 len = state->end + 1 - state->start;
1365                 bool do_list = !is_free_space_inode(root, inode);
1366
1367                 if (*bits & EXTENT_FIRST_DELALLOC)
1368                         *bits &= ~EXTENT_FIRST_DELALLOC;
1369                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1370                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1371
1372                 if (*bits & EXTENT_DO_ACCOUNTING)
1373                         btrfs_delalloc_release_metadata(inode, len);
1374
1375                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1376                     && do_list)
1377                         btrfs_free_reserved_data_space(inode, len);
1378
1379                 spin_lock(&root->fs_info->delalloc_lock);
1380                 root->fs_info->delalloc_bytes -= len;
1381                 BTRFS_I(inode)->delalloc_bytes -= len;
1382
1383                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1384                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1385                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1386                 }
1387                 spin_unlock(&root->fs_info->delalloc_lock);
1388         }
1389         return 0;
1390 }
1391
1392 /*
1393  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1394  * we don't create bios that span stripes or chunks
1395  */
1396 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1397                          size_t size, struct bio *bio,
1398                          unsigned long bio_flags)
1399 {
1400         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1401         struct btrfs_mapping_tree *map_tree;
1402         u64 logical = (u64)bio->bi_sector << 9;
1403         u64 length = 0;
1404         u64 map_length;
1405         int ret;
1406
1407         if (bio_flags & EXTENT_BIO_COMPRESSED)
1408                 return 0;
1409
1410         length = bio->bi_size;
1411         map_tree = &root->fs_info->mapping_tree;
1412         map_length = length;
1413         ret = btrfs_map_block(map_tree, READ, logical,
1414                               &map_length, NULL, 0);
1415
1416         if (map_length < length + size)
1417                 return 1;
1418         return ret;
1419 }
1420
1421 /*
1422  * in order to insert checksums into the metadata in large chunks,
1423  * we wait until bio submission time.   All the pages in the bio are
1424  * checksummed and sums are attached onto the ordered extent record.
1425  *
1426  * At IO completion time the cums attached on the ordered extent record
1427  * are inserted into the btree
1428  */
1429 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1430                                     struct bio *bio, int mirror_num,
1431                                     unsigned long bio_flags,
1432                                     u64 bio_offset)
1433 {
1434         struct btrfs_root *root = BTRFS_I(inode)->root;
1435         int ret = 0;
1436
1437         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1438         BUG_ON(ret);
1439         return 0;
1440 }
1441
1442 /*
1443  * in order to insert checksums into the metadata in large chunks,
1444  * we wait until bio submission time.   All the pages in the bio are
1445  * checksummed and sums are attached onto the ordered extent record.
1446  *
1447  * At IO completion time the cums attached on the ordered extent record
1448  * are inserted into the btree
1449  */
1450 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1451                           int mirror_num, unsigned long bio_flags,
1452                           u64 bio_offset)
1453 {
1454         struct btrfs_root *root = BTRFS_I(inode)->root;
1455         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1456 }
1457
1458 /*
1459  * extent_io.c submission hook. This does the right thing for csum calculation
1460  * on write, or reading the csums from the tree before a read
1461  */
1462 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1463                           int mirror_num, unsigned long bio_flags,
1464                           u64 bio_offset)
1465 {
1466         struct btrfs_root *root = BTRFS_I(inode)->root;
1467         int ret = 0;
1468         int skip_sum;
1469
1470         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1471
1472         if (is_free_space_inode(root, inode))
1473                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1474         else
1475                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1476         BUG_ON(ret);
1477
1478         if (!(rw & REQ_WRITE)) {
1479                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1480                         return btrfs_submit_compressed_read(inode, bio,
1481                                                     mirror_num, bio_flags);
1482                 } else if (!skip_sum) {
1483                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1484                         if (ret)
1485                                 return ret;
1486                 }
1487                 goto mapit;
1488         } else if (!skip_sum) {
1489                 /* csum items have already been cloned */
1490                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1491                         goto mapit;
1492                 /* we're doing a write, do the async checksumming */
1493                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1494                                    inode, rw, bio, mirror_num,
1495                                    bio_flags, bio_offset,
1496                                    __btrfs_submit_bio_start,
1497                                    __btrfs_submit_bio_done);
1498         }
1499
1500 mapit:
1501         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1502 }
1503
1504 /*
1505  * given a list of ordered sums record them in the inode.  This happens
1506  * at IO completion time based on sums calculated at bio submission time.
1507  */
1508 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1509                              struct inode *inode, u64 file_offset,
1510                              struct list_head *list)
1511 {
1512         struct btrfs_ordered_sum *sum;
1513
1514         btrfs_set_trans_block_group(trans, inode);
1515
1516         list_for_each_entry(sum, list, list) {
1517                 btrfs_csum_file_blocks(trans,
1518                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1519         }
1520         return 0;
1521 }
1522
1523 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1524                               struct extent_state **cached_state)
1525 {
1526         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1527                 WARN_ON(1);
1528         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1529                                    cached_state, GFP_NOFS);
1530 }
1531
1532 /* see btrfs_writepage_start_hook for details on why this is required */
1533 struct btrfs_writepage_fixup {
1534         struct page *page;
1535         struct btrfs_work work;
1536 };
1537
1538 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1539 {
1540         struct btrfs_writepage_fixup *fixup;
1541         struct btrfs_ordered_extent *ordered;
1542         struct extent_state *cached_state = NULL;
1543         struct page *page;
1544         struct inode *inode;
1545         u64 page_start;
1546         u64 page_end;
1547
1548         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1549         page = fixup->page;
1550 again:
1551         lock_page(page);
1552         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1553                 ClearPageChecked(page);
1554                 goto out_page;
1555         }
1556
1557         inode = page->mapping->host;
1558         page_start = page_offset(page);
1559         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1560
1561         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1562                          &cached_state, GFP_NOFS);
1563
1564         /* already ordered? We're done */
1565         if (PagePrivate2(page))
1566                 goto out;
1567
1568         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1569         if (ordered) {
1570                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1571                                      page_end, &cached_state, GFP_NOFS);
1572                 unlock_page(page);
1573                 btrfs_start_ordered_extent(inode, ordered, 1);
1574                 goto again;
1575         }
1576
1577         BUG();
1578         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1579         ClearPageChecked(page);
1580 out:
1581         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1582                              &cached_state, GFP_NOFS);
1583 out_page:
1584         unlock_page(page);
1585         page_cache_release(page);
1586         kfree(fixup);
1587 }
1588
1589 /*
1590  * There are a few paths in the higher layers of the kernel that directly
1591  * set the page dirty bit without asking the filesystem if it is a
1592  * good idea.  This causes problems because we want to make sure COW
1593  * properly happens and the data=ordered rules are followed.
1594  *
1595  * In our case any range that doesn't have the ORDERED bit set
1596  * hasn't been properly setup for IO.  We kick off an async process
1597  * to fix it up.  The async helper will wait for ordered extents, set
1598  * the delalloc bit and make it safe to write the page.
1599  */
1600 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1601 {
1602         struct inode *inode = page->mapping->host;
1603         struct btrfs_writepage_fixup *fixup;
1604         struct btrfs_root *root = BTRFS_I(inode)->root;
1605
1606         /* this page is properly in the ordered list */
1607         if (TestClearPagePrivate2(page))
1608                 return 0;
1609
1610         if (PageChecked(page))
1611                 return -EAGAIN;
1612
1613         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1614         if (!fixup)
1615                 return -EAGAIN;
1616
1617         SetPageChecked(page);
1618         page_cache_get(page);
1619         fixup->work.func = btrfs_writepage_fixup_worker;
1620         fixup->page = page;
1621         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1622         return -EAGAIN;
1623 }
1624
1625 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1626                                        struct inode *inode, u64 file_pos,
1627                                        u64 disk_bytenr, u64 disk_num_bytes,
1628                                        u64 num_bytes, u64 ram_bytes,
1629                                        u8 compression, u8 encryption,
1630                                        u16 other_encoding, int extent_type)
1631 {
1632         struct btrfs_root *root = BTRFS_I(inode)->root;
1633         struct btrfs_file_extent_item *fi;
1634         struct btrfs_path *path;
1635         struct extent_buffer *leaf;
1636         struct btrfs_key ins;
1637         u64 hint;
1638         int ret;
1639
1640         path = btrfs_alloc_path();
1641         BUG_ON(!path);
1642
1643         path->leave_spinning = 1;
1644
1645         /*
1646          * we may be replacing one extent in the tree with another.
1647          * The new extent is pinned in the extent map, and we don't want
1648          * to drop it from the cache until it is completely in the btree.
1649          *
1650          * So, tell btrfs_drop_extents to leave this extent in the cache.
1651          * the caller is expected to unpin it and allow it to be merged
1652          * with the others.
1653          */
1654         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1655                                  &hint, 0);
1656         BUG_ON(ret);
1657
1658         ins.objectid = btrfs_ino(inode);
1659         ins.offset = file_pos;
1660         ins.type = BTRFS_EXTENT_DATA_KEY;
1661         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1662         BUG_ON(ret);
1663         leaf = path->nodes[0];
1664         fi = btrfs_item_ptr(leaf, path->slots[0],
1665                             struct btrfs_file_extent_item);
1666         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1667         btrfs_set_file_extent_type(leaf, fi, extent_type);
1668         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1669         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1670         btrfs_set_file_extent_offset(leaf, fi, 0);
1671         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1672         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1673         btrfs_set_file_extent_compression(leaf, fi, compression);
1674         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1675         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1676
1677         btrfs_unlock_up_safe(path, 1);
1678         btrfs_set_lock_blocking(leaf);
1679
1680         btrfs_mark_buffer_dirty(leaf);
1681
1682         inode_add_bytes(inode, num_bytes);
1683
1684         ins.objectid = disk_bytenr;
1685         ins.offset = disk_num_bytes;
1686         ins.type = BTRFS_EXTENT_ITEM_KEY;
1687         ret = btrfs_alloc_reserved_file_extent(trans, root,
1688                                         root->root_key.objectid,
1689                                         btrfs_ino(inode), file_pos, &ins);
1690         BUG_ON(ret);
1691         btrfs_free_path(path);
1692
1693         return 0;
1694 }
1695
1696 /*
1697  * helper function for btrfs_finish_ordered_io, this
1698  * just reads in some of the csum leaves to prime them into ram
1699  * before we start the transaction.  It limits the amount of btree
1700  * reads required while inside the transaction.
1701  */
1702 /* as ordered data IO finishes, this gets called so we can finish
1703  * an ordered extent if the range of bytes in the file it covers are
1704  * fully written.
1705  */
1706 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1707 {
1708         struct btrfs_root *root = BTRFS_I(inode)->root;
1709         struct btrfs_trans_handle *trans = NULL;
1710         struct btrfs_ordered_extent *ordered_extent = NULL;
1711         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1712         struct extent_state *cached_state = NULL;
1713         int compress_type = 0;
1714         int ret;
1715         bool nolock;
1716
1717         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1718                                              end - start + 1);
1719         if (!ret)
1720                 return 0;
1721         BUG_ON(!ordered_extent);
1722
1723         nolock = is_free_space_inode(root, inode);
1724
1725         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1726                 BUG_ON(!list_empty(&ordered_extent->list));
1727                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1728                 if (!ret) {
1729                         if (nolock)
1730                                 trans = btrfs_join_transaction_nolock(root, 1);
1731                         else
1732                                 trans = btrfs_join_transaction(root, 1);
1733                         BUG_ON(IS_ERR(trans));
1734                         btrfs_set_trans_block_group(trans, inode);
1735                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1736                         ret = btrfs_update_inode(trans, root, inode);
1737                         BUG_ON(ret);
1738                 }
1739                 goto out;
1740         }
1741
1742         lock_extent_bits(io_tree, ordered_extent->file_offset,
1743                          ordered_extent->file_offset + ordered_extent->len - 1,
1744                          0, &cached_state, GFP_NOFS);
1745
1746         if (nolock)
1747                 trans = btrfs_join_transaction_nolock(root, 1);
1748         else
1749                 trans = btrfs_join_transaction(root, 1);
1750         BUG_ON(IS_ERR(trans));
1751         btrfs_set_trans_block_group(trans, inode);
1752         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1753
1754         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1755                 compress_type = ordered_extent->compress_type;
1756         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1757                 BUG_ON(compress_type);
1758                 ret = btrfs_mark_extent_written(trans, inode,
1759                                                 ordered_extent->file_offset,
1760                                                 ordered_extent->file_offset +
1761                                                 ordered_extent->len);
1762                 BUG_ON(ret);
1763         } else {
1764                 BUG_ON(root == root->fs_info->tree_root);
1765                 ret = insert_reserved_file_extent(trans, inode,
1766                                                 ordered_extent->file_offset,
1767                                                 ordered_extent->start,
1768                                                 ordered_extent->disk_len,
1769                                                 ordered_extent->len,
1770                                                 ordered_extent->len,
1771                                                 compress_type, 0, 0,
1772                                                 BTRFS_FILE_EXTENT_REG);
1773                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1774                                    ordered_extent->file_offset,
1775                                    ordered_extent->len);
1776                 BUG_ON(ret);
1777         }
1778         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1779                              ordered_extent->file_offset +
1780                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1781
1782         add_pending_csums(trans, inode, ordered_extent->file_offset,
1783                           &ordered_extent->list);
1784
1785         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1786         if (!ret) {
1787                 ret = btrfs_update_inode(trans, root, inode);
1788                 BUG_ON(ret);
1789         }
1790         ret = 0;
1791 out:
1792         if (nolock) {
1793                 if (trans)
1794                         btrfs_end_transaction_nolock(trans, root);
1795         } else {
1796                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1797                 if (trans)
1798                         btrfs_end_transaction(trans, root);
1799         }
1800
1801         /* once for us */
1802         btrfs_put_ordered_extent(ordered_extent);
1803         /* once for the tree */
1804         btrfs_put_ordered_extent(ordered_extent);
1805
1806         return 0;
1807 }
1808
1809 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1810                                 struct extent_state *state, int uptodate)
1811 {
1812         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1813
1814         ClearPagePrivate2(page);
1815         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1816 }
1817
1818 /*
1819  * When IO fails, either with EIO or csum verification fails, we
1820  * try other mirrors that might have a good copy of the data.  This
1821  * io_failure_record is used to record state as we go through all the
1822  * mirrors.  If another mirror has good data, the page is set up to date
1823  * and things continue.  If a good mirror can't be found, the original
1824  * bio end_io callback is called to indicate things have failed.
1825  */
1826 struct io_failure_record {
1827         struct page *page;
1828         u64 start;
1829         u64 len;
1830         u64 logical;
1831         unsigned long bio_flags;
1832         int last_mirror;
1833 };
1834
1835 static int btrfs_io_failed_hook(struct bio *failed_bio,
1836                          struct page *page, u64 start, u64 end,
1837                          struct extent_state *state)
1838 {
1839         struct io_failure_record *failrec = NULL;
1840         u64 private;
1841         struct extent_map *em;
1842         struct inode *inode = page->mapping->host;
1843         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1844         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1845         struct bio *bio;
1846         int num_copies;
1847         int ret;
1848         int rw;
1849         u64 logical;
1850
1851         ret = get_state_private(failure_tree, start, &private);
1852         if (ret) {
1853                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1854                 if (!failrec)
1855                         return -ENOMEM;
1856                 failrec->start = start;
1857                 failrec->len = end - start + 1;
1858                 failrec->last_mirror = 0;
1859                 failrec->bio_flags = 0;
1860
1861                 read_lock(&em_tree->lock);
1862                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1863                 if (em->start > start || em->start + em->len < start) {
1864                         free_extent_map(em);
1865                         em = NULL;
1866                 }
1867                 read_unlock(&em_tree->lock);
1868
1869                 if (IS_ERR_OR_NULL(em)) {
1870                         kfree(failrec);
1871                         return -EIO;
1872                 }
1873                 logical = start - em->start;
1874                 logical = em->block_start + logical;
1875                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1876                         logical = em->block_start;
1877                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1878                         extent_set_compress_type(&failrec->bio_flags,
1879                                                  em->compress_type);
1880                 }
1881                 failrec->logical = logical;
1882                 free_extent_map(em);
1883                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1884                                 EXTENT_DIRTY, GFP_NOFS);
1885                 set_state_private(failure_tree, start,
1886                                  (u64)(unsigned long)failrec);
1887         } else {
1888                 failrec = (struct io_failure_record *)(unsigned long)private;
1889         }
1890         num_copies = btrfs_num_copies(
1891                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1892                               failrec->logical, failrec->len);
1893         failrec->last_mirror++;
1894         if (!state) {
1895                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1896                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1897                                                     failrec->start,
1898                                                     EXTENT_LOCKED);
1899                 if (state && state->start != failrec->start)
1900                         state = NULL;
1901                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1902         }
1903         if (!state || failrec->last_mirror > num_copies) {
1904                 set_state_private(failure_tree, failrec->start, 0);
1905                 clear_extent_bits(failure_tree, failrec->start,
1906                                   failrec->start + failrec->len - 1,
1907                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1908                 kfree(failrec);
1909                 return -EIO;
1910         }
1911         bio = bio_alloc(GFP_NOFS, 1);
1912         bio->bi_private = state;
1913         bio->bi_end_io = failed_bio->bi_end_io;
1914         bio->bi_sector = failrec->logical >> 9;
1915         bio->bi_bdev = failed_bio->bi_bdev;
1916         bio->bi_size = 0;
1917
1918         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1919         if (failed_bio->bi_rw & REQ_WRITE)
1920                 rw = WRITE;
1921         else
1922                 rw = READ;
1923
1924         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1925                                                       failrec->last_mirror,
1926                                                       failrec->bio_flags, 0);
1927         return ret;
1928 }
1929
1930 /*
1931  * each time an IO finishes, we do a fast check in the IO failure tree
1932  * to see if we need to process or clean up an io_failure_record
1933  */
1934 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1935 {
1936         u64 private;
1937         u64 private_failure;
1938         struct io_failure_record *failure;
1939         int ret;
1940
1941         private = 0;
1942         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1943                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1944                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1945                                         start, &private_failure);
1946                 if (ret == 0) {
1947                         failure = (struct io_failure_record *)(unsigned long)
1948                                    private_failure;
1949                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1950                                           failure->start, 0);
1951                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1952                                           failure->start,
1953                                           failure->start + failure->len - 1,
1954                                           EXTENT_DIRTY | EXTENT_LOCKED,
1955                                           GFP_NOFS);
1956                         kfree(failure);
1957                 }
1958         }
1959         return 0;
1960 }
1961
1962 /*
1963  * when reads are done, we need to check csums to verify the data is correct
1964  * if there's a match, we allow the bio to finish.  If not, we go through
1965  * the io_failure_record routines to find good copies
1966  */
1967 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1968                                struct extent_state *state)
1969 {
1970         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1971         struct inode *inode = page->mapping->host;
1972         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1973         char *kaddr;
1974         u64 private = ~(u32)0;
1975         int ret;
1976         struct btrfs_root *root = BTRFS_I(inode)->root;
1977         u32 csum = ~(u32)0;
1978
1979         if (PageChecked(page)) {
1980                 ClearPageChecked(page);
1981                 goto good;
1982         }
1983
1984         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1985                 return 0;
1986
1987         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1988             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1989                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1990                                   GFP_NOFS);
1991                 return 0;
1992         }
1993
1994         if (state && state->start == start) {
1995                 private = state->private;
1996                 ret = 0;
1997         } else {
1998                 ret = get_state_private(io_tree, start, &private);
1999         }
2000         kaddr = kmap_atomic(page, KM_USER0);
2001         if (ret)
2002                 goto zeroit;
2003
2004         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2005         btrfs_csum_final(csum, (char *)&csum);
2006         if (csum != private)
2007                 goto zeroit;
2008
2009         kunmap_atomic(kaddr, KM_USER0);
2010 good:
2011         /* if the io failure tree for this inode is non-empty,
2012          * check to see if we've recovered from a failed IO
2013          */
2014         btrfs_clean_io_failures(inode, start);
2015         return 0;
2016
2017 zeroit:
2018         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2019                        "private %llu\n",
2020                        (unsigned long long)btrfs_ino(page->mapping->host),
2021                        (unsigned long long)start, csum,
2022                        (unsigned long long)private);
2023         memset(kaddr + offset, 1, end - start + 1);
2024         flush_dcache_page(page);
2025         kunmap_atomic(kaddr, KM_USER0);
2026         if (private == 0)
2027                 return 0;
2028         return -EIO;
2029 }
2030
2031 struct delayed_iput {
2032         struct list_head list;
2033         struct inode *inode;
2034 };
2035
2036 void btrfs_add_delayed_iput(struct inode *inode)
2037 {
2038         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2039         struct delayed_iput *delayed;
2040
2041         if (atomic_add_unless(&inode->i_count, -1, 1))
2042                 return;
2043
2044         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2045         delayed->inode = inode;
2046
2047         spin_lock(&fs_info->delayed_iput_lock);
2048         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2049         spin_unlock(&fs_info->delayed_iput_lock);
2050 }
2051
2052 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2053 {
2054         LIST_HEAD(list);
2055         struct btrfs_fs_info *fs_info = root->fs_info;
2056         struct delayed_iput *delayed;
2057         int empty;
2058
2059         spin_lock(&fs_info->delayed_iput_lock);
2060         empty = list_empty(&fs_info->delayed_iputs);
2061         spin_unlock(&fs_info->delayed_iput_lock);
2062         if (empty)
2063                 return;
2064
2065         down_read(&root->fs_info->cleanup_work_sem);
2066         spin_lock(&fs_info->delayed_iput_lock);
2067         list_splice_init(&fs_info->delayed_iputs, &list);
2068         spin_unlock(&fs_info->delayed_iput_lock);
2069
2070         while (!list_empty(&list)) {
2071                 delayed = list_entry(list.next, struct delayed_iput, list);
2072                 list_del(&delayed->list);
2073                 iput(delayed->inode);
2074                 kfree(delayed);
2075         }
2076         up_read(&root->fs_info->cleanup_work_sem);
2077 }
2078
2079 /*
2080  * calculate extra metadata reservation when snapshotting a subvolume
2081  * contains orphan files.
2082  */
2083 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2084                                 struct btrfs_pending_snapshot *pending,
2085                                 u64 *bytes_to_reserve)
2086 {
2087         struct btrfs_root *root;
2088         struct btrfs_block_rsv *block_rsv;
2089         u64 num_bytes;
2090         int index;
2091
2092         root = pending->root;
2093         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2094                 return;
2095
2096         block_rsv = root->orphan_block_rsv;
2097
2098         /* orphan block reservation for the snapshot */
2099         num_bytes = block_rsv->size;
2100
2101         /*
2102          * after the snapshot is created, COWing tree blocks may use more
2103          * space than it frees. So we should make sure there is enough
2104          * reserved space.
2105          */
2106         index = trans->transid & 0x1;
2107         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2108                 num_bytes += block_rsv->size -
2109                              (block_rsv->reserved + block_rsv->freed[index]);
2110         }
2111
2112         *bytes_to_reserve += num_bytes;
2113 }
2114
2115 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2116                                 struct btrfs_pending_snapshot *pending)
2117 {
2118         struct btrfs_root *root = pending->root;
2119         struct btrfs_root *snap = pending->snap;
2120         struct btrfs_block_rsv *block_rsv;
2121         u64 num_bytes;
2122         int index;
2123         int ret;
2124
2125         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2126                 return;
2127
2128         /* refill source subvolume's orphan block reservation */
2129         block_rsv = root->orphan_block_rsv;
2130         index = trans->transid & 0x1;
2131         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2132                 num_bytes = block_rsv->size -
2133                             (block_rsv->reserved + block_rsv->freed[index]);
2134                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2135                                               root->orphan_block_rsv,
2136                                               num_bytes);
2137                 BUG_ON(ret);
2138         }
2139
2140         /* setup orphan block reservation for the snapshot */
2141         block_rsv = btrfs_alloc_block_rsv(snap);
2142         BUG_ON(!block_rsv);
2143
2144         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2145         snap->orphan_block_rsv = block_rsv;
2146
2147         num_bytes = root->orphan_block_rsv->size;
2148         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2149                                       block_rsv, num_bytes);
2150         BUG_ON(ret);
2151
2152 #if 0
2153         /* insert orphan item for the snapshot */
2154         WARN_ON(!root->orphan_item_inserted);
2155         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2156                                        snap->root_key.objectid);
2157         BUG_ON(ret);
2158         snap->orphan_item_inserted = 1;
2159 #endif
2160 }
2161
2162 enum btrfs_orphan_cleanup_state {
2163         ORPHAN_CLEANUP_STARTED  = 1,
2164         ORPHAN_CLEANUP_DONE     = 2,
2165 };
2166
2167 /*
2168  * This is called in transaction commmit time. If there are no orphan
2169  * files in the subvolume, it removes orphan item and frees block_rsv
2170  * structure.
2171  */
2172 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2173                               struct btrfs_root *root)
2174 {
2175         int ret;
2176
2177         if (!list_empty(&root->orphan_list) ||
2178             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2179                 return;
2180
2181         if (root->orphan_item_inserted &&
2182             btrfs_root_refs(&root->root_item) > 0) {
2183                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2184                                             root->root_key.objectid);
2185                 BUG_ON(ret);
2186                 root->orphan_item_inserted = 0;
2187         }
2188
2189         if (root->orphan_block_rsv) {
2190                 WARN_ON(root->orphan_block_rsv->size > 0);
2191                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2192                 root->orphan_block_rsv = NULL;
2193         }
2194 }
2195
2196 /*
2197  * This creates an orphan entry for the given inode in case something goes
2198  * wrong in the middle of an unlink/truncate.
2199  *
2200  * NOTE: caller of this function should reserve 5 units of metadata for
2201  *       this function.
2202  */
2203 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2204 {
2205         struct btrfs_root *root = BTRFS_I(inode)->root;
2206         struct btrfs_block_rsv *block_rsv = NULL;
2207         int reserve = 0;
2208         int insert = 0;
2209         int ret;
2210
2211         if (!root->orphan_block_rsv) {
2212                 block_rsv = btrfs_alloc_block_rsv(root);
2213                 BUG_ON(!block_rsv);
2214         }
2215
2216         spin_lock(&root->orphan_lock);
2217         if (!root->orphan_block_rsv) {
2218                 root->orphan_block_rsv = block_rsv;
2219         } else if (block_rsv) {
2220                 btrfs_free_block_rsv(root, block_rsv);
2221                 block_rsv = NULL;
2222         }
2223
2224         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2225                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2226 #if 0
2227                 /*
2228                  * For proper ENOSPC handling, we should do orphan
2229                  * cleanup when mounting. But this introduces backward
2230                  * compatibility issue.
2231                  */
2232                 if (!xchg(&root->orphan_item_inserted, 1))
2233                         insert = 2;
2234                 else
2235                         insert = 1;
2236 #endif
2237                 insert = 1;
2238         }
2239
2240         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2241                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2242                 reserve = 1;
2243         }
2244         spin_unlock(&root->orphan_lock);
2245
2246         if (block_rsv)
2247                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2248
2249         /* grab metadata reservation from transaction handle */
2250         if (reserve) {
2251                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2252                 BUG_ON(ret);
2253         }
2254
2255         /* insert an orphan item to track this unlinked/truncated file */
2256         if (insert >= 1) {
2257                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2258                 BUG_ON(ret);
2259         }
2260
2261         /* insert an orphan item to track subvolume contains orphan files */
2262         if (insert >= 2) {
2263                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2264                                                root->root_key.objectid);
2265                 BUG_ON(ret);
2266         }
2267         return 0;
2268 }
2269
2270 /*
2271  * We have done the truncate/delete so we can go ahead and remove the orphan
2272  * item for this particular inode.
2273  */
2274 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2275 {
2276         struct btrfs_root *root = BTRFS_I(inode)->root;
2277         int delete_item = 0;
2278         int release_rsv = 0;
2279         int ret = 0;
2280
2281         spin_lock(&root->orphan_lock);
2282         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2283                 list_del_init(&BTRFS_I(inode)->i_orphan);
2284                 delete_item = 1;
2285         }
2286
2287         if (BTRFS_I(inode)->orphan_meta_reserved) {
2288                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2289                 release_rsv = 1;
2290         }
2291         spin_unlock(&root->orphan_lock);
2292
2293         if (trans && delete_item) {
2294                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2295                 BUG_ON(ret);
2296         }
2297
2298         if (release_rsv)
2299                 btrfs_orphan_release_metadata(inode);
2300
2301         return 0;
2302 }
2303
2304 /*
2305  * this cleans up any orphans that may be left on the list from the last use
2306  * of this root.
2307  */
2308 int btrfs_orphan_cleanup(struct btrfs_root *root)
2309 {
2310         struct btrfs_path *path;
2311         struct extent_buffer *leaf;
2312         struct btrfs_key key, found_key;
2313         struct btrfs_trans_handle *trans;
2314         struct inode *inode;
2315         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2316
2317         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2318                 return 0;
2319
2320         path = btrfs_alloc_path();
2321         if (!path) {
2322                 ret = -ENOMEM;
2323                 goto out;
2324         }
2325         path->reada = -1;
2326
2327         key.objectid = BTRFS_ORPHAN_OBJECTID;
2328         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2329         key.offset = (u64)-1;
2330
2331         while (1) {
2332                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2333                 if (ret < 0)
2334                         goto out;
2335
2336                 /*
2337                  * if ret == 0 means we found what we were searching for, which
2338                  * is weird, but possible, so only screw with path if we didn't
2339                  * find the key and see if we have stuff that matches
2340                  */
2341                 if (ret > 0) {
2342                         ret = 0;
2343                         if (path->slots[0] == 0)
2344                                 break;
2345                         path->slots[0]--;
2346                 }
2347
2348                 /* pull out the item */
2349                 leaf = path->nodes[0];
2350                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2351
2352                 /* make sure the item matches what we want */
2353                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2354                         break;
2355                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2356                         break;
2357
2358                 /* release the path since we're done with it */
2359                 btrfs_release_path(path);
2360
2361                 /*
2362                  * this is where we are basically btrfs_lookup, without the
2363                  * crossing root thing.  we store the inode number in the
2364                  * offset of the orphan item.
2365                  */
2366                 found_key.objectid = found_key.offset;
2367                 found_key.type = BTRFS_INODE_ITEM_KEY;
2368                 found_key.offset = 0;
2369                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2370                 if (IS_ERR(inode)) {
2371                         ret = PTR_ERR(inode);
2372                         goto out;
2373                 }
2374
2375                 /*
2376                  * add this inode to the orphan list so btrfs_orphan_del does
2377                  * the proper thing when we hit it
2378                  */
2379                 spin_lock(&root->orphan_lock);
2380                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2381                 spin_unlock(&root->orphan_lock);
2382
2383                 /*
2384                  * if this is a bad inode, means we actually succeeded in
2385                  * removing the inode, but not the orphan record, which means
2386                  * we need to manually delete the orphan since iput will just
2387                  * do a destroy_inode
2388                  */
2389                 if (is_bad_inode(inode)) {
2390                         trans = btrfs_start_transaction(root, 0);
2391                         if (IS_ERR(trans)) {
2392                                 ret = PTR_ERR(trans);
2393                                 goto out;
2394                         }
2395                         btrfs_orphan_del(trans, inode);
2396                         btrfs_end_transaction(trans, root);
2397                         iput(inode);
2398                         continue;
2399                 }
2400
2401                 /* if we have links, this was a truncate, lets do that */
2402                 if (inode->i_nlink) {
2403                         if (!S_ISREG(inode->i_mode)) {
2404                                 WARN_ON(1);
2405                                 iput(inode);
2406                                 continue;
2407                         }
2408                         nr_truncate++;
2409                         ret = btrfs_truncate(inode);
2410                 } else {
2411                         nr_unlink++;
2412                 }
2413
2414                 /* this will do delete_inode and everything for us */
2415                 iput(inode);
2416                 if (ret)
2417                         goto out;
2418         }
2419         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2420
2421         if (root->orphan_block_rsv)
2422                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2423                                         (u64)-1);
2424
2425         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2426                 trans = btrfs_join_transaction(root, 1);
2427                 if (!IS_ERR(trans))
2428                         btrfs_end_transaction(trans, root);
2429         }
2430
2431         if (nr_unlink)
2432                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2433         if (nr_truncate)
2434                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2435
2436 out:
2437         if (ret)
2438                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2439         btrfs_free_path(path);
2440         return ret;
2441 }
2442
2443 /*
2444  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2445  * don't find any xattrs, we know there can't be any acls.
2446  *
2447  * slot is the slot the inode is in, objectid is the objectid of the inode
2448  */
2449 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2450                                           int slot, u64 objectid)
2451 {
2452         u32 nritems = btrfs_header_nritems(leaf);
2453         struct btrfs_key found_key;
2454         int scanned = 0;
2455
2456         slot++;
2457         while (slot < nritems) {
2458                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2459
2460                 /* we found a different objectid, there must not be acls */
2461                 if (found_key.objectid != objectid)
2462                         return 0;
2463
2464                 /* we found an xattr, assume we've got an acl */
2465                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2466                         return 1;
2467
2468                 /*
2469                  * we found a key greater than an xattr key, there can't
2470                  * be any acls later on
2471                  */
2472                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2473                         return 0;
2474
2475                 slot++;
2476                 scanned++;
2477
2478                 /*
2479                  * it goes inode, inode backrefs, xattrs, extents,
2480                  * so if there are a ton of hard links to an inode there can
2481                  * be a lot of backrefs.  Don't waste time searching too hard,
2482                  * this is just an optimization
2483                  */
2484                 if (scanned >= 8)
2485                         break;
2486         }
2487         /* we hit the end of the leaf before we found an xattr or
2488          * something larger than an xattr.  We have to assume the inode
2489          * has acls
2490          */
2491         return 1;
2492 }
2493
2494 /*
2495  * read an inode from the btree into the in-memory inode
2496  */
2497 static void btrfs_read_locked_inode(struct inode *inode)
2498 {
2499         struct btrfs_path *path;
2500         struct extent_buffer *leaf;
2501         struct btrfs_inode_item *inode_item;
2502         struct btrfs_timespec *tspec;
2503         struct btrfs_root *root = BTRFS_I(inode)->root;
2504         struct btrfs_key location;
2505         int maybe_acls;
2506         u64 alloc_group_block;
2507         u32 rdev;
2508         int ret;
2509
2510         path = btrfs_alloc_path();
2511         BUG_ON(!path);
2512         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2513
2514         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2515         if (ret)
2516                 goto make_bad;
2517
2518         leaf = path->nodes[0];
2519         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2520                                     struct btrfs_inode_item);
2521
2522         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2523         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2524         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2525         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2526         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2527
2528         tspec = btrfs_inode_atime(inode_item);
2529         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2530         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2531
2532         tspec = btrfs_inode_mtime(inode_item);
2533         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2534         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2535
2536         tspec = btrfs_inode_ctime(inode_item);
2537         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2538         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2539
2540         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2541         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2542         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2543         inode->i_generation = BTRFS_I(inode)->generation;
2544         inode->i_rdev = 0;
2545         rdev = btrfs_inode_rdev(leaf, inode_item);
2546
2547         BTRFS_I(inode)->index_cnt = (u64)-1;
2548         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2549
2550         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2551
2552         /*
2553          * try to precache a NULL acl entry for files that don't have
2554          * any xattrs or acls
2555          */
2556         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2557                                            btrfs_ino(inode));
2558         if (!maybe_acls)
2559                 cache_no_acl(inode);
2560
2561         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2562                                                 alloc_group_block, 0);
2563         btrfs_free_path(path);
2564         inode_item = NULL;
2565
2566         switch (inode->i_mode & S_IFMT) {
2567         case S_IFREG:
2568                 inode->i_mapping->a_ops = &btrfs_aops;
2569                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2570                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2571                 inode->i_fop = &btrfs_file_operations;
2572                 inode->i_op = &btrfs_file_inode_operations;
2573                 break;
2574         case S_IFDIR:
2575                 inode->i_fop = &btrfs_dir_file_operations;
2576                 if (root == root->fs_info->tree_root)
2577                         inode->i_op = &btrfs_dir_ro_inode_operations;
2578                 else
2579                         inode->i_op = &btrfs_dir_inode_operations;
2580                 break;
2581         case S_IFLNK:
2582                 inode->i_op = &btrfs_symlink_inode_operations;
2583                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2584                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2585                 break;
2586         default:
2587                 inode->i_op = &btrfs_special_inode_operations;
2588                 init_special_inode(inode, inode->i_mode, rdev);
2589                 break;
2590         }
2591
2592         btrfs_update_iflags(inode);
2593         return;
2594
2595 make_bad:
2596         btrfs_free_path(path);
2597         make_bad_inode(inode);
2598 }
2599
2600 /*
2601  * given a leaf and an inode, copy the inode fields into the leaf
2602  */
2603 static void fill_inode_item(struct btrfs_trans_handle *trans,
2604                             struct extent_buffer *leaf,
2605                             struct btrfs_inode_item *item,
2606                             struct inode *inode)
2607 {
2608         if (!leaf->map_token)
2609                 map_private_extent_buffer(leaf, (unsigned long)item,
2610                                           sizeof(struct btrfs_inode_item),
2611                                           &leaf->map_token, &leaf->kaddr,
2612                                           &leaf->map_start, &leaf->map_len,
2613                                           KM_USER1);
2614
2615         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2616         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2617         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2618         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2619         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2620
2621         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2622                                inode->i_atime.tv_sec);
2623         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2624                                 inode->i_atime.tv_nsec);
2625
2626         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2627                                inode->i_mtime.tv_sec);
2628         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2629                                 inode->i_mtime.tv_nsec);
2630
2631         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2632                                inode->i_ctime.tv_sec);
2633         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2634                                 inode->i_ctime.tv_nsec);
2635
2636         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2637         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2638         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2639         btrfs_set_inode_transid(leaf, item, trans->transid);
2640         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2641         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2642         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2643
2644         if (leaf->map_token) {
2645                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2646                 leaf->map_token = NULL;
2647         }
2648 }
2649
2650 /*
2651  * copy everything in the in-memory inode into the btree.
2652  */
2653 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2654                                 struct btrfs_root *root, struct inode *inode)
2655 {
2656         struct btrfs_inode_item *inode_item;
2657         struct btrfs_path *path;
2658         struct extent_buffer *leaf;
2659         int ret;
2660
2661         /*
2662          * If root is tree root, it means this inode is used to
2663          * store free space information. And these inodes are updated
2664          * when committing the transaction, so they needn't delaye to
2665          * be updated, or deadlock will occured.
2666          */
2667         if (!is_free_space_inode(root, inode)) {
2668                 ret = btrfs_delayed_update_inode(trans, root, inode);
2669                 if (!ret)
2670                         btrfs_set_inode_last_trans(trans, inode);
2671                 return ret;
2672         }
2673
2674         path = btrfs_alloc_path();
2675         if (!path)
2676                 return -ENOMEM;
2677
2678         path->leave_spinning = 1;
2679         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2680                                  1);
2681         if (ret) {
2682                 if (ret > 0)
2683                         ret = -ENOENT;
2684                 goto failed;
2685         }
2686
2687         btrfs_unlock_up_safe(path, 1);
2688         leaf = path->nodes[0];
2689         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2690                                     struct btrfs_inode_item);
2691
2692         fill_inode_item(trans, leaf, inode_item, inode);
2693         btrfs_mark_buffer_dirty(leaf);
2694         btrfs_set_inode_last_trans(trans, inode);
2695         ret = 0;
2696 failed:
2697         btrfs_free_path(path);
2698         return ret;
2699 }
2700
2701 /*
2702  * unlink helper that gets used here in inode.c and in the tree logging
2703  * recovery code.  It remove a link in a directory with a given name, and
2704  * also drops the back refs in the inode to the directory
2705  */
2706 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2707                                 struct btrfs_root *root,
2708                                 struct inode *dir, struct inode *inode,
2709                                 const char *name, int name_len)
2710 {
2711         struct btrfs_path *path;
2712         int ret = 0;
2713         struct extent_buffer *leaf;
2714         struct btrfs_dir_item *di;
2715         struct btrfs_key key;
2716         u64 index;
2717         u64 ino = btrfs_ino(inode);
2718         u64 dir_ino = btrfs_ino(dir);
2719
2720         path = btrfs_alloc_path();
2721         if (!path) {
2722                 ret = -ENOMEM;
2723                 goto out;
2724         }
2725
2726         path->leave_spinning = 1;
2727         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2728                                     name, name_len, -1);
2729         if (IS_ERR(di)) {
2730                 ret = PTR_ERR(di);
2731                 goto err;
2732         }
2733         if (!di) {
2734                 ret = -ENOENT;
2735                 goto err;
2736         }
2737         leaf = path->nodes[0];
2738         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2739         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2740         if (ret)
2741                 goto err;
2742         btrfs_release_path(path);
2743
2744         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2745                                   dir_ino, &index);
2746         if (ret) {
2747                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2748                        "inode %llu parent %llu\n", name_len, name,
2749                        (unsigned long long)ino, (unsigned long long)dir_ino);
2750                 goto err;
2751         }
2752
2753         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2754         if (ret)
2755                 goto err;
2756
2757         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2758                                          inode, dir_ino);
2759         BUG_ON(ret != 0 && ret != -ENOENT);
2760
2761         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2762                                            dir, index);
2763         if (ret == -ENOENT)
2764                 ret = 0;
2765 err:
2766         btrfs_free_path(path);
2767         if (ret)
2768                 goto out;
2769
2770         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2771         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2772         btrfs_update_inode(trans, root, dir);
2773 out:
2774         return ret;
2775 }
2776
2777 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2778                        struct btrfs_root *root,
2779                        struct inode *dir, struct inode *inode,
2780                        const char *name, int name_len)
2781 {
2782         int ret;
2783         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2784         if (!ret) {
2785                 btrfs_drop_nlink(inode);
2786                 ret = btrfs_update_inode(trans, root, inode);
2787         }
2788         return ret;
2789 }
2790
2791
2792 /* helper to check if there is any shared block in the path */
2793 static int check_path_shared(struct btrfs_root *root,
2794                              struct btrfs_path *path)
2795 {
2796         struct extent_buffer *eb;
2797         int level;
2798         u64 refs = 1;
2799
2800         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2801                 int ret;
2802
2803                 if (!path->nodes[level])
2804                         break;
2805                 eb = path->nodes[level];
2806                 if (!btrfs_block_can_be_shared(root, eb))
2807                         continue;
2808                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2809                                                &refs, NULL);
2810                 if (refs > 1)
2811                         return 1;
2812         }
2813         return 0;
2814 }
2815
2816 /*
2817  * helper to start transaction for unlink and rmdir.
2818  *
2819  * unlink and rmdir are special in btrfs, they do not always free space.
2820  * so in enospc case, we should make sure they will free space before
2821  * allowing them to use the global metadata reservation.
2822  */
2823 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2824                                                        struct dentry *dentry)
2825 {
2826         struct btrfs_trans_handle *trans;
2827         struct btrfs_root *root = BTRFS_I(dir)->root;
2828         struct btrfs_path *path;
2829         struct btrfs_inode_ref *ref;
2830         struct btrfs_dir_item *di;
2831         struct inode *inode = dentry->d_inode;
2832         u64 index;
2833         int check_link = 1;
2834         int err = -ENOSPC;
2835         int ret;
2836         u64 ino = btrfs_ino(inode);
2837         u64 dir_ino = btrfs_ino(dir);
2838
2839         trans = btrfs_start_transaction(root, 10);
2840         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2841                 return trans;
2842
2843         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2844                 return ERR_PTR(-ENOSPC);
2845
2846         /* check if there is someone else holds reference */
2847         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2848                 return ERR_PTR(-ENOSPC);
2849
2850         if (atomic_read(&inode->i_count) > 2)
2851                 return ERR_PTR(-ENOSPC);
2852
2853         if (xchg(&root->fs_info->enospc_unlink, 1))
2854                 return ERR_PTR(-ENOSPC);
2855
2856         path = btrfs_alloc_path();
2857         if (!path) {
2858                 root->fs_info->enospc_unlink = 0;
2859                 return ERR_PTR(-ENOMEM);
2860         }
2861
2862         trans = btrfs_start_transaction(root, 0);
2863         if (IS_ERR(trans)) {
2864                 btrfs_free_path(path);
2865                 root->fs_info->enospc_unlink = 0;
2866                 return trans;
2867         }
2868
2869         path->skip_locking = 1;
2870         path->search_commit_root = 1;
2871
2872         ret = btrfs_lookup_inode(trans, root, path,
2873                                 &BTRFS_I(dir)->location, 0);
2874         if (ret < 0) {
2875                 err = ret;
2876                 goto out;
2877         }
2878         if (ret == 0) {
2879                 if (check_path_shared(root, path))
2880                         goto out;
2881         } else {
2882                 check_link = 0;
2883         }
2884         btrfs_release_path(path);
2885
2886         ret = btrfs_lookup_inode(trans, root, path,
2887                                 &BTRFS_I(inode)->location, 0);
2888         if (ret < 0) {
2889                 err = ret;
2890                 goto out;
2891         }
2892         if (ret == 0) {
2893                 if (check_path_shared(root, path))
2894                         goto out;
2895         } else {
2896                 check_link = 0;
2897         }
2898         btrfs_release_path(path);
2899
2900         if (ret == 0 && S_ISREG(inode->i_mode)) {
2901                 ret = btrfs_lookup_file_extent(trans, root, path,
2902                                                ino, (u64)-1, 0);
2903                 if (ret < 0) {
2904                         err = ret;
2905                         goto out;
2906                 }
2907                 BUG_ON(ret == 0);
2908                 if (check_path_shared(root, path))
2909                         goto out;
2910                 btrfs_release_path(path);
2911         }
2912
2913         if (!check_link) {
2914                 err = 0;
2915                 goto out;
2916         }
2917
2918         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2919                                 dentry->d_name.name, dentry->d_name.len, 0);
2920         if (IS_ERR(di)) {
2921                 err = PTR_ERR(di);
2922                 goto out;
2923         }
2924         if (di) {
2925                 if (check_path_shared(root, path))
2926                         goto out;
2927         } else {
2928                 err = 0;
2929                 goto out;
2930         }
2931         btrfs_release_path(path);
2932
2933         ref = btrfs_lookup_inode_ref(trans, root, path,
2934                                 dentry->d_name.name, dentry->d_name.len,
2935                                 ino, dir_ino, 0);
2936         if (IS_ERR(ref)) {
2937                 err = PTR_ERR(ref);
2938                 goto out;
2939         }
2940         BUG_ON(!ref);
2941         if (check_path_shared(root, path))
2942                 goto out;
2943         index = btrfs_inode_ref_index(path->nodes[0], ref);
2944         btrfs_release_path(path);
2945
2946         /*
2947          * This is a commit root search, if we can lookup inode item and other
2948          * relative items in the commit root, it means the transaction of
2949          * dir/file creation has been committed, and the dir index item that we
2950          * delay to insert has also been inserted into the commit root. So
2951          * we needn't worry about the delayed insertion of the dir index item
2952          * here.
2953          */
2954         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2955                                 dentry->d_name.name, dentry->d_name.len, 0);
2956         if (IS_ERR(di)) {
2957                 err = PTR_ERR(di);
2958                 goto out;
2959         }
2960         BUG_ON(ret == -ENOENT);
2961         if (check_path_shared(root, path))
2962                 goto out;
2963
2964         err = 0;
2965 out:
2966         btrfs_free_path(path);
2967         if (err) {
2968                 btrfs_end_transaction(trans, root);
2969                 root->fs_info->enospc_unlink = 0;
2970                 return ERR_PTR(err);
2971         }
2972
2973         trans->block_rsv = &root->fs_info->global_block_rsv;
2974         return trans;
2975 }
2976
2977 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2978                                struct btrfs_root *root)
2979 {
2980         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2981                 BUG_ON(!root->fs_info->enospc_unlink);
2982                 root->fs_info->enospc_unlink = 0;
2983         }
2984         btrfs_end_transaction_throttle(trans, root);
2985 }
2986
2987 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2988 {
2989         struct btrfs_root *root = BTRFS_I(dir)->root;
2990         struct btrfs_trans_handle *trans;
2991         struct inode *inode = dentry->d_inode;
2992         int ret;
2993         unsigned long nr = 0;
2994
2995         trans = __unlink_start_trans(dir, dentry);
2996         if (IS_ERR(trans))
2997                 return PTR_ERR(trans);
2998
2999         btrfs_set_trans_block_group(trans, dir);
3000
3001         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3002
3003         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3004                                  dentry->d_name.name, dentry->d_name.len);
3005         BUG_ON(ret);
3006
3007         if (inode->i_nlink == 0) {
3008                 ret = btrfs_orphan_add(trans, inode);
3009                 BUG_ON(ret);
3010         }
3011
3012         nr = trans->blocks_used;
3013         __unlink_end_trans(trans, root);
3014         btrfs_btree_balance_dirty(root, nr);
3015         return ret;
3016 }
3017
3018 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3019                         struct btrfs_root *root,
3020                         struct inode *dir, u64 objectid,
3021                         const char *name, int name_len)
3022 {
3023         struct btrfs_path *path;
3024         struct extent_buffer *leaf;
3025         struct btrfs_dir_item *di;
3026         struct btrfs_key key;
3027         u64 index;
3028         int ret;
3029         u64 dir_ino = btrfs_ino(dir);
3030
3031         path = btrfs_alloc_path();
3032         if (!path)
3033                 return -ENOMEM;
3034
3035         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3036                                    name, name_len, -1);
3037         BUG_ON(IS_ERR_OR_NULL(di));
3038
3039         leaf = path->nodes[0];
3040         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3041         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3042         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3043         BUG_ON(ret);
3044         btrfs_release_path(path);
3045
3046         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3047                                  objectid, root->root_key.objectid,
3048                                  dir_ino, &index, name, name_len);
3049         if (ret < 0) {
3050                 BUG_ON(ret != -ENOENT);
3051                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3052                                                  name, name_len);
3053                 BUG_ON(IS_ERR_OR_NULL(di));
3054
3055                 leaf = path->nodes[0];
3056                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3057                 btrfs_release_path(path);
3058                 index = key.offset;
3059         }
3060         btrfs_release_path(path);
3061
3062         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3063         BUG_ON(ret);
3064
3065         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3066         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3067         ret = btrfs_update_inode(trans, root, dir);
3068         BUG_ON(ret);
3069
3070         return 0;
3071 }
3072
3073 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3074 {
3075         struct inode *inode = dentry->d_inode;
3076         int err = 0;
3077         struct btrfs_root *root = BTRFS_I(dir)->root;
3078         struct btrfs_trans_handle *trans;
3079         unsigned long nr = 0;
3080
3081         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3082             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3083                 return -ENOTEMPTY;
3084
3085         trans = __unlink_start_trans(dir, dentry);
3086         if (IS_ERR(trans))
3087                 return PTR_ERR(trans);
3088
3089         btrfs_set_trans_block_group(trans, dir);
3090
3091         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3092                 err = btrfs_unlink_subvol(trans, root, dir,
3093                                           BTRFS_I(inode)->location.objectid,
3094                                           dentry->d_name.name,
3095                                           dentry->d_name.len);
3096                 goto out;
3097         }
3098
3099         err = btrfs_orphan_add(trans, inode);
3100         if (err)
3101                 goto out;
3102
3103         /* now the directory is empty */
3104         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3105                                  dentry->d_name.name, dentry->d_name.len);
3106         if (!err)
3107                 btrfs_i_size_write(inode, 0);
3108 out:
3109         nr = trans->blocks_used;
3110         __unlink_end_trans(trans, root);
3111         btrfs_btree_balance_dirty(root, nr);
3112
3113         return err;
3114 }
3115
3116 /*
3117  * this can truncate away extent items, csum items and directory items.
3118  * It starts at a high offset and removes keys until it can't find
3119  * any higher than new_size
3120  *
3121  * csum items that cross the new i_size are truncated to the new size
3122  * as well.
3123  *
3124  * min_type is the minimum key type to truncate down to.  If set to 0, this
3125  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3126  */
3127 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3128                                struct btrfs_root *root,
3129                                struct inode *inode,
3130                                u64 new_size, u32 min_type)
3131 {
3132         struct btrfs_path *path;
3133         struct extent_buffer *leaf;
3134         struct btrfs_file_extent_item *fi;
3135         struct btrfs_key key;
3136         struct btrfs_key found_key;
3137         u64 extent_start = 0;
3138         u64 extent_num_bytes = 0;
3139         u64 extent_offset = 0;
3140         u64 item_end = 0;
3141         u64 mask = root->sectorsize - 1;
3142         u32 found_type = (u8)-1;
3143         int found_extent;
3144         int del_item;
3145         int pending_del_nr = 0;
3146         int pending_del_slot = 0;
3147         int extent_type = -1;
3148         int encoding;
3149         int ret;
3150         int err = 0;
3151         u64 ino = btrfs_ino(inode);
3152
3153         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3154
3155         if (root->ref_cows || root == root->fs_info->tree_root)
3156                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3157
3158         /*
3159          * This function is also used to drop the items in the log tree before
3160          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3161          * it is used to drop the loged items. So we shouldn't kill the delayed
3162          * items.
3163          */
3164         if (min_type == 0 && root == BTRFS_I(inode)->root)
3165                 btrfs_kill_delayed_inode_items(inode);
3166
3167         path = btrfs_alloc_path();
3168         BUG_ON(!path);
3169         path->reada = -1;
3170
3171         key.objectid = ino;
3172         key.offset = (u64)-1;
3173         key.type = (u8)-1;
3174
3175 search_again:
3176         path->leave_spinning = 1;
3177         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3178         if (ret < 0) {
3179                 err = ret;
3180                 goto out;
3181         }
3182
3183         if (ret > 0) {
3184                 /* there are no items in the tree for us to truncate, we're
3185                  * done
3186                  */
3187                 if (path->slots[0] == 0)
3188                         goto out;
3189                 path->slots[0]--;
3190         }
3191
3192         while (1) {
3193                 fi = NULL;
3194                 leaf = path->nodes[0];
3195                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3196                 found_type = btrfs_key_type(&found_key);
3197                 encoding = 0;
3198
3199                 if (found_key.objectid != ino)
3200                         break;
3201
3202                 if (found_type < min_type)
3203                         break;
3204
3205                 item_end = found_key.offset;
3206                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3207                         fi = btrfs_item_ptr(leaf, path->slots[0],
3208                                             struct btrfs_file_extent_item);
3209                         extent_type = btrfs_file_extent_type(leaf, fi);
3210                         encoding = btrfs_file_extent_compression(leaf, fi);
3211                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3212                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3213
3214                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3215                                 item_end +=
3216                                     btrfs_file_extent_num_bytes(leaf, fi);
3217                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3218                                 item_end += btrfs_file_extent_inline_len(leaf,
3219                                                                          fi);
3220                         }
3221                         item_end--;
3222                 }
3223                 if (found_type > min_type) {
3224                         del_item = 1;
3225                 } else {
3226                         if (item_end < new_size)
3227                                 break;
3228                         if (found_key.offset >= new_size)
3229                                 del_item = 1;
3230                         else
3231                                 del_item = 0;
3232                 }
3233                 found_extent = 0;
3234                 /* FIXME, shrink the extent if the ref count is only 1 */
3235                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3236                         goto delete;
3237
3238                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3239                         u64 num_dec;
3240                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3241                         if (!del_item && !encoding) {
3242                                 u64 orig_num_bytes =
3243                                         btrfs_file_extent_num_bytes(leaf, fi);
3244                                 extent_num_bytes = new_size -
3245                                         found_key.offset + root->sectorsize - 1;
3246                                 extent_num_bytes = extent_num_bytes &
3247                                         ~((u64)root->sectorsize - 1);
3248                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3249                                                          extent_num_bytes);
3250                                 num_dec = (orig_num_bytes -
3251                                            extent_num_bytes);
3252                                 if (root->ref_cows && extent_start != 0)
3253                                         inode_sub_bytes(inode, num_dec);
3254                                 btrfs_mark_buffer_dirty(leaf);
3255                         } else {
3256                                 extent_num_bytes =
3257                                         btrfs_file_extent_disk_num_bytes(leaf,
3258                                                                          fi);
3259                                 extent_offset = found_key.offset -
3260                                         btrfs_file_extent_offset(leaf, fi);
3261
3262                                 /* FIXME blocksize != 4096 */
3263                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3264                                 if (extent_start != 0) {
3265                                         found_extent = 1;
3266                                         if (root->ref_cows)
3267                                                 inode_sub_bytes(inode, num_dec);
3268                                 }
3269                         }
3270                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3271                         /*
3272                          * we can't truncate inline items that have had
3273                          * special encodings
3274                          */
3275                         if (!del_item &&
3276                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3277                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3278                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3279                                 u32 size = new_size - found_key.offset;
3280
3281                                 if (root->ref_cows) {
3282                                         inode_sub_bytes(inode, item_end + 1 -
3283                                                         new_size);
3284                                 }
3285                                 size =
3286                                     btrfs_file_extent_calc_inline_size(size);
3287                                 ret = btrfs_truncate_item(trans, root, path,
3288                                                           size, 1);
3289                                 BUG_ON(ret);
3290                         } else if (root->ref_cows) {
3291                                 inode_sub_bytes(inode, item_end + 1 -
3292                                                 found_key.offset);
3293                         }
3294                 }
3295 delete:
3296                 if (del_item) {
3297                         if (!pending_del_nr) {
3298                                 /* no pending yet, add ourselves */
3299                                 pending_del_slot = path->slots[0];
3300                                 pending_del_nr = 1;
3301                         } else if (pending_del_nr &&
3302                                    path->slots[0] + 1 == pending_del_slot) {
3303                                 /* hop on the pending chunk */
3304                                 pending_del_nr++;
3305                                 pending_del_slot = path->slots[0];
3306                         } else {
3307                                 BUG();
3308                         }
3309                 } else {
3310                         break;
3311                 }
3312                 if (found_extent && (root->ref_cows ||
3313                                      root == root->fs_info->tree_root)) {
3314                         btrfs_set_path_blocking(path);
3315                         ret = btrfs_free_extent(trans, root, extent_start,
3316                                                 extent_num_bytes, 0,
3317                                                 btrfs_header_owner(leaf),
3318                                                 ino, extent_offset);
3319                         BUG_ON(ret);
3320                 }
3321
3322                 if (found_type == BTRFS_INODE_ITEM_KEY)
3323                         break;
3324
3325                 if (path->slots[0] == 0 ||
3326                     path->slots[0] != pending_del_slot) {
3327                         if (root->ref_cows &&
3328                             BTRFS_I(inode)->location.objectid !=
3329                                                 BTRFS_FREE_INO_OBJECTID) {
3330                                 err = -EAGAIN;
3331                                 goto out;
3332                         }
3333                         if (pending_del_nr) {
3334                                 ret = btrfs_del_items(trans, root, path,
3335                                                 pending_del_slot,
3336                                                 pending_del_nr);
3337                                 BUG_ON(ret);
3338                                 pending_del_nr = 0;
3339                         }
3340                         btrfs_release_path(path);
3341                         goto search_again;
3342                 } else {
3343                         path->slots[0]--;
3344                 }
3345         }
3346 out:
3347         if (pending_del_nr) {
3348                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3349                                       pending_del_nr);
3350                 BUG_ON(ret);
3351         }
3352         btrfs_free_path(path);
3353         return err;
3354 }
3355
3356 /*
3357  * taken from block_truncate_page, but does cow as it zeros out
3358  * any bytes left in the last page in the file.
3359  */
3360 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3361 {
3362         struct inode *inode = mapping->host;
3363         struct btrfs_root *root = BTRFS_I(inode)->root;
3364         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3365         struct btrfs_ordered_extent *ordered;
3366         struct extent_state *cached_state = NULL;
3367         char *kaddr;
3368         u32 blocksize = root->sectorsize;
3369         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3370         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3371         struct page *page;
3372         int ret = 0;
3373         u64 page_start;
3374         u64 page_end;
3375
3376         if ((offset & (blocksize - 1)) == 0)
3377                 goto out;
3378         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3379         if (ret)
3380                 goto out;
3381
3382         ret = -ENOMEM;
3383 again:
3384         page = grab_cache_page(mapping, index);
3385         if (!page) {
3386                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3387                 goto out;
3388         }
3389
3390         page_start = page_offset(page);
3391         page_end = page_start + PAGE_CACHE_SIZE - 1;
3392
3393         if (!PageUptodate(page)) {
3394                 ret = btrfs_readpage(NULL, page);
3395                 lock_page(page);
3396                 if (page->mapping != mapping) {
3397                         unlock_page(page);
3398                         page_cache_release(page);
3399                         goto again;
3400                 }
3401                 if (!PageUptodate(page)) {
3402                         ret = -EIO;
3403                         goto out_unlock;
3404                 }
3405         }
3406         wait_on_page_writeback(page);
3407
3408         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3409                          GFP_NOFS);
3410         set_page_extent_mapped(page);
3411
3412         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3413         if (ordered) {
3414                 unlock_extent_cached(io_tree, page_start, page_end,
3415                                      &cached_state, GFP_NOFS);
3416                 unlock_page(page);
3417                 page_cache_release(page);
3418                 btrfs_start_ordered_extent(inode, ordered, 1);
3419                 btrfs_put_ordered_extent(ordered);
3420                 goto again;
3421         }
3422
3423         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3424                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3425                           0, 0, &cached_state, GFP_NOFS);
3426
3427         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3428                                         &cached_state);
3429         if (ret) {
3430                 unlock_extent_cached(io_tree, page_start, page_end,
3431                                      &cached_state, GFP_NOFS);
3432                 goto out_unlock;
3433         }
3434
3435         ret = 0;
3436         if (offset != PAGE_CACHE_SIZE) {
3437                 kaddr = kmap(page);
3438                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3439                 flush_dcache_page(page);
3440                 kunmap(page);
3441         }
3442         ClearPageChecked(page);
3443         set_page_dirty(page);
3444         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3445                              GFP_NOFS);
3446
3447 out_unlock:
3448         if (ret)
3449                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3450         unlock_page(page);
3451         page_cache_release(page);
3452 out:
3453         return ret;
3454 }
3455
3456 /*
3457  * This function puts in dummy file extents for the area we're creating a hole
3458  * for.  So if we are truncating this file to a larger size we need to insert
3459  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3460  * the range between oldsize and size
3461  */
3462 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3463 {
3464         struct btrfs_trans_handle *trans;
3465         struct btrfs_root *root = BTRFS_I(inode)->root;
3466         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3467         struct extent_map *em = NULL;
3468         struct extent_state *cached_state = NULL;
3469         u64 mask = root->sectorsize - 1;
3470         u64 hole_start = (oldsize + mask) & ~mask;
3471         u64 block_end = (size + mask) & ~mask;
3472         u64 last_byte;
3473         u64 cur_offset;
3474         u64 hole_size;
3475         int err = 0;
3476
3477         if (size <= hole_start)
3478                 return 0;
3479
3480         while (1) {
3481                 struct btrfs_ordered_extent *ordered;
3482                 btrfs_wait_ordered_range(inode, hole_start,
3483                                          block_end - hole_start);
3484                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3485                                  &cached_state, GFP_NOFS);
3486                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3487                 if (!ordered)
3488                         break;
3489                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3490                                      &cached_state, GFP_NOFS);
3491                 btrfs_put_ordered_extent(ordered);
3492         }
3493
3494         cur_offset = hole_start;
3495         while (1) {
3496                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3497                                 block_end - cur_offset, 0);
3498                 BUG_ON(IS_ERR_OR_NULL(em));
3499                 last_byte = min(extent_map_end(em), block_end);
3500                 last_byte = (last_byte + mask) & ~mask;
3501                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3502                         u64 hint_byte = 0;
3503                         hole_size = last_byte - cur_offset;
3504
3505                         trans = btrfs_start_transaction(root, 2);
3506                         if (IS_ERR(trans)) {
3507                                 err = PTR_ERR(trans);
3508                                 break;
3509                         }
3510                         btrfs_set_trans_block_group(trans, inode);
3511
3512                         err = btrfs_drop_extents(trans, inode, cur_offset,
3513                                                  cur_offset + hole_size,
3514                                                  &hint_byte, 1);
3515                         if (err)
3516                                 break;
3517
3518                         err = btrfs_insert_file_extent(trans, root,
3519                                         btrfs_ino(inode), cur_offset, 0,
3520                                         0, hole_size, 0, hole_size,
3521                                         0, 0, 0);
3522                         if (err)
3523                                 break;
3524
3525                         btrfs_drop_extent_cache(inode, hole_start,
3526                                         last_byte - 1, 0);
3527
3528                         btrfs_end_transaction(trans, root);
3529                 }
3530                 free_extent_map(em);
3531                 em = NULL;
3532                 cur_offset = last_byte;
3533                 if (cur_offset >= block_end)
3534                         break;
3535         }
3536
3537         free_extent_map(em);
3538         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3539                              GFP_NOFS);
3540         return err;
3541 }
3542
3543 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3544 {
3545         loff_t oldsize = i_size_read(inode);
3546         int ret;
3547
3548         if (newsize == oldsize)
3549                 return 0;
3550
3551         if (newsize > oldsize) {
3552                 i_size_write(inode, newsize);
3553                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3554                 truncate_pagecache(inode, oldsize, newsize);
3555                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3556                 if (ret) {
3557                         btrfs_setsize(inode, oldsize);
3558                         return ret;
3559                 }
3560
3561                 mark_inode_dirty(inode);
3562         } else {
3563
3564                 /*
3565                  * We're truncating a file that used to have good data down to
3566                  * zero. Make sure it gets into the ordered flush list so that
3567                  * any new writes get down to disk quickly.
3568                  */
3569                 if (newsize == 0)
3570                         BTRFS_I(inode)->ordered_data_close = 1;
3571
3572                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3573                 truncate_setsize(inode, newsize);
3574                 ret = btrfs_truncate(inode);
3575         }
3576
3577         return ret;
3578 }
3579
3580 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3581 {
3582         struct inode *inode = dentry->d_inode;
3583         struct btrfs_root *root = BTRFS_I(inode)->root;
3584         int err;
3585
3586         if (btrfs_root_readonly(root))
3587                 return -EROFS;
3588
3589         err = inode_change_ok(inode, attr);
3590         if (err)
3591                 return err;
3592
3593         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3594                 err = btrfs_setsize(inode, attr->ia_size);
3595                 if (err)
3596                         return err;
3597         }
3598
3599         if (attr->ia_valid) {
3600                 setattr_copy(inode, attr);
3601                 mark_inode_dirty(inode);
3602
3603                 if (attr->ia_valid & ATTR_MODE)
3604                         err = btrfs_acl_chmod(inode);
3605         }
3606
3607         return err;
3608 }
3609
3610 void btrfs_evict_inode(struct inode *inode)
3611 {
3612         struct btrfs_trans_handle *trans;
3613         struct btrfs_root *root = BTRFS_I(inode)->root;
3614         unsigned long nr;
3615         int ret;
3616
3617         trace_btrfs_inode_evict(inode);
3618
3619         truncate_inode_pages(&inode->i_data, 0);
3620         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3621                                is_free_space_inode(root, inode)))
3622                 goto no_delete;
3623
3624         if (is_bad_inode(inode)) {
3625                 btrfs_orphan_del(NULL, inode);
3626                 goto no_delete;
3627         }
3628         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3629         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3630
3631         if (root->fs_info->log_root_recovering) {
3632                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3633                 goto no_delete;
3634         }
3635
3636         if (inode->i_nlink > 0) {
3637                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3638                 goto no_delete;
3639         }
3640
3641         btrfs_i_size_write(inode, 0);
3642
3643         while (1) {
3644                 trans = btrfs_start_transaction(root, 0);
3645                 BUG_ON(IS_ERR(trans));
3646                 btrfs_set_trans_block_group(trans, inode);
3647                 trans->block_rsv = root->orphan_block_rsv;
3648
3649                 ret = btrfs_block_rsv_check(trans, root,
3650                                             root->orphan_block_rsv, 0, 5);
3651                 if (ret) {
3652                         BUG_ON(ret != -EAGAIN);
3653                         ret = btrfs_commit_transaction(trans, root);
3654                         BUG_ON(ret);
3655                         continue;
3656                 }
3657
3658                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3659                 if (ret != -EAGAIN)
3660                         break;
3661
3662                 nr = trans->blocks_used;
3663                 btrfs_end_transaction(trans, root);
3664                 trans = NULL;
3665                 btrfs_btree_balance_dirty(root, nr);
3666
3667         }
3668
3669         if (ret == 0) {
3670                 ret = btrfs_orphan_del(trans, inode);
3671                 BUG_ON(ret);
3672         }
3673
3674         if (!(root == root->fs_info->tree_root ||
3675               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3676                 btrfs_return_ino(root, btrfs_ino(inode));
3677
3678         nr = trans->blocks_used;
3679         btrfs_end_transaction(trans, root);
3680         btrfs_btree_balance_dirty(root, nr);
3681 no_delete:
3682         end_writeback(inode);
3683         return;
3684 }
3685
3686 /*
3687  * this returns the key found in the dir entry in the location pointer.
3688  * If no dir entries were found, location->objectid is 0.
3689  */
3690 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3691                                struct btrfs_key *location)
3692 {
3693         const char *name = dentry->d_name.name;
3694         int namelen = dentry->d_name.len;
3695         struct btrfs_dir_item *di;
3696         struct btrfs_path *path;
3697         struct btrfs_root *root = BTRFS_I(dir)->root;
3698         int ret = 0;
3699
3700         path = btrfs_alloc_path();
3701         BUG_ON(!path);
3702
3703         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3704                                     namelen, 0);
3705         if (IS_ERR(di))
3706                 ret = PTR_ERR(di);
3707
3708         if (IS_ERR_OR_NULL(di))
3709                 goto out_err;
3710
3711         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3712 out:
3713         btrfs_free_path(path);
3714         return ret;
3715 out_err:
3716         location->objectid = 0;
3717         goto out;
3718 }
3719
3720 /*
3721  * when we hit a tree root in a directory, the btrfs part of the inode
3722  * needs to be changed to reflect the root directory of the tree root.  This
3723  * is kind of like crossing a mount point.
3724  */
3725 static int fixup_tree_root_location(struct btrfs_root *root,
3726                                     struct inode *dir,
3727                                     struct dentry *dentry,
3728                                     struct btrfs_key *location,
3729                                     struct btrfs_root **sub_root)
3730 {
3731         struct btrfs_path *path;
3732         struct btrfs_root *new_root;
3733         struct btrfs_root_ref *ref;
3734         struct extent_buffer *leaf;
3735         int ret;
3736         int err = 0;
3737
3738         path = btrfs_alloc_path();
3739         if (!path) {
3740                 err = -ENOMEM;
3741                 goto out;
3742         }
3743
3744         err = -ENOENT;
3745         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3746                                   BTRFS_I(dir)->root->root_key.objectid,
3747                                   location->objectid);
3748         if (ret) {
3749                 if (ret < 0)
3750                         err = ret;
3751                 goto out;
3752         }
3753
3754         leaf = path->nodes[0];
3755         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3756         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3757             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3758                 goto out;
3759
3760         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3761                                    (unsigned long)(ref + 1),
3762                                    dentry->d_name.len);
3763         if (ret)
3764                 goto out;
3765
3766         btrfs_release_path(path);
3767
3768         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3769         if (IS_ERR(new_root)) {
3770                 err = PTR_ERR(new_root);
3771                 goto out;
3772         }
3773
3774         if (btrfs_root_refs(&new_root->root_item) == 0) {
3775                 err = -ENOENT;
3776                 goto out;
3777         }
3778
3779         *sub_root = new_root;
3780         location->objectid = btrfs_root_dirid(&new_root->root_item);
3781         location->type = BTRFS_INODE_ITEM_KEY;
3782         location->offset = 0;
3783         err = 0;
3784 out:
3785         btrfs_free_path(path);
3786         return err;
3787 }
3788
3789 static void inode_tree_add(struct inode *inode)
3790 {
3791         struct btrfs_root *root = BTRFS_I(inode)->root;
3792         struct btrfs_inode *entry;
3793         struct rb_node **p;
3794         struct rb_node *parent;
3795         u64 ino = btrfs_ino(inode);
3796 again:
3797         p = &root->inode_tree.rb_node;
3798         parent = NULL;
3799
3800         if (inode_unhashed(inode))
3801                 return;
3802
3803         spin_lock(&root->inode_lock);
3804         while (*p) {
3805                 parent = *p;
3806                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3807
3808                 if (ino < btrfs_ino(&entry->vfs_inode))
3809                         p = &parent->rb_left;
3810                 else if (ino > btrfs_ino(&entry->vfs_inode))
3811                         p = &parent->rb_right;
3812                 else {
3813                         WARN_ON(!(entry->vfs_inode.i_state &
3814                                   (I_WILL_FREE | I_FREEING)));
3815                         rb_erase(parent, &root->inode_tree);
3816                         RB_CLEAR_NODE(parent);
3817                         spin_unlock(&root->inode_lock);
3818                         goto again;
3819                 }
3820         }
3821         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3822         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3823         spin_unlock(&root->inode_lock);
3824 }
3825
3826 static void inode_tree_del(struct inode *inode)
3827 {
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         int empty = 0;
3830
3831         spin_lock(&root->inode_lock);
3832         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3833                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3834                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3835                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3836         }
3837         spin_unlock(&root->inode_lock);
3838
3839         /*
3840          * Free space cache has inodes in the tree root, but the tree root has a
3841          * root_refs of 0, so this could end up dropping the tree root as a
3842          * snapshot, so we need the extra !root->fs_info->tree_root check to
3843          * make sure we don't drop it.
3844          */
3845         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3846             root != root->fs_info->tree_root) {
3847                 synchronize_srcu(&root->fs_info->subvol_srcu);
3848                 spin_lock(&root->inode_lock);
3849                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3850                 spin_unlock(&root->inode_lock);
3851                 if (empty)
3852                         btrfs_add_dead_root(root);
3853         }
3854 }
3855
3856 int btrfs_invalidate_inodes(struct btrfs_root *root)
3857 {
3858         struct rb_node *node;
3859         struct rb_node *prev;
3860         struct btrfs_inode *entry;
3861         struct inode *inode;
3862         u64 objectid = 0;
3863
3864         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3865
3866         spin_lock(&root->inode_lock);
3867 again:
3868         node = root->inode_tree.rb_node;
3869         prev = NULL;
3870         while (node) {
3871                 prev = node;
3872                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3873
3874                 if (objectid < btrfs_ino(&entry->vfs_inode))
3875                         node = node->rb_left;
3876                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3877                         node = node->rb_right;
3878                 else
3879                         break;
3880         }
3881         if (!node) {
3882                 while (prev) {
3883                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3884                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3885                                 node = prev;
3886                                 break;
3887                         }
3888                         prev = rb_next(prev);
3889                 }
3890         }
3891         while (node) {
3892                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3893                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3894                 inode = igrab(&entry->vfs_inode);
3895                 if (inode) {
3896                         spin_unlock(&root->inode_lock);
3897                         if (atomic_read(&inode->i_count) > 1)
3898                                 d_prune_aliases(inode);
3899                         /*
3900                          * btrfs_drop_inode will have it removed from
3901                          * the inode cache when its usage count
3902                          * hits zero.
3903                          */
3904                         iput(inode);
3905                         cond_resched();
3906                         spin_lock(&root->inode_lock);
3907                         goto again;
3908                 }
3909
3910                 if (cond_resched_lock(&root->inode_lock))
3911                         goto again;
3912
3913                 node = rb_next(node);
3914         }
3915         spin_unlock(&root->inode_lock);
3916         return 0;
3917 }
3918
3919 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3920 {
3921         struct btrfs_iget_args *args = p;
3922         inode->i_ino = args->ino;
3923         BTRFS_I(inode)->root = args->root;
3924         btrfs_set_inode_space_info(args->root, inode);
3925         return 0;
3926 }
3927
3928 static int btrfs_find_actor(struct inode *inode, void *opaque)
3929 {
3930         struct btrfs_iget_args *args = opaque;
3931         return args->ino == btrfs_ino(inode) &&
3932                 args->root == BTRFS_I(inode)->root;
3933 }
3934
3935 static struct inode *btrfs_iget_locked(struct super_block *s,
3936                                        u64 objectid,
3937                                        struct btrfs_root *root)
3938 {
3939         struct inode *inode;
3940         struct btrfs_iget_args args;
3941         args.ino = objectid;
3942         args.root = root;
3943
3944         inode = iget5_locked(s, objectid, btrfs_find_actor,
3945                              btrfs_init_locked_inode,
3946                              (void *)&args);
3947         return inode;
3948 }
3949
3950 /* Get an inode object given its location and corresponding root.
3951  * Returns in *is_new if the inode was read from disk
3952  */
3953 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3954                          struct btrfs_root *root, int *new)
3955 {
3956         struct inode *inode;
3957
3958         inode = btrfs_iget_locked(s, location->objectid, root);
3959         if (!inode)
3960                 return ERR_PTR(-ENOMEM);
3961
3962         if (inode->i_state & I_NEW) {
3963                 BTRFS_I(inode)->root = root;
3964                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3965                 btrfs_read_locked_inode(inode);
3966                 inode_tree_add(inode);
3967                 unlock_new_inode(inode);
3968                 if (new)
3969                         *new = 1;
3970         }
3971
3972         return inode;
3973 }
3974
3975 static struct inode *new_simple_dir(struct super_block *s,
3976                                     struct btrfs_key *key,
3977                                     struct btrfs_root *root)
3978 {
3979         struct inode *inode = new_inode(s);
3980
3981         if (!inode)
3982                 return ERR_PTR(-ENOMEM);
3983
3984         BTRFS_I(inode)->root = root;
3985         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3986         BTRFS_I(inode)->dummy_inode = 1;
3987
3988         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3989         inode->i_op = &simple_dir_inode_operations;
3990         inode->i_fop = &simple_dir_operations;
3991         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3992         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3993
3994         return inode;
3995 }
3996
3997 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3998 {
3999         struct inode *inode;
4000         struct btrfs_root *root = BTRFS_I(dir)->root;
4001         struct btrfs_root *sub_root = root;
4002         struct btrfs_key location;
4003         int index;
4004         int ret;
4005
4006         if (dentry->d_name.len > BTRFS_NAME_LEN)
4007                 return ERR_PTR(-ENAMETOOLONG);
4008
4009         ret = btrfs_inode_by_name(dir, dentry, &location);
4010
4011         if (ret < 0)
4012                 return ERR_PTR(ret);
4013
4014         if (location.objectid == 0)
4015                 return NULL;
4016
4017         if (location.type == BTRFS_INODE_ITEM_KEY) {
4018                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4019                 return inode;
4020         }
4021
4022         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4023
4024         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4025         ret = fixup_tree_root_location(root, dir, dentry,
4026                                        &location, &sub_root);
4027         if (ret < 0) {
4028                 if (ret != -ENOENT)
4029                         inode = ERR_PTR(ret);
4030                 else
4031                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4032         } else {
4033                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4034         }
4035         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4036
4037         if (!IS_ERR(inode) && root != sub_root) {
4038                 down_read(&root->fs_info->cleanup_work_sem);
4039                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4040                         ret = btrfs_orphan_cleanup(sub_root);
4041                 up_read(&root->fs_info->cleanup_work_sem);
4042                 if (ret)
4043                         inode = ERR_PTR(ret);
4044         }
4045
4046         return inode;
4047 }
4048
4049 static int btrfs_dentry_delete(const struct dentry *dentry)
4050 {
4051         struct btrfs_root *root;
4052
4053         if (!dentry->d_inode && !IS_ROOT(dentry))
4054                 dentry = dentry->d_parent;
4055
4056         if (dentry->d_inode) {
4057                 root = BTRFS_I(dentry->d_inode)->root;
4058                 if (btrfs_root_refs(&root->root_item) == 0)
4059                         return 1;
4060         }
4061         return 0;
4062 }
4063
4064 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4065                                    struct nameidata *nd)
4066 {
4067         struct inode *inode;
4068
4069         inode = btrfs_lookup_dentry(dir, dentry);
4070         if (IS_ERR(inode))
4071                 return ERR_CAST(inode);
4072
4073         return d_splice_alias(inode, dentry);
4074 }
4075
4076 unsigned char btrfs_filetype_table[] = {
4077         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4078 };
4079
4080 static int btrfs_real_readdir(struct file *filp, void *dirent,
4081                               filldir_t filldir)
4082 {
4083         struct inode *inode = filp->f_dentry->d_inode;
4084         struct btrfs_root *root = BTRFS_I(inode)->root;
4085         struct btrfs_item *item;
4086         struct btrfs_dir_item *di;
4087         struct btrfs_key key;
4088         struct btrfs_key found_key;
4089         struct btrfs_path *path;
4090         struct list_head ins_list;
4091         struct list_head del_list;
4092         int ret;
4093         struct extent_buffer *leaf;
4094         int slot;
4095         unsigned char d_type;
4096         int over = 0;
4097         u32 di_cur;
4098         u32 di_total;
4099         u32 di_len;
4100         int key_type = BTRFS_DIR_INDEX_KEY;
4101         char tmp_name[32];
4102         char *name_ptr;
4103         int name_len;
4104         int is_curr = 0;        /* filp->f_pos points to the current index? */
4105
4106         /* FIXME, use a real flag for deciding about the key type */
4107         if (root->fs_info->tree_root == root)
4108                 key_type = BTRFS_DIR_ITEM_KEY;
4109
4110         /* special case for "." */
4111         if (filp->f_pos == 0) {
4112                 over = filldir(dirent, ".", 1, 1, btrfs_ino(inode), DT_DIR);
4113                 if (over)
4114                         return 0;
4115                 filp->f_pos = 1;
4116         }
4117         /* special case for .., just use the back ref */
4118         if (filp->f_pos == 1) {
4119                 u64 pino = parent_ino(filp->f_path.dentry);
4120                 over = filldir(dirent, "..", 2,
4121                                2, pino, DT_DIR);
4122                 if (over)
4123                         return 0;
4124                 filp->f_pos = 2;
4125         }
4126         path = btrfs_alloc_path();
4127         if (!path)
4128                 return -ENOMEM;
4129         path->reada = 2;
4130
4131         if (key_type == BTRFS_DIR_INDEX_KEY) {
4132                 INIT_LIST_HEAD(&ins_list);
4133                 INIT_LIST_HEAD(&del_list);
4134                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4135         }
4136
4137         btrfs_set_key_type(&key, key_type);
4138         key.offset = filp->f_pos;
4139         key.objectid = btrfs_ino(inode);
4140
4141         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4142         if (ret < 0)
4143                 goto err;
4144
4145         while (1) {
4146                 leaf = path->nodes[0];
4147                 slot = path->slots[0];
4148                 if (slot >= btrfs_header_nritems(leaf)) {
4149                         ret = btrfs_next_leaf(root, path);
4150                         if (ret < 0)
4151                                 goto err;
4152                         else if (ret > 0)
4153                                 break;
4154                         continue;
4155                 }
4156
4157                 item = btrfs_item_nr(leaf, slot);
4158                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4159
4160                 if (found_key.objectid != key.objectid)
4161                         break;
4162                 if (btrfs_key_type(&found_key) != key_type)
4163                         break;
4164                 if (found_key.offset < filp->f_pos)
4165                         goto next;
4166                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4167                     btrfs_should_delete_dir_index(&del_list,
4168                                                   found_key.offset))
4169                         goto next;
4170
4171                 filp->f_pos = found_key.offset;
4172                 is_curr = 1;
4173
4174                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4175                 di_cur = 0;
4176                 di_total = btrfs_item_size(leaf, item);
4177
4178                 while (di_cur < di_total) {
4179                         struct btrfs_key location;
4180
4181                         if (verify_dir_item(root, leaf, di))
4182                                 break;
4183
4184                         name_len = btrfs_dir_name_len(leaf, di);
4185                         if (name_len <= sizeof(tmp_name)) {
4186                                 name_ptr = tmp_name;
4187                         } else {
4188                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4189                                 if (!name_ptr) {
4190                                         ret = -ENOMEM;
4191                                         goto err;
4192                                 }
4193                         }
4194                         read_extent_buffer(leaf, name_ptr,
4195                                            (unsigned long)(di + 1), name_len);
4196
4197                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4198                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4199
4200                         /* is this a reference to our own snapshot? If so
4201                          * skip it
4202                          */
4203                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4204                             location.objectid == root->root_key.objectid) {
4205                                 over = 0;
4206                                 goto skip;
4207                         }
4208                         over = filldir(dirent, name_ptr, name_len,
4209                                        found_key.offset, location.objectid,
4210                                        d_type);
4211
4212 skip:
4213                         if (name_ptr != tmp_name)
4214                                 kfree(name_ptr);
4215
4216                         if (over)
4217                                 goto nopos;
4218                         di_len = btrfs_dir_name_len(leaf, di) +
4219                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4220                         di_cur += di_len;
4221                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4222                 }
4223 next:
4224                 path->slots[0]++;
4225         }
4226
4227         if (key_type == BTRFS_DIR_INDEX_KEY) {
4228                 if (is_curr)
4229                         filp->f_pos++;
4230                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4231                                                       &ins_list);
4232                 if (ret)
4233                         goto nopos;
4234         }
4235
4236         /* Reached end of directory/root. Bump pos past the last item. */
4237         if (key_type == BTRFS_DIR_INDEX_KEY)
4238                 /*
4239                  * 32-bit glibc will use getdents64, but then strtol -
4240                  * so the last number we can serve is this.
4241                  */
4242                 filp->f_pos = 0x7fffffff;
4243         else
4244                 filp->f_pos++;
4245 nopos:
4246         ret = 0;
4247 err:
4248         if (key_type == BTRFS_DIR_INDEX_KEY)
4249                 btrfs_put_delayed_items(&ins_list, &del_list);
4250         btrfs_free_path(path);
4251         return ret;
4252 }
4253
4254 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4255 {
4256         struct btrfs_root *root = BTRFS_I(inode)->root;
4257         struct btrfs_trans_handle *trans;
4258         int ret = 0;
4259         bool nolock = false;
4260
4261         if (BTRFS_I(inode)->dummy_inode)
4262                 return 0;
4263
4264         smp_mb();
4265         if (root->fs_info->closing && is_free_space_inode(root, inode))
4266                 nolock = true;
4267
4268         if (wbc->sync_mode == WB_SYNC_ALL) {
4269                 if (nolock)
4270                         trans = btrfs_join_transaction_nolock(root, 1);
4271                 else
4272                         trans = btrfs_join_transaction(root, 1);
4273                 if (IS_ERR(trans))
4274                         return PTR_ERR(trans);
4275                 btrfs_set_trans_block_group(trans, inode);
4276                 if (nolock)
4277                         ret = btrfs_end_transaction_nolock(trans, root);
4278                 else
4279                         ret = btrfs_commit_transaction(trans, root);
4280         }
4281         return ret;
4282 }
4283
4284 /*
4285  * This is somewhat expensive, updating the tree every time the
4286  * inode changes.  But, it is most likely to find the inode in cache.
4287  * FIXME, needs more benchmarking...there are no reasons other than performance
4288  * to keep or drop this code.
4289  */
4290 void btrfs_dirty_inode(struct inode *inode)
4291 {
4292         struct btrfs_root *root = BTRFS_I(inode)->root;
4293         struct btrfs_trans_handle *trans;
4294         int ret;
4295
4296         if (BTRFS_I(inode)->dummy_inode)
4297                 return;
4298
4299         trans = btrfs_join_transaction(root, 1);
4300         BUG_ON(IS_ERR(trans));
4301         btrfs_set_trans_block_group(trans, inode);
4302
4303         ret = btrfs_update_inode(trans, root, inode);
4304         if (ret && ret == -ENOSPC) {
4305                 /* whoops, lets try again with the full transaction */
4306                 btrfs_end_transaction(trans, root);
4307                 trans = btrfs_start_transaction(root, 1);
4308                 if (IS_ERR(trans)) {
4309                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4310                                        "dirty  inode %llu error %ld\n",
4311                                        (unsigned long long)btrfs_ino(inode),
4312                                        PTR_ERR(trans));
4313                         return;
4314                 }
4315                 btrfs_set_trans_block_group(trans, inode);
4316
4317                 ret = btrfs_update_inode(trans, root, inode);
4318                 if (ret) {
4319                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4320                                        "dirty  inode %llu error %d\n",
4321                                        (unsigned long long)btrfs_ino(inode),
4322                                        ret);
4323                 }
4324         }
4325         btrfs_end_transaction(trans, root);
4326         if (BTRFS_I(inode)->delayed_node)
4327                 btrfs_balance_delayed_items(root);
4328 }
4329
4330 /*
4331  * find the highest existing sequence number in a directory
4332  * and then set the in-memory index_cnt variable to reflect
4333  * free sequence numbers
4334  */
4335 static int btrfs_set_inode_index_count(struct inode *inode)
4336 {
4337         struct btrfs_root *root = BTRFS_I(inode)->root;
4338         struct btrfs_key key, found_key;
4339         struct btrfs_path *path;
4340         struct extent_buffer *leaf;
4341         int ret;
4342
4343         key.objectid = btrfs_ino(inode);
4344         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4345         key.offset = (u64)-1;
4346
4347         path = btrfs_alloc_path();
4348         if (!path)
4349                 return -ENOMEM;
4350
4351         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4352         if (ret < 0)
4353                 goto out;
4354         /* FIXME: we should be able to handle this */
4355         if (ret == 0)
4356                 goto out;
4357         ret = 0;
4358
4359         /*
4360          * MAGIC NUMBER EXPLANATION:
4361          * since we search a directory based on f_pos we have to start at 2
4362          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4363          * else has to start at 2
4364          */
4365         if (path->slots[0] == 0) {
4366                 BTRFS_I(inode)->index_cnt = 2;
4367                 goto out;
4368         }
4369
4370         path->slots[0]--;
4371
4372         leaf = path->nodes[0];
4373         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4374
4375         if (found_key.objectid != btrfs_ino(inode) ||
4376             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4377                 BTRFS_I(inode)->index_cnt = 2;
4378                 goto out;
4379         }
4380
4381         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4382 out:
4383         btrfs_free_path(path);
4384         return ret;
4385 }
4386
4387 /*
4388  * helper to find a free sequence number in a given directory.  This current
4389  * code is very simple, later versions will do smarter things in the btree
4390  */
4391 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4392 {
4393         int ret = 0;
4394
4395         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4396                 ret = btrfs_inode_delayed_dir_index_count(dir);
4397                 if (ret) {
4398                         ret = btrfs_set_inode_index_count(dir);
4399                         if (ret)
4400                                 return ret;
4401                 }
4402         }
4403
4404         *index = BTRFS_I(dir)->index_cnt;
4405         BTRFS_I(dir)->index_cnt++;
4406
4407         return ret;
4408 }
4409
4410 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4411                                      struct btrfs_root *root,
4412                                      struct inode *dir,
4413                                      const char *name, int name_len,
4414                                      u64 ref_objectid, u64 objectid,
4415                                      u64 alloc_hint, int mode, u64 *index)
4416 {
4417         struct inode *inode;
4418         struct btrfs_inode_item *inode_item;
4419         struct btrfs_key *location;
4420         struct btrfs_path *path;
4421         struct btrfs_inode_ref *ref;
4422         struct btrfs_key key[2];
4423         u32 sizes[2];
4424         unsigned long ptr;
4425         int ret;
4426         int owner;
4427
4428         path = btrfs_alloc_path();
4429         BUG_ON(!path);
4430
4431         inode = new_inode(root->fs_info->sb);
4432         if (!inode) {
4433                 btrfs_free_path(path);
4434                 return ERR_PTR(-ENOMEM);
4435         }
4436
4437         /*
4438          * we have to initialize this early, so we can reclaim the inode
4439          * number if we fail afterwards in this function.
4440          */
4441         inode->i_ino = objectid;
4442
4443         if (dir) {
4444                 trace_btrfs_inode_request(dir);
4445
4446                 ret = btrfs_set_inode_index(dir, index);
4447                 if (ret) {
4448                         btrfs_free_path(path);
4449                         iput(inode);
4450                         return ERR_PTR(ret);
4451                 }
4452         }
4453         /*
4454          * index_cnt is ignored for everything but a dir,
4455          * btrfs_get_inode_index_count has an explanation for the magic
4456          * number
4457          */
4458         BTRFS_I(inode)->index_cnt = 2;
4459         BTRFS_I(inode)->root = root;
4460         BTRFS_I(inode)->generation = trans->transid;
4461         inode->i_generation = BTRFS_I(inode)->generation;
4462         btrfs_set_inode_space_info(root, inode);
4463
4464         if (mode & S_IFDIR)
4465                 owner = 0;
4466         else
4467                 owner = 1;
4468         BTRFS_I(inode)->block_group =
4469                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4470
4471         key[0].objectid = objectid;
4472         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4473         key[0].offset = 0;
4474
4475         key[1].objectid = objectid;
4476         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4477         key[1].offset = ref_objectid;
4478
4479         sizes[0] = sizeof(struct btrfs_inode_item);
4480         sizes[1] = name_len + sizeof(*ref);
4481
4482         path->leave_spinning = 1;
4483         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4484         if (ret != 0)
4485                 goto fail;
4486
4487         inode_init_owner(inode, dir, mode);
4488         inode_set_bytes(inode, 0);
4489         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4490         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4491                                   struct btrfs_inode_item);
4492         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4493
4494         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4495                              struct btrfs_inode_ref);
4496         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4497         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4498         ptr = (unsigned long)(ref + 1);
4499         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4500
4501         btrfs_mark_buffer_dirty(path->nodes[0]);
4502         btrfs_free_path(path);
4503
4504         location = &BTRFS_I(inode)->location;
4505         location->objectid = objectid;
4506         location->offset = 0;
4507         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4508
4509         btrfs_inherit_iflags(inode, dir);
4510
4511         if ((mode & S_IFREG)) {
4512                 if (btrfs_test_opt(root, NODATASUM))
4513                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4514                 if (btrfs_test_opt(root, NODATACOW) ||
4515                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4516                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4517         }
4518
4519         insert_inode_hash(inode);
4520         inode_tree_add(inode);
4521
4522         trace_btrfs_inode_new(inode);
4523
4524         return inode;
4525 fail:
4526         if (dir)
4527                 BTRFS_I(dir)->index_cnt--;
4528         btrfs_free_path(path);
4529         iput(inode);
4530         return ERR_PTR(ret);
4531 }
4532
4533 static inline u8 btrfs_inode_type(struct inode *inode)
4534 {
4535         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4536 }
4537
4538 /*
4539  * utility function to add 'inode' into 'parent_inode' with
4540  * a give name and a given sequence number.
4541  * if 'add_backref' is true, also insert a backref from the
4542  * inode to the parent directory.
4543  */
4544 int btrfs_add_link(struct btrfs_trans_handle *trans,
4545                    struct inode *parent_inode, struct inode *inode,
4546                    const char *name, int name_len, int add_backref, u64 index)
4547 {
4548         int ret = 0;
4549         struct btrfs_key key;
4550         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4551         u64 ino = btrfs_ino(inode);
4552         u64 parent_ino = btrfs_ino(parent_inode);
4553
4554         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4555                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4556         } else {
4557                 key.objectid = ino;
4558                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4559                 key.offset = 0;
4560         }
4561
4562         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4563                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4564                                          key.objectid, root->root_key.objectid,
4565                                          parent_ino, index, name, name_len);
4566         } else if (add_backref) {
4567                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4568                                              parent_ino, index);
4569         }
4570
4571         if (ret == 0) {
4572                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4573                                             parent_inode, &key,
4574                                             btrfs_inode_type(inode), index);
4575                 BUG_ON(ret);
4576
4577                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4578                                    name_len * 2);
4579                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4580                 ret = btrfs_update_inode(trans, root, parent_inode);
4581         }
4582         return ret;
4583 }
4584
4585 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4586                             struct inode *dir, struct dentry *dentry,
4587                             struct inode *inode, int backref, u64 index)
4588 {
4589         int err = btrfs_add_link(trans, dir, inode,
4590                                  dentry->d_name.name, dentry->d_name.len,
4591                                  backref, index);
4592         if (!err) {
4593                 d_instantiate(dentry, inode);
4594                 return 0;
4595         }
4596         if (err > 0)
4597                 err = -EEXIST;
4598         return err;
4599 }
4600
4601 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4602                         int mode, dev_t rdev)
4603 {
4604         struct btrfs_trans_handle *trans;
4605         struct btrfs_root *root = BTRFS_I(dir)->root;
4606         struct inode *inode = NULL;
4607         int err;
4608         int drop_inode = 0;
4609         u64 objectid;
4610         unsigned long nr = 0;
4611         u64 index = 0;
4612
4613         if (!new_valid_dev(rdev))
4614                 return -EINVAL;
4615
4616         /*
4617          * 2 for inode item and ref
4618          * 2 for dir items
4619          * 1 for xattr if selinux is on
4620          */
4621         trans = btrfs_start_transaction(root, 5);
4622         if (IS_ERR(trans))
4623                 return PTR_ERR(trans);
4624
4625         btrfs_set_trans_block_group(trans, dir);
4626
4627         err = btrfs_find_free_ino(root, &objectid);
4628         if (err)
4629                 goto out_unlock;
4630
4631         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4632                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4633                                 BTRFS_I(dir)->block_group, mode, &index);
4634         if (IS_ERR(inode)) {
4635                 err = PTR_ERR(inode);
4636                 goto out_unlock;
4637         }
4638
4639         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4640         if (err) {
4641                 drop_inode = 1;
4642                 goto out_unlock;
4643         }
4644
4645         btrfs_set_trans_block_group(trans, inode);
4646         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4647         if (err)
4648                 drop_inode = 1;
4649         else {
4650                 inode->i_op = &btrfs_special_inode_operations;
4651                 init_special_inode(inode, inode->i_mode, rdev);
4652                 btrfs_update_inode(trans, root, inode);
4653         }
4654         btrfs_update_inode_block_group(trans, inode);
4655         btrfs_update_inode_block_group(trans, dir);
4656 out_unlock:
4657         nr = trans->blocks_used;
4658         btrfs_end_transaction_throttle(trans, root);
4659         btrfs_btree_balance_dirty(root, nr);
4660         if (drop_inode) {
4661                 inode_dec_link_count(inode);
4662                 iput(inode);
4663         }
4664         return err;
4665 }
4666
4667 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4668                         int mode, struct nameidata *nd)
4669 {
4670         struct btrfs_trans_handle *trans;
4671         struct btrfs_root *root = BTRFS_I(dir)->root;
4672         struct inode *inode = NULL;
4673         int drop_inode = 0;
4674         int err;
4675         unsigned long nr = 0;
4676         u64 objectid;
4677         u64 index = 0;
4678
4679         /*
4680          * 2 for inode item and ref
4681          * 2 for dir items
4682          * 1 for xattr if selinux is on
4683          */
4684         trans = btrfs_start_transaction(root, 5);
4685         if (IS_ERR(trans))
4686                 return PTR_ERR(trans);
4687
4688         btrfs_set_trans_block_group(trans, dir);
4689
4690         err = btrfs_find_free_ino(root, &objectid);
4691         if (err)
4692                 goto out_unlock;
4693
4694         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4695                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4696                                 BTRFS_I(dir)->block_group, mode, &index);
4697         if (IS_ERR(inode)) {
4698                 err = PTR_ERR(inode);
4699                 goto out_unlock;
4700         }
4701
4702         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4703         if (err) {
4704                 drop_inode = 1;
4705                 goto out_unlock;
4706         }
4707
4708         btrfs_set_trans_block_group(trans, inode);
4709         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4710         if (err)
4711                 drop_inode = 1;
4712         else {
4713                 inode->i_mapping->a_ops = &btrfs_aops;
4714                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4715                 inode->i_fop = &btrfs_file_operations;
4716                 inode->i_op = &btrfs_file_inode_operations;
4717                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4718         }
4719         btrfs_update_inode_block_group(trans, inode);
4720         btrfs_update_inode_block_group(trans, dir);
4721 out_unlock:
4722         nr = trans->blocks_used;
4723         btrfs_end_transaction_throttle(trans, root);
4724         if (drop_inode) {
4725                 inode_dec_link_count(inode);
4726                 iput(inode);
4727         }
4728         btrfs_btree_balance_dirty(root, nr);
4729         return err;
4730 }
4731
4732 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4733                       struct dentry *dentry)
4734 {
4735         struct btrfs_trans_handle *trans;
4736         struct btrfs_root *root = BTRFS_I(dir)->root;
4737         struct inode *inode = old_dentry->d_inode;
4738         u64 index;
4739         unsigned long nr = 0;
4740         int err;
4741         int drop_inode = 0;
4742
4743         /* do not allow sys_link's with other subvols of the same device */
4744         if (root->objectid != BTRFS_I(inode)->root->objectid)
4745                 return -EXDEV;
4746
4747         if (inode->i_nlink == ~0U)
4748                 return -EMLINK;
4749
4750         err = btrfs_set_inode_index(dir, &index);
4751         if (err)
4752                 goto fail;
4753
4754         /*
4755          * 2 items for inode and inode ref
4756          * 2 items for dir items
4757          * 1 item for parent inode
4758          */
4759         trans = btrfs_start_transaction(root, 5);
4760         if (IS_ERR(trans)) {
4761                 err = PTR_ERR(trans);
4762                 goto fail;
4763         }
4764
4765         btrfs_inc_nlink(inode);
4766         inode->i_ctime = CURRENT_TIME;
4767
4768         btrfs_set_trans_block_group(trans, dir);
4769         ihold(inode);
4770
4771         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4772
4773         if (err) {
4774                 drop_inode = 1;
4775         } else {
4776                 struct dentry *parent = dget_parent(dentry);
4777                 btrfs_update_inode_block_group(trans, dir);
4778                 err = btrfs_update_inode(trans, root, inode);
4779                 BUG_ON(err);
4780                 btrfs_log_new_name(trans, inode, NULL, parent);
4781                 dput(parent);
4782         }
4783
4784         nr = trans->blocks_used;
4785         btrfs_end_transaction_throttle(trans, root);
4786 fail:
4787         if (drop_inode) {
4788                 inode_dec_link_count(inode);
4789                 iput(inode);
4790         }
4791         btrfs_btree_balance_dirty(root, nr);
4792         return err;
4793 }
4794
4795 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4796 {
4797         struct inode *inode = NULL;
4798         struct btrfs_trans_handle *trans;
4799         struct btrfs_root *root = BTRFS_I(dir)->root;
4800         int err = 0;
4801         int drop_on_err = 0;
4802         u64 objectid = 0;
4803         u64 index = 0;
4804         unsigned long nr = 1;
4805
4806         /*
4807          * 2 items for inode and ref
4808          * 2 items for dir items
4809          * 1 for xattr if selinux is on
4810          */
4811         trans = btrfs_start_transaction(root, 5);
4812         if (IS_ERR(trans))
4813                 return PTR_ERR(trans);
4814         btrfs_set_trans_block_group(trans, dir);
4815
4816         err = btrfs_find_free_ino(root, &objectid);
4817         if (err)
4818                 goto out_fail;
4819
4820         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4821                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4822                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4823                                 &index);
4824         if (IS_ERR(inode)) {
4825                 err = PTR_ERR(inode);
4826                 goto out_fail;
4827         }
4828
4829         drop_on_err = 1;
4830
4831         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4832         if (err)
4833                 goto out_fail;
4834
4835         inode->i_op = &btrfs_dir_inode_operations;
4836         inode->i_fop = &btrfs_dir_file_operations;
4837         btrfs_set_trans_block_group(trans, inode);
4838
4839         btrfs_i_size_write(inode, 0);
4840         err = btrfs_update_inode(trans, root, inode);
4841         if (err)
4842                 goto out_fail;
4843
4844         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4845                              dentry->d_name.len, 0, index);
4846         if (err)
4847                 goto out_fail;
4848
4849         d_instantiate(dentry, inode);
4850         drop_on_err = 0;
4851         btrfs_update_inode_block_group(trans, inode);
4852         btrfs_update_inode_block_group(trans, dir);
4853
4854 out_fail:
4855         nr = trans->blocks_used;
4856         btrfs_end_transaction_throttle(trans, root);
4857         if (drop_on_err)
4858                 iput(inode);
4859         btrfs_btree_balance_dirty(root, nr);
4860         return err;
4861 }
4862
4863 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4864  * and an extent that you want to insert, deal with overlap and insert
4865  * the new extent into the tree.
4866  */
4867 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4868                                 struct extent_map *existing,
4869                                 struct extent_map *em,
4870                                 u64 map_start, u64 map_len)
4871 {
4872         u64 start_diff;
4873
4874         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4875         start_diff = map_start - em->start;
4876         em->start = map_start;
4877         em->len = map_len;
4878         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4879             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4880                 em->block_start += start_diff;
4881                 em->block_len -= start_diff;
4882         }
4883         return add_extent_mapping(em_tree, em);
4884 }
4885
4886 static noinline int uncompress_inline(struct btrfs_path *path,
4887                                       struct inode *inode, struct page *page,
4888                                       size_t pg_offset, u64 extent_offset,
4889                                       struct btrfs_file_extent_item *item)
4890 {
4891         int ret;
4892         struct extent_buffer *leaf = path->nodes[0];
4893         char *tmp;
4894         size_t max_size;
4895         unsigned long inline_size;
4896         unsigned long ptr;
4897         int compress_type;
4898
4899         WARN_ON(pg_offset != 0);
4900         compress_type = btrfs_file_extent_compression(leaf, item);
4901         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4902         inline_size = btrfs_file_extent_inline_item_len(leaf,
4903                                         btrfs_item_nr(leaf, path->slots[0]));
4904         tmp = kmalloc(inline_size, GFP_NOFS);
4905         if (!tmp)
4906                 return -ENOMEM;
4907         ptr = btrfs_file_extent_inline_start(item);
4908
4909         read_extent_buffer(leaf, tmp, ptr, inline_size);
4910
4911         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4912         ret = btrfs_decompress(compress_type, tmp, page,
4913                                extent_offset, inline_size, max_size);
4914         if (ret) {
4915                 char *kaddr = kmap_atomic(page, KM_USER0);
4916                 unsigned long copy_size = min_t(u64,
4917                                   PAGE_CACHE_SIZE - pg_offset,
4918                                   max_size - extent_offset);
4919                 memset(kaddr + pg_offset, 0, copy_size);
4920                 kunmap_atomic(kaddr, KM_USER0);
4921         }
4922         kfree(tmp);
4923         return 0;
4924 }
4925
4926 /*
4927  * a bit scary, this does extent mapping from logical file offset to the disk.
4928  * the ugly parts come from merging extents from the disk with the in-ram
4929  * representation.  This gets more complex because of the data=ordered code,
4930  * where the in-ram extents might be locked pending data=ordered completion.
4931  *
4932  * This also copies inline extents directly into the page.
4933  */
4934
4935 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4936                                     size_t pg_offset, u64 start, u64 len,
4937                                     int create)
4938 {
4939         int ret;
4940         int err = 0;
4941         u64 bytenr;
4942         u64 extent_start = 0;
4943         u64 extent_end = 0;
4944         u64 objectid = btrfs_ino(inode);
4945         u32 found_type;
4946         struct btrfs_path *path = NULL;
4947         struct btrfs_root *root = BTRFS_I(inode)->root;
4948         struct btrfs_file_extent_item *item;
4949         struct extent_buffer *leaf;
4950         struct btrfs_key found_key;
4951         struct extent_map *em = NULL;
4952         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4953         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4954         struct btrfs_trans_handle *trans = NULL;
4955         int compress_type;
4956
4957 again:
4958         read_lock(&em_tree->lock);
4959         em = lookup_extent_mapping(em_tree, start, len);
4960         if (em)
4961                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4962         read_unlock(&em_tree->lock);
4963
4964         if (em) {
4965                 if (em->start > start || em->start + em->len <= start)
4966                         free_extent_map(em);
4967                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4968                         free_extent_map(em);
4969                 else
4970                         goto out;
4971         }
4972         em = alloc_extent_map();
4973         if (!em) {
4974                 err = -ENOMEM;
4975                 goto out;
4976         }
4977         em->bdev = root->fs_info->fs_devices->latest_bdev;
4978         em->start = EXTENT_MAP_HOLE;
4979         em->orig_start = EXTENT_MAP_HOLE;
4980         em->len = (u64)-1;
4981         em->block_len = (u64)-1;
4982
4983         if (!path) {
4984                 path = btrfs_alloc_path();
4985                 BUG_ON(!path);
4986         }
4987
4988         ret = btrfs_lookup_file_extent(trans, root, path,
4989                                        objectid, start, trans != NULL);
4990         if (ret < 0) {
4991                 err = ret;
4992                 goto out;
4993         }
4994
4995         if (ret != 0) {
4996                 if (path->slots[0] == 0)
4997                         goto not_found;
4998                 path->slots[0]--;
4999         }
5000
5001         leaf = path->nodes[0];
5002         item = btrfs_item_ptr(leaf, path->slots[0],
5003                               struct btrfs_file_extent_item);
5004         /* are we inside the extent that was found? */
5005         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5006         found_type = btrfs_key_type(&found_key);
5007         if (found_key.objectid != objectid ||
5008             found_type != BTRFS_EXTENT_DATA_KEY) {
5009                 goto not_found;
5010         }
5011
5012         found_type = btrfs_file_extent_type(leaf, item);
5013         extent_start = found_key.offset;
5014         compress_type = btrfs_file_extent_compression(leaf, item);
5015         if (found_type == BTRFS_FILE_EXTENT_REG ||
5016             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5017                 extent_end = extent_start +
5018                        btrfs_file_extent_num_bytes(leaf, item);
5019         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5020                 size_t size;
5021                 size = btrfs_file_extent_inline_len(leaf, item);
5022                 extent_end = (extent_start + size + root->sectorsize - 1) &
5023                         ~((u64)root->sectorsize - 1);
5024         }
5025
5026         if (start >= extent_end) {
5027                 path->slots[0]++;
5028                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5029                         ret = btrfs_next_leaf(root, path);
5030                         if (ret < 0) {
5031                                 err = ret;
5032                                 goto out;
5033                         }
5034                         if (ret > 0)
5035                                 goto not_found;
5036                         leaf = path->nodes[0];
5037                 }
5038                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5039                 if (found_key.objectid != objectid ||
5040                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5041                         goto not_found;
5042                 if (start + len <= found_key.offset)
5043                         goto not_found;
5044                 em->start = start;
5045                 em->len = found_key.offset - start;
5046                 goto not_found_em;
5047         }
5048
5049         if (found_type == BTRFS_FILE_EXTENT_REG ||
5050             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5051                 em->start = extent_start;
5052                 em->len = extent_end - extent_start;
5053                 em->orig_start = extent_start -
5054                                  btrfs_file_extent_offset(leaf, item);
5055                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5056                 if (bytenr == 0) {
5057                         em->block_start = EXTENT_MAP_HOLE;
5058                         goto insert;
5059                 }
5060                 if (compress_type != BTRFS_COMPRESS_NONE) {
5061                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5062                         em->compress_type = compress_type;
5063                         em->block_start = bytenr;
5064                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5065                                                                          item);
5066                 } else {
5067                         bytenr += btrfs_file_extent_offset(leaf, item);
5068                         em->block_start = bytenr;
5069                         em->block_len = em->len;
5070                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5071                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5072                 }
5073                 goto insert;
5074         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5075                 unsigned long ptr;
5076                 char *map;
5077                 size_t size;
5078                 size_t extent_offset;
5079                 size_t copy_size;
5080
5081                 em->block_start = EXTENT_MAP_INLINE;
5082                 if (!page || create) {
5083                         em->start = extent_start;
5084                         em->len = extent_end - extent_start;
5085                         goto out;
5086                 }
5087
5088                 size = btrfs_file_extent_inline_len(leaf, item);
5089                 extent_offset = page_offset(page) + pg_offset - extent_start;
5090                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5091                                 size - extent_offset);
5092                 em->start = extent_start + extent_offset;
5093                 em->len = (copy_size + root->sectorsize - 1) &
5094                         ~((u64)root->sectorsize - 1);
5095                 em->orig_start = EXTENT_MAP_INLINE;
5096                 if (compress_type) {
5097                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5098                         em->compress_type = compress_type;
5099                 }
5100                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5101                 if (create == 0 && !PageUptodate(page)) {
5102                         if (btrfs_file_extent_compression(leaf, item) !=
5103                             BTRFS_COMPRESS_NONE) {
5104                                 ret = uncompress_inline(path, inode, page,
5105                                                         pg_offset,
5106                                                         extent_offset, item);
5107                                 BUG_ON(ret);
5108                         } else {
5109                                 map = kmap(page);
5110                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5111                                                    copy_size);
5112                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5113                                         memset(map + pg_offset + copy_size, 0,
5114                                                PAGE_CACHE_SIZE - pg_offset -
5115                                                copy_size);
5116                                 }
5117                                 kunmap(page);
5118                         }
5119                         flush_dcache_page(page);
5120                 } else if (create && PageUptodate(page)) {
5121                         WARN_ON(1);
5122                         if (!trans) {
5123                                 kunmap(page);
5124                                 free_extent_map(em);
5125                                 em = NULL;
5126                                 btrfs_release_path(path);
5127                                 trans = btrfs_join_transaction(root, 1);
5128                                 if (IS_ERR(trans))
5129                                         return ERR_CAST(trans);
5130                                 goto again;
5131                         }
5132                         map = kmap(page);
5133                         write_extent_buffer(leaf, map + pg_offset, ptr,
5134                                             copy_size);
5135                         kunmap(page);
5136                         btrfs_mark_buffer_dirty(leaf);
5137                 }
5138                 set_extent_uptodate(io_tree, em->start,
5139                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5140                 goto insert;
5141         } else {
5142                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5143                 WARN_ON(1);
5144         }
5145 not_found:
5146         em->start = start;
5147         em->len = len;
5148 not_found_em:
5149         em->block_start = EXTENT_MAP_HOLE;
5150         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5151 insert:
5152         btrfs_release_path(path);
5153         if (em->start > start || extent_map_end(em) <= start) {
5154                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5155                        "[%llu %llu]\n", (unsigned long long)em->start,
5156                        (unsigned long long)em->len,
5157                        (unsigned long long)start,
5158                        (unsigned long long)len);
5159                 err = -EIO;
5160                 goto out;
5161         }
5162
5163         err = 0;
5164         write_lock(&em_tree->lock);
5165         ret = add_extent_mapping(em_tree, em);
5166         /* it is possible that someone inserted the extent into the tree
5167          * while we had the lock dropped.  It is also possible that
5168          * an overlapping map exists in the tree
5169          */
5170         if (ret == -EEXIST) {
5171                 struct extent_map *existing;
5172
5173                 ret = 0;
5174
5175                 existing = lookup_extent_mapping(em_tree, start, len);
5176                 if (existing && (existing->start > start ||
5177                     existing->start + existing->len <= start)) {
5178                         free_extent_map(existing);
5179                         existing = NULL;
5180                 }
5181                 if (!existing) {
5182                         existing = lookup_extent_mapping(em_tree, em->start,
5183                                                          em->len);
5184                         if (existing) {
5185                                 err = merge_extent_mapping(em_tree, existing,
5186                                                            em, start,
5187                                                            root->sectorsize);
5188                                 free_extent_map(existing);
5189                                 if (err) {
5190                                         free_extent_map(em);
5191                                         em = NULL;
5192                                 }
5193                         } else {
5194                                 err = -EIO;
5195                                 free_extent_map(em);
5196                                 em = NULL;
5197                         }
5198                 } else {
5199                         free_extent_map(em);
5200                         em = existing;
5201                         err = 0;
5202                 }
5203         }
5204         write_unlock(&em_tree->lock);
5205 out:
5206
5207         trace_btrfs_get_extent(root, em);
5208
5209         if (path)
5210                 btrfs_free_path(path);
5211         if (trans) {
5212                 ret = btrfs_end_transaction(trans, root);
5213                 if (!err)
5214                         err = ret;
5215         }
5216         if (err) {
5217                 free_extent_map(em);
5218                 return ERR_PTR(err);
5219         }
5220         return em;
5221 }
5222
5223 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5224                                            size_t pg_offset, u64 start, u64 len,
5225                                            int create)
5226 {
5227         struct extent_map *em;
5228         struct extent_map *hole_em = NULL;
5229         u64 range_start = start;
5230         u64 end;
5231         u64 found;
5232         u64 found_end;
5233         int err = 0;
5234
5235         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5236         if (IS_ERR(em))
5237                 return em;
5238         if (em) {
5239                 /*
5240                  * if our em maps to a hole, there might
5241                  * actually be delalloc bytes behind it
5242                  */
5243                 if (em->block_start != EXTENT_MAP_HOLE)
5244                         return em;
5245                 else
5246                         hole_em = em;
5247         }
5248
5249         /* check to see if we've wrapped (len == -1 or similar) */
5250         end = start + len;
5251         if (end < start)
5252                 end = (u64)-1;
5253         else
5254                 end -= 1;
5255
5256         em = NULL;
5257
5258         /* ok, we didn't find anything, lets look for delalloc */
5259         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5260                                  end, len, EXTENT_DELALLOC, 1);
5261         found_end = range_start + found;
5262         if (found_end < range_start)
5263                 found_end = (u64)-1;
5264
5265         /*
5266          * we didn't find anything useful, return
5267          * the original results from get_extent()
5268          */
5269         if (range_start > end || found_end <= start) {
5270                 em = hole_em;
5271                 hole_em = NULL;
5272                 goto out;
5273         }
5274
5275         /* adjust the range_start to make sure it doesn't
5276          * go backwards from the start they passed in
5277          */
5278         range_start = max(start,range_start);
5279         found = found_end - range_start;
5280
5281         if (found > 0) {
5282                 u64 hole_start = start;
5283                 u64 hole_len = len;
5284
5285                 em = alloc_extent_map();
5286                 if (!em) {
5287                         err = -ENOMEM;
5288                         goto out;
5289                 }
5290                 /*
5291                  * when btrfs_get_extent can't find anything it
5292                  * returns one huge hole
5293                  *
5294                  * make sure what it found really fits our range, and
5295                  * adjust to make sure it is based on the start from
5296                  * the caller
5297                  */
5298                 if (hole_em) {
5299                         u64 calc_end = extent_map_end(hole_em);
5300
5301                         if (calc_end <= start || (hole_em->start > end)) {
5302                                 free_extent_map(hole_em);
5303                                 hole_em = NULL;
5304                         } else {
5305                                 hole_start = max(hole_em->start, start);
5306                                 hole_len = calc_end - hole_start;
5307                         }
5308                 }
5309                 em->bdev = NULL;
5310                 if (hole_em && range_start > hole_start) {
5311                         /* our hole starts before our delalloc, so we
5312                          * have to return just the parts of the hole
5313                          * that go until  the delalloc starts
5314                          */
5315                         em->len = min(hole_len,
5316                                       range_start - hole_start);
5317                         em->start = hole_start;
5318                         em->orig_start = hole_start;
5319                         /*
5320                          * don't adjust block start at all,
5321                          * it is fixed at EXTENT_MAP_HOLE
5322                          */
5323                         em->block_start = hole_em->block_start;
5324                         em->block_len = hole_len;
5325                 } else {
5326                         em->start = range_start;
5327                         em->len = found;
5328                         em->orig_start = range_start;
5329                         em->block_start = EXTENT_MAP_DELALLOC;
5330                         em->block_len = found;
5331                 }
5332         } else if (hole_em) {
5333                 return hole_em;
5334         }
5335 out:
5336
5337         free_extent_map(hole_em);
5338         if (err) {
5339                 free_extent_map(em);
5340                 return ERR_PTR(err);
5341         }
5342         return em;
5343 }
5344
5345 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5346                                                   struct extent_map *em,
5347                                                   u64 start, u64 len)
5348 {
5349         struct btrfs_root *root = BTRFS_I(inode)->root;
5350         struct btrfs_trans_handle *trans;
5351         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5352         struct btrfs_key ins;
5353         u64 alloc_hint;
5354         int ret;
5355         bool insert = false;
5356
5357         /*
5358          * Ok if the extent map we looked up is a hole and is for the exact
5359          * range we want, there is no reason to allocate a new one, however if
5360          * it is not right then we need to free this one and drop the cache for
5361          * our range.
5362          */
5363         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5364             em->len != len) {
5365                 free_extent_map(em);
5366                 em = NULL;
5367                 insert = true;
5368                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5369         }
5370
5371         trans = btrfs_join_transaction(root, 0);
5372         if (IS_ERR(trans))
5373                 return ERR_CAST(trans);
5374
5375         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5376
5377         alloc_hint = get_extent_allocation_hint(inode, start, len);
5378         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5379                                    alloc_hint, (u64)-1, &ins, 1);
5380         if (ret) {
5381                 em = ERR_PTR(ret);
5382                 goto out;
5383         }
5384
5385         if (!em) {
5386                 em = alloc_extent_map();
5387                 if (!em) {
5388                         em = ERR_PTR(-ENOMEM);
5389                         goto out;
5390                 }
5391         }
5392
5393         em->start = start;
5394         em->orig_start = em->start;
5395         em->len = ins.offset;
5396
5397         em->block_start = ins.objectid;
5398         em->block_len = ins.offset;
5399         em->bdev = root->fs_info->fs_devices->latest_bdev;
5400
5401         /*
5402          * We need to do this because if we're using the original em we searched
5403          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5404          */
5405         em->flags = 0;
5406         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5407
5408         while (insert) {
5409                 write_lock(&em_tree->lock);
5410                 ret = add_extent_mapping(em_tree, em);
5411                 write_unlock(&em_tree->lock);
5412                 if (ret != -EEXIST)
5413                         break;
5414                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5415         }
5416
5417         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5418                                            ins.offset, ins.offset, 0);
5419         if (ret) {
5420                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5421                 em = ERR_PTR(ret);
5422         }
5423 out:
5424         btrfs_end_transaction(trans, root);
5425         return em;
5426 }
5427
5428 /*
5429  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5430  * block must be cow'd
5431  */
5432 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5433                                       struct inode *inode, u64 offset, u64 len)
5434 {
5435         struct btrfs_path *path;
5436         int ret;
5437         struct extent_buffer *leaf;
5438         struct btrfs_root *root = BTRFS_I(inode)->root;
5439         struct btrfs_file_extent_item *fi;
5440         struct btrfs_key key;
5441         u64 disk_bytenr;
5442         u64 backref_offset;
5443         u64 extent_end;
5444         u64 num_bytes;
5445         int slot;
5446         int found_type;
5447
5448         path = btrfs_alloc_path();
5449         if (!path)
5450                 return -ENOMEM;
5451
5452         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5453                                        offset, 0);
5454         if (ret < 0)
5455                 goto out;
5456
5457         slot = path->slots[0];
5458         if (ret == 1) {
5459                 if (slot == 0) {
5460                         /* can't find the item, must cow */
5461                         ret = 0;
5462                         goto out;
5463                 }
5464                 slot--;
5465         }
5466         ret = 0;
5467         leaf = path->nodes[0];
5468         btrfs_item_key_to_cpu(leaf, &key, slot);
5469         if (key.objectid != btrfs_ino(inode) ||
5470             key.type != BTRFS_EXTENT_DATA_KEY) {
5471                 /* not our file or wrong item type, must cow */
5472                 goto out;
5473         }
5474
5475         if (key.offset > offset) {
5476                 /* Wrong offset, must cow */
5477                 goto out;
5478         }
5479
5480         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5481         found_type = btrfs_file_extent_type(leaf, fi);
5482         if (found_type != BTRFS_FILE_EXTENT_REG &&
5483             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5484                 /* not a regular extent, must cow */
5485                 goto out;
5486         }
5487         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5488         backref_offset = btrfs_file_extent_offset(leaf, fi);
5489
5490         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5491         if (extent_end < offset + len) {
5492                 /* extent doesn't include our full range, must cow */
5493                 goto out;
5494         }
5495
5496         if (btrfs_extent_readonly(root, disk_bytenr))
5497                 goto out;
5498
5499         /*
5500          * look for other files referencing this extent, if we
5501          * find any we must cow
5502          */
5503         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5504                                   key.offset - backref_offset, disk_bytenr))
5505                 goto out;
5506
5507         /*
5508          * adjust disk_bytenr and num_bytes to cover just the bytes
5509          * in this extent we are about to write.  If there
5510          * are any csums in that range we have to cow in order
5511          * to keep the csums correct
5512          */
5513         disk_bytenr += backref_offset;
5514         disk_bytenr += offset - key.offset;
5515         num_bytes = min(offset + len, extent_end) - offset;
5516         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5517                                 goto out;
5518         /*
5519          * all of the above have passed, it is safe to overwrite this extent
5520          * without cow
5521          */
5522         ret = 1;
5523 out:
5524         btrfs_free_path(path);
5525         return ret;
5526 }
5527
5528 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5529                                    struct buffer_head *bh_result, int create)
5530 {
5531         struct extent_map *em;
5532         struct btrfs_root *root = BTRFS_I(inode)->root;
5533         u64 start = iblock << inode->i_blkbits;
5534         u64 len = bh_result->b_size;
5535         struct btrfs_trans_handle *trans;
5536
5537         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5538         if (IS_ERR(em))
5539                 return PTR_ERR(em);
5540
5541         /*
5542          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5543          * io.  INLINE is special, and we could probably kludge it in here, but
5544          * it's still buffered so for safety lets just fall back to the generic
5545          * buffered path.
5546          *
5547          * For COMPRESSED we _have_ to read the entire extent in so we can
5548          * decompress it, so there will be buffering required no matter what we
5549          * do, so go ahead and fallback to buffered.
5550          *
5551          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5552          * to buffered IO.  Don't blame me, this is the price we pay for using
5553          * the generic code.
5554          */
5555         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5556             em->block_start == EXTENT_MAP_INLINE) {
5557                 free_extent_map(em);
5558                 return -ENOTBLK;
5559         }
5560
5561         /* Just a good old fashioned hole, return */
5562         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5563                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5564                 free_extent_map(em);
5565                 /* DIO will do one hole at a time, so just unlock a sector */
5566                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5567                               start + root->sectorsize - 1, GFP_NOFS);
5568                 return 0;
5569         }
5570
5571         /*
5572          * We don't allocate a new extent in the following cases
5573          *
5574          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5575          * existing extent.
5576          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5577          * just use the extent.
5578          *
5579          */
5580         if (!create) {
5581                 len = em->len - (start - em->start);
5582                 goto map;
5583         }
5584
5585         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5586             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5587              em->block_start != EXTENT_MAP_HOLE)) {
5588                 int type;
5589                 int ret;
5590                 u64 block_start;
5591
5592                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5593                         type = BTRFS_ORDERED_PREALLOC;
5594                 else
5595                         type = BTRFS_ORDERED_NOCOW;
5596                 len = min(len, em->len - (start - em->start));
5597                 block_start = em->block_start + (start - em->start);
5598
5599                 /*
5600                  * we're not going to log anything, but we do need
5601                  * to make sure the current transaction stays open
5602                  * while we look for nocow cross refs
5603                  */
5604                 trans = btrfs_join_transaction(root, 0);
5605                 if (IS_ERR(trans))
5606                         goto must_cow;
5607
5608                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5609                         ret = btrfs_add_ordered_extent_dio(inode, start,
5610                                            block_start, len, len, type);
5611                         btrfs_end_transaction(trans, root);
5612                         if (ret) {
5613                                 free_extent_map(em);
5614                                 return ret;
5615                         }
5616                         goto unlock;
5617                 }
5618                 btrfs_end_transaction(trans, root);
5619         }
5620 must_cow:
5621         /*
5622          * this will cow the extent, reset the len in case we changed
5623          * it above
5624          */
5625         len = bh_result->b_size;
5626         em = btrfs_new_extent_direct(inode, em, start, len);
5627         if (IS_ERR(em))
5628                 return PTR_ERR(em);
5629         len = min(len, em->len - (start - em->start));
5630 unlock:
5631         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5632                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5633                           0, NULL, GFP_NOFS);
5634 map:
5635         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5636                 inode->i_blkbits;
5637         bh_result->b_size = len;
5638         bh_result->b_bdev = em->bdev;
5639         set_buffer_mapped(bh_result);
5640         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5641                 set_buffer_new(bh_result);
5642
5643         free_extent_map(em);
5644
5645         return 0;
5646 }
5647
5648 struct btrfs_dio_private {
5649         struct inode *inode;
5650         u64 logical_offset;
5651         u64 disk_bytenr;
5652         u64 bytes;
5653         u32 *csums;
5654         void *private;
5655
5656         /* number of bios pending for this dio */
5657         atomic_t pending_bios;
5658
5659         /* IO errors */
5660         int errors;
5661
5662         struct bio *orig_bio;
5663 };
5664
5665 static void btrfs_endio_direct_read(struct bio *bio, int err)
5666 {
5667         struct btrfs_dio_private *dip = bio->bi_private;
5668         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5669         struct bio_vec *bvec = bio->bi_io_vec;
5670         struct inode *inode = dip->inode;
5671         struct btrfs_root *root = BTRFS_I(inode)->root;
5672         u64 start;
5673         u32 *private = dip->csums;
5674
5675         start = dip->logical_offset;
5676         do {
5677                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5678                         struct page *page = bvec->bv_page;
5679                         char *kaddr;
5680                         u32 csum = ~(u32)0;
5681                         unsigned long flags;
5682
5683                         local_irq_save(flags);
5684                         kaddr = kmap_atomic(page, KM_IRQ0);
5685                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5686                                                csum, bvec->bv_len);
5687                         btrfs_csum_final(csum, (char *)&csum);
5688                         kunmap_atomic(kaddr, KM_IRQ0);
5689                         local_irq_restore(flags);
5690
5691                         flush_dcache_page(bvec->bv_page);
5692                         if (csum != *private) {
5693                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5694                                       " %llu csum %u private %u\n",
5695                                       (unsigned long long)btrfs_ino(inode),
5696                                       (unsigned long long)start,
5697                                       csum, *private);
5698                                 err = -EIO;
5699                         }
5700                 }
5701
5702                 start += bvec->bv_len;
5703                 private++;
5704                 bvec++;
5705         } while (bvec <= bvec_end);
5706
5707         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5708                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5709         bio->bi_private = dip->private;
5710
5711         kfree(dip->csums);
5712         kfree(dip);
5713
5714         /* If we had a csum failure make sure to clear the uptodate flag */
5715         if (err)
5716                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5717         dio_end_io(bio, err);
5718 }
5719
5720 static void btrfs_endio_direct_write(struct bio *bio, int err)
5721 {
5722         struct btrfs_dio_private *dip = bio->bi_private;
5723         struct inode *inode = dip->inode;
5724         struct btrfs_root *root = BTRFS_I(inode)->root;
5725         struct btrfs_trans_handle *trans;
5726         struct btrfs_ordered_extent *ordered = NULL;
5727         struct extent_state *cached_state = NULL;
5728         u64 ordered_offset = dip->logical_offset;
5729         u64 ordered_bytes = dip->bytes;
5730         int ret;
5731
5732         if (err)
5733                 goto out_done;
5734 again:
5735         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5736                                                    &ordered_offset,
5737                                                    ordered_bytes);
5738         if (!ret)
5739                 goto out_test;
5740
5741         BUG_ON(!ordered);
5742
5743         trans = btrfs_join_transaction(root, 1);
5744         if (IS_ERR(trans)) {
5745                 err = -ENOMEM;
5746                 goto out;
5747         }
5748         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5749
5750         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5751                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5752                 if (!ret)
5753                         ret = btrfs_update_inode(trans, root, inode);
5754                 err = ret;
5755                 goto out;
5756         }
5757
5758         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5759                          ordered->file_offset + ordered->len - 1, 0,
5760                          &cached_state, GFP_NOFS);
5761
5762         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5763                 ret = btrfs_mark_extent_written(trans, inode,
5764                                                 ordered->file_offset,
5765                                                 ordered->file_offset +
5766                                                 ordered->len);
5767                 if (ret) {
5768                         err = ret;
5769                         goto out_unlock;
5770                 }
5771         } else {
5772                 ret = insert_reserved_file_extent(trans, inode,
5773                                                   ordered->file_offset,
5774                                                   ordered->start,
5775                                                   ordered->disk_len,
5776                                                   ordered->len,
5777                                                   ordered->len,
5778                                                   0, 0, 0,
5779                                                   BTRFS_FILE_EXTENT_REG);
5780                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5781                                    ordered->file_offset, ordered->len);
5782                 if (ret) {
5783                         err = ret;
5784                         WARN_ON(1);
5785                         goto out_unlock;
5786                 }
5787         }
5788
5789         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5790         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5791         if (!ret)
5792                 btrfs_update_inode(trans, root, inode);
5793         ret = 0;
5794 out_unlock:
5795         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5796                              ordered->file_offset + ordered->len - 1,
5797                              &cached_state, GFP_NOFS);
5798 out:
5799         btrfs_delalloc_release_metadata(inode, ordered->len);
5800         btrfs_end_transaction(trans, root);
5801         ordered_offset = ordered->file_offset + ordered->len;
5802         btrfs_put_ordered_extent(ordered);
5803         btrfs_put_ordered_extent(ordered);
5804
5805 out_test:
5806         /*
5807          * our bio might span multiple ordered extents.  If we haven't
5808          * completed the accounting for the whole dio, go back and try again
5809          */
5810         if (ordered_offset < dip->logical_offset + dip->bytes) {
5811                 ordered_bytes = dip->logical_offset + dip->bytes -
5812                         ordered_offset;
5813                 goto again;
5814         }
5815 out_done:
5816         bio->bi_private = dip->private;
5817
5818         kfree(dip->csums);
5819         kfree(dip);
5820
5821         /* If we had an error make sure to clear the uptodate flag */
5822         if (err)
5823                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5824         dio_end_io(bio, err);
5825 }
5826
5827 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5828                                     struct bio *bio, int mirror_num,
5829                                     unsigned long bio_flags, u64 offset)
5830 {
5831         int ret;
5832         struct btrfs_root *root = BTRFS_I(inode)->root;
5833         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5834         BUG_ON(ret);
5835         return 0;
5836 }
5837
5838 static void btrfs_end_dio_bio(struct bio *bio, int err)
5839 {
5840         struct btrfs_dio_private *dip = bio->bi_private;
5841
5842         if (err) {
5843                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5844                       "sector %#Lx len %u err no %d\n",
5845                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5846                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5847                 dip->errors = 1;
5848
5849                 /*
5850                  * before atomic variable goto zero, we must make sure
5851                  * dip->errors is perceived to be set.
5852                  */
5853                 smp_mb__before_atomic_dec();
5854         }
5855
5856         /* if there are more bios still pending for this dio, just exit */
5857         if (!atomic_dec_and_test(&dip->pending_bios))
5858                 goto out;
5859
5860         if (dip->errors)
5861                 bio_io_error(dip->orig_bio);
5862         else {
5863                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5864                 bio_endio(dip->orig_bio, 0);
5865         }
5866 out:
5867         bio_put(bio);
5868 }
5869
5870 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5871                                        u64 first_sector, gfp_t gfp_flags)
5872 {
5873         int nr_vecs = bio_get_nr_vecs(bdev);
5874         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5875 }
5876
5877 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5878                                          int rw, u64 file_offset, int skip_sum,
5879                                          u32 *csums, int async_submit)
5880 {
5881         int write = rw & REQ_WRITE;
5882         struct btrfs_root *root = BTRFS_I(inode)->root;
5883         int ret;
5884
5885         bio_get(bio);
5886         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5887         if (ret)
5888                 goto err;
5889
5890         if (skip_sum)
5891                 goto map;
5892
5893         if (write && async_submit) {
5894                 ret = btrfs_wq_submit_bio(root->fs_info,
5895                                    inode, rw, bio, 0, 0,
5896                                    file_offset,
5897                                    __btrfs_submit_bio_start_direct_io,
5898                                    __btrfs_submit_bio_done);
5899                 goto err;
5900         } else if (write) {
5901                 /*
5902                  * If we aren't doing async submit, calculate the csum of the
5903                  * bio now.
5904                  */
5905                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5906                 if (ret)
5907                         goto err;
5908         } else if (!skip_sum) {
5909                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5910                                           file_offset, csums);
5911                 if (ret)
5912                         goto err;
5913         }
5914
5915 map:
5916         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5917 err:
5918         bio_put(bio);
5919         return ret;
5920 }
5921
5922 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5923                                     int skip_sum)
5924 {
5925         struct inode *inode = dip->inode;
5926         struct btrfs_root *root = BTRFS_I(inode)->root;
5927         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5928         struct bio *bio;
5929         struct bio *orig_bio = dip->orig_bio;
5930         struct bio_vec *bvec = orig_bio->bi_io_vec;
5931         u64 start_sector = orig_bio->bi_sector;
5932         u64 file_offset = dip->logical_offset;
5933         u64 submit_len = 0;
5934         u64 map_length;
5935         int nr_pages = 0;
5936         u32 *csums = dip->csums;
5937         int ret = 0;
5938         int async_submit = 0;
5939         int write = rw & REQ_WRITE;
5940
5941         map_length = orig_bio->bi_size;
5942         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5943                               &map_length, NULL, 0);
5944         if (ret) {
5945                 bio_put(orig_bio);
5946                 return -EIO;
5947         }
5948
5949         if (map_length >= orig_bio->bi_size) {
5950                 bio = orig_bio;
5951                 goto submit;
5952         }
5953
5954         async_submit = 1;
5955         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5956         if (!bio)
5957                 return -ENOMEM;
5958         bio->bi_private = dip;
5959         bio->bi_end_io = btrfs_end_dio_bio;
5960         atomic_inc(&dip->pending_bios);
5961
5962         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5963                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5964                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5965                                  bvec->bv_offset) < bvec->bv_len)) {
5966                         /*
5967                          * inc the count before we submit the bio so
5968                          * we know the end IO handler won't happen before
5969                          * we inc the count. Otherwise, the dip might get freed
5970                          * before we're done setting it up
5971                          */
5972                         atomic_inc(&dip->pending_bios);
5973                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5974                                                      file_offset, skip_sum,
5975                                                      csums, async_submit);
5976                         if (ret) {
5977                                 bio_put(bio);
5978                                 atomic_dec(&dip->pending_bios);
5979                                 goto out_err;
5980                         }
5981
5982                         /* Write's use the ordered csums */
5983                         if (!write && !skip_sum)
5984                                 csums = csums + nr_pages;
5985                         start_sector += submit_len >> 9;
5986                         file_offset += submit_len;
5987
5988                         submit_len = 0;
5989                         nr_pages = 0;
5990
5991                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5992                                                   start_sector, GFP_NOFS);
5993                         if (!bio)
5994                                 goto out_err;
5995                         bio->bi_private = dip;
5996                         bio->bi_end_io = btrfs_end_dio_bio;
5997
5998                         map_length = orig_bio->bi_size;
5999                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6000                                               &map_length, NULL, 0);
6001                         if (ret) {
6002                                 bio_put(bio);
6003                                 goto out_err;
6004                         }
6005                 } else {
6006                         submit_len += bvec->bv_len;
6007                         nr_pages ++;
6008                         bvec++;
6009                 }
6010         }
6011
6012 submit:
6013         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6014                                      csums, async_submit);
6015         if (!ret)
6016                 return 0;
6017
6018         bio_put(bio);
6019 out_err:
6020         dip->errors = 1;
6021         /*
6022          * before atomic variable goto zero, we must
6023          * make sure dip->errors is perceived to be set.
6024          */
6025         smp_mb__before_atomic_dec();
6026         if (atomic_dec_and_test(&dip->pending_bios))
6027                 bio_io_error(dip->orig_bio);
6028
6029         /* bio_end_io() will handle error, so we needn't return it */
6030         return 0;
6031 }
6032
6033 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6034                                 loff_t file_offset)
6035 {
6036         struct btrfs_root *root = BTRFS_I(inode)->root;
6037         struct btrfs_dio_private *dip;
6038         struct bio_vec *bvec = bio->bi_io_vec;
6039         int skip_sum;
6040         int write = rw & REQ_WRITE;
6041         int ret = 0;
6042
6043         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6044
6045         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6046         if (!dip) {
6047                 ret = -ENOMEM;
6048                 goto free_ordered;
6049         }
6050         dip->csums = NULL;
6051
6052         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6053         if (!write && !skip_sum) {
6054                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6055                 if (!dip->csums) {
6056                         kfree(dip);
6057                         ret = -ENOMEM;
6058                         goto free_ordered;
6059                 }
6060         }
6061
6062         dip->private = bio->bi_private;
6063         dip->inode = inode;
6064         dip->logical_offset = file_offset;
6065
6066         dip->bytes = 0;
6067         do {
6068                 dip->bytes += bvec->bv_len;
6069                 bvec++;
6070         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6071
6072         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6073         bio->bi_private = dip;
6074         dip->errors = 0;
6075         dip->orig_bio = bio;
6076         atomic_set(&dip->pending_bios, 0);
6077
6078         if (write)
6079                 bio->bi_end_io = btrfs_endio_direct_write;
6080         else
6081                 bio->bi_end_io = btrfs_endio_direct_read;
6082
6083         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6084         if (!ret)
6085                 return;
6086 free_ordered:
6087         /*
6088          * If this is a write, we need to clean up the reserved space and kill
6089          * the ordered extent.
6090          */
6091         if (write) {
6092                 struct btrfs_ordered_extent *ordered;
6093                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6094                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6095                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6096                         btrfs_free_reserved_extent(root, ordered->start,
6097                                                    ordered->disk_len);
6098                 btrfs_put_ordered_extent(ordered);
6099                 btrfs_put_ordered_extent(ordered);
6100         }
6101         bio_endio(bio, ret);
6102 }
6103
6104 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6105                         const struct iovec *iov, loff_t offset,
6106                         unsigned long nr_segs)
6107 {
6108         int seg;
6109         int i;
6110         size_t size;
6111         unsigned long addr;
6112         unsigned blocksize_mask = root->sectorsize - 1;
6113         ssize_t retval = -EINVAL;
6114         loff_t end = offset;
6115
6116         if (offset & blocksize_mask)
6117                 goto out;
6118
6119         /* Check the memory alignment.  Blocks cannot straddle pages */
6120         for (seg = 0; seg < nr_segs; seg++) {
6121                 addr = (unsigned long)iov[seg].iov_base;
6122                 size = iov[seg].iov_len;
6123                 end += size;
6124                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6125                         goto out;
6126
6127                 /* If this is a write we don't need to check anymore */
6128                 if (rw & WRITE)
6129                         continue;
6130
6131                 /*
6132                  * Check to make sure we don't have duplicate iov_base's in this
6133                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6134                  * when reading back.
6135                  */
6136                 for (i = seg + 1; i < nr_segs; i++) {
6137                         if (iov[seg].iov_base == iov[i].iov_base)
6138                                 goto out;
6139                 }
6140         }
6141         retval = 0;
6142 out:
6143         return retval;
6144 }
6145 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6146                         const struct iovec *iov, loff_t offset,
6147                         unsigned long nr_segs)
6148 {
6149         struct file *file = iocb->ki_filp;
6150         struct inode *inode = file->f_mapping->host;
6151         struct btrfs_ordered_extent *ordered;
6152         struct extent_state *cached_state = NULL;
6153         u64 lockstart, lockend;
6154         ssize_t ret;
6155         int writing = rw & WRITE;
6156         int write_bits = 0;
6157         size_t count = iov_length(iov, nr_segs);
6158
6159         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6160                             offset, nr_segs)) {
6161                 return 0;
6162         }
6163
6164         lockstart = offset;
6165         lockend = offset + count - 1;
6166
6167         if (writing) {
6168                 ret = btrfs_delalloc_reserve_space(inode, count);
6169                 if (ret)
6170                         goto out;
6171         }
6172
6173         while (1) {
6174                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6175                                  0, &cached_state, GFP_NOFS);
6176                 /*
6177                  * We're concerned with the entire range that we're going to be
6178                  * doing DIO to, so we need to make sure theres no ordered
6179                  * extents in this range.
6180                  */
6181                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6182                                                      lockend - lockstart + 1);
6183                 if (!ordered)
6184                         break;
6185                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6186                                      &cached_state, GFP_NOFS);
6187                 btrfs_start_ordered_extent(inode, ordered, 1);
6188                 btrfs_put_ordered_extent(ordered);
6189                 cond_resched();
6190         }
6191
6192         /*
6193          * we don't use btrfs_set_extent_delalloc because we don't want
6194          * the dirty or uptodate bits
6195          */
6196         if (writing) {
6197                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6198                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6199                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6200                                      GFP_NOFS);
6201                 if (ret) {
6202                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6203                                          lockend, EXTENT_LOCKED | write_bits,
6204                                          1, 0, &cached_state, GFP_NOFS);
6205                         goto out;
6206                 }
6207         }
6208
6209         free_extent_state(cached_state);
6210         cached_state = NULL;
6211
6212         ret = __blockdev_direct_IO(rw, iocb, inode,
6213                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6214                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6215                    btrfs_submit_direct, 0);
6216
6217         if (ret < 0 && ret != -EIOCBQUEUED) {
6218                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6219                               offset + iov_length(iov, nr_segs) - 1,
6220                               EXTENT_LOCKED | write_bits, 1, 0,
6221                               &cached_state, GFP_NOFS);
6222         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6223                 /*
6224                  * We're falling back to buffered, unlock the section we didn't
6225                  * do IO on.
6226                  */
6227                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6228                               offset + iov_length(iov, nr_segs) - 1,
6229                               EXTENT_LOCKED | write_bits, 1, 0,
6230                               &cached_state, GFP_NOFS);
6231         }
6232 out:
6233         free_extent_state(cached_state);
6234         return ret;
6235 }
6236
6237 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6238                 __u64 start, __u64 len)
6239 {
6240         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6241 }
6242
6243 int btrfs_readpage(struct file *file, struct page *page)
6244 {
6245         struct extent_io_tree *tree;
6246         tree = &BTRFS_I(page->mapping->host)->io_tree;
6247         return extent_read_full_page(tree, page, btrfs_get_extent);
6248 }
6249
6250 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6251 {
6252         struct extent_io_tree *tree;
6253
6254
6255         if (current->flags & PF_MEMALLOC) {
6256                 redirty_page_for_writepage(wbc, page);
6257                 unlock_page(page);
6258                 return 0;
6259         }
6260         tree = &BTRFS_I(page->mapping->host)->io_tree;
6261         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6262 }
6263
6264 int btrfs_writepages(struct address_space *mapping,
6265                      struct writeback_control *wbc)
6266 {
6267         struct extent_io_tree *tree;
6268
6269         tree = &BTRFS_I(mapping->host)->io_tree;
6270         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6271 }
6272
6273 static int
6274 btrfs_readpages(struct file *file, struct address_space *mapping,
6275                 struct list_head *pages, unsigned nr_pages)
6276 {
6277         struct extent_io_tree *tree;
6278         tree = &BTRFS_I(mapping->host)->io_tree;
6279         return extent_readpages(tree, mapping, pages, nr_pages,
6280                                 btrfs_get_extent);
6281 }
6282 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6283 {
6284         struct extent_io_tree *tree;
6285         struct extent_map_tree *map;
6286         int ret;
6287
6288         tree = &BTRFS_I(page->mapping->host)->io_tree;
6289         map = &BTRFS_I(page->mapping->host)->extent_tree;
6290         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6291         if (ret == 1) {
6292                 ClearPagePrivate(page);
6293                 set_page_private(page, 0);
6294                 page_cache_release(page);
6295         }
6296         return ret;
6297 }
6298
6299 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6300 {
6301         if (PageWriteback(page) || PageDirty(page))
6302                 return 0;
6303         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6304 }
6305
6306 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6307 {
6308         struct extent_io_tree *tree;
6309         struct btrfs_ordered_extent *ordered;
6310         struct extent_state *cached_state = NULL;
6311         u64 page_start = page_offset(page);
6312         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6313
6314
6315         /*
6316          * we have the page locked, so new writeback can't start,
6317          * and the dirty bit won't be cleared while we are here.
6318          *
6319          * Wait for IO on this page so that we can safely clear
6320          * the PagePrivate2 bit and do ordered accounting
6321          */
6322         wait_on_page_writeback(page);
6323
6324         tree = &BTRFS_I(page->mapping->host)->io_tree;
6325         if (offset) {
6326                 btrfs_releasepage(page, GFP_NOFS);
6327                 return;
6328         }
6329         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6330                          GFP_NOFS);
6331         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6332                                            page_offset(page));
6333         if (ordered) {
6334                 /*
6335                  * IO on this page will never be started, so we need
6336                  * to account for any ordered extents now
6337                  */
6338                 clear_extent_bit(tree, page_start, page_end,
6339                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6340                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6341                                  &cached_state, GFP_NOFS);
6342                 /*
6343                  * whoever cleared the private bit is responsible
6344                  * for the finish_ordered_io
6345                  */
6346                 if (TestClearPagePrivate2(page)) {
6347                         btrfs_finish_ordered_io(page->mapping->host,
6348                                                 page_start, page_end);
6349                 }
6350                 btrfs_put_ordered_extent(ordered);
6351                 cached_state = NULL;
6352                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6353                                  GFP_NOFS);
6354         }
6355         clear_extent_bit(tree, page_start, page_end,
6356                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6357                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6358         __btrfs_releasepage(page, GFP_NOFS);
6359
6360         ClearPageChecked(page);
6361         if (PagePrivate(page)) {
6362                 ClearPagePrivate(page);
6363                 set_page_private(page, 0);
6364                 page_cache_release(page);
6365         }
6366 }
6367
6368 /*
6369  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6370  * called from a page fault handler when a page is first dirtied. Hence we must
6371  * be careful to check for EOF conditions here. We set the page up correctly
6372  * for a written page which means we get ENOSPC checking when writing into
6373  * holes and correct delalloc and unwritten extent mapping on filesystems that
6374  * support these features.
6375  *
6376  * We are not allowed to take the i_mutex here so we have to play games to
6377  * protect against truncate races as the page could now be beyond EOF.  Because
6378  * vmtruncate() writes the inode size before removing pages, once we have the
6379  * page lock we can determine safely if the page is beyond EOF. If it is not
6380  * beyond EOF, then the page is guaranteed safe against truncation until we
6381  * unlock the page.
6382  */
6383 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6384 {
6385         struct page *page = vmf->page;
6386         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6387         struct btrfs_root *root = BTRFS_I(inode)->root;
6388         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6389         struct btrfs_ordered_extent *ordered;
6390         struct extent_state *cached_state = NULL;
6391         char *kaddr;
6392         unsigned long zero_start;
6393         loff_t size;
6394         int ret;
6395         u64 page_start;
6396         u64 page_end;
6397
6398         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6399         if (ret) {
6400                 if (ret == -ENOMEM)
6401                         ret = VM_FAULT_OOM;
6402                 else /* -ENOSPC, -EIO, etc */
6403                         ret = VM_FAULT_SIGBUS;
6404                 goto out;
6405         }
6406
6407         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6408 again:
6409         lock_page(page);
6410         size = i_size_read(inode);
6411         page_start = page_offset(page);
6412         page_end = page_start + PAGE_CACHE_SIZE - 1;
6413
6414         if ((page->mapping != inode->i_mapping) ||
6415             (page_start >= size)) {
6416                 /* page got truncated out from underneath us */
6417                 goto out_unlock;
6418         }
6419         wait_on_page_writeback(page);
6420
6421         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6422                          GFP_NOFS);
6423         set_page_extent_mapped(page);
6424
6425         /*
6426          * we can't set the delalloc bits if there are pending ordered
6427          * extents.  Drop our locks and wait for them to finish
6428          */
6429         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6430         if (ordered) {
6431                 unlock_extent_cached(io_tree, page_start, page_end,
6432                                      &cached_state, GFP_NOFS);
6433                 unlock_page(page);
6434                 btrfs_start_ordered_extent(inode, ordered, 1);
6435                 btrfs_put_ordered_extent(ordered);
6436                 goto again;
6437         }
6438
6439         /*
6440          * XXX - page_mkwrite gets called every time the page is dirtied, even
6441          * if it was already dirty, so for space accounting reasons we need to
6442          * clear any delalloc bits for the range we are fixing to save.  There
6443          * is probably a better way to do this, but for now keep consistent with
6444          * prepare_pages in the normal write path.
6445          */
6446         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6447                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6448                           0, 0, &cached_state, GFP_NOFS);
6449
6450         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6451                                         &cached_state);
6452         if (ret) {
6453                 unlock_extent_cached(io_tree, page_start, page_end,
6454                                      &cached_state, GFP_NOFS);
6455                 ret = VM_FAULT_SIGBUS;
6456                 goto out_unlock;
6457         }
6458         ret = 0;
6459
6460         /* page is wholly or partially inside EOF */
6461         if (page_start + PAGE_CACHE_SIZE > size)
6462                 zero_start = size & ~PAGE_CACHE_MASK;
6463         else
6464                 zero_start = PAGE_CACHE_SIZE;
6465
6466         if (zero_start != PAGE_CACHE_SIZE) {
6467                 kaddr = kmap(page);
6468                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6469                 flush_dcache_page(page);
6470                 kunmap(page);
6471         }
6472         ClearPageChecked(page);
6473         set_page_dirty(page);
6474         SetPageUptodate(page);
6475
6476         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6477         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6478
6479         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6480
6481 out_unlock:
6482         if (!ret)
6483                 return VM_FAULT_LOCKED;
6484         unlock_page(page);
6485         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6486 out:
6487         return ret;
6488 }
6489
6490 static int btrfs_truncate(struct inode *inode)
6491 {
6492         struct btrfs_root *root = BTRFS_I(inode)->root;
6493         int ret;
6494         int err = 0;
6495         struct btrfs_trans_handle *trans;
6496         unsigned long nr;
6497         u64 mask = root->sectorsize - 1;
6498
6499         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6500         if (ret)
6501                 return ret;
6502
6503         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6504         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6505
6506         trans = btrfs_start_transaction(root, 5);
6507         if (IS_ERR(trans))
6508                 return PTR_ERR(trans);
6509
6510         btrfs_set_trans_block_group(trans, inode);
6511
6512         ret = btrfs_orphan_add(trans, inode);
6513         if (ret) {
6514                 btrfs_end_transaction(trans, root);
6515                 return ret;
6516         }
6517
6518         nr = trans->blocks_used;
6519         btrfs_end_transaction(trans, root);
6520         btrfs_btree_balance_dirty(root, nr);
6521
6522         /* Now start a transaction for the truncate */
6523         trans = btrfs_start_transaction(root, 0);
6524         if (IS_ERR(trans))
6525                 return PTR_ERR(trans);
6526         btrfs_set_trans_block_group(trans, inode);
6527         trans->block_rsv = root->orphan_block_rsv;
6528
6529         /*
6530          * setattr is responsible for setting the ordered_data_close flag,
6531          * but that is only tested during the last file release.  That
6532          * could happen well after the next commit, leaving a great big
6533          * window where new writes may get lost if someone chooses to write
6534          * to this file after truncating to zero
6535          *
6536          * The inode doesn't have any dirty data here, and so if we commit
6537          * this is a noop.  If someone immediately starts writing to the inode
6538          * it is very likely we'll catch some of their writes in this
6539          * transaction, and the commit will find this file on the ordered
6540          * data list with good things to send down.
6541          *
6542          * This is a best effort solution, there is still a window where
6543          * using truncate to replace the contents of the file will
6544          * end up with a zero length file after a crash.
6545          */
6546         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6547                 btrfs_add_ordered_operation(trans, root, inode);
6548
6549         while (1) {
6550                 if (!trans) {
6551                         trans = btrfs_start_transaction(root, 0);
6552                         if (IS_ERR(trans))
6553                                 return PTR_ERR(trans);
6554                         btrfs_set_trans_block_group(trans, inode);
6555                         trans->block_rsv = root->orphan_block_rsv;
6556                 }
6557
6558                 ret = btrfs_block_rsv_check(trans, root,
6559                                             root->orphan_block_rsv, 0, 5);
6560                 if (ret == -EAGAIN) {
6561                         ret = btrfs_commit_transaction(trans, root);
6562                         if (ret)
6563                                 return ret;
6564                         trans = NULL;
6565                         continue;
6566                 } else if (ret) {
6567                         err = ret;
6568                         break;
6569                 }
6570
6571                 ret = btrfs_truncate_inode_items(trans, root, inode,
6572                                                  inode->i_size,
6573                                                  BTRFS_EXTENT_DATA_KEY);
6574                 if (ret != -EAGAIN) {
6575                         err = ret;
6576                         break;
6577                 }
6578
6579                 ret = btrfs_update_inode(trans, root, inode);
6580                 if (ret) {
6581                         err = ret;
6582                         break;
6583                 }
6584
6585                 nr = trans->blocks_used;
6586                 btrfs_end_transaction(trans, root);
6587                 trans = NULL;
6588                 btrfs_btree_balance_dirty(root, nr);
6589         }
6590
6591         if (ret == 0 && inode->i_nlink > 0) {
6592                 ret = btrfs_orphan_del(trans, inode);
6593                 if (ret)
6594                         err = ret;
6595         } else if (ret && inode->i_nlink > 0) {
6596                 /*
6597                  * Failed to do the truncate, remove us from the in memory
6598                  * orphan list.
6599                  */
6600                 ret = btrfs_orphan_del(NULL, inode);
6601         }
6602
6603         ret = btrfs_update_inode(trans, root, inode);
6604         if (ret && !err)
6605                 err = ret;
6606
6607         nr = trans->blocks_used;
6608         ret = btrfs_end_transaction_throttle(trans, root);
6609         if (ret && !err)
6610                 err = ret;
6611         btrfs_btree_balance_dirty(root, nr);
6612
6613         return err;
6614 }
6615
6616 /*
6617  * create a new subvolume directory/inode (helper for the ioctl).
6618  */
6619 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6620                              struct btrfs_root *new_root,
6621                              u64 new_dirid, u64 alloc_hint)
6622 {
6623         struct inode *inode;
6624         int err;
6625         u64 index = 0;
6626
6627         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6628                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6629         if (IS_ERR(inode))
6630                 return PTR_ERR(inode);
6631         inode->i_op = &btrfs_dir_inode_operations;
6632         inode->i_fop = &btrfs_dir_file_operations;
6633
6634         inode->i_nlink = 1;
6635         btrfs_i_size_write(inode, 0);
6636
6637         err = btrfs_update_inode(trans, new_root, inode);
6638         BUG_ON(err);
6639
6640         iput(inode);
6641         return 0;
6642 }
6643
6644 /* helper function for file defrag and space balancing.  This
6645  * forces readahead on a given range of bytes in an inode
6646  */
6647 unsigned long btrfs_force_ra(struct address_space *mapping,
6648                               struct file_ra_state *ra, struct file *file,
6649                               pgoff_t offset, pgoff_t last_index)
6650 {
6651         pgoff_t req_size = last_index - offset + 1;
6652
6653         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6654         return offset + req_size;
6655 }
6656
6657 struct inode *btrfs_alloc_inode(struct super_block *sb)
6658 {
6659         struct btrfs_inode *ei;
6660         struct inode *inode;
6661
6662         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6663         if (!ei)
6664                 return NULL;
6665
6666         ei->root = NULL;
6667         ei->space_info = NULL;
6668         ei->generation = 0;
6669         ei->sequence = 0;
6670         ei->last_trans = 0;
6671         ei->last_sub_trans = 0;
6672         ei->logged_trans = 0;
6673         ei->delalloc_bytes = 0;
6674         ei->reserved_bytes = 0;
6675         ei->disk_i_size = 0;
6676         ei->flags = 0;
6677         ei->index_cnt = (u64)-1;
6678         ei->last_unlink_trans = 0;
6679
6680         atomic_set(&ei->outstanding_extents, 0);
6681         atomic_set(&ei->reserved_extents, 0);
6682
6683         ei->ordered_data_close = 0;
6684         ei->orphan_meta_reserved = 0;
6685         ei->dummy_inode = 0;
6686         ei->force_compress = BTRFS_COMPRESS_NONE;
6687
6688         ei->delayed_node = NULL;
6689
6690         inode = &ei->vfs_inode;
6691         extent_map_tree_init(&ei->extent_tree);
6692         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6693         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6694         mutex_init(&ei->log_mutex);
6695         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6696         INIT_LIST_HEAD(&ei->i_orphan);
6697         INIT_LIST_HEAD(&ei->delalloc_inodes);
6698         INIT_LIST_HEAD(&ei->ordered_operations);
6699         RB_CLEAR_NODE(&ei->rb_node);
6700
6701         return inode;
6702 }
6703
6704 static void btrfs_i_callback(struct rcu_head *head)
6705 {
6706         struct inode *inode = container_of(head, struct inode, i_rcu);
6707         INIT_LIST_HEAD(&inode->i_dentry);
6708         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6709 }
6710
6711 void btrfs_destroy_inode(struct inode *inode)
6712 {
6713         struct btrfs_ordered_extent *ordered;
6714         struct btrfs_root *root = BTRFS_I(inode)->root;
6715
6716         WARN_ON(!list_empty(&inode->i_dentry));
6717         WARN_ON(inode->i_data.nrpages);
6718         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6719         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6720
6721         /*
6722          * This can happen where we create an inode, but somebody else also
6723          * created the same inode and we need to destroy the one we already
6724          * created.
6725          */
6726         if (!root)
6727                 goto free;
6728
6729         /*
6730          * Make sure we're properly removed from the ordered operation
6731          * lists.
6732          */
6733         smp_mb();
6734         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6735                 spin_lock(&root->fs_info->ordered_extent_lock);
6736                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6737                 spin_unlock(&root->fs_info->ordered_extent_lock);
6738         }
6739
6740         if (root == root->fs_info->tree_root) {
6741                 struct btrfs_block_group_cache *block_group;
6742
6743                 block_group = btrfs_lookup_block_group(root->fs_info,
6744                                                 BTRFS_I(inode)->block_group);
6745                 if (block_group && block_group->inode == inode) {
6746                         spin_lock(&block_group->lock);
6747                         block_group->inode = NULL;
6748                         spin_unlock(&block_group->lock);
6749                         btrfs_put_block_group(block_group);
6750                 } else if (block_group) {
6751                         btrfs_put_block_group(block_group);
6752                 }
6753         }
6754
6755         spin_lock(&root->orphan_lock);
6756         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6757                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6758                        (unsigned long long)btrfs_ino(inode));
6759                 list_del_init(&BTRFS_I(inode)->i_orphan);
6760         }
6761         spin_unlock(&root->orphan_lock);
6762
6763         while (1) {
6764                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6765                 if (!ordered)
6766                         break;
6767                 else {
6768                         printk(KERN_ERR "btrfs found ordered "
6769                                "extent %llu %llu on inode cleanup\n",
6770                                (unsigned long long)ordered->file_offset,
6771                                (unsigned long long)ordered->len);
6772                         btrfs_remove_ordered_extent(inode, ordered);
6773                         btrfs_put_ordered_extent(ordered);
6774                         btrfs_put_ordered_extent(ordered);
6775                 }
6776         }
6777         inode_tree_del(inode);
6778         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6779 free:
6780         btrfs_remove_delayed_node(inode);
6781         call_rcu(&inode->i_rcu, btrfs_i_callback);
6782 }
6783
6784 int btrfs_drop_inode(struct inode *inode)
6785 {
6786         struct btrfs_root *root = BTRFS_I(inode)->root;
6787
6788         if (btrfs_root_refs(&root->root_item) == 0 &&
6789             !is_free_space_inode(root, inode))
6790                 return 1;
6791         else
6792                 return generic_drop_inode(inode);
6793 }
6794
6795 static void init_once(void *foo)
6796 {
6797         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6798
6799         inode_init_once(&ei->vfs_inode);
6800 }
6801
6802 void btrfs_destroy_cachep(void)
6803 {
6804         if (btrfs_inode_cachep)
6805                 kmem_cache_destroy(btrfs_inode_cachep);
6806         if (btrfs_trans_handle_cachep)
6807                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6808         if (btrfs_transaction_cachep)
6809                 kmem_cache_destroy(btrfs_transaction_cachep);
6810         if (btrfs_path_cachep)
6811                 kmem_cache_destroy(btrfs_path_cachep);
6812         if (btrfs_free_space_cachep)
6813                 kmem_cache_destroy(btrfs_free_space_cachep);
6814 }
6815
6816 int btrfs_init_cachep(void)
6817 {
6818         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6819                         sizeof(struct btrfs_inode), 0,
6820                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6821         if (!btrfs_inode_cachep)
6822                 goto fail;
6823
6824         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6825                         sizeof(struct btrfs_trans_handle), 0,
6826                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6827         if (!btrfs_trans_handle_cachep)
6828                 goto fail;
6829
6830         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6831                         sizeof(struct btrfs_transaction), 0,
6832                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6833         if (!btrfs_transaction_cachep)
6834                 goto fail;
6835
6836         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6837                         sizeof(struct btrfs_path), 0,
6838                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6839         if (!btrfs_path_cachep)
6840                 goto fail;
6841
6842         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6843                         sizeof(struct btrfs_free_space), 0,
6844                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6845         if (!btrfs_free_space_cachep)
6846                 goto fail;
6847
6848         return 0;
6849 fail:
6850         btrfs_destroy_cachep();
6851         return -ENOMEM;
6852 }
6853
6854 static int btrfs_getattr(struct vfsmount *mnt,
6855                          struct dentry *dentry, struct kstat *stat)
6856 {
6857         struct inode *inode = dentry->d_inode;
6858         generic_fillattr(inode, stat);
6859         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6860         stat->blksize = PAGE_CACHE_SIZE;
6861         stat->blocks = (inode_get_bytes(inode) +
6862                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6863         return 0;
6864 }
6865
6866 /*
6867  * If a file is moved, it will inherit the cow and compression flags of the new
6868  * directory.
6869  */
6870 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6871 {
6872         struct btrfs_inode *b_dir = BTRFS_I(dir);
6873         struct btrfs_inode *b_inode = BTRFS_I(inode);
6874
6875         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6876                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6877         else
6878                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6879
6880         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6881                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6882         else
6883                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6884 }
6885
6886 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6887                            struct inode *new_dir, struct dentry *new_dentry)
6888 {
6889         struct btrfs_trans_handle *trans;
6890         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6891         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6892         struct inode *new_inode = new_dentry->d_inode;
6893         struct inode *old_inode = old_dentry->d_inode;
6894         struct timespec ctime = CURRENT_TIME;
6895         u64 index = 0;
6896         u64 root_objectid;
6897         int ret;
6898         u64 old_ino = btrfs_ino(old_inode);
6899
6900         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6901                 return -EPERM;
6902
6903         /* we only allow rename subvolume link between subvolumes */
6904         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6905                 return -EXDEV;
6906
6907         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6908             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6909                 return -ENOTEMPTY;
6910
6911         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6912             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6913                 return -ENOTEMPTY;
6914         /*
6915          * we're using rename to replace one file with another.
6916          * and the replacement file is large.  Start IO on it now so
6917          * we don't add too much work to the end of the transaction
6918          */
6919         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6920             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6921                 filemap_flush(old_inode->i_mapping);
6922
6923         /* close the racy window with snapshot create/destroy ioctl */
6924         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6925                 down_read(&root->fs_info->subvol_sem);
6926         /*
6927          * We want to reserve the absolute worst case amount of items.  So if
6928          * both inodes are subvols and we need to unlink them then that would
6929          * require 4 item modifications, but if they are both normal inodes it
6930          * would require 5 item modifications, so we'll assume their normal
6931          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6932          * should cover the worst case number of items we'll modify.
6933          */
6934         trans = btrfs_start_transaction(root, 20);
6935         if (IS_ERR(trans)) {
6936                 ret = PTR_ERR(trans);
6937                 goto out_notrans;
6938         }
6939
6940         btrfs_set_trans_block_group(trans, new_dir);
6941
6942         if (dest != root)
6943                 btrfs_record_root_in_trans(trans, dest);
6944
6945         ret = btrfs_set_inode_index(new_dir, &index);
6946         if (ret)
6947                 goto out_fail;
6948
6949         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6950                 /* force full log commit if subvolume involved. */
6951                 root->fs_info->last_trans_log_full_commit = trans->transid;
6952         } else {
6953                 ret = btrfs_insert_inode_ref(trans, dest,
6954                                              new_dentry->d_name.name,
6955                                              new_dentry->d_name.len,
6956                                              old_ino,
6957                                              btrfs_ino(new_dir), index);
6958                 if (ret)
6959                         goto out_fail;
6960                 /*
6961                  * this is an ugly little race, but the rename is required
6962                  * to make sure that if we crash, the inode is either at the
6963                  * old name or the new one.  pinning the log transaction lets
6964                  * us make sure we don't allow a log commit to come in after
6965                  * we unlink the name but before we add the new name back in.
6966                  */
6967                 btrfs_pin_log_trans(root);
6968         }
6969         /*
6970          * make sure the inode gets flushed if it is replacing
6971          * something.
6972          */
6973         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
6974                 btrfs_add_ordered_operation(trans, root, old_inode);
6975
6976         old_dir->i_ctime = old_dir->i_mtime = ctime;
6977         new_dir->i_ctime = new_dir->i_mtime = ctime;
6978         old_inode->i_ctime = ctime;
6979
6980         if (old_dentry->d_parent != new_dentry->d_parent)
6981                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6982
6983         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6984                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6985                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6986                                         old_dentry->d_name.name,
6987                                         old_dentry->d_name.len);
6988         } else {
6989                 ret = __btrfs_unlink_inode(trans, root, old_dir,
6990                                         old_dentry->d_inode,
6991                                         old_dentry->d_name.name,
6992                                         old_dentry->d_name.len);
6993                 if (!ret)
6994                         ret = btrfs_update_inode(trans, root, old_inode);
6995         }
6996         BUG_ON(ret);
6997
6998         if (new_inode) {
6999                 new_inode->i_ctime = CURRENT_TIME;
7000                 if (unlikely(btrfs_ino(new_inode) ==
7001                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7002                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7003                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7004                                                 root_objectid,
7005                                                 new_dentry->d_name.name,
7006                                                 new_dentry->d_name.len);
7007                         BUG_ON(new_inode->i_nlink == 0);
7008                 } else {
7009                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7010                                                  new_dentry->d_inode,
7011                                                  new_dentry->d_name.name,
7012                                                  new_dentry->d_name.len);
7013                 }
7014                 BUG_ON(ret);
7015                 if (new_inode->i_nlink == 0) {
7016                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7017                         BUG_ON(ret);
7018                 }
7019         }
7020
7021         fixup_inode_flags(new_dir, old_inode);
7022
7023         ret = btrfs_add_link(trans, new_dir, old_inode,
7024                              new_dentry->d_name.name,
7025                              new_dentry->d_name.len, 0, index);
7026         BUG_ON(ret);
7027
7028         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7029                 struct dentry *parent = dget_parent(new_dentry);
7030                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7031                 dput(parent);
7032                 btrfs_end_log_trans(root);
7033         }
7034 out_fail:
7035         btrfs_end_transaction_throttle(trans, root);
7036 out_notrans:
7037         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7038                 up_read(&root->fs_info->subvol_sem);
7039
7040         return ret;
7041 }
7042
7043 /*
7044  * some fairly slow code that needs optimization. This walks the list
7045  * of all the inodes with pending delalloc and forces them to disk.
7046  */
7047 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7048 {
7049         struct list_head *head = &root->fs_info->delalloc_inodes;
7050         struct btrfs_inode *binode;
7051         struct inode *inode;
7052
7053         if (root->fs_info->sb->s_flags & MS_RDONLY)
7054                 return -EROFS;
7055
7056         spin_lock(&root->fs_info->delalloc_lock);
7057         while (!list_empty(head)) {
7058                 binode = list_entry(head->next, struct btrfs_inode,
7059                                     delalloc_inodes);
7060                 inode = igrab(&binode->vfs_inode);
7061                 if (!inode)
7062                         list_del_init(&binode->delalloc_inodes);
7063                 spin_unlock(&root->fs_info->delalloc_lock);
7064                 if (inode) {
7065                         filemap_flush(inode->i_mapping);
7066                         if (delay_iput)
7067                                 btrfs_add_delayed_iput(inode);
7068                         else
7069                                 iput(inode);
7070                 }
7071                 cond_resched();
7072                 spin_lock(&root->fs_info->delalloc_lock);
7073         }
7074         spin_unlock(&root->fs_info->delalloc_lock);
7075
7076         /* the filemap_flush will queue IO into the worker threads, but
7077          * we have to make sure the IO is actually started and that
7078          * ordered extents get created before we return
7079          */
7080         atomic_inc(&root->fs_info->async_submit_draining);
7081         while (atomic_read(&root->fs_info->nr_async_submits) ||
7082               atomic_read(&root->fs_info->async_delalloc_pages)) {
7083                 wait_event(root->fs_info->async_submit_wait,
7084                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7085                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7086         }
7087         atomic_dec(&root->fs_info->async_submit_draining);
7088         return 0;
7089 }
7090
7091 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7092                          const char *symname)
7093 {
7094         struct btrfs_trans_handle *trans;
7095         struct btrfs_root *root = BTRFS_I(dir)->root;
7096         struct btrfs_path *path;
7097         struct btrfs_key key;
7098         struct inode *inode = NULL;
7099         int err;
7100         int drop_inode = 0;
7101         u64 objectid;
7102         u64 index = 0 ;
7103         int name_len;
7104         int datasize;
7105         unsigned long ptr;
7106         struct btrfs_file_extent_item *ei;
7107         struct extent_buffer *leaf;
7108         unsigned long nr = 0;
7109
7110         name_len = strlen(symname) + 1;
7111         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7112                 return -ENAMETOOLONG;
7113
7114         /*
7115          * 2 items for inode item and ref
7116          * 2 items for dir items
7117          * 1 item for xattr if selinux is on
7118          */
7119         trans = btrfs_start_transaction(root, 5);
7120         if (IS_ERR(trans))
7121                 return PTR_ERR(trans);
7122
7123         btrfs_set_trans_block_group(trans, dir);
7124
7125         err = btrfs_find_free_ino(root, &objectid);
7126         if (err)
7127                 goto out_unlock;
7128
7129         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7130                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7131                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7132                                 &index);
7133         if (IS_ERR(inode)) {
7134                 err = PTR_ERR(inode);
7135                 goto out_unlock;
7136         }
7137
7138         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7139         if (err) {
7140                 drop_inode = 1;
7141                 goto out_unlock;
7142         }
7143
7144         btrfs_set_trans_block_group(trans, inode);
7145         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7146         if (err)
7147                 drop_inode = 1;
7148         else {
7149                 inode->i_mapping->a_ops = &btrfs_aops;
7150                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7151                 inode->i_fop = &btrfs_file_operations;
7152                 inode->i_op = &btrfs_file_inode_operations;
7153                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7154         }
7155         btrfs_update_inode_block_group(trans, inode);
7156         btrfs_update_inode_block_group(trans, dir);
7157         if (drop_inode)
7158                 goto out_unlock;
7159
7160         path = btrfs_alloc_path();
7161         BUG_ON(!path);
7162         key.objectid = btrfs_ino(inode);
7163         key.offset = 0;
7164         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7165         datasize = btrfs_file_extent_calc_inline_size(name_len);
7166         err = btrfs_insert_empty_item(trans, root, path, &key,
7167                                       datasize);
7168         if (err) {
7169                 drop_inode = 1;
7170                 goto out_unlock;
7171         }
7172         leaf = path->nodes[0];
7173         ei = btrfs_item_ptr(leaf, path->slots[0],
7174                             struct btrfs_file_extent_item);
7175         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7176         btrfs_set_file_extent_type(leaf, ei,
7177                                    BTRFS_FILE_EXTENT_INLINE);
7178         btrfs_set_file_extent_encryption(leaf, ei, 0);
7179         btrfs_set_file_extent_compression(leaf, ei, 0);
7180         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7181         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7182
7183         ptr = btrfs_file_extent_inline_start(ei);
7184         write_extent_buffer(leaf, symname, ptr, name_len);
7185         btrfs_mark_buffer_dirty(leaf);
7186         btrfs_free_path(path);
7187
7188         inode->i_op = &btrfs_symlink_inode_operations;
7189         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7190         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7191         inode_set_bytes(inode, name_len);
7192         btrfs_i_size_write(inode, name_len - 1);
7193         err = btrfs_update_inode(trans, root, inode);
7194         if (err)
7195                 drop_inode = 1;
7196
7197 out_unlock:
7198         nr = trans->blocks_used;
7199         btrfs_end_transaction_throttle(trans, root);
7200         if (drop_inode) {
7201                 inode_dec_link_count(inode);
7202                 iput(inode);
7203         }
7204         btrfs_btree_balance_dirty(root, nr);
7205         return err;
7206 }
7207
7208 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7209                                        u64 start, u64 num_bytes, u64 min_size,
7210                                        loff_t actual_len, u64 *alloc_hint,
7211                                        struct btrfs_trans_handle *trans)
7212 {
7213         struct btrfs_root *root = BTRFS_I(inode)->root;
7214         struct btrfs_key ins;
7215         u64 cur_offset = start;
7216         u64 i_size;
7217         int ret = 0;
7218         bool own_trans = true;
7219
7220         if (trans)
7221                 own_trans = false;
7222         while (num_bytes > 0) {
7223                 if (own_trans) {
7224                         trans = btrfs_start_transaction(root, 3);
7225                         if (IS_ERR(trans)) {
7226                                 ret = PTR_ERR(trans);
7227                                 break;
7228                         }
7229                 }
7230
7231                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7232                                            0, *alloc_hint, (u64)-1, &ins, 1);
7233                 if (ret) {
7234                         if (own_trans)
7235                                 btrfs_end_transaction(trans, root);
7236                         break;
7237                 }
7238
7239                 ret = insert_reserved_file_extent(trans, inode,
7240                                                   cur_offset, ins.objectid,
7241                                                   ins.offset, ins.offset,
7242                                                   ins.offset, 0, 0, 0,
7243                                                   BTRFS_FILE_EXTENT_PREALLOC);
7244                 BUG_ON(ret);
7245                 btrfs_drop_extent_cache(inode, cur_offset,
7246                                         cur_offset + ins.offset -1, 0);
7247
7248                 num_bytes -= ins.offset;
7249                 cur_offset += ins.offset;
7250                 *alloc_hint = ins.objectid + ins.offset;
7251
7252                 inode->i_ctime = CURRENT_TIME;
7253                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7254                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7255                     (actual_len > inode->i_size) &&
7256                     (cur_offset > inode->i_size)) {
7257                         if (cur_offset > actual_len)
7258                                 i_size = actual_len;
7259                         else
7260                                 i_size = cur_offset;
7261                         i_size_write(inode, i_size);
7262                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7263                 }
7264
7265                 ret = btrfs_update_inode(trans, root, inode);
7266                 BUG_ON(ret);
7267
7268                 if (own_trans)
7269                         btrfs_end_transaction(trans, root);
7270         }
7271         return ret;
7272 }
7273
7274 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7275                               u64 start, u64 num_bytes, u64 min_size,
7276                               loff_t actual_len, u64 *alloc_hint)
7277 {
7278         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7279                                            min_size, actual_len, alloc_hint,
7280                                            NULL);
7281 }
7282
7283 int btrfs_prealloc_file_range_trans(struct inode *inode,
7284                                     struct btrfs_trans_handle *trans, int mode,
7285                                     u64 start, u64 num_bytes, u64 min_size,
7286                                     loff_t actual_len, u64 *alloc_hint)
7287 {
7288         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7289                                            min_size, actual_len, alloc_hint, trans);
7290 }
7291
7292 static int btrfs_set_page_dirty(struct page *page)
7293 {
7294         return __set_page_dirty_nobuffers(page);
7295 }
7296
7297 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7298 {
7299         struct btrfs_root *root = BTRFS_I(inode)->root;
7300
7301         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7302                 return -EROFS;
7303         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7304                 return -EACCES;
7305         return generic_permission(inode, mask, flags, btrfs_check_acl);
7306 }
7307
7308 static const struct inode_operations btrfs_dir_inode_operations = {
7309         .getattr        = btrfs_getattr,
7310         .lookup         = btrfs_lookup,
7311         .create         = btrfs_create,
7312         .unlink         = btrfs_unlink,
7313         .link           = btrfs_link,
7314         .mkdir          = btrfs_mkdir,
7315         .rmdir          = btrfs_rmdir,
7316         .rename         = btrfs_rename,
7317         .symlink        = btrfs_symlink,
7318         .setattr        = btrfs_setattr,
7319         .mknod          = btrfs_mknod,
7320         .setxattr       = btrfs_setxattr,
7321         .getxattr       = btrfs_getxattr,
7322         .listxattr      = btrfs_listxattr,
7323         .removexattr    = btrfs_removexattr,
7324         .permission     = btrfs_permission,
7325 };
7326 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7327         .lookup         = btrfs_lookup,
7328         .permission     = btrfs_permission,
7329 };
7330
7331 static const struct file_operations btrfs_dir_file_operations = {
7332         .llseek         = generic_file_llseek,
7333         .read           = generic_read_dir,
7334         .readdir        = btrfs_real_readdir,
7335         .unlocked_ioctl = btrfs_ioctl,
7336 #ifdef CONFIG_COMPAT
7337         .compat_ioctl   = btrfs_ioctl,
7338 #endif
7339         .release        = btrfs_release_file,
7340         .fsync          = btrfs_sync_file,
7341 };
7342
7343 static struct extent_io_ops btrfs_extent_io_ops = {
7344         .fill_delalloc = run_delalloc_range,
7345         .submit_bio_hook = btrfs_submit_bio_hook,
7346         .merge_bio_hook = btrfs_merge_bio_hook,
7347         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7348         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7349         .writepage_start_hook = btrfs_writepage_start_hook,
7350         .readpage_io_failed_hook = btrfs_io_failed_hook,
7351         .set_bit_hook = btrfs_set_bit_hook,
7352         .clear_bit_hook = btrfs_clear_bit_hook,
7353         .merge_extent_hook = btrfs_merge_extent_hook,
7354         .split_extent_hook = btrfs_split_extent_hook,
7355 };
7356
7357 /*
7358  * btrfs doesn't support the bmap operation because swapfiles
7359  * use bmap to make a mapping of extents in the file.  They assume
7360  * these extents won't change over the life of the file and they
7361  * use the bmap result to do IO directly to the drive.
7362  *
7363  * the btrfs bmap call would return logical addresses that aren't
7364  * suitable for IO and they also will change frequently as COW
7365  * operations happen.  So, swapfile + btrfs == corruption.
7366  *
7367  * For now we're avoiding this by dropping bmap.
7368  */
7369 static const struct address_space_operations btrfs_aops = {
7370         .readpage       = btrfs_readpage,
7371         .writepage      = btrfs_writepage,
7372         .writepages     = btrfs_writepages,
7373         .readpages      = btrfs_readpages,
7374         .direct_IO      = btrfs_direct_IO,
7375         .invalidatepage = btrfs_invalidatepage,
7376         .releasepage    = btrfs_releasepage,
7377         .set_page_dirty = btrfs_set_page_dirty,
7378         .error_remove_page = generic_error_remove_page,
7379 };
7380
7381 static const struct address_space_operations btrfs_symlink_aops = {
7382         .readpage       = btrfs_readpage,
7383         .writepage      = btrfs_writepage,
7384         .invalidatepage = btrfs_invalidatepage,
7385         .releasepage    = btrfs_releasepage,
7386 };
7387
7388 static const struct inode_operations btrfs_file_inode_operations = {
7389         .getattr        = btrfs_getattr,
7390         .setattr        = btrfs_setattr,
7391         .setxattr       = btrfs_setxattr,
7392         .getxattr       = btrfs_getxattr,
7393         .listxattr      = btrfs_listxattr,
7394         .removexattr    = btrfs_removexattr,
7395         .permission     = btrfs_permission,
7396         .fiemap         = btrfs_fiemap,
7397 };
7398 static const struct inode_operations btrfs_special_inode_operations = {
7399         .getattr        = btrfs_getattr,
7400         .setattr        = btrfs_setattr,
7401         .permission     = btrfs_permission,
7402         .setxattr       = btrfs_setxattr,
7403         .getxattr       = btrfs_getxattr,
7404         .listxattr      = btrfs_listxattr,
7405         .removexattr    = btrfs_removexattr,
7406 };
7407 static const struct inode_operations btrfs_symlink_inode_operations = {
7408         .readlink       = generic_readlink,
7409         .follow_link    = page_follow_link_light,
7410         .put_link       = page_put_link,
7411         .getattr        = btrfs_getattr,
7412         .permission     = btrfs_permission,
7413         .setxattr       = btrfs_setxattr,
7414         .getxattr       = btrfs_getxattr,
7415         .listxattr      = btrfs_listxattr,
7416         .removexattr    = btrfs_removexattr,
7417 };
7418
7419 const struct dentry_operations btrfs_dentry_operations = {
7420         .d_delete       = btrfs_dentry_delete,
7421 };