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_throttle(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 encoding;
3013         int ret;
3014         int err = 0;
3015         u64 ino = btrfs_ino(inode);
3016
3017         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3018
3019         path = btrfs_alloc_path();
3020         if (!path)
3021                 return -ENOMEM;
3022         path->reada = -1;
3023
3024         if (root->ref_cows || root == root->fs_info->tree_root)
3025                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3026
3027         /*
3028          * This function is also used to drop the items in the log tree before
3029          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3030          * it is used to drop the loged items. So we shouldn't kill the delayed
3031          * items.
3032          */
3033         if (min_type == 0 && root == BTRFS_I(inode)->root)
3034                 btrfs_kill_delayed_inode_items(inode);
3035
3036         key.objectid = ino;
3037         key.offset = (u64)-1;
3038         key.type = (u8)-1;
3039
3040 search_again:
3041         path->leave_spinning = 1;
3042         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3043         if (ret < 0) {
3044                 err = ret;
3045                 goto out;
3046         }
3047
3048         if (ret > 0) {
3049                 /* there are no items in the tree for us to truncate, we're
3050                  * done
3051                  */
3052                 if (path->slots[0] == 0)
3053                         goto out;
3054                 path->slots[0]--;
3055         }
3056
3057         while (1) {
3058                 fi = NULL;
3059                 leaf = path->nodes[0];
3060                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3061                 found_type = btrfs_key_type(&found_key);
3062                 encoding = 0;
3063
3064                 if (found_key.objectid != ino)
3065                         break;
3066
3067                 if (found_type < min_type)
3068                         break;
3069
3070                 item_end = found_key.offset;
3071                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3072                         fi = btrfs_item_ptr(leaf, path->slots[0],
3073                                             struct btrfs_file_extent_item);
3074                         extent_type = btrfs_file_extent_type(leaf, fi);
3075                         encoding = btrfs_file_extent_compression(leaf, fi);
3076                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3077                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3078
3079                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3080                                 item_end +=
3081                                     btrfs_file_extent_num_bytes(leaf, fi);
3082                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3083                                 item_end += btrfs_file_extent_inline_len(leaf,
3084                                                                          fi);
3085                         }
3086                         item_end--;
3087                 }
3088                 if (found_type > min_type) {
3089                         del_item = 1;
3090                 } else {
3091                         if (item_end < new_size)
3092                                 break;
3093                         if (found_key.offset >= new_size)
3094                                 del_item = 1;
3095                         else
3096                                 del_item = 0;
3097                 }
3098                 found_extent = 0;
3099                 /* FIXME, shrink the extent if the ref count is only 1 */
3100                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3101                         goto delete;
3102
3103                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3104                         u64 num_dec;
3105                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3106                         if (!del_item && !encoding) {
3107                                 u64 orig_num_bytes =
3108                                         btrfs_file_extent_num_bytes(leaf, fi);
3109                                 extent_num_bytes = new_size -
3110                                         found_key.offset + root->sectorsize - 1;
3111                                 extent_num_bytes = extent_num_bytes &
3112                                         ~((u64)root->sectorsize - 1);
3113                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3114                                                          extent_num_bytes);
3115                                 num_dec = (orig_num_bytes -
3116                                            extent_num_bytes);
3117                                 if (root->ref_cows && extent_start != 0)
3118                                         inode_sub_bytes(inode, num_dec);
3119                                 btrfs_mark_buffer_dirty(leaf);
3120                         } else {
3121                                 extent_num_bytes =
3122                                         btrfs_file_extent_disk_num_bytes(leaf,
3123                                                                          fi);
3124                                 extent_offset = found_key.offset -
3125                                         btrfs_file_extent_offset(leaf, fi);
3126
3127                                 /* FIXME blocksize != 4096 */
3128                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3129                                 if (extent_start != 0) {
3130                                         found_extent = 1;
3131                                         if (root->ref_cows)
3132                                                 inode_sub_bytes(inode, num_dec);
3133                                 }
3134                         }
3135                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3136                         /*
3137                          * we can't truncate inline items that have had
3138                          * special encodings
3139                          */
3140                         if (!del_item &&
3141                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3142                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3143                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3144                                 u32 size = new_size - found_key.offset;
3145
3146                                 if (root->ref_cows) {
3147                                         inode_sub_bytes(inode, item_end + 1 -
3148                                                         new_size);
3149                                 }
3150                                 size =
3151                                     btrfs_file_extent_calc_inline_size(size);
3152                                 ret = btrfs_truncate_item(trans, root, path,
3153                                                           size, 1);
3154                         } else if (root->ref_cows) {
3155                                 inode_sub_bytes(inode, item_end + 1 -
3156                                                 found_key.offset);
3157                         }
3158                 }
3159 delete:
3160                 if (del_item) {
3161                         if (!pending_del_nr) {
3162                                 /* no pending yet, add ourselves */
3163                                 pending_del_slot = path->slots[0];
3164                                 pending_del_nr = 1;
3165                         } else if (pending_del_nr &&
3166                                    path->slots[0] + 1 == pending_del_slot) {
3167                                 /* hop on the pending chunk */
3168                                 pending_del_nr++;
3169                                 pending_del_slot = path->slots[0];
3170                         } else {
3171                                 BUG();
3172                         }
3173                 } else {
3174                         break;
3175                 }
3176                 if (found_extent && (root->ref_cows ||
3177                                      root == root->fs_info->tree_root)) {
3178                         btrfs_set_path_blocking(path);
3179                         ret = btrfs_free_extent(trans, root, extent_start,
3180                                                 extent_num_bytes, 0,
3181                                                 btrfs_header_owner(leaf),
3182                                                 ino, extent_offset);
3183                         BUG_ON(ret);
3184                 }
3185
3186                 if (found_type == BTRFS_INODE_ITEM_KEY)
3187                         break;
3188
3189                 if (path->slots[0] == 0 ||
3190                     path->slots[0] != pending_del_slot) {
3191                         if (root->ref_cows &&
3192                             BTRFS_I(inode)->location.objectid !=
3193                                                 BTRFS_FREE_INO_OBJECTID) {
3194                                 err = -EAGAIN;
3195                                 goto out;
3196                         }
3197                         if (pending_del_nr) {
3198                                 ret = btrfs_del_items(trans, root, path,
3199                                                 pending_del_slot,
3200                                                 pending_del_nr);
3201                                 BUG_ON(ret);
3202                                 pending_del_nr = 0;
3203                         }
3204                         btrfs_release_path(path);
3205                         goto search_again;
3206                 } else {
3207                         path->slots[0]--;
3208                 }
3209         }
3210 out:
3211         if (pending_del_nr) {
3212                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3213                                       pending_del_nr);
3214                 BUG_ON(ret);
3215         }
3216         btrfs_free_path(path);
3217         return err;
3218 }
3219
3220 /*
3221  * taken from block_truncate_page, but does cow as it zeros out
3222  * any bytes left in the last page in the file.
3223  */
3224 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3225 {
3226         struct inode *inode = mapping->host;
3227         struct btrfs_root *root = BTRFS_I(inode)->root;
3228         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3229         struct btrfs_ordered_extent *ordered;
3230         struct extent_state *cached_state = NULL;
3231         char *kaddr;
3232         u32 blocksize = root->sectorsize;
3233         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3234         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3235         struct page *page;
3236         gfp_t mask = btrfs_alloc_write_mask(mapping);
3237         int ret = 0;
3238         u64 page_start;
3239         u64 page_end;
3240
3241         if ((offset & (blocksize - 1)) == 0)
3242                 goto out;
3243         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3244         if (ret)
3245                 goto out;
3246
3247         ret = -ENOMEM;
3248 again:
3249         page = find_or_create_page(mapping, index, mask);
3250         if (!page) {
3251                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3252                 goto out;
3253         }
3254
3255         page_start = page_offset(page);
3256         page_end = page_start + PAGE_CACHE_SIZE - 1;
3257
3258         if (!PageUptodate(page)) {
3259                 ret = btrfs_readpage(NULL, page);
3260                 lock_page(page);
3261                 if (page->mapping != mapping) {
3262                         unlock_page(page);
3263                         page_cache_release(page);
3264                         goto again;
3265                 }
3266                 if (!PageUptodate(page)) {
3267                         ret = -EIO;
3268                         goto out_unlock;
3269                 }
3270         }
3271         wait_on_page_writeback(page);
3272
3273         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3274                          GFP_NOFS);
3275         set_page_extent_mapped(page);
3276
3277         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3278         if (ordered) {
3279                 unlock_extent_cached(io_tree, page_start, page_end,
3280                                      &cached_state, GFP_NOFS);
3281                 unlock_page(page);
3282                 page_cache_release(page);
3283                 btrfs_start_ordered_extent(inode, ordered, 1);
3284                 btrfs_put_ordered_extent(ordered);
3285                 goto again;
3286         }
3287
3288         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3289                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3290                           0, 0, &cached_state, GFP_NOFS);
3291
3292         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3293                                         &cached_state);
3294         if (ret) {
3295                 unlock_extent_cached(io_tree, page_start, page_end,
3296                                      &cached_state, GFP_NOFS);
3297                 goto out_unlock;
3298         }
3299
3300         ret = 0;
3301         if (offset != PAGE_CACHE_SIZE) {
3302                 kaddr = kmap(page);
3303                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3304                 flush_dcache_page(page);
3305                 kunmap(page);
3306         }
3307         ClearPageChecked(page);
3308         set_page_dirty(page);
3309         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3310                              GFP_NOFS);
3311
3312 out_unlock:
3313         if (ret)
3314                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3315         unlock_page(page);
3316         page_cache_release(page);
3317 out:
3318         return ret;
3319 }
3320
3321 /*
3322  * This function puts in dummy file extents for the area we're creating a hole
3323  * for.  So if we are truncating this file to a larger size we need to insert
3324  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3325  * the range between oldsize and size
3326  */
3327 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3328 {
3329         struct btrfs_trans_handle *trans;
3330         struct btrfs_root *root = BTRFS_I(inode)->root;
3331         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3332         struct extent_map *em = NULL;
3333         struct extent_state *cached_state = NULL;
3334         u64 mask = root->sectorsize - 1;
3335         u64 hole_start = (oldsize + mask) & ~mask;
3336         u64 block_end = (size + mask) & ~mask;
3337         u64 last_byte;
3338         u64 cur_offset;
3339         u64 hole_size;
3340         int err = 0;
3341
3342         if (size <= hole_start)
3343                 return 0;
3344
3345         while (1) {
3346                 struct btrfs_ordered_extent *ordered;
3347                 btrfs_wait_ordered_range(inode, hole_start,
3348                                          block_end - hole_start);
3349                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3350                                  &cached_state, GFP_NOFS);
3351                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3352                 if (!ordered)
3353                         break;
3354                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3355                                      &cached_state, GFP_NOFS);
3356                 btrfs_put_ordered_extent(ordered);
3357         }
3358
3359         cur_offset = hole_start;
3360         while (1) {
3361                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3362                                 block_end - cur_offset, 0);
3363                 BUG_ON(IS_ERR_OR_NULL(em));
3364                 last_byte = min(extent_map_end(em), block_end);
3365                 last_byte = (last_byte + mask) & ~mask;
3366                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3367                         u64 hint_byte = 0;
3368                         hole_size = last_byte - cur_offset;
3369
3370                         trans = btrfs_start_transaction(root, 3);
3371                         if (IS_ERR(trans)) {
3372                                 err = PTR_ERR(trans);
3373                                 break;
3374                         }
3375
3376                         err = btrfs_drop_extents(trans, inode, cur_offset,
3377                                                  cur_offset + hole_size,
3378                                                  &hint_byte, 1);
3379                         if (err) {
3380                                 btrfs_update_inode(trans, root, inode);
3381                                 btrfs_end_transaction(trans, root);
3382                                 break;
3383                         }
3384
3385                         err = btrfs_insert_file_extent(trans, root,
3386                                         btrfs_ino(inode), cur_offset, 0,
3387                                         0, hole_size, 0, hole_size,
3388                                         0, 0, 0);
3389                         if (err) {
3390                                 btrfs_update_inode(trans, root, inode);
3391                                 btrfs_end_transaction(trans, root);
3392                                 break;
3393                         }
3394
3395                         btrfs_drop_extent_cache(inode, hole_start,
3396                                         last_byte - 1, 0);
3397
3398                         btrfs_update_inode(trans, root, inode);
3399                         btrfs_end_transaction(trans, root);
3400                 }
3401                 free_extent_map(em);
3402                 em = NULL;
3403                 cur_offset = last_byte;
3404                 if (cur_offset >= block_end)
3405                         break;
3406         }
3407
3408         free_extent_map(em);
3409         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3410                              GFP_NOFS);
3411         return err;
3412 }
3413
3414 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3415 {
3416         struct btrfs_root *root = BTRFS_I(inode)->root;
3417         struct btrfs_trans_handle *trans;
3418         loff_t oldsize = i_size_read(inode);
3419         int ret;
3420
3421         if (newsize == oldsize)
3422                 return 0;
3423
3424         if (newsize > oldsize) {
3425                 truncate_pagecache(inode, oldsize, newsize);
3426                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3427                 if (ret)
3428                         return ret;
3429
3430                 trans = btrfs_start_transaction(root, 1);
3431                 if (IS_ERR(trans))
3432                         return PTR_ERR(trans);
3433
3434                 i_size_write(inode, newsize);
3435                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3436                 ret = btrfs_update_inode(trans, root, inode);
3437                 btrfs_end_transaction_throttle(trans, root);
3438         } else {
3439
3440                 /*
3441                  * We're truncating a file that used to have good data down to
3442                  * zero. Make sure it gets into the ordered flush list so that
3443                  * any new writes get down to disk quickly.
3444                  */
3445                 if (newsize == 0)
3446                         BTRFS_I(inode)->ordered_data_close = 1;
3447
3448                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3449                 truncate_setsize(inode, newsize);
3450                 ret = btrfs_truncate(inode);
3451         }
3452
3453         return ret;
3454 }
3455
3456 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3457 {
3458         struct inode *inode = dentry->d_inode;
3459         struct btrfs_root *root = BTRFS_I(inode)->root;
3460         int err;
3461
3462         if (btrfs_root_readonly(root))
3463                 return -EROFS;
3464
3465         err = inode_change_ok(inode, attr);
3466         if (err)
3467                 return err;
3468
3469         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3470                 err = btrfs_setsize(inode, attr->ia_size);
3471                 if (err)
3472                         return err;
3473         }
3474
3475         if (attr->ia_valid) {
3476                 setattr_copy(inode, attr);
3477                 err = btrfs_dirty_inode(inode);
3478
3479                 if (!err && attr->ia_valid & ATTR_MODE)
3480                         err = btrfs_acl_chmod(inode);
3481         }
3482
3483         return err;
3484 }
3485
3486 void btrfs_evict_inode(struct inode *inode)
3487 {
3488         struct btrfs_trans_handle *trans;
3489         struct btrfs_root *root = BTRFS_I(inode)->root;
3490         struct btrfs_block_rsv *rsv, *global_rsv;
3491         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3492         unsigned long nr;
3493         int ret;
3494
3495         trace_btrfs_inode_evict(inode);
3496
3497         truncate_inode_pages(&inode->i_data, 0);
3498         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3499                                btrfs_is_free_space_inode(root, inode)))
3500                 goto no_delete;
3501
3502         if (is_bad_inode(inode)) {
3503                 btrfs_orphan_del(NULL, inode);
3504                 goto no_delete;
3505         }
3506         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3507         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3508
3509         if (root->fs_info->log_root_recovering) {
3510                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3511                 goto no_delete;
3512         }
3513
3514         if (inode->i_nlink > 0) {
3515                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3516                 goto no_delete;
3517         }
3518
3519         rsv = btrfs_alloc_block_rsv(root);
3520         if (!rsv) {
3521                 btrfs_orphan_del(NULL, inode);
3522                 goto no_delete;
3523         }
3524         rsv->size = min_size;
3525         global_rsv = &root->fs_info->global_block_rsv;
3526
3527         btrfs_i_size_write(inode, 0);
3528
3529         /*
3530          * This is a bit simpler than btrfs_truncate since
3531          *
3532          * 1) We've already reserved our space for our orphan item in the
3533          *    unlink.
3534          * 2) We're going to delete the inode item, so we don't need to update
3535          *    it at all.
3536          *
3537          * So we just need to reserve some slack space in case we add bytes when
3538          * doing the truncate.
3539          */
3540         while (1) {
3541                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3542
3543                 /*
3544                  * Try and steal from the global reserve since we will
3545                  * likely not use this space anyway, we want to try as
3546                  * hard as possible to get this to work.
3547                  */
3548                 if (ret)
3549                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3550
3551                 if (ret) {
3552                         printk(KERN_WARNING "Could not get space for a "
3553                                "delete, will truncate on mount %d\n", ret);
3554                         btrfs_orphan_del(NULL, inode);
3555                         btrfs_free_block_rsv(root, rsv);
3556                         goto no_delete;
3557                 }
3558
3559                 trans = btrfs_start_transaction(root, 0);
3560                 if (IS_ERR(trans)) {
3561                         btrfs_orphan_del(NULL, inode);
3562                         btrfs_free_block_rsv(root, rsv);
3563                         goto no_delete;
3564                 }
3565
3566                 trans->block_rsv = rsv;
3567
3568                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3569                 if (ret != -EAGAIN)
3570                         break;
3571
3572                 nr = trans->blocks_used;
3573                 btrfs_end_transaction(trans, root);
3574                 trans = NULL;
3575                 btrfs_btree_balance_dirty(root, nr);
3576         }
3577
3578         btrfs_free_block_rsv(root, rsv);
3579
3580         if (ret == 0) {
3581                 trans->block_rsv = root->orphan_block_rsv;
3582                 ret = btrfs_orphan_del(trans, inode);
3583                 BUG_ON(ret);
3584         }
3585
3586         trans->block_rsv = &root->fs_info->trans_block_rsv;
3587         if (!(root == root->fs_info->tree_root ||
3588               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3589                 btrfs_return_ino(root, btrfs_ino(inode));
3590
3591         nr = trans->blocks_used;
3592         btrfs_end_transaction(trans, root);
3593         btrfs_btree_balance_dirty(root, nr);
3594 no_delete:
3595         end_writeback(inode);
3596         return;
3597 }
3598
3599 /*
3600  * this returns the key found in the dir entry in the location pointer.
3601  * If no dir entries were found, location->objectid is 0.
3602  */
3603 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3604                                struct btrfs_key *location)
3605 {
3606         const char *name = dentry->d_name.name;
3607         int namelen = dentry->d_name.len;
3608         struct btrfs_dir_item *di;
3609         struct btrfs_path *path;
3610         struct btrfs_root *root = BTRFS_I(dir)->root;
3611         int ret = 0;
3612
3613         path = btrfs_alloc_path();
3614         if (!path)
3615                 return -ENOMEM;
3616
3617         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3618                                     namelen, 0);
3619         if (IS_ERR(di))
3620                 ret = PTR_ERR(di);
3621
3622         if (IS_ERR_OR_NULL(di))
3623                 goto out_err;
3624
3625         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3626 out:
3627         btrfs_free_path(path);
3628         return ret;
3629 out_err:
3630         location->objectid = 0;
3631         goto out;
3632 }
3633
3634 /*
3635  * when we hit a tree root in a directory, the btrfs part of the inode
3636  * needs to be changed to reflect the root directory of the tree root.  This
3637  * is kind of like crossing a mount point.
3638  */
3639 static int fixup_tree_root_location(struct btrfs_root *root,
3640                                     struct inode *dir,
3641                                     struct dentry *dentry,
3642                                     struct btrfs_key *location,
3643                                     struct btrfs_root **sub_root)
3644 {
3645         struct btrfs_path *path;
3646         struct btrfs_root *new_root;
3647         struct btrfs_root_ref *ref;
3648         struct extent_buffer *leaf;
3649         int ret;
3650         int err = 0;
3651
3652         path = btrfs_alloc_path();
3653         if (!path) {
3654                 err = -ENOMEM;
3655                 goto out;
3656         }
3657
3658         err = -ENOENT;
3659         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3660                                   BTRFS_I(dir)->root->root_key.objectid,
3661                                   location->objectid);
3662         if (ret) {
3663                 if (ret < 0)
3664                         err = ret;
3665                 goto out;
3666         }
3667
3668         leaf = path->nodes[0];
3669         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3670         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3671             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3672                 goto out;
3673
3674         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3675                                    (unsigned long)(ref + 1),
3676                                    dentry->d_name.len);
3677         if (ret)
3678                 goto out;
3679
3680         btrfs_release_path(path);
3681
3682         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3683         if (IS_ERR(new_root)) {
3684                 err = PTR_ERR(new_root);
3685                 goto out;
3686         }
3687
3688         if (btrfs_root_refs(&new_root->root_item) == 0) {
3689                 err = -ENOENT;
3690                 goto out;
3691         }
3692
3693         *sub_root = new_root;
3694         location->objectid = btrfs_root_dirid(&new_root->root_item);
3695         location->type = BTRFS_INODE_ITEM_KEY;
3696         location->offset = 0;
3697         err = 0;
3698 out:
3699         btrfs_free_path(path);
3700         return err;
3701 }
3702
3703 static void inode_tree_add(struct inode *inode)
3704 {
3705         struct btrfs_root *root = BTRFS_I(inode)->root;
3706         struct btrfs_inode *entry;
3707         struct rb_node **p;
3708         struct rb_node *parent;
3709         u64 ino = btrfs_ino(inode);
3710 again:
3711         p = &root->inode_tree.rb_node;
3712         parent = NULL;
3713
3714         if (inode_unhashed(inode))
3715                 return;
3716
3717         spin_lock(&root->inode_lock);
3718         while (*p) {
3719                 parent = *p;
3720                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3721
3722                 if (ino < btrfs_ino(&entry->vfs_inode))
3723                         p = &parent->rb_left;
3724                 else if (ino > btrfs_ino(&entry->vfs_inode))
3725                         p = &parent->rb_right;
3726                 else {
3727                         WARN_ON(!(entry->vfs_inode.i_state &
3728                                   (I_WILL_FREE | I_FREEING)));
3729                         rb_erase(parent, &root->inode_tree);
3730                         RB_CLEAR_NODE(parent);
3731                         spin_unlock(&root->inode_lock);
3732                         goto again;
3733                 }
3734         }
3735         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3736         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3737         spin_unlock(&root->inode_lock);
3738 }
3739
3740 static void inode_tree_del(struct inode *inode)
3741 {
3742         struct btrfs_root *root = BTRFS_I(inode)->root;
3743         int empty = 0;
3744
3745         spin_lock(&root->inode_lock);
3746         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3747                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3748                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3749                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3750         }
3751         spin_unlock(&root->inode_lock);
3752
3753         /*
3754          * Free space cache has inodes in the tree root, but the tree root has a
3755          * root_refs of 0, so this could end up dropping the tree root as a
3756          * snapshot, so we need the extra !root->fs_info->tree_root check to
3757          * make sure we don't drop it.
3758          */
3759         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3760             root != root->fs_info->tree_root) {
3761                 synchronize_srcu(&root->fs_info->subvol_srcu);
3762                 spin_lock(&root->inode_lock);
3763                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3764                 spin_unlock(&root->inode_lock);
3765                 if (empty)
3766                         btrfs_add_dead_root(root);
3767         }
3768 }
3769
3770 int btrfs_invalidate_inodes(struct btrfs_root *root)
3771 {
3772         struct rb_node *node;
3773         struct rb_node *prev;
3774         struct btrfs_inode *entry;
3775         struct inode *inode;
3776         u64 objectid = 0;
3777
3778         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3779
3780         spin_lock(&root->inode_lock);
3781 again:
3782         node = root->inode_tree.rb_node;
3783         prev = NULL;
3784         while (node) {
3785                 prev = node;
3786                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3787
3788                 if (objectid < btrfs_ino(&entry->vfs_inode))
3789                         node = node->rb_left;
3790                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3791                         node = node->rb_right;
3792                 else
3793                         break;
3794         }
3795         if (!node) {
3796                 while (prev) {
3797                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3798                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3799                                 node = prev;
3800                                 break;
3801                         }
3802                         prev = rb_next(prev);
3803                 }
3804         }
3805         while (node) {
3806                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3807                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3808                 inode = igrab(&entry->vfs_inode);
3809                 if (inode) {
3810                         spin_unlock(&root->inode_lock);
3811                         if (atomic_read(&inode->i_count) > 1)
3812                                 d_prune_aliases(inode);
3813                         /*
3814                          * btrfs_drop_inode will have it removed from
3815                          * the inode cache when its usage count
3816                          * hits zero.
3817                          */
3818                         iput(inode);
3819                         cond_resched();
3820                         spin_lock(&root->inode_lock);
3821                         goto again;
3822                 }
3823
3824                 if (cond_resched_lock(&root->inode_lock))
3825                         goto again;
3826
3827                 node = rb_next(node);
3828         }
3829         spin_unlock(&root->inode_lock);
3830         return 0;
3831 }
3832
3833 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3834 {
3835         struct btrfs_iget_args *args = p;
3836         inode->i_ino = args->ino;
3837         BTRFS_I(inode)->root = args->root;
3838         btrfs_set_inode_space_info(args->root, inode);
3839         return 0;
3840 }
3841
3842 static int btrfs_find_actor(struct inode *inode, void *opaque)
3843 {
3844         struct btrfs_iget_args *args = opaque;
3845         return args->ino == btrfs_ino(inode) &&
3846                 args->root == BTRFS_I(inode)->root;
3847 }
3848
3849 static struct inode *btrfs_iget_locked(struct super_block *s,
3850                                        u64 objectid,
3851                                        struct btrfs_root *root)
3852 {
3853         struct inode *inode;
3854         struct btrfs_iget_args args;
3855         args.ino = objectid;
3856         args.root = root;
3857
3858         inode = iget5_locked(s, objectid, btrfs_find_actor,
3859                              btrfs_init_locked_inode,
3860                              (void *)&args);
3861         return inode;
3862 }
3863
3864 /* Get an inode object given its location and corresponding root.
3865  * Returns in *is_new if the inode was read from disk
3866  */
3867 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3868                          struct btrfs_root *root, int *new)
3869 {
3870         struct inode *inode;
3871
3872         inode = btrfs_iget_locked(s, location->objectid, root);
3873         if (!inode)
3874                 return ERR_PTR(-ENOMEM);
3875
3876         if (inode->i_state & I_NEW) {
3877                 BTRFS_I(inode)->root = root;
3878                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3879                 btrfs_read_locked_inode(inode);
3880                 if (!is_bad_inode(inode)) {
3881                         inode_tree_add(inode);
3882                         unlock_new_inode(inode);
3883                         if (new)
3884                                 *new = 1;
3885                 } else {
3886                         unlock_new_inode(inode);
3887                         iput(inode);
3888                         inode = ERR_PTR(-ESTALE);
3889                 }
3890         }
3891
3892         return inode;
3893 }
3894
3895 static struct inode *new_simple_dir(struct super_block *s,
3896                                     struct btrfs_key *key,
3897                                     struct btrfs_root *root)
3898 {
3899         struct inode *inode = new_inode(s);
3900
3901         if (!inode)
3902                 return ERR_PTR(-ENOMEM);
3903
3904         BTRFS_I(inode)->root = root;
3905         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3906         BTRFS_I(inode)->dummy_inode = 1;
3907
3908         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3909         inode->i_op = &simple_dir_inode_operations;
3910         inode->i_fop = &simple_dir_operations;
3911         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3912         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3913
3914         return inode;
3915 }
3916
3917 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3918 {
3919         struct inode *inode;
3920         struct btrfs_root *root = BTRFS_I(dir)->root;
3921         struct btrfs_root *sub_root = root;
3922         struct btrfs_key location;
3923         int index;
3924         int ret = 0;
3925
3926         if (dentry->d_name.len > BTRFS_NAME_LEN)
3927                 return ERR_PTR(-ENAMETOOLONG);
3928
3929         if (unlikely(d_need_lookup(dentry))) {
3930                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
3931                 kfree(dentry->d_fsdata);
3932                 dentry->d_fsdata = NULL;
3933                 /* This thing is hashed, drop it for now */
3934                 d_drop(dentry);
3935         } else {
3936                 ret = btrfs_inode_by_name(dir, dentry, &location);
3937         }
3938
3939         if (ret < 0)
3940                 return ERR_PTR(ret);
3941
3942         if (location.objectid == 0)
3943                 return NULL;
3944
3945         if (location.type == BTRFS_INODE_ITEM_KEY) {
3946                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
3947                 return inode;
3948         }
3949
3950         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
3951
3952         index = srcu_read_lock(&root->fs_info->subvol_srcu);
3953         ret = fixup_tree_root_location(root, dir, dentry,
3954                                        &location, &sub_root);
3955         if (ret < 0) {
3956                 if (ret != -ENOENT)
3957                         inode = ERR_PTR(ret);
3958                 else
3959                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
3960         } else {
3961                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
3962         }
3963         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
3964
3965         if (!IS_ERR(inode) && root != sub_root) {
3966                 down_read(&root->fs_info->cleanup_work_sem);
3967                 if (!(inode->i_sb->s_flags & MS_RDONLY))
3968                         ret = btrfs_orphan_cleanup(sub_root);
3969                 up_read(&root->fs_info->cleanup_work_sem);
3970                 if (ret)
3971                         inode = ERR_PTR(ret);
3972         }
3973
3974         return inode;
3975 }
3976
3977 static int btrfs_dentry_delete(const struct dentry *dentry)
3978 {
3979         struct btrfs_root *root;
3980
3981         if (!dentry->d_inode && !IS_ROOT(dentry))
3982                 dentry = dentry->d_parent;
3983
3984         if (dentry->d_inode) {
3985                 root = BTRFS_I(dentry->d_inode)->root;
3986                 if (btrfs_root_refs(&root->root_item) == 0)
3987                         return 1;
3988         }
3989         return 0;
3990 }
3991
3992 static void btrfs_dentry_release(struct dentry *dentry)
3993 {
3994         if (dentry->d_fsdata)
3995                 kfree(dentry->d_fsdata);
3996 }
3997
3998 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
3999                                    struct nameidata *nd)
4000 {
4001         struct dentry *ret;
4002
4003         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4004         if (unlikely(d_need_lookup(dentry))) {
4005                 spin_lock(&dentry->d_lock);
4006                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4007                 spin_unlock(&dentry->d_lock);
4008         }
4009         return ret;
4010 }
4011
4012 unsigned char btrfs_filetype_table[] = {
4013         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4014 };
4015
4016 static int btrfs_real_readdir(struct file *filp, void *dirent,
4017                               filldir_t filldir)
4018 {
4019         struct inode *inode = filp->f_dentry->d_inode;
4020         struct btrfs_root *root = BTRFS_I(inode)->root;
4021         struct btrfs_item *item;
4022         struct btrfs_dir_item *di;
4023         struct btrfs_key key;
4024         struct btrfs_key found_key;
4025         struct btrfs_path *path;
4026         struct list_head ins_list;
4027         struct list_head del_list;
4028         struct qstr q;
4029         int ret;
4030         struct extent_buffer *leaf;
4031         int slot;
4032         unsigned char d_type;
4033         int over = 0;
4034         u32 di_cur;
4035         u32 di_total;
4036         u32 di_len;
4037         int key_type = BTRFS_DIR_INDEX_KEY;
4038         char tmp_name[32];
4039         char *name_ptr;
4040         int name_len;
4041         int is_curr = 0;        /* filp->f_pos points to the current index? */
4042
4043         /* FIXME, use a real flag for deciding about the key type */
4044         if (root->fs_info->tree_root == root)
4045                 key_type = BTRFS_DIR_ITEM_KEY;
4046
4047         /* special case for "." */
4048         if (filp->f_pos == 0) {
4049                 over = filldir(dirent, ".", 1,
4050                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4051                 if (over)
4052                         return 0;
4053                 filp->f_pos = 1;
4054         }
4055         /* special case for .., just use the back ref */
4056         if (filp->f_pos == 1) {
4057                 u64 pino = parent_ino(filp->f_path.dentry);
4058                 over = filldir(dirent, "..", 2,
4059                                filp->f_pos, pino, DT_DIR);
4060                 if (over)
4061                         return 0;
4062                 filp->f_pos = 2;
4063         }
4064         path = btrfs_alloc_path();
4065         if (!path)
4066                 return -ENOMEM;
4067
4068         path->reada = 1;
4069
4070         if (key_type == BTRFS_DIR_INDEX_KEY) {
4071                 INIT_LIST_HEAD(&ins_list);
4072                 INIT_LIST_HEAD(&del_list);
4073                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4074         }
4075
4076         btrfs_set_key_type(&key, key_type);
4077         key.offset = filp->f_pos;
4078         key.objectid = btrfs_ino(inode);
4079
4080         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4081         if (ret < 0)
4082                 goto err;
4083
4084         while (1) {
4085                 leaf = path->nodes[0];
4086                 slot = path->slots[0];
4087                 if (slot >= btrfs_header_nritems(leaf)) {
4088                         ret = btrfs_next_leaf(root, path);
4089                         if (ret < 0)
4090                                 goto err;
4091                         else if (ret > 0)
4092                                 break;
4093                         continue;
4094                 }
4095
4096                 item = btrfs_item_nr(leaf, slot);
4097                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4098
4099                 if (found_key.objectid != key.objectid)
4100                         break;
4101                 if (btrfs_key_type(&found_key) != key_type)
4102                         break;
4103                 if (found_key.offset < filp->f_pos)
4104                         goto next;
4105                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4106                     btrfs_should_delete_dir_index(&del_list,
4107                                                   found_key.offset))
4108                         goto next;
4109
4110                 filp->f_pos = found_key.offset;
4111                 is_curr = 1;
4112
4113                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4114                 di_cur = 0;
4115                 di_total = btrfs_item_size(leaf, item);
4116
4117                 while (di_cur < di_total) {
4118                         struct btrfs_key location;
4119                         struct dentry *tmp;
4120
4121                         if (verify_dir_item(root, leaf, di))
4122                                 break;
4123
4124                         name_len = btrfs_dir_name_len(leaf, di);
4125                         if (name_len <= sizeof(tmp_name)) {
4126                                 name_ptr = tmp_name;
4127                         } else {
4128                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4129                                 if (!name_ptr) {
4130                                         ret = -ENOMEM;
4131                                         goto err;
4132                                 }
4133                         }
4134                         read_extent_buffer(leaf, name_ptr,
4135                                            (unsigned long)(di + 1), name_len);
4136
4137                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4138                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4139
4140                         q.name = name_ptr;
4141                         q.len = name_len;
4142                         q.hash = full_name_hash(q.name, q.len);
4143                         tmp = d_lookup(filp->f_dentry, &q);
4144                         if (!tmp) {
4145                                 struct btrfs_key *newkey;
4146
4147                                 newkey = kzalloc(sizeof(struct btrfs_key),
4148                                                  GFP_NOFS);
4149                                 if (!newkey)
4150                                         goto no_dentry;
4151                                 tmp = d_alloc(filp->f_dentry, &q);
4152                                 if (!tmp) {
4153                                         kfree(newkey);
4154                                         dput(tmp);
4155                                         goto no_dentry;
4156                                 }
4157                                 memcpy(newkey, &location,
4158                                        sizeof(struct btrfs_key));
4159                                 tmp->d_fsdata = newkey;
4160                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4161                                 d_rehash(tmp);
4162                                 dput(tmp);
4163                         } else {
4164                                 dput(tmp);
4165                         }
4166 no_dentry:
4167                         /* is this a reference to our own snapshot? If so
4168                          * skip it
4169                          */
4170                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4171                             location.objectid == root->root_key.objectid) {
4172                                 over = 0;
4173                                 goto skip;
4174                         }
4175                         over = filldir(dirent, name_ptr, name_len,
4176                                        found_key.offset, location.objectid,
4177                                        d_type);
4178
4179 skip:
4180                         if (name_ptr != tmp_name)
4181                                 kfree(name_ptr);
4182
4183                         if (over)
4184                                 goto nopos;
4185                         di_len = btrfs_dir_name_len(leaf, di) +
4186                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4187                         di_cur += di_len;
4188                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4189                 }
4190 next:
4191                 path->slots[0]++;
4192         }
4193
4194         if (key_type == BTRFS_DIR_INDEX_KEY) {
4195                 if (is_curr)
4196                         filp->f_pos++;
4197                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4198                                                       &ins_list);
4199                 if (ret)
4200                         goto nopos;
4201         }
4202
4203         /* Reached end of directory/root. Bump pos past the last item. */
4204         if (key_type == BTRFS_DIR_INDEX_KEY)
4205                 /*
4206                  * 32-bit glibc will use getdents64, but then strtol -
4207                  * so the last number we can serve is this.
4208                  */
4209                 filp->f_pos = 0x7fffffff;
4210         else
4211                 filp->f_pos++;
4212 nopos:
4213         ret = 0;
4214 err:
4215         if (key_type == BTRFS_DIR_INDEX_KEY)
4216                 btrfs_put_delayed_items(&ins_list, &del_list);
4217         btrfs_free_path(path);
4218         return ret;
4219 }
4220
4221 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4222 {
4223         struct btrfs_root *root = BTRFS_I(inode)->root;
4224         struct btrfs_trans_handle *trans;
4225         int ret = 0;
4226         bool nolock = false;
4227
4228         if (BTRFS_I(inode)->dummy_inode)
4229                 return 0;
4230
4231         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4232                 nolock = true;
4233
4234         if (wbc->sync_mode == WB_SYNC_ALL) {
4235                 if (nolock)
4236                         trans = btrfs_join_transaction_nolock(root);
4237                 else
4238                         trans = btrfs_join_transaction(root);
4239                 if (IS_ERR(trans))
4240                         return PTR_ERR(trans);
4241                 if (nolock)
4242                         ret = btrfs_end_transaction_nolock(trans, root);
4243                 else
4244                         ret = btrfs_commit_transaction(trans, root);
4245         }
4246         return ret;
4247 }
4248
4249 /*
4250  * This is somewhat expensive, updating the tree every time the
4251  * inode changes.  But, it is most likely to find the inode in cache.
4252  * FIXME, needs more benchmarking...there are no reasons other than performance
4253  * to keep or drop this code.
4254  */
4255 int btrfs_dirty_inode(struct inode *inode)
4256 {
4257         struct btrfs_root *root = BTRFS_I(inode)->root;
4258         struct btrfs_trans_handle *trans;
4259         int ret;
4260
4261         if (BTRFS_I(inode)->dummy_inode)
4262                 return 0;
4263
4264         trans = btrfs_join_transaction(root);
4265         if (IS_ERR(trans))
4266                 return PTR_ERR(trans);
4267
4268         ret = btrfs_update_inode(trans, root, inode);
4269         if (ret && ret == -ENOSPC) {
4270                 /* whoops, lets try again with the full transaction */
4271                 btrfs_end_transaction(trans, root);
4272                 trans = btrfs_start_transaction(root, 1);
4273                 if (IS_ERR(trans))
4274                         return PTR_ERR(trans);
4275
4276                 ret = btrfs_update_inode(trans, root, inode);
4277         }
4278         btrfs_end_transaction(trans, root);
4279         if (BTRFS_I(inode)->delayed_node)
4280                 btrfs_balance_delayed_items(root);
4281
4282         return ret;
4283 }
4284
4285 /*
4286  * This is a copy of file_update_time.  We need this so we can return error on
4287  * ENOSPC for updating the inode in the case of file write and mmap writes.
4288  */
4289 int btrfs_update_time(struct file *file)
4290 {
4291         struct inode *inode = file->f_path.dentry->d_inode;
4292         struct timespec now;
4293         int ret;
4294         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4295
4296         /* First try to exhaust all avenues to not sync */
4297         if (IS_NOCMTIME(inode))
4298                 return 0;
4299
4300         now = current_fs_time(inode->i_sb);
4301         if (!timespec_equal(&inode->i_mtime, &now))
4302                 sync_it = S_MTIME;
4303
4304         if (!timespec_equal(&inode->i_ctime, &now))
4305                 sync_it |= S_CTIME;
4306
4307         if (IS_I_VERSION(inode))
4308                 sync_it |= S_VERSION;
4309
4310         if (!sync_it)
4311                 return 0;
4312
4313         /* Finally allowed to write? Takes lock. */
4314         if (mnt_want_write_file(file))
4315                 return 0;
4316
4317         /* Only change inode inside the lock region */
4318         if (sync_it & S_VERSION)
4319                 inode_inc_iversion(inode);
4320         if (sync_it & S_CTIME)
4321                 inode->i_ctime = now;
4322         if (sync_it & S_MTIME)
4323                 inode->i_mtime = now;
4324         ret = btrfs_dirty_inode(inode);
4325         if (!ret)
4326                 mark_inode_dirty_sync(inode);
4327         mnt_drop_write(file->f_path.mnt);
4328         return ret;
4329 }
4330
4331 /*
4332  * find the highest existing sequence number in a directory
4333  * and then set the in-memory index_cnt variable to reflect
4334  * free sequence numbers
4335  */
4336 static int btrfs_set_inode_index_count(struct inode *inode)
4337 {
4338         struct btrfs_root *root = BTRFS_I(inode)->root;
4339         struct btrfs_key key, found_key;
4340         struct btrfs_path *path;
4341         struct extent_buffer *leaf;
4342         int ret;
4343
4344         key.objectid = btrfs_ino(inode);
4345         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4346         key.offset = (u64)-1;
4347
4348         path = btrfs_alloc_path();
4349         if (!path)
4350                 return -ENOMEM;
4351
4352         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4353         if (ret < 0)
4354                 goto out;
4355         /* FIXME: we should be able to handle this */
4356         if (ret == 0)
4357                 goto out;
4358         ret = 0;
4359
4360         /*
4361          * MAGIC NUMBER EXPLANATION:
4362          * since we search a directory based on f_pos we have to start at 2
4363          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4364          * else has to start at 2
4365          */
4366         if (path->slots[0] == 0) {
4367                 BTRFS_I(inode)->index_cnt = 2;
4368                 goto out;
4369         }
4370
4371         path->slots[0]--;
4372
4373         leaf = path->nodes[0];
4374         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4375
4376         if (found_key.objectid != btrfs_ino(inode) ||
4377             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4378                 BTRFS_I(inode)->index_cnt = 2;
4379                 goto out;
4380         }
4381
4382         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4383 out:
4384         btrfs_free_path(path);
4385         return ret;
4386 }
4387
4388 /*
4389  * helper to find a free sequence number in a given directory.  This current
4390  * code is very simple, later versions will do smarter things in the btree
4391  */
4392 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4393 {
4394         int ret = 0;
4395
4396         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4397                 ret = btrfs_inode_delayed_dir_index_count(dir);
4398                 if (ret) {
4399                         ret = btrfs_set_inode_index_count(dir);
4400                         if (ret)
4401                                 return ret;
4402                 }
4403         }
4404
4405         *index = BTRFS_I(dir)->index_cnt;
4406         BTRFS_I(dir)->index_cnt++;
4407
4408         return ret;
4409 }
4410
4411 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4412                                      struct btrfs_root *root,
4413                                      struct inode *dir,
4414                                      const char *name, int name_len,
4415                                      u64 ref_objectid, u64 objectid, int mode,
4416                                      u64 *index)
4417 {
4418         struct inode *inode;
4419         struct btrfs_inode_item *inode_item;
4420         struct btrfs_key *location;
4421         struct btrfs_path *path;
4422         struct btrfs_inode_ref *ref;
4423         struct btrfs_key key[2];
4424         u32 sizes[2];
4425         unsigned long ptr;
4426         int ret;
4427         int owner;
4428
4429         path = btrfs_alloc_path();
4430         if (!path)
4431                 return ERR_PTR(-ENOMEM);
4432
4433         inode = new_inode(root->fs_info->sb);
4434         if (!inode) {
4435                 btrfs_free_path(path);
4436                 return ERR_PTR(-ENOMEM);
4437         }
4438
4439         /*
4440          * we have to initialize this early, so we can reclaim the inode
4441          * number if we fail afterwards in this function.
4442          */
4443         inode->i_ino = objectid;
4444
4445         if (dir) {
4446                 trace_btrfs_inode_request(dir);
4447
4448                 ret = btrfs_set_inode_index(dir, index);
4449                 if (ret) {
4450                         btrfs_free_path(path);
4451                         iput(inode);
4452                         return ERR_PTR(ret);
4453                 }
4454         }
4455         /*
4456          * index_cnt is ignored for everything but a dir,
4457          * btrfs_get_inode_index_count has an explanation for the magic
4458          * number
4459          */
4460         BTRFS_I(inode)->index_cnt = 2;
4461         BTRFS_I(inode)->root = root;
4462         BTRFS_I(inode)->generation = trans->transid;
4463         inode->i_generation = BTRFS_I(inode)->generation;
4464         btrfs_set_inode_space_info(root, inode);
4465
4466         if (S_ISDIR(mode))
4467                 owner = 0;
4468         else
4469                 owner = 1;
4470
4471         key[0].objectid = objectid;
4472         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4473         key[0].offset = 0;
4474
4475         key[1].objectid = objectid;
4476         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4477         key[1].offset = ref_objectid;
4478
4479         sizes[0] = sizeof(struct btrfs_inode_item);
4480         sizes[1] = name_len + sizeof(*ref);
4481
4482         path->leave_spinning = 1;
4483         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4484         if (ret != 0)
4485                 goto fail;
4486
4487         inode_init_owner(inode, dir, mode);
4488         inode_set_bytes(inode, 0);
4489         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4490         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4491                                   struct btrfs_inode_item);
4492         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4493
4494         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4495                              struct btrfs_inode_ref);
4496         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4497         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4498         ptr = (unsigned long)(ref + 1);
4499         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4500
4501         btrfs_mark_buffer_dirty(path->nodes[0]);
4502         btrfs_free_path(path);
4503
4504         location = &BTRFS_I(inode)->location;
4505         location->objectid = objectid;
4506         location->offset = 0;
4507         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4508
4509         btrfs_inherit_iflags(inode, dir);
4510
4511         if (S_ISREG(mode)) {
4512                 if (btrfs_test_opt(root, NODATASUM))
4513                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4514                 if (btrfs_test_opt(root, NODATACOW) ||
4515                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4516                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4517         }
4518
4519         insert_inode_hash(inode);
4520         inode_tree_add(inode);
4521
4522         trace_btrfs_inode_new(inode);
4523         btrfs_set_inode_last_trans(trans, inode);
4524
4525         return inode;
4526 fail:
4527         if (dir)
4528                 BTRFS_I(dir)->index_cnt--;
4529         btrfs_free_path(path);
4530         iput(inode);
4531         return ERR_PTR(ret);
4532 }
4533
4534 static inline u8 btrfs_inode_type(struct inode *inode)
4535 {
4536         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4537 }
4538
4539 /*
4540  * utility function to add 'inode' into 'parent_inode' with
4541  * a give name and a given sequence number.
4542  * if 'add_backref' is true, also insert a backref from the
4543  * inode to the parent directory.
4544  */
4545 int btrfs_add_link(struct btrfs_trans_handle *trans,
4546                    struct inode *parent_inode, struct inode *inode,
4547                    const char *name, int name_len, int add_backref, u64 index)
4548 {
4549         int ret = 0;
4550         struct btrfs_key key;
4551         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4552         u64 ino = btrfs_ino(inode);
4553         u64 parent_ino = btrfs_ino(parent_inode);
4554
4555         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4556                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4557         } else {
4558                 key.objectid = ino;
4559                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4560                 key.offset = 0;
4561         }
4562
4563         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4564                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4565                                          key.objectid, root->root_key.objectid,
4566                                          parent_ino, index, name, name_len);
4567         } else if (add_backref) {
4568                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4569                                              parent_ino, index);
4570         }
4571
4572         if (ret == 0) {
4573                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4574                                             parent_inode, &key,
4575                                             btrfs_inode_type(inode), index);
4576                 BUG_ON(ret);
4577
4578                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4579                                    name_len * 2);
4580                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4581                 ret = btrfs_update_inode(trans, root, parent_inode);
4582         }
4583         return ret;
4584 }
4585
4586 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4587                             struct inode *dir, struct dentry *dentry,
4588                             struct inode *inode, int backref, u64 index)
4589 {
4590         int err = btrfs_add_link(trans, dir, inode,
4591                                  dentry->d_name.name, dentry->d_name.len,
4592                                  backref, index);
4593         if (!err) {
4594                 d_instantiate(dentry, inode);
4595                 return 0;
4596         }
4597         if (err > 0)
4598                 err = -EEXIST;
4599         return err;
4600 }
4601
4602 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4603                         int mode, dev_t rdev)
4604 {
4605         struct btrfs_trans_handle *trans;
4606         struct btrfs_root *root = BTRFS_I(dir)->root;
4607         struct inode *inode = NULL;
4608         int err;
4609         int drop_inode = 0;
4610         u64 objectid;
4611         unsigned long nr = 0;
4612         u64 index = 0;
4613
4614         if (!new_valid_dev(rdev))
4615                 return -EINVAL;
4616
4617         /*
4618          * 2 for inode item and ref
4619          * 2 for dir items
4620          * 1 for xattr if selinux is on
4621          */
4622         trans = btrfs_start_transaction(root, 5);
4623         if (IS_ERR(trans))
4624                 return PTR_ERR(trans);
4625
4626         err = btrfs_find_free_ino(root, &objectid);
4627         if (err)
4628                 goto out_unlock;
4629
4630         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4631                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4632                                 mode, &index);
4633         if (IS_ERR(inode)) {
4634                 err = PTR_ERR(inode);
4635                 goto out_unlock;
4636         }
4637
4638         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4639         if (err) {
4640                 drop_inode = 1;
4641                 goto out_unlock;
4642         }
4643
4644         /*
4645         * If the active LSM wants to access the inode during
4646         * d_instantiate it needs these. Smack checks to see
4647         * if the filesystem supports xattrs by looking at the
4648         * ops vector.
4649         */
4650
4651         inode->i_op = &btrfs_special_inode_operations;
4652         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4653         if (err)
4654                 drop_inode = 1;
4655         else {
4656                 init_special_inode(inode, inode->i_mode, rdev);
4657                 btrfs_update_inode(trans, root, inode);
4658         }
4659 out_unlock:
4660         nr = trans->blocks_used;
4661         btrfs_end_transaction_throttle(trans, root);
4662         btrfs_btree_balance_dirty(root, nr);
4663         if (drop_inode) {
4664                 inode_dec_link_count(inode);
4665                 iput(inode);
4666         }
4667         return err;
4668 }
4669
4670 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4671                         int mode, struct nameidata *nd)
4672 {
4673         struct btrfs_trans_handle *trans;
4674         struct btrfs_root *root = BTRFS_I(dir)->root;
4675         struct inode *inode = NULL;
4676         int drop_inode = 0;
4677         int err;
4678         unsigned long nr = 0;
4679         u64 objectid;
4680         u64 index = 0;
4681
4682         /*
4683          * 2 for inode item and ref
4684          * 2 for dir items
4685          * 1 for xattr if selinux is on
4686          */
4687         trans = btrfs_start_transaction(root, 5);
4688         if (IS_ERR(trans))
4689                 return PTR_ERR(trans);
4690
4691         err = btrfs_find_free_ino(root, &objectid);
4692         if (err)
4693                 goto out_unlock;
4694
4695         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4696                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4697                                 mode, &index);
4698         if (IS_ERR(inode)) {
4699                 err = PTR_ERR(inode);
4700                 goto out_unlock;
4701         }
4702
4703         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4704         if (err) {
4705                 drop_inode = 1;
4706                 goto out_unlock;
4707         }
4708
4709         /*
4710         * If the active LSM wants to access the inode during
4711         * d_instantiate it needs these. Smack checks to see
4712         * if the filesystem supports xattrs by looking at the
4713         * ops vector.
4714         */
4715         inode->i_fop = &btrfs_file_operations;
4716         inode->i_op = &btrfs_file_inode_operations;
4717
4718         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4719         if (err)
4720                 drop_inode = 1;
4721         else {
4722                 inode->i_mapping->a_ops = &btrfs_aops;
4723                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4724                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4725         }
4726 out_unlock:
4727         nr = trans->blocks_used;
4728         btrfs_end_transaction_throttle(trans, root);
4729         if (drop_inode) {
4730                 inode_dec_link_count(inode);
4731                 iput(inode);
4732         }
4733         btrfs_btree_balance_dirty(root, nr);
4734         return err;
4735 }
4736
4737 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4738                       struct dentry *dentry)
4739 {
4740         struct btrfs_trans_handle *trans;
4741         struct btrfs_root *root = BTRFS_I(dir)->root;
4742         struct inode *inode = old_dentry->d_inode;
4743         u64 index;
4744         unsigned long nr = 0;
4745         int err;
4746         int drop_inode = 0;
4747
4748         /* do not allow sys_link's with other subvols of the same device */
4749         if (root->objectid != BTRFS_I(inode)->root->objectid)
4750                 return -EXDEV;
4751
4752         if (inode->i_nlink == ~0U)
4753                 return -EMLINK;
4754
4755         err = btrfs_set_inode_index(dir, &index);
4756         if (err)
4757                 goto fail;
4758
4759         /*
4760          * 2 items for inode and inode ref
4761          * 2 items for dir items
4762          * 1 item for parent inode
4763          */
4764         trans = btrfs_start_transaction(root, 5);
4765         if (IS_ERR(trans)) {
4766                 err = PTR_ERR(trans);
4767                 goto fail;
4768         }
4769
4770         btrfs_inc_nlink(inode);
4771         inode->i_ctime = CURRENT_TIME;
4772         ihold(inode);
4773
4774         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4775
4776         if (err) {
4777                 drop_inode = 1;
4778         } else {
4779                 struct dentry *parent = dentry->d_parent;
4780                 err = btrfs_update_inode(trans, root, inode);
4781                 BUG_ON(err);
4782                 btrfs_log_new_name(trans, inode, NULL, parent);
4783         }
4784
4785         nr = trans->blocks_used;
4786         btrfs_end_transaction_throttle(trans, root);
4787 fail:
4788         if (drop_inode) {
4789                 inode_dec_link_count(inode);
4790                 iput(inode);
4791         }
4792         btrfs_btree_balance_dirty(root, nr);
4793         return err;
4794 }
4795
4796 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4797 {
4798         struct inode *inode = NULL;
4799         struct btrfs_trans_handle *trans;
4800         struct btrfs_root *root = BTRFS_I(dir)->root;
4801         int err = 0;
4802         int drop_on_err = 0;
4803         u64 objectid = 0;
4804         u64 index = 0;
4805         unsigned long nr = 1;
4806
4807         /*
4808          * 2 items for inode and ref
4809          * 2 items for dir items
4810          * 1 for xattr if selinux is on
4811          */
4812         trans = btrfs_start_transaction(root, 5);
4813         if (IS_ERR(trans))
4814                 return PTR_ERR(trans);
4815
4816         err = btrfs_find_free_ino(root, &objectid);
4817         if (err)
4818                 goto out_fail;
4819
4820         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4821                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4822                                 S_IFDIR | mode, &index);
4823         if (IS_ERR(inode)) {
4824                 err = PTR_ERR(inode);
4825                 goto out_fail;
4826         }
4827
4828         drop_on_err = 1;
4829
4830         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4831         if (err)
4832                 goto out_fail;
4833
4834         inode->i_op = &btrfs_dir_inode_operations;
4835         inode->i_fop = &btrfs_dir_file_operations;
4836
4837         btrfs_i_size_write(inode, 0);
4838         err = btrfs_update_inode(trans, root, inode);
4839         if (err)
4840                 goto out_fail;
4841
4842         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4843                              dentry->d_name.len, 0, index);
4844         if (err)
4845                 goto out_fail;
4846
4847         d_instantiate(dentry, inode);
4848         drop_on_err = 0;
4849
4850 out_fail:
4851         nr = trans->blocks_used;
4852         btrfs_end_transaction_throttle(trans, root);
4853         if (drop_on_err)
4854                 iput(inode);
4855         btrfs_btree_balance_dirty(root, nr);
4856         return err;
4857 }
4858
4859 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4860  * and an extent that you want to insert, deal with overlap and insert
4861  * the new extent into the tree.
4862  */
4863 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4864                                 struct extent_map *existing,
4865                                 struct extent_map *em,
4866                                 u64 map_start, u64 map_len)
4867 {
4868         u64 start_diff;
4869
4870         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4871         start_diff = map_start - em->start;
4872         em->start = map_start;
4873         em->len = map_len;
4874         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4875             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4876                 em->block_start += start_diff;
4877                 em->block_len -= start_diff;
4878         }
4879         return add_extent_mapping(em_tree, em);
4880 }
4881
4882 static noinline int uncompress_inline(struct btrfs_path *path,
4883                                       struct inode *inode, struct page *page,
4884                                       size_t pg_offset, u64 extent_offset,
4885                                       struct btrfs_file_extent_item *item)
4886 {
4887         int ret;
4888         struct extent_buffer *leaf = path->nodes[0];
4889         char *tmp;
4890         size_t max_size;
4891         unsigned long inline_size;
4892         unsigned long ptr;
4893         int compress_type;
4894
4895         WARN_ON(pg_offset != 0);
4896         compress_type = btrfs_file_extent_compression(leaf, item);
4897         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4898         inline_size = btrfs_file_extent_inline_item_len(leaf,
4899                                         btrfs_item_nr(leaf, path->slots[0]));
4900         tmp = kmalloc(inline_size, GFP_NOFS);
4901         if (!tmp)
4902                 return -ENOMEM;
4903         ptr = btrfs_file_extent_inline_start(item);
4904
4905         read_extent_buffer(leaf, tmp, ptr, inline_size);
4906
4907         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4908         ret = btrfs_decompress(compress_type, tmp, page,
4909                                extent_offset, inline_size, max_size);
4910         if (ret) {
4911                 char *kaddr = kmap_atomic(page, KM_USER0);
4912                 unsigned long copy_size = min_t(u64,
4913                                   PAGE_CACHE_SIZE - pg_offset,
4914                                   max_size - extent_offset);
4915                 memset(kaddr + pg_offset, 0, copy_size);
4916                 kunmap_atomic(kaddr, KM_USER0);
4917         }
4918         kfree(tmp);
4919         return 0;
4920 }
4921
4922 /*
4923  * a bit scary, this does extent mapping from logical file offset to the disk.
4924  * the ugly parts come from merging extents from the disk with the in-ram
4925  * representation.  This gets more complex because of the data=ordered code,
4926  * where the in-ram extents might be locked pending data=ordered completion.
4927  *
4928  * This also copies inline extents directly into the page.
4929  */
4930
4931 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4932                                     size_t pg_offset, u64 start, u64 len,
4933                                     int create)
4934 {
4935         int ret;
4936         int err = 0;
4937         u64 bytenr;
4938         u64 extent_start = 0;
4939         u64 extent_end = 0;
4940         u64 objectid = btrfs_ino(inode);
4941         u32 found_type;
4942         struct btrfs_path *path = NULL;
4943         struct btrfs_root *root = BTRFS_I(inode)->root;
4944         struct btrfs_file_extent_item *item;
4945         struct extent_buffer *leaf;
4946         struct btrfs_key found_key;
4947         struct extent_map *em = NULL;
4948         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4949         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4950         struct btrfs_trans_handle *trans = NULL;
4951         int compress_type;
4952
4953 again:
4954         read_lock(&em_tree->lock);
4955         em = lookup_extent_mapping(em_tree, start, len);
4956         if (em)
4957                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4958         read_unlock(&em_tree->lock);
4959
4960         if (em) {
4961                 if (em->start > start || em->start + em->len <= start)
4962                         free_extent_map(em);
4963                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4964                         free_extent_map(em);
4965                 else
4966                         goto out;
4967         }
4968         em = alloc_extent_map();
4969         if (!em) {
4970                 err = -ENOMEM;
4971                 goto out;
4972         }
4973         em->bdev = root->fs_info->fs_devices->latest_bdev;
4974         em->start = EXTENT_MAP_HOLE;
4975         em->orig_start = EXTENT_MAP_HOLE;
4976         em->len = (u64)-1;
4977         em->block_len = (u64)-1;
4978
4979         if (!path) {
4980                 path = btrfs_alloc_path();
4981                 if (!path) {
4982                         err = -ENOMEM;
4983                         goto out;
4984                 }
4985                 /*
4986                  * Chances are we'll be called again, so go ahead and do
4987                  * readahead
4988                  */
4989                 path->reada = 1;
4990         }
4991
4992         ret = btrfs_lookup_file_extent(trans, root, path,
4993                                        objectid, start, trans != NULL);
4994         if (ret < 0) {
4995                 err = ret;
4996                 goto out;
4997         }
4998
4999         if (ret != 0) {
5000                 if (path->slots[0] == 0)
5001                         goto not_found;
5002                 path->slots[0]--;
5003         }
5004
5005         leaf = path->nodes[0];
5006         item = btrfs_item_ptr(leaf, path->slots[0],
5007                               struct btrfs_file_extent_item);
5008         /* are we inside the extent that was found? */
5009         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5010         found_type = btrfs_key_type(&found_key);
5011         if (found_key.objectid != objectid ||
5012             found_type != BTRFS_EXTENT_DATA_KEY) {
5013                 goto not_found;
5014         }
5015
5016         found_type = btrfs_file_extent_type(leaf, item);
5017         extent_start = found_key.offset;
5018         compress_type = btrfs_file_extent_compression(leaf, item);
5019         if (found_type == BTRFS_FILE_EXTENT_REG ||
5020             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5021                 extent_end = extent_start +
5022                        btrfs_file_extent_num_bytes(leaf, item);
5023         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5024                 size_t size;
5025                 size = btrfs_file_extent_inline_len(leaf, item);
5026                 extent_end = (extent_start + size + root->sectorsize - 1) &
5027                         ~((u64)root->sectorsize - 1);
5028         }
5029
5030         if (start >= extent_end) {
5031                 path->slots[0]++;
5032                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5033                         ret = btrfs_next_leaf(root, path);
5034                         if (ret < 0) {
5035                                 err = ret;
5036                                 goto out;
5037                         }
5038                         if (ret > 0)
5039                                 goto not_found;
5040                         leaf = path->nodes[0];
5041                 }
5042                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5043                 if (found_key.objectid != objectid ||
5044                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5045                         goto not_found;
5046                 if (start + len <= found_key.offset)
5047                         goto not_found;
5048                 em->start = start;
5049                 em->len = found_key.offset - start;
5050                 goto not_found_em;
5051         }
5052
5053         if (found_type == BTRFS_FILE_EXTENT_REG ||
5054             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5055                 em->start = extent_start;
5056                 em->len = extent_end - extent_start;
5057                 em->orig_start = extent_start -
5058                                  btrfs_file_extent_offset(leaf, item);
5059                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5060                 if (bytenr == 0) {
5061                         em->block_start = EXTENT_MAP_HOLE;
5062                         goto insert;
5063                 }
5064                 if (compress_type != BTRFS_COMPRESS_NONE) {
5065                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5066                         em->compress_type = compress_type;
5067                         em->block_start = bytenr;
5068                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5069                                                                          item);
5070                 } else {
5071                         bytenr += btrfs_file_extent_offset(leaf, item);
5072                         em->block_start = bytenr;
5073                         em->block_len = em->len;
5074                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5075                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5076                 }
5077                 goto insert;
5078         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5079                 unsigned long ptr;
5080                 char *map;
5081                 size_t size;
5082                 size_t extent_offset;
5083                 size_t copy_size;
5084
5085                 em->block_start = EXTENT_MAP_INLINE;
5086                 if (!page || create) {
5087                         em->start = extent_start;
5088                         em->len = extent_end - extent_start;
5089                         goto out;
5090                 }
5091
5092                 size = btrfs_file_extent_inline_len(leaf, item);
5093                 extent_offset = page_offset(page) + pg_offset - extent_start;
5094                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5095                                 size - extent_offset);
5096                 em->start = extent_start + extent_offset;
5097                 em->len = (copy_size + root->sectorsize - 1) &
5098                         ~((u64)root->sectorsize - 1);
5099                 em->orig_start = EXTENT_MAP_INLINE;
5100                 if (compress_type) {
5101                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5102                         em->compress_type = compress_type;
5103                 }
5104                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5105                 if (create == 0 && !PageUptodate(page)) {
5106                         if (btrfs_file_extent_compression(leaf, item) !=
5107                             BTRFS_COMPRESS_NONE) {
5108                                 ret = uncompress_inline(path, inode, page,
5109                                                         pg_offset,
5110                                                         extent_offset, item);
5111                                 BUG_ON(ret);
5112                         } else {
5113                                 map = kmap(page);
5114                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5115                                                    copy_size);
5116                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5117                                         memset(map + pg_offset + copy_size, 0,
5118                                                PAGE_CACHE_SIZE - pg_offset -
5119                                                copy_size);
5120                                 }
5121                                 kunmap(page);
5122                         }
5123                         flush_dcache_page(page);
5124                 } else if (create && PageUptodate(page)) {
5125                         WARN_ON(1);
5126                         if (!trans) {
5127                                 kunmap(page);
5128                                 free_extent_map(em);
5129                                 em = NULL;
5130
5131                                 btrfs_release_path(path);
5132                                 trans = btrfs_join_transaction(root);
5133
5134                                 if (IS_ERR(trans))
5135                                         return ERR_CAST(trans);
5136                                 goto again;
5137                         }
5138                         map = kmap(page);
5139                         write_extent_buffer(leaf, map + pg_offset, ptr,
5140                                             copy_size);
5141                         kunmap(page);
5142                         btrfs_mark_buffer_dirty(leaf);
5143                 }
5144                 set_extent_uptodate(io_tree, em->start,
5145                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5146                 goto insert;
5147         } else {
5148                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5149                 WARN_ON(1);
5150         }
5151 not_found:
5152         em->start = start;
5153         em->len = len;
5154 not_found_em:
5155         em->block_start = EXTENT_MAP_HOLE;
5156         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5157 insert:
5158         btrfs_release_path(path);
5159         if (em->start > start || extent_map_end(em) <= start) {
5160                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5161                        "[%llu %llu]\n", (unsigned long long)em->start,
5162                        (unsigned long long)em->len,
5163                        (unsigned long long)start,
5164                        (unsigned long long)len);
5165                 err = -EIO;
5166                 goto out;
5167         }
5168
5169         err = 0;
5170         write_lock(&em_tree->lock);
5171         ret = add_extent_mapping(em_tree, em);
5172         /* it is possible that someone inserted the extent into the tree
5173          * while we had the lock dropped.  It is also possible that
5174          * an overlapping map exists in the tree
5175          */
5176         if (ret == -EEXIST) {
5177                 struct extent_map *existing;
5178
5179                 ret = 0;
5180
5181                 existing = lookup_extent_mapping(em_tree, start, len);
5182                 if (existing && (existing->start > start ||
5183                     existing->start + existing->len <= start)) {
5184                         free_extent_map(existing);
5185                         existing = NULL;
5186                 }
5187                 if (!existing) {
5188                         existing = lookup_extent_mapping(em_tree, em->start,
5189                                                          em->len);
5190                         if (existing) {
5191                                 err = merge_extent_mapping(em_tree, existing,
5192                                                            em, start,
5193                                                            root->sectorsize);
5194                                 free_extent_map(existing);
5195                                 if (err) {
5196                                         free_extent_map(em);
5197                                         em = NULL;
5198                                 }
5199                         } else {
5200                                 err = -EIO;
5201                                 free_extent_map(em);
5202                                 em = NULL;
5203                         }
5204                 } else {
5205                         free_extent_map(em);
5206                         em = existing;
5207                         err = 0;
5208                 }
5209         }
5210         write_unlock(&em_tree->lock);
5211 out:
5212
5213         trace_btrfs_get_extent(root, em);
5214
5215         if (path)
5216                 btrfs_free_path(path);
5217         if (trans) {
5218                 ret = btrfs_end_transaction(trans, root);
5219                 if (!err)
5220                         err = ret;
5221         }
5222         if (err) {
5223                 free_extent_map(em);
5224                 return ERR_PTR(err);
5225         }
5226         return em;
5227 }
5228
5229 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5230                                            size_t pg_offset, u64 start, u64 len,
5231                                            int create)
5232 {
5233         struct extent_map *em;
5234         struct extent_map *hole_em = NULL;
5235         u64 range_start = start;
5236         u64 end;
5237         u64 found;
5238         u64 found_end;
5239         int err = 0;
5240
5241         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5242         if (IS_ERR(em))
5243                 return em;
5244         if (em) {
5245                 /*
5246                  * if our em maps to a hole, there might
5247                  * actually be delalloc bytes behind it
5248                  */
5249                 if (em->block_start != EXTENT_MAP_HOLE)
5250                         return em;
5251                 else
5252                         hole_em = em;
5253         }
5254
5255         /* check to see if we've wrapped (len == -1 or similar) */
5256         end = start + len;
5257         if (end < start)
5258                 end = (u64)-1;
5259         else
5260                 end -= 1;
5261
5262         em = NULL;
5263
5264         /* ok, we didn't find anything, lets look for delalloc */
5265         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5266                                  end, len, EXTENT_DELALLOC, 1);
5267         found_end = range_start + found;
5268         if (found_end < range_start)
5269                 found_end = (u64)-1;
5270
5271         /*
5272          * we didn't find anything useful, return
5273          * the original results from get_extent()
5274          */
5275         if (range_start > end || found_end <= start) {
5276                 em = hole_em;
5277                 hole_em = NULL;
5278                 goto out;
5279         }
5280
5281         /* adjust the range_start to make sure it doesn't
5282          * go backwards from the start they passed in
5283          */
5284         range_start = max(start,range_start);
5285         found = found_end - range_start;
5286
5287         if (found > 0) {
5288                 u64 hole_start = start;
5289                 u64 hole_len = len;
5290
5291                 em = alloc_extent_map();
5292                 if (!em) {
5293                         err = -ENOMEM;
5294                         goto out;
5295                 }
5296                 /*
5297                  * when btrfs_get_extent can't find anything it
5298                  * returns one huge hole
5299                  *
5300                  * make sure what it found really fits our range, and
5301                  * adjust to make sure it is based on the start from
5302                  * the caller
5303                  */
5304                 if (hole_em) {
5305                         u64 calc_end = extent_map_end(hole_em);
5306
5307                         if (calc_end <= start || (hole_em->start > end)) {
5308                                 free_extent_map(hole_em);
5309                                 hole_em = NULL;
5310                         } else {
5311                                 hole_start = max(hole_em->start, start);
5312                                 hole_len = calc_end - hole_start;
5313                         }
5314                 }
5315                 em->bdev = NULL;
5316                 if (hole_em && range_start > hole_start) {
5317                         /* our hole starts before our delalloc, so we
5318                          * have to return just the parts of the hole
5319                          * that go until  the delalloc starts
5320                          */
5321                         em->len = min(hole_len,
5322                                       range_start - hole_start);
5323                         em->start = hole_start;
5324                         em->orig_start = hole_start;
5325                         /*
5326                          * don't adjust block start at all,
5327                          * it is fixed at EXTENT_MAP_HOLE
5328                          */
5329                         em->block_start = hole_em->block_start;
5330                         em->block_len = hole_len;
5331                 } else {
5332                         em->start = range_start;
5333                         em->len = found;
5334                         em->orig_start = range_start;
5335                         em->block_start = EXTENT_MAP_DELALLOC;
5336                         em->block_len = found;
5337                 }
5338         } else if (hole_em) {
5339                 return hole_em;
5340         }
5341 out:
5342
5343         free_extent_map(hole_em);
5344         if (err) {
5345                 free_extent_map(em);
5346                 return ERR_PTR(err);
5347         }
5348         return em;
5349 }
5350
5351 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5352                                                   struct extent_map *em,
5353                                                   u64 start, u64 len)
5354 {
5355         struct btrfs_root *root = BTRFS_I(inode)->root;
5356         struct btrfs_trans_handle *trans;
5357         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5358         struct btrfs_key ins;
5359         u64 alloc_hint;
5360         int ret;
5361         bool insert = false;
5362
5363         /*
5364          * Ok if the extent map we looked up is a hole and is for the exact
5365          * range we want, there is no reason to allocate a new one, however if
5366          * it is not right then we need to free this one and drop the cache for
5367          * our range.
5368          */
5369         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5370             em->len != len) {
5371                 free_extent_map(em);
5372                 em = NULL;
5373                 insert = true;
5374                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5375         }
5376
5377         trans = btrfs_join_transaction(root);
5378         if (IS_ERR(trans))
5379                 return ERR_CAST(trans);
5380
5381         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5382                 btrfs_add_inode_defrag(trans, inode);
5383
5384         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5385
5386         alloc_hint = get_extent_allocation_hint(inode, start, len);
5387         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5388                                    alloc_hint, (u64)-1, &ins, 1);
5389         if (ret) {
5390                 em = ERR_PTR(ret);
5391                 goto out;
5392         }
5393
5394         if (!em) {
5395                 em = alloc_extent_map();
5396                 if (!em) {
5397                         em = ERR_PTR(-ENOMEM);
5398                         goto out;
5399                 }
5400         }
5401
5402         em->start = start;
5403         em->orig_start = em->start;
5404         em->len = ins.offset;
5405
5406         em->block_start = ins.objectid;
5407         em->block_len = ins.offset;
5408         em->bdev = root->fs_info->fs_devices->latest_bdev;
5409
5410         /*
5411          * We need to do this because if we're using the original em we searched
5412          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5413          */
5414         em->flags = 0;
5415         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5416
5417         while (insert) {
5418                 write_lock(&em_tree->lock);
5419                 ret = add_extent_mapping(em_tree, em);
5420                 write_unlock(&em_tree->lock);
5421                 if (ret != -EEXIST)
5422                         break;
5423                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5424         }
5425
5426         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5427                                            ins.offset, ins.offset, 0);
5428         if (ret) {
5429                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5430                 em = ERR_PTR(ret);
5431         }
5432 out:
5433         btrfs_end_transaction(trans, root);
5434         return em;
5435 }
5436
5437 /*
5438  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5439  * block must be cow'd
5440  */
5441 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5442                                       struct inode *inode, u64 offset, u64 len)
5443 {
5444         struct btrfs_path *path;
5445         int ret;
5446         struct extent_buffer *leaf;
5447         struct btrfs_root *root = BTRFS_I(inode)->root;
5448         struct btrfs_file_extent_item *fi;
5449         struct btrfs_key key;
5450         u64 disk_bytenr;
5451         u64 backref_offset;
5452         u64 extent_end;
5453         u64 num_bytes;
5454         int slot;
5455         int found_type;
5456
5457         path = btrfs_alloc_path();
5458         if (!path)
5459                 return -ENOMEM;
5460
5461         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5462                                        offset, 0);
5463         if (ret < 0)
5464                 goto out;
5465
5466         slot = path->slots[0];
5467         if (ret == 1) {
5468                 if (slot == 0) {
5469                         /* can't find the item, must cow */
5470                         ret = 0;
5471                         goto out;
5472                 }
5473                 slot--;
5474         }
5475         ret = 0;
5476         leaf = path->nodes[0];
5477         btrfs_item_key_to_cpu(leaf, &key, slot);
5478         if (key.objectid != btrfs_ino(inode) ||
5479             key.type != BTRFS_EXTENT_DATA_KEY) {
5480                 /* not our file or wrong item type, must cow */
5481                 goto out;
5482         }
5483
5484         if (key.offset > offset) {
5485                 /* Wrong offset, must cow */
5486                 goto out;
5487         }
5488
5489         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5490         found_type = btrfs_file_extent_type(leaf, fi);
5491         if (found_type != BTRFS_FILE_EXTENT_REG &&
5492             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5493                 /* not a regular extent, must cow */
5494                 goto out;
5495         }
5496         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5497         backref_offset = btrfs_file_extent_offset(leaf, fi);
5498
5499         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5500         if (extent_end < offset + len) {
5501                 /* extent doesn't include our full range, must cow */
5502                 goto out;
5503         }
5504
5505         if (btrfs_extent_readonly(root, disk_bytenr))
5506                 goto out;
5507
5508         /*
5509          * look for other files referencing this extent, if we
5510          * find any we must cow
5511          */
5512         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5513                                   key.offset - backref_offset, disk_bytenr))
5514                 goto out;
5515
5516         /*
5517          * adjust disk_bytenr and num_bytes to cover just the bytes
5518          * in this extent we are about to write.  If there
5519          * are any csums in that range we have to cow in order
5520          * to keep the csums correct
5521          */
5522         disk_bytenr += backref_offset;
5523         disk_bytenr += offset - key.offset;
5524         num_bytes = min(offset + len, extent_end) - offset;
5525         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5526                                 goto out;
5527         /*
5528          * all of the above have passed, it is safe to overwrite this extent
5529          * without cow
5530          */
5531         ret = 1;
5532 out:
5533         btrfs_free_path(path);
5534         return ret;
5535 }
5536
5537 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5538                                    struct buffer_head *bh_result, int create)
5539 {
5540         struct extent_map *em;
5541         struct btrfs_root *root = BTRFS_I(inode)->root;
5542         u64 start = iblock << inode->i_blkbits;
5543         u64 len = bh_result->b_size;
5544         struct btrfs_trans_handle *trans;
5545
5546         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5547         if (IS_ERR(em))
5548                 return PTR_ERR(em);
5549
5550         /*
5551          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5552          * io.  INLINE is special, and we could probably kludge it in here, but
5553          * it's still buffered so for safety lets just fall back to the generic
5554          * buffered path.
5555          *
5556          * For COMPRESSED we _have_ to read the entire extent in so we can
5557          * decompress it, so there will be buffering required no matter what we
5558          * do, so go ahead and fallback to buffered.
5559          *
5560          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5561          * to buffered IO.  Don't blame me, this is the price we pay for using
5562          * the generic code.
5563          */
5564         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5565             em->block_start == EXTENT_MAP_INLINE) {
5566                 free_extent_map(em);
5567                 return -ENOTBLK;
5568         }
5569
5570         /* Just a good old fashioned hole, return */
5571         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5572                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5573                 free_extent_map(em);
5574                 /* DIO will do one hole at a time, so just unlock a sector */
5575                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5576                               start + root->sectorsize - 1, GFP_NOFS);
5577                 return 0;
5578         }
5579
5580         /*
5581          * We don't allocate a new extent in the following cases
5582          *
5583          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5584          * existing extent.
5585          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5586          * just use the extent.
5587          *
5588          */
5589         if (!create) {
5590                 len = em->len - (start - em->start);
5591                 goto map;
5592         }
5593
5594         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5595             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5596              em->block_start != EXTENT_MAP_HOLE)) {
5597                 int type;
5598                 int ret;
5599                 u64 block_start;
5600
5601                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5602                         type = BTRFS_ORDERED_PREALLOC;
5603                 else
5604                         type = BTRFS_ORDERED_NOCOW;
5605                 len = min(len, em->len - (start - em->start));
5606                 block_start = em->block_start + (start - em->start);
5607
5608                 /*
5609                  * we're not going to log anything, but we do need
5610                  * to make sure the current transaction stays open
5611                  * while we look for nocow cross refs
5612                  */
5613                 trans = btrfs_join_transaction(root);
5614                 if (IS_ERR(trans))
5615                         goto must_cow;
5616
5617                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5618                         ret = btrfs_add_ordered_extent_dio(inode, start,
5619                                            block_start, len, len, type);
5620                         btrfs_end_transaction(trans, root);
5621                         if (ret) {
5622                                 free_extent_map(em);
5623                                 return ret;
5624                         }
5625                         goto unlock;
5626                 }
5627                 btrfs_end_transaction(trans, root);
5628         }
5629 must_cow:
5630         /*
5631          * this will cow the extent, reset the len in case we changed
5632          * it above
5633          */
5634         len = bh_result->b_size;
5635         em = btrfs_new_extent_direct(inode, em, start, len);
5636         if (IS_ERR(em))
5637                 return PTR_ERR(em);
5638         len = min(len, em->len - (start - em->start));
5639 unlock:
5640         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5641                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5642                           0, NULL, GFP_NOFS);
5643 map:
5644         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5645                 inode->i_blkbits;
5646         bh_result->b_size = len;
5647         bh_result->b_bdev = em->bdev;
5648         set_buffer_mapped(bh_result);
5649         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5650                 set_buffer_new(bh_result);
5651
5652         free_extent_map(em);
5653
5654         return 0;
5655 }
5656
5657 struct btrfs_dio_private {
5658         struct inode *inode;
5659         u64 logical_offset;
5660         u64 disk_bytenr;
5661         u64 bytes;
5662         u32 *csums;
5663         void *private;
5664
5665         /* number of bios pending for this dio */
5666         atomic_t pending_bios;
5667
5668         /* IO errors */
5669         int errors;
5670
5671         struct bio *orig_bio;
5672 };
5673
5674 static void btrfs_endio_direct_read(struct bio *bio, int err)
5675 {
5676         struct btrfs_dio_private *dip = bio->bi_private;
5677         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5678         struct bio_vec *bvec = bio->bi_io_vec;
5679         struct inode *inode = dip->inode;
5680         struct btrfs_root *root = BTRFS_I(inode)->root;
5681         u64 start;
5682         u32 *private = dip->csums;
5683
5684         start = dip->logical_offset;
5685         do {
5686                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5687                         struct page *page = bvec->bv_page;
5688                         char *kaddr;
5689                         u32 csum = ~(u32)0;
5690                         unsigned long flags;
5691
5692                         local_irq_save(flags);
5693                         kaddr = kmap_atomic(page, KM_IRQ0);
5694                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5695                                                csum, bvec->bv_len);
5696                         btrfs_csum_final(csum, (char *)&csum);
5697                         kunmap_atomic(kaddr, KM_IRQ0);
5698                         local_irq_restore(flags);
5699
5700                         flush_dcache_page(bvec->bv_page);
5701                         if (csum != *private) {
5702                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5703                                       " %llu csum %u private %u\n",
5704                                       (unsigned long long)btrfs_ino(inode),
5705                                       (unsigned long long)start,
5706                                       csum, *private);
5707                                 err = -EIO;
5708                         }
5709                 }
5710
5711                 start += bvec->bv_len;
5712                 private++;
5713                 bvec++;
5714         } while (bvec <= bvec_end);
5715
5716         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5717                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5718         bio->bi_private = dip->private;
5719
5720         kfree(dip->csums);
5721         kfree(dip);
5722
5723         /* If we had a csum failure make sure to clear the uptodate flag */
5724         if (err)
5725                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5726         dio_end_io(bio, err);
5727 }
5728
5729 static void btrfs_endio_direct_write(struct bio *bio, int err)
5730 {
5731         struct btrfs_dio_private *dip = bio->bi_private;
5732         struct inode *inode = dip->inode;
5733         struct btrfs_root *root = BTRFS_I(inode)->root;
5734         struct btrfs_trans_handle *trans;
5735         struct btrfs_ordered_extent *ordered = NULL;
5736         struct extent_state *cached_state = NULL;
5737         u64 ordered_offset = dip->logical_offset;
5738         u64 ordered_bytes = dip->bytes;
5739         int ret;
5740
5741         if (err)
5742                 goto out_done;
5743 again:
5744         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5745                                                    &ordered_offset,
5746                                                    ordered_bytes);
5747         if (!ret)
5748                 goto out_test;
5749
5750         BUG_ON(!ordered);
5751
5752         trans = btrfs_join_transaction(root);
5753         if (IS_ERR(trans)) {
5754                 err = -ENOMEM;
5755                 goto out;
5756         }
5757         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5758
5759         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5760                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5761                 if (!ret)
5762                         err = btrfs_update_inode_fallback(trans, root, inode);
5763                 goto out;
5764         }
5765
5766         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5767                          ordered->file_offset + ordered->len - 1, 0,
5768                          &cached_state, GFP_NOFS);
5769
5770         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5771                 ret = btrfs_mark_extent_written(trans, inode,
5772                                                 ordered->file_offset,
5773                                                 ordered->file_offset +
5774                                                 ordered->len);
5775                 if (ret) {
5776                         err = ret;
5777                         goto out_unlock;
5778                 }
5779         } else {
5780                 ret = insert_reserved_file_extent(trans, inode,
5781                                                   ordered->file_offset,
5782                                                   ordered->start,
5783                                                   ordered->disk_len,
5784                                                   ordered->len,
5785                                                   ordered->len,
5786                                                   0, 0, 0,
5787                                                   BTRFS_FILE_EXTENT_REG);
5788                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5789                                    ordered->file_offset, ordered->len);
5790                 if (ret) {
5791                         err = ret;
5792                         WARN_ON(1);
5793                         goto out_unlock;
5794                 }
5795         }
5796
5797         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5798         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5799         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5800                 btrfs_update_inode_fallback(trans, root, inode);
5801         ret = 0;
5802 out_unlock:
5803         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5804                              ordered->file_offset + ordered->len - 1,
5805                              &cached_state, GFP_NOFS);
5806 out:
5807         btrfs_delalloc_release_metadata(inode, ordered->len);
5808         btrfs_end_transaction(trans, root);
5809         ordered_offset = ordered->file_offset + ordered->len;
5810         btrfs_put_ordered_extent(ordered);
5811         btrfs_put_ordered_extent(ordered);
5812
5813 out_test:
5814         /*
5815          * our bio might span multiple ordered extents.  If we haven't
5816          * completed the accounting for the whole dio, go back and try again
5817          */
5818         if (ordered_offset < dip->logical_offset + dip->bytes) {
5819                 ordered_bytes = dip->logical_offset + dip->bytes -
5820                         ordered_offset;
5821                 goto again;
5822         }
5823 out_done:
5824         bio->bi_private = dip->private;
5825
5826         kfree(dip->csums);
5827         kfree(dip);
5828
5829         /* If we had an error make sure to clear the uptodate flag */
5830         if (err)
5831                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5832         dio_end_io(bio, err);
5833 }
5834
5835 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5836                                     struct bio *bio, int mirror_num,
5837                                     unsigned long bio_flags, u64 offset)
5838 {
5839         int ret;
5840         struct btrfs_root *root = BTRFS_I(inode)->root;
5841         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5842         BUG_ON(ret);
5843         return 0;
5844 }
5845
5846 static void btrfs_end_dio_bio(struct bio *bio, int err)
5847 {
5848         struct btrfs_dio_private *dip = bio->bi_private;
5849
5850         if (err) {
5851                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5852                       "sector %#Lx len %u err no %d\n",
5853                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5854                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5855                 dip->errors = 1;
5856
5857                 /*
5858                  * before atomic variable goto zero, we must make sure
5859                  * dip->errors is perceived to be set.
5860                  */
5861                 smp_mb__before_atomic_dec();
5862         }
5863
5864         /* if there are more bios still pending for this dio, just exit */
5865         if (!atomic_dec_and_test(&dip->pending_bios))
5866                 goto out;
5867
5868         if (dip->errors)
5869                 bio_io_error(dip->orig_bio);
5870         else {
5871                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5872                 bio_endio(dip->orig_bio, 0);
5873         }
5874 out:
5875         bio_put(bio);
5876 }
5877
5878 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5879                                        u64 first_sector, gfp_t gfp_flags)
5880 {
5881         int nr_vecs = bio_get_nr_vecs(bdev);
5882         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5883 }
5884
5885 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5886                                          int rw, u64 file_offset, int skip_sum,
5887                                          u32 *csums, int async_submit)
5888 {
5889         int write = rw & REQ_WRITE;
5890         struct btrfs_root *root = BTRFS_I(inode)->root;
5891         int ret;
5892
5893         bio_get(bio);
5894         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5895         if (ret)
5896                 goto err;
5897
5898         if (skip_sum)
5899                 goto map;
5900
5901         if (write && async_submit) {
5902                 ret = btrfs_wq_submit_bio(root->fs_info,
5903                                    inode, rw, bio, 0, 0,
5904                                    file_offset,
5905                                    __btrfs_submit_bio_start_direct_io,
5906                                    __btrfs_submit_bio_done);
5907                 goto err;
5908         } else if (write) {
5909                 /*
5910                  * If we aren't doing async submit, calculate the csum of the
5911                  * bio now.
5912                  */
5913                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5914                 if (ret)
5915                         goto err;
5916         } else if (!skip_sum) {
5917                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5918                                           file_offset, csums);
5919                 if (ret)
5920                         goto err;
5921         }
5922
5923 map:
5924         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5925 err:
5926         bio_put(bio);
5927         return ret;
5928 }
5929
5930 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5931                                     int skip_sum)
5932 {
5933         struct inode *inode = dip->inode;
5934         struct btrfs_root *root = BTRFS_I(inode)->root;
5935         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5936         struct bio *bio;
5937         struct bio *orig_bio = dip->orig_bio;
5938         struct bio_vec *bvec = orig_bio->bi_io_vec;
5939         u64 start_sector = orig_bio->bi_sector;
5940         u64 file_offset = dip->logical_offset;
5941         u64 submit_len = 0;
5942         u64 map_length;
5943         int nr_pages = 0;
5944         u32 *csums = dip->csums;
5945         int ret = 0;
5946         int async_submit = 0;
5947         int write = rw & REQ_WRITE;
5948
5949         map_length = orig_bio->bi_size;
5950         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5951                               &map_length, NULL, 0);
5952         if (ret) {
5953                 bio_put(orig_bio);
5954                 return -EIO;
5955         }
5956
5957         if (map_length >= orig_bio->bi_size) {
5958                 bio = orig_bio;
5959                 goto submit;
5960         }
5961
5962         async_submit = 1;
5963         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5964         if (!bio)
5965                 return -ENOMEM;
5966         bio->bi_private = dip;
5967         bio->bi_end_io = btrfs_end_dio_bio;
5968         atomic_inc(&dip->pending_bios);
5969
5970         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5971                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5972                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5973                                  bvec->bv_offset) < bvec->bv_len)) {
5974                         /*
5975                          * inc the count before we submit the bio so
5976                          * we know the end IO handler won't happen before
5977                          * we inc the count. Otherwise, the dip might get freed
5978                          * before we're done setting it up
5979                          */
5980                         atomic_inc(&dip->pending_bios);
5981                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5982                                                      file_offset, skip_sum,
5983                                                      csums, async_submit);
5984                         if (ret) {
5985                                 bio_put(bio);
5986                                 atomic_dec(&dip->pending_bios);
5987                                 goto out_err;
5988                         }
5989
5990                         /* Write's use the ordered csums */
5991                         if (!write && !skip_sum)
5992                                 csums = csums + nr_pages;
5993                         start_sector += submit_len >> 9;
5994                         file_offset += submit_len;
5995
5996                         submit_len = 0;
5997                         nr_pages = 0;
5998
5999                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6000                                                   start_sector, GFP_NOFS);
6001                         if (!bio)
6002                                 goto out_err;
6003                         bio->bi_private = dip;
6004                         bio->bi_end_io = btrfs_end_dio_bio;
6005
6006                         map_length = orig_bio->bi_size;
6007                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6008                                               &map_length, NULL, 0);
6009                         if (ret) {
6010                                 bio_put(bio);
6011                                 goto out_err;
6012                         }
6013                 } else {
6014                         submit_len += bvec->bv_len;
6015                         nr_pages ++;
6016                         bvec++;
6017                 }
6018         }
6019
6020 submit:
6021         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6022                                      csums, async_submit);
6023         if (!ret)
6024                 return 0;
6025
6026         bio_put(bio);
6027 out_err:
6028         dip->errors = 1;
6029         /*
6030          * before atomic variable goto zero, we must
6031          * make sure dip->errors is perceived to be set.
6032          */
6033         smp_mb__before_atomic_dec();
6034         if (atomic_dec_and_test(&dip->pending_bios))
6035                 bio_io_error(dip->orig_bio);
6036
6037         /* bio_end_io() will handle error, so we needn't return it */
6038         return 0;
6039 }
6040
6041 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6042                                 loff_t file_offset)
6043 {
6044         struct btrfs_root *root = BTRFS_I(inode)->root;
6045         struct btrfs_dio_private *dip;
6046         struct bio_vec *bvec = bio->bi_io_vec;
6047         int skip_sum;
6048         int write = rw & REQ_WRITE;
6049         int ret = 0;
6050
6051         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6052
6053         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6054         if (!dip) {
6055                 ret = -ENOMEM;
6056                 goto free_ordered;
6057         }
6058         dip->csums = NULL;
6059
6060         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6061         if (!write && !skip_sum) {
6062                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6063                 if (!dip->csums) {
6064                         kfree(dip);
6065                         ret = -ENOMEM;
6066                         goto free_ordered;
6067                 }
6068         }
6069
6070         dip->private = bio->bi_private;
6071         dip->inode = inode;
6072         dip->logical_offset = file_offset;
6073
6074         dip->bytes = 0;
6075         do {
6076                 dip->bytes += bvec->bv_len;
6077                 bvec++;
6078         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6079
6080         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6081         bio->bi_private = dip;
6082         dip->errors = 0;
6083         dip->orig_bio = bio;
6084         atomic_set(&dip->pending_bios, 0);
6085
6086         if (write)
6087                 bio->bi_end_io = btrfs_endio_direct_write;
6088         else
6089                 bio->bi_end_io = btrfs_endio_direct_read;
6090
6091         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6092         if (!ret)
6093                 return;
6094 free_ordered:
6095         /*
6096          * If this is a write, we need to clean up the reserved space and kill
6097          * the ordered extent.
6098          */
6099         if (write) {
6100                 struct btrfs_ordered_extent *ordered;
6101                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6102                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6103                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6104                         btrfs_free_reserved_extent(root, ordered->start,
6105                                                    ordered->disk_len);
6106                 btrfs_put_ordered_extent(ordered);
6107                 btrfs_put_ordered_extent(ordered);
6108         }
6109         bio_endio(bio, ret);
6110 }
6111
6112 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6113                         const struct iovec *iov, loff_t offset,
6114                         unsigned long nr_segs)
6115 {
6116         int seg;
6117         int i;
6118         size_t size;
6119         unsigned long addr;
6120         unsigned blocksize_mask = root->sectorsize - 1;
6121         ssize_t retval = -EINVAL;
6122         loff_t end = offset;
6123
6124         if (offset & blocksize_mask)
6125                 goto out;
6126
6127         /* Check the memory alignment.  Blocks cannot straddle pages */
6128         for (seg = 0; seg < nr_segs; seg++) {
6129                 addr = (unsigned long)iov[seg].iov_base;
6130                 size = iov[seg].iov_len;
6131                 end += size;
6132                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6133                         goto out;
6134
6135                 /* If this is a write we don't need to check anymore */
6136                 if (rw & WRITE)
6137                         continue;
6138
6139                 /*
6140                  * Check to make sure we don't have duplicate iov_base's in this
6141                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6142                  * when reading back.
6143                  */
6144                 for (i = seg + 1; i < nr_segs; i++) {
6145                         if (iov[seg].iov_base == iov[i].iov_base)
6146                                 goto out;
6147                 }
6148         }
6149         retval = 0;
6150 out:
6151         return retval;
6152 }
6153 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6154                         const struct iovec *iov, loff_t offset,
6155                         unsigned long nr_segs)
6156 {
6157         struct file *file = iocb->ki_filp;
6158         struct inode *inode = file->f_mapping->host;
6159         struct btrfs_ordered_extent *ordered;
6160         struct extent_state *cached_state = NULL;
6161         u64 lockstart, lockend;
6162         ssize_t ret;
6163         int writing = rw & WRITE;
6164         int write_bits = 0;
6165         size_t count = iov_length(iov, nr_segs);
6166
6167         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6168                             offset, nr_segs)) {
6169                 return 0;
6170         }
6171
6172         lockstart = offset;
6173         lockend = offset + count - 1;
6174
6175         if (writing) {
6176                 ret = btrfs_delalloc_reserve_space(inode, count);
6177                 if (ret)
6178                         goto out;
6179         }
6180
6181         while (1) {
6182                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6183                                  0, &cached_state, GFP_NOFS);
6184                 /*
6185                  * We're concerned with the entire range that we're going to be
6186                  * doing DIO to, so we need to make sure theres no ordered
6187                  * extents in this range.
6188                  */
6189                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6190                                                      lockend - lockstart + 1);
6191                 if (!ordered)
6192                         break;
6193                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6194                                      &cached_state, GFP_NOFS);
6195                 btrfs_start_ordered_extent(inode, ordered, 1);
6196                 btrfs_put_ordered_extent(ordered);
6197                 cond_resched();
6198         }
6199
6200         /*
6201          * we don't use btrfs_set_extent_delalloc because we don't want
6202          * the dirty or uptodate bits
6203          */
6204         if (writing) {
6205                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6206                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6207                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6208                                      GFP_NOFS);
6209                 if (ret) {
6210                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6211                                          lockend, EXTENT_LOCKED | write_bits,
6212                                          1, 0, &cached_state, GFP_NOFS);
6213                         goto out;
6214                 }
6215         }
6216
6217         free_extent_state(cached_state);
6218         cached_state = NULL;
6219
6220         ret = __blockdev_direct_IO(rw, iocb, inode,
6221                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6222                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6223                    btrfs_submit_direct, 0);
6224
6225         if (ret < 0 && ret != -EIOCBQUEUED) {
6226                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6227                               offset + iov_length(iov, nr_segs) - 1,
6228                               EXTENT_LOCKED | write_bits, 1, 0,
6229                               &cached_state, GFP_NOFS);
6230         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6231                 /*
6232                  * We're falling back to buffered, unlock the section we didn't
6233                  * do IO on.
6234                  */
6235                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6236                               offset + iov_length(iov, nr_segs) - 1,
6237                               EXTENT_LOCKED | write_bits, 1, 0,
6238                               &cached_state, GFP_NOFS);
6239         }
6240 out:
6241         free_extent_state(cached_state);
6242         return ret;
6243 }
6244
6245 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6246                 __u64 start, __u64 len)
6247 {
6248         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6249 }
6250
6251 int btrfs_readpage(struct file *file, struct page *page)
6252 {
6253         struct extent_io_tree *tree;
6254         tree = &BTRFS_I(page->mapping->host)->io_tree;
6255         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6256 }
6257
6258 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6259 {
6260         struct extent_io_tree *tree;
6261
6262
6263         if (current->flags & PF_MEMALLOC) {
6264                 redirty_page_for_writepage(wbc, page);
6265                 unlock_page(page);
6266                 return 0;
6267         }
6268         tree = &BTRFS_I(page->mapping->host)->io_tree;
6269         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6270 }
6271
6272 int btrfs_writepages(struct address_space *mapping,
6273                      struct writeback_control *wbc)
6274 {
6275         struct extent_io_tree *tree;
6276
6277         tree = &BTRFS_I(mapping->host)->io_tree;
6278         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6279 }
6280
6281 static int
6282 btrfs_readpages(struct file *file, struct address_space *mapping,
6283                 struct list_head *pages, unsigned nr_pages)
6284 {
6285         struct extent_io_tree *tree;
6286         tree = &BTRFS_I(mapping->host)->io_tree;
6287         return extent_readpages(tree, mapping, pages, nr_pages,
6288                                 btrfs_get_extent);
6289 }
6290 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6291 {
6292         struct extent_io_tree *tree;
6293         struct extent_map_tree *map;
6294         int ret;
6295
6296         tree = &BTRFS_I(page->mapping->host)->io_tree;
6297         map = &BTRFS_I(page->mapping->host)->extent_tree;
6298         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6299         if (ret == 1) {
6300                 ClearPagePrivate(page);
6301                 set_page_private(page, 0);
6302                 page_cache_release(page);
6303         }
6304         return ret;
6305 }
6306
6307 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6308 {
6309         if (PageWriteback(page) || PageDirty(page))
6310                 return 0;
6311         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6312 }
6313
6314 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6315 {
6316         struct extent_io_tree *tree;
6317         struct btrfs_ordered_extent *ordered;
6318         struct extent_state *cached_state = NULL;
6319         u64 page_start = page_offset(page);
6320         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6321
6322
6323         /*
6324          * we have the page locked, so new writeback can't start,
6325          * and the dirty bit won't be cleared while we are here.
6326          *
6327          * Wait for IO on this page so that we can safely clear
6328          * the PagePrivate2 bit and do ordered accounting
6329          */
6330         wait_on_page_writeback(page);
6331
6332         tree = &BTRFS_I(page->mapping->host)->io_tree;
6333         if (offset) {
6334                 btrfs_releasepage(page, GFP_NOFS);
6335                 return;
6336         }
6337         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6338                          GFP_NOFS);
6339         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6340                                            page_offset(page));
6341         if (ordered) {
6342                 /*
6343                  * IO on this page will never be started, so we need
6344                  * to account for any ordered extents now
6345                  */
6346                 clear_extent_bit(tree, page_start, page_end,
6347                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6348                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6349                                  &cached_state, GFP_NOFS);
6350                 /*
6351                  * whoever cleared the private bit is responsible
6352                  * for the finish_ordered_io
6353                  */
6354                 if (TestClearPagePrivate2(page)) {
6355                         btrfs_finish_ordered_io(page->mapping->host,
6356                                                 page_start, page_end);
6357                 }
6358                 btrfs_put_ordered_extent(ordered);
6359                 cached_state = NULL;
6360                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6361                                  GFP_NOFS);
6362         }
6363         clear_extent_bit(tree, page_start, page_end,
6364                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6365                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6366         __btrfs_releasepage(page, GFP_NOFS);
6367
6368         ClearPageChecked(page);
6369         if (PagePrivate(page)) {
6370                 ClearPagePrivate(page);
6371                 set_page_private(page, 0);
6372                 page_cache_release(page);
6373         }
6374 }
6375
6376 /*
6377  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6378  * called from a page fault handler when a page is first dirtied. Hence we must
6379  * be careful to check for EOF conditions here. We set the page up correctly
6380  * for a written page which means we get ENOSPC checking when writing into
6381  * holes and correct delalloc and unwritten extent mapping on filesystems that
6382  * support these features.
6383  *
6384  * We are not allowed to take the i_mutex here so we have to play games to
6385  * protect against truncate races as the page could now be beyond EOF.  Because
6386  * vmtruncate() writes the inode size before removing pages, once we have the
6387  * page lock we can determine safely if the page is beyond EOF. If it is not
6388  * beyond EOF, then the page is guaranteed safe against truncation until we
6389  * unlock the page.
6390  */
6391 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6392 {
6393         struct page *page = vmf->page;
6394         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6395         struct btrfs_root *root = BTRFS_I(inode)->root;
6396         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6397         struct btrfs_ordered_extent *ordered;
6398         struct extent_state *cached_state = NULL;
6399         char *kaddr;
6400         unsigned long zero_start;
6401         loff_t size;
6402         int ret;
6403         u64 page_start;
6404         u64 page_end;
6405
6406         /* Need this to keep space reservations serialized */
6407         mutex_lock(&inode->i_mutex);
6408         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6409         mutex_unlock(&inode->i_mutex);
6410         if (!ret)
6411                 ret = btrfs_update_time(vma->vm_file);
6412         if (ret) {
6413                 if (ret == -ENOMEM)
6414                         ret = VM_FAULT_OOM;
6415                 else /* -ENOSPC, -EIO, etc */
6416                         ret = VM_FAULT_SIGBUS;
6417                 goto out;
6418         }
6419
6420         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6421 again:
6422         lock_page(page);
6423         size = i_size_read(inode);
6424         page_start = page_offset(page);
6425         page_end = page_start + PAGE_CACHE_SIZE - 1;
6426
6427         if ((page->mapping != inode->i_mapping) ||
6428             (page_start >= size)) {
6429                 /* page got truncated out from underneath us */
6430                 goto out_unlock;
6431         }
6432         wait_on_page_writeback(page);
6433
6434         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6435                          GFP_NOFS);
6436         set_page_extent_mapped(page);
6437
6438         /*
6439          * we can't set the delalloc bits if there are pending ordered
6440          * extents.  Drop our locks and wait for them to finish
6441          */
6442         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6443         if (ordered) {
6444                 unlock_extent_cached(io_tree, page_start, page_end,
6445                                      &cached_state, GFP_NOFS);
6446                 unlock_page(page);
6447                 btrfs_start_ordered_extent(inode, ordered, 1);
6448                 btrfs_put_ordered_extent(ordered);
6449                 goto again;
6450         }
6451
6452         /*
6453          * XXX - page_mkwrite gets called every time the page is dirtied, even
6454          * if it was already dirty, so for space accounting reasons we need to
6455          * clear any delalloc bits for the range we are fixing to save.  There
6456          * is probably a better way to do this, but for now keep consistent with
6457          * prepare_pages in the normal write path.
6458          */
6459         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6460                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6461                           0, 0, &cached_state, GFP_NOFS);
6462
6463         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6464                                         &cached_state);
6465         if (ret) {
6466                 unlock_extent_cached(io_tree, page_start, page_end,
6467                                      &cached_state, GFP_NOFS);
6468                 ret = VM_FAULT_SIGBUS;
6469                 goto out_unlock;
6470         }
6471         ret = 0;
6472
6473         /* page is wholly or partially inside EOF */
6474         if (page_start + PAGE_CACHE_SIZE > size)
6475                 zero_start = size & ~PAGE_CACHE_MASK;
6476         else
6477                 zero_start = PAGE_CACHE_SIZE;
6478
6479         if (zero_start != PAGE_CACHE_SIZE) {
6480                 kaddr = kmap(page);
6481                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6482                 flush_dcache_page(page);
6483                 kunmap(page);
6484         }
6485         ClearPageChecked(page);
6486         set_page_dirty(page);
6487         SetPageUptodate(page);
6488
6489         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6490         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6491
6492         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6493
6494 out_unlock:
6495         if (!ret)
6496                 return VM_FAULT_LOCKED;
6497         unlock_page(page);
6498         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6499 out:
6500         return ret;
6501 }
6502
6503 static int btrfs_truncate(struct inode *inode)
6504 {
6505         struct btrfs_root *root = BTRFS_I(inode)->root;
6506         struct btrfs_block_rsv *rsv;
6507         int ret;
6508         int err = 0;
6509         struct btrfs_trans_handle *trans;
6510         unsigned long nr;
6511         u64 mask = root->sectorsize - 1;
6512         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6513
6514         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6515         if (ret)
6516                 return ret;
6517
6518         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6519         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6520
6521         /*
6522          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6523          * 3 things going on here
6524          *
6525          * 1) We need to reserve space for our orphan item and the space to
6526          * delete our orphan item.  Lord knows we don't want to have a dangling
6527          * orphan item because we didn't reserve space to remove it.
6528          *
6529          * 2) We need to reserve space to update our inode.
6530          *
6531          * 3) We need to have something to cache all the space that is going to
6532          * be free'd up by the truncate operation, but also have some slack
6533          * space reserved in case it uses space during the truncate (thank you
6534          * very much snapshotting).
6535          *
6536          * And we need these to all be seperate.  The fact is we can use alot of
6537          * space doing the truncate, and we have no earthly idea how much space
6538          * we will use, so we need the truncate reservation to be seperate so it
6539          * doesn't end up using space reserved for updating the inode or
6540          * removing the orphan item.  We also need to be able to stop the
6541          * transaction and start a new one, which means we need to be able to
6542          * update the inode several times, and we have no idea of knowing how
6543          * many times that will be, so we can't just reserve 1 item for the
6544          * entirety of the opration, so that has to be done seperately as well.
6545          * Then there is the orphan item, which does indeed need to be held on
6546          * to for the whole operation, and we need nobody to touch this reserved
6547          * space except the orphan code.
6548          *
6549          * So that leaves us with
6550          *
6551          * 1) root->orphan_block_rsv - for the orphan deletion.
6552          * 2) rsv - for the truncate reservation, which we will steal from the
6553          * transaction reservation.
6554          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6555          * updating the inode.
6556          */
6557         rsv = btrfs_alloc_block_rsv(root);
6558         if (!rsv)
6559                 return -ENOMEM;
6560         rsv->size = min_size;
6561
6562         /*
6563          * 1 for the truncate slack space
6564          * 1 for the orphan item we're going to add
6565          * 1 for the orphan item deletion
6566          * 1 for updating the inode.
6567          */
6568         trans = btrfs_start_transaction(root, 4);
6569         if (IS_ERR(trans)) {
6570                 err = PTR_ERR(trans);
6571                 goto out;
6572         }
6573
6574         /* Migrate the slack space for the truncate to our reserve */
6575         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6576                                       min_size);
6577         BUG_ON(ret);
6578
6579         ret = btrfs_orphan_add(trans, inode);
6580         if (ret) {
6581                 btrfs_end_transaction(trans, root);
6582                 goto out;
6583         }
6584
6585         /*
6586          * setattr is responsible for setting the ordered_data_close flag,
6587          * but that is only tested during the last file release.  That
6588          * could happen well after the next commit, leaving a great big
6589          * window where new writes may get lost if someone chooses to write
6590          * to this file after truncating to zero
6591          *
6592          * The inode doesn't have any dirty data here, and so if we commit
6593          * this is a noop.  If someone immediately starts writing to the inode
6594          * it is very likely we'll catch some of their writes in this
6595          * transaction, and the commit will find this file on the ordered
6596          * data list with good things to send down.
6597          *
6598          * This is a best effort solution, there is still a window where
6599          * using truncate to replace the contents of the file will
6600          * end up with a zero length file after a crash.
6601          */
6602         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6603                 btrfs_add_ordered_operation(trans, root, inode);
6604
6605         while (1) {
6606                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6607                 if (ret) {
6608                         /*
6609                          * This can only happen with the original transaction we
6610                          * started above, every other time we shouldn't have a
6611                          * transaction started yet.
6612                          */
6613                         if (ret == -EAGAIN)
6614                                 goto end_trans;
6615                         err = ret;
6616                         break;
6617                 }
6618
6619                 if (!trans) {
6620                         /* Just need the 1 for updating the inode */
6621                         trans = btrfs_start_transaction(root, 1);
6622                         if (IS_ERR(trans)) {
6623                                 ret = err = PTR_ERR(trans);
6624                                 trans = NULL;
6625                                 break;
6626                         }
6627                 }
6628
6629                 trans->block_rsv = rsv;
6630
6631                 ret = btrfs_truncate_inode_items(trans, root, inode,
6632                                                  inode->i_size,
6633                                                  BTRFS_EXTENT_DATA_KEY);
6634                 if (ret != -EAGAIN) {
6635                         err = ret;
6636                         break;
6637                 }
6638
6639                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6640                 ret = btrfs_update_inode(trans, root, inode);
6641                 if (ret) {
6642                         err = ret;
6643                         break;
6644                 }
6645 end_trans:
6646                 nr = trans->blocks_used;
6647                 btrfs_end_transaction(trans, root);
6648                 trans = NULL;
6649                 btrfs_btree_balance_dirty(root, nr);
6650         }
6651
6652         if (ret == 0 && inode->i_nlink > 0) {
6653                 trans->block_rsv = root->orphan_block_rsv;
6654                 ret = btrfs_orphan_del(trans, inode);
6655                 if (ret)
6656                         err = ret;
6657         } else if (ret && inode->i_nlink > 0) {
6658                 /*
6659                  * Failed to do the truncate, remove us from the in memory
6660                  * orphan list.
6661                  */
6662                 ret = btrfs_orphan_del(NULL, inode);
6663         }
6664
6665         if (trans) {
6666                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6667                 ret = btrfs_update_inode(trans, root, inode);
6668                 if (ret && !err)
6669                         err = ret;
6670
6671                 nr = trans->blocks_used;
6672                 ret = btrfs_end_transaction_throttle(trans, root);
6673                 btrfs_btree_balance_dirty(root, nr);
6674         }
6675
6676 out:
6677         btrfs_free_block_rsv(root, rsv);
6678
6679         if (ret && !err)
6680                 err = ret;
6681
6682         return err;
6683 }
6684
6685 /*
6686  * create a new subvolume directory/inode (helper for the ioctl).
6687  */
6688 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6689                              struct btrfs_root *new_root, u64 new_dirid)
6690 {
6691         struct inode *inode;
6692         int err;
6693         u64 index = 0;
6694
6695         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6696                                 new_dirid, S_IFDIR | 0700, &index);
6697         if (IS_ERR(inode))
6698                 return PTR_ERR(inode);
6699         inode->i_op = &btrfs_dir_inode_operations;
6700         inode->i_fop = &btrfs_dir_file_operations;
6701
6702         set_nlink(inode, 1);
6703         btrfs_i_size_write(inode, 0);
6704
6705         err = btrfs_update_inode(trans, new_root, inode);
6706         BUG_ON(err);
6707
6708         iput(inode);
6709         return 0;
6710 }
6711
6712 struct inode *btrfs_alloc_inode(struct super_block *sb)
6713 {
6714         struct btrfs_inode *ei;
6715         struct inode *inode;
6716
6717         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6718         if (!ei)
6719                 return NULL;
6720
6721         ei->root = NULL;
6722         ei->space_info = NULL;
6723         ei->generation = 0;
6724         ei->sequence = 0;
6725         ei->last_trans = 0;
6726         ei->last_sub_trans = 0;
6727         ei->logged_trans = 0;
6728         ei->delalloc_bytes = 0;
6729         ei->disk_i_size = 0;
6730         ei->flags = 0;
6731         ei->csum_bytes = 0;
6732         ei->index_cnt = (u64)-1;
6733         ei->last_unlink_trans = 0;
6734
6735         spin_lock_init(&ei->lock);
6736         ei->outstanding_extents = 0;
6737         ei->reserved_extents = 0;
6738
6739         ei->ordered_data_close = 0;
6740         ei->orphan_meta_reserved = 0;
6741         ei->dummy_inode = 0;
6742         ei->in_defrag = 0;
6743         ei->delalloc_meta_reserved = 0;
6744         ei->force_compress = BTRFS_COMPRESS_NONE;
6745
6746         ei->delayed_node = NULL;
6747
6748         inode = &ei->vfs_inode;
6749         extent_map_tree_init(&ei->extent_tree);
6750         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6751         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6752         mutex_init(&ei->log_mutex);
6753         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6754         INIT_LIST_HEAD(&ei->i_orphan);
6755         INIT_LIST_HEAD(&ei->delalloc_inodes);
6756         INIT_LIST_HEAD(&ei->ordered_operations);
6757         RB_CLEAR_NODE(&ei->rb_node);
6758
6759         return inode;
6760 }
6761
6762 static void btrfs_i_callback(struct rcu_head *head)
6763 {
6764         struct inode *inode = container_of(head, struct inode, i_rcu);
6765         INIT_LIST_HEAD(&inode->i_dentry);
6766         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6767 }
6768
6769 void btrfs_destroy_inode(struct inode *inode)
6770 {
6771         struct btrfs_ordered_extent *ordered;
6772         struct btrfs_root *root = BTRFS_I(inode)->root;
6773
6774         WARN_ON(!list_empty(&inode->i_dentry));
6775         WARN_ON(inode->i_data.nrpages);
6776         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6777         WARN_ON(BTRFS_I(inode)->reserved_extents);
6778         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6779         WARN_ON(BTRFS_I(inode)->csum_bytes);
6780
6781         /*
6782          * This can happen where we create an inode, but somebody else also
6783          * created the same inode and we need to destroy the one we already
6784          * created.
6785          */
6786         if (!root)
6787                 goto free;
6788
6789         /*
6790          * Make sure we're properly removed from the ordered operation
6791          * lists.
6792          */
6793         smp_mb();
6794         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6795                 spin_lock(&root->fs_info->ordered_extent_lock);
6796                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6797                 spin_unlock(&root->fs_info->ordered_extent_lock);
6798         }
6799
6800         spin_lock(&root->orphan_lock);
6801         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6802                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6803                        (unsigned long long)btrfs_ino(inode));
6804                 list_del_init(&BTRFS_I(inode)->i_orphan);
6805         }
6806         spin_unlock(&root->orphan_lock);
6807
6808         while (1) {
6809                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6810                 if (!ordered)
6811                         break;
6812                 else {
6813                         printk(KERN_ERR "btrfs found ordered "
6814                                "extent %llu %llu on inode cleanup\n",
6815                                (unsigned long long)ordered->file_offset,
6816                                (unsigned long long)ordered->len);
6817                         btrfs_remove_ordered_extent(inode, ordered);
6818                         btrfs_put_ordered_extent(ordered);
6819                         btrfs_put_ordered_extent(ordered);
6820                 }
6821         }
6822         inode_tree_del(inode);
6823         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6824 free:
6825         btrfs_remove_delayed_node(inode);
6826         call_rcu(&inode->i_rcu, btrfs_i_callback);
6827 }
6828
6829 int btrfs_drop_inode(struct inode *inode)
6830 {
6831         struct btrfs_root *root = BTRFS_I(inode)->root;
6832
6833         if (btrfs_root_refs(&root->root_item) == 0 &&
6834             !btrfs_is_free_space_inode(root, inode))
6835                 return 1;
6836         else
6837                 return generic_drop_inode(inode);
6838 }
6839
6840 static void init_once(void *foo)
6841 {
6842         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6843
6844         inode_init_once(&ei->vfs_inode);
6845 }
6846
6847 void btrfs_destroy_cachep(void)
6848 {
6849         if (btrfs_inode_cachep)
6850                 kmem_cache_destroy(btrfs_inode_cachep);
6851         if (btrfs_trans_handle_cachep)
6852                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6853         if (btrfs_transaction_cachep)
6854                 kmem_cache_destroy(btrfs_transaction_cachep);
6855         if (btrfs_path_cachep)
6856                 kmem_cache_destroy(btrfs_path_cachep);
6857         if (btrfs_free_space_cachep)
6858                 kmem_cache_destroy(btrfs_free_space_cachep);
6859 }
6860
6861 int btrfs_init_cachep(void)
6862 {
6863         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6864                         sizeof(struct btrfs_inode), 0,
6865                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6866         if (!btrfs_inode_cachep)
6867                 goto fail;
6868
6869         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6870                         sizeof(struct btrfs_trans_handle), 0,
6871                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6872         if (!btrfs_trans_handle_cachep)
6873                 goto fail;
6874
6875         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6876                         sizeof(struct btrfs_transaction), 0,
6877                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6878         if (!btrfs_transaction_cachep)
6879                 goto fail;
6880
6881         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6882                         sizeof(struct btrfs_path), 0,
6883                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6884         if (!btrfs_path_cachep)
6885                 goto fail;
6886
6887         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6888                         sizeof(struct btrfs_free_space), 0,
6889                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6890         if (!btrfs_free_space_cachep)
6891                 goto fail;
6892
6893         return 0;
6894 fail:
6895         btrfs_destroy_cachep();
6896         return -ENOMEM;
6897 }
6898
6899 static int btrfs_getattr(struct vfsmount *mnt,
6900                          struct dentry *dentry, struct kstat *stat)
6901 {
6902         struct inode *inode = dentry->d_inode;
6903         u32 blocksize = inode->i_sb->s_blocksize;
6904
6905         generic_fillattr(inode, stat);
6906         stat->dev = BTRFS_I(inode)->root->anon_dev;
6907         stat->blksize = PAGE_CACHE_SIZE;
6908         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
6909                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
6910         return 0;
6911 }
6912
6913 /*
6914  * If a file is moved, it will inherit the cow and compression flags of the new
6915  * directory.
6916  */
6917 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6918 {
6919         struct btrfs_inode *b_dir = BTRFS_I(dir);
6920         struct btrfs_inode *b_inode = BTRFS_I(inode);
6921
6922         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6923                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6924         else
6925                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6926
6927         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6928                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6929         else
6930                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6931 }
6932
6933 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6934                            struct inode *new_dir, struct dentry *new_dentry)
6935 {
6936         struct btrfs_trans_handle *trans;
6937         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6938         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6939         struct inode *new_inode = new_dentry->d_inode;
6940         struct inode *old_inode = old_dentry->d_inode;
6941         struct timespec ctime = CURRENT_TIME;
6942         u64 index = 0;
6943         u64 root_objectid;
6944         int ret;
6945         u64 old_ino = btrfs_ino(old_inode);
6946
6947         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6948                 return -EPERM;
6949
6950         /* we only allow rename subvolume link between subvolumes */
6951         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6952                 return -EXDEV;
6953
6954         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6955             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6956                 return -ENOTEMPTY;
6957
6958         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6959             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6960                 return -ENOTEMPTY;
6961         /*
6962          * we're using rename to replace one file with another.
6963          * and the replacement file is large.  Start IO on it now so
6964          * we don't add too much work to the end of the transaction
6965          */
6966         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6967             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6968                 filemap_flush(old_inode->i_mapping);
6969
6970         /* close the racy window with snapshot create/destroy ioctl */
6971         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6972                 down_read(&root->fs_info->subvol_sem);
6973         /*
6974          * We want to reserve the absolute worst case amount of items.  So if
6975          * both inodes are subvols and we need to unlink them then that would
6976          * require 4 item modifications, but if they are both normal inodes it
6977          * would require 5 item modifications, so we'll assume their normal
6978          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6979          * should cover the worst case number of items we'll modify.
6980          */
6981         trans = btrfs_start_transaction(root, 20);
6982         if (IS_ERR(trans)) {
6983                 ret = PTR_ERR(trans);
6984                 goto out_notrans;
6985         }
6986
6987         if (dest != root)
6988                 btrfs_record_root_in_trans(trans, dest);
6989
6990         ret = btrfs_set_inode_index(new_dir, &index);
6991         if (ret)
6992                 goto out_fail;
6993
6994         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6995                 /* force full log commit if subvolume involved. */
6996                 root->fs_info->last_trans_log_full_commit = trans->transid;
6997         } else {
6998                 ret = btrfs_insert_inode_ref(trans, dest,
6999                                              new_dentry->d_name.name,
7000                                              new_dentry->d_name.len,
7001                                              old_ino,
7002                                              btrfs_ino(new_dir), index);
7003                 if (ret)
7004                         goto out_fail;
7005                 /*
7006                  * this is an ugly little race, but the rename is required
7007                  * to make sure that if we crash, the inode is either at the
7008                  * old name or the new one.  pinning the log transaction lets
7009                  * us make sure we don't allow a log commit to come in after
7010                  * we unlink the name but before we add the new name back in.
7011                  */
7012                 btrfs_pin_log_trans(root);
7013         }
7014         /*
7015          * make sure the inode gets flushed if it is replacing
7016          * something.
7017          */
7018         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7019                 btrfs_add_ordered_operation(trans, root, old_inode);
7020
7021         old_dir->i_ctime = old_dir->i_mtime = ctime;
7022         new_dir->i_ctime = new_dir->i_mtime = ctime;
7023         old_inode->i_ctime = ctime;
7024
7025         if (old_dentry->d_parent != new_dentry->d_parent)
7026                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7027
7028         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7029                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7030                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7031                                         old_dentry->d_name.name,
7032                                         old_dentry->d_name.len);
7033         } else {
7034                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7035                                         old_dentry->d_inode,
7036                                         old_dentry->d_name.name,
7037                                         old_dentry->d_name.len);
7038                 if (!ret)
7039                         ret = btrfs_update_inode(trans, root, old_inode);
7040         }
7041         BUG_ON(ret);
7042
7043         if (new_inode) {
7044                 new_inode->i_ctime = CURRENT_TIME;
7045                 if (unlikely(btrfs_ino(new_inode) ==
7046                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7047                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7048                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7049                                                 root_objectid,
7050                                                 new_dentry->d_name.name,
7051                                                 new_dentry->d_name.len);
7052                         BUG_ON(new_inode->i_nlink == 0);
7053                 } else {
7054                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7055                                                  new_dentry->d_inode,
7056                                                  new_dentry->d_name.name,
7057                                                  new_dentry->d_name.len);
7058                 }
7059                 BUG_ON(ret);
7060                 if (new_inode->i_nlink == 0) {
7061                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7062                         BUG_ON(ret);
7063                 }
7064         }
7065
7066         fixup_inode_flags(new_dir, old_inode);
7067
7068         ret = btrfs_add_link(trans, new_dir, old_inode,
7069                              new_dentry->d_name.name,
7070                              new_dentry->d_name.len, 0, index);
7071         BUG_ON(ret);
7072
7073         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7074                 struct dentry *parent = new_dentry->d_parent;
7075                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7076                 btrfs_end_log_trans(root);
7077         }
7078 out_fail:
7079         btrfs_end_transaction_throttle(trans, root);
7080 out_notrans:
7081         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7082                 up_read(&root->fs_info->subvol_sem);
7083
7084         return ret;
7085 }
7086
7087 /*
7088  * some fairly slow code that needs optimization. This walks the list
7089  * of all the inodes with pending delalloc and forces them to disk.
7090  */
7091 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7092 {
7093         struct list_head *head = &root->fs_info->delalloc_inodes;
7094         struct btrfs_inode *binode;
7095         struct inode *inode;
7096
7097         if (root->fs_info->sb->s_flags & MS_RDONLY)
7098                 return -EROFS;
7099
7100         spin_lock(&root->fs_info->delalloc_lock);
7101         while (!list_empty(head)) {
7102                 binode = list_entry(head->next, struct btrfs_inode,
7103                                     delalloc_inodes);
7104                 inode = igrab(&binode->vfs_inode);
7105                 if (!inode)
7106                         list_del_init(&binode->delalloc_inodes);
7107                 spin_unlock(&root->fs_info->delalloc_lock);
7108                 if (inode) {
7109                         filemap_flush(inode->i_mapping);
7110                         if (delay_iput)
7111                                 btrfs_add_delayed_iput(inode);
7112                         else
7113                                 iput(inode);
7114                 }
7115                 cond_resched();
7116                 spin_lock(&root->fs_info->delalloc_lock);
7117         }
7118         spin_unlock(&root->fs_info->delalloc_lock);
7119
7120         /* the filemap_flush will queue IO into the worker threads, but
7121          * we have to make sure the IO is actually started and that
7122          * ordered extents get created before we return
7123          */
7124         atomic_inc(&root->fs_info->async_submit_draining);
7125         while (atomic_read(&root->fs_info->nr_async_submits) ||
7126               atomic_read(&root->fs_info->async_delalloc_pages)) {
7127                 wait_event(root->fs_info->async_submit_wait,
7128                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7129                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7130         }
7131         atomic_dec(&root->fs_info->async_submit_draining);
7132         return 0;
7133 }
7134
7135 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7136                          const char *symname)
7137 {
7138         struct btrfs_trans_handle *trans;
7139         struct btrfs_root *root = BTRFS_I(dir)->root;
7140         struct btrfs_path *path;
7141         struct btrfs_key key;
7142         struct inode *inode = NULL;
7143         int err;
7144         int drop_inode = 0;
7145         u64 objectid;
7146         u64 index = 0 ;
7147         int name_len;
7148         int datasize;
7149         unsigned long ptr;
7150         struct btrfs_file_extent_item *ei;
7151         struct extent_buffer *leaf;
7152         unsigned long nr = 0;
7153
7154         name_len = strlen(symname) + 1;
7155         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7156                 return -ENAMETOOLONG;
7157
7158         /*
7159          * 2 items for inode item and ref
7160          * 2 items for dir items
7161          * 1 item for xattr if selinux is on
7162          */
7163         trans = btrfs_start_transaction(root, 5);
7164         if (IS_ERR(trans))
7165                 return PTR_ERR(trans);
7166
7167         err = btrfs_find_free_ino(root, &objectid);
7168         if (err)
7169                 goto out_unlock;
7170
7171         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7172                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7173                                 S_IFLNK|S_IRWXUGO, &index);
7174         if (IS_ERR(inode)) {
7175                 err = PTR_ERR(inode);
7176                 goto out_unlock;
7177         }
7178
7179         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7180         if (err) {
7181                 drop_inode = 1;
7182                 goto out_unlock;
7183         }
7184
7185         /*
7186         * If the active LSM wants to access the inode during
7187         * d_instantiate it needs these. Smack checks to see
7188         * if the filesystem supports xattrs by looking at the
7189         * ops vector.
7190         */
7191         inode->i_fop = &btrfs_file_operations;
7192         inode->i_op = &btrfs_file_inode_operations;
7193
7194         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7195         if (err)
7196                 drop_inode = 1;
7197         else {
7198                 inode->i_mapping->a_ops = &btrfs_aops;
7199                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7200                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7201         }
7202         if (drop_inode)
7203                 goto out_unlock;
7204
7205         path = btrfs_alloc_path();
7206         if (!path) {
7207                 err = -ENOMEM;
7208                 drop_inode = 1;
7209                 goto out_unlock;
7210         }
7211         key.objectid = btrfs_ino(inode);
7212         key.offset = 0;
7213         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7214         datasize = btrfs_file_extent_calc_inline_size(name_len);
7215         err = btrfs_insert_empty_item(trans, root, path, &key,
7216                                       datasize);
7217         if (err) {
7218                 drop_inode = 1;
7219                 btrfs_free_path(path);
7220                 goto out_unlock;
7221         }
7222         leaf = path->nodes[0];
7223         ei = btrfs_item_ptr(leaf, path->slots[0],
7224                             struct btrfs_file_extent_item);
7225         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7226         btrfs_set_file_extent_type(leaf, ei,
7227                                    BTRFS_FILE_EXTENT_INLINE);
7228         btrfs_set_file_extent_encryption(leaf, ei, 0);
7229         btrfs_set_file_extent_compression(leaf, ei, 0);
7230         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7231         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7232
7233         ptr = btrfs_file_extent_inline_start(ei);
7234         write_extent_buffer(leaf, symname, ptr, name_len);
7235         btrfs_mark_buffer_dirty(leaf);
7236         btrfs_free_path(path);
7237
7238         inode->i_op = &btrfs_symlink_inode_operations;
7239         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7240         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7241         inode_set_bytes(inode, name_len);
7242         btrfs_i_size_write(inode, name_len - 1);
7243         err = btrfs_update_inode(trans, root, inode);
7244         if (err)
7245                 drop_inode = 1;
7246
7247 out_unlock:
7248         nr = trans->blocks_used;
7249         btrfs_end_transaction_throttle(trans, root);
7250         if (drop_inode) {
7251                 inode_dec_link_count(inode);
7252                 iput(inode);
7253         }
7254         btrfs_btree_balance_dirty(root, nr);
7255         return err;
7256 }
7257
7258 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7259                                        u64 start, u64 num_bytes, u64 min_size,
7260                                        loff_t actual_len, u64 *alloc_hint,
7261                                        struct btrfs_trans_handle *trans)
7262 {
7263         struct btrfs_root *root = BTRFS_I(inode)->root;
7264         struct btrfs_key ins;
7265         u64 cur_offset = start;
7266         u64 i_size;
7267         int ret = 0;
7268         bool own_trans = true;
7269
7270         if (trans)
7271                 own_trans = false;
7272         while (num_bytes > 0) {
7273                 if (own_trans) {
7274                         trans = btrfs_start_transaction(root, 3);
7275                         if (IS_ERR(trans)) {
7276                                 ret = PTR_ERR(trans);
7277                                 break;
7278                         }
7279                 }
7280
7281                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7282                                            0, *alloc_hint, (u64)-1, &ins, 1);
7283                 if (ret) {
7284                         if (own_trans)
7285                                 btrfs_end_transaction(trans, root);
7286                         break;
7287                 }
7288
7289                 ret = insert_reserved_file_extent(trans, inode,
7290                                                   cur_offset, ins.objectid,
7291                                                   ins.offset, ins.offset,
7292                                                   ins.offset, 0, 0, 0,
7293                                                   BTRFS_FILE_EXTENT_PREALLOC);
7294                 BUG_ON(ret);
7295                 btrfs_drop_extent_cache(inode, cur_offset,
7296                                         cur_offset + ins.offset -1, 0);
7297
7298                 num_bytes -= ins.offset;
7299                 cur_offset += ins.offset;
7300                 *alloc_hint = ins.objectid + ins.offset;
7301
7302                 inode->i_ctime = CURRENT_TIME;
7303                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7304                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7305                     (actual_len > inode->i_size) &&
7306                     (cur_offset > inode->i_size)) {
7307                         if (cur_offset > actual_len)
7308                                 i_size = actual_len;
7309                         else
7310                                 i_size = cur_offset;
7311                         i_size_write(inode, i_size);
7312                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7313                 }
7314
7315                 ret = btrfs_update_inode(trans, root, inode);
7316                 BUG_ON(ret);
7317
7318                 if (own_trans)
7319                         btrfs_end_transaction(trans, root);
7320         }
7321         return ret;
7322 }
7323
7324 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7325                               u64 start, u64 num_bytes, u64 min_size,
7326                               loff_t actual_len, u64 *alloc_hint)
7327 {
7328         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7329                                            min_size, actual_len, alloc_hint,
7330                                            NULL);
7331 }
7332
7333 int btrfs_prealloc_file_range_trans(struct inode *inode,
7334                                     struct btrfs_trans_handle *trans, int mode,
7335                                     u64 start, u64 num_bytes, u64 min_size,
7336                                     loff_t actual_len, u64 *alloc_hint)
7337 {
7338         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7339                                            min_size, actual_len, alloc_hint, trans);
7340 }
7341
7342 static int btrfs_set_page_dirty(struct page *page)
7343 {
7344         return __set_page_dirty_nobuffers(page);
7345 }
7346
7347 static int btrfs_permission(struct inode *inode, int mask)
7348 {
7349         struct btrfs_root *root = BTRFS_I(inode)->root;
7350         umode_t mode = inode->i_mode;
7351
7352         if (mask & MAY_WRITE &&
7353             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7354                 if (btrfs_root_readonly(root))
7355                         return -EROFS;
7356                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7357                         return -EACCES;
7358         }
7359         return generic_permission(inode, mask);
7360 }
7361
7362 static const struct inode_operations btrfs_dir_inode_operations = {
7363         .getattr        = btrfs_getattr,
7364         .lookup         = btrfs_lookup,
7365         .create         = btrfs_create,
7366         .unlink         = btrfs_unlink,
7367         .link           = btrfs_link,
7368         .mkdir          = btrfs_mkdir,
7369         .rmdir          = btrfs_rmdir,
7370         .rename         = btrfs_rename,
7371         .symlink        = btrfs_symlink,
7372         .setattr        = btrfs_setattr,
7373         .mknod          = btrfs_mknod,
7374         .setxattr       = btrfs_setxattr,
7375         .getxattr       = btrfs_getxattr,
7376         .listxattr      = btrfs_listxattr,
7377         .removexattr    = btrfs_removexattr,
7378         .permission     = btrfs_permission,
7379         .get_acl        = btrfs_get_acl,
7380 };
7381 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7382         .lookup         = btrfs_lookup,
7383         .permission     = btrfs_permission,
7384         .get_acl        = btrfs_get_acl,
7385 };
7386
7387 static const struct file_operations btrfs_dir_file_operations = {
7388         .llseek         = generic_file_llseek,
7389         .read           = generic_read_dir,
7390         .readdir        = btrfs_real_readdir,
7391         .unlocked_ioctl = btrfs_ioctl,
7392 #ifdef CONFIG_COMPAT
7393         .compat_ioctl   = btrfs_ioctl,
7394 #endif
7395         .release        = btrfs_release_file,
7396         .fsync          = btrfs_sync_file,
7397 };
7398
7399 static struct extent_io_ops btrfs_extent_io_ops = {
7400         .fill_delalloc = run_delalloc_range,
7401         .submit_bio_hook = btrfs_submit_bio_hook,
7402         .merge_bio_hook = btrfs_merge_bio_hook,
7403         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7404         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7405         .writepage_start_hook = btrfs_writepage_start_hook,
7406         .set_bit_hook = btrfs_set_bit_hook,
7407         .clear_bit_hook = btrfs_clear_bit_hook,
7408         .merge_extent_hook = btrfs_merge_extent_hook,
7409         .split_extent_hook = btrfs_split_extent_hook,
7410 };
7411
7412 /*
7413  * btrfs doesn't support the bmap operation because swapfiles
7414  * use bmap to make a mapping of extents in the file.  They assume
7415  * these extents won't change over the life of the file and they
7416  * use the bmap result to do IO directly to the drive.
7417  *
7418  * the btrfs bmap call would return logical addresses that aren't
7419  * suitable for IO and they also will change frequently as COW
7420  * operations happen.  So, swapfile + btrfs == corruption.
7421  *
7422  * For now we're avoiding this by dropping bmap.
7423  */
7424 static const struct address_space_operations btrfs_aops = {
7425         .readpage       = btrfs_readpage,
7426         .writepage      = btrfs_writepage,
7427         .writepages     = btrfs_writepages,
7428         .readpages      = btrfs_readpages,
7429         .direct_IO      = btrfs_direct_IO,
7430         .invalidatepage = btrfs_invalidatepage,
7431         .releasepage    = btrfs_releasepage,
7432         .set_page_dirty = btrfs_set_page_dirty,
7433         .error_remove_page = generic_error_remove_page,
7434 };
7435
7436 static const struct address_space_operations btrfs_symlink_aops = {
7437         .readpage       = btrfs_readpage,
7438         .writepage      = btrfs_writepage,
7439         .invalidatepage = btrfs_invalidatepage,
7440         .releasepage    = btrfs_releasepage,
7441 };
7442
7443 static const struct inode_operations btrfs_file_inode_operations = {
7444         .getattr        = btrfs_getattr,
7445         .setattr        = btrfs_setattr,
7446         .setxattr       = btrfs_setxattr,
7447         .getxattr       = btrfs_getxattr,
7448         .listxattr      = btrfs_listxattr,
7449         .removexattr    = btrfs_removexattr,
7450         .permission     = btrfs_permission,
7451         .fiemap         = btrfs_fiemap,
7452         .get_acl        = btrfs_get_acl,
7453 };
7454 static const struct inode_operations btrfs_special_inode_operations = {
7455         .getattr        = btrfs_getattr,
7456         .setattr        = btrfs_setattr,
7457         .permission     = btrfs_permission,
7458         .setxattr       = btrfs_setxattr,
7459         .getxattr       = btrfs_getxattr,
7460         .listxattr      = btrfs_listxattr,
7461         .removexattr    = btrfs_removexattr,
7462         .get_acl        = btrfs_get_acl,
7463 };
7464 static const struct inode_operations btrfs_symlink_inode_operations = {
7465         .readlink       = generic_readlink,
7466         .follow_link    = page_follow_link_light,
7467         .put_link       = page_put_link,
7468         .getattr        = btrfs_getattr,
7469         .setattr        = btrfs_setattr,
7470         .permission     = btrfs_permission,
7471         .setxattr       = btrfs_setxattr,
7472         .getxattr       = btrfs_getxattr,
7473         .listxattr      = btrfs_listxattr,
7474         .removexattr    = btrfs_removexattr,
7475         .get_acl        = btrfs_get_acl,
7476 };
7477
7478 const struct dentry_operations btrfs_dentry_operations = {
7479         .d_delete       = btrfs_dentry_delete,
7480         .d_release      = btrfs_dentry_release,
7481 };