fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         BUG_ON(ret);
 154         if (ret) {
 155                 err = ret;
 156                 goto fail;
 157         }
 158         leaf = path->nodes[0];
 159         ei = btrfs_item_ptr(leaf, path->slots[0],
 160                             struct btrfs_file_extent_item);
 161         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 162         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 163         btrfs_set_file_extent_encryption(leaf, ei, 0);
 164         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 165         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 166         ptr = btrfs_file_extent_inline_start(ei);
 167
 168         if (compress_type != BTRFS_COMPRESS_NONE) {
 169                 struct page *cpage;
 170                 int i = 0;
 171                 while (compressed_size > 0) {
 172                         cpage = compressed_pages[i];
 173                         cur_size = min_t(unsigned long, compressed_size,
 174                                        PAGE_CACHE_SIZE);
 175
 176                         kaddr = kmap_atomic(cpage, KM_USER0);
 177                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 178                         kunmap_atomic(kaddr, KM_USER0);
 179
 180                         i++;
 181                         ptr += cur_size;
 182                         compressed_size -= cur_size;
 183                 }
 184                 btrfs_set_file_extent_compression(leaf, ei,
 185                                                   compress_type);
 186         } else {
 187                 page = find_get_page(inode->i_mapping,
 188                                      start >> PAGE_CACHE_SHIFT);
 189                 btrfs_set_file_extent_compression(leaf, ei, 0);
 190                 kaddr = kmap_atomic(page, KM_USER0);
 191                 offset = start & (PAGE_CACHE_SIZE - 1);
 192                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 193                 kunmap_atomic(kaddr, KM_USER0);
 194                 page_cache_release(page);
 195         }
 196         btrfs_mark_buffer_dirty(leaf);
 197         btrfs_free_path(path);
 198
 199         /*
 200          * we're an inline extent, so nobody can
 201          * extend the file past i_size without locking
 202          * a page we already have locked.
 203          *
 204          * We must do any isize and inode updates
 205          * before we unlock the pages.  Otherwise we
 206          * could end up racing with unlink.
 207          */
 208         BTRFS_I(inode)->disk_i_size = inode->i_size;
 209         btrfs_update_inode(trans, root, inode);
 210
 211         return 0;
 212 fail:
 213         btrfs_free_path(path);
 214         return err;
 215 }
 216
 217
 218 /*
 219  * conditionally insert an inline extent into the file.  This
 220  * does the checks required to make sure the data is small enough
 221  * to fit as an inline extent.
 222  */
 223 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 224                                  struct btrfs_root *root,
 225                                  struct inode *inode, u64 start, u64 end,
 226                                  size_t compressed_size, int compress_type,
 227                                  struct page **compressed_pages)
 228 {
 229         u64 isize = i_size_read(inode);
 230         u64 actual_end = min(end + 1, isize);
 231         u64 inline_len = actual_end - start;
 232         u64 aligned_end = (end + root->sectorsize - 1) &
 233                         ~((u64)root->sectorsize - 1);
 234         u64 hint_byte;
 235         u64 data_len = inline_len;
 236         int ret;
 237
 238         if (compressed_size)
 239                 data_len = compressed_size;
 240
 241         if (start > 0 ||
 242             actual_end >= PAGE_CACHE_SIZE ||
 243             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 244             (!compressed_size &&
 245             (actual_end & (root->sectorsize - 1)) == 0) ||
 246             end + 1 < isize ||
 247             data_len > root->fs_info->max_inline) {
 248                 return 1;
 249         }
 250
 251         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 252                                  &hint_byte, 1);
 253         BUG_ON(ret);
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         BUG_ON(ret);
 261         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 262         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 263         return 0;
 264 }
 265
 266 struct async_extent {
 267         u64 start;
 268         u64 ram_size;
 269         u64 compressed_size;
 270         struct page **pages;
 271         unsigned long nr_pages;
 272         int compress_type;
 273         struct list_head list;
 274 };
 275
 276 struct async_cow {
 277         struct inode *inode;
 278         struct btrfs_root *root;
 279         struct page *locked_page;
 280         u64 start;
 281         u64 end;
 282         struct list_head extents;
 283         struct btrfs_work work;
 284 };
 285
 286 static noinline int add_async_extent(struct async_cow *cow,
 287                                      u64 start, u64 ram_size,
 288                                      u64 compressed_size,
 289                                      struct page **pages,
 290                                      unsigned long nr_pages,
 291                                      int compress_type)
 292 {
 293         struct async_extent *async_extent;
 294
 295         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 296         BUG_ON(!async_extent);
 297         async_extent->start = start;
 298         async_extent->ram_size = ram_size;
 299         async_extent->compressed_size = compressed_size;
 300         async_extent->pages = pages;
 301         async_extent->nr_pages = nr_pages;
 302         async_extent->compress_type = compress_type;
 303         list_add_tail(&async_extent->list, &cow->extents);
 304         return 0;
 305 }
 306
 307 /*
 308  * we create compressed extents in two phases.  The first
 309  * phase compresses a range of pages that have already been
 310  * locked (both pages and state bits are locked).
 311  *
 312  * This is done inside an ordered work queue, and the compression
 313  * is spread across many cpus.  The actual IO submission is step
 314  * two, and the ordered work queue takes care of making sure that
 315  * happens in the same order things were put onto the queue by
 316  * writepages and friends.
 317  *
 318  * If this code finds it can't get good compression, it puts an
 319  * entry onto the work queue to write the uncompressed bytes.  This
 320  * makes sure that both compressed inodes and uncompressed inodes
 321  * are written in the same order that pdflush sent them down.
 322  */
 323 static noinline int compress_file_range(struct inode *inode,
 324                                         struct page *locked_page,
 325                                         u64 start, u64 end,
 326                                         struct async_cow *async_cow,
 327                                         int *num_added)
 328 {
 329         struct btrfs_root *root = BTRFS_I(inode)->root;
 330         struct btrfs_trans_handle *trans;
 331         u64 num_bytes;
 332         u64 blocksize = root->sectorsize;
 333         u64 actual_end;
 334         u64 isize = i_size_read(inode);
 335         int ret = 0;
 336         struct page **pages = NULL;
 337         unsigned long nr_pages;
 338         unsigned long nr_pages_ret = 0;
 339         unsigned long total_compressed = 0;
 340         unsigned long total_in = 0;
 341         unsigned long max_compressed = 128 * 1024;
 342         unsigned long max_uncompressed = 128 * 1024;
 343         int i;
 344         int will_compress;
 345         int compress_type = root->fs_info->compress_type;
 346
 347         /* if this is a small write inside eof, kick off a defragbot */
 348         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 349                 btrfs_add_inode_defrag(NULL, inode);
 350
 351         actual_end = min_t(u64, isize, end + 1);
 352 again:
 353         will_compress = 0;
 354         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 355         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 356
 357         /*
 358          * we don't want to send crud past the end of i_size through
 359          * compression, that's just a waste of CPU time.  So, if the
 360          * end of the file is before the start of our current
 361          * requested range of bytes, we bail out to the uncompressed
 362          * cleanup code that can deal with all of this.
 363          *
 364          * It isn't really the fastest way to fix things, but this is a
 365          * very uncommon corner.
 366          */
 367         if (actual_end <= start)
 368                 goto cleanup_and_bail_uncompressed;
 369
 370         total_compressed = actual_end - start;
 371
 372         /* we want to make sure that amount of ram required to uncompress
 373          * an extent is reasonable, so we limit the total size in ram
 374          * of a compressed extent to 128k.  This is a crucial number
 375          * because it also controls how easily we can spread reads across
 376          * cpus for decompression.
 377          *
 378          * We also want to make sure the amount of IO required to do
 379          * a random read is reasonably small, so we limit the size of
 380          * a compressed extent to 128k.
 381          */
 382         total_compressed = min(total_compressed, max_uncompressed);
 383         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 384         num_bytes = max(blocksize,  num_bytes);
 385         total_in = 0;
 386         ret = 0;
 387
 388         /*
 389          * we do compression for mount -o compress and when the
 390          * inode has not been flagged as nocompress.  This flag can
 391          * change at any time if we discover bad compression ratios.
 392          */
 393         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 394             (btrfs_test_opt(root, COMPRESS) ||
 395              (BTRFS_I(inode)->force_compress) ||
 396              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 397                 WARN_ON(pages);
 398                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 399                 if (!pages) {
 400                         /* just bail out to the uncompressed code */
 401                         goto cont;
 402                 }
 403
 404                 if (BTRFS_I(inode)->force_compress)
 405                         compress_type = BTRFS_I(inode)->force_compress;
 406
 407                 ret = btrfs_compress_pages(compress_type,
 408                                            inode->i_mapping, start,
 409                                            total_compressed, pages,
 410                                            nr_pages, &nr_pages_ret,
 411                                            &total_in,
 412                                            &total_compressed,
 413                                            max_compressed);
 414
 415                 if (!ret) {
 416                         unsigned long offset = total_compressed &
 417                                 (PAGE_CACHE_SIZE - 1);
 418                         struct page *page = pages[nr_pages_ret - 1];
 419                         char *kaddr;
 420
 421                         /* zero the tail end of the last page, we might be
 422                          * sending it down to disk
 423                          */
 424                         if (offset) {
 425                                 kaddr = kmap_atomic(page, KM_USER0);
 426                                 memset(kaddr + offset, 0,
 427                                        PAGE_CACHE_SIZE - offset);
 428                                 kunmap_atomic(kaddr, KM_USER0);
 429                         }
 430                         will_compress = 1;
 431                 }
 432         }
 433 cont:
 434         if (start == 0) {
 435                 trans = btrfs_join_transaction(root);
 436                 BUG_ON(IS_ERR(trans));
 437                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 438
 439                 /* lets try to make an inline extent */
 440                 if (ret || total_in < (actual_end - start)) {
 441                         /* we didn't compress the entire range, try
 442                          * to make an uncompressed inline extent.
 443                          */
 444                         ret = cow_file_range_inline(trans, root, inode,
 445                                                     start, end, 0, 0, NULL);
 446                 } else {
 447                         /* try making a compressed inline extent */
 448                         ret = cow_file_range_inline(trans, root, inode,
 449                                                     start, end,
 450                                                     total_compressed,
 451                                                     compress_type, pages);
 452                 }
 453                 if (ret == 0) {
 454                         /*
 455                          * inline extent creation worked, we don't need
 456                          * to create any more async work items.  Unlock
 457                          * and free up our temp pages.
 458                          */
 459                         extent_clear_unlock_delalloc(inode,
 460                              &BTRFS_I(inode)->io_tree,
 461                              start, end, NULL,
 462                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 463                              EXTENT_CLEAR_DELALLOC |
 464                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 465
 466                         btrfs_end_transaction(trans, root);
 467                         goto free_pages_out;
 468                 }
 469                 btrfs_end_transaction(trans, root);
 470         }
 471
 472         if (will_compress) {
 473                 /*
 474                  * we aren't doing an inline extent round the compressed size
 475                  * up to a block size boundary so the allocator does sane
 476                  * things
 477                  */
 478                 total_compressed = (total_compressed + blocksize - 1) &
 479                         ~(blocksize - 1);
 480
 481                 /*
 482                  * one last check to make sure the compression is really a
 483                  * win, compare the page count read with the blocks on disk
 484                  */
 485                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 486                         ~(PAGE_CACHE_SIZE - 1);
 487                 if (total_compressed >= total_in) {
 488                         will_compress = 0;
 489                 } else {
 490                         num_bytes = total_in;
 491                 }
 492         }
 493         if (!will_compress && pages) {
 494                 /*
 495                  * the compression code ran but failed to make things smaller,
 496                  * free any pages it allocated and our page pointer array
 497                  */
 498                 for (i = 0; i < nr_pages_ret; i++) {
 499                         WARN_ON(pages[i]->mapping);
 500                         page_cache_release(pages[i]);
 501                 }
 502                 kfree(pages);
 503                 pages = NULL;
 504                 total_compressed = 0;
 505                 nr_pages_ret = 0;
 506
 507                 /* flag the file so we don't compress in the future */
 508                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 509                     !(BTRFS_I(inode)->force_compress)) {
 510                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 511                 }
 512         }
 513         if (will_compress) {
 514                 *num_added += 1;
 515
 516                 /* the async work queues will take care of doing actual
 517                  * allocation on disk for these compressed pages,
 518                  * and will submit them to the elevator.
 519                  */
 520                 add_async_extent(async_cow, start, num_bytes,
 521                                  total_compressed, pages, nr_pages_ret,
 522                                  compress_type);
 523
 524                 if (start + num_bytes < end) {
 525                         start += num_bytes;
 526                         pages = NULL;
 527                         cond_resched();
 528                         goto again;
 529                 }
 530         } else {
 531 cleanup_and_bail_uncompressed:
 532                 /*
 533                  * No compression, but we still need to write the pages in
 534                  * the file we've been given so far.  redirty the locked
 535                  * page if it corresponds to our extent and set things up
 536                  * for the async work queue to run cow_file_range to do
 537                  * the normal delalloc dance
 538                  */
 539                 if (page_offset(locked_page) >= start &&
 540                     page_offset(locked_page) <= end) {
 541                         __set_page_dirty_nobuffers(locked_page);
 542                         /* unlocked later on in the async handlers */
 543                 }
 544                 add_async_extent(async_cow, start, end - start + 1,
 545                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 546                 *num_added += 1;
 547         }
 548
 549 out:
 550         return 0;
 551
 552 free_pages_out:
 553         for (i = 0; i < nr_pages_ret; i++) {
 554                 WARN_ON(pages[i]->mapping);
 555                 page_cache_release(pages[i]);
 556         }
 557         kfree(pages);
 558
 559         goto out;
 560 }
 561
 562 /*
 563  * phase two of compressed writeback.  This is the ordered portion
 564  * of the code, which only gets called in the order the work was
 565  * queued.  We walk all the async extents created by compress_file_range
 566  * and send them down to the disk.
 567  */
 568 static noinline int submit_compressed_extents(struct inode *inode,
 569                                               struct async_cow *async_cow)
 570 {
 571         struct async_extent *async_extent;
 572         u64 alloc_hint = 0;
 573         struct btrfs_trans_handle *trans;
 574         struct btrfs_key ins;
 575         struct extent_map *em;
 576         struct btrfs_root *root = BTRFS_I(inode)->root;
 577         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 578         struct extent_io_tree *io_tree;
 579         int ret = 0;
 580
 581         if (list_empty(&async_cow->extents))
 582                 return 0;
 583
 584
 585         while (!list_empty(&async_cow->extents)) {
 586                 async_extent = list_entry(async_cow->extents.next,
 587                                           struct async_extent, list);
 588                 list_del(&async_extent->list);
 589
 590                 io_tree = &BTRFS_I(inode)->io_tree;
 591
 592 retry:
 593                 /* did the compression code fall back to uncompressed IO? */
 594                 if (!async_extent->pages) {
 595                         int page_started = 0;
 596                         unsigned long nr_written = 0;
 597
 598                         lock_extent(io_tree, async_extent->start,
 599                                          async_extent->start +
 600                                          async_extent->ram_size - 1, GFP_NOFS);
 601
 602                         /* allocate blocks */
 603                         ret = cow_file_range(inode, async_cow->locked_page,
 604                                              async_extent->start,
 605                                              async_extent->start +
 606                                              async_extent->ram_size - 1,
 607                                              &page_started, &nr_written, 0);
 608
 609                         /*
 610                          * if page_started, cow_file_range inserted an
 611                          * inline extent and took care of all the unlocking
 612                          * and IO for us.  Otherwise, we need to submit
 613                          * all those pages down to the drive.
 614                          */
 615                         if (!page_started && !ret)
 616                                 extent_write_locked_range(io_tree,
 617                                                   inode, async_extent->start,
 618                                                   async_extent->start +
 619                                                   async_extent->ram_size - 1,
 620                                                   btrfs_get_extent,
 621                                                   WB_SYNC_ALL);
 622                         kfree(async_extent);
 623                         cond_resched();
 624                         continue;
 625                 }
 626
 627                 lock_extent(io_tree, async_extent->start,
 628                             async_extent->start + async_extent->ram_size - 1,
 629                             GFP_NOFS);
 630
 631                 trans = btrfs_join_transaction(root);
 632                 BUG_ON(IS_ERR(trans));
 633                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 634                 ret = btrfs_reserve_extent(trans, root,
 635                                            async_extent->compressed_size,
 636                                            async_extent->compressed_size,
 637                                            0, alloc_hint,
 638                                            (u64)-1, &ins, 1);
 639                 btrfs_end_transaction(trans, root);
 640
 641                 if (ret) {
 642                         int i;
 643                         for (i = 0; i < async_extent->nr_pages; i++) {
 644                                 WARN_ON(async_extent->pages[i]->mapping);
 645                                 page_cache_release(async_extent->pages[i]);
 646                         }
 647                         kfree(async_extent->pages);
 648                         async_extent->nr_pages = 0;
 649                         async_extent->pages = NULL;
 650                         unlock_extent(io_tree, async_extent->start,
 651                                       async_extent->start +
 652                                       async_extent->ram_size - 1, GFP_NOFS);
 653                         goto retry;
 654                 }
 655
 656                 /*
 657                  * here we're doing allocation and writeback of the
 658                  * compressed pages
 659                  */
 660                 btrfs_drop_extent_cache(inode, async_extent->start,
 661                                         async_extent->start +
 662                                         async_extent->ram_size - 1, 0);
 663
 664                 em = alloc_extent_map();
 665                 BUG_ON(!em);
 666                 em->start = async_extent->start;
 667                 em->len = async_extent->ram_size;
 668                 em->orig_start = em->start;
 669
 670                 em->block_start = ins.objectid;
 671                 em->block_len = ins.offset;
 672                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 673                 em->compress_type = async_extent->compress_type;
 674                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 675                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 676
 677                 while (1) {
 678                         write_lock(&em_tree->lock);
 679                         ret = add_extent_mapping(em_tree, em);
 680                         write_unlock(&em_tree->lock);
 681                         if (ret != -EEXIST) {
 682                                 free_extent_map(em);
 683                                 break;
 684                         }
 685                         btrfs_drop_extent_cache(inode, async_extent->start,
 686                                                 async_extent->start +
 687                                                 async_extent->ram_size - 1, 0);
 688                 }
 689
 690                 ret = btrfs_add_ordered_extent_compress(inode,
 691                                                 async_extent->start,
 692                                                 ins.objectid,
 693                                                 async_extent->ram_size,
 694                                                 ins.offset,
 695                                                 BTRFS_ORDERED_COMPRESSED,
 696                                                 async_extent->compress_type);
 697                 BUG_ON(ret);
 698
 699                 /*
 700                  * clear dirty, set writeback and unlock the pages.
 701                  */
 702                 extent_clear_unlock_delalloc(inode,
 703                                 &BTRFS_I(inode)->io_tree,
 704                                 async_extent->start,
 705                                 async_extent->start +
 706                                 async_extent->ram_size - 1,
 707                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 708                                 EXTENT_CLEAR_UNLOCK |
 709                                 EXTENT_CLEAR_DELALLOC |
 710                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 711
 712                 ret = btrfs_submit_compressed_write(inode,
 713                                     async_extent->start,
 714                                     async_extent->ram_size,
 715                                     ins.objectid,
 716                                     ins.offset, async_extent->pages,
 717                                     async_extent->nr_pages);
 718
 719                 BUG_ON(ret);
 720                 alloc_hint = ins.objectid + ins.offset;
 721                 kfree(async_extent);
 722                 cond_resched();
 723         }
 724
 725         return 0;
 726 }
 727
 728 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 729                                       u64 num_bytes)
 730 {
 731         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 732         struct extent_map *em;
 733         u64 alloc_hint = 0;
 734
 735         read_lock(&em_tree->lock);
 736         em = search_extent_mapping(em_tree, start, num_bytes);
 737         if (em) {
 738                 /*
 739                  * if block start isn't an actual block number then find the
 740                  * first block in this inode and use that as a hint.  If that
 741                  * block is also bogus then just don't worry about it.
 742                  */
 743                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 744                         free_extent_map(em);
 745                         em = search_extent_mapping(em_tree, 0, 0);
 746                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 747                                 alloc_hint = em->block_start;
 748                         if (em)
 749                                 free_extent_map(em);
 750                 } else {
 751                         alloc_hint = em->block_start;
 752                         free_extent_map(em);
 753                 }
 754         }
 755         read_unlock(&em_tree->lock);
 756
 757         return alloc_hint;
 758 }
 759
 760 /*
 761  * when extent_io.c finds a delayed allocation range in the file,
 762  * the call backs end up in this code.  The basic idea is to
 763  * allocate extents on disk for the range, and create ordered data structs
 764  * in ram to track those extents.
 765  *
 766  * locked_page is the page that writepage had locked already.  We use
 767  * it to make sure we don't do extra locks or unlocks.
 768  *
 769  * *page_started is set to one if we unlock locked_page and do everything
 770  * required to start IO on it.  It may be clean and already done with
 771  * IO when we return.
 772  */
 773 static noinline int cow_file_range(struct inode *inode,
 774                                    struct page *locked_page,
 775                                    u64 start, u64 end, int *page_started,
 776                                    unsigned long *nr_written,
 777                                    int unlock)
 778 {
 779         struct btrfs_root *root = BTRFS_I(inode)->root;
 780         struct btrfs_trans_handle *trans;
 781         u64 alloc_hint = 0;
 782         u64 num_bytes;
 783         unsigned long ram_size;
 784         u64 disk_num_bytes;
 785         u64 cur_alloc_size;
 786         u64 blocksize = root->sectorsize;
 787         struct btrfs_key ins;
 788         struct extent_map *em;
 789         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 790         int ret = 0;
 791
 792         BUG_ON(btrfs_is_free_space_inode(root, inode));
 793         trans = btrfs_join_transaction(root);
 794         BUG_ON(IS_ERR(trans));
 795         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 796
 797         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 798         num_bytes = max(blocksize,  num_bytes);
 799         disk_num_bytes = num_bytes;
 800         ret = 0;
 801
 802         /* if this is a small write inside eof, kick off defrag */
 803         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 804                 btrfs_add_inode_defrag(trans, inode);
 805
 806         if (start == 0) {
 807                 /* lets try to make an inline extent */
 808                 ret = cow_file_range_inline(trans, root, inode,
 809                                             start, end, 0, 0, NULL);
 810                 if (ret == 0) {
 811                         extent_clear_unlock_delalloc(inode,
 812                                      &BTRFS_I(inode)->io_tree,
 813                                      start, end, NULL,
 814                                      EXTENT_CLEAR_UNLOCK_PAGE |
 815                                      EXTENT_CLEAR_UNLOCK |
 816                                      EXTENT_CLEAR_DELALLOC |
 817                                      EXTENT_CLEAR_DIRTY |
 818                                      EXTENT_SET_WRITEBACK |
 819                                      EXTENT_END_WRITEBACK);
 820
 821                         *nr_written = *nr_written +
 822                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 823                         *page_started = 1;
 824                         ret = 0;
 825                         goto out;
 826                 }
 827         }
 828
 829         BUG_ON(disk_num_bytes >
 830                btrfs_super_total_bytes(root->fs_info->super_copy));
 831
 832         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 833         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 834
 835         while (disk_num_bytes > 0) {
 836                 unsigned long op;
 837
 838                 cur_alloc_size = disk_num_bytes;
 839                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 840                                            root->sectorsize, 0, alloc_hint,
 841                                            (u64)-1, &ins, 1);
 842                 BUG_ON(ret);
 843
 844                 em = alloc_extent_map();
 845                 BUG_ON(!em);
 846                 em->start = start;
 847                 em->orig_start = em->start;
 848                 ram_size = ins.offset;
 849                 em->len = ins.offset;
 850
 851                 em->block_start = ins.objectid;
 852                 em->block_len = ins.offset;
 853                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 854                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 855
 856                 while (1) {
 857                         write_lock(&em_tree->lock);
 858                         ret = add_extent_mapping(em_tree, em);
 859                         write_unlock(&em_tree->lock);
 860                         if (ret != -EEXIST) {
 861                                 free_extent_map(em);
 862                                 break;
 863                         }
 864                         btrfs_drop_extent_cache(inode, start,
 865                                                 start + ram_size - 1, 0);
 866                 }
 867
 868                 cur_alloc_size = ins.offset;
 869                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 870                                                ram_size, cur_alloc_size, 0);
 871                 BUG_ON(ret);
 872
 873                 if (root->root_key.objectid ==
 874                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 875                         ret = btrfs_reloc_clone_csums(inode, start,
 876                                                       cur_alloc_size);
 877                         BUG_ON(ret);
 878                 }
 879
 880                 if (disk_num_bytes < cur_alloc_size)
 881                         break;
 882
 883                 /* we're not doing compressed IO, don't unlock the first
 884                  * page (which the caller expects to stay locked), don't
 885                  * clear any dirty bits and don't set any writeback bits
 886                  *
 887                  * Do set the Private2 bit so we know this page was properly
 888                  * setup for writepage
 889                  */
 890                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 891                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 892                         EXTENT_SET_PRIVATE2;
 893
 894                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 895                                              start, start + ram_size - 1,
 896                                              locked_page, op);
 897                 disk_num_bytes -= cur_alloc_size;
 898                 num_bytes -= cur_alloc_size;
 899                 alloc_hint = ins.objectid + ins.offset;
 900                 start += cur_alloc_size;
 901         }
 902 out:
 903         ret = 0;
 904         btrfs_end_transaction(trans, root);
 905
 906         return ret;
 907 }
 908
 909 /*
 910  * work queue call back to started compression on a file and pages
 911  */
 912 static noinline void async_cow_start(struct btrfs_work *work)
 913 {
 914         struct async_cow *async_cow;
 915         int num_added = 0;
 916         async_cow = container_of(work, struct async_cow, work);
 917
 918         compress_file_range(async_cow->inode, async_cow->locked_page,
 919                             async_cow->start, async_cow->end, async_cow,
 920                             &num_added);
 921         if (num_added == 0)
 922                 async_cow->inode = NULL;
 923 }
 924
 925 /*
 926  * work queue call back to submit previously compressed pages
 927  */
 928 static noinline void async_cow_submit(struct btrfs_work *work)
 929 {
 930         struct async_cow *async_cow;
 931         struct btrfs_root *root;
 932         unsigned long nr_pages;
 933
 934         async_cow = container_of(work, struct async_cow, work);
 935
 936         root = async_cow->root;
 937         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 938                 PAGE_CACHE_SHIFT;
 939
 940         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 941
 942         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 943             5 * 1042 * 1024 &&
 944             waitqueue_active(&root->fs_info->async_submit_wait))
 945                 wake_up(&root->fs_info->async_submit_wait);
 946
 947         if (async_cow->inode)
 948                 submit_compressed_extents(async_cow->inode, async_cow);
 949 }
 950
 951 static noinline void async_cow_free(struct btrfs_work *work)
 952 {
 953         struct async_cow *async_cow;
 954         async_cow = container_of(work, struct async_cow, work);
 955         kfree(async_cow);
 956 }
 957
 958 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 959                                 u64 start, u64 end, int *page_started,
 960                                 unsigned long *nr_written)
 961 {
 962         struct async_cow *async_cow;
 963         struct btrfs_root *root = BTRFS_I(inode)->root;
 964         unsigned long nr_pages;
 965         u64 cur_end;
 966         int limit = 10 * 1024 * 1042;
 967
 968         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 969                          1, 0, NULL, GFP_NOFS);
 970         while (start < end) {
 971                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 972                 BUG_ON(!async_cow);
 973                 async_cow->inode = inode;
 974                 async_cow->root = root;
 975                 async_cow->locked_page = locked_page;
 976                 async_cow->start = start;
 977
 978                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 979                         cur_end = end;
 980                 else
 981                         cur_end = min(end, start + 512 * 1024 - 1);
 982
 983                 async_cow->end = cur_end;
 984                 INIT_LIST_HEAD(&async_cow->extents);
 985
 986                 async_cow->work.func = async_cow_start;
 987                 async_cow->work.ordered_func = async_cow_submit;
 988                 async_cow->work.ordered_free = async_cow_free;
 989                 async_cow->work.flags = 0;
 990
 991                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 992                         PAGE_CACHE_SHIFT;
 993                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 994
 995                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 996                                    &async_cow->work);
 997
 998                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 999                         wait_event(root->fs_info->async_submit_wait,
1000                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1001                             limit));
1002                 }
1003
1004                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1005                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1006                         wait_event(root->fs_info->async_submit_wait,
1007                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1008                            0));
1009                 }
1010
1011                 *nr_written += nr_pages;
1012                 start = cur_end + 1;
1013         }
1014         *page_started = 1;
1015         return 0;
1016 }
1017
1018 static noinline int csum_exist_in_range(struct btrfs_root *root,
1019                                         u64 bytenr, u64 num_bytes)
1020 {
1021         int ret;
1022         struct btrfs_ordered_sum *sums;
1023         LIST_HEAD(list);
1024
1025         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1026                                        bytenr + num_bytes - 1, &list, 0);
1027         if (ret == 0 && list_empty(&list))
1028                 return 0;
1029
1030         while (!list_empty(&list)) {
1031                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1032                 list_del(&sums->list);
1033                 kfree(sums);
1034         }
1035         return 1;
1036 }
1037
1038 /*
1039  * when nowcow writeback call back.  This checks for snapshots or COW copies
1040  * of the extents that exist in the file, and COWs the file as required.
1041  *
1042  * If no cow copies or snapshots exist, we write directly to the existing
1043  * blocks on disk
1044  */
1045 static noinline int run_delalloc_nocow(struct inode *inode,
1046                                        struct page *locked_page,
1047                               u64 start, u64 end, int *page_started, int force,
1048                               unsigned long *nr_written)
1049 {
1050         struct btrfs_root *root = BTRFS_I(inode)->root;
1051         struct btrfs_trans_handle *trans;
1052         struct extent_buffer *leaf;
1053         struct btrfs_path *path;
1054         struct btrfs_file_extent_item *fi;
1055         struct btrfs_key found_key;
1056         u64 cow_start;
1057         u64 cur_offset;
1058         u64 extent_end;
1059         u64 extent_offset;
1060         u64 disk_bytenr;
1061         u64 num_bytes;
1062         int extent_type;
1063         int ret;
1064         int type;
1065         int nocow;
1066         int check_prev = 1;
1067         bool nolock;
1068         u64 ino = btrfs_ino(inode);
1069
1070         path = btrfs_alloc_path();
1071         if (!path)
1072                 return -ENOMEM;
1073
1074         nolock = btrfs_is_free_space_inode(root, inode);
1075
1076         if (nolock)
1077                 trans = btrfs_join_transaction_nolock(root);
1078         else
1079                 trans = btrfs_join_transaction(root);
1080
1081         BUG_ON(IS_ERR(trans));
1082         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1083
1084         cow_start = (u64)-1;
1085         cur_offset = start;
1086         while (1) {
1087                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1088                                                cur_offset, 0);
1089                 BUG_ON(ret < 0);
1090                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1091                         leaf = path->nodes[0];
1092                         btrfs_item_key_to_cpu(leaf, &found_key,
1093                                               path->slots[0] - 1);
1094                         if (found_key.objectid == ino &&
1095                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1096                                 path->slots[0]--;
1097                 }
1098                 check_prev = 0;
1099 next_slot:
1100                 leaf = path->nodes[0];
1101                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1102                         ret = btrfs_next_leaf(root, path);
1103                         if (ret < 0)
1104                                 BUG_ON(1);
1105                         if (ret > 0)
1106                                 break;
1107                         leaf = path->nodes[0];
1108                 }
1109
1110                 nocow = 0;
1111                 disk_bytenr = 0;
1112                 num_bytes = 0;
1113                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1114
1115                 if (found_key.objectid > ino ||
1116                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1117                     found_key.offset > end)
1118                         break;
1119
1120                 if (found_key.offset > cur_offset) {
1121                         extent_end = found_key.offset;
1122                         extent_type = 0;
1123                         goto out_check;
1124                 }
1125
1126                 fi = btrfs_item_ptr(leaf, path->slots[0],
1127                                     struct btrfs_file_extent_item);
1128                 extent_type = btrfs_file_extent_type(leaf, fi);
1129
1130                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1131                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1132                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1133                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1134                         extent_end = found_key.offset +
1135                                 btrfs_file_extent_num_bytes(leaf, fi);
1136                         if (extent_end <= start) {
1137                                 path->slots[0]++;
1138                                 goto next_slot;
1139                         }
1140                         if (disk_bytenr == 0)
1141                                 goto out_check;
1142                         if (btrfs_file_extent_compression(leaf, fi) ||
1143                             btrfs_file_extent_encryption(leaf, fi) ||
1144                             btrfs_file_extent_other_encoding(leaf, fi))
1145                                 goto out_check;
1146                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1147                                 goto out_check;
1148                         if (btrfs_extent_readonly(root, disk_bytenr))
1149                                 goto out_check;
1150                         if (btrfs_cross_ref_exist(trans, root, ino,
1151                                                   found_key.offset -
1152                                                   extent_offset, disk_bytenr))
1153                                 goto out_check;
1154                         disk_bytenr += extent_offset;
1155                         disk_bytenr += cur_offset - found_key.offset;
1156                         num_bytes = min(end + 1, extent_end) - cur_offset;
1157                         /*
1158                          * force cow if csum exists in the range.
1159                          * this ensure that csum for a given extent are
1160                          * either valid or do not exist.
1161                          */
1162                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1163                                 goto out_check;
1164                         nocow = 1;
1165                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1166                         extent_end = found_key.offset +
1167                                 btrfs_file_extent_inline_len(leaf, fi);
1168                         extent_end = ALIGN(extent_end, root->sectorsize);
1169                 } else {
1170                         BUG_ON(1);
1171                 }
1172 out_check:
1173                 if (extent_end <= start) {
1174                         path->slots[0]++;
1175                         goto next_slot;
1176                 }
1177                 if (!nocow) {
1178                         if (cow_start == (u64)-1)
1179                                 cow_start = cur_offset;
1180                         cur_offset = extent_end;
1181                         if (cur_offset > end)
1182                                 break;
1183                         path->slots[0]++;
1184                         goto next_slot;
1185                 }
1186
1187                 btrfs_release_path(path);
1188                 if (cow_start != (u64)-1) {
1189                         ret = cow_file_range(inode, locked_page, cow_start,
1190                                         found_key.offset - 1, page_started,
1191                                         nr_written, 1);
1192                         BUG_ON(ret);
1193                         cow_start = (u64)-1;
1194                 }
1195
1196                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1197                         struct extent_map *em;
1198                         struct extent_map_tree *em_tree;
1199                         em_tree = &BTRFS_I(inode)->extent_tree;
1200                         em = alloc_extent_map();
1201                         BUG_ON(!em);
1202                         em->start = cur_offset;
1203                         em->orig_start = em->start;
1204                         em->len = num_bytes;
1205                         em->block_len = num_bytes;
1206                         em->block_start = disk_bytenr;
1207                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1208                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1209                         while (1) {
1210                                 write_lock(&em_tree->lock);
1211                                 ret = add_extent_mapping(em_tree, em);
1212                                 write_unlock(&em_tree->lock);
1213                                 if (ret != -EEXIST) {
1214                                         free_extent_map(em);
1215                                         break;
1216                                 }
1217                                 btrfs_drop_extent_cache(inode, em->start,
1218                                                 em->start + em->len - 1, 0);
1219                         }
1220                         type = BTRFS_ORDERED_PREALLOC;
1221                 } else {
1222                         type = BTRFS_ORDERED_NOCOW;
1223                 }
1224
1225                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1226                                                num_bytes, num_bytes, type);
1227                 BUG_ON(ret);
1228
1229                 if (root->root_key.objectid ==
1230                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1231                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1232                                                       num_bytes);
1233                         BUG_ON(ret);
1234                 }
1235
1236                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1237                                 cur_offset, cur_offset + num_bytes - 1,
1238                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1239                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1240                                 EXTENT_SET_PRIVATE2);
1241                 cur_offset = extent_end;
1242                 if (cur_offset > end)
1243                         break;
1244         }
1245         btrfs_release_path(path);
1246
1247         if (cur_offset <= end && cow_start == (u64)-1)
1248                 cow_start = cur_offset;
1249         if (cow_start != (u64)-1) {
1250                 ret = cow_file_range(inode, locked_page, cow_start, end,
1251                                      page_started, nr_written, 1);
1252                 BUG_ON(ret);
1253         }
1254
1255         if (nolock) {
1256                 ret = btrfs_end_transaction_nolock(trans, root);
1257                 BUG_ON(ret);
1258         } else {
1259                 ret = btrfs_end_transaction(trans, root);
1260                 BUG_ON(ret);
1261         }
1262         btrfs_free_path(path);
1263         return 0;
1264 }
1265
1266 /*
1267  * extent_io.c call back to do delayed allocation processing
1268  */
1269 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1270                               u64 start, u64 end, int *page_started,
1271                               unsigned long *nr_written)
1272 {
1273         int ret;
1274         struct btrfs_root *root = BTRFS_I(inode)->root;
1275
1276         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 1, nr_written);
1279         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1280                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1281                                          page_started, 0, nr_written);
1282         else if (!btrfs_test_opt(root, COMPRESS) &&
1283                  !(BTRFS_I(inode)->force_compress) &&
1284                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1285                 ret = cow_file_range(inode, locked_page, start, end,
1286                                       page_started, nr_written, 1);
1287         else
1288                 ret = cow_file_range_async(inode, locked_page, start, end,
1289                                            page_started, nr_written);
1290         return ret;
1291 }
1292
1293 static void btrfs_split_extent_hook(struct inode *inode,
1294                                     struct extent_state *orig, u64 split)
1295 {
1296         /* not delalloc, ignore it */
1297         if (!(orig->state & EXTENT_DELALLOC))
1298                 return;
1299
1300         spin_lock(&BTRFS_I(inode)->lock);
1301         BTRFS_I(inode)->outstanding_extents++;
1302         spin_unlock(&BTRFS_I(inode)->lock);
1303 }
1304
1305 /*
1306  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1307  * extents so we can keep track of new extents that are just merged onto old
1308  * extents, such as when we are doing sequential writes, so we can properly
1309  * account for the metadata space we'll need.
1310  */
1311 static void btrfs_merge_extent_hook(struct inode *inode,
1312                                     struct extent_state *new,
1313                                     struct extent_state *other)
1314 {
1315         /* not delalloc, ignore it */
1316         if (!(other->state & EXTENT_DELALLOC))
1317                 return;
1318
1319         spin_lock(&BTRFS_I(inode)->lock);
1320         BTRFS_I(inode)->outstanding_extents--;
1321         spin_unlock(&BTRFS_I(inode)->lock);
1322 }
1323
1324 /*
1325  * extent_io.c set_bit_hook, used to track delayed allocation
1326  * bytes in this file, and to maintain the list of inodes that
1327  * have pending delalloc work to be done.
1328  */
1329 static void btrfs_set_bit_hook(struct inode *inode,
1330                                struct extent_state *state, int *bits)
1331 {
1332
1333         /*
1334          * set_bit and clear bit hooks normally require _irqsave/restore
1335          * but in this case, we are only testing for the DELALLOC
1336          * bit, which is only set or cleared with irqs on
1337          */
1338         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1339                 struct btrfs_root *root = BTRFS_I(inode)->root;
1340                 u64 len = state->end + 1 - state->start;
1341                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1342
1343                 if (*bits & EXTENT_FIRST_DELALLOC) {
1344                         *bits &= ~EXTENT_FIRST_DELALLOC;
1345                 } else {
1346                         spin_lock(&BTRFS_I(inode)->lock);
1347                         BTRFS_I(inode)->outstanding_extents++;
1348                         spin_unlock(&BTRFS_I(inode)->lock);
1349                 }
1350
1351                 spin_lock(&root->fs_info->delalloc_lock);
1352                 BTRFS_I(inode)->delalloc_bytes += len;
1353                 root->fs_info->delalloc_bytes += len;
1354                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1355                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1356                                       &root->fs_info->delalloc_inodes);
1357                 }
1358                 spin_unlock(&root->fs_info->delalloc_lock);
1359         }
1360 }
1361
1362 /*
1363  * extent_io.c clear_bit_hook, see set_bit_hook for why
1364  */
1365 static void btrfs_clear_bit_hook(struct inode *inode,
1366                                  struct extent_state *state, int *bits)
1367 {
1368         /*
1369          * set_bit and clear bit hooks normally require _irqsave/restore
1370          * but in this case, we are only testing for the DELALLOC
1371          * bit, which is only set or cleared with irqs on
1372          */
1373         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1374                 struct btrfs_root *root = BTRFS_I(inode)->root;
1375                 u64 len = state->end + 1 - state->start;
1376                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1377
1378                 if (*bits & EXTENT_FIRST_DELALLOC) {
1379                         *bits &= ~EXTENT_FIRST_DELALLOC;
1380                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1381                         spin_lock(&BTRFS_I(inode)->lock);
1382                         BTRFS_I(inode)->outstanding_extents--;
1383                         spin_unlock(&BTRFS_I(inode)->lock);
1384                 }
1385
1386                 if (*bits & EXTENT_DO_ACCOUNTING)
1387                         btrfs_delalloc_release_metadata(inode, len);
1388
1389                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1390                     && do_list)
1391                         btrfs_free_reserved_data_space(inode, len);
1392
1393                 spin_lock(&root->fs_info->delalloc_lock);
1394                 root->fs_info->delalloc_bytes -= len;
1395                 BTRFS_I(inode)->delalloc_bytes -= len;
1396
1397                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1398                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1399                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1400                 }
1401                 spin_unlock(&root->fs_info->delalloc_lock);
1402         }
1403 }
1404
1405 /*
1406  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1407  * we don't create bios that span stripes or chunks
1408  */
1409 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1410                          size_t size, struct bio *bio,
1411                          unsigned long bio_flags)
1412 {
1413         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1414         struct btrfs_mapping_tree *map_tree;
1415         u64 logical = (u64)bio->bi_sector << 9;
1416         u64 length = 0;
1417         u64 map_length;
1418         int ret;
1419
1420         if (bio_flags & EXTENT_BIO_COMPRESSED)
1421                 return 0;
1422
1423         length = bio->bi_size;
1424         map_tree = &root->fs_info->mapping_tree;
1425         map_length = length;
1426         ret = btrfs_map_block(map_tree, READ, logical,
1427                               &map_length, NULL, 0);
1428
1429         if (map_length < length + size)
1430                 return 1;
1431         return ret;
1432 }
1433
1434 /*
1435  * in order to insert checksums into the metadata in large chunks,
1436  * we wait until bio submission time.   All the pages in the bio are
1437  * checksummed and sums are attached onto the ordered extent record.
1438  *
1439  * At IO completion time the cums attached on the ordered extent record
1440  * are inserted into the btree
1441  */
1442 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1443                                     struct bio *bio, int mirror_num,
1444                                     unsigned long bio_flags,
1445                                     u64 bio_offset)
1446 {
1447         struct btrfs_root *root = BTRFS_I(inode)->root;
1448         int ret = 0;
1449
1450         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1451         BUG_ON(ret);
1452         return 0;
1453 }
1454
1455 /*
1456  * in order to insert checksums into the metadata in large chunks,
1457  * we wait until bio submission time.   All the pages in the bio are
1458  * checksummed and sums are attached onto the ordered extent record.
1459  *
1460  * At IO completion time the cums attached on the ordered extent record
1461  * are inserted into the btree
1462  */
1463 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1464                           int mirror_num, unsigned long bio_flags,
1465                           u64 bio_offset)
1466 {
1467         struct btrfs_root *root = BTRFS_I(inode)->root;
1468         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1469 }
1470
1471 /*
1472  * extent_io.c submission hook. This does the right thing for csum calculation
1473  * on write, or reading the csums from the tree before a read
1474  */
1475 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1476                           int mirror_num, unsigned long bio_flags,
1477                           u64 bio_offset)
1478 {
1479         struct btrfs_root *root = BTRFS_I(inode)->root;
1480         int ret = 0;
1481         int skip_sum;
1482
1483         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1484
1485         if (btrfs_is_free_space_inode(root, inode))
1486                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1487         else
1488                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1489         BUG_ON(ret);
1490
1491         if (!(rw & REQ_WRITE)) {
1492                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1493                         return btrfs_submit_compressed_read(inode, bio,
1494                                                     mirror_num, bio_flags);
1495                 } else if (!skip_sum) {
1496                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1497                         if (ret)
1498                                 return ret;
1499                 }
1500                 goto mapit;
1501         } else if (!skip_sum) {
1502                 /* csum items have already been cloned */
1503                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1504                         goto mapit;
1505                 /* we're doing a write, do the async checksumming */
1506                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1507                                    inode, rw, bio, mirror_num,
1508                                    bio_flags, bio_offset,
1509                                    __btrfs_submit_bio_start,
1510                                    __btrfs_submit_bio_done);
1511         }
1512
1513 mapit:
1514         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1515 }
1516
1517 /*
1518  * given a list of ordered sums record them in the inode.  This happens
1519  * at IO completion time based on sums calculated at bio submission time.
1520  */
1521 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1522                              struct inode *inode, u64 file_offset,
1523                              struct list_head *list)
1524 {
1525         struct btrfs_ordered_sum *sum;
1526
1527         list_for_each_entry(sum, list, list) {
1528                 btrfs_csum_file_blocks(trans,
1529                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1530         }
1531         return 0;
1532 }
1533
1534 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1535                               struct extent_state **cached_state)
1536 {
1537         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1538                 WARN_ON(1);
1539         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1540                                    cached_state, GFP_NOFS);
1541 }
1542
1543 /* see btrfs_writepage_start_hook for details on why this is required */
1544 struct btrfs_writepage_fixup {
1545         struct page *page;
1546         struct btrfs_work work;
1547 };
1548
1549 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1550 {
1551         struct btrfs_writepage_fixup *fixup;
1552         struct btrfs_ordered_extent *ordered;
1553         struct extent_state *cached_state = NULL;
1554         struct page *page;
1555         struct inode *inode;
1556         u64 page_start;
1557         u64 page_end;
1558
1559         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1560         page = fixup->page;
1561 again:
1562         lock_page(page);
1563         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1564                 ClearPageChecked(page);
1565                 goto out_page;
1566         }
1567
1568         inode = page->mapping->host;
1569         page_start = page_offset(page);
1570         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1571
1572         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1573                          &cached_state, GFP_NOFS);
1574
1575         /* already ordered? We're done */
1576         if (PagePrivate2(page))
1577                 goto out;
1578
1579         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1580         if (ordered) {
1581                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1582                                      page_end, &cached_state, GFP_NOFS);
1583                 unlock_page(page);
1584                 btrfs_start_ordered_extent(inode, ordered, 1);
1585                 goto again;
1586         }
1587
1588         BUG();
1589         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1590         ClearPageChecked(page);
1591 out:
1592         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1593                              &cached_state, GFP_NOFS);
1594 out_page:
1595         unlock_page(page);
1596         page_cache_release(page);
1597         kfree(fixup);
1598 }
1599
1600 /*
1601  * There are a few paths in the higher layers of the kernel that directly
1602  * set the page dirty bit without asking the filesystem if it is a
1603  * good idea.  This causes problems because we want to make sure COW
1604  * properly happens and the data=ordered rules are followed.
1605  *
1606  * In our case any range that doesn't have the ORDERED bit set
1607  * hasn't been properly setup for IO.  We kick off an async process
1608  * to fix it up.  The async helper will wait for ordered extents, set
1609  * the delalloc bit and make it safe to write the page.
1610  */
1611 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1612 {
1613         struct inode *inode = page->mapping->host;
1614         struct btrfs_writepage_fixup *fixup;
1615         struct btrfs_root *root = BTRFS_I(inode)->root;
1616
1617         /* this page is properly in the ordered list */
1618         if (TestClearPagePrivate2(page))
1619                 return 0;
1620
1621         if (PageChecked(page))
1622                 return -EAGAIN;
1623
1624         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1625         if (!fixup)
1626                 return -EAGAIN;
1627
1628         SetPageChecked(page);
1629         page_cache_get(page);
1630         fixup->work.func = btrfs_writepage_fixup_worker;
1631         fixup->page = page;
1632         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1633         return -EAGAIN;
1634 }
1635
1636 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1637                                        struct inode *inode, u64 file_pos,
1638                                        u64 disk_bytenr, u64 disk_num_bytes,
1639                                        u64 num_bytes, u64 ram_bytes,
1640                                        u8 compression, u8 encryption,
1641                                        u16 other_encoding, int extent_type)
1642 {
1643         struct btrfs_root *root = BTRFS_I(inode)->root;
1644         struct btrfs_file_extent_item *fi;
1645         struct btrfs_path *path;
1646         struct extent_buffer *leaf;
1647         struct btrfs_key ins;
1648         u64 hint;
1649         int ret;
1650
1651         path = btrfs_alloc_path();
1652         if (!path)
1653                 return -ENOMEM;
1654
1655         path->leave_spinning = 1;
1656
1657         /*
1658          * we may be replacing one extent in the tree with another.
1659          * The new extent is pinned in the extent map, and we don't want
1660          * to drop it from the cache until it is completely in the btree.
1661          *
1662          * So, tell btrfs_drop_extents to leave this extent in the cache.
1663          * the caller is expected to unpin it and allow it to be merged
1664          * with the others.
1665          */
1666         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1667                                  &hint, 0);
1668         BUG_ON(ret);
1669
1670         ins.objectid = btrfs_ino(inode);
1671         ins.offset = file_pos;
1672         ins.type = BTRFS_EXTENT_DATA_KEY;
1673         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1674         BUG_ON(ret);
1675         leaf = path->nodes[0];
1676         fi = btrfs_item_ptr(leaf, path->slots[0],
1677                             struct btrfs_file_extent_item);
1678         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1679         btrfs_set_file_extent_type(leaf, fi, extent_type);
1680         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1681         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1682         btrfs_set_file_extent_offset(leaf, fi, 0);
1683         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1684         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1685         btrfs_set_file_extent_compression(leaf, fi, compression);
1686         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1687         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1688
1689         btrfs_unlock_up_safe(path, 1);
1690         btrfs_set_lock_blocking(leaf);
1691
1692         btrfs_mark_buffer_dirty(leaf);
1693
1694         inode_add_bytes(inode, num_bytes);
1695
1696         ins.objectid = disk_bytenr;
1697         ins.offset = disk_num_bytes;
1698         ins.type = BTRFS_EXTENT_ITEM_KEY;
1699         ret = btrfs_alloc_reserved_file_extent(trans, root,
1700                                         root->root_key.objectid,
1701                                         btrfs_ino(inode), file_pos, &ins);
1702         BUG_ON(ret);
1703         btrfs_free_path(path);
1704
1705         return 0;
1706 }
1707
1708 /*
1709  * helper function for btrfs_finish_ordered_io, this
1710  * just reads in some of the csum leaves to prime them into ram
1711  * before we start the transaction.  It limits the amount of btree
1712  * reads required while inside the transaction.
1713  */
1714 /* as ordered data IO finishes, this gets called so we can finish
1715  * an ordered extent if the range of bytes in the file it covers are
1716  * fully written.
1717  */
1718 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1719 {
1720         struct btrfs_root *root = BTRFS_I(inode)->root;
1721         struct btrfs_trans_handle *trans = NULL;
1722         struct btrfs_ordered_extent *ordered_extent = NULL;
1723         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1724         struct extent_state *cached_state = NULL;
1725         int compress_type = 0;
1726         int ret;
1727         bool nolock;
1728
1729         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1730                                              end - start + 1);
1731         if (!ret)
1732                 return 0;
1733         BUG_ON(!ordered_extent);
1734
1735         nolock = btrfs_is_free_space_inode(root, inode);
1736
1737         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1738                 BUG_ON(!list_empty(&ordered_extent->list));
1739                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1740                 if (!ret) {
1741                         if (nolock)
1742                                 trans = btrfs_join_transaction_nolock(root);
1743                         else
1744                                 trans = btrfs_join_transaction(root);
1745                         BUG_ON(IS_ERR(trans));
1746                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1747                         ret = btrfs_update_inode_fallback(trans, root, inode);
1748                         BUG_ON(ret);
1749                 }
1750                 goto out;
1751         }
1752
1753         lock_extent_bits(io_tree, ordered_extent->file_offset,
1754                          ordered_extent->file_offset + ordered_extent->len - 1,
1755                          0, &cached_state, GFP_NOFS);
1756
1757         if (nolock)
1758                 trans = btrfs_join_transaction_nolock(root);
1759         else
1760                 trans = btrfs_join_transaction(root);
1761         BUG_ON(IS_ERR(trans));
1762         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1763
1764         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1765                 compress_type = ordered_extent->compress_type;
1766         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1767                 BUG_ON(compress_type);
1768                 ret = btrfs_mark_extent_written(trans, inode,
1769                                                 ordered_extent->file_offset,
1770                                                 ordered_extent->file_offset +
1771                                                 ordered_extent->len);
1772                 BUG_ON(ret);
1773         } else {
1774                 BUG_ON(root == root->fs_info->tree_root);
1775                 ret = insert_reserved_file_extent(trans, inode,
1776                                                 ordered_extent->file_offset,
1777                                                 ordered_extent->start,
1778                                                 ordered_extent->disk_len,
1779                                                 ordered_extent->len,
1780                                                 ordered_extent->len,
1781                                                 compress_type, 0, 0,
1782                                                 BTRFS_FILE_EXTENT_REG);
1783                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1784                                    ordered_extent->file_offset,
1785                                    ordered_extent->len);
1786                 BUG_ON(ret);
1787         }
1788         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1789                              ordered_extent->file_offset +
1790                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1791
1792         add_pending_csums(trans, inode, ordered_extent->file_offset,
1793                           &ordered_extent->list);
1794
1795         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1796         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1797                 ret = btrfs_update_inode_fallback(trans, root, inode);
1798                 BUG_ON(ret);
1799         }
1800         ret = 0;
1801 out:
1802         if (root != root->fs_info->tree_root)
1803                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1804         if (trans) {
1805                 if (nolock)
1806                         btrfs_end_transaction_nolock(trans, root);
1807                 else
1808                         btrfs_end_transaction(trans, root);
1809         }
1810
1811         /* once for us */
1812         btrfs_put_ordered_extent(ordered_extent);
1813         /* once for the tree */
1814         btrfs_put_ordered_extent(ordered_extent);
1815
1816         return 0;
1817 }
1818
1819 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1820                                 struct extent_state *state, int uptodate)
1821 {
1822         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1823
1824         ClearPagePrivate2(page);
1825         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1826 }
1827
1828 /*
1829  * when reads are done, we need to check csums to verify the data is correct
1830  * if there's a match, we allow the bio to finish.  If not, the code in
1831  * extent_io.c will try to find good copies for us.
1832  */
1833 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1834                                struct extent_state *state)
1835 {
1836         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1837         struct inode *inode = page->mapping->host;
1838         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1839         char *kaddr;
1840         u64 private = ~(u32)0;
1841         int ret;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843         u32 csum = ~(u32)0;
1844
1845         if (PageChecked(page)) {
1846                 ClearPageChecked(page);
1847                 goto good;
1848         }
1849
1850         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1851                 goto good;
1852
1853         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1854             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1855                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1856                                   GFP_NOFS);
1857                 return 0;
1858         }
1859
1860         if (state && state->start == start) {
1861                 private = state->private;
1862                 ret = 0;
1863         } else {
1864                 ret = get_state_private(io_tree, start, &private);
1865         }
1866         kaddr = kmap_atomic(page, KM_USER0);
1867         if (ret)
1868                 goto zeroit;
1869
1870         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1871         btrfs_csum_final(csum, (char *)&csum);
1872         if (csum != private)
1873                 goto zeroit;
1874
1875         kunmap_atomic(kaddr, KM_USER0);
1876 good:
1877         return 0;
1878
1879 zeroit:
1880         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1881                        "private %llu\n",
1882                        (unsigned long long)btrfs_ino(page->mapping->host),
1883                        (unsigned long long)start, csum,
1884                        (unsigned long long)private);
1885         memset(kaddr + offset, 1, end - start + 1);
1886         flush_dcache_page(page);
1887         kunmap_atomic(kaddr, KM_USER0);
1888         if (private == 0)
1889                 return 0;
1890         return -EIO;
1891 }
1892
1893 struct delayed_iput {
1894         struct list_head list;
1895         struct inode *inode;
1896 };
1897
1898 void btrfs_add_delayed_iput(struct inode *inode)
1899 {
1900         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1901         struct delayed_iput *delayed;
1902
1903         if (atomic_add_unless(&inode->i_count, -1, 1))
1904                 return;
1905
1906         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1907         delayed->inode = inode;
1908
1909         spin_lock(&fs_info->delayed_iput_lock);
1910         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
1911         spin_unlock(&fs_info->delayed_iput_lock);
1912 }
1913
1914 void btrfs_run_delayed_iputs(struct btrfs_root *root)
1915 {
1916         LIST_HEAD(list);
1917         struct btrfs_fs_info *fs_info = root->fs_info;
1918         struct delayed_iput *delayed;
1919         int empty;
1920
1921         spin_lock(&fs_info->delayed_iput_lock);
1922         empty = list_empty(&fs_info->delayed_iputs);
1923         spin_unlock(&fs_info->delayed_iput_lock);
1924         if (empty)
1925                 return;
1926
1927         down_read(&root->fs_info->cleanup_work_sem);
1928         spin_lock(&fs_info->delayed_iput_lock);
1929         list_splice_init(&fs_info->delayed_iputs, &list);
1930         spin_unlock(&fs_info->delayed_iput_lock);
1931
1932         while (!list_empty(&list)) {
1933                 delayed = list_entry(list.next, struct delayed_iput, list);
1934                 list_del(&delayed->list);
1935                 iput(delayed->inode);
1936                 kfree(delayed);
1937         }
1938         up_read(&root->fs_info->cleanup_work_sem);
1939 }
1940
1941 enum btrfs_orphan_cleanup_state {
1942         ORPHAN_CLEANUP_STARTED  = 1,
1943         ORPHAN_CLEANUP_DONE     = 2,
1944 };
1945
1946 /*
1947  * This is called in transaction commmit time. If there are no orphan
1948  * files in the subvolume, it removes orphan item and frees block_rsv
1949  * structure.
1950  */
1951 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
1952                               struct btrfs_root *root)
1953 {
1954         int ret;
1955
1956         if (!list_empty(&root->orphan_list) ||
1957             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
1958                 return;
1959
1960         if (root->orphan_item_inserted &&
1961             btrfs_root_refs(&root->root_item) > 0) {
1962                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
1963                                             root->root_key.objectid);
1964                 BUG_ON(ret);
1965                 root->orphan_item_inserted = 0;
1966         }
1967
1968         if (root->orphan_block_rsv) {
1969                 WARN_ON(root->orphan_block_rsv->size > 0);
1970                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
1971                 root->orphan_block_rsv = NULL;
1972         }
1973 }
1974
1975 /*
1976  * This creates an orphan entry for the given inode in case something goes
1977  * wrong in the middle of an unlink/truncate.
1978  *
1979  * NOTE: caller of this function should reserve 5 units of metadata for
1980  *       this function.
1981  */
1982 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
1983 {
1984         struct btrfs_root *root = BTRFS_I(inode)->root;
1985         struct btrfs_block_rsv *block_rsv = NULL;
1986         int reserve = 0;
1987         int insert = 0;
1988         int ret;
1989
1990         if (!root->orphan_block_rsv) {
1991                 block_rsv = btrfs_alloc_block_rsv(root);
1992                 if (!block_rsv)
1993                         return -ENOMEM;
1994         }
1995
1996         spin_lock(&root->orphan_lock);
1997         if (!root->orphan_block_rsv) {
1998                 root->orphan_block_rsv = block_rsv;
1999         } else if (block_rsv) {
2000                 btrfs_free_block_rsv(root, block_rsv);
2001                 block_rsv = NULL;
2002         }
2003
2004         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2005                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2006 #if 0
2007                 /*
2008                  * For proper ENOSPC handling, we should do orphan
2009                  * cleanup when mounting. But this introduces backward
2010                  * compatibility issue.
2011                  */
2012                 if (!xchg(&root->orphan_item_inserted, 1))
2013                         insert = 2;
2014                 else
2015                         insert = 1;
2016 #endif
2017                 insert = 1;
2018         }
2019
2020         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2021                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2022                 reserve = 1;
2023         }
2024         spin_unlock(&root->orphan_lock);
2025
2026         /* grab metadata reservation from transaction handle */
2027         if (reserve) {
2028                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2029                 BUG_ON(ret);
2030         }
2031
2032         /* insert an orphan item to track this unlinked/truncated file */
2033         if (insert >= 1) {
2034                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2035                 BUG_ON(ret && ret != -EEXIST);
2036         }
2037
2038         /* insert an orphan item to track subvolume contains orphan files */
2039         if (insert >= 2) {
2040                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2041                                                root->root_key.objectid);
2042                 BUG_ON(ret);
2043         }
2044         return 0;
2045 }
2046
2047 /*
2048  * We have done the truncate/delete so we can go ahead and remove the orphan
2049  * item for this particular inode.
2050  */
2051 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2052 {
2053         struct btrfs_root *root = BTRFS_I(inode)->root;
2054         int delete_item = 0;
2055         int release_rsv = 0;
2056         int ret = 0;
2057
2058         spin_lock(&root->orphan_lock);
2059         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2060                 list_del_init(&BTRFS_I(inode)->i_orphan);
2061                 delete_item = 1;
2062         }
2063
2064         if (BTRFS_I(inode)->orphan_meta_reserved) {
2065                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2066                 release_rsv = 1;
2067         }
2068         spin_unlock(&root->orphan_lock);
2069
2070         if (trans && delete_item) {
2071                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2072                 BUG_ON(ret);
2073         }
2074
2075         if (release_rsv)
2076                 btrfs_orphan_release_metadata(inode);
2077
2078         return 0;
2079 }
2080
2081 /*
2082  * this cleans up any orphans that may be left on the list from the last use
2083  * of this root.
2084  */
2085 int btrfs_orphan_cleanup(struct btrfs_root *root)
2086 {
2087         struct btrfs_path *path;
2088         struct extent_buffer *leaf;
2089         struct btrfs_key key, found_key;
2090         struct btrfs_trans_handle *trans;
2091         struct inode *inode;
2092         u64 last_objectid = 0;
2093         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2094
2095         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2096                 return 0;
2097
2098         path = btrfs_alloc_path();
2099         if (!path) {
2100                 ret = -ENOMEM;
2101                 goto out;
2102         }
2103         path->reada = -1;
2104
2105         key.objectid = BTRFS_ORPHAN_OBJECTID;
2106         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2107         key.offset = (u64)-1;
2108
2109         while (1) {
2110                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2111                 if (ret < 0)
2112                         goto out;
2113
2114                 /*
2115                  * if ret == 0 means we found what we were searching for, which
2116                  * is weird, but possible, so only screw with path if we didn't
2117                  * find the key and see if we have stuff that matches
2118                  */
2119                 if (ret > 0) {
2120                         ret = 0;
2121                         if (path->slots[0] == 0)
2122                                 break;
2123                         path->slots[0]--;
2124                 }
2125
2126                 /* pull out the item */
2127                 leaf = path->nodes[0];
2128                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2129
2130                 /* make sure the item matches what we want */
2131                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2132                         break;
2133                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2134                         break;
2135
2136                 /* release the path since we're done with it */
2137                 btrfs_release_path(path);
2138
2139                 /*
2140                  * this is where we are basically btrfs_lookup, without the
2141                  * crossing root thing.  we store the inode number in the
2142                  * offset of the orphan item.
2143                  */
2144
2145                 if (found_key.offset == last_objectid) {
2146                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2147                                "stopping orphan cleanup\n");
2148                         ret = -EINVAL;
2149                         goto out;
2150                 }
2151
2152                 last_objectid = found_key.offset;
2153
2154                 found_key.objectid = found_key.offset;
2155                 found_key.type = BTRFS_INODE_ITEM_KEY;
2156                 found_key.offset = 0;
2157                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2158                 ret = PTR_RET(inode);
2159                 if (ret && ret != -ESTALE)
2160                         goto out;
2161
2162                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2163                         struct btrfs_root *dead_root;
2164                         struct btrfs_fs_info *fs_info = root->fs_info;
2165                         int is_dead_root = 0;
2166
2167                         /*
2168                          * this is an orphan in the tree root. Currently these
2169                          * could come from 2 sources:
2170                          *  a) a snapshot deletion in progress
2171                          *  b) a free space cache inode
2172                          * We need to distinguish those two, as the snapshot
2173                          * orphan must not get deleted.
2174                          * find_dead_roots already ran before us, so if this
2175                          * is a snapshot deletion, we should find the root
2176                          * in the dead_roots list
2177                          */
2178                         spin_lock(&fs_info->trans_lock);
2179                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2180                                             root_list) {
2181                                 if (dead_root->root_key.objectid ==
2182                                     found_key.objectid) {
2183                                         is_dead_root = 1;
2184                                         break;
2185                                 }
2186                         }
2187                         spin_unlock(&fs_info->trans_lock);
2188                         if (is_dead_root) {
2189                                 /* prevent this orphan from being found again */
2190                                 key.offset = found_key.objectid - 1;
2191                                 continue;
2192                         }
2193                 }
2194                 /*
2195                  * Inode is already gone but the orphan item is still there,
2196                  * kill the orphan item.
2197                  */
2198                 if (ret == -ESTALE) {
2199                         trans = btrfs_start_transaction(root, 1);
2200                         if (IS_ERR(trans)) {
2201                                 ret = PTR_ERR(trans);
2202                                 goto out;
2203                         }
2204                         ret = btrfs_del_orphan_item(trans, root,
2205                                                     found_key.objectid);
2206                         BUG_ON(ret);
2207                         btrfs_end_transaction(trans, root);
2208                         continue;
2209                 }
2210
2211                 /*
2212                  * add this inode to the orphan list so btrfs_orphan_del does
2213                  * the proper thing when we hit it
2214                  */
2215                 spin_lock(&root->orphan_lock);
2216                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2217                 spin_unlock(&root->orphan_lock);
2218
2219                 /* if we have links, this was a truncate, lets do that */
2220                 if (inode->i_nlink) {
2221                         if (!S_ISREG(inode->i_mode)) {
2222                                 WARN_ON(1);
2223                                 iput(inode);
2224                                 continue;
2225                         }
2226                         nr_truncate++;
2227                         /*
2228                          * Need to hold the imutex for reservation purposes, not
2229                          * a huge deal here but I have a WARN_ON in
2230                          * btrfs_delalloc_reserve_space to catch offenders.
2231                          */
2232                         mutex_lock(&inode->i_mutex);
2233                         ret = btrfs_truncate(inode);
2234                         mutex_unlock(&inode->i_mutex);
2235                 } else {
2236                         nr_unlink++;
2237                 }
2238
2239                 /* this will do delete_inode and everything for us */
2240                 iput(inode);
2241                 if (ret)
2242                         goto out;
2243         }
2244         /* release the path since we're done with it */
2245         btrfs_release_path(path);
2246
2247         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2248
2249         if (root->orphan_block_rsv)
2250                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2251                                         (u64)-1);
2252
2253         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2254                 trans = btrfs_join_transaction(root);
2255                 if (!IS_ERR(trans))
2256                         btrfs_end_transaction(trans, root);
2257         }
2258
2259         if (nr_unlink)
2260                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2261         if (nr_truncate)
2262                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2263
2264 out:
2265         if (ret)
2266                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2267         btrfs_free_path(path);
2268         return ret;
2269 }
2270
2271 /*
2272  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2273  * don't find any xattrs, we know there can't be any acls.
2274  *
2275  * slot is the slot the inode is in, objectid is the objectid of the inode
2276  */
2277 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2278                                           int slot, u64 objectid)
2279 {
2280         u32 nritems = btrfs_header_nritems(leaf);
2281         struct btrfs_key found_key;
2282         int scanned = 0;
2283
2284         slot++;
2285         while (slot < nritems) {
2286                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2287
2288                 /* we found a different objectid, there must not be acls */
2289                 if (found_key.objectid != objectid)
2290                         return 0;
2291
2292                 /* we found an xattr, assume we've got an acl */
2293                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2294                         return 1;
2295
2296                 /*
2297                  * we found a key greater than an xattr key, there can't
2298                  * be any acls later on
2299                  */
2300                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2301                         return 0;
2302
2303                 slot++;
2304                 scanned++;
2305
2306                 /*
2307                  * it goes inode, inode backrefs, xattrs, extents,
2308                  * so if there are a ton of hard links to an inode there can
2309                  * be a lot of backrefs.  Don't waste time searching too hard,
2310                  * this is just an optimization
2311                  */
2312                 if (scanned >= 8)
2313                         break;
2314         }
2315         /* we hit the end of the leaf before we found an xattr or
2316          * something larger than an xattr.  We have to assume the inode
2317          * has acls
2318          */
2319         return 1;
2320 }
2321
2322 /*
2323  * read an inode from the btree into the in-memory inode
2324  */
2325 static void btrfs_read_locked_inode(struct inode *inode)
2326 {
2327         struct btrfs_path *path;
2328         struct extent_buffer *leaf;
2329         struct btrfs_inode_item *inode_item;
2330         struct btrfs_timespec *tspec;
2331         struct btrfs_root *root = BTRFS_I(inode)->root;
2332         struct btrfs_key location;
2333         int maybe_acls;
2334         u32 rdev;
2335         int ret;
2336         bool filled = false;
2337
2338         ret = btrfs_fill_inode(inode, &rdev);
2339         if (!ret)
2340                 filled = true;
2341
2342         path = btrfs_alloc_path();
2343         if (!path)
2344                 goto make_bad;
2345
2346         path->leave_spinning = 1;
2347         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2348
2349         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2350         if (ret)
2351                 goto make_bad;
2352
2353         leaf = path->nodes[0];
2354
2355         if (filled)
2356                 goto cache_acl;
2357
2358         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2359                                     struct btrfs_inode_item);
2360         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2361         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2362         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2363         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2364         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2365
2366         tspec = btrfs_inode_atime(inode_item);
2367         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2368         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2369
2370         tspec = btrfs_inode_mtime(inode_item);
2371         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2372         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2373
2374         tspec = btrfs_inode_ctime(inode_item);
2375         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2376         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2377
2378         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2379         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2380         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2381         inode->i_generation = BTRFS_I(inode)->generation;
2382         inode->i_rdev = 0;
2383         rdev = btrfs_inode_rdev(leaf, inode_item);
2384
2385         BTRFS_I(inode)->index_cnt = (u64)-1;
2386         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2387 cache_acl:
2388         /*
2389          * try to precache a NULL acl entry for files that don't have
2390          * any xattrs or acls
2391          */
2392         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2393                                            btrfs_ino(inode));
2394         if (!maybe_acls)
2395                 cache_no_acl(inode);
2396
2397         btrfs_free_path(path);
2398
2399         switch (inode->i_mode & S_IFMT) {
2400         case S_IFREG:
2401                 inode->i_mapping->a_ops = &btrfs_aops;
2402                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2403                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2404                 inode->i_fop = &btrfs_file_operations;
2405                 inode->i_op = &btrfs_file_inode_operations;
2406                 break;
2407         case S_IFDIR:
2408                 inode->i_fop = &btrfs_dir_file_operations;
2409                 if (root == root->fs_info->tree_root)
2410                         inode->i_op = &btrfs_dir_ro_inode_operations;
2411                 else
2412                         inode->i_op = &btrfs_dir_inode_operations;
2413                 break;
2414         case S_IFLNK:
2415                 inode->i_op = &btrfs_symlink_inode_operations;
2416                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2417                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2418                 break;
2419         default:
2420                 inode->i_op = &btrfs_special_inode_operations;
2421                 init_special_inode(inode, inode->i_mode, rdev);
2422                 break;
2423         }
2424
2425         btrfs_update_iflags(inode);
2426         return;
2427
2428 make_bad:
2429         btrfs_free_path(path);
2430         make_bad_inode(inode);
2431 }
2432
2433 /*
2434  * given a leaf and an inode, copy the inode fields into the leaf
2435  */
2436 static void fill_inode_item(struct btrfs_trans_handle *trans,
2437                             struct extent_buffer *leaf,
2438                             struct btrfs_inode_item *item,
2439                             struct inode *inode)
2440 {
2441         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2442         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2443         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2444         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2445         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2446
2447         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2448                                inode->i_atime.tv_sec);
2449         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2450                                 inode->i_atime.tv_nsec);
2451
2452         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2453                                inode->i_mtime.tv_sec);
2454         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2455                                 inode->i_mtime.tv_nsec);
2456
2457         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2458                                inode->i_ctime.tv_sec);
2459         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2460                                 inode->i_ctime.tv_nsec);
2461
2462         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2463         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2464         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2465         btrfs_set_inode_transid(leaf, item, trans->transid);
2466         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2467         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2468         btrfs_set_inode_block_group(leaf, item, 0);
2469 }
2470
2471 /*
2472  * copy everything in the in-memory inode into the btree.
2473  */
2474 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2475                                 struct btrfs_root *root, struct inode *inode)
2476 {
2477         struct btrfs_inode_item *inode_item;
2478         struct btrfs_path *path;
2479         struct extent_buffer *leaf;
2480         int ret;
2481
2482         path = btrfs_alloc_path();
2483         if (!path)
2484                 return -ENOMEM;
2485
2486         path->leave_spinning = 1;
2487         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2488                                  1);
2489         if (ret) {
2490                 if (ret > 0)
2491                         ret = -ENOENT;
2492                 goto failed;
2493         }
2494
2495         btrfs_unlock_up_safe(path, 1);
2496         leaf = path->nodes[0];
2497         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2498                                     struct btrfs_inode_item);
2499
2500         fill_inode_item(trans, leaf, inode_item, inode);
2501         btrfs_mark_buffer_dirty(leaf);
2502         btrfs_set_inode_last_trans(trans, inode);
2503         ret = 0;
2504 failed:
2505         btrfs_free_path(path);
2506         return ret;
2507 }
2508
2509 /*
2510  * copy everything in the in-memory inode into the btree.
2511  */
2512 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2513                                 struct btrfs_root *root, struct inode *inode)
2514 {
2515         int ret;
2516
2517         /*
2518          * If the inode is a free space inode, we can deadlock during commit
2519          * if we put it into the delayed code.
2520          *
2521          * The data relocation inode should also be directly updated
2522          * without delay
2523          */
2524         if (!btrfs_is_free_space_inode(root, inode)
2525             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2526                 ret = btrfs_delayed_update_inode(trans, root, inode);
2527                 if (!ret)
2528                         btrfs_set_inode_last_trans(trans, inode);
2529                 return ret;
2530         }
2531
2532         return btrfs_update_inode_item(trans, root, inode);
2533 }
2534
2535 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2536                                 struct btrfs_root *root, struct inode *inode)
2537 {
2538         int ret;
2539
2540         ret = btrfs_update_inode(trans, root, inode);
2541         if (ret == -ENOSPC)
2542                 return btrfs_update_inode_item(trans, root, inode);
2543         return ret;
2544 }
2545
2546 /*
2547  * unlink helper that gets used here in inode.c and in the tree logging
2548  * recovery code.  It remove a link in a directory with a given name, and
2549  * also drops the back refs in the inode to the directory
2550  */
2551 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2552                                 struct btrfs_root *root,
2553                                 struct inode *dir, struct inode *inode,
2554                                 const char *name, int name_len)
2555 {
2556         struct btrfs_path *path;
2557         int ret = 0;
2558         struct extent_buffer *leaf;
2559         struct btrfs_dir_item *di;
2560         struct btrfs_key key;
2561         u64 index;
2562         u64 ino = btrfs_ino(inode);
2563         u64 dir_ino = btrfs_ino(dir);
2564
2565         path = btrfs_alloc_path();
2566         if (!path) {
2567                 ret = -ENOMEM;
2568                 goto out;
2569         }
2570
2571         path->leave_spinning = 1;
2572         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2573                                     name, name_len, -1);
2574         if (IS_ERR(di)) {
2575                 ret = PTR_ERR(di);
2576                 goto err;
2577         }
2578         if (!di) {
2579                 ret = -ENOENT;
2580                 goto err;
2581         }
2582         leaf = path->nodes[0];
2583         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2584         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2585         if (ret)
2586                 goto err;
2587         btrfs_release_path(path);
2588
2589         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2590                                   dir_ino, &index);
2591         if (ret) {
2592                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2593                        "inode %llu parent %llu\n", name_len, name,
2594                        (unsigned long long)ino, (unsigned long long)dir_ino);
2595                 goto err;
2596         }
2597
2598         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2599         if (ret)
2600                 goto err;
2601
2602         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2603                                          inode, dir_ino);
2604         BUG_ON(ret != 0 && ret != -ENOENT);
2605
2606         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2607                                            dir, index);
2608         if (ret == -ENOENT)
2609                 ret = 0;
2610 err:
2611         btrfs_free_path(path);
2612         if (ret)
2613                 goto out;
2614
2615         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2616         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2617         btrfs_update_inode(trans, root, dir);
2618 out:
2619         return ret;
2620 }
2621
2622 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2623                        struct btrfs_root *root,
2624                        struct inode *dir, struct inode *inode,
2625                        const char *name, int name_len)
2626 {
2627         int ret;
2628         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2629         if (!ret) {
2630                 btrfs_drop_nlink(inode);
2631                 ret = btrfs_update_inode(trans, root, inode);
2632         }
2633         return ret;
2634 }
2635
2636
2637 /* helper to check if there is any shared block in the path */
2638 static int check_path_shared(struct btrfs_root *root,
2639                              struct btrfs_path *path)
2640 {
2641         struct extent_buffer *eb;
2642         int level;
2643         u64 refs = 1;
2644
2645         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2646                 int ret;
2647
2648                 if (!path->nodes[level])
2649                         break;
2650                 eb = path->nodes[level];
2651                 if (!btrfs_block_can_be_shared(root, eb))
2652                         continue;
2653                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2654                                                &refs, NULL);
2655                 if (refs > 1)
2656                         return 1;
2657         }
2658         return 0;
2659 }
2660
2661 /*
2662  * helper to start transaction for unlink and rmdir.
2663  *
2664  * unlink and rmdir are special in btrfs, they do not always free space.
2665  * so in enospc case, we should make sure they will free space before
2666  * allowing them to use the global metadata reservation.
2667  */
2668 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2669                                                        struct dentry *dentry)
2670 {
2671         struct btrfs_trans_handle *trans;
2672         struct btrfs_root *root = BTRFS_I(dir)->root;
2673         struct btrfs_path *path;
2674         struct btrfs_inode_ref *ref;
2675         struct btrfs_dir_item *di;
2676         struct inode *inode = dentry->d_inode;
2677         u64 index;
2678         int check_link = 1;
2679         int err = -ENOSPC;
2680         int ret;
2681         u64 ino = btrfs_ino(inode);
2682         u64 dir_ino = btrfs_ino(dir);
2683
2684         /*
2685          * 1 for the possible orphan item
2686          * 1 for the dir item
2687          * 1 for the dir index
2688          * 1 for the inode ref
2689          * 1 for the inode ref in the tree log
2690          * 2 for the dir entries in the log
2691          * 1 for the inode
2692          */
2693         trans = btrfs_start_transaction(root, 8);
2694         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2695                 return trans;
2696
2697         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2698                 return ERR_PTR(-ENOSPC);
2699
2700         /* check if there is someone else holds reference */
2701         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2702                 return ERR_PTR(-ENOSPC);
2703
2704         if (atomic_read(&inode->i_count) > 2)
2705                 return ERR_PTR(-ENOSPC);
2706
2707         if (xchg(&root->fs_info->enospc_unlink, 1))
2708                 return ERR_PTR(-ENOSPC);
2709
2710         path = btrfs_alloc_path();
2711         if (!path) {
2712                 root->fs_info->enospc_unlink = 0;
2713                 return ERR_PTR(-ENOMEM);
2714         }
2715
2716         /* 1 for the orphan item */
2717         trans = btrfs_start_transaction(root, 1);
2718         if (IS_ERR(trans)) {
2719                 btrfs_free_path(path);
2720                 root->fs_info->enospc_unlink = 0;
2721                 return trans;
2722         }
2723
2724         path->skip_locking = 1;
2725         path->search_commit_root = 1;
2726
2727         ret = btrfs_lookup_inode(trans, root, path,
2728                                 &BTRFS_I(dir)->location, 0);
2729         if (ret < 0) {
2730                 err = ret;
2731                 goto out;
2732         }
2733         if (ret == 0) {
2734                 if (check_path_shared(root, path))
2735                         goto out;
2736         } else {
2737                 check_link = 0;
2738         }
2739         btrfs_release_path(path);
2740
2741         ret = btrfs_lookup_inode(trans, root, path,
2742                                 &BTRFS_I(inode)->location, 0);
2743         if (ret < 0) {
2744                 err = ret;
2745                 goto out;
2746         }
2747         if (ret == 0) {
2748                 if (check_path_shared(root, path))
2749                         goto out;
2750         } else {
2751                 check_link = 0;
2752         }
2753         btrfs_release_path(path);
2754
2755         if (ret == 0 && S_ISREG(inode->i_mode)) {
2756                 ret = btrfs_lookup_file_extent(trans, root, path,
2757                                                ino, (u64)-1, 0);
2758                 if (ret < 0) {
2759                         err = ret;
2760                         goto out;
2761                 }
2762                 BUG_ON(ret == 0);
2763                 if (check_path_shared(root, path))
2764                         goto out;
2765                 btrfs_release_path(path);
2766         }
2767
2768         if (!check_link) {
2769                 err = 0;
2770                 goto out;
2771         }
2772
2773         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2774                                 dentry->d_name.name, dentry->d_name.len, 0);
2775         if (IS_ERR(di)) {
2776                 err = PTR_ERR(di);
2777                 goto out;
2778         }
2779         if (di) {
2780                 if (check_path_shared(root, path))
2781                         goto out;
2782         } else {
2783                 err = 0;
2784                 goto out;
2785         }
2786         btrfs_release_path(path);
2787
2788         ref = btrfs_lookup_inode_ref(trans, root, path,
2789                                 dentry->d_name.name, dentry->d_name.len,
2790                                 ino, dir_ino, 0);
2791         if (IS_ERR(ref)) {
2792                 err = PTR_ERR(ref);
2793                 goto out;
2794         }
2795         BUG_ON(!ref);
2796         if (check_path_shared(root, path))
2797                 goto out;
2798         index = btrfs_inode_ref_index(path->nodes[0], ref);
2799         btrfs_release_path(path);
2800
2801         /*
2802          * This is a commit root search, if we can lookup inode item and other
2803          * relative items in the commit root, it means the transaction of
2804          * dir/file creation has been committed, and the dir index item that we
2805          * delay to insert has also been inserted into the commit root. So
2806          * we needn't worry about the delayed insertion of the dir index item
2807          * here.
2808          */
2809         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2810                                 dentry->d_name.name, dentry->d_name.len, 0);
2811         if (IS_ERR(di)) {
2812                 err = PTR_ERR(di);
2813                 goto out;
2814         }
2815         BUG_ON(ret == -ENOENT);
2816         if (check_path_shared(root, path))
2817                 goto out;
2818
2819         err = 0;
2820 out:
2821         btrfs_free_path(path);
2822         /* Migrate the orphan reservation over */
2823         if (!err)
2824                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2825                                 &root->fs_info->global_block_rsv,
2826                                 trans->bytes_reserved);
2827
2828         if (err) {
2829                 btrfs_end_transaction(trans, root);
2830                 root->fs_info->enospc_unlink = 0;
2831                 return ERR_PTR(err);
2832         }
2833
2834         trans->block_rsv = &root->fs_info->global_block_rsv;
2835         return trans;
2836 }
2837
2838 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2839                                struct btrfs_root *root)
2840 {
2841         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2842                 btrfs_block_rsv_release(root, trans->block_rsv,
2843                                         trans->bytes_reserved);
2844                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2845                 BUG_ON(!root->fs_info->enospc_unlink);
2846                 root->fs_info->enospc_unlink = 0;
2847         }
2848         btrfs_end_transaction(trans, root);
2849 }
2850
2851 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2852 {
2853         struct btrfs_root *root = BTRFS_I(dir)->root;
2854         struct btrfs_trans_handle *trans;
2855         struct inode *inode = dentry->d_inode;
2856         int ret;
2857         unsigned long nr = 0;
2858
2859         trans = __unlink_start_trans(dir, dentry);
2860         if (IS_ERR(trans))
2861                 return PTR_ERR(trans);
2862
2863         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2864
2865         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2866                                  dentry->d_name.name, dentry->d_name.len);
2867         if (ret)
2868                 goto out;
2869
2870         if (inode->i_nlink == 0) {
2871                 ret = btrfs_orphan_add(trans, inode);
2872                 if (ret)
2873                         goto out;
2874         }
2875
2876 out:
2877         nr = trans->blocks_used;
2878         __unlink_end_trans(trans, root);
2879         btrfs_btree_balance_dirty(root, nr);
2880         return ret;
2881 }
2882
2883 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2884                         struct btrfs_root *root,
2885                         struct inode *dir, u64 objectid,
2886                         const char *name, int name_len)
2887 {
2888         struct btrfs_path *path;
2889         struct extent_buffer *leaf;
2890         struct btrfs_dir_item *di;
2891         struct btrfs_key key;
2892         u64 index;
2893         int ret;
2894         u64 dir_ino = btrfs_ino(dir);
2895
2896         path = btrfs_alloc_path();
2897         if (!path)
2898                 return -ENOMEM;
2899
2900         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2901                                    name, name_len, -1);
2902         BUG_ON(IS_ERR_OR_NULL(di));
2903
2904         leaf = path->nodes[0];
2905         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2906         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2907         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2908         BUG_ON(ret);
2909         btrfs_release_path(path);
2910
2911         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2912                                  objectid, root->root_key.objectid,
2913                                  dir_ino, &index, name, name_len);
2914         if (ret < 0) {
2915                 BUG_ON(ret != -ENOENT);
2916                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2917                                                  name, name_len);
2918                 BUG_ON(IS_ERR_OR_NULL(di));
2919
2920                 leaf = path->nodes[0];
2921                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2922                 btrfs_release_path(path);
2923                 index = key.offset;
2924         }
2925         btrfs_release_path(path);
2926
2927         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2928         BUG_ON(ret);
2929
2930         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2931         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2932         ret = btrfs_update_inode(trans, root, dir);
2933         BUG_ON(ret);
2934
2935         btrfs_free_path(path);
2936         return 0;
2937 }
2938
2939 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2940 {
2941         struct inode *inode = dentry->d_inode;
2942         int err = 0;
2943         struct btrfs_root *root = BTRFS_I(dir)->root;
2944         struct btrfs_trans_handle *trans;
2945         unsigned long nr = 0;
2946
2947         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2948             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
2949                 return -ENOTEMPTY;
2950
2951         trans = __unlink_start_trans(dir, dentry);
2952         if (IS_ERR(trans))
2953                 return PTR_ERR(trans);
2954
2955         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2956                 err = btrfs_unlink_subvol(trans, root, dir,
2957                                           BTRFS_I(inode)->location.objectid,
2958                                           dentry->d_name.name,
2959                                           dentry->d_name.len);
2960                 goto out;
2961         }
2962
2963         err = btrfs_orphan_add(trans, inode);
2964         if (err)
2965                 goto out;
2966
2967         /* now the directory is empty */
2968         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2969                                  dentry->d_name.name, dentry->d_name.len);
2970         if (!err)
2971                 btrfs_i_size_write(inode, 0);
2972 out:
2973         nr = trans->blocks_used;
2974         __unlink_end_trans(trans, root);
2975         btrfs_btree_balance_dirty(root, nr);
2976
2977         return err;
2978 }
2979
2980 /*
2981  * this can truncate away extent items, csum items and directory items.
2982  * It starts at a high offset and removes keys until it can't find
2983  * any higher than new_size
2984  *
2985  * csum items that cross the new i_size are truncated to the new size
2986  * as well.
2987  *
2988  * min_type is the minimum key type to truncate down to.  If set to 0, this
2989  * will kill all the items on this inode, including the INODE_ITEM_KEY.
2990  */
2991 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
2992                                struct btrfs_root *root,
2993                                struct inode *inode,
2994                                u64 new_size, u32 min_type)
2995 {
2996         struct btrfs_path *path;
2997         struct extent_buffer *leaf;
2998         struct btrfs_file_extent_item *fi;
2999         struct btrfs_key key;
3000         struct btrfs_key found_key;
3001         u64 extent_start = 0;
3002         u64 extent_num_bytes = 0;
3003         u64 extent_offset = 0;
3004         u64 item_end = 0;
3005         u64 mask = root->sectorsize - 1;
3006         u32 found_type = (u8)-1;
3007         int found_extent;
3008         int del_item;
3009         int pending_del_nr = 0;
3010         int pending_del_slot = 0;
3011         int extent_type = -1;
3012         int ret;
3013         int err = 0;
3014         u64 ino = btrfs_ino(inode);
3015
3016         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3017
3018         path = btrfs_alloc_path();
3019         if (!path)
3020                 return -ENOMEM;
3021         path->reada = -1;
3022
3023         if (root->ref_cows || root == root->fs_info->tree_root)
3024                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3025
3026         /*
3027          * This function is also used to drop the items in the log tree before
3028          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3029          * it is used to drop the loged items. So we shouldn't kill the delayed
3030          * items.
3031          */
3032         if (min_type == 0 && root == BTRFS_I(inode)->root)
3033                 btrfs_kill_delayed_inode_items(inode);
3034
3035         key.objectid = ino;
3036         key.offset = (u64)-1;
3037         key.type = (u8)-1;
3038
3039 search_again:
3040         path->leave_spinning = 1;
3041         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3042         if (ret < 0) {
3043                 err = ret;
3044                 goto out;
3045         }
3046
3047         if (ret > 0) {
3048                 /* there are no items in the tree for us to truncate, we're
3049                  * done
3050                  */
3051                 if (path->slots[0] == 0)
3052                         goto out;
3053                 path->slots[0]--;
3054         }
3055
3056         while (1) {
3057                 fi = NULL;
3058                 leaf = path->nodes[0];
3059                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3060                 found_type = btrfs_key_type(&found_key);
3061
3062                 if (found_key.objectid != ino)
3063                         break;
3064
3065                 if (found_type < min_type)
3066                         break;
3067
3068                 item_end = found_key.offset;
3069                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3070                         fi = btrfs_item_ptr(leaf, path->slots[0],
3071                                             struct btrfs_file_extent_item);
3072                         extent_type = btrfs_file_extent_type(leaf, fi);
3073                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3074                                 item_end +=
3075                                     btrfs_file_extent_num_bytes(leaf, fi);
3076                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3077                                 item_end += btrfs_file_extent_inline_len(leaf,
3078                                                                          fi);
3079                         }
3080                         item_end--;
3081                 }
3082                 if (found_type > min_type) {
3083                         del_item = 1;
3084                 } else {
3085                         if (item_end < new_size)
3086                                 break;
3087                         if (found_key.offset >= new_size)
3088                                 del_item = 1;
3089                         else
3090                                 del_item = 0;
3091                 }
3092                 found_extent = 0;
3093                 /* FIXME, shrink the extent if the ref count is only 1 */
3094                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3095                         goto delete;
3096
3097                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3098                         u64 num_dec;
3099                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3100                         if (!del_item) {
3101                                 u64 orig_num_bytes =
3102                                         btrfs_file_extent_num_bytes(leaf, fi);
3103                                 extent_num_bytes = new_size -
3104                                         found_key.offset + root->sectorsize - 1;
3105                                 extent_num_bytes = extent_num_bytes &
3106                                         ~((u64)root->sectorsize - 1);
3107                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3108                                                          extent_num_bytes);
3109                                 num_dec = (orig_num_bytes -
3110                                            extent_num_bytes);
3111                                 if (root->ref_cows && extent_start != 0)
3112                                         inode_sub_bytes(inode, num_dec);
3113                                 btrfs_mark_buffer_dirty(leaf);
3114                         } else {
3115                                 extent_num_bytes =
3116                                         btrfs_file_extent_disk_num_bytes(leaf,
3117                                                                          fi);
3118                                 extent_offset = found_key.offset -
3119                                         btrfs_file_extent_offset(leaf, fi);
3120
3121                                 /* FIXME blocksize != 4096 */
3122                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3123                                 if (extent_start != 0) {
3124                                         found_extent = 1;
3125                                         if (root->ref_cows)
3126                                                 inode_sub_bytes(inode, num_dec);
3127                                 }
3128                         }
3129                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3130                         /*
3131                          * we can't truncate inline items that have had
3132                          * special encodings
3133                          */
3134                         if (!del_item &&
3135                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3136                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3137                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3138                                 u32 size = new_size - found_key.offset;
3139
3140                                 if (root->ref_cows) {
3141                                         inode_sub_bytes(inode, item_end + 1 -
3142                                                         new_size);
3143                                 }
3144                                 size =
3145                                     btrfs_file_extent_calc_inline_size(size);
3146                                 ret = btrfs_truncate_item(trans, root, path,
3147                                                           size, 1);
3148                         } else if (root->ref_cows) {
3149                                 inode_sub_bytes(inode, item_end + 1 -
3150                                                 found_key.offset);
3151                         }
3152                 }
3153 delete:
3154                 if (del_item) {
3155                         if (!pending_del_nr) {
3156                                 /* no pending yet, add ourselves */
3157                                 pending_del_slot = path->slots[0];
3158                                 pending_del_nr = 1;
3159                         } else if (pending_del_nr &&
3160                                    path->slots[0] + 1 == pending_del_slot) {
3161                                 /* hop on the pending chunk */
3162                                 pending_del_nr++;
3163                                 pending_del_slot = path->slots[0];
3164                         } else {
3165                                 BUG();
3166                         }
3167                 } else {
3168                         break;
3169                 }
3170                 if (found_extent && (root->ref_cows ||
3171                                      root == root->fs_info->tree_root)) {
3172                         btrfs_set_path_blocking(path);
3173                         ret = btrfs_free_extent(trans, root, extent_start,
3174                                                 extent_num_bytes, 0,
3175                                                 btrfs_header_owner(leaf),
3176                                                 ino, extent_offset, 0);
3177                         BUG_ON(ret);
3178                 }
3179
3180                 if (found_type == BTRFS_INODE_ITEM_KEY)
3181                         break;
3182
3183                 if (path->slots[0] == 0 ||
3184                     path->slots[0] != pending_del_slot) {
3185                         if (root->ref_cows &&
3186                             BTRFS_I(inode)->location.objectid !=
3187                                                 BTRFS_FREE_INO_OBJECTID) {
3188                                 err = -EAGAIN;
3189                                 goto out;
3190                         }
3191                         if (pending_del_nr) {
3192                                 ret = btrfs_del_items(trans, root, path,
3193                                                 pending_del_slot,
3194                                                 pending_del_nr);
3195                                 BUG_ON(ret);
3196                                 pending_del_nr = 0;
3197                         }
3198                         btrfs_release_path(path);
3199                         goto search_again;
3200                 } else {
3201                         path->slots[0]--;
3202                 }
3203         }
3204 out:
3205         if (pending_del_nr) {
3206                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3207                                       pending_del_nr);
3208                 BUG_ON(ret);
3209         }
3210         btrfs_free_path(path);
3211         return err;
3212 }
3213
3214 /*
3215  * taken from block_truncate_page, but does cow as it zeros out
3216  * any bytes left in the last page in the file.
3217  */
3218 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3219 {
3220         struct inode *inode = mapping->host;
3221         struct btrfs_root *root = BTRFS_I(inode)->root;
3222         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3223         struct btrfs_ordered_extent *ordered;
3224         struct extent_state *cached_state = NULL;
3225         char *kaddr;
3226         u32 blocksize = root->sectorsize;
3227         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3228         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3229         struct page *page;
3230         gfp_t mask = btrfs_alloc_write_mask(mapping);
3231         int ret = 0;
3232         u64 page_start;
3233         u64 page_end;
3234
3235         if ((offset & (blocksize - 1)) == 0)
3236                 goto out;
3237         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3238         if (ret)
3239                 goto out;
3240
3241         ret = -ENOMEM;
3242 again:
3243         page = find_or_create_page(mapping, index, mask);
3244         if (!page) {
3245                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3246                 goto out;
3247         }
3248
3249         page_start = page_offset(page);
3250         page_end = page_start + PAGE_CACHE_SIZE - 1;
3251
3252         if (!PageUptodate(page)) {
3253                 ret = btrfs_readpage(NULL, page);
3254                 lock_page(page);
3255                 if (page->mapping != mapping) {
3256                         unlock_page(page);
3257                         page_cache_release(page);
3258                         goto again;
3259                 }
3260                 if (!PageUptodate(page)) {
3261                         ret = -EIO;
3262                         goto out_unlock;
3263                 }
3264         }
3265         wait_on_page_writeback(page);
3266
3267         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3268                          GFP_NOFS);
3269         set_page_extent_mapped(page);
3270
3271         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3272         if (ordered) {
3273                 unlock_extent_cached(io_tree, page_start, page_end,
3274                                      &cached_state, GFP_NOFS);
3275                 unlock_page(page);
3276                 page_cache_release(page);
3277                 btrfs_start_ordered_extent(inode, ordered, 1);
3278                 btrfs_put_ordered_extent(ordered);
3279                 goto again;
3280         }
3281
3282         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3283                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3284                           0, 0, &cached_state, GFP_NOFS);
3285
3286         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3287                                         &cached_state);
3288         if (ret) {
3289                 unlock_extent_cached(io_tree, page_start, page_end,
3290                                      &cached_state, GFP_NOFS);
3291                 goto out_unlock;
3292         }
3293
3294         ret = 0;
3295         if (offset != PAGE_CACHE_SIZE) {
3296                 kaddr = kmap(page);
3297                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3298                 flush_dcache_page(page);
3299                 kunmap(page);
3300         }
3301         ClearPageChecked(page);
3302         set_page_dirty(page);
3303         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3304                              GFP_NOFS);
3305
3306 out_unlock:
3307         if (ret)
3308                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3309         unlock_page(page);
3310         page_cache_release(page);
3311 out:
3312         return ret;
3313 }
3314
3315 /*
3316  * This function puts in dummy file extents for the area we're creating a hole
3317  * for.  So if we are truncating this file to a larger size we need to insert
3318  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3319  * the range between oldsize and size
3320  */
3321 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3322 {
3323         struct btrfs_trans_handle *trans;
3324         struct btrfs_root *root = BTRFS_I(inode)->root;
3325         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3326         struct extent_map *em = NULL;
3327         struct extent_state *cached_state = NULL;
3328         u64 mask = root->sectorsize - 1;
3329         u64 hole_start = (oldsize + mask) & ~mask;
3330         u64 block_end = (size + mask) & ~mask;
3331         u64 last_byte;
3332         u64 cur_offset;
3333         u64 hole_size;
3334         int err = 0;
3335
3336         if (size <= hole_start)
3337                 return 0;
3338
3339         while (1) {
3340                 struct btrfs_ordered_extent *ordered;
3341                 btrfs_wait_ordered_range(inode, hole_start,
3342                                          block_end - hole_start);
3343                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3344                                  &cached_state, GFP_NOFS);
3345                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3346                 if (!ordered)
3347                         break;
3348                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3349                                      &cached_state, GFP_NOFS);
3350                 btrfs_put_ordered_extent(ordered);
3351         }
3352
3353         cur_offset = hole_start;
3354         while (1) {
3355                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3356                                 block_end - cur_offset, 0);
3357                 BUG_ON(IS_ERR_OR_NULL(em));
3358                 last_byte = min(extent_map_end(em), block_end);
3359                 last_byte = (last_byte + mask) & ~mask;
3360                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3361                         u64 hint_byte = 0;
3362                         hole_size = last_byte - cur_offset;
3363
3364                         trans = btrfs_start_transaction(root, 3);
3365                         if (IS_ERR(trans)) {
3366                                 err = PTR_ERR(trans);
3367                                 break;
3368                         }
3369
3370                         err = btrfs_drop_extents(trans, inode, cur_offset,
3371                                                  cur_offset + hole_size,
3372                                                  &hint_byte, 1);
3373                         if (err) {
3374                                 btrfs_update_inode(trans, root, inode);
3375                                 btrfs_end_transaction(trans, root);
3376                                 break;
3377                         }
3378
3379                         err = btrfs_insert_file_extent(trans, root,
3380                                         btrfs_ino(inode), cur_offset, 0,
3381                                         0, hole_size, 0, hole_size,
3382                                         0, 0, 0);
3383                         if (err) {
3384                                 btrfs_update_inode(trans, root, inode);
3385                                 btrfs_end_transaction(trans, root);
3386                                 break;
3387                         }
3388
3389                         btrfs_drop_extent_cache(inode, hole_start,
3390                                         last_byte - 1, 0);
3391
3392                         btrfs_update_inode(trans, root, inode);
3393                         btrfs_end_transaction(trans, root);
3394                 }
3395                 free_extent_map(em);
3396                 em = NULL;
3397                 cur_offset = last_byte;
3398                 if (cur_offset >= block_end)
3399                         break;
3400         }
3401
3402         free_extent_map(em);
3403         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3404                              GFP_NOFS);
3405         return err;
3406 }
3407
3408 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3409 {
3410         struct btrfs_root *root = BTRFS_I(inode)->root;
3411         struct btrfs_trans_handle *trans;
3412         loff_t oldsize = i_size_read(inode);
3413         int ret;
3414
3415         if (newsize == oldsize)
3416                 return 0;
3417
3418         if (newsize > oldsize) {
3419                 truncate_pagecache(inode, oldsize, newsize);
3420                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3421                 if (ret)
3422                         return ret;
3423
3424                 trans = btrfs_start_transaction(root, 1);
3425                 if (IS_ERR(trans))
3426                         return PTR_ERR(trans);
3427
3428                 i_size_write(inode, newsize);
3429                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3430                 ret = btrfs_update_inode(trans, root, inode);
3431                 btrfs_end_transaction(trans, root);
3432         } else {
3433
3434                 /*
3435                  * We're truncating a file that used to have good data down to
3436                  * zero. Make sure it gets into the ordered flush list so that
3437                  * any new writes get down to disk quickly.
3438                  */
3439                 if (newsize == 0)
3440                         BTRFS_I(inode)->ordered_data_close = 1;
3441
3442                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3443                 truncate_setsize(inode, newsize);
3444                 ret = btrfs_truncate(inode);
3445         }
3446
3447         return ret;
3448 }
3449
3450 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3451 {
3452         struct inode *inode = dentry->d_inode;
3453         struct btrfs_root *root = BTRFS_I(inode)->root;
3454         int err;
3455
3456         if (btrfs_root_readonly(root))
3457                 return -EROFS;
3458
3459         err = inode_change_ok(inode, attr);
3460         if (err)
3461                 return err;
3462
3463         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3464                 err = btrfs_setsize(inode, attr->ia_size);
3465                 if (err)
3466                         return err;
3467         }
3468
3469         if (attr->ia_valid) {
3470                 setattr_copy(inode, attr);
3471                 err = btrfs_dirty_inode(inode);
3472
3473                 if (!err && attr->ia_valid & ATTR_MODE)
3474                         err = btrfs_acl_chmod(inode);
3475         }
3476
3477         return err;
3478 }
3479
3480 void btrfs_evict_inode(struct inode *inode)
3481 {
3482         struct btrfs_trans_handle *trans;
3483         struct btrfs_root *root = BTRFS_I(inode)->root;
3484         struct btrfs_block_rsv *rsv, *global_rsv;
3485         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3486         unsigned long nr;
3487         int ret;
3488
3489         trace_btrfs_inode_evict(inode);
3490
3491         truncate_inode_pages(&inode->i_data, 0);
3492         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3493                                btrfs_is_free_space_inode(root, inode)))
3494                 goto no_delete;
3495
3496         if (is_bad_inode(inode)) {
3497                 btrfs_orphan_del(NULL, inode);
3498                 goto no_delete;
3499         }
3500         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3501         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3502
3503         if (root->fs_info->log_root_recovering) {
3504                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3505                 goto no_delete;
3506         }
3507
3508         if (inode->i_nlink > 0) {
3509                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3510                 goto no_delete;
3511         }
3512
3513         rsv = btrfs_alloc_block_rsv(root);
3514         if (!rsv) {
3515                 btrfs_orphan_del(NULL, inode);
3516                 goto no_delete;
3517         }
3518         rsv->size = min_size;
3519         global_rsv = &root->fs_info->global_block_rsv;
3520
3521         btrfs_i_size_write(inode, 0);
3522
3523         /*
3524          * This is a bit simpler than btrfs_truncate since
3525          *
3526          * 1) We've already reserved our space for our orphan item in the
3527          *    unlink.
3528          * 2) We're going to delete the inode item, so we don't need to update
3529          *    it at all.
3530          *
3531          * So we just need to reserve some slack space in case we add bytes when
3532          * doing the truncate.
3533          */
3534         while (1) {
3535                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3536
3537                 /*
3538                  * Try and steal from the global reserve since we will
3539                  * likely not use this space anyway, we want to try as
3540                  * hard as possible to get this to work.
3541                  */
3542                 if (ret)
3543                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3544
3545                 if (ret) {
3546                         printk(KERN_WARNING "Could not get space for a "
3547                                "delete, will truncate on mount %d\n", ret);
3548                         btrfs_orphan_del(NULL, inode);
3549                         btrfs_free_block_rsv(root, rsv);
3550                         goto no_delete;
3551                 }
3552
3553                 trans = btrfs_start_transaction(root, 0);
3554                 if (IS_ERR(trans)) {
3555                         btrfs_orphan_del(NULL, inode);
3556                         btrfs_free_block_rsv(root, rsv);
3557                         goto no_delete;
3558                 }
3559
3560                 trans->block_rsv = rsv;
3561
3562                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3563                 if (ret != -EAGAIN)
3564                         break;
3565
3566                 nr = trans->blocks_used;
3567                 btrfs_end_transaction(trans, root);
3568                 trans = NULL;
3569                 btrfs_btree_balance_dirty(root, nr);
3570         }
3571
3572         btrfs_free_block_rsv(root, rsv);
3573
3574         if (ret == 0) {
3575                 trans->block_rsv = root->orphan_block_rsv;
3576                 ret = btrfs_orphan_del(trans, inode);
3577                 BUG_ON(ret);
3578         }
3579
3580         trans->block_rsv = &root->fs_info->trans_block_rsv;
3581         if (!(root == root->fs_info->tree_root ||
3582               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3583                 btrfs_return_ino(root, btrfs_ino(inode));
3584
3585         nr = trans->blocks_used;
3586         btrfs_end_transaction(trans, root);
3587         btrfs_btree_balance_dirty(root, nr);
3588 no_delete:
3589         end_writeback(inode);
3590         return;
3591 }
3592
3593 /*
3594  * this returns the key found in the dir entry in the location pointer.
3595  * If no dir entries were found, location->objectid is 0.
3596  */
3597 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3598                                struct btrfs_key *location)
3599 {
3600         const char *name = dentry->d_name.name;
3601         int namelen = dentry->d_name.len;
3602         struct btrfs_dir_item *di;
3603         struct btrfs_path *path;
3604         struct btrfs_root *root = BTRFS_I(dir)->root;
3605         int ret = 0;
3606
3607         path = btrfs_alloc_path();
3608         if (!path)
3609                 return -ENOMEM;
3610
3611         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3612                                     namelen, 0);
3613         if (IS_ERR(di))
3614                 ret = PTR_ERR(di);
3615
3616         if (IS_ERR_OR_NULL(di))
3617                 goto out_err;
3618
3619         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3620 out:
3621         btrfs_free_path(path);
3622         return ret;
3623 out_err:
3624         location->objectid = 0;
3625         goto out;
3626 }
3627
3628 /*
3629  * when we hit a tree root in a directory, the btrfs part of the inode
3630  * needs to be changed to reflect the root directory of the tree root.  This
3631  * is kind of like crossing a mount point.
3632  */
3633 static int fixup_tree_root_location(struct btrfs_root *root,
3634                                     struct inode *dir,
3635                                     struct dentry *dentry,
3636                                     struct btrfs_key *location,
3637                                     struct btrfs_root **sub_root)
3638 {
3639         struct btrfs_path *path;
3640         struct btrfs_root *new_root;
3641         struct btrfs_root_ref *ref;
3642         struct extent_buffer *leaf;
3643         int ret;
3644         int err = 0;
3645
3646         path = btrfs_alloc_path();
3647         if (!path) {
3648                 err = -ENOMEM;
3649                 goto out;
3650         }
3651
3652         err = -ENOENT;
3653         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3654                                   BTRFS_I(dir)->root->root_key.objectid,
3655                                   location->objectid);
3656         if (ret) {
3657                 if (ret < 0)
3658                         err = ret;
3659                 goto out;
3660         }
3661
3662         leaf = path->nodes[0];
3663         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3664         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3665             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3666                 goto out;
3667
3668         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3669                                    (unsigned long)(ref + 1),
3670                                    dentry->d_name.len);
3671         if (ret)
3672                 goto out;
3673
3674         btrfs_release_path(path);
3675
3676         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3677         if (IS_ERR(new_root)) {
3678                 err = PTR_ERR(new_root);
3679                 goto out;
3680         }
3681
3682         if (btrfs_root_refs(&new_root->root_item) == 0) {
3683                 err = -ENOENT;
3684                 goto out;
3685         }
3686
3687         *sub_root = new_root;
3688         location->objectid = btrfs_root_dirid(&new_root->root_item);
3689         location->type = BTRFS_INODE_ITEM_KEY;
3690         location->offset = 0;
3691         err = 0;
3692 out:
3693         btrfs_free_path(path);
3694         return err;
3695 }
3696
3697 static void inode_tree_add(struct inode *inode)
3698 {
3699         struct btrfs_root *root = BTRFS_I(inode)->root;
3700         struct btrfs_inode *entry;
3701         struct rb_node **p;
3702         struct rb_node *parent;
3703         u64 ino = btrfs_ino(inode);
3704 again:
3705         p = &root->inode_tree.rb_node;
3706         parent = NULL;
3707
3708         if (inode_unhashed(inode))
3709                 return;
3710
3711         spin_lock(&root->inode_lock);
3712         while (*p) {
3713                 parent = *p;
3714                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3715
3716                 if (ino < btrfs_ino(&entry->vfs_inode))
3717                         p = &parent->rb_left;
3718                 else if (ino > btrfs_ino(&entry->vfs_inode))
3719                         p = &parent->rb_right;
3720                 else {
3721                         WARN_ON(!(entry->vfs_inode.i_state &
3722                                   (I_WILL_FREE | I_FREEING)));
3723                         rb_erase(parent, &root->inode_tree);
3724                         RB_CLEAR_NODE(parent);
3725                         spin_unlock(&root->inode_lock);
3726                         goto again;
3727                 }
3728         }
3729         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3730         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3731         spin_unlock(&root->inode_lock);
3732 }
3733
3734 static void inode_tree_del(struct inode *inode)
3735 {
3736         struct btrfs_root *root = BTRFS_I(inode)->root;
3737         int empty = 0;
3738
3739         spin_lock(&root->inode_lock);
3740         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3741                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3742                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3743                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3744         }
3745         spin_unlock(&root->inode_lock);
3746
3747         /*
3748          * Free space cache has inodes in the tree root, but the tree root has a
3749          * root_refs of 0, so this could end up dropping the tree root as a
3750          * snapshot, so we need the extra !root->fs_info->tree_root check to
3751          * make sure we don't drop it.
3752          */
3753         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3754             root != root->fs_info->tree_root) {
3755                 synchronize_srcu(&root->fs_info->subvol_srcu);
3756                 spin_lock(&root->inode_lock);
3757                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3758                 spin_unlock(&root->inode_lock);
3759                 if (empty)
3760                         btrfs_add_dead_root(root);
3761         }
3762 }
3763
3764 int btrfs_invalidate_inodes(struct btrfs_root *root)
3765 {
3766         struct rb_node *node;
3767         struct rb_node *prev;
3768         struct btrfs_inode *entry;
3769         struct inode *inode;
3770         u64 objectid = 0;
3771
3772         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3773
3774         spin_lock(&root->inode_lock);
3775 again:
3776         node = root->inode_tree.rb_node;
3777         prev = NULL;
3778         while (node) {
3779                 prev = node;
3780                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3781
3782                 if (objectid < btrfs_ino(&entry->vfs_inode))
3783                         node = node->rb_left;
3784                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3785                         node = node->rb_right;
3786                 else
3787                         break;
3788         }
3789         if (!node) {
3790                 while (prev) {
3791                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3792                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3793                                 node = prev;
3794                                 break;
3795                         }
3796                         prev = rb_next(prev);
3797                 }
3798         }
3799         while (node) {
3800                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3801                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3802                 inode = igrab(&entry->vfs_inode);
3803                 if (inode) {
3804                         spin_unlock(&root->inode_lock);
3805                         if (atomic_read(&inode->i_count) > 1)
3806                                 d_prune_aliases(inode);
3807                         /*
3808                          * btrfs_drop_inode will have it removed from
3809                          * the inode cache when its usage count
3810                          * hits zero.
3811                          */
3812                         iput(inode);
3813                         cond_resched();
3814                         spin_lock(&root->inode_lock);
3815                         goto again;
3816                 }
3817
3818                 if (cond_resched_lock(&root->inode_lock))
3819                         goto again;
3820
3821                 node = rb_next(node);
3822         }
3823         spin_unlock(&root->inode_lock);
3824         return 0;
3825 }
3826
3827 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3828 {
3829         struct btrfs_iget_args *args = p;
3830         inode->i_ino = args->ino;
3831         BTRFS_I(inode)->root = args->root;
3832         btrfs_set_inode_space_info(args->root, inode);
3833         return 0;
3834 }
3835
3836 static int btrfs_find_actor(struct inode *inode, void *opaque)
3837 {
3838         struct btrfs_iget_args *args = opaque;
3839         return args->ino == btrfs_ino(inode) &&
3840                 args->root == BTRFS_I(inode)->root;
3841 }
3842
3843 static struct inode *btrfs_iget_locked(struct super_block *s,
3844                                        u64 objectid,
3845                                        struct btrfs_root *root)
3846 {
3847         struct inode *inode;
3848         struct btrfs_iget_args args;
3849         args.ino = objectid;
3850         args.root = root;
3851
3852         inode = iget5_locked(s, objectid, btrfs_find_actor,
3853                              btrfs_init_locked_inode,
3854                              (void *)&args);
3855         return inode;
3856 }
3857
3858 /* Get an inode object given its location and corresponding root.
3859  * Returns in *is_new if the inode was read from disk
3860  */
3861 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3862                          struct btrfs_root *root, int *new)
3863 {
3864         struct inode *inode;
3865
3866         inode = btrfs_iget_locked(s, location->objectid, root);
3867         if (!inode)
3868                 return ERR_PTR(-ENOMEM);
3869
3870         if (inode->i_state & I_NEW) {
3871                 BTRFS_I(inode)->root = root;
3872                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3873                 btrfs_read_locked_inode(inode);
3874                 if (!is_bad_inode(inode)) {
3875                         inode_tree_add(inode);
3876                         unlock_new_inode(inode);
3877                         if (new)
3878                                 *new = 1;
3879                 } else {
3880                         unlock_new_inode(inode);
3881                         iput(inode);
3882                         inode = ERR_PTR(-ESTALE);
3883                 }
3884         }
3885
3886         return inode;
3887 }
3888
3889 static struct inode *new_simple_dir(struct super_block *s,
3890                                     struct btrfs_key *key,
3891                                     struct btrfs_root *root)
3892 {
3893         struct inode *inode = new_inode(s);
3894
3895         if (!inode)
3896                 return ERR_PTR(-ENOMEM);
3897
3898         BTRFS_I(inode)->root = root;
3899         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3900         BTRFS_I(inode)->dummy_inode = 1;
3901
3902         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3903         inode->i_op = &simple_dir_inode_operations;
3904         inode->i_fop = &simple_dir_operations;
3905         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3906         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3907
3908         return inode;
3909 }
3910
3911 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3912 {
3913         struct inode *inode;
3914         struct btrfs_root *root = BTRFS_I(dir)->root;
3915         struct btrfs_root *sub_root = root;
3916         struct btrfs_key location;
3917         int index;
3918         int ret = 0;
3919
3920         if (dentry->d_name.len > BTRFS_NAME_LEN)
3921                 return ERR_PTR(-ENAMETOOLONG);
3922
3923         if (unlikely(d_need_lookup(dentry))) {
3924                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3925                 kfree(dentry->d_fsdata);
3926                 dentry->d_fsdata = NULL;
3927                 /* This thing is hashed, drop it for now */
3928                 d_drop(dentry);
3929         } else {
3930                 ret = btrfs_inode_by_name(dir, dentry, &location);
3931         }
3932
3933         if (ret < 0)
3934                 return ERR_PTR(ret);
3935
3936         if (location.objectid == 0)
3937                 return NULL;
3938
3939         if (location.type == BTRFS_INODE_ITEM_KEY) {
3940                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3941                 return inode;
3942         }
3943
3944         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3945
3946         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3947         ret = fixup_tree_root_location(root, dir, dentry,
3948                                        &location, &sub_root);
3949         if (ret < 0) {
3950                 if (ret != -ENOENT)
3951                         inode = ERR_PTR(ret);
3952                 else
3953                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3954         } else {
3955                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3956         }
3957         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3958
3959         if (!IS_ERR(inode) && root != sub_root) {
3960                 down_read(&root->fs_info->cleanup_work_sem);
3961                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3962                         ret = btrfs_orphan_cleanup(sub_root);
3963                 up_read(&root->fs_info->cleanup_work_sem);
3964                 if (ret)
3965                         inode = ERR_PTR(ret);
3966         }
3967
3968         return inode;
3969 }
3970
3971 static int btrfs_dentry_delete(const struct dentry *dentry)
3972 {
3973         struct btrfs_root *root;
3974
3975         if (!dentry->d_inode && !IS_ROOT(dentry))
3976                 dentry = dentry->d_parent;
3977
3978         if (dentry->d_inode) {
3979                 root = BTRFS_I(dentry->d_inode)->root;
3980                 if (btrfs_root_refs(&root->root_item) == 0)
3981                         return 1;
3982         }
3983         return 0;
3984 }
3985
3986 static void btrfs_dentry_release(struct dentry *dentry)
3987 {
3988         if (dentry->d_fsdata)
3989                 kfree(dentry->d_fsdata);
3990 }
3991
3992 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3993                                    struct nameidata *nd)
3994 {
3995         struct dentry *ret;
3996
3997         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
3998         if (unlikely(d_need_lookup(dentry))) {
3999                 spin_lock(&dentry->d_lock);
4000                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4001                 spin_unlock(&dentry->d_lock);
4002         }
4003         return ret;
4004 }
4005
4006 unsigned char btrfs_filetype_table[] = {
4007         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4008 };
4009
4010 static int btrfs_real_readdir(struct file *filp, void *dirent,
4011                               filldir_t filldir)
4012 {
4013         struct inode *inode = filp->f_dentry->d_inode;
4014         struct btrfs_root *root = BTRFS_I(inode)->root;
4015         struct btrfs_item *item;
4016         struct btrfs_dir_item *di;
4017         struct btrfs_key key;
4018         struct btrfs_key found_key;
4019         struct btrfs_path *path;
4020         struct list_head ins_list;
4021         struct list_head del_list;
4022         struct qstr q;
4023         int ret;
4024         struct extent_buffer *leaf;
4025         int slot;
4026         unsigned char d_type;
4027         int over = 0;
4028         u32 di_cur;
4029         u32 di_total;
4030         u32 di_len;
4031         int key_type = BTRFS_DIR_INDEX_KEY;
4032         char tmp_name[32];
4033         char *name_ptr;
4034         int name_len;
4035         int is_curr = 0;        /* filp->f_pos points to the current index? */
4036
4037         /* FIXME, use a real flag for deciding about the key type */
4038         if (root->fs_info->tree_root == root)
4039                 key_type = BTRFS_DIR_ITEM_KEY;
4040
4041         /* special case for "." */
4042         if (filp->f_pos == 0) {
4043                 over = filldir(dirent, ".", 1,
4044                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4045                 if (over)
4046                         return 0;
4047                 filp->f_pos = 1;
4048         }
4049         /* special case for .., just use the back ref */
4050         if (filp->f_pos == 1) {
4051                 u64 pino = parent_ino(filp->f_path.dentry);
4052                 over = filldir(dirent, "..", 2,
4053                                filp->f_pos, pino, DT_DIR);
4054                 if (over)
4055                         return 0;
4056                 filp->f_pos = 2;
4057         }
4058         path = btrfs_alloc_path();
4059         if (!path)
4060                 return -ENOMEM;
4061
4062         path->reada = 1;
4063
4064         if (key_type == BTRFS_DIR_INDEX_KEY) {
4065                 INIT_LIST_HEAD(&ins_list);
4066                 INIT_LIST_HEAD(&del_list);
4067                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4068         }
4069
4070         btrfs_set_key_type(&key, key_type);
4071         key.offset = filp->f_pos;
4072         key.objectid = btrfs_ino(inode);
4073
4074         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4075         if (ret < 0)
4076                 goto err;
4077
4078         while (1) {
4079                 leaf = path->nodes[0];
4080                 slot = path->slots[0];
4081                 if (slot >= btrfs_header_nritems(leaf)) {
4082                         ret = btrfs_next_leaf(root, path);
4083                         if (ret < 0)
4084                                 goto err;
4085                         else if (ret > 0)
4086                                 break;
4087                         continue;
4088                 }
4089
4090                 item = btrfs_item_nr(leaf, slot);
4091                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4092
4093                 if (found_key.objectid != key.objectid)
4094                         break;
4095                 if (btrfs_key_type(&found_key) != key_type)
4096                         break;
4097                 if (found_key.offset < filp->f_pos)
4098                         goto next;
4099                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4100                     btrfs_should_delete_dir_index(&del_list,
4101                                                   found_key.offset))
4102                         goto next;
4103
4104                 filp->f_pos = found_key.offset;
4105                 is_curr = 1;
4106
4107                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4108                 di_cur = 0;
4109                 di_total = btrfs_item_size(leaf, item);
4110
4111                 while (di_cur < di_total) {
4112                         struct btrfs_key location;
4113                         struct dentry *tmp;
4114
4115                         if (verify_dir_item(root, leaf, di))
4116                                 break;
4117
4118                         name_len = btrfs_dir_name_len(leaf, di);
4119                         if (name_len <= sizeof(tmp_name)) {
4120                                 name_ptr = tmp_name;
4121                         } else {
4122                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4123                                 if (!name_ptr) {
4124                                         ret = -ENOMEM;
4125                                         goto err;
4126                                 }
4127                         }
4128                         read_extent_buffer(leaf, name_ptr,
4129                                            (unsigned long)(di + 1), name_len);
4130
4131                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4132                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4133
4134                         q.name = name_ptr;
4135                         q.len = name_len;
4136                         q.hash = full_name_hash(q.name, q.len);
4137                         tmp = d_lookup(filp->f_dentry, &q);
4138                         if (!tmp) {
4139                                 struct btrfs_key *newkey;
4140
4141                                 newkey = kzalloc(sizeof(struct btrfs_key),
4142                                                  GFP_NOFS);
4143                                 if (!newkey)
4144                                         goto no_dentry;
4145                                 tmp = d_alloc(filp->f_dentry, &q);
4146                                 if (!tmp) {
4147                                         kfree(newkey);
4148                                         dput(tmp);
4149                                         goto no_dentry;
4150                                 }
4151                                 memcpy(newkey, &location,
4152                                        sizeof(struct btrfs_key));
4153                                 tmp->d_fsdata = newkey;
4154                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4155                                 d_rehash(tmp);
4156                                 dput(tmp);
4157                         } else {
4158                                 dput(tmp);
4159                         }
4160 no_dentry:
4161                         /* is this a reference to our own snapshot? If so
4162                          * skip it
4163                          */
4164                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4165                             location.objectid == root->root_key.objectid) {
4166                                 over = 0;
4167                                 goto skip;
4168                         }
4169                         over = filldir(dirent, name_ptr, name_len,
4170                                        found_key.offset, location.objectid,
4171                                        d_type);
4172
4173 skip:
4174                         if (name_ptr != tmp_name)
4175                                 kfree(name_ptr);
4176
4177                         if (over)
4178                                 goto nopos;
4179                         di_len = btrfs_dir_name_len(leaf, di) +
4180                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4181                         di_cur += di_len;
4182                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4183                 }
4184 next:
4185                 path->slots[0]++;
4186         }
4187
4188         if (key_type == BTRFS_DIR_INDEX_KEY) {
4189                 if (is_curr)
4190                         filp->f_pos++;
4191                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4192                                                       &ins_list);
4193                 if (ret)
4194                         goto nopos;
4195         }
4196
4197         /* Reached end of directory/root. Bump pos past the last item. */
4198         if (key_type == BTRFS_DIR_INDEX_KEY)
4199                 /*
4200                  * 32-bit glibc will use getdents64, but then strtol -
4201                  * so the last number we can serve is this.
4202                  */
4203                 filp->f_pos = 0x7fffffff;
4204         else
4205                 filp->f_pos++;
4206 nopos:
4207         ret = 0;
4208 err:
4209         if (key_type == BTRFS_DIR_INDEX_KEY)
4210                 btrfs_put_delayed_items(&ins_list, &del_list);
4211         btrfs_free_path(path);
4212         return ret;
4213 }
4214
4215 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4216 {
4217         struct btrfs_root *root = BTRFS_I(inode)->root;
4218         struct btrfs_trans_handle *trans;
4219         int ret = 0;
4220         bool nolock = false;
4221
4222         if (BTRFS_I(inode)->dummy_inode)
4223                 return 0;
4224
4225         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4226                 nolock = true;
4227
4228         if (wbc->sync_mode == WB_SYNC_ALL) {
4229                 if (nolock)
4230                         trans = btrfs_join_transaction_nolock(root);
4231                 else
4232                         trans = btrfs_join_transaction(root);
4233                 if (IS_ERR(trans))
4234                         return PTR_ERR(trans);
4235                 if (nolock)
4236                         ret = btrfs_end_transaction_nolock(trans, root);
4237                 else
4238                         ret = btrfs_commit_transaction(trans, root);
4239         }
4240         return ret;
4241 }
4242
4243 /*
4244  * This is somewhat expensive, updating the tree every time the
4245  * inode changes.  But, it is most likely to find the inode in cache.
4246  * FIXME, needs more benchmarking...there are no reasons other than performance
4247  * to keep or drop this code.
4248  */
4249 int btrfs_dirty_inode(struct inode *inode)
4250 {
4251         struct btrfs_root *root = BTRFS_I(inode)->root;
4252         struct btrfs_trans_handle *trans;
4253         int ret;
4254
4255         if (BTRFS_I(inode)->dummy_inode)
4256                 return 0;
4257
4258         trans = btrfs_join_transaction(root);
4259         if (IS_ERR(trans))
4260                 return PTR_ERR(trans);
4261
4262         ret = btrfs_update_inode(trans, root, inode);
4263         if (ret && ret == -ENOSPC) {
4264                 /* whoops, lets try again with the full transaction */
4265                 btrfs_end_transaction(trans, root);
4266                 trans = btrfs_start_transaction(root, 1);
4267                 if (IS_ERR(trans))
4268                         return PTR_ERR(trans);
4269
4270                 ret = btrfs_update_inode(trans, root, inode);
4271         }
4272         btrfs_end_transaction(trans, root);
4273         if (BTRFS_I(inode)->delayed_node)
4274                 btrfs_balance_delayed_items(root);
4275
4276         return ret;
4277 }
4278
4279 /*
4280  * This is a copy of file_update_time.  We need this so we can return error on
4281  * ENOSPC for updating the inode in the case of file write and mmap writes.
4282  */
4283 int btrfs_update_time(struct file *file)
4284 {
4285         struct inode *inode = file->f_path.dentry->d_inode;
4286         struct timespec now;
4287         int ret;
4288         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4289
4290         /* First try to exhaust all avenues to not sync */
4291         if (IS_NOCMTIME(inode))
4292                 return 0;
4293
4294         now = current_fs_time(inode->i_sb);
4295         if (!timespec_equal(&inode->i_mtime, &now))
4296                 sync_it = S_MTIME;
4297
4298         if (!timespec_equal(&inode->i_ctime, &now))
4299                 sync_it |= S_CTIME;
4300
4301         if (IS_I_VERSION(inode))
4302                 sync_it |= S_VERSION;
4303
4304         if (!sync_it)
4305                 return 0;
4306
4307         /* Finally allowed to write? Takes lock. */
4308         if (mnt_want_write_file(file))
4309                 return 0;
4310
4311         /* Only change inode inside the lock region */
4312         if (sync_it & S_VERSION)
4313                 inode_inc_iversion(inode);
4314         if (sync_it & S_CTIME)
4315                 inode->i_ctime = now;
4316         if (sync_it & S_MTIME)
4317                 inode->i_mtime = now;
4318         ret = btrfs_dirty_inode(inode);
4319         if (!ret)
4320                 mark_inode_dirty_sync(inode);
4321         mnt_drop_write(file->f_path.mnt);
4322         return ret;
4323 }
4324
4325 /*
4326  * find the highest existing sequence number in a directory
4327  * and then set the in-memory index_cnt variable to reflect
4328  * free sequence numbers
4329  */
4330 static int btrfs_set_inode_index_count(struct inode *inode)
4331 {
4332         struct btrfs_root *root = BTRFS_I(inode)->root;
4333         struct btrfs_key key, found_key;
4334         struct btrfs_path *path;
4335         struct extent_buffer *leaf;
4336         int ret;
4337
4338         key.objectid = btrfs_ino(inode);
4339         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4340         key.offset = (u64)-1;
4341
4342         path = btrfs_alloc_path();
4343         if (!path)
4344                 return -ENOMEM;
4345
4346         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4347         if (ret < 0)
4348                 goto out;
4349         /* FIXME: we should be able to handle this */
4350         if (ret == 0)
4351                 goto out;
4352         ret = 0;
4353
4354         /*
4355          * MAGIC NUMBER EXPLANATION:
4356          * since we search a directory based on f_pos we have to start at 2
4357          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4358          * else has to start at 2
4359          */
4360         if (path->slots[0] == 0) {
4361                 BTRFS_I(inode)->index_cnt = 2;
4362                 goto out;
4363         }
4364
4365         path->slots[0]--;
4366
4367         leaf = path->nodes[0];
4368         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4369
4370         if (found_key.objectid != btrfs_ino(inode) ||
4371             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4372                 BTRFS_I(inode)->index_cnt = 2;
4373                 goto out;
4374         }
4375
4376         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4377 out:
4378         btrfs_free_path(path);
4379         return ret;
4380 }
4381
4382 /*
4383  * helper to find a free sequence number in a given directory.  This current
4384  * code is very simple, later versions will do smarter things in the btree
4385  */
4386 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4387 {
4388         int ret = 0;
4389
4390         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4391                 ret = btrfs_inode_delayed_dir_index_count(dir);
4392                 if (ret) {
4393                         ret = btrfs_set_inode_index_count(dir);
4394                         if (ret)
4395                                 return ret;
4396                 }
4397         }
4398
4399         *index = BTRFS_I(dir)->index_cnt;
4400         BTRFS_I(dir)->index_cnt++;
4401
4402         return ret;
4403 }
4404
4405 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4406                                      struct btrfs_root *root,
4407                                      struct inode *dir,
4408                                      const char *name, int name_len,
4409                                      u64 ref_objectid, u64 objectid, int mode,
4410                                      u64 *index)
4411 {
4412         struct inode *inode;
4413         struct btrfs_inode_item *inode_item;
4414         struct btrfs_key *location;
4415         struct btrfs_path *path;
4416         struct btrfs_inode_ref *ref;
4417         struct btrfs_key key[2];
4418         u32 sizes[2];
4419         unsigned long ptr;
4420         int ret;
4421         int owner;
4422
4423         path = btrfs_alloc_path();
4424         if (!path)
4425                 return ERR_PTR(-ENOMEM);
4426
4427         inode = new_inode(root->fs_info->sb);
4428         if (!inode) {
4429                 btrfs_free_path(path);
4430                 return ERR_PTR(-ENOMEM);
4431         }
4432
4433         /*
4434          * we have to initialize this early, so we can reclaim the inode
4435          * number if we fail afterwards in this function.
4436          */
4437         inode->i_ino = objectid;
4438
4439         if (dir) {
4440                 trace_btrfs_inode_request(dir);
4441
4442                 ret = btrfs_set_inode_index(dir, index);
4443                 if (ret) {
4444                         btrfs_free_path(path);
4445                         iput(inode);
4446                         return ERR_PTR(ret);
4447                 }
4448         }
4449         /*
4450          * index_cnt is ignored for everything but a dir,
4451          * btrfs_get_inode_index_count has an explanation for the magic
4452          * number
4453          */
4454         BTRFS_I(inode)->index_cnt = 2;
4455         BTRFS_I(inode)->root = root;
4456         BTRFS_I(inode)->generation = trans->transid;
4457         inode->i_generation = BTRFS_I(inode)->generation;
4458         btrfs_set_inode_space_info(root, inode);
4459
4460         if (S_ISDIR(mode))
4461                 owner = 0;
4462         else
4463                 owner = 1;
4464
4465         key[0].objectid = objectid;
4466         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4467         key[0].offset = 0;
4468
4469         key[1].objectid = objectid;
4470         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4471         key[1].offset = ref_objectid;
4472
4473         sizes[0] = sizeof(struct btrfs_inode_item);
4474         sizes[1] = name_len + sizeof(*ref);
4475
4476         path->leave_spinning = 1;
4477         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4478         if (ret != 0)
4479                 goto fail;
4480
4481         inode_init_owner(inode, dir, mode);
4482         inode_set_bytes(inode, 0);
4483         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4484         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4485                                   struct btrfs_inode_item);
4486         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4487
4488         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4489                              struct btrfs_inode_ref);
4490         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4491         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4492         ptr = (unsigned long)(ref + 1);
4493         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4494
4495         btrfs_mark_buffer_dirty(path->nodes[0]);
4496         btrfs_free_path(path);
4497
4498         location = &BTRFS_I(inode)->location;
4499         location->objectid = objectid;
4500         location->offset = 0;
4501         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4502
4503         btrfs_inherit_iflags(inode, dir);
4504
4505         if (S_ISREG(mode)) {
4506                 if (btrfs_test_opt(root, NODATASUM))
4507                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4508                 if (btrfs_test_opt(root, NODATACOW) ||
4509                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4510                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4511         }
4512
4513         insert_inode_hash(inode);
4514         inode_tree_add(inode);
4515
4516         trace_btrfs_inode_new(inode);
4517         btrfs_set_inode_last_trans(trans, inode);
4518
4519         return inode;
4520 fail:
4521         if (dir)
4522                 BTRFS_I(dir)->index_cnt--;
4523         btrfs_free_path(path);
4524         iput(inode);
4525         return ERR_PTR(ret);
4526 }
4527
4528 static inline u8 btrfs_inode_type(struct inode *inode)
4529 {
4530         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4531 }
4532
4533 /*
4534  * utility function to add 'inode' into 'parent_inode' with
4535  * a give name and a given sequence number.
4536  * if 'add_backref' is true, also insert a backref from the
4537  * inode to the parent directory.
4538  */
4539 int btrfs_add_link(struct btrfs_trans_handle *trans,
4540                    struct inode *parent_inode, struct inode *inode,
4541                    const char *name, int name_len, int add_backref, u64 index)
4542 {
4543         int ret = 0;
4544         struct btrfs_key key;
4545         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4546         u64 ino = btrfs_ino(inode);
4547         u64 parent_ino = btrfs_ino(parent_inode);
4548
4549         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4550                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4551         } else {
4552                 key.objectid = ino;
4553                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4554                 key.offset = 0;
4555         }
4556
4557         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4558                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4559                                          key.objectid, root->root_key.objectid,
4560                                          parent_ino, index, name, name_len);
4561         } else if (add_backref) {
4562                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4563                                              parent_ino, index);
4564         }
4565
4566         if (ret == 0) {
4567                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4568                                             parent_inode, &key,
4569                                             btrfs_inode_type(inode), index);
4570                 BUG_ON(ret);
4571
4572                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4573                                    name_len * 2);
4574                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4575                 ret = btrfs_update_inode(trans, root, parent_inode);
4576         }
4577         return ret;
4578 }
4579
4580 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4581                             struct inode *dir, struct dentry *dentry,
4582                             struct inode *inode, int backref, u64 index)
4583 {
4584         int err = btrfs_add_link(trans, dir, inode,
4585                                  dentry->d_name.name, dentry->d_name.len,
4586                                  backref, index);
4587         if (err > 0)
4588                 err = -EEXIST;
4589         return err;
4590 }
4591
4592 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4593                         int mode, dev_t rdev)
4594 {
4595         struct btrfs_trans_handle *trans;
4596         struct btrfs_root *root = BTRFS_I(dir)->root;
4597         struct inode *inode = NULL;
4598         int err;
4599         int drop_inode = 0;
4600         u64 objectid;
4601         unsigned long nr = 0;
4602         u64 index = 0;
4603
4604         if (!new_valid_dev(rdev))
4605                 return -EINVAL;
4606
4607         /*
4608          * 2 for inode item and ref
4609          * 2 for dir items
4610          * 1 for xattr if selinux is on
4611          */
4612         trans = btrfs_start_transaction(root, 5);
4613         if (IS_ERR(trans))
4614                 return PTR_ERR(trans);
4615
4616         err = btrfs_find_free_ino(root, &objectid);
4617         if (err)
4618                 goto out_unlock;
4619
4620         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4621                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4622                                 mode, &index);
4623         if (IS_ERR(inode)) {
4624                 err = PTR_ERR(inode);
4625                 goto out_unlock;
4626         }
4627
4628         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4629         if (err) {
4630                 drop_inode = 1;
4631                 goto out_unlock;
4632         }
4633
4634         /*
4635         * If the active LSM wants to access the inode during
4636         * d_instantiate it needs these. Smack checks to see
4637         * if the filesystem supports xattrs by looking at the
4638         * ops vector.
4639         */
4640
4641         inode->i_op = &btrfs_special_inode_operations;
4642         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4643         if (err)
4644                 drop_inode = 1;
4645         else {
4646                 init_special_inode(inode, inode->i_mode, rdev);
4647                 btrfs_update_inode(trans, root, inode);
4648                 d_instantiate(dentry, inode);
4649         }
4650 out_unlock:
4651         nr = trans->blocks_used;
4652         btrfs_end_transaction(trans, root);
4653         btrfs_btree_balance_dirty(root, nr);
4654         if (drop_inode) {
4655                 inode_dec_link_count(inode);
4656                 iput(inode);
4657         }
4658         return err;
4659 }
4660
4661 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4662                         int mode, struct nameidata *nd)
4663 {
4664         struct btrfs_trans_handle *trans;
4665         struct btrfs_root *root = BTRFS_I(dir)->root;
4666         struct inode *inode = NULL;
4667         int drop_inode = 0;
4668         int err;
4669         unsigned long nr = 0;
4670         u64 objectid;
4671         u64 index = 0;
4672
4673         /*
4674          * 2 for inode item and ref
4675          * 2 for dir items
4676          * 1 for xattr if selinux is on
4677          */
4678         trans = btrfs_start_transaction(root, 5);
4679         if (IS_ERR(trans))
4680                 return PTR_ERR(trans);
4681
4682         err = btrfs_find_free_ino(root, &objectid);
4683         if (err)
4684                 goto out_unlock;
4685
4686         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4687                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4688                                 mode, &index);
4689         if (IS_ERR(inode)) {
4690                 err = PTR_ERR(inode);
4691                 goto out_unlock;
4692         }
4693
4694         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4695         if (err) {
4696                 drop_inode = 1;
4697                 goto out_unlock;
4698         }
4699
4700         /*
4701         * If the active LSM wants to access the inode during
4702         * d_instantiate it needs these. Smack checks to see
4703         * if the filesystem supports xattrs by looking at the
4704         * ops vector.
4705         */
4706         inode->i_fop = &btrfs_file_operations;
4707         inode->i_op = &btrfs_file_inode_operations;
4708
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                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4716                 d_instantiate(dentry, inode);
4717         }
4718 out_unlock:
4719         nr = trans->blocks_used;
4720         btrfs_end_transaction(trans, root);
4721         if (drop_inode) {
4722                 inode_dec_link_count(inode);
4723                 iput(inode);
4724         }
4725         btrfs_btree_balance_dirty(root, nr);
4726         return err;
4727 }
4728
4729 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4730                       struct dentry *dentry)
4731 {
4732         struct btrfs_trans_handle *trans;
4733         struct btrfs_root *root = BTRFS_I(dir)->root;
4734         struct inode *inode = old_dentry->d_inode;
4735         u64 index;
4736         unsigned long nr = 0;
4737         int err;
4738         int drop_inode = 0;
4739
4740         /* do not allow sys_link's with other subvols of the same device */
4741         if (root->objectid != BTRFS_I(inode)->root->objectid)
4742                 return -EXDEV;
4743
4744         if (inode->i_nlink == ~0U)
4745                 return -EMLINK;
4746
4747         err = btrfs_set_inode_index(dir, &index);
4748         if (err)
4749                 goto fail;
4750
4751         /*
4752          * 2 items for inode and inode ref
4753          * 2 items for dir items
4754          * 1 item for parent inode
4755          */
4756         trans = btrfs_start_transaction(root, 5);
4757         if (IS_ERR(trans)) {
4758                 err = PTR_ERR(trans);
4759                 goto fail;
4760         }
4761
4762         btrfs_inc_nlink(inode);
4763         inode->i_ctime = CURRENT_TIME;
4764         ihold(inode);
4765
4766         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4767
4768         if (err) {
4769                 drop_inode = 1;
4770         } else {
4771                 struct dentry *parent = dentry->d_parent;
4772                 err = btrfs_update_inode(trans, root, inode);
4773                 BUG_ON(err);
4774                 d_instantiate(dentry, inode);
4775                 btrfs_log_new_name(trans, inode, NULL, parent);
4776         }
4777
4778         nr = trans->blocks_used;
4779         btrfs_end_transaction(trans, root);
4780 fail:
4781         if (drop_inode) {
4782                 inode_dec_link_count(inode);
4783                 iput(inode);
4784         }
4785         btrfs_btree_balance_dirty(root, nr);
4786         return err;
4787 }
4788
4789 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4790 {
4791         struct inode *inode = NULL;
4792         struct btrfs_trans_handle *trans;
4793         struct btrfs_root *root = BTRFS_I(dir)->root;
4794         int err = 0;
4795         int drop_on_err = 0;
4796         u64 objectid = 0;
4797         u64 index = 0;
4798         unsigned long nr = 1;
4799
4800         /*
4801          * 2 items for inode and ref
4802          * 2 items for dir items
4803          * 1 for xattr if selinux is on
4804          */
4805         trans = btrfs_start_transaction(root, 5);
4806         if (IS_ERR(trans))
4807                 return PTR_ERR(trans);
4808
4809         err = btrfs_find_free_ino(root, &objectid);
4810         if (err)
4811                 goto out_fail;
4812
4813         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4814                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4815                                 S_IFDIR | mode, &index);
4816         if (IS_ERR(inode)) {
4817                 err = PTR_ERR(inode);
4818                 goto out_fail;
4819         }
4820
4821         drop_on_err = 1;
4822
4823         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4824         if (err)
4825                 goto out_fail;
4826
4827         inode->i_op = &btrfs_dir_inode_operations;
4828         inode->i_fop = &btrfs_dir_file_operations;
4829
4830         btrfs_i_size_write(inode, 0);
4831         err = btrfs_update_inode(trans, root, inode);
4832         if (err)
4833                 goto out_fail;
4834
4835         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4836                              dentry->d_name.len, 0, index);
4837         if (err)
4838                 goto out_fail;
4839
4840         d_instantiate(dentry, inode);
4841         drop_on_err = 0;
4842
4843 out_fail:
4844         nr = trans->blocks_used;
4845         btrfs_end_transaction(trans, root);
4846         if (drop_on_err)
4847                 iput(inode);
4848         btrfs_btree_balance_dirty(root, nr);
4849         return err;
4850 }
4851
4852 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4853  * and an extent that you want to insert, deal with overlap and insert
4854  * the new extent into the tree.
4855  */
4856 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4857                                 struct extent_map *existing,
4858                                 struct extent_map *em,
4859                                 u64 map_start, u64 map_len)
4860 {
4861         u64 start_diff;
4862
4863         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4864         start_diff = map_start - em->start;
4865         em->start = map_start;
4866         em->len = map_len;
4867         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4868             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4869                 em->block_start += start_diff;
4870                 em->block_len -= start_diff;
4871         }
4872         return add_extent_mapping(em_tree, em);
4873 }
4874
4875 static noinline int uncompress_inline(struct btrfs_path *path,
4876                                       struct inode *inode, struct page *page,
4877                                       size_t pg_offset, u64 extent_offset,
4878                                       struct btrfs_file_extent_item *item)
4879 {
4880         int ret;
4881         struct extent_buffer *leaf = path->nodes[0];
4882         char *tmp;
4883         size_t max_size;
4884         unsigned long inline_size;
4885         unsigned long ptr;
4886         int compress_type;
4887
4888         WARN_ON(pg_offset != 0);
4889         compress_type = btrfs_file_extent_compression(leaf, item);
4890         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4891         inline_size = btrfs_file_extent_inline_item_len(leaf,
4892                                         btrfs_item_nr(leaf, path->slots[0]));
4893         tmp = kmalloc(inline_size, GFP_NOFS);
4894         if (!tmp)
4895                 return -ENOMEM;
4896         ptr = btrfs_file_extent_inline_start(item);
4897
4898         read_extent_buffer(leaf, tmp, ptr, inline_size);
4899
4900         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4901         ret = btrfs_decompress(compress_type, tmp, page,
4902                                extent_offset, inline_size, max_size);
4903         if (ret) {
4904                 char *kaddr = kmap_atomic(page, KM_USER0);
4905                 unsigned long copy_size = min_t(u64,
4906                                   PAGE_CACHE_SIZE - pg_offset,
4907                                   max_size - extent_offset);
4908                 memset(kaddr + pg_offset, 0, copy_size);
4909                 kunmap_atomic(kaddr, KM_USER0);
4910         }
4911         kfree(tmp);
4912         return 0;
4913 }
4914
4915 /*
4916  * a bit scary, this does extent mapping from logical file offset to the disk.
4917  * the ugly parts come from merging extents from the disk with the in-ram
4918  * representation.  This gets more complex because of the data=ordered code,
4919  * where the in-ram extents might be locked pending data=ordered completion.
4920  *
4921  * This also copies inline extents directly into the page.
4922  */
4923
4924 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4925                                     size_t pg_offset, u64 start, u64 len,
4926                                     int create)
4927 {
4928         int ret;
4929         int err = 0;
4930         u64 bytenr;
4931         u64 extent_start = 0;
4932         u64 extent_end = 0;
4933         u64 objectid = btrfs_ino(inode);
4934         u32 found_type;
4935         struct btrfs_path *path = NULL;
4936         struct btrfs_root *root = BTRFS_I(inode)->root;
4937         struct btrfs_file_extent_item *item;
4938         struct extent_buffer *leaf;
4939         struct btrfs_key found_key;
4940         struct extent_map *em = NULL;
4941         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4942         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4943         struct btrfs_trans_handle *trans = NULL;
4944         int compress_type;
4945
4946 again:
4947         read_lock(&em_tree->lock);
4948         em = lookup_extent_mapping(em_tree, start, len);
4949         if (em)
4950                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4951         read_unlock(&em_tree->lock);
4952
4953         if (em) {
4954                 if (em->start > start || em->start + em->len <= start)
4955                         free_extent_map(em);
4956                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4957                         free_extent_map(em);
4958                 else
4959                         goto out;
4960         }
4961         em = alloc_extent_map();
4962         if (!em) {
4963                 err = -ENOMEM;
4964                 goto out;
4965         }
4966         em->bdev = root->fs_info->fs_devices->latest_bdev;
4967         em->start = EXTENT_MAP_HOLE;
4968         em->orig_start = EXTENT_MAP_HOLE;
4969         em->len = (u64)-1;
4970         em->block_len = (u64)-1;
4971
4972         if (!path) {
4973                 path = btrfs_alloc_path();
4974                 if (!path) {
4975                         err = -ENOMEM;
4976                         goto out;
4977                 }
4978                 /*
4979                  * Chances are we'll be called again, so go ahead and do
4980                  * readahead
4981                  */
4982                 path->reada = 1;
4983         }
4984
4985         ret = btrfs_lookup_file_extent(trans, root, path,
4986                                        objectid, start, trans != NULL);
4987         if (ret < 0) {
4988                 err = ret;
4989                 goto out;
4990         }
4991
4992         if (ret != 0) {
4993                 if (path->slots[0] == 0)
4994                         goto not_found;
4995                 path->slots[0]--;
4996         }
4997
4998         leaf = path->nodes[0];
4999         item = btrfs_item_ptr(leaf, path->slots[0],
5000                               struct btrfs_file_extent_item);
5001         /* are we inside the extent that was found? */
5002         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5003         found_type = btrfs_key_type(&found_key);
5004         if (found_key.objectid != objectid ||
5005             found_type != BTRFS_EXTENT_DATA_KEY) {
5006                 goto not_found;
5007         }
5008
5009         found_type = btrfs_file_extent_type(leaf, item);
5010         extent_start = found_key.offset;
5011         compress_type = btrfs_file_extent_compression(leaf, item);
5012         if (found_type == BTRFS_FILE_EXTENT_REG ||
5013             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5014                 extent_end = extent_start +
5015                        btrfs_file_extent_num_bytes(leaf, item);
5016         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5017                 size_t size;
5018                 size = btrfs_file_extent_inline_len(leaf, item);
5019                 extent_end = (extent_start + size + root->sectorsize - 1) &
5020                         ~((u64)root->sectorsize - 1);
5021         }
5022
5023         if (start >= extent_end) {
5024                 path->slots[0]++;
5025                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5026                         ret = btrfs_next_leaf(root, path);
5027                         if (ret < 0) {
5028                                 err = ret;
5029                                 goto out;
5030                         }
5031                         if (ret > 0)
5032                                 goto not_found;
5033                         leaf = path->nodes[0];
5034                 }
5035                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5036                 if (found_key.objectid != objectid ||
5037                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5038                         goto not_found;
5039                 if (start + len <= found_key.offset)
5040                         goto not_found;
5041                 em->start = start;
5042                 em->len = found_key.offset - start;
5043                 goto not_found_em;
5044         }
5045
5046         if (found_type == BTRFS_FILE_EXTENT_REG ||
5047             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5048                 em->start = extent_start;
5049                 em->len = extent_end - extent_start;
5050                 em->orig_start = extent_start -
5051                                  btrfs_file_extent_offset(leaf, item);
5052                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5053                 if (bytenr == 0) {
5054                         em->block_start = EXTENT_MAP_HOLE;
5055                         goto insert;
5056                 }
5057                 if (compress_type != BTRFS_COMPRESS_NONE) {
5058                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5059                         em->compress_type = compress_type;
5060                         em->block_start = bytenr;
5061                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5062                                                                          item);
5063                 } else {
5064                         bytenr += btrfs_file_extent_offset(leaf, item);
5065                         em->block_start = bytenr;
5066                         em->block_len = em->len;
5067                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5068                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5069                 }
5070                 goto insert;
5071         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5072                 unsigned long ptr;
5073                 char *map;
5074                 size_t size;
5075                 size_t extent_offset;
5076                 size_t copy_size;
5077
5078                 em->block_start = EXTENT_MAP_INLINE;
5079                 if (!page || create) {
5080                         em->start = extent_start;
5081                         em->len = extent_end - extent_start;
5082                         goto out;
5083                 }
5084
5085                 size = btrfs_file_extent_inline_len(leaf, item);
5086                 extent_offset = page_offset(page) + pg_offset - extent_start;
5087                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5088                                 size - extent_offset);
5089                 em->start = extent_start + extent_offset;
5090                 em->len = (copy_size + root->sectorsize - 1) &
5091                         ~((u64)root->sectorsize - 1);
5092                 em->orig_start = EXTENT_MAP_INLINE;
5093                 if (compress_type) {
5094                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5095                         em->compress_type = compress_type;
5096                 }
5097                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5098                 if (create == 0 && !PageUptodate(page)) {
5099                         if (btrfs_file_extent_compression(leaf, item) !=
5100                             BTRFS_COMPRESS_NONE) {
5101                                 ret = uncompress_inline(path, inode, page,
5102                                                         pg_offset,
5103                                                         extent_offset, item);
5104                                 BUG_ON(ret);
5105                         } else {
5106                                 map = kmap(page);
5107                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5108                                                    copy_size);
5109                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5110                                         memset(map + pg_offset + copy_size, 0,
5111                                                PAGE_CACHE_SIZE - pg_offset -
5112                                                copy_size);
5113                                 }
5114                                 kunmap(page);
5115                         }
5116                         flush_dcache_page(page);
5117                 } else if (create && PageUptodate(page)) {
5118                         BUG();
5119                         if (!trans) {
5120                                 kunmap(page);
5121                                 free_extent_map(em);
5122                                 em = NULL;
5123
5124                                 btrfs_release_path(path);
5125                                 trans = btrfs_join_transaction(root);
5126
5127                                 if (IS_ERR(trans))
5128                                         return ERR_CAST(trans);
5129                                 goto again;
5130                         }
5131                         map = kmap(page);
5132                         write_extent_buffer(leaf, map + pg_offset, ptr,
5133                                             copy_size);
5134                         kunmap(page);
5135                         btrfs_mark_buffer_dirty(leaf);
5136                 }
5137                 set_extent_uptodate(io_tree, em->start,
5138                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5139                 goto insert;
5140         } else {
5141                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5142                 WARN_ON(1);
5143         }
5144 not_found:
5145         em->start = start;
5146         em->len = len;
5147 not_found_em:
5148         em->block_start = EXTENT_MAP_HOLE;
5149         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5150 insert:
5151         btrfs_release_path(path);
5152         if (em->start > start || extent_map_end(em) <= start) {
5153                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5154                        "[%llu %llu]\n", (unsigned long long)em->start,
5155                        (unsigned long long)em->len,
5156                        (unsigned long long)start,
5157                        (unsigned long long)len);
5158                 err = -EIO;
5159                 goto out;
5160         }
5161
5162         err = 0;
5163         write_lock(&em_tree->lock);
5164         ret = add_extent_mapping(em_tree, em);
5165         /* it is possible that someone inserted the extent into the tree
5166          * while we had the lock dropped.  It is also possible that
5167          * an overlapping map exists in the tree
5168          */
5169         if (ret == -EEXIST) {
5170                 struct extent_map *existing;
5171
5172                 ret = 0;
5173
5174                 existing = lookup_extent_mapping(em_tree, start, len);
5175                 if (existing && (existing->start > start ||
5176                     existing->start + existing->len <= start)) {
5177                         free_extent_map(existing);
5178                         existing = NULL;
5179                 }
5180                 if (!existing) {
5181                         existing = lookup_extent_mapping(em_tree, em->start,
5182                                                          em->len);
5183                         if (existing) {
5184                                 err = merge_extent_mapping(em_tree, existing,
5185                                                            em, start,
5186                                                            root->sectorsize);
5187                                 free_extent_map(existing);
5188                                 if (err) {
5189                                         free_extent_map(em);
5190                                         em = NULL;
5191                                 }
5192                         } else {
5193                                 err = -EIO;
5194                                 free_extent_map(em);
5195                                 em = NULL;
5196                         }
5197                 } else {
5198                         free_extent_map(em);
5199                         em = existing;
5200                         err = 0;
5201                 }
5202         }
5203         write_unlock(&em_tree->lock);
5204 out:
5205
5206         trace_btrfs_get_extent(root, em);
5207
5208         if (path)
5209                 btrfs_free_path(path);
5210         if (trans) {
5211                 ret = btrfs_end_transaction(trans, root);
5212                 if (!err)
5213                         err = ret;
5214         }
5215         if (err) {
5216                 free_extent_map(em);
5217                 return ERR_PTR(err);
5218         }
5219         return em;
5220 }
5221
5222 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5223                                            size_t pg_offset, u64 start, u64 len,
5224                                            int create)
5225 {
5226         struct extent_map *em;
5227         struct extent_map *hole_em = NULL;
5228         u64 range_start = start;
5229         u64 end;
5230         u64 found;
5231         u64 found_end;
5232         int err = 0;
5233
5234         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5235         if (IS_ERR(em))
5236                 return em;
5237         if (em) {
5238                 /*
5239                  * if our em maps to a hole, there might
5240                  * actually be delalloc bytes behind it
5241                  */
5242                 if (em->block_start != EXTENT_MAP_HOLE)
5243                         return em;
5244                 else
5245                         hole_em = em;
5246         }
5247
5248         /* check to see if we've wrapped (len == -1 or similar) */
5249         end = start + len;
5250         if (end < start)
5251                 end = (u64)-1;
5252         else
5253                 end -= 1;
5254
5255         em = NULL;
5256
5257         /* ok, we didn't find anything, lets look for delalloc */
5258         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5259                                  end, len, EXTENT_DELALLOC, 1);
5260         found_end = range_start + found;
5261         if (found_end < range_start)
5262                 found_end = (u64)-1;
5263
5264         /*
5265          * we didn't find anything useful, return
5266          * the original results from get_extent()
5267          */
5268         if (range_start > end || found_end <= start) {
5269                 em = hole_em;
5270                 hole_em = NULL;
5271                 goto out;
5272         }
5273
5274         /* adjust the range_start to make sure it doesn't
5275          * go backwards from the start they passed in
5276          */
5277         range_start = max(start,range_start);
5278         found = found_end - range_start;
5279
5280         if (found > 0) {
5281                 u64 hole_start = start;
5282                 u64 hole_len = len;
5283
5284                 em = alloc_extent_map();
5285                 if (!em) {
5286                         err = -ENOMEM;
5287                         goto out;
5288                 }
5289                 /*
5290                  * when btrfs_get_extent can't find anything it
5291                  * returns one huge hole
5292                  *
5293                  * make sure what it found really fits our range, and
5294                  * adjust to make sure it is based on the start from
5295                  * the caller
5296                  */
5297                 if (hole_em) {
5298                         u64 calc_end = extent_map_end(hole_em);
5299
5300                         if (calc_end <= start || (hole_em->start > end)) {
5301                                 free_extent_map(hole_em);
5302                                 hole_em = NULL;
5303                         } else {
5304                                 hole_start = max(hole_em->start, start);
5305                                 hole_len = calc_end - hole_start;
5306                         }
5307                 }
5308                 em->bdev = NULL;
5309                 if (hole_em && range_start > hole_start) {
5310                         /* our hole starts before our delalloc, so we
5311                          * have to return just the parts of the hole
5312                          * that go until  the delalloc starts
5313                          */
5314                         em->len = min(hole_len,
5315                                       range_start - hole_start);
5316                         em->start = hole_start;
5317                         em->orig_start = hole_start;
5318                         /*
5319                          * don't adjust block start at all,
5320                          * it is fixed at EXTENT_MAP_HOLE
5321                          */
5322                         em->block_start = hole_em->block_start;
5323                         em->block_len = hole_len;
5324                 } else {
5325                         em->start = range_start;
5326                         em->len = found;
5327                         em->orig_start = range_start;
5328                         em->block_start = EXTENT_MAP_DELALLOC;
5329                         em->block_len = found;
5330                 }
5331         } else if (hole_em) {
5332                 return hole_em;
5333         }
5334 out:
5335
5336         free_extent_map(hole_em);
5337         if (err) {
5338                 free_extent_map(em);
5339                 return ERR_PTR(err);
5340         }
5341         return em;
5342 }
5343
5344 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5345                                                   struct extent_map *em,
5346                                                   u64 start, u64 len)
5347 {
5348         struct btrfs_root *root = BTRFS_I(inode)->root;
5349         struct btrfs_trans_handle *trans;
5350         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5351         struct btrfs_key ins;
5352         u64 alloc_hint;
5353         int ret;
5354         bool insert = false;
5355
5356         /*
5357          * Ok if the extent map we looked up is a hole and is for the exact
5358          * range we want, there is no reason to allocate a new one, however if
5359          * it is not right then we need to free this one and drop the cache for
5360          * our range.
5361          */
5362         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5363             em->len != len) {
5364                 free_extent_map(em);
5365                 em = NULL;
5366                 insert = true;
5367                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5368         }
5369
5370         trans = btrfs_join_transaction(root);
5371         if (IS_ERR(trans))
5372                 return ERR_CAST(trans);
5373
5374         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5375                 btrfs_add_inode_defrag(trans, inode);
5376
5377         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5378
5379         alloc_hint = get_extent_allocation_hint(inode, start, len);
5380         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5381                                    alloc_hint, (u64)-1, &ins, 1);
5382         if (ret) {
5383                 em = ERR_PTR(ret);
5384                 goto out;
5385         }
5386
5387         if (!em) {
5388                 em = alloc_extent_map();
5389                 if (!em) {
5390                         em = ERR_PTR(-ENOMEM);
5391                         goto out;
5392                 }
5393         }
5394
5395         em->start = start;
5396         em->orig_start = em->start;
5397         em->len = ins.offset;
5398
5399         em->block_start = ins.objectid;
5400         em->block_len = ins.offset;
5401         em->bdev = root->fs_info->fs_devices->latest_bdev;
5402
5403         /*
5404          * We need to do this because if we're using the original em we searched
5405          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5406          */
5407         em->flags = 0;
5408         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5409
5410         while (insert) {
5411                 write_lock(&em_tree->lock);
5412                 ret = add_extent_mapping(em_tree, em);
5413                 write_unlock(&em_tree->lock);
5414                 if (ret != -EEXIST)
5415                         break;
5416                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5417         }
5418
5419         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5420                                            ins.offset, ins.offset, 0);
5421         if (ret) {
5422                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5423                 em = ERR_PTR(ret);
5424         }
5425 out:
5426         btrfs_end_transaction(trans, root);
5427         return em;
5428 }
5429
5430 /*
5431  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5432  * block must be cow'd
5433  */
5434 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5435                                       struct inode *inode, u64 offset, u64 len)
5436 {
5437         struct btrfs_path *path;
5438         int ret;
5439         struct extent_buffer *leaf;
5440         struct btrfs_root *root = BTRFS_I(inode)->root;
5441         struct btrfs_file_extent_item *fi;
5442         struct btrfs_key key;
5443         u64 disk_bytenr;
5444         u64 backref_offset;
5445         u64 extent_end;
5446         u64 num_bytes;
5447         int slot;
5448         int found_type;
5449
5450         path = btrfs_alloc_path();
5451         if (!path)
5452                 return -ENOMEM;
5453
5454         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5455                                        offset, 0);
5456         if (ret < 0)
5457                 goto out;
5458
5459         slot = path->slots[0];
5460         if (ret == 1) {
5461                 if (slot == 0) {
5462                         /* can't find the item, must cow */
5463                         ret = 0;
5464                         goto out;
5465                 }
5466                 slot--;
5467         }
5468         ret = 0;
5469         leaf = path->nodes[0];
5470         btrfs_item_key_to_cpu(leaf, &key, slot);
5471         if (key.objectid != btrfs_ino(inode) ||
5472             key.type != BTRFS_EXTENT_DATA_KEY) {
5473                 /* not our file or wrong item type, must cow */
5474                 goto out;
5475         }
5476
5477         if (key.offset > offset) {
5478                 /* Wrong offset, must cow */
5479                 goto out;
5480         }
5481
5482         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5483         found_type = btrfs_file_extent_type(leaf, fi);
5484         if (found_type != BTRFS_FILE_EXTENT_REG &&
5485             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5486                 /* not a regular extent, must cow */
5487                 goto out;
5488         }
5489         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5490         backref_offset = btrfs_file_extent_offset(leaf, fi);
5491
5492         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5493         if (extent_end < offset + len) {
5494                 /* extent doesn't include our full range, must cow */
5495                 goto out;
5496         }
5497
5498         if (btrfs_extent_readonly(root, disk_bytenr))
5499                 goto out;
5500
5501         /*
5502          * look for other files referencing this extent, if we
5503          * find any we must cow
5504          */
5505         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5506                                   key.offset - backref_offset, disk_bytenr))
5507                 goto out;
5508
5509         /*
5510          * adjust disk_bytenr and num_bytes to cover just the bytes
5511          * in this extent we are about to write.  If there
5512          * are any csums in that range we have to cow in order
5513          * to keep the csums correct
5514          */
5515         disk_bytenr += backref_offset;
5516         disk_bytenr += offset - key.offset;
5517         num_bytes = min(offset + len, extent_end) - offset;
5518         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5519                                 goto out;
5520         /*
5521          * all of the above have passed, it is safe to overwrite this extent
5522          * without cow
5523          */
5524         ret = 1;
5525 out:
5526         btrfs_free_path(path);
5527         return ret;
5528 }
5529
5530 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5531                                    struct buffer_head *bh_result, int create)
5532 {
5533         struct extent_map *em;
5534         struct btrfs_root *root = BTRFS_I(inode)->root;
5535         u64 start = iblock << inode->i_blkbits;
5536         u64 len = bh_result->b_size;
5537         struct btrfs_trans_handle *trans;
5538
5539         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5540         if (IS_ERR(em))
5541                 return PTR_ERR(em);
5542
5543         /*
5544          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5545          * io.  INLINE is special, and we could probably kludge it in here, but
5546          * it's still buffered so for safety lets just fall back to the generic
5547          * buffered path.
5548          *
5549          * For COMPRESSED we _have_ to read the entire extent in so we can
5550          * decompress it, so there will be buffering required no matter what we
5551          * do, so go ahead and fallback to buffered.
5552          *
5553          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5554          * to buffered IO.  Don't blame me, this is the price we pay for using
5555          * the generic code.
5556          */
5557         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5558             em->block_start == EXTENT_MAP_INLINE) {
5559                 free_extent_map(em);
5560                 return -ENOTBLK;
5561         }
5562
5563         /* Just a good old fashioned hole, return */
5564         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5565                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5566                 free_extent_map(em);
5567                 /* DIO will do one hole at a time, so just unlock a sector */
5568                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5569                               start + root->sectorsize - 1, GFP_NOFS);
5570                 return 0;
5571         }
5572
5573         /*
5574          * We don't allocate a new extent in the following cases
5575          *
5576          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5577          * existing extent.
5578          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5579          * just use the extent.
5580          *
5581          */
5582         if (!create) {
5583                 len = em->len - (start - em->start);
5584                 goto map;
5585         }
5586
5587         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5588             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5589              em->block_start != EXTENT_MAP_HOLE)) {
5590                 int type;
5591                 int ret;
5592                 u64 block_start;
5593
5594                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5595                         type = BTRFS_ORDERED_PREALLOC;
5596                 else
5597                         type = BTRFS_ORDERED_NOCOW;
5598                 len = min(len, em->len - (start - em->start));
5599                 block_start = em->block_start + (start - em->start);
5600
5601                 /*
5602                  * we're not going to log anything, but we do need
5603                  * to make sure the current transaction stays open
5604                  * while we look for nocow cross refs
5605                  */
5606                 trans = btrfs_join_transaction(root);
5607                 if (IS_ERR(trans))
5608                         goto must_cow;
5609
5610                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5611                         ret = btrfs_add_ordered_extent_dio(inode, start,
5612                                            block_start, len, len, type);
5613                         btrfs_end_transaction(trans, root);
5614                         if (ret) {
5615                                 free_extent_map(em);
5616                                 return ret;
5617                         }
5618                         goto unlock;
5619                 }
5620                 btrfs_end_transaction(trans, root);
5621         }
5622 must_cow:
5623         /*
5624          * this will cow the extent, reset the len in case we changed
5625          * it above
5626          */
5627         len = bh_result->b_size;
5628         em = btrfs_new_extent_direct(inode, em, start, len);
5629         if (IS_ERR(em))
5630                 return PTR_ERR(em);
5631         len = min(len, em->len - (start - em->start));
5632 unlock:
5633         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5634                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5635                           0, NULL, GFP_NOFS);
5636 map:
5637         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5638                 inode->i_blkbits;
5639         bh_result->b_size = len;
5640         bh_result->b_bdev = em->bdev;
5641         set_buffer_mapped(bh_result);
5642         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5643                 set_buffer_new(bh_result);
5644
5645         free_extent_map(em);
5646
5647         return 0;
5648 }
5649
5650 struct btrfs_dio_private {
5651         struct inode *inode;
5652         u64 logical_offset;
5653         u64 disk_bytenr;
5654         u64 bytes;
5655         u32 *csums;
5656         void *private;
5657
5658         /* number of bios pending for this dio */
5659         atomic_t pending_bios;
5660
5661         /* IO errors */
5662         int errors;
5663
5664         struct bio *orig_bio;
5665 };
5666
5667 static void btrfs_endio_direct_read(struct bio *bio, int err)
5668 {
5669         struct btrfs_dio_private *dip = bio->bi_private;
5670         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5671         struct bio_vec *bvec = bio->bi_io_vec;
5672         struct inode *inode = dip->inode;
5673         struct btrfs_root *root = BTRFS_I(inode)->root;
5674         u64 start;
5675         u32 *private = dip->csums;
5676
5677         start = dip->logical_offset;
5678         do {
5679                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5680                         struct page *page = bvec->bv_page;
5681                         char *kaddr;
5682                         u32 csum = ~(u32)0;
5683                         unsigned long flags;
5684
5685                         local_irq_save(flags);
5686                         kaddr = kmap_atomic(page, KM_IRQ0);
5687                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5688                                                csum, bvec->bv_len);
5689                         btrfs_csum_final(csum, (char *)&csum);
5690                         kunmap_atomic(kaddr, KM_IRQ0);
5691                         local_irq_restore(flags);
5692
5693                         flush_dcache_page(bvec->bv_page);
5694                         if (csum != *private) {
5695                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5696                                       " %llu csum %u private %u\n",
5697                                       (unsigned long long)btrfs_ino(inode),
5698                                       (unsigned long long)start,
5699                                       csum, *private);
5700                                 err = -EIO;
5701                         }
5702                 }
5703
5704                 start += bvec->bv_len;
5705                 private++;
5706                 bvec++;
5707         } while (bvec <= bvec_end);
5708
5709         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5710                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5711         bio->bi_private = dip->private;
5712
5713         kfree(dip->csums);
5714         kfree(dip);
5715
5716         /* If we had a csum failure make sure to clear the uptodate flag */
5717         if (err)
5718                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5719         dio_end_io(bio, err);
5720 }
5721
5722 static void btrfs_endio_direct_write(struct bio *bio, int err)
5723 {
5724         struct btrfs_dio_private *dip = bio->bi_private;
5725         struct inode *inode = dip->inode;
5726         struct btrfs_root *root = BTRFS_I(inode)->root;
5727         struct btrfs_trans_handle *trans;
5728         struct btrfs_ordered_extent *ordered = NULL;
5729         struct extent_state *cached_state = NULL;
5730         u64 ordered_offset = dip->logical_offset;
5731         u64 ordered_bytes = dip->bytes;
5732         int ret;
5733
5734         if (err)
5735                 goto out_done;
5736 again:
5737         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5738                                                    &ordered_offset,
5739                                                    ordered_bytes);
5740         if (!ret)
5741                 goto out_test;
5742
5743         BUG_ON(!ordered);
5744
5745         trans = btrfs_join_transaction(root);
5746         if (IS_ERR(trans)) {
5747                 err = -ENOMEM;
5748                 goto out;
5749         }
5750         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5751
5752         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5753                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5754                 if (!ret)
5755                         err = btrfs_update_inode_fallback(trans, root, inode);
5756                 goto out;
5757         }
5758
5759         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5760                          ordered->file_offset + ordered->len - 1, 0,
5761                          &cached_state, GFP_NOFS);
5762
5763         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5764                 ret = btrfs_mark_extent_written(trans, inode,
5765                                                 ordered->file_offset,
5766                                                 ordered->file_offset +
5767                                                 ordered->len);
5768                 if (ret) {
5769                         err = ret;
5770                         goto out_unlock;
5771                 }
5772         } else {
5773                 ret = insert_reserved_file_extent(trans, inode,
5774                                                   ordered->file_offset,
5775                                                   ordered->start,
5776                                                   ordered->disk_len,
5777                                                   ordered->len,
5778                                                   ordered->len,
5779                                                   0, 0, 0,
5780                                                   BTRFS_FILE_EXTENT_REG);
5781                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5782                                    ordered->file_offset, ordered->len);
5783                 if (ret) {
5784                         err = ret;
5785                         WARN_ON(1);
5786                         goto out_unlock;
5787                 }
5788         }
5789
5790         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5791         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5792         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5793                 btrfs_update_inode_fallback(trans, root, inode);
5794         ret = 0;
5795 out_unlock:
5796         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5797                              ordered->file_offset + ordered->len - 1,
5798                              &cached_state, GFP_NOFS);
5799 out:
5800         btrfs_delalloc_release_metadata(inode, ordered->len);
5801         btrfs_end_transaction(trans, root);
5802         ordered_offset = ordered->file_offset + ordered->len;
5803         btrfs_put_ordered_extent(ordered);
5804         btrfs_put_ordered_extent(ordered);
5805
5806 out_test:
5807         /*
5808          * our bio might span multiple ordered extents.  If we haven't
5809          * completed the accounting for the whole dio, go back and try again
5810          */
5811         if (ordered_offset < dip->logical_offset + dip->bytes) {
5812                 ordered_bytes = dip->logical_offset + dip->bytes -
5813                         ordered_offset;
5814                 goto again;
5815         }
5816 out_done:
5817         bio->bi_private = dip->private;
5818
5819         kfree(dip->csums);
5820         kfree(dip);
5821
5822         /* If we had an error make sure to clear the uptodate flag */
5823         if (err)
5824                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5825         dio_end_io(bio, err);
5826 }
5827
5828 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5829                                     struct bio *bio, int mirror_num,
5830                                     unsigned long bio_flags, u64 offset)
5831 {
5832         int ret;
5833         struct btrfs_root *root = BTRFS_I(inode)->root;
5834         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5835         BUG_ON(ret);
5836         return 0;
5837 }
5838
5839 static void btrfs_end_dio_bio(struct bio *bio, int err)
5840 {
5841         struct btrfs_dio_private *dip = bio->bi_private;
5842
5843         if (err) {
5844                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5845                       "sector %#Lx len %u err no %d\n",
5846                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5847                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5848                 dip->errors = 1;
5849
5850                 /*
5851                  * before atomic variable goto zero, we must make sure
5852                  * dip->errors is perceived to be set.
5853                  */
5854                 smp_mb__before_atomic_dec();
5855         }
5856
5857         /* if there are more bios still pending for this dio, just exit */
5858         if (!atomic_dec_and_test(&dip->pending_bios))
5859                 goto out;
5860
5861         if (dip->errors)
5862                 bio_io_error(dip->orig_bio);
5863         else {
5864                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5865                 bio_endio(dip->orig_bio, 0);
5866         }
5867 out:
5868         bio_put(bio);
5869 }
5870
5871 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5872                                        u64 first_sector, gfp_t gfp_flags)
5873 {
5874         int nr_vecs = bio_get_nr_vecs(bdev);
5875         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5876 }
5877
5878 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5879                                          int rw, u64 file_offset, int skip_sum,
5880                                          u32 *csums, int async_submit)
5881 {
5882         int write = rw & REQ_WRITE;
5883         struct btrfs_root *root = BTRFS_I(inode)->root;
5884         int ret;
5885
5886         bio_get(bio);
5887         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5888         if (ret)
5889                 goto err;
5890
5891         if (skip_sum)
5892                 goto map;
5893
5894         if (write && async_submit) {
5895                 ret = btrfs_wq_submit_bio(root->fs_info,
5896                                    inode, rw, bio, 0, 0,
5897                                    file_offset,
5898                                    __btrfs_submit_bio_start_direct_io,
5899                                    __btrfs_submit_bio_done);
5900                 goto err;
5901         } else if (write) {
5902                 /*
5903                  * If we aren't doing async submit, calculate the csum of the
5904                  * bio now.
5905                  */
5906                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5907                 if (ret)
5908                         goto err;
5909         } else if (!skip_sum) {
5910                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5911                                           file_offset, csums);
5912                 if (ret)
5913                         goto err;
5914         }
5915
5916 map:
5917         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5918 err:
5919         bio_put(bio);
5920         return ret;
5921 }
5922
5923 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5924                                     int skip_sum)
5925 {
5926         struct inode *inode = dip->inode;
5927         struct btrfs_root *root = BTRFS_I(inode)->root;
5928         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5929         struct bio *bio;
5930         struct bio *orig_bio = dip->orig_bio;
5931         struct bio_vec *bvec = orig_bio->bi_io_vec;
5932         u64 start_sector = orig_bio->bi_sector;
5933         u64 file_offset = dip->logical_offset;
5934         u64 submit_len = 0;
5935         u64 map_length;
5936         int nr_pages = 0;
5937         u32 *csums = dip->csums;
5938         int ret = 0;
5939         int async_submit = 0;
5940         int write = rw & REQ_WRITE;
5941
5942         map_length = orig_bio->bi_size;
5943         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5944                               &map_length, NULL, 0);
5945         if (ret) {
5946                 bio_put(orig_bio);
5947                 return -EIO;
5948         }
5949
5950         if (map_length >= orig_bio->bi_size) {
5951                 bio = orig_bio;
5952                 goto submit;
5953         }
5954
5955         async_submit = 1;
5956         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5957         if (!bio)
5958                 return -ENOMEM;
5959         bio->bi_private = dip;
5960         bio->bi_end_io = btrfs_end_dio_bio;
5961         atomic_inc(&dip->pending_bios);
5962
5963         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5964                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5965                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5966                                  bvec->bv_offset) < bvec->bv_len)) {
5967                         /*
5968                          * inc the count before we submit the bio so
5969                          * we know the end IO handler won't happen before
5970                          * we inc the count. Otherwise, the dip might get freed
5971                          * before we're done setting it up
5972                          */
5973                         atomic_inc(&dip->pending_bios);
5974                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5975                                                      file_offset, skip_sum,
5976                                                      csums, async_submit);
5977                         if (ret) {
5978                                 bio_put(bio);
5979                                 atomic_dec(&dip->pending_bios);
5980                                 goto out_err;
5981                         }
5982
5983                         /* Write's use the ordered csums */
5984                         if (!write && !skip_sum)
5985                                 csums = csums + nr_pages;
5986                         start_sector += submit_len >> 9;
5987                         file_offset += submit_len;
5988
5989                         submit_len = 0;
5990                         nr_pages = 0;
5991
5992                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5993                                                   start_sector, GFP_NOFS);
5994                         if (!bio)
5995                                 goto out_err;
5996                         bio->bi_private = dip;
5997                         bio->bi_end_io = btrfs_end_dio_bio;
5998
5999                         map_length = orig_bio->bi_size;
6000                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6001                                               &map_length, NULL, 0);
6002                         if (ret) {
6003                                 bio_put(bio);
6004                                 goto out_err;
6005                         }
6006                 } else {
6007                         submit_len += bvec->bv_len;
6008                         nr_pages ++;
6009                         bvec++;
6010                 }
6011         }
6012
6013 submit:
6014         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6015                                      csums, async_submit);
6016         if (!ret)
6017                 return 0;
6018
6019         bio_put(bio);
6020 out_err:
6021         dip->errors = 1;
6022         /*
6023          * before atomic variable goto zero, we must
6024          * make sure dip->errors is perceived to be set.
6025          */
6026         smp_mb__before_atomic_dec();
6027         if (atomic_dec_and_test(&dip->pending_bios))
6028                 bio_io_error(dip->orig_bio);
6029
6030         /* bio_end_io() will handle error, so we needn't return it */
6031         return 0;
6032 }
6033
6034 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6035                                 loff_t file_offset)
6036 {
6037         struct btrfs_root *root = BTRFS_I(inode)->root;
6038         struct btrfs_dio_private *dip;
6039         struct bio_vec *bvec = bio->bi_io_vec;
6040         int skip_sum;
6041         int write = rw & REQ_WRITE;
6042         int ret = 0;
6043
6044         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6045
6046         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6047         if (!dip) {
6048                 ret = -ENOMEM;
6049                 goto free_ordered;
6050         }
6051         dip->csums = NULL;
6052
6053         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6054         if (!write && !skip_sum) {
6055                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6056                 if (!dip->csums) {
6057                         kfree(dip);
6058                         ret = -ENOMEM;
6059                         goto free_ordered;
6060                 }
6061         }
6062
6063         dip->private = bio->bi_private;
6064         dip->inode = inode;
6065         dip->logical_offset = file_offset;
6066
6067         dip->bytes = 0;
6068         do {
6069                 dip->bytes += bvec->bv_len;
6070                 bvec++;
6071         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6072
6073         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6074         bio->bi_private = dip;
6075         dip->errors = 0;
6076         dip->orig_bio = bio;
6077         atomic_set(&dip->pending_bios, 0);
6078
6079         if (write)
6080                 bio->bi_end_io = btrfs_endio_direct_write;
6081         else
6082                 bio->bi_end_io = btrfs_endio_direct_read;
6083
6084         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6085         if (!ret)
6086                 return;
6087 free_ordered:
6088         /*
6089          * If this is a write, we need to clean up the reserved space and kill
6090          * the ordered extent.
6091          */
6092         if (write) {
6093                 struct btrfs_ordered_extent *ordered;
6094                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6095                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6096                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6097                         btrfs_free_reserved_extent(root, ordered->start,
6098                                                    ordered->disk_len);
6099                 btrfs_put_ordered_extent(ordered);
6100                 btrfs_put_ordered_extent(ordered);
6101         }
6102         bio_endio(bio, ret);
6103 }
6104
6105 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6106                         const struct iovec *iov, loff_t offset,
6107                         unsigned long nr_segs)
6108 {
6109         int seg;
6110         int i;
6111         size_t size;
6112         unsigned long addr;
6113         unsigned blocksize_mask = root->sectorsize - 1;
6114         ssize_t retval = -EINVAL;
6115         loff_t end = offset;
6116
6117         if (offset & blocksize_mask)
6118                 goto out;
6119
6120         /* Check the memory alignment.  Blocks cannot straddle pages */
6121         for (seg = 0; seg < nr_segs; seg++) {
6122                 addr = (unsigned long)iov[seg].iov_base;
6123                 size = iov[seg].iov_len;
6124                 end += size;
6125                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6126                         goto out;
6127
6128                 /* If this is a write we don't need to check anymore */
6129                 if (rw & WRITE)
6130                         continue;
6131
6132                 /*
6133                  * Check to make sure we don't have duplicate iov_base's in this
6134                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6135                  * when reading back.
6136                  */
6137                 for (i = seg + 1; i < nr_segs; i++) {
6138                         if (iov[seg].iov_base == iov[i].iov_base)
6139                                 goto out;
6140                 }
6141         }
6142         retval = 0;
6143 out:
6144         return retval;
6145 }
6146 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6147                         const struct iovec *iov, loff_t offset,
6148                         unsigned long nr_segs)
6149 {
6150         struct file *file = iocb->ki_filp;
6151         struct inode *inode = file->f_mapping->host;
6152         struct btrfs_ordered_extent *ordered;
6153         struct extent_state *cached_state = NULL;
6154         u64 lockstart, lockend;
6155         ssize_t ret;
6156         int writing = rw & WRITE;
6157         int write_bits = 0;
6158         size_t count = iov_length(iov, nr_segs);
6159
6160         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6161                             offset, nr_segs)) {
6162                 return 0;
6163         }
6164
6165         lockstart = offset;
6166         lockend = offset + count - 1;
6167
6168         if (writing) {
6169                 ret = btrfs_delalloc_reserve_space(inode, count);
6170                 if (ret)
6171                         goto out;
6172         }
6173
6174         while (1) {
6175                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6176                                  0, &cached_state, GFP_NOFS);
6177                 /*
6178                  * We're concerned with the entire range that we're going to be
6179                  * doing DIO to, so we need to make sure theres no ordered
6180                  * extents in this range.
6181                  */
6182                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6183                                                      lockend - lockstart + 1);
6184                 if (!ordered)
6185                         break;
6186                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6187                                      &cached_state, GFP_NOFS);
6188                 btrfs_start_ordered_extent(inode, ordered, 1);
6189                 btrfs_put_ordered_extent(ordered);
6190                 cond_resched();
6191         }
6192
6193         /*
6194          * we don't use btrfs_set_extent_delalloc because we don't want
6195          * the dirty or uptodate bits
6196          */
6197         if (writing) {
6198                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6199                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6200                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6201                                      GFP_NOFS);
6202                 if (ret) {
6203                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6204                                          lockend, EXTENT_LOCKED | write_bits,
6205                                          1, 0, &cached_state, GFP_NOFS);
6206                         goto out;
6207                 }
6208         }
6209
6210         free_extent_state(cached_state);
6211         cached_state = NULL;
6212
6213         ret = __blockdev_direct_IO(rw, iocb, inode,
6214                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6215                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6216                    btrfs_submit_direct, 0);
6217
6218         if (ret < 0 && ret != -EIOCBQUEUED) {
6219                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6220                               offset + iov_length(iov, nr_segs) - 1,
6221                               EXTENT_LOCKED | write_bits, 1, 0,
6222                               &cached_state, GFP_NOFS);
6223         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6224                 /*
6225                  * We're falling back to buffered, unlock the section we didn't
6226                  * do IO on.
6227                  */
6228                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6229                               offset + iov_length(iov, nr_segs) - 1,
6230                               EXTENT_LOCKED | write_bits, 1, 0,
6231                               &cached_state, GFP_NOFS);
6232         }
6233 out:
6234         free_extent_state(cached_state);
6235         return ret;
6236 }
6237
6238 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6239                 __u64 start, __u64 len)
6240 {
6241         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6242 }
6243
6244 int btrfs_readpage(struct file *file, struct page *page)
6245 {
6246         struct extent_io_tree *tree;
6247         tree = &BTRFS_I(page->mapping->host)->io_tree;
6248         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6249 }
6250
6251 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6252 {
6253         struct extent_io_tree *tree;
6254
6255
6256         if (current->flags & PF_MEMALLOC) {
6257                 redirty_page_for_writepage(wbc, page);
6258                 unlock_page(page);
6259                 return 0;
6260         }
6261         tree = &BTRFS_I(page->mapping->host)->io_tree;
6262         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6263 }
6264
6265 int btrfs_writepages(struct address_space *mapping,
6266                      struct writeback_control *wbc)
6267 {
6268         struct extent_io_tree *tree;
6269
6270         tree = &BTRFS_I(mapping->host)->io_tree;
6271         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6272 }
6273
6274 static int
6275 btrfs_readpages(struct file *file, struct address_space *mapping,
6276                 struct list_head *pages, unsigned nr_pages)
6277 {
6278         struct extent_io_tree *tree;
6279         tree = &BTRFS_I(mapping->host)->io_tree;
6280         return extent_readpages(tree, mapping, pages, nr_pages,
6281                                 btrfs_get_extent);
6282 }
6283 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6284 {
6285         struct extent_io_tree *tree;
6286         struct extent_map_tree *map;
6287         int ret;
6288
6289         tree = &BTRFS_I(page->mapping->host)->io_tree;
6290         map = &BTRFS_I(page->mapping->host)->extent_tree;
6291         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6292         if (ret == 1) {
6293                 ClearPagePrivate(page);
6294                 set_page_private(page, 0);
6295                 page_cache_release(page);
6296         }
6297         return ret;
6298 }
6299
6300 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6301 {
6302         if (PageWriteback(page) || PageDirty(page))
6303                 return 0;
6304         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6305 }
6306
6307 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6308 {
6309         struct extent_io_tree *tree;
6310         struct btrfs_ordered_extent *ordered;
6311         struct extent_state *cached_state = NULL;
6312         u64 page_start = page_offset(page);
6313         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6314
6315
6316         /*
6317          * we have the page locked, so new writeback can't start,
6318          * and the dirty bit won't be cleared while we are here.
6319          *
6320          * Wait for IO on this page so that we can safely clear
6321          * the PagePrivate2 bit and do ordered accounting
6322          */
6323         wait_on_page_writeback(page);
6324
6325         tree = &BTRFS_I(page->mapping->host)->io_tree;
6326         if (offset) {
6327                 btrfs_releasepage(page, GFP_NOFS);
6328                 return;
6329         }
6330         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6331                          GFP_NOFS);
6332         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6333                                            page_offset(page));
6334         if (ordered) {
6335                 /*
6336                  * IO on this page will never be started, so we need
6337                  * to account for any ordered extents now
6338                  */
6339                 clear_extent_bit(tree, page_start, page_end,
6340                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6341                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6342                                  &cached_state, GFP_NOFS);
6343                 /*
6344                  * whoever cleared the private bit is responsible
6345                  * for the finish_ordered_io
6346                  */
6347                 if (TestClearPagePrivate2(page)) {
6348                         btrfs_finish_ordered_io(page->mapping->host,
6349                                                 page_start, page_end);
6350                 }
6351                 btrfs_put_ordered_extent(ordered);
6352                 cached_state = NULL;
6353                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6354                                  GFP_NOFS);
6355         }
6356         clear_extent_bit(tree, page_start, page_end,
6357                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6358                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6359         __btrfs_releasepage(page, GFP_NOFS);
6360
6361         ClearPageChecked(page);
6362         if (PagePrivate(page)) {
6363                 ClearPagePrivate(page);
6364                 set_page_private(page, 0);
6365                 page_cache_release(page);
6366         }
6367 }
6368
6369 /*
6370  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6371  * called from a page fault handler when a page is first dirtied. Hence we must
6372  * be careful to check for EOF conditions here. We set the page up correctly
6373  * for a written page which means we get ENOSPC checking when writing into
6374  * holes and correct delalloc and unwritten extent mapping on filesystems that
6375  * support these features.
6376  *
6377  * We are not allowed to take the i_mutex here so we have to play games to
6378  * protect against truncate races as the page could now be beyond EOF.  Because
6379  * vmtruncate() writes the inode size before removing pages, once we have the
6380  * page lock we can determine safely if the page is beyond EOF. If it is not
6381  * beyond EOF, then the page is guaranteed safe against truncation until we
6382  * unlock the page.
6383  */
6384 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6385 {
6386         struct page *page = vmf->page;
6387         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6388         struct btrfs_root *root = BTRFS_I(inode)->root;
6389         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6390         struct btrfs_ordered_extent *ordered;
6391         struct extent_state *cached_state = NULL;
6392         char *kaddr;
6393         unsigned long zero_start;
6394         loff_t size;
6395         int ret;
6396         u64 page_start;
6397         u64 page_end;
6398
6399         /* Need this to keep space reservations serialized */
6400         mutex_lock(&inode->i_mutex);
6401         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6402         mutex_unlock(&inode->i_mutex);
6403         if (!ret)
6404                 ret = btrfs_update_time(vma->vm_file);
6405         if (ret) {
6406                 if (ret == -ENOMEM)
6407                         ret = VM_FAULT_OOM;
6408                 else /* -ENOSPC, -EIO, etc */
6409                         ret = VM_FAULT_SIGBUS;
6410                 goto out;
6411         }
6412
6413         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6414 again:
6415         lock_page(page);
6416         size = i_size_read(inode);
6417         page_start = page_offset(page);
6418         page_end = page_start + PAGE_CACHE_SIZE - 1;
6419
6420         if ((page->mapping != inode->i_mapping) ||
6421             (page_start >= size)) {
6422                 /* page got truncated out from underneath us */
6423                 goto out_unlock;
6424         }
6425         wait_on_page_writeback(page);
6426
6427         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6428                          GFP_NOFS);
6429         set_page_extent_mapped(page);
6430
6431         /*
6432          * we can't set the delalloc bits if there are pending ordered
6433          * extents.  Drop our locks and wait for them to finish
6434          */
6435         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6436         if (ordered) {
6437                 unlock_extent_cached(io_tree, page_start, page_end,
6438                                      &cached_state, GFP_NOFS);
6439                 unlock_page(page);
6440                 btrfs_start_ordered_extent(inode, ordered, 1);
6441                 btrfs_put_ordered_extent(ordered);
6442                 goto again;
6443         }
6444
6445         /*
6446          * XXX - page_mkwrite gets called every time the page is dirtied, even
6447          * if it was already dirty, so for space accounting reasons we need to
6448          * clear any delalloc bits for the range we are fixing to save.  There
6449          * is probably a better way to do this, but for now keep consistent with
6450          * prepare_pages in the normal write path.
6451          */
6452         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6453                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6454                           0, 0, &cached_state, GFP_NOFS);
6455
6456         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6457                                         &cached_state);
6458         if (ret) {
6459                 unlock_extent_cached(io_tree, page_start, page_end,
6460                                      &cached_state, GFP_NOFS);
6461                 ret = VM_FAULT_SIGBUS;
6462                 goto out_unlock;
6463         }
6464         ret = 0;
6465
6466         /* page is wholly or partially inside EOF */
6467         if (page_start + PAGE_CACHE_SIZE > size)
6468                 zero_start = size & ~PAGE_CACHE_MASK;
6469         else
6470                 zero_start = PAGE_CACHE_SIZE;
6471
6472         if (zero_start != PAGE_CACHE_SIZE) {
6473                 kaddr = kmap(page);
6474                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6475                 flush_dcache_page(page);
6476                 kunmap(page);
6477         }
6478         ClearPageChecked(page);
6479         set_page_dirty(page);
6480         SetPageUptodate(page);
6481
6482         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6483         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6484
6485         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6486
6487 out_unlock:
6488         if (!ret)
6489                 return VM_FAULT_LOCKED;
6490         unlock_page(page);
6491 out:
6492         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6493         return ret;
6494 }
6495
6496 static int btrfs_truncate(struct inode *inode)
6497 {
6498         struct btrfs_root *root = BTRFS_I(inode)->root;
6499         struct btrfs_block_rsv *rsv;
6500         int ret;
6501         int err = 0;
6502         struct btrfs_trans_handle *trans;
6503         unsigned long nr;
6504         u64 mask = root->sectorsize - 1;
6505         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6506
6507         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6508         if (ret)
6509                 return ret;
6510
6511         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6512         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6513
6514         /*
6515          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6516          * 3 things going on here
6517          *
6518          * 1) We need to reserve space for our orphan item and the space to
6519          * delete our orphan item.  Lord knows we don't want to have a dangling
6520          * orphan item because we didn't reserve space to remove it.
6521          *
6522          * 2) We need to reserve space to update our inode.
6523          *
6524          * 3) We need to have something to cache all the space that is going to
6525          * be free'd up by the truncate operation, but also have some slack
6526          * space reserved in case it uses space during the truncate (thank you
6527          * very much snapshotting).
6528          *
6529          * And we need these to all be seperate.  The fact is we can use alot of
6530          * space doing the truncate, and we have no earthly idea how much space
6531          * we will use, so we need the truncate reservation to be seperate so it
6532          * doesn't end up using space reserved for updating the inode or
6533          * removing the orphan item.  We also need to be able to stop the
6534          * transaction and start a new one, which means we need to be able to
6535          * update the inode several times, and we have no idea of knowing how
6536          * many times that will be, so we can't just reserve 1 item for the
6537          * entirety of the opration, so that has to be done seperately as well.
6538          * Then there is the orphan item, which does indeed need to be held on
6539          * to for the whole operation, and we need nobody to touch this reserved
6540          * space except the orphan code.
6541          *
6542          * So that leaves us with
6543          *
6544          * 1) root->orphan_block_rsv - for the orphan deletion.
6545          * 2) rsv - for the truncate reservation, which we will steal from the
6546          * transaction reservation.
6547          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6548          * updating the inode.
6549          */
6550         rsv = btrfs_alloc_block_rsv(root);
6551         if (!rsv)
6552                 return -ENOMEM;
6553         rsv->size = min_size;
6554
6555         /*
6556          * 1 for the truncate slack space
6557          * 1 for the orphan item we're going to add
6558          * 1 for the orphan item deletion
6559          * 1 for updating the inode.
6560          */
6561         trans = btrfs_start_transaction(root, 4);
6562         if (IS_ERR(trans)) {
6563                 err = PTR_ERR(trans);
6564                 goto out;
6565         }
6566
6567         /* Migrate the slack space for the truncate to our reserve */
6568         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6569                                       min_size);
6570         BUG_ON(ret);
6571
6572         ret = btrfs_orphan_add(trans, inode);
6573         if (ret) {
6574                 btrfs_end_transaction(trans, root);
6575                 goto out;
6576         }
6577
6578         /*
6579          * setattr is responsible for setting the ordered_data_close flag,
6580          * but that is only tested during the last file release.  That
6581          * could happen well after the next commit, leaving a great big
6582          * window where new writes may get lost if someone chooses to write
6583          * to this file after truncating to zero
6584          *
6585          * The inode doesn't have any dirty data here, and so if we commit
6586          * this is a noop.  If someone immediately starts writing to the inode
6587          * it is very likely we'll catch some of their writes in this
6588          * transaction, and the commit will find this file on the ordered
6589          * data list with good things to send down.
6590          *
6591          * This is a best effort solution, there is still a window where
6592          * using truncate to replace the contents of the file will
6593          * end up with a zero length file after a crash.
6594          */
6595         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6596                 btrfs_add_ordered_operation(trans, root, inode);
6597
6598         while (1) {
6599                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6600                 if (ret) {
6601                         /*
6602                          * This can only happen with the original transaction we
6603                          * started above, every other time we shouldn't have a
6604                          * transaction started yet.
6605                          */
6606                         if (ret == -EAGAIN)
6607                                 goto end_trans;
6608                         err = ret;
6609                         break;
6610                 }
6611
6612                 if (!trans) {
6613                         /* Just need the 1 for updating the inode */
6614                         trans = btrfs_start_transaction(root, 1);
6615                         if (IS_ERR(trans)) {
6616                                 ret = err = PTR_ERR(trans);
6617                                 trans = NULL;
6618                                 break;
6619                         }
6620                 }
6621
6622                 trans->block_rsv = rsv;
6623
6624                 ret = btrfs_truncate_inode_items(trans, root, inode,
6625                                                  inode->i_size,
6626                                                  BTRFS_EXTENT_DATA_KEY);
6627                 if (ret != -EAGAIN) {
6628                         err = ret;
6629                         break;
6630                 }
6631
6632                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6633                 ret = btrfs_update_inode(trans, root, inode);
6634                 if (ret) {
6635                         err = ret;
6636                         break;
6637                 }
6638 end_trans:
6639                 nr = trans->blocks_used;
6640                 btrfs_end_transaction(trans, root);
6641                 trans = NULL;
6642                 btrfs_btree_balance_dirty(root, nr);
6643         }
6644
6645         if (ret == 0 && inode->i_nlink > 0) {
6646                 trans->block_rsv = root->orphan_block_rsv;
6647                 ret = btrfs_orphan_del(trans, inode);
6648                 if (ret)
6649                         err = ret;
6650         } else if (ret && inode->i_nlink > 0) {
6651                 /*
6652                  * Failed to do the truncate, remove us from the in memory
6653                  * orphan list.
6654                  */
6655                 ret = btrfs_orphan_del(NULL, inode);
6656         }
6657
6658         if (trans) {
6659                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6660                 ret = btrfs_update_inode(trans, root, inode);
6661                 if (ret && !err)
6662                         err = ret;
6663
6664                 nr = trans->blocks_used;
6665                 ret = btrfs_end_transaction(trans, root);
6666                 btrfs_btree_balance_dirty(root, nr);
6667         }
6668
6669 out:
6670         btrfs_free_block_rsv(root, rsv);
6671
6672         if (ret && !err)
6673                 err = ret;
6674
6675         return err;
6676 }
6677
6678 /*
6679  * create a new subvolume directory/inode (helper for the ioctl).
6680  */
6681 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6682                              struct btrfs_root *new_root, u64 new_dirid)
6683 {
6684         struct inode *inode;
6685         int err;
6686         u64 index = 0;
6687
6688         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6689                                 new_dirid, S_IFDIR | 0700, &index);
6690         if (IS_ERR(inode))
6691                 return PTR_ERR(inode);
6692         inode->i_op = &btrfs_dir_inode_operations;
6693         inode->i_fop = &btrfs_dir_file_operations;
6694
6695         set_nlink(inode, 1);
6696         btrfs_i_size_write(inode, 0);
6697
6698         err = btrfs_update_inode(trans, new_root, inode);
6699         BUG_ON(err);
6700
6701         iput(inode);
6702         return 0;
6703 }
6704
6705 struct inode *btrfs_alloc_inode(struct super_block *sb)
6706 {
6707         struct btrfs_inode *ei;
6708         struct inode *inode;
6709
6710         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6711         if (!ei)
6712                 return NULL;
6713
6714         ei->root = NULL;
6715         ei->space_info = NULL;
6716         ei->generation = 0;
6717         ei->sequence = 0;
6718         ei->last_trans = 0;
6719         ei->last_sub_trans = 0;
6720         ei->logged_trans = 0;
6721         ei->delalloc_bytes = 0;
6722         ei->disk_i_size = 0;
6723         ei->flags = 0;
6724         ei->csum_bytes = 0;
6725         ei->index_cnt = (u64)-1;
6726         ei->last_unlink_trans = 0;
6727
6728         spin_lock_init(&ei->lock);
6729         ei->outstanding_extents = 0;
6730         ei->reserved_extents = 0;
6731
6732         ei->ordered_data_close = 0;
6733         ei->orphan_meta_reserved = 0;
6734         ei->dummy_inode = 0;
6735         ei->in_defrag = 0;
6736         ei->delalloc_meta_reserved = 0;
6737         ei->force_compress = BTRFS_COMPRESS_NONE;
6738
6739         ei->delayed_node = NULL;
6740
6741         inode = &ei->vfs_inode;
6742         extent_map_tree_init(&ei->extent_tree);
6743         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6744         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6745         mutex_init(&ei->log_mutex);
6746         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6747         INIT_LIST_HEAD(&ei->i_orphan);
6748         INIT_LIST_HEAD(&ei->delalloc_inodes);
6749         INIT_LIST_HEAD(&ei->ordered_operations);
6750         RB_CLEAR_NODE(&ei->rb_node);
6751
6752         return inode;
6753 }
6754
6755 static void btrfs_i_callback(struct rcu_head *head)
6756 {
6757         struct inode *inode = container_of(head, struct inode, i_rcu);
6758         INIT_LIST_HEAD(&inode->i_dentry);
6759         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6760 }
6761
6762 void btrfs_destroy_inode(struct inode *inode)
6763 {
6764         struct btrfs_ordered_extent *ordered;
6765         struct btrfs_root *root = BTRFS_I(inode)->root;
6766
6767         WARN_ON(!list_empty(&inode->i_dentry));
6768         WARN_ON(inode->i_data.nrpages);
6769         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6770         WARN_ON(BTRFS_I(inode)->reserved_extents);
6771         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6772         WARN_ON(BTRFS_I(inode)->csum_bytes);
6773
6774         /*
6775          * This can happen where we create an inode, but somebody else also
6776          * created the same inode and we need to destroy the one we already
6777          * created.
6778          */
6779         if (!root)
6780                 goto free;
6781
6782         /*
6783          * Make sure we're properly removed from the ordered operation
6784          * lists.
6785          */
6786         smp_mb();
6787         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6788                 spin_lock(&root->fs_info->ordered_extent_lock);
6789                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6790                 spin_unlock(&root->fs_info->ordered_extent_lock);
6791         }
6792
6793         spin_lock(&root->orphan_lock);
6794         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6795                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6796                        (unsigned long long)btrfs_ino(inode));
6797                 list_del_init(&BTRFS_I(inode)->i_orphan);
6798         }
6799         spin_unlock(&root->orphan_lock);
6800
6801         while (1) {
6802                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6803                 if (!ordered)
6804                         break;
6805                 else {
6806                         printk(KERN_ERR "btrfs found ordered "
6807                                "extent %llu %llu on inode cleanup\n",
6808                                (unsigned long long)ordered->file_offset,
6809                                (unsigned long long)ordered->len);
6810                         btrfs_remove_ordered_extent(inode, ordered);
6811                         btrfs_put_ordered_extent(ordered);
6812                         btrfs_put_ordered_extent(ordered);
6813                 }
6814         }
6815         inode_tree_del(inode);
6816         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6817 free:
6818         btrfs_remove_delayed_node(inode);
6819         call_rcu(&inode->i_rcu, btrfs_i_callback);
6820 }
6821
6822 int btrfs_drop_inode(struct inode *inode)
6823 {
6824         struct btrfs_root *root = BTRFS_I(inode)->root;
6825
6826         if (btrfs_root_refs(&root->root_item) == 0 &&
6827             !btrfs_is_free_space_inode(root, inode))
6828                 return 1;
6829         else
6830                 return generic_drop_inode(inode);
6831 }
6832
6833 static void init_once(void *foo)
6834 {
6835         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6836
6837         inode_init_once(&ei->vfs_inode);
6838 }
6839
6840 void btrfs_destroy_cachep(void)
6841 {
6842         if (btrfs_inode_cachep)
6843                 kmem_cache_destroy(btrfs_inode_cachep);
6844         if (btrfs_trans_handle_cachep)
6845                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6846         if (btrfs_transaction_cachep)
6847                 kmem_cache_destroy(btrfs_transaction_cachep);
6848         if (btrfs_path_cachep)
6849                 kmem_cache_destroy(btrfs_path_cachep);
6850         if (btrfs_free_space_cachep)
6851                 kmem_cache_destroy(btrfs_free_space_cachep);
6852 }
6853
6854 int btrfs_init_cachep(void)
6855 {
6856         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6857                         sizeof(struct btrfs_inode), 0,
6858                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6859         if (!btrfs_inode_cachep)
6860                 goto fail;
6861
6862         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6863                         sizeof(struct btrfs_trans_handle), 0,
6864                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6865         if (!btrfs_trans_handle_cachep)
6866                 goto fail;
6867
6868         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6869                         sizeof(struct btrfs_transaction), 0,
6870                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6871         if (!btrfs_transaction_cachep)
6872                 goto fail;
6873
6874         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6875                         sizeof(struct btrfs_path), 0,
6876                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6877         if (!btrfs_path_cachep)
6878                 goto fail;
6879
6880         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6881                         sizeof(struct btrfs_free_space), 0,
6882                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6883         if (!btrfs_free_space_cachep)
6884                 goto fail;
6885
6886         return 0;
6887 fail:
6888         btrfs_destroy_cachep();
6889         return -ENOMEM;
6890 }
6891
6892 static int btrfs_getattr(struct vfsmount *mnt,
6893                          struct dentry *dentry, struct kstat *stat)
6894 {
6895         struct inode *inode = dentry->d_inode;
6896         u32 blocksize = inode->i_sb->s_blocksize;
6897
6898         generic_fillattr(inode, stat);
6899         stat->dev = BTRFS_I(inode)->root->anon_dev;
6900         stat->blksize = PAGE_CACHE_SIZE;
6901         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6902                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6903         return 0;
6904 }
6905
6906 /*
6907  * If a file is moved, it will inherit the cow and compression flags of the new
6908  * directory.
6909  */
6910 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6911 {
6912         struct btrfs_inode *b_dir = BTRFS_I(dir);
6913         struct btrfs_inode *b_inode = BTRFS_I(inode);
6914
6915         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6916                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6917         else
6918                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6919
6920         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6921                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6922         else
6923                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6924 }
6925
6926 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6927                            struct inode *new_dir, struct dentry *new_dentry)
6928 {
6929         struct btrfs_trans_handle *trans;
6930         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6931         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6932         struct inode *new_inode = new_dentry->d_inode;
6933         struct inode *old_inode = old_dentry->d_inode;
6934         struct timespec ctime = CURRENT_TIME;
6935         u64 index = 0;
6936         u64 root_objectid;
6937         int ret;
6938         u64 old_ino = btrfs_ino(old_inode);
6939
6940         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6941                 return -EPERM;
6942
6943         /* we only allow rename subvolume link between subvolumes */
6944         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6945                 return -EXDEV;
6946
6947         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6948             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6949                 return -ENOTEMPTY;
6950
6951         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6952             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6953                 return -ENOTEMPTY;
6954         /*
6955          * we're using rename to replace one file with another.
6956          * and the replacement file is large.  Start IO on it now so
6957          * we don't add too much work to the end of the transaction
6958          */
6959         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6960             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6961                 filemap_flush(old_inode->i_mapping);
6962
6963         /* close the racy window with snapshot create/destroy ioctl */
6964         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6965                 down_read(&root->fs_info->subvol_sem);
6966         /*
6967          * We want to reserve the absolute worst case amount of items.  So if
6968          * both inodes are subvols and we need to unlink them then that would
6969          * require 4 item modifications, but if they are both normal inodes it
6970          * would require 5 item modifications, so we'll assume their normal
6971          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6972          * should cover the worst case number of items we'll modify.
6973          */
6974         trans = btrfs_start_transaction(root, 20);
6975         if (IS_ERR(trans)) {
6976                 ret = PTR_ERR(trans);
6977                 goto out_notrans;
6978         }
6979
6980         if (dest != root)
6981                 btrfs_record_root_in_trans(trans, dest);
6982
6983         ret = btrfs_set_inode_index(new_dir, &index);
6984         if (ret)
6985                 goto out_fail;
6986
6987         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6988                 /* force full log commit if subvolume involved. */
6989                 root->fs_info->last_trans_log_full_commit = trans->transid;
6990         } else {
6991                 ret = btrfs_insert_inode_ref(trans, dest,
6992                                              new_dentry->d_name.name,
6993                                              new_dentry->d_name.len,
6994                                              old_ino,
6995                                              btrfs_ino(new_dir), index);
6996                 if (ret)
6997                         goto out_fail;
6998                 /*
6999                  * this is an ugly little race, but the rename is required
7000                  * to make sure that if we crash, the inode is either at the
7001                  * old name or the new one.  pinning the log transaction lets
7002                  * us make sure we don't allow a log commit to come in after
7003                  * we unlink the name but before we add the new name back in.
7004                  */
7005                 btrfs_pin_log_trans(root);
7006         }
7007         /*
7008          * make sure the inode gets flushed if it is replacing
7009          * something.
7010          */
7011         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7012                 btrfs_add_ordered_operation(trans, root, old_inode);
7013
7014         old_dir->i_ctime = old_dir->i_mtime = ctime;
7015         new_dir->i_ctime = new_dir->i_mtime = ctime;
7016         old_inode->i_ctime = ctime;
7017
7018         if (old_dentry->d_parent != new_dentry->d_parent)
7019                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7020
7021         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7022                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7023                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7024                                         old_dentry->d_name.name,
7025                                         old_dentry->d_name.len);
7026         } else {
7027                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7028                                         old_dentry->d_inode,
7029                                         old_dentry->d_name.name,
7030                                         old_dentry->d_name.len);
7031                 if (!ret)
7032                         ret = btrfs_update_inode(trans, root, old_inode);
7033         }
7034         BUG_ON(ret);
7035
7036         if (new_inode) {
7037                 new_inode->i_ctime = CURRENT_TIME;
7038                 if (unlikely(btrfs_ino(new_inode) ==
7039                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7040                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7041                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7042                                                 root_objectid,
7043                                                 new_dentry->d_name.name,
7044                                                 new_dentry->d_name.len);
7045                         BUG_ON(new_inode->i_nlink == 0);
7046                 } else {
7047                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7048                                                  new_dentry->d_inode,
7049                                                  new_dentry->d_name.name,
7050                                                  new_dentry->d_name.len);
7051                 }
7052                 BUG_ON(ret);
7053                 if (new_inode->i_nlink == 0) {
7054                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7055                         BUG_ON(ret);
7056                 }
7057         }
7058
7059         fixup_inode_flags(new_dir, old_inode);
7060
7061         ret = btrfs_add_link(trans, new_dir, old_inode,
7062                              new_dentry->d_name.name,
7063                              new_dentry->d_name.len, 0, index);
7064         BUG_ON(ret);
7065
7066         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7067                 struct dentry *parent = new_dentry->d_parent;
7068                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7069                 btrfs_end_log_trans(root);
7070         }
7071 out_fail:
7072         btrfs_end_transaction(trans, root);
7073 out_notrans:
7074         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7075                 up_read(&root->fs_info->subvol_sem);
7076
7077         return ret;
7078 }
7079
7080 /*
7081  * some fairly slow code that needs optimization. This walks the list
7082  * of all the inodes with pending delalloc and forces them to disk.
7083  */
7084 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7085 {
7086         struct list_head *head = &root->fs_info->delalloc_inodes;
7087         struct btrfs_inode *binode;
7088         struct inode *inode;
7089
7090         if (root->fs_info->sb->s_flags & MS_RDONLY)
7091                 return -EROFS;
7092
7093         spin_lock(&root->fs_info->delalloc_lock);
7094         while (!list_empty(head)) {
7095                 binode = list_entry(head->next, struct btrfs_inode,
7096                                     delalloc_inodes);
7097                 inode = igrab(&binode->vfs_inode);
7098                 if (!inode)
7099                         list_del_init(&binode->delalloc_inodes);
7100                 spin_unlock(&root->fs_info->delalloc_lock);
7101                 if (inode) {
7102                         filemap_flush(inode->i_mapping);
7103                         if (delay_iput)
7104                                 btrfs_add_delayed_iput(inode);
7105                         else
7106                                 iput(inode);
7107                 }
7108                 cond_resched();
7109                 spin_lock(&root->fs_info->delalloc_lock);
7110         }
7111         spin_unlock(&root->fs_info->delalloc_lock);
7112
7113         /* the filemap_flush will queue IO into the worker threads, but
7114          * we have to make sure the IO is actually started and that
7115          * ordered extents get created before we return
7116          */
7117         atomic_inc(&root->fs_info->async_submit_draining);
7118         while (atomic_read(&root->fs_info->nr_async_submits) ||
7119               atomic_read(&root->fs_info->async_delalloc_pages)) {
7120                 wait_event(root->fs_info->async_submit_wait,
7121                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7122                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7123         }
7124         atomic_dec(&root->fs_info->async_submit_draining);
7125         return 0;
7126 }
7127
7128 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7129                          const char *symname)
7130 {
7131         struct btrfs_trans_handle *trans;
7132         struct btrfs_root *root = BTRFS_I(dir)->root;
7133         struct btrfs_path *path;
7134         struct btrfs_key key;
7135         struct inode *inode = NULL;
7136         int err;
7137         int drop_inode = 0;
7138         u64 objectid;
7139         u64 index = 0 ;
7140         int name_len;
7141         int datasize;
7142         unsigned long ptr;
7143         struct btrfs_file_extent_item *ei;
7144         struct extent_buffer *leaf;
7145         unsigned long nr = 0;
7146
7147         name_len = strlen(symname) + 1;
7148         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7149                 return -ENAMETOOLONG;
7150
7151         /*
7152          * 2 items for inode item and ref
7153          * 2 items for dir items
7154          * 1 item for xattr if selinux is on
7155          */
7156         trans = btrfs_start_transaction(root, 5);
7157         if (IS_ERR(trans))
7158                 return PTR_ERR(trans);
7159
7160         err = btrfs_find_free_ino(root, &objectid);
7161         if (err)
7162                 goto out_unlock;
7163
7164         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7165                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7166                                 S_IFLNK|S_IRWXUGO, &index);
7167         if (IS_ERR(inode)) {
7168                 err = PTR_ERR(inode);
7169                 goto out_unlock;
7170         }
7171
7172         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7173         if (err) {
7174                 drop_inode = 1;
7175                 goto out_unlock;
7176         }
7177
7178         /*
7179         * If the active LSM wants to access the inode during
7180         * d_instantiate it needs these. Smack checks to see
7181         * if the filesystem supports xattrs by looking at the
7182         * ops vector.
7183         */
7184         inode->i_fop = &btrfs_file_operations;
7185         inode->i_op = &btrfs_file_inode_operations;
7186
7187         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7188         if (err)
7189                 drop_inode = 1;
7190         else {
7191                 inode->i_mapping->a_ops = &btrfs_aops;
7192                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7193                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7194         }
7195         if (drop_inode)
7196                 goto out_unlock;
7197
7198         path = btrfs_alloc_path();
7199         if (!path) {
7200                 err = -ENOMEM;
7201                 drop_inode = 1;
7202                 goto out_unlock;
7203         }
7204         key.objectid = btrfs_ino(inode);
7205         key.offset = 0;
7206         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7207         datasize = btrfs_file_extent_calc_inline_size(name_len);
7208         err = btrfs_insert_empty_item(trans, root, path, &key,
7209                                       datasize);
7210         if (err) {
7211                 drop_inode = 1;
7212                 btrfs_free_path(path);
7213                 goto out_unlock;
7214         }
7215         leaf = path->nodes[0];
7216         ei = btrfs_item_ptr(leaf, path->slots[0],
7217                             struct btrfs_file_extent_item);
7218         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7219         btrfs_set_file_extent_type(leaf, ei,
7220                                    BTRFS_FILE_EXTENT_INLINE);
7221         btrfs_set_file_extent_encryption(leaf, ei, 0);
7222         btrfs_set_file_extent_compression(leaf, ei, 0);
7223         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7224         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7225
7226         ptr = btrfs_file_extent_inline_start(ei);
7227         write_extent_buffer(leaf, symname, ptr, name_len);
7228         btrfs_mark_buffer_dirty(leaf);
7229         btrfs_free_path(path);
7230
7231         inode->i_op = &btrfs_symlink_inode_operations;
7232         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7233         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7234         inode_set_bytes(inode, name_len);
7235         btrfs_i_size_write(inode, name_len - 1);
7236         err = btrfs_update_inode(trans, root, inode);
7237         if (err)
7238                 drop_inode = 1;
7239
7240 out_unlock:
7241         if (!err)
7242                 d_instantiate(dentry, inode);
7243         nr = trans->blocks_used;
7244         btrfs_end_transaction(trans, root);
7245         if (drop_inode) {
7246                 inode_dec_link_count(inode);
7247                 iput(inode);
7248         }
7249         btrfs_btree_balance_dirty(root, nr);
7250         return err;
7251 }
7252
7253 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7254                                        u64 start, u64 num_bytes, u64 min_size,
7255                                        loff_t actual_len, u64 *alloc_hint,
7256                                        struct btrfs_trans_handle *trans)
7257 {
7258         struct btrfs_root *root = BTRFS_I(inode)->root;
7259         struct btrfs_key ins;
7260         u64 cur_offset = start;
7261         u64 i_size;
7262         int ret = 0;
7263         bool own_trans = true;
7264
7265         if (trans)
7266                 own_trans = false;
7267         while (num_bytes > 0) {
7268                 if (own_trans) {
7269                         trans = btrfs_start_transaction(root, 3);
7270                         if (IS_ERR(trans)) {
7271                                 ret = PTR_ERR(trans);
7272                                 break;
7273                         }
7274                 }
7275
7276                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7277                                            0, *alloc_hint, (u64)-1, &ins, 1);
7278                 if (ret) {
7279                         if (own_trans)
7280                                 btrfs_end_transaction(trans, root);
7281                         break;
7282                 }
7283
7284                 ret = insert_reserved_file_extent(trans, inode,
7285                                                   cur_offset, ins.objectid,
7286                                                   ins.offset, ins.offset,
7287                                                   ins.offset, 0, 0, 0,
7288                                                   BTRFS_FILE_EXTENT_PREALLOC);
7289                 BUG_ON(ret);
7290                 btrfs_drop_extent_cache(inode, cur_offset,
7291                                         cur_offset + ins.offset -1, 0);
7292
7293                 num_bytes -= ins.offset;
7294                 cur_offset += ins.offset;
7295                 *alloc_hint = ins.objectid + ins.offset;
7296
7297                 inode->i_ctime = CURRENT_TIME;
7298                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7299                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7300                     (actual_len > inode->i_size) &&
7301                     (cur_offset > inode->i_size)) {
7302                         if (cur_offset > actual_len)
7303                                 i_size = actual_len;
7304                         else
7305                                 i_size = cur_offset;
7306                         i_size_write(inode, i_size);
7307                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7308                 }
7309
7310                 ret = btrfs_update_inode(trans, root, inode);
7311                 BUG_ON(ret);
7312
7313                 if (own_trans)
7314                         btrfs_end_transaction(trans, root);
7315         }
7316         return ret;
7317 }
7318
7319 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7320                               u64 start, u64 num_bytes, u64 min_size,
7321                               loff_t actual_len, u64 *alloc_hint)
7322 {
7323         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7324                                            min_size, actual_len, alloc_hint,
7325                                            NULL);
7326 }
7327
7328 int btrfs_prealloc_file_range_trans(struct inode *inode,
7329                                     struct btrfs_trans_handle *trans, int mode,
7330                                     u64 start, u64 num_bytes, u64 min_size,
7331                                     loff_t actual_len, u64 *alloc_hint)
7332 {
7333         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7334                                            min_size, actual_len, alloc_hint, trans);
7335 }
7336
7337 static int btrfs_set_page_dirty(struct page *page)
7338 {
7339         return __set_page_dirty_nobuffers(page);
7340 }
7341
7342 static int btrfs_permission(struct inode *inode, int mask)
7343 {
7344         struct btrfs_root *root = BTRFS_I(inode)->root;
7345         umode_t mode = inode->i_mode;
7346
7347         if (mask & MAY_WRITE &&
7348             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7349                 if (btrfs_root_readonly(root))
7350                         return -EROFS;
7351                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7352                         return -EACCES;
7353         }
7354         return generic_permission(inode, mask);
7355 }
7356
7357 static const struct inode_operations btrfs_dir_inode_operations = {
7358         .getattr        = btrfs_getattr,
7359         .lookup         = btrfs_lookup,
7360         .create         = btrfs_create,
7361         .unlink         = btrfs_unlink,
7362         .link           = btrfs_link,
7363         .mkdir          = btrfs_mkdir,
7364         .rmdir          = btrfs_rmdir,
7365         .rename         = btrfs_rename,
7366         .symlink        = btrfs_symlink,
7367         .setattr        = btrfs_setattr,
7368         .mknod          = btrfs_mknod,
7369         .setxattr       = btrfs_setxattr,
7370         .getxattr       = btrfs_getxattr,
7371         .listxattr      = btrfs_listxattr,
7372         .removexattr    = btrfs_removexattr,
7373         .permission     = btrfs_permission,
7374         .get_acl        = btrfs_get_acl,
7375 };
7376 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7377         .lookup         = btrfs_lookup,
7378         .permission     = btrfs_permission,
7379         .get_acl        = btrfs_get_acl,
7380 };
7381
7382 static const struct file_operations btrfs_dir_file_operations = {
7383         .llseek         = generic_file_llseek,
7384         .read           = generic_read_dir,
7385         .readdir        = btrfs_real_readdir,
7386         .unlocked_ioctl = btrfs_ioctl,
7387 #ifdef CONFIG_COMPAT
7388         .compat_ioctl   = btrfs_ioctl,
7389 #endif
7390         .release        = btrfs_release_file,
7391         .fsync          = btrfs_sync_file,
7392 };
7393
7394 static struct extent_io_ops btrfs_extent_io_ops = {
7395         .fill_delalloc = run_delalloc_range,
7396         .submit_bio_hook = btrfs_submit_bio_hook,
7397         .merge_bio_hook = btrfs_merge_bio_hook,
7398         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7399         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7400         .writepage_start_hook = btrfs_writepage_start_hook,
7401         .set_bit_hook = btrfs_set_bit_hook,
7402         .clear_bit_hook = btrfs_clear_bit_hook,
7403         .merge_extent_hook = btrfs_merge_extent_hook,
7404         .split_extent_hook = btrfs_split_extent_hook,
7405 };
7406
7407 /*
7408  * btrfs doesn't support the bmap operation because swapfiles
7409  * use bmap to make a mapping of extents in the file.  They assume
7410  * these extents won't change over the life of the file and they
7411  * use the bmap result to do IO directly to the drive.
7412  *
7413  * the btrfs bmap call would return logical addresses that aren't
7414  * suitable for IO and they also will change frequently as COW
7415  * operations happen.  So, swapfile + btrfs == corruption.
7416  *
7417  * For now we're avoiding this by dropping bmap.
7418  */
7419 static const struct address_space_operations btrfs_aops = {
7420         .readpage       = btrfs_readpage,
7421         .writepage      = btrfs_writepage,
7422         .writepages     = btrfs_writepages,
7423         .readpages      = btrfs_readpages,
7424         .direct_IO      = btrfs_direct_IO,
7425         .invalidatepage = btrfs_invalidatepage,
7426         .releasepage    = btrfs_releasepage,
7427         .set_page_dirty = btrfs_set_page_dirty,
7428         .error_remove_page = generic_error_remove_page,
7429 };
7430
7431 static const struct address_space_operations btrfs_symlink_aops = {
7432         .readpage       = btrfs_readpage,
7433         .writepage      = btrfs_writepage,
7434         .invalidatepage = btrfs_invalidatepage,
7435         .releasepage    = btrfs_releasepage,
7436 };
7437
7438 static const struct inode_operations btrfs_file_inode_operations = {
7439         .getattr        = btrfs_getattr,
7440         .setattr        = btrfs_setattr,
7441         .setxattr       = btrfs_setxattr,
7442         .getxattr       = btrfs_getxattr,
7443         .listxattr      = btrfs_listxattr,
7444         .removexattr    = btrfs_removexattr,
7445         .permission     = btrfs_permission,
7446         .fiemap         = btrfs_fiemap,
7447         .get_acl        = btrfs_get_acl,
7448 };
7449 static const struct inode_operations btrfs_special_inode_operations = {
7450         .getattr        = btrfs_getattr,
7451         .setattr        = btrfs_setattr,
7452         .permission     = btrfs_permission,
7453         .setxattr       = btrfs_setxattr,
7454         .getxattr       = btrfs_getxattr,
7455         .listxattr      = btrfs_listxattr,
7456         .removexattr    = btrfs_removexattr,
7457         .get_acl        = btrfs_get_acl,
7458 };
7459 static const struct inode_operations btrfs_symlink_inode_operations = {
7460         .readlink       = generic_readlink,
7461         .follow_link    = page_follow_link_light,
7462         .put_link       = page_put_link,
7463         .getattr        = btrfs_getattr,
7464         .setattr        = btrfs_setattr,
7465         .permission     = btrfs_permission,
7466         .setxattr       = btrfs_setxattr,
7467         .getxattr       = btrfs_getxattr,
7468         .listxattr      = btrfs_listxattr,
7469         .removexattr    = btrfs_removexattr,
7470         .get_acl        = btrfs_get_acl,
7471 };
7472
7473 const struct dentry_operations btrfs_dentry_operations = {
7474         .d_delete       = btrfs_dentry_delete,
7475         .d_release      = btrfs_dentry_release,
7476 };