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 "compat.h"
  41 #include "ctree.h"
  42 #include "disk-io.h"
  43 #include "transaction.h"
  44 #include "btrfs_inode.h"
  45 #include "ioctl.h"
  46 #include "print-tree.h"
  47 #include "volumes.h"
  48 #include "ordered-data.h"
  49 #include "xattr.h"
  50 #include "tree-log.h"
  51 #include "compression.h"
  52 #include "locking.h"
  53
  54 struct btrfs_iget_args {
  55         u64 ino;
  56         struct btrfs_root *root;
  57 };
  58
  59 static const struct inode_operations btrfs_dir_inode_operations;
  60 static const struct inode_operations btrfs_symlink_inode_operations;
  61 static const struct inode_operations btrfs_dir_ro_inode_operations;
  62 static const struct inode_operations btrfs_special_inode_operations;
  63 static const struct inode_operations btrfs_file_inode_operations;
  64 static const struct address_space_operations btrfs_aops;
  65 static const struct address_space_operations btrfs_symlink_aops;
  66 static const struct file_operations btrfs_dir_file_operations;
  67 static struct extent_io_ops btrfs_extent_io_ops;
  68
  69 static struct kmem_cache *btrfs_inode_cachep;
  70 struct kmem_cache *btrfs_trans_handle_cachep;
  71 struct kmem_cache *btrfs_transaction_cachep;
  72 struct kmem_cache *btrfs_path_cachep;
  73
  74 #define S_SHIFT 12
  75 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  76         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  77         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  78         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  79         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  80         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  81         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  82         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  83 };
  84
  85 static void btrfs_truncate(struct inode *inode);
  86 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  87 static noinline int cow_file_range(struct inode *inode,
  88                                    struct page *locked_page,
  89                                    u64 start, u64 end, int *page_started,
  90                                    unsigned long *nr_written, int unlock);
  91
  92 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  93                                      struct inode *inode,  struct inode *dir)
  94 {
  95         int err;
  96
  97         err = btrfs_init_acl(trans, inode, dir);
  98         if (!err)
  99                 err = btrfs_xattr_security_init(trans, inode, dir);
 100         return err;
 101 }
 102
 103 /*
 104  * this does all the hard work for inserting an inline extent into
 105  * the btree.  The caller should have done a btrfs_drop_extents so that
 106  * no overlapping inline items exist in the btree
 107  */
 108 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 109                                 struct btrfs_root *root, struct inode *inode,
 110                                 u64 start, size_t size, size_t compressed_size,
 111                                 struct page **compressed_pages)
 112 {
 113         struct btrfs_key key;
 114         struct btrfs_path *path;
 115         struct extent_buffer *leaf;
 116         struct page *page = NULL;
 117         char *kaddr;
 118         unsigned long ptr;
 119         struct btrfs_file_extent_item *ei;
 120         int err = 0;
 121         int ret;
 122         size_t cur_size = size;
 123         size_t datasize;
 124         unsigned long offset;
 125         int use_compress = 0;
 126
 127         if (compressed_size && compressed_pages) {
 128                 use_compress = 1;
 129                 cur_size = compressed_size;
 130         }
 131
 132         path = btrfs_alloc_path();
 133         if (!path)
 134                 return -ENOMEM;
 135
 136         path->leave_spinning = 1;
 137         btrfs_set_trans_block_group(trans, inode);
 138
 139         key.objectid = inode->i_ino;
 140         key.offset = start;
 141         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 142         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 143
 144         inode_add_bytes(inode, size);
 145         ret = btrfs_insert_empty_item(trans, root, path, &key,
 146                                       datasize);
 147         BUG_ON(ret);
 148         if (ret) {
 149                 err = ret;
 150                 goto fail;
 151         }
 152         leaf = path->nodes[0];
 153         ei = btrfs_item_ptr(leaf, path->slots[0],
 154                             struct btrfs_file_extent_item);
 155         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 156         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 157         btrfs_set_file_extent_encryption(leaf, ei, 0);
 158         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 159         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 160         ptr = btrfs_file_extent_inline_start(ei);
 161
 162         if (use_compress) {
 163                 struct page *cpage;
 164                 int i = 0;
 165                 while (compressed_size > 0) {
 166                         cpage = compressed_pages[i];
 167                         cur_size = min_t(unsigned long, compressed_size,
 168                                        PAGE_CACHE_SIZE);
 169
 170                         kaddr = kmap_atomic(cpage, KM_USER0);
 171                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 172                         kunmap_atomic(kaddr, KM_USER0);
 173
 174                         i++;
 175                         ptr += cur_size;
 176                         compressed_size -= cur_size;
 177                 }
 178                 btrfs_set_file_extent_compression(leaf, ei,
 179                                                   BTRFS_COMPRESS_ZLIB);
 180         } else {
 181                 page = find_get_page(inode->i_mapping,
 182                                      start >> PAGE_CACHE_SHIFT);
 183                 btrfs_set_file_extent_compression(leaf, ei, 0);
 184                 kaddr = kmap_atomic(page, KM_USER0);
 185                 offset = start & (PAGE_CACHE_SIZE - 1);
 186                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 187                 kunmap_atomic(kaddr, KM_USER0);
 188                 page_cache_release(page);
 189         }
 190         btrfs_mark_buffer_dirty(leaf);
 191         btrfs_free_path(path);
 192
 193         /*
 194          * we're an inline extent, so nobody can
 195          * extend the file past i_size without locking
 196          * a page we already have locked.
 197          *
 198          * We must do any isize and inode updates
 199          * before we unlock the pages.  Otherwise we
 200          * could end up racing with unlink.
 201          */
 202         BTRFS_I(inode)->disk_i_size = inode->i_size;
 203         btrfs_update_inode(trans, root, inode);
 204
 205         return 0;
 206 fail:
 207         btrfs_free_path(path);
 208         return err;
 209 }
 210
 211
 212 /*
 213  * conditionally insert an inline extent into the file.  This
 214  * does the checks required to make sure the data is small enough
 215  * to fit as an inline extent.
 216  */
 217 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 218                                  struct btrfs_root *root,
 219                                  struct inode *inode, u64 start, u64 end,
 220                                  size_t compressed_size,
 221                                  struct page **compressed_pages)
 222 {
 223         u64 isize = i_size_read(inode);
 224         u64 actual_end = min(end + 1, isize);
 225         u64 inline_len = actual_end - start;
 226         u64 aligned_end = (end + root->sectorsize - 1) &
 227                         ~((u64)root->sectorsize - 1);
 228         u64 hint_byte;
 229         u64 data_len = inline_len;
 230         int ret;
 231
 232         if (compressed_size)
 233                 data_len = compressed_size;
 234
 235         if (start > 0 ||
 236             actual_end >= PAGE_CACHE_SIZE ||
 237             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 238             (!compressed_size &&
 239             (actual_end & (root->sectorsize - 1)) == 0) ||
 240             end + 1 < isize ||
 241             data_len > root->fs_info->max_inline) {
 242                 return 1;
 243         }
 244
 245         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 246                                  &hint_byte, 1);
 247         BUG_ON(ret);
 248
 249         if (isize > actual_end)
 250                 inline_len = min_t(u64, isize, actual_end);
 251         ret = insert_inline_extent(trans, root, inode, start,
 252                                    inline_len, compressed_size,
 253                                    compressed_pages);
 254         BUG_ON(ret);
 255         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 256         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 257         return 0;
 258 }
 259
 260 struct async_extent {
 261         u64 start;
 262         u64 ram_size;
 263         u64 compressed_size;
 264         struct page **pages;
 265         unsigned long nr_pages;
 266         struct list_head list;
 267 };
 268
 269 struct async_cow {
 270         struct inode *inode;
 271         struct btrfs_root *root;
 272         struct page *locked_page;
 273         u64 start;
 274         u64 end;
 275         struct list_head extents;
 276         struct btrfs_work work;
 277 };
 278
 279 static noinline int add_async_extent(struct async_cow *cow,
 280                                      u64 start, u64 ram_size,
 281                                      u64 compressed_size,
 282                                      struct page **pages,
 283                                      unsigned long nr_pages)
 284 {
 285         struct async_extent *async_extent;
 286
 287         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 288         async_extent->start = start;
 289         async_extent->ram_size = ram_size;
 290         async_extent->compressed_size = compressed_size;
 291         async_extent->pages = pages;
 292         async_extent->nr_pages = nr_pages;
 293         list_add_tail(&async_extent->list, &cow->extents);
 294         return 0;
 295 }
 296
 297 /*
 298  * we create compressed extents in two phases.  The first
 299  * phase compresses a range of pages that have already been
 300  * locked (both pages and state bits are locked).
 301  *
 302  * This is done inside an ordered work queue, and the compression
 303  * is spread across many cpus.  The actual IO submission is step
 304  * two, and the ordered work queue takes care of making sure that
 305  * happens in the same order things were put onto the queue by
 306  * writepages and friends.
 307  *
 308  * If this code finds it can't get good compression, it puts an
 309  * entry onto the work queue to write the uncompressed bytes.  This
 310  * makes sure that both compressed inodes and uncompressed inodes
 311  * are written in the same order that pdflush sent them down.
 312  */
 313 static noinline int compress_file_range(struct inode *inode,
 314                                         struct page *locked_page,
 315                                         u64 start, u64 end,
 316                                         struct async_cow *async_cow,
 317                                         int *num_added)
 318 {
 319         struct btrfs_root *root = BTRFS_I(inode)->root;
 320         struct btrfs_trans_handle *trans;
 321         u64 num_bytes;
 322         u64 blocksize = root->sectorsize;
 323         u64 actual_end;
 324         u64 isize = i_size_read(inode);
 325         int ret = 0;
 326         struct page **pages = NULL;
 327         unsigned long nr_pages;
 328         unsigned long nr_pages_ret = 0;
 329         unsigned long total_compressed = 0;
 330         unsigned long total_in = 0;
 331         unsigned long max_compressed = 128 * 1024;
 332         unsigned long max_uncompressed = 128 * 1024;
 333         int i;
 334         int will_compress;
 335
 336         actual_end = min_t(u64, isize, end + 1);
 337 again:
 338         will_compress = 0;
 339         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 340         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 341
 342         /*
 343          * we don't want to send crud past the end of i_size through
 344          * compression, that's just a waste of CPU time.  So, if the
 345          * end of the file is before the start of our current
 346          * requested range of bytes, we bail out to the uncompressed
 347          * cleanup code that can deal with all of this.
 348          *
 349          * It isn't really the fastest way to fix things, but this is a
 350          * very uncommon corner.
 351          */
 352         if (actual_end <= start)
 353                 goto cleanup_and_bail_uncompressed;
 354
 355         total_compressed = actual_end - start;
 356
 357         /* we want to make sure that amount of ram required to uncompress
 358          * an extent is reasonable, so we limit the total size in ram
 359          * of a compressed extent to 128k.  This is a crucial number
 360          * because it also controls how easily we can spread reads across
 361          * cpus for decompression.
 362          *
 363          * We also want to make sure the amount of IO required to do
 364          * a random read is reasonably small, so we limit the size of
 365          * a compressed extent to 128k.
 366          */
 367         total_compressed = min(total_compressed, max_uncompressed);
 368         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 369         num_bytes = max(blocksize,  num_bytes);
 370         total_in = 0;
 371         ret = 0;
 372
 373         /*
 374          * we do compression for mount -o compress and when the
 375          * inode has not been flagged as nocompress.  This flag can
 376          * change at any time if we discover bad compression ratios.
 377          */
 378         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 379             (btrfs_test_opt(root, COMPRESS) ||
 380              (BTRFS_I(inode)->force_compress))) {
 381                 WARN_ON(pages);
 382                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 383
 384                 ret = btrfs_zlib_compress_pages(inode->i_mapping, start,
 385                                                 total_compressed, pages,
 386                                                 nr_pages, &nr_pages_ret,
 387                                                 &total_in,
 388                                                 &total_compressed,
 389                                                 max_compressed);
 390
 391                 if (!ret) {
 392                         unsigned long offset = total_compressed &
 393                                 (PAGE_CACHE_SIZE - 1);
 394                         struct page *page = pages[nr_pages_ret - 1];
 395                         char *kaddr;
 396
 397                         /* zero the tail end of the last page, we might be
 398                          * sending it down to disk
 399                          */
 400                         if (offset) {
 401                                 kaddr = kmap_atomic(page, KM_USER0);
 402                                 memset(kaddr + offset, 0,
 403                                        PAGE_CACHE_SIZE - offset);
 404                                 kunmap_atomic(kaddr, KM_USER0);
 405                         }
 406                         will_compress = 1;
 407                 }
 408         }
 409         if (start == 0) {
 410                 trans = btrfs_join_transaction(root, 1);
 411                 BUG_ON(!trans);
 412                 btrfs_set_trans_block_group(trans, inode);
 413                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 414
 415                 /* lets try to make an inline extent */
 416                 if (ret || total_in < (actual_end - start)) {
 417                         /* we didn't compress the entire range, try
 418                          * to make an uncompressed inline extent.
 419                          */
 420                         ret = cow_file_range_inline(trans, root, inode,
 421                                                     start, end, 0, NULL);
 422                 } else {
 423                         /* try making a compressed inline extent */
 424                         ret = cow_file_range_inline(trans, root, inode,
 425                                                     start, end,
 426                                                     total_compressed, pages);
 427                 }
 428                 if (ret == 0) {
 429                         /*
 430                          * inline extent creation worked, we don't need
 431                          * to create any more async work items.  Unlock
 432                          * and free up our temp pages.
 433                          */
 434                         extent_clear_unlock_delalloc(inode,
 435                              &BTRFS_I(inode)->io_tree,
 436                              start, end, NULL,
 437                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 438                              EXTENT_CLEAR_DELALLOC |
 439                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 440
 441                         btrfs_end_transaction(trans, root);
 442                         goto free_pages_out;
 443                 }
 444                 btrfs_end_transaction(trans, root);
 445         }
 446
 447         if (will_compress) {
 448                 /*
 449                  * we aren't doing an inline extent round the compressed size
 450                  * up to a block size boundary so the allocator does sane
 451                  * things
 452                  */
 453                 total_compressed = (total_compressed + blocksize - 1) &
 454                         ~(blocksize - 1);
 455
 456                 /*
 457                  * one last check to make sure the compression is really a
 458                  * win, compare the page count read with the blocks on disk
 459                  */
 460                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 461                         ~(PAGE_CACHE_SIZE - 1);
 462                 if (total_compressed >= total_in) {
 463                         will_compress = 0;
 464                 } else {
 465                         num_bytes = total_in;
 466                 }
 467         }
 468         if (!will_compress && pages) {
 469                 /*
 470                  * the compression code ran but failed to make things smaller,
 471                  * free any pages it allocated and our page pointer array
 472                  */
 473                 for (i = 0; i < nr_pages_ret; i++) {
 474                         WARN_ON(pages[i]->mapping);
 475                         page_cache_release(pages[i]);
 476                 }
 477                 kfree(pages);
 478                 pages = NULL;
 479                 total_compressed = 0;
 480                 nr_pages_ret = 0;
 481
 482                 /* flag the file so we don't compress in the future */
 483                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 484                     !(BTRFS_I(inode)->force_compress)) {
 485                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 486                 }
 487         }
 488         if (will_compress) {
 489                 *num_added += 1;
 490
 491                 /* the async work queues will take care of doing actual
 492                  * allocation on disk for these compressed pages,
 493                  * and will submit them to the elevator.
 494                  */
 495                 add_async_extent(async_cow, start, num_bytes,
 496                                  total_compressed, pages, nr_pages_ret);
 497
 498                 if (start + num_bytes < end) {
 499                         start += num_bytes;
 500                         pages = NULL;
 501                         cond_resched();
 502                         goto again;
 503                 }
 504         } else {
 505 cleanup_and_bail_uncompressed:
 506                 /*
 507                  * No compression, but we still need to write the pages in
 508                  * the file we've been given so far.  redirty the locked
 509                  * page if it corresponds to our extent and set things up
 510                  * for the async work queue to run cow_file_range to do
 511                  * the normal delalloc dance
 512                  */
 513                 if (page_offset(locked_page) >= start &&
 514                     page_offset(locked_page) <= end) {
 515                         __set_page_dirty_nobuffers(locked_page);
 516                         /* unlocked later on in the async handlers */
 517                 }
 518                 add_async_extent(async_cow, start, end - start + 1, 0, NULL, 0);
 519                 *num_added += 1;
 520         }
 521
 522 out:
 523         return 0;
 524
 525 free_pages_out:
 526         for (i = 0; i < nr_pages_ret; i++) {
 527                 WARN_ON(pages[i]->mapping);
 528                 page_cache_release(pages[i]);
 529         }
 530         kfree(pages);
 531
 532         goto out;
 533 }
 534
 535 /*
 536  * phase two of compressed writeback.  This is the ordered portion
 537  * of the code, which only gets called in the order the work was
 538  * queued.  We walk all the async extents created by compress_file_range
 539  * and send them down to the disk.
 540  */
 541 static noinline int submit_compressed_extents(struct inode *inode,
 542                                               struct async_cow *async_cow)
 543 {
 544         struct async_extent *async_extent;
 545         u64 alloc_hint = 0;
 546         struct btrfs_trans_handle *trans;
 547         struct btrfs_key ins;
 548         struct extent_map *em;
 549         struct btrfs_root *root = BTRFS_I(inode)->root;
 550         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 551         struct extent_io_tree *io_tree;
 552         int ret = 0;
 553
 554         if (list_empty(&async_cow->extents))
 555                 return 0;
 556
 557
 558         while (!list_empty(&async_cow->extents)) {
 559                 async_extent = list_entry(async_cow->extents.next,
 560                                           struct async_extent, list);
 561                 list_del(&async_extent->list);
 562
 563                 io_tree = &BTRFS_I(inode)->io_tree;
 564
 565 retry:
 566                 /* did the compression code fall back to uncompressed IO? */
 567                 if (!async_extent->pages) {
 568                         int page_started = 0;
 569                         unsigned long nr_written = 0;
 570
 571                         lock_extent(io_tree, async_extent->start,
 572                                          async_extent->start +
 573                                          async_extent->ram_size - 1, GFP_NOFS);
 574
 575                         /* allocate blocks */
 576                         ret = cow_file_range(inode, async_cow->locked_page,
 577                                              async_extent->start,
 578                                              async_extent->start +
 579                                              async_extent->ram_size - 1,
 580                                              &page_started, &nr_written, 0);
 581
 582                         /*
 583                          * if page_started, cow_file_range inserted an
 584                          * inline extent and took care of all the unlocking
 585                          * and IO for us.  Otherwise, we need to submit
 586                          * all those pages down to the drive.
 587                          */
 588                         if (!page_started && !ret)
 589                                 extent_write_locked_range(io_tree,
 590                                                   inode, async_extent->start,
 591                                                   async_extent->start +
 592                                                   async_extent->ram_size - 1,
 593                                                   btrfs_get_extent,
 594                                                   WB_SYNC_ALL);
 595                         kfree(async_extent);
 596                         cond_resched();
 597                         continue;
 598                 }
 599
 600                 lock_extent(io_tree, async_extent->start,
 601                             async_extent->start + async_extent->ram_size - 1,
 602                             GFP_NOFS);
 603
 604                 trans = btrfs_join_transaction(root, 1);
 605                 ret = btrfs_reserve_extent(trans, root,
 606                                            async_extent->compressed_size,
 607                                            async_extent->compressed_size,
 608                                            0, alloc_hint,
 609                                            (u64)-1, &ins, 1);
 610                 btrfs_end_transaction(trans, root);
 611
 612                 if (ret) {
 613                         int i;
 614                         for (i = 0; i < async_extent->nr_pages; i++) {
 615                                 WARN_ON(async_extent->pages[i]->mapping);
 616                                 page_cache_release(async_extent->pages[i]);
 617                         }
 618                         kfree(async_extent->pages);
 619                         async_extent->nr_pages = 0;
 620                         async_extent->pages = NULL;
 621                         unlock_extent(io_tree, async_extent->start,
 622                                       async_extent->start +
 623                                       async_extent->ram_size - 1, GFP_NOFS);
 624                         goto retry;
 625                 }
 626
 627                 /*
 628                  * here we're doing allocation and writeback of the
 629                  * compressed pages
 630                  */
 631                 btrfs_drop_extent_cache(inode, async_extent->start,
 632                                         async_extent->start +
 633                                         async_extent->ram_size - 1, 0);
 634
 635                 em = alloc_extent_map(GFP_NOFS);
 636                 em->start = async_extent->start;
 637                 em->len = async_extent->ram_size;
 638                 em->orig_start = em->start;
 639
 640                 em->block_start = ins.objectid;
 641                 em->block_len = ins.offset;
 642                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 643                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 644                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 645
 646                 while (1) {
 647                         write_lock(&em_tree->lock);
 648                         ret = add_extent_mapping(em_tree, em);
 649                         write_unlock(&em_tree->lock);
 650                         if (ret != -EEXIST) {
 651                                 free_extent_map(em);
 652                                 break;
 653                         }
 654                         btrfs_drop_extent_cache(inode, async_extent->start,
 655                                                 async_extent->start +
 656                                                 async_extent->ram_size - 1, 0);
 657                 }
 658
 659                 ret = btrfs_add_ordered_extent(inode, async_extent->start,
 660                                                ins.objectid,
 661                                                async_extent->ram_size,
 662                                                ins.offset,
 663                                                BTRFS_ORDERED_COMPRESSED);
 664                 BUG_ON(ret);
 665
 666                 /*
 667                  * clear dirty, set writeback and unlock the pages.
 668                  */
 669                 extent_clear_unlock_delalloc(inode,
 670                                 &BTRFS_I(inode)->io_tree,
 671                                 async_extent->start,
 672                                 async_extent->start +
 673                                 async_extent->ram_size - 1,
 674                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 675                                 EXTENT_CLEAR_UNLOCK |
 676                                 EXTENT_CLEAR_DELALLOC |
 677                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 678
 679                 ret = btrfs_submit_compressed_write(inode,
 680                                     async_extent->start,
 681                                     async_extent->ram_size,
 682                                     ins.objectid,
 683                                     ins.offset, async_extent->pages,
 684                                     async_extent->nr_pages);
 685
 686                 BUG_ON(ret);
 687                 alloc_hint = ins.objectid + ins.offset;
 688                 kfree(async_extent);
 689                 cond_resched();
 690         }
 691
 692         return 0;
 693 }
 694
 695 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 696                                       u64 num_bytes)
 697 {
 698         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 699         struct extent_map *em;
 700         u64 alloc_hint = 0;
 701
 702         read_lock(&em_tree->lock);
 703         em = search_extent_mapping(em_tree, start, num_bytes);
 704         if (em) {
 705                 /*
 706                  * if block start isn't an actual block number then find the
 707                  * first block in this inode and use that as a hint.  If that
 708                  * block is also bogus then just don't worry about it.
 709                  */
 710                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 711                         free_extent_map(em);
 712                         em = search_extent_mapping(em_tree, 0, 0);
 713                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 714                                 alloc_hint = em->block_start;
 715                         if (em)
 716                                 free_extent_map(em);
 717                 } else {
 718                         alloc_hint = em->block_start;
 719                         free_extent_map(em);
 720                 }
 721         }
 722         read_unlock(&em_tree->lock);
 723
 724         return alloc_hint;
 725 }
 726
 727 /*
 728  * when extent_io.c finds a delayed allocation range in the file,
 729  * the call backs end up in this code.  The basic idea is to
 730  * allocate extents on disk for the range, and create ordered data structs
 731  * in ram to track those extents.
 732  *
 733  * locked_page is the page that writepage had locked already.  We use
 734  * it to make sure we don't do extra locks or unlocks.
 735  *
 736  * *page_started is set to one if we unlock locked_page and do everything
 737  * required to start IO on it.  It may be clean and already done with
 738  * IO when we return.
 739  */
 740 static noinline int cow_file_range(struct inode *inode,
 741                                    struct page *locked_page,
 742                                    u64 start, u64 end, int *page_started,
 743                                    unsigned long *nr_written,
 744                                    int unlock)
 745 {
 746         struct btrfs_root *root = BTRFS_I(inode)->root;
 747         struct btrfs_trans_handle *trans;
 748         u64 alloc_hint = 0;
 749         u64 num_bytes;
 750         unsigned long ram_size;
 751         u64 disk_num_bytes;
 752         u64 cur_alloc_size;
 753         u64 blocksize = root->sectorsize;
 754         struct btrfs_key ins;
 755         struct extent_map *em;
 756         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 757         int ret = 0;
 758
 759         BUG_ON(root == root->fs_info->tree_root);
 760         trans = btrfs_join_transaction(root, 1);
 761         BUG_ON(!trans);
 762         btrfs_set_trans_block_group(trans, inode);
 763         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 764
 765         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 766         num_bytes = max(blocksize,  num_bytes);
 767         disk_num_bytes = num_bytes;
 768         ret = 0;
 769
 770         if (start == 0) {
 771                 /* lets try to make an inline extent */
 772                 ret = cow_file_range_inline(trans, root, inode,
 773                                             start, end, 0, NULL);
 774                 if (ret == 0) {
 775                         extent_clear_unlock_delalloc(inode,
 776                                      &BTRFS_I(inode)->io_tree,
 777                                      start, end, NULL,
 778                                      EXTENT_CLEAR_UNLOCK_PAGE |
 779                                      EXTENT_CLEAR_UNLOCK |
 780                                      EXTENT_CLEAR_DELALLOC |
 781                                      EXTENT_CLEAR_DIRTY |
 782                                      EXTENT_SET_WRITEBACK |
 783                                      EXTENT_END_WRITEBACK);
 784
 785                         *nr_written = *nr_written +
 786                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 787                         *page_started = 1;
 788                         ret = 0;
 789                         goto out;
 790                 }
 791         }
 792
 793         BUG_ON(disk_num_bytes >
 794                btrfs_super_total_bytes(&root->fs_info->super_copy));
 795
 796         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 797         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 798
 799         while (disk_num_bytes > 0) {
 800                 unsigned long op;
 801
 802                 cur_alloc_size = disk_num_bytes;
 803                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 804                                            root->sectorsize, 0, alloc_hint,
 805                                            (u64)-1, &ins, 1);
 806                 BUG_ON(ret);
 807
 808                 em = alloc_extent_map(GFP_NOFS);
 809                 em->start = start;
 810                 em->orig_start = em->start;
 811                 ram_size = ins.offset;
 812                 em->len = ins.offset;
 813
 814                 em->block_start = ins.objectid;
 815                 em->block_len = ins.offset;
 816                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 817                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 818
 819                 while (1) {
 820                         write_lock(&em_tree->lock);
 821                         ret = add_extent_mapping(em_tree, em);
 822                         write_unlock(&em_tree->lock);
 823                         if (ret != -EEXIST) {
 824                                 free_extent_map(em);
 825                                 break;
 826                         }
 827                         btrfs_drop_extent_cache(inode, start,
 828                                                 start + ram_size - 1, 0);
 829                 }
 830
 831                 cur_alloc_size = ins.offset;
 832                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 833                                                ram_size, cur_alloc_size, 0);
 834                 BUG_ON(ret);
 835
 836                 if (root->root_key.objectid ==
 837                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 838                         ret = btrfs_reloc_clone_csums(inode, start,
 839                                                       cur_alloc_size);
 840                         BUG_ON(ret);
 841                 }
 842
 843                 if (disk_num_bytes < cur_alloc_size)
 844                         break;
 845
 846                 /* we're not doing compressed IO, don't unlock the first
 847                  * page (which the caller expects to stay locked), don't
 848                  * clear any dirty bits and don't set any writeback bits
 849                  *
 850                  * Do set the Private2 bit so we know this page was properly
 851                  * setup for writepage
 852                  */
 853                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 854                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 855                         EXTENT_SET_PRIVATE2;
 856
 857                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 858                                              start, start + ram_size - 1,
 859                                              locked_page, op);
 860                 disk_num_bytes -= cur_alloc_size;
 861                 num_bytes -= cur_alloc_size;
 862                 alloc_hint = ins.objectid + ins.offset;
 863                 start += cur_alloc_size;
 864         }
 865 out:
 866         ret = 0;
 867         btrfs_end_transaction(trans, root);
 868
 869         return ret;
 870 }
 871
 872 /*
 873  * work queue call back to started compression on a file and pages
 874  */
 875 static noinline void async_cow_start(struct btrfs_work *work)
 876 {
 877         struct async_cow *async_cow;
 878         int num_added = 0;
 879         async_cow = container_of(work, struct async_cow, work);
 880
 881         compress_file_range(async_cow->inode, async_cow->locked_page,
 882                             async_cow->start, async_cow->end, async_cow,
 883                             &num_added);
 884         if (num_added == 0)
 885                 async_cow->inode = NULL;
 886 }
 887
 888 /*
 889  * work queue call back to submit previously compressed pages
 890  */
 891 static noinline void async_cow_submit(struct btrfs_work *work)
 892 {
 893         struct async_cow *async_cow;
 894         struct btrfs_root *root;
 895         unsigned long nr_pages;
 896
 897         async_cow = container_of(work, struct async_cow, work);
 898
 899         root = async_cow->root;
 900         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 901                 PAGE_CACHE_SHIFT;
 902
 903         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 904
 905         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 906             5 * 1042 * 1024 &&
 907             waitqueue_active(&root->fs_info->async_submit_wait))
 908                 wake_up(&root->fs_info->async_submit_wait);
 909
 910         if (async_cow->inode)
 911                 submit_compressed_extents(async_cow->inode, async_cow);
 912 }
 913
 914 static noinline void async_cow_free(struct btrfs_work *work)
 915 {
 916         struct async_cow *async_cow;
 917         async_cow = container_of(work, struct async_cow, work);
 918         kfree(async_cow);
 919 }
 920
 921 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 922                                 u64 start, u64 end, int *page_started,
 923                                 unsigned long *nr_written)
 924 {
 925         struct async_cow *async_cow;
 926         struct btrfs_root *root = BTRFS_I(inode)->root;
 927         unsigned long nr_pages;
 928         u64 cur_end;
 929         int limit = 10 * 1024 * 1042;
 930
 931         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 932                          1, 0, NULL, GFP_NOFS);
 933         while (start < end) {
 934                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 935                 async_cow->inode = inode;
 936                 async_cow->root = root;
 937                 async_cow->locked_page = locked_page;
 938                 async_cow->start = start;
 939
 940                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 941                         cur_end = end;
 942                 else
 943                         cur_end = min(end, start + 512 * 1024 - 1);
 944
 945                 async_cow->end = cur_end;
 946                 INIT_LIST_HEAD(&async_cow->extents);
 947
 948                 async_cow->work.func = async_cow_start;
 949                 async_cow->work.ordered_func = async_cow_submit;
 950                 async_cow->work.ordered_free = async_cow_free;
 951                 async_cow->work.flags = 0;
 952
 953                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 954                         PAGE_CACHE_SHIFT;
 955                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 956
 957                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 958                                    &async_cow->work);
 959
 960                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 961                         wait_event(root->fs_info->async_submit_wait,
 962                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 963                             limit));
 964                 }
 965
 966                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 967                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 968                         wait_event(root->fs_info->async_submit_wait,
 969                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 970                            0));
 971                 }
 972
 973                 *nr_written += nr_pages;
 974                 start = cur_end + 1;
 975         }
 976         *page_started = 1;
 977         return 0;
 978 }
 979
 980 static noinline int csum_exist_in_range(struct btrfs_root *root,
 981                                         u64 bytenr, u64 num_bytes)
 982 {
 983         int ret;
 984         struct btrfs_ordered_sum *sums;
 985         LIST_HEAD(list);
 986
 987         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
 988                                        bytenr + num_bytes - 1, &list);
 989         if (ret == 0 && list_empty(&list))
 990                 return 0;
 991
 992         while (!list_empty(&list)) {
 993                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
 994                 list_del(&sums->list);
 995                 kfree(sums);
 996         }
 997         return 1;
 998 }
 999
1000 /*
1001  * when nowcow writeback call back.  This checks for snapshots or COW copies
1002  * of the extents that exist in the file, and COWs the file as required.
1003  *
1004  * If no cow copies or snapshots exist, we write directly to the existing
1005  * blocks on disk
1006  */
1007 static noinline int run_delalloc_nocow(struct inode *inode,
1008                                        struct page *locked_page,
1009                               u64 start, u64 end, int *page_started, int force,
1010                               unsigned long *nr_written)
1011 {
1012         struct btrfs_root *root = BTRFS_I(inode)->root;
1013         struct btrfs_trans_handle *trans;
1014         struct extent_buffer *leaf;
1015         struct btrfs_path *path;
1016         struct btrfs_file_extent_item *fi;
1017         struct btrfs_key found_key;
1018         u64 cow_start;
1019         u64 cur_offset;
1020         u64 extent_end;
1021         u64 extent_offset;
1022         u64 disk_bytenr;
1023         u64 num_bytes;
1024         int extent_type;
1025         int ret;
1026         int type;
1027         int nocow;
1028         int check_prev = 1;
1029         bool nolock = false;
1030
1031         path = btrfs_alloc_path();
1032         BUG_ON(!path);
1033         if (root == root->fs_info->tree_root) {
1034                 nolock = true;
1035                 trans = btrfs_join_transaction_nolock(root, 1);
1036         } else {
1037                 trans = btrfs_join_transaction(root, 1);
1038         }
1039         BUG_ON(!trans);
1040
1041         cow_start = (u64)-1;
1042         cur_offset = start;
1043         while (1) {
1044                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1045                                                cur_offset, 0);
1046                 BUG_ON(ret < 0);
1047                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1048                         leaf = path->nodes[0];
1049                         btrfs_item_key_to_cpu(leaf, &found_key,
1050                                               path->slots[0] - 1);
1051                         if (found_key.objectid == inode->i_ino &&
1052                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1053                                 path->slots[0]--;
1054                 }
1055                 check_prev = 0;
1056 next_slot:
1057                 leaf = path->nodes[0];
1058                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1059                         ret = btrfs_next_leaf(root, path);
1060                         if (ret < 0)
1061                                 BUG_ON(1);
1062                         if (ret > 0)
1063                                 break;
1064                         leaf = path->nodes[0];
1065                 }
1066
1067                 nocow = 0;
1068                 disk_bytenr = 0;
1069                 num_bytes = 0;
1070                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1071
1072                 if (found_key.objectid > inode->i_ino ||
1073                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1074                     found_key.offset > end)
1075                         break;
1076
1077                 if (found_key.offset > cur_offset) {
1078                         extent_end = found_key.offset;
1079                         extent_type = 0;
1080                         goto out_check;
1081                 }
1082
1083                 fi = btrfs_item_ptr(leaf, path->slots[0],
1084                                     struct btrfs_file_extent_item);
1085                 extent_type = btrfs_file_extent_type(leaf, fi);
1086
1087                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1088                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1089                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1090                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1091                         extent_end = found_key.offset +
1092                                 btrfs_file_extent_num_bytes(leaf, fi);
1093                         if (extent_end <= start) {
1094                                 path->slots[0]++;
1095                                 goto next_slot;
1096                         }
1097                         if (disk_bytenr == 0)
1098                                 goto out_check;
1099                         if (btrfs_file_extent_compression(leaf, fi) ||
1100                             btrfs_file_extent_encryption(leaf, fi) ||
1101                             btrfs_file_extent_other_encoding(leaf, fi))
1102                                 goto out_check;
1103                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1104                                 goto out_check;
1105                         if (btrfs_extent_readonly(root, disk_bytenr))
1106                                 goto out_check;
1107                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1108                                                   found_key.offset -
1109                                                   extent_offset, disk_bytenr))
1110                                 goto out_check;
1111                         disk_bytenr += extent_offset;
1112                         disk_bytenr += cur_offset - found_key.offset;
1113                         num_bytes = min(end + 1, extent_end) - cur_offset;
1114                         /*
1115                          * force cow if csum exists in the range.
1116                          * this ensure that csum for a given extent are
1117                          * either valid or do not exist.
1118                          */
1119                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1120                                 goto out_check;
1121                         nocow = 1;
1122                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1123                         extent_end = found_key.offset +
1124                                 btrfs_file_extent_inline_len(leaf, fi);
1125                         extent_end = ALIGN(extent_end, root->sectorsize);
1126                 } else {
1127                         BUG_ON(1);
1128                 }
1129 out_check:
1130                 if (extent_end <= start) {
1131                         path->slots[0]++;
1132                         goto next_slot;
1133                 }
1134                 if (!nocow) {
1135                         if (cow_start == (u64)-1)
1136                                 cow_start = cur_offset;
1137                         cur_offset = extent_end;
1138                         if (cur_offset > end)
1139                                 break;
1140                         path->slots[0]++;
1141                         goto next_slot;
1142                 }
1143
1144                 btrfs_release_path(root, path);
1145                 if (cow_start != (u64)-1) {
1146                         ret = cow_file_range(inode, locked_page, cow_start,
1147                                         found_key.offset - 1, page_started,
1148                                         nr_written, 1);
1149                         BUG_ON(ret);
1150                         cow_start = (u64)-1;
1151                 }
1152
1153                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1154                         struct extent_map *em;
1155                         struct extent_map_tree *em_tree;
1156                         em_tree = &BTRFS_I(inode)->extent_tree;
1157                         em = alloc_extent_map(GFP_NOFS);
1158                         em->start = cur_offset;
1159                         em->orig_start = em->start;
1160                         em->len = num_bytes;
1161                         em->block_len = num_bytes;
1162                         em->block_start = disk_bytenr;
1163                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1164                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1165                         while (1) {
1166                                 write_lock(&em_tree->lock);
1167                                 ret = add_extent_mapping(em_tree, em);
1168                                 write_unlock(&em_tree->lock);
1169                                 if (ret != -EEXIST) {
1170                                         free_extent_map(em);
1171                                         break;
1172                                 }
1173                                 btrfs_drop_extent_cache(inode, em->start,
1174                                                 em->start + em->len - 1, 0);
1175                         }
1176                         type = BTRFS_ORDERED_PREALLOC;
1177                 } else {
1178                         type = BTRFS_ORDERED_NOCOW;
1179                 }
1180
1181                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1182                                                num_bytes, num_bytes, type);
1183                 BUG_ON(ret);
1184
1185                 if (root->root_key.objectid ==
1186                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1187                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1188                                                       num_bytes);
1189                         BUG_ON(ret);
1190                 }
1191
1192                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1193                                 cur_offset, cur_offset + num_bytes - 1,
1194                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1195                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1196                                 EXTENT_SET_PRIVATE2);
1197                 cur_offset = extent_end;
1198                 if (cur_offset > end)
1199                         break;
1200         }
1201         btrfs_release_path(root, path);
1202
1203         if (cur_offset <= end && cow_start == (u64)-1)
1204                 cow_start = cur_offset;
1205         if (cow_start != (u64)-1) {
1206                 ret = cow_file_range(inode, locked_page, cow_start, end,
1207                                      page_started, nr_written, 1);
1208                 BUG_ON(ret);
1209         }
1210
1211         if (nolock) {
1212                 ret = btrfs_end_transaction_nolock(trans, root);
1213                 BUG_ON(ret);
1214         } else {
1215                 ret = btrfs_end_transaction(trans, root);
1216                 BUG_ON(ret);
1217         }
1218         btrfs_free_path(path);
1219         return 0;
1220 }
1221
1222 /*
1223  * extent_io.c call back to do delayed allocation processing
1224  */
1225 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1226                               u64 start, u64 end, int *page_started,
1227                               unsigned long *nr_written)
1228 {
1229         int ret;
1230         struct btrfs_root *root = BTRFS_I(inode)->root;
1231
1232         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1233                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1234                                          page_started, 1, nr_written);
1235         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1236                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1237                                          page_started, 0, nr_written);
1238         else if (!btrfs_test_opt(root, COMPRESS) &&
1239                  !(BTRFS_I(inode)->force_compress))
1240                 ret = cow_file_range(inode, locked_page, start, end,
1241                                       page_started, nr_written, 1);
1242         else
1243                 ret = cow_file_range_async(inode, locked_page, start, end,
1244                                            page_started, nr_written);
1245         return ret;
1246 }
1247
1248 static int btrfs_split_extent_hook(struct inode *inode,
1249                                    struct extent_state *orig, u64 split)
1250 {
1251         /* not delalloc, ignore it */
1252         if (!(orig->state & EXTENT_DELALLOC))
1253                 return 0;
1254
1255         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1256         return 0;
1257 }
1258
1259 /*
1260  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1261  * extents so we can keep track of new extents that are just merged onto old
1262  * extents, such as when we are doing sequential writes, so we can properly
1263  * account for the metadata space we'll need.
1264  */
1265 static int btrfs_merge_extent_hook(struct inode *inode,
1266                                    struct extent_state *new,
1267                                    struct extent_state *other)
1268 {
1269         /* not delalloc, ignore it */
1270         if (!(other->state & EXTENT_DELALLOC))
1271                 return 0;
1272
1273         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1274         return 0;
1275 }
1276
1277 /*
1278  * extent_io.c set_bit_hook, used to track delayed allocation
1279  * bytes in this file, and to maintain the list of inodes that
1280  * have pending delalloc work to be done.
1281  */
1282 static int btrfs_set_bit_hook(struct inode *inode,
1283                               struct extent_state *state, int *bits)
1284 {
1285
1286         /*
1287          * set_bit and clear bit hooks normally require _irqsave/restore
1288          * but in this case, we are only testeing for the DELALLOC
1289          * bit, which is only set or cleared with irqs on
1290          */
1291         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1292                 struct btrfs_root *root = BTRFS_I(inode)->root;
1293                 u64 len = state->end + 1 - state->start;
1294                 int do_list = (root->root_key.objectid !=
1295                                BTRFS_ROOT_TREE_OBJECTID);
1296
1297                 if (*bits & EXTENT_FIRST_DELALLOC)
1298                         *bits &= ~EXTENT_FIRST_DELALLOC;
1299                 else
1300                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1301
1302                 spin_lock(&root->fs_info->delalloc_lock);
1303                 BTRFS_I(inode)->delalloc_bytes += len;
1304                 root->fs_info->delalloc_bytes += len;
1305                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1306                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1307                                       &root->fs_info->delalloc_inodes);
1308                 }
1309                 spin_unlock(&root->fs_info->delalloc_lock);
1310         }
1311         return 0;
1312 }
1313
1314 /*
1315  * extent_io.c clear_bit_hook, see set_bit_hook for why
1316  */
1317 static int btrfs_clear_bit_hook(struct inode *inode,
1318                                 struct extent_state *state, int *bits)
1319 {
1320         /*
1321          * set_bit and clear bit hooks normally require _irqsave/restore
1322          * but in this case, we are only testeing for the DELALLOC
1323          * bit, which is only set or cleared with irqs on
1324          */
1325         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1326                 struct btrfs_root *root = BTRFS_I(inode)->root;
1327                 u64 len = state->end + 1 - state->start;
1328                 int do_list = (root->root_key.objectid !=
1329                                BTRFS_ROOT_TREE_OBJECTID);
1330
1331                 if (*bits & EXTENT_FIRST_DELALLOC)
1332                         *bits &= ~EXTENT_FIRST_DELALLOC;
1333                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1334                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1335
1336                 if (*bits & EXTENT_DO_ACCOUNTING)
1337                         btrfs_delalloc_release_metadata(inode, len);
1338
1339                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1340                     && do_list)
1341                         btrfs_free_reserved_data_space(inode, len);
1342
1343                 spin_lock(&root->fs_info->delalloc_lock);
1344                 root->fs_info->delalloc_bytes -= len;
1345                 BTRFS_I(inode)->delalloc_bytes -= len;
1346
1347                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1348                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1349                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1350                 }
1351                 spin_unlock(&root->fs_info->delalloc_lock);
1352         }
1353         return 0;
1354 }
1355
1356 /*
1357  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1358  * we don't create bios that span stripes or chunks
1359  */
1360 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1361                          size_t size, struct bio *bio,
1362                          unsigned long bio_flags)
1363 {
1364         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1365         struct btrfs_mapping_tree *map_tree;
1366         u64 logical = (u64)bio->bi_sector << 9;
1367         u64 length = 0;
1368         u64 map_length;
1369         int ret;
1370
1371         if (bio_flags & EXTENT_BIO_COMPRESSED)
1372                 return 0;
1373
1374         length = bio->bi_size;
1375         map_tree = &root->fs_info->mapping_tree;
1376         map_length = length;
1377         ret = btrfs_map_block(map_tree, READ, logical,
1378                               &map_length, NULL, 0);
1379
1380         if (map_length < length + size)
1381                 return 1;
1382         return ret;
1383 }
1384
1385 /*
1386  * in order to insert checksums into the metadata in large chunks,
1387  * we wait until bio submission time.   All the pages in the bio are
1388  * checksummed and sums are attached onto the ordered extent record.
1389  *
1390  * At IO completion time the cums attached on the ordered extent record
1391  * are inserted into the btree
1392  */
1393 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1394                                     struct bio *bio, int mirror_num,
1395                                     unsigned long bio_flags,
1396                                     u64 bio_offset)
1397 {
1398         struct btrfs_root *root = BTRFS_I(inode)->root;
1399         int ret = 0;
1400
1401         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1402         BUG_ON(ret);
1403         return 0;
1404 }
1405
1406 /*
1407  * in order to insert checksums into the metadata in large chunks,
1408  * we wait until bio submission time.   All the pages in the bio are
1409  * checksummed and sums are attached onto the ordered extent record.
1410  *
1411  * At IO completion time the cums attached on the ordered extent record
1412  * are inserted into the btree
1413  */
1414 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1415                           int mirror_num, unsigned long bio_flags,
1416                           u64 bio_offset)
1417 {
1418         struct btrfs_root *root = BTRFS_I(inode)->root;
1419         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1420 }
1421
1422 /*
1423  * extent_io.c submission hook. This does the right thing for csum calculation
1424  * on write, or reading the csums from the tree before a read
1425  */
1426 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1427                           int mirror_num, unsigned long bio_flags,
1428                           u64 bio_offset)
1429 {
1430         struct btrfs_root *root = BTRFS_I(inode)->root;
1431         int ret = 0;
1432         int skip_sum;
1433
1434         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1435
1436         if (root == root->fs_info->tree_root)
1437                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1438         else
1439                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1440         BUG_ON(ret);
1441
1442         if (!(rw & REQ_WRITE)) {
1443                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1444                         return btrfs_submit_compressed_read(inode, bio,
1445                                                     mirror_num, bio_flags);
1446                 } else if (!skip_sum)
1447                         btrfs_lookup_bio_sums(root, inode, bio, NULL);
1448                 goto mapit;
1449         } else if (!skip_sum) {
1450                 /* csum items have already been cloned */
1451                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1452                         goto mapit;
1453                 /* we're doing a write, do the async checksumming */
1454                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1455                                    inode, rw, bio, mirror_num,
1456                                    bio_flags, bio_offset,
1457                                    __btrfs_submit_bio_start,
1458                                    __btrfs_submit_bio_done);
1459         }
1460
1461 mapit:
1462         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1463 }
1464
1465 /*
1466  * given a list of ordered sums record them in the inode.  This happens
1467  * at IO completion time based on sums calculated at bio submission time.
1468  */
1469 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1470                              struct inode *inode, u64 file_offset,
1471                              struct list_head *list)
1472 {
1473         struct btrfs_ordered_sum *sum;
1474
1475         btrfs_set_trans_block_group(trans, inode);
1476
1477         list_for_each_entry(sum, list, list) {
1478                 btrfs_csum_file_blocks(trans,
1479                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1480         }
1481         return 0;
1482 }
1483
1484 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1485                               struct extent_state **cached_state)
1486 {
1487         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1488                 WARN_ON(1);
1489         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1490                                    cached_state, GFP_NOFS);
1491 }
1492
1493 /* see btrfs_writepage_start_hook for details on why this is required */
1494 struct btrfs_writepage_fixup {
1495         struct page *page;
1496         struct btrfs_work work;
1497 };
1498
1499 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1500 {
1501         struct btrfs_writepage_fixup *fixup;
1502         struct btrfs_ordered_extent *ordered;
1503         struct extent_state *cached_state = NULL;
1504         struct page *page;
1505         struct inode *inode;
1506         u64 page_start;
1507         u64 page_end;
1508
1509         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1510         page = fixup->page;
1511 again:
1512         lock_page(page);
1513         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1514                 ClearPageChecked(page);
1515                 goto out_page;
1516         }
1517
1518         inode = page->mapping->host;
1519         page_start = page_offset(page);
1520         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1521
1522         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1523                          &cached_state, GFP_NOFS);
1524
1525         /* already ordered? We're done */
1526         if (PagePrivate2(page))
1527                 goto out;
1528
1529         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1530         if (ordered) {
1531                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1532                                      page_end, &cached_state, GFP_NOFS);
1533                 unlock_page(page);
1534                 btrfs_start_ordered_extent(inode, ordered, 1);
1535                 goto again;
1536         }
1537
1538         BUG();
1539         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1540         ClearPageChecked(page);
1541 out:
1542         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1543                              &cached_state, GFP_NOFS);
1544 out_page:
1545         unlock_page(page);
1546         page_cache_release(page);
1547 }
1548
1549 /*
1550  * There are a few paths in the higher layers of the kernel that directly
1551  * set the page dirty bit without asking the filesystem if it is a
1552  * good idea.  This causes problems because we want to make sure COW
1553  * properly happens and the data=ordered rules are followed.
1554  *
1555  * In our case any range that doesn't have the ORDERED bit set
1556  * hasn't been properly setup for IO.  We kick off an async process
1557  * to fix it up.  The async helper will wait for ordered extents, set
1558  * the delalloc bit and make it safe to write the page.
1559  */
1560 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1561 {
1562         struct inode *inode = page->mapping->host;
1563         struct btrfs_writepage_fixup *fixup;
1564         struct btrfs_root *root = BTRFS_I(inode)->root;
1565
1566         /* this page is properly in the ordered list */
1567         if (TestClearPagePrivate2(page))
1568                 return 0;
1569
1570         if (PageChecked(page))
1571                 return -EAGAIN;
1572
1573         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1574         if (!fixup)
1575                 return -EAGAIN;
1576
1577         SetPageChecked(page);
1578         page_cache_get(page);
1579         fixup->work.func = btrfs_writepage_fixup_worker;
1580         fixup->page = page;
1581         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1582         return -EAGAIN;
1583 }
1584
1585 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1586                                        struct inode *inode, u64 file_pos,
1587                                        u64 disk_bytenr, u64 disk_num_bytes,
1588                                        u64 num_bytes, u64 ram_bytes,
1589                                        u8 compression, u8 encryption,
1590                                        u16 other_encoding, int extent_type)
1591 {
1592         struct btrfs_root *root = BTRFS_I(inode)->root;
1593         struct btrfs_file_extent_item *fi;
1594         struct btrfs_path *path;
1595         struct extent_buffer *leaf;
1596         struct btrfs_key ins;
1597         u64 hint;
1598         int ret;
1599
1600         path = btrfs_alloc_path();
1601         BUG_ON(!path);
1602
1603         path->leave_spinning = 1;
1604
1605         /*
1606          * we may be replacing one extent in the tree with another.
1607          * The new extent is pinned in the extent map, and we don't want
1608          * to drop it from the cache until it is completely in the btree.
1609          *
1610          * So, tell btrfs_drop_extents to leave this extent in the cache.
1611          * the caller is expected to unpin it and allow it to be merged
1612          * with the others.
1613          */
1614         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1615                                  &hint, 0);
1616         BUG_ON(ret);
1617
1618         ins.objectid = inode->i_ino;
1619         ins.offset = file_pos;
1620         ins.type = BTRFS_EXTENT_DATA_KEY;
1621         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1622         BUG_ON(ret);
1623         leaf = path->nodes[0];
1624         fi = btrfs_item_ptr(leaf, path->slots[0],
1625                             struct btrfs_file_extent_item);
1626         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1627         btrfs_set_file_extent_type(leaf, fi, extent_type);
1628         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1629         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1630         btrfs_set_file_extent_offset(leaf, fi, 0);
1631         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1632         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1633         btrfs_set_file_extent_compression(leaf, fi, compression);
1634         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1635         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1636
1637         btrfs_unlock_up_safe(path, 1);
1638         btrfs_set_lock_blocking(leaf);
1639
1640         btrfs_mark_buffer_dirty(leaf);
1641
1642         inode_add_bytes(inode, num_bytes);
1643
1644         ins.objectid = disk_bytenr;
1645         ins.offset = disk_num_bytes;
1646         ins.type = BTRFS_EXTENT_ITEM_KEY;
1647         ret = btrfs_alloc_reserved_file_extent(trans, root,
1648                                         root->root_key.objectid,
1649                                         inode->i_ino, file_pos, &ins);
1650         BUG_ON(ret);
1651         btrfs_free_path(path);
1652
1653         return 0;
1654 }
1655
1656 /*
1657  * helper function for btrfs_finish_ordered_io, this
1658  * just reads in some of the csum leaves to prime them into ram
1659  * before we start the transaction.  It limits the amount of btree
1660  * reads required while inside the transaction.
1661  */
1662 /* as ordered data IO finishes, this gets called so we can finish
1663  * an ordered extent if the range of bytes in the file it covers are
1664  * fully written.
1665  */
1666 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1667 {
1668         struct btrfs_root *root = BTRFS_I(inode)->root;
1669         struct btrfs_trans_handle *trans = NULL;
1670         struct btrfs_ordered_extent *ordered_extent = NULL;
1671         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1672         struct extent_state *cached_state = NULL;
1673         int compressed = 0;
1674         int ret;
1675         bool nolock = false;
1676
1677         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1678                                              end - start + 1);
1679         if (!ret)
1680                 return 0;
1681         BUG_ON(!ordered_extent);
1682
1683         nolock = (root == root->fs_info->tree_root);
1684
1685         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1686                 BUG_ON(!list_empty(&ordered_extent->list));
1687                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1688                 if (!ret) {
1689                         if (nolock)
1690                                 trans = btrfs_join_transaction_nolock(root, 1);
1691                         else
1692                                 trans = btrfs_join_transaction(root, 1);
1693                         BUG_ON(!trans);
1694                         btrfs_set_trans_block_group(trans, inode);
1695                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1696                         ret = btrfs_update_inode(trans, root, inode);
1697                         BUG_ON(ret);
1698                 }
1699                 goto out;
1700         }
1701
1702         lock_extent_bits(io_tree, ordered_extent->file_offset,
1703                          ordered_extent->file_offset + ordered_extent->len - 1,
1704                          0, &cached_state, GFP_NOFS);
1705
1706         if (nolock)
1707                 trans = btrfs_join_transaction_nolock(root, 1);
1708         else
1709                 trans = btrfs_join_transaction(root, 1);
1710         btrfs_set_trans_block_group(trans, inode);
1711         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1712
1713         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1714                 compressed = 1;
1715         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1716                 BUG_ON(compressed);
1717                 ret = btrfs_mark_extent_written(trans, inode,
1718                                                 ordered_extent->file_offset,
1719                                                 ordered_extent->file_offset +
1720                                                 ordered_extent->len);
1721                 BUG_ON(ret);
1722         } else {
1723                 BUG_ON(root == root->fs_info->tree_root);
1724                 ret = insert_reserved_file_extent(trans, inode,
1725                                                 ordered_extent->file_offset,
1726                                                 ordered_extent->start,
1727                                                 ordered_extent->disk_len,
1728                                                 ordered_extent->len,
1729                                                 ordered_extent->len,
1730                                                 compressed, 0, 0,
1731                                                 BTRFS_FILE_EXTENT_REG);
1732                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1733                                    ordered_extent->file_offset,
1734                                    ordered_extent->len);
1735                 BUG_ON(ret);
1736         }
1737         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1738                              ordered_extent->file_offset +
1739                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1740
1741         add_pending_csums(trans, inode, ordered_extent->file_offset,
1742                           &ordered_extent->list);
1743
1744         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1745         ret = btrfs_update_inode(trans, root, inode);
1746         BUG_ON(ret);
1747 out:
1748         if (nolock) {
1749                 if (trans)
1750                         btrfs_end_transaction_nolock(trans, root);
1751         } else {
1752                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1753                 if (trans)
1754                         btrfs_end_transaction(trans, root);
1755         }
1756
1757         /* once for us */
1758         btrfs_put_ordered_extent(ordered_extent);
1759         /* once for the tree */
1760         btrfs_put_ordered_extent(ordered_extent);
1761
1762         return 0;
1763 }
1764
1765 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1766                                 struct extent_state *state, int uptodate)
1767 {
1768         ClearPagePrivate2(page);
1769         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1770 }
1771
1772 /*
1773  * When IO fails, either with EIO or csum verification fails, we
1774  * try other mirrors that might have a good copy of the data.  This
1775  * io_failure_record is used to record state as we go through all the
1776  * mirrors.  If another mirror has good data, the page is set up to date
1777  * and things continue.  If a good mirror can't be found, the original
1778  * bio end_io callback is called to indicate things have failed.
1779  */
1780 struct io_failure_record {
1781         struct page *page;
1782         u64 start;
1783         u64 len;
1784         u64 logical;
1785         unsigned long bio_flags;
1786         int last_mirror;
1787 };
1788
1789 static int btrfs_io_failed_hook(struct bio *failed_bio,
1790                          struct page *page, u64 start, u64 end,
1791                          struct extent_state *state)
1792 {
1793         struct io_failure_record *failrec = NULL;
1794         u64 private;
1795         struct extent_map *em;
1796         struct inode *inode = page->mapping->host;
1797         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1798         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1799         struct bio *bio;
1800         int num_copies;
1801         int ret;
1802         int rw;
1803         u64 logical;
1804
1805         ret = get_state_private(failure_tree, start, &private);
1806         if (ret) {
1807                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1808                 if (!failrec)
1809                         return -ENOMEM;
1810                 failrec->start = start;
1811                 failrec->len = end - start + 1;
1812                 failrec->last_mirror = 0;
1813                 failrec->bio_flags = 0;
1814
1815                 read_lock(&em_tree->lock);
1816                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1817                 if (em->start > start || em->start + em->len < start) {
1818                         free_extent_map(em);
1819                         em = NULL;
1820                 }
1821                 read_unlock(&em_tree->lock);
1822
1823                 if (!em || IS_ERR(em)) {
1824                         kfree(failrec);
1825                         return -EIO;
1826                 }
1827                 logical = start - em->start;
1828                 logical = em->block_start + logical;
1829                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1830                         logical = em->block_start;
1831                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1832                 }
1833                 failrec->logical = logical;
1834                 free_extent_map(em);
1835                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1836                                 EXTENT_DIRTY, GFP_NOFS);
1837                 set_state_private(failure_tree, start,
1838                                  (u64)(unsigned long)failrec);
1839         } else {
1840                 failrec = (struct io_failure_record *)(unsigned long)private;
1841         }
1842         num_copies = btrfs_num_copies(
1843                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1844                               failrec->logical, failrec->len);
1845         failrec->last_mirror++;
1846         if (!state) {
1847                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1848                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1849                                                     failrec->start,
1850                                                     EXTENT_LOCKED);
1851                 if (state && state->start != failrec->start)
1852                         state = NULL;
1853                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1854         }
1855         if (!state || failrec->last_mirror > num_copies) {
1856                 set_state_private(failure_tree, failrec->start, 0);
1857                 clear_extent_bits(failure_tree, failrec->start,
1858                                   failrec->start + failrec->len - 1,
1859                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1860                 kfree(failrec);
1861                 return -EIO;
1862         }
1863         bio = bio_alloc(GFP_NOFS, 1);
1864         bio->bi_private = state;
1865         bio->bi_end_io = failed_bio->bi_end_io;
1866         bio->bi_sector = failrec->logical >> 9;
1867         bio->bi_bdev = failed_bio->bi_bdev;
1868         bio->bi_size = 0;
1869
1870         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1871         if (failed_bio->bi_rw & REQ_WRITE)
1872                 rw = WRITE;
1873         else
1874                 rw = READ;
1875
1876         BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1877                                                       failrec->last_mirror,
1878                                                       failrec->bio_flags, 0);
1879         return 0;
1880 }
1881
1882 /*
1883  * each time an IO finishes, we do a fast check in the IO failure tree
1884  * to see if we need to process or clean up an io_failure_record
1885  */
1886 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1887 {
1888         u64 private;
1889         u64 private_failure;
1890         struct io_failure_record *failure;
1891         int ret;
1892
1893         private = 0;
1894         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1895                              (u64)-1, 1, EXTENT_DIRTY)) {
1896                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1897                                         start, &private_failure);
1898                 if (ret == 0) {
1899                         failure = (struct io_failure_record *)(unsigned long)
1900                                    private_failure;
1901                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1902                                           failure->start, 0);
1903                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1904                                           failure->start,
1905                                           failure->start + failure->len - 1,
1906                                           EXTENT_DIRTY | EXTENT_LOCKED,
1907                                           GFP_NOFS);
1908                         kfree(failure);
1909                 }
1910         }
1911         return 0;
1912 }
1913
1914 /*
1915  * when reads are done, we need to check csums to verify the data is correct
1916  * if there's a match, we allow the bio to finish.  If not, we go through
1917  * the io_failure_record routines to find good copies
1918  */
1919 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1920                                struct extent_state *state)
1921 {
1922         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1923         struct inode *inode = page->mapping->host;
1924         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1925         char *kaddr;
1926         u64 private = ~(u32)0;
1927         int ret;
1928         struct btrfs_root *root = BTRFS_I(inode)->root;
1929         u32 csum = ~(u32)0;
1930
1931         if (PageChecked(page)) {
1932                 ClearPageChecked(page);
1933                 goto good;
1934         }
1935
1936         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1937                 return 0;
1938
1939         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1940             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1941                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1942                                   GFP_NOFS);
1943                 return 0;
1944         }
1945
1946         if (state && state->start == start) {
1947                 private = state->private;
1948                 ret = 0;
1949         } else {
1950                 ret = get_state_private(io_tree, start, &private);
1951         }
1952         kaddr = kmap_atomic(page, KM_USER0);
1953         if (ret)
1954                 goto zeroit;
1955
1956         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1957         btrfs_csum_final(csum, (char *)&csum);
1958         if (csum != private)
1959                 goto zeroit;
1960
1961         kunmap_atomic(kaddr, KM_USER0);
1962 good:
1963         /* if the io failure tree for this inode is non-empty,
1964          * check to see if we've recovered from a failed IO
1965          */
1966         btrfs_clean_io_failures(inode, start);
1967         return 0;
1968
1969 zeroit:
1970         if (printk_ratelimit()) {
1971                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
1972                        "private %llu\n", page->mapping->host->i_ino,
1973                        (unsigned long long)start, csum,
1974                        (unsigned long long)private);
1975         }
1976         memset(kaddr + offset, 1, end - start + 1);
1977         flush_dcache_page(page);
1978         kunmap_atomic(kaddr, KM_USER0);
1979         if (private == 0)
1980                 return 0;
1981         return -EIO;
1982 }
1983
1984 struct delayed_iput {
1985         struct list_head list;
1986         struct inode *inode;
1987 };
1988
1989 void btrfs_add_delayed_iput(struct inode *inode)
1990 {
1991         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
1992         struct delayed_iput *delayed;
1993
1994         if (atomic_add_unless(&inode->i_count, -1, 1))
1995                 return;
1996
1997         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
1998         delayed->inode = inode;
1999
2000         spin_lock(&fs_info->delayed_iput_lock);
2001         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2002         spin_unlock(&fs_info->delayed_iput_lock);
2003 }
2004
2005 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2006 {
2007         LIST_HEAD(list);
2008         struct btrfs_fs_info *fs_info = root->fs_info;
2009         struct delayed_iput *delayed;
2010         int empty;
2011
2012         spin_lock(&fs_info->delayed_iput_lock);
2013         empty = list_empty(&fs_info->delayed_iputs);
2014         spin_unlock(&fs_info->delayed_iput_lock);
2015         if (empty)
2016                 return;
2017
2018         down_read(&root->fs_info->cleanup_work_sem);
2019         spin_lock(&fs_info->delayed_iput_lock);
2020         list_splice_init(&fs_info->delayed_iputs, &list);
2021         spin_unlock(&fs_info->delayed_iput_lock);
2022
2023         while (!list_empty(&list)) {
2024                 delayed = list_entry(list.next, struct delayed_iput, list);
2025                 list_del(&delayed->list);
2026                 iput(delayed->inode);
2027                 kfree(delayed);
2028         }
2029         up_read(&root->fs_info->cleanup_work_sem);
2030 }
2031
2032 /*
2033  * calculate extra metadata reservation when snapshotting a subvolume
2034  * contains orphan files.
2035  */
2036 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2037                                 struct btrfs_pending_snapshot *pending,
2038                                 u64 *bytes_to_reserve)
2039 {
2040         struct btrfs_root *root;
2041         struct btrfs_block_rsv *block_rsv;
2042         u64 num_bytes;
2043         int index;
2044
2045         root = pending->root;
2046         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2047                 return;
2048
2049         block_rsv = root->orphan_block_rsv;
2050
2051         /* orphan block reservation for the snapshot */
2052         num_bytes = block_rsv->size;
2053
2054         /*
2055          * after the snapshot is created, COWing tree blocks may use more
2056          * space than it frees. So we should make sure there is enough
2057          * reserved space.
2058          */
2059         index = trans->transid & 0x1;
2060         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2061                 num_bytes += block_rsv->size -
2062                              (block_rsv->reserved + block_rsv->freed[index]);
2063         }
2064
2065         *bytes_to_reserve += num_bytes;
2066 }
2067
2068 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2069                                 struct btrfs_pending_snapshot *pending)
2070 {
2071         struct btrfs_root *root = pending->root;
2072         struct btrfs_root *snap = pending->snap;
2073         struct btrfs_block_rsv *block_rsv;
2074         u64 num_bytes;
2075         int index;
2076         int ret;
2077
2078         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2079                 return;
2080
2081         /* refill source subvolume's orphan block reservation */
2082         block_rsv = root->orphan_block_rsv;
2083         index = trans->transid & 0x1;
2084         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2085                 num_bytes = block_rsv->size -
2086                             (block_rsv->reserved + block_rsv->freed[index]);
2087                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2088                                               root->orphan_block_rsv,
2089                                               num_bytes);
2090                 BUG_ON(ret);
2091         }
2092
2093         /* setup orphan block reservation for the snapshot */
2094         block_rsv = btrfs_alloc_block_rsv(snap);
2095         BUG_ON(!block_rsv);
2096
2097         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2098         snap->orphan_block_rsv = block_rsv;
2099
2100         num_bytes = root->orphan_block_rsv->size;
2101         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2102                                       block_rsv, num_bytes);
2103         BUG_ON(ret);
2104
2105 #if 0
2106         /* insert orphan item for the snapshot */
2107         WARN_ON(!root->orphan_item_inserted);
2108         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2109                                        snap->root_key.objectid);
2110         BUG_ON(ret);
2111         snap->orphan_item_inserted = 1;
2112 #endif
2113 }
2114
2115 enum btrfs_orphan_cleanup_state {
2116         ORPHAN_CLEANUP_STARTED  = 1,
2117         ORPHAN_CLEANUP_DONE     = 2,
2118 };
2119
2120 /*
2121  * This is called in transaction commmit time. If there are no orphan
2122  * files in the subvolume, it removes orphan item and frees block_rsv
2123  * structure.
2124  */
2125 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2126                               struct btrfs_root *root)
2127 {
2128         int ret;
2129
2130         if (!list_empty(&root->orphan_list) ||
2131             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2132                 return;
2133
2134         if (root->orphan_item_inserted &&
2135             btrfs_root_refs(&root->root_item) > 0) {
2136                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2137                                             root->root_key.objectid);
2138                 BUG_ON(ret);
2139                 root->orphan_item_inserted = 0;
2140         }
2141
2142         if (root->orphan_block_rsv) {
2143                 WARN_ON(root->orphan_block_rsv->size > 0);
2144                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2145                 root->orphan_block_rsv = NULL;
2146         }
2147 }
2148
2149 /*
2150  * This creates an orphan entry for the given inode in case something goes
2151  * wrong in the middle of an unlink/truncate.
2152  *
2153  * NOTE: caller of this function should reserve 5 units of metadata for
2154  *       this function.
2155  */
2156 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2157 {
2158         struct btrfs_root *root = BTRFS_I(inode)->root;
2159         struct btrfs_block_rsv *block_rsv = NULL;
2160         int reserve = 0;
2161         int insert = 0;
2162         int ret;
2163
2164         if (!root->orphan_block_rsv) {
2165                 block_rsv = btrfs_alloc_block_rsv(root);
2166                 BUG_ON(!block_rsv);
2167         }
2168
2169         spin_lock(&root->orphan_lock);
2170         if (!root->orphan_block_rsv) {
2171                 root->orphan_block_rsv = block_rsv;
2172         } else if (block_rsv) {
2173                 btrfs_free_block_rsv(root, block_rsv);
2174                 block_rsv = NULL;
2175         }
2176
2177         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2178                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2179 #if 0
2180                 /*
2181                  * For proper ENOSPC handling, we should do orphan
2182                  * cleanup when mounting. But this introduces backward
2183                  * compatibility issue.
2184                  */
2185                 if (!xchg(&root->orphan_item_inserted, 1))
2186                         insert = 2;
2187                 else
2188                         insert = 1;
2189 #endif
2190                 insert = 1;
2191         } else {
2192                 WARN_ON(!BTRFS_I(inode)->orphan_meta_reserved);
2193         }
2194
2195         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2196                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2197                 reserve = 1;
2198         }
2199         spin_unlock(&root->orphan_lock);
2200
2201         if (block_rsv)
2202                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2203
2204         /* grab metadata reservation from transaction handle */
2205         if (reserve) {
2206                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2207                 BUG_ON(ret);
2208         }
2209
2210         /* insert an orphan item to track this unlinked/truncated file */
2211         if (insert >= 1) {
2212                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2213                 BUG_ON(ret);
2214         }
2215
2216         /* insert an orphan item to track subvolume contains orphan files */
2217         if (insert >= 2) {
2218                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2219                                                root->root_key.objectid);
2220                 BUG_ON(ret);
2221         }
2222         return 0;
2223 }
2224
2225 /*
2226  * We have done the truncate/delete so we can go ahead and remove the orphan
2227  * item for this particular inode.
2228  */
2229 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2230 {
2231         struct btrfs_root *root = BTRFS_I(inode)->root;
2232         int delete_item = 0;
2233         int release_rsv = 0;
2234         int ret = 0;
2235
2236         spin_lock(&root->orphan_lock);
2237         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2238                 list_del_init(&BTRFS_I(inode)->i_orphan);
2239                 delete_item = 1;
2240         }
2241
2242         if (BTRFS_I(inode)->orphan_meta_reserved) {
2243                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2244                 release_rsv = 1;
2245         }
2246         spin_unlock(&root->orphan_lock);
2247
2248         if (trans && delete_item) {
2249                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2250                 BUG_ON(ret);
2251         }
2252
2253         if (release_rsv)
2254                 btrfs_orphan_release_metadata(inode);
2255
2256         return 0;
2257 }
2258
2259 /*
2260  * this cleans up any orphans that may be left on the list from the last use
2261  * of this root.
2262  */
2263 void btrfs_orphan_cleanup(struct btrfs_root *root)
2264 {
2265         struct btrfs_path *path;
2266         struct extent_buffer *leaf;
2267         struct btrfs_key key, found_key;
2268         struct btrfs_trans_handle *trans;
2269         struct inode *inode;
2270         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2271
2272         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2273                 return;
2274
2275         path = btrfs_alloc_path();
2276         BUG_ON(!path);
2277         path->reada = -1;
2278
2279         key.objectid = BTRFS_ORPHAN_OBJECTID;
2280         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2281         key.offset = (u64)-1;
2282
2283         while (1) {
2284                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2285                 if (ret < 0) {
2286                         printk(KERN_ERR "Error searching slot for orphan: %d"
2287                                "\n", ret);
2288                         break;
2289                 }
2290
2291                 /*
2292                  * if ret == 0 means we found what we were searching for, which
2293                  * is weird, but possible, so only screw with path if we didnt
2294                  * find the key and see if we have stuff that matches
2295                  */
2296                 if (ret > 0) {
2297                         if (path->slots[0] == 0)
2298                                 break;
2299                         path->slots[0]--;
2300                 }
2301
2302                 /* pull out the item */
2303                 leaf = path->nodes[0];
2304                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2305
2306                 /* make sure the item matches what we want */
2307                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2308                         break;
2309                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2310                         break;
2311
2312                 /* release the path since we're done with it */
2313                 btrfs_release_path(root, path);
2314
2315                 /*
2316                  * this is where we are basically btrfs_lookup, without the
2317                  * crossing root thing.  we store the inode number in the
2318                  * offset of the orphan item.
2319                  */
2320                 found_key.objectid = found_key.offset;
2321                 found_key.type = BTRFS_INODE_ITEM_KEY;
2322                 found_key.offset = 0;
2323                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2324                 BUG_ON(IS_ERR(inode));
2325
2326                 /*
2327                  * add this inode to the orphan list so btrfs_orphan_del does
2328                  * the proper thing when we hit it
2329                  */
2330                 spin_lock(&root->orphan_lock);
2331                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2332                 spin_unlock(&root->orphan_lock);
2333
2334                 /*
2335                  * if this is a bad inode, means we actually succeeded in
2336                  * removing the inode, but not the orphan record, which means
2337                  * we need to manually delete the orphan since iput will just
2338                  * do a destroy_inode
2339                  */
2340                 if (is_bad_inode(inode)) {
2341                         trans = btrfs_start_transaction(root, 0);
2342                         btrfs_orphan_del(trans, inode);
2343                         btrfs_end_transaction(trans, root);
2344                         iput(inode);
2345                         continue;
2346                 }
2347
2348                 /* if we have links, this was a truncate, lets do that */
2349                 if (inode->i_nlink) {
2350                         nr_truncate++;
2351                         btrfs_truncate(inode);
2352                 } else {
2353                         nr_unlink++;
2354                 }
2355
2356                 /* this will do delete_inode and everything for us */
2357                 iput(inode);
2358         }
2359         btrfs_free_path(path);
2360
2361         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2362
2363         if (root->orphan_block_rsv)
2364                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2365                                         (u64)-1);
2366
2367         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2368                 trans = btrfs_join_transaction(root, 1);
2369                 btrfs_end_transaction(trans, root);
2370         }
2371
2372         if (nr_unlink)
2373                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2374         if (nr_truncate)
2375                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2376 }
2377
2378 /*
2379  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2380  * don't find any xattrs, we know there can't be any acls.
2381  *
2382  * slot is the slot the inode is in, objectid is the objectid of the inode
2383  */
2384 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2385                                           int slot, u64 objectid)
2386 {
2387         u32 nritems = btrfs_header_nritems(leaf);
2388         struct btrfs_key found_key;
2389         int scanned = 0;
2390
2391         slot++;
2392         while (slot < nritems) {
2393                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2394
2395                 /* we found a different objectid, there must not be acls */
2396                 if (found_key.objectid != objectid)
2397                         return 0;
2398
2399                 /* we found an xattr, assume we've got an acl */
2400                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2401                         return 1;
2402
2403                 /*
2404                  * we found a key greater than an xattr key, there can't
2405                  * be any acls later on
2406                  */
2407                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2408                         return 0;
2409
2410                 slot++;
2411                 scanned++;
2412
2413                 /*
2414                  * it goes inode, inode backrefs, xattrs, extents,
2415                  * so if there are a ton of hard links to an inode there can
2416                  * be a lot of backrefs.  Don't waste time searching too hard,
2417                  * this is just an optimization
2418                  */
2419                 if (scanned >= 8)
2420                         break;
2421         }
2422         /* we hit the end of the leaf before we found an xattr or
2423          * something larger than an xattr.  We have to assume the inode
2424          * has acls
2425          */
2426         return 1;
2427 }
2428
2429 /*
2430  * read an inode from the btree into the in-memory inode
2431  */
2432 static void btrfs_read_locked_inode(struct inode *inode)
2433 {
2434         struct btrfs_path *path;
2435         struct extent_buffer *leaf;
2436         struct btrfs_inode_item *inode_item;
2437         struct btrfs_timespec *tspec;
2438         struct btrfs_root *root = BTRFS_I(inode)->root;
2439         struct btrfs_key location;
2440         int maybe_acls;
2441         u64 alloc_group_block;
2442         u32 rdev;
2443         int ret;
2444
2445         path = btrfs_alloc_path();
2446         BUG_ON(!path);
2447         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2448
2449         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2450         if (ret)
2451                 goto make_bad;
2452
2453         leaf = path->nodes[0];
2454         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2455                                     struct btrfs_inode_item);
2456
2457         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2458         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2459         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2460         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2461         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2462
2463         tspec = btrfs_inode_atime(inode_item);
2464         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2465         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2466
2467         tspec = btrfs_inode_mtime(inode_item);
2468         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2469         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2470
2471         tspec = btrfs_inode_ctime(inode_item);
2472         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2473         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2474
2475         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2476         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2477         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2478         inode->i_generation = BTRFS_I(inode)->generation;
2479         inode->i_rdev = 0;
2480         rdev = btrfs_inode_rdev(leaf, inode_item);
2481
2482         BTRFS_I(inode)->index_cnt = (u64)-1;
2483         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2484
2485         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2486
2487         /*
2488          * try to precache a NULL acl entry for files that don't have
2489          * any xattrs or acls
2490          */
2491         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2492         if (!maybe_acls)
2493                 cache_no_acl(inode);
2494
2495         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2496                                                 alloc_group_block, 0);
2497         btrfs_free_path(path);
2498         inode_item = NULL;
2499
2500         switch (inode->i_mode & S_IFMT) {
2501         case S_IFREG:
2502                 inode->i_mapping->a_ops = &btrfs_aops;
2503                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2504                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2505                 inode->i_fop = &btrfs_file_operations;
2506                 inode->i_op = &btrfs_file_inode_operations;
2507                 break;
2508         case S_IFDIR:
2509                 inode->i_fop = &btrfs_dir_file_operations;
2510                 if (root == root->fs_info->tree_root)
2511                         inode->i_op = &btrfs_dir_ro_inode_operations;
2512                 else
2513                         inode->i_op = &btrfs_dir_inode_operations;
2514                 break;
2515         case S_IFLNK:
2516                 inode->i_op = &btrfs_symlink_inode_operations;
2517                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2518                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2519                 break;
2520         default:
2521                 inode->i_op = &btrfs_special_inode_operations;
2522                 init_special_inode(inode, inode->i_mode, rdev);
2523                 break;
2524         }
2525
2526         btrfs_update_iflags(inode);
2527         return;
2528
2529 make_bad:
2530         btrfs_free_path(path);
2531         make_bad_inode(inode);
2532 }
2533
2534 /*
2535  * given a leaf and an inode, copy the inode fields into the leaf
2536  */
2537 static void fill_inode_item(struct btrfs_trans_handle *trans,
2538                             struct extent_buffer *leaf,
2539                             struct btrfs_inode_item *item,
2540                             struct inode *inode)
2541 {
2542         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2543         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2544         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2545         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2546         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2547
2548         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2549                                inode->i_atime.tv_sec);
2550         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2551                                 inode->i_atime.tv_nsec);
2552
2553         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2554                                inode->i_mtime.tv_sec);
2555         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2556                                 inode->i_mtime.tv_nsec);
2557
2558         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2559                                inode->i_ctime.tv_sec);
2560         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2561                                 inode->i_ctime.tv_nsec);
2562
2563         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2564         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2565         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2566         btrfs_set_inode_transid(leaf, item, trans->transid);
2567         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2568         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2569         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2570 }
2571
2572 /*
2573  * copy everything in the in-memory inode into the btree.
2574  */
2575 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2576                                 struct btrfs_root *root, struct inode *inode)
2577 {
2578         struct btrfs_inode_item *inode_item;
2579         struct btrfs_path *path;
2580         struct extent_buffer *leaf;
2581         int ret;
2582
2583         path = btrfs_alloc_path();
2584         BUG_ON(!path);
2585         path->leave_spinning = 1;
2586         ret = btrfs_lookup_inode(trans, root, path,
2587                                  &BTRFS_I(inode)->location, 1);
2588         if (ret) {
2589                 if (ret > 0)
2590                         ret = -ENOENT;
2591                 goto failed;
2592         }
2593
2594         btrfs_unlock_up_safe(path, 1);
2595         leaf = path->nodes[0];
2596         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2597                                   struct btrfs_inode_item);
2598
2599         fill_inode_item(trans, leaf, inode_item, inode);
2600         btrfs_mark_buffer_dirty(leaf);
2601         btrfs_set_inode_last_trans(trans, inode);
2602         ret = 0;
2603 failed:
2604         btrfs_free_path(path);
2605         return ret;
2606 }
2607
2608
2609 /*
2610  * unlink helper that gets used here in inode.c and in the tree logging
2611  * recovery code.  It remove a link in a directory with a given name, and
2612  * also drops the back refs in the inode to the directory
2613  */
2614 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2615                        struct btrfs_root *root,
2616                        struct inode *dir, struct inode *inode,
2617                        const char *name, int name_len)
2618 {
2619         struct btrfs_path *path;
2620         int ret = 0;
2621         struct extent_buffer *leaf;
2622         struct btrfs_dir_item *di;
2623         struct btrfs_key key;
2624         u64 index;
2625
2626         path = btrfs_alloc_path();
2627         if (!path) {
2628                 ret = -ENOMEM;
2629                 goto err;
2630         }
2631
2632         path->leave_spinning = 1;
2633         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2634                                     name, name_len, -1);
2635         if (IS_ERR(di)) {
2636                 ret = PTR_ERR(di);
2637                 goto err;
2638         }
2639         if (!di) {
2640                 ret = -ENOENT;
2641                 goto err;
2642         }
2643         leaf = path->nodes[0];
2644         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2645         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2646         if (ret)
2647                 goto err;
2648         btrfs_release_path(root, path);
2649
2650         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2651                                   inode->i_ino,
2652                                   dir->i_ino, &index);
2653         if (ret) {
2654                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2655                        "inode %lu parent %lu\n", name_len, name,
2656                        inode->i_ino, dir->i_ino);
2657                 goto err;
2658         }
2659
2660         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2661                                          index, name, name_len, -1);
2662         if (IS_ERR(di)) {
2663                 ret = PTR_ERR(di);
2664                 goto err;
2665         }
2666         if (!di) {
2667                 ret = -ENOENT;
2668                 goto err;
2669         }
2670         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2671         btrfs_release_path(root, path);
2672
2673         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2674                                          inode, dir->i_ino);
2675         BUG_ON(ret != 0 && ret != -ENOENT);
2676
2677         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2678                                            dir, index);
2679         if (ret == -ENOENT)
2680                 ret = 0;
2681 err:
2682         btrfs_free_path(path);
2683         if (ret)
2684                 goto out;
2685
2686         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2687         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2688         btrfs_update_inode(trans, root, dir);
2689         btrfs_drop_nlink(inode);
2690         ret = btrfs_update_inode(trans, root, inode);
2691 out:
2692         return ret;
2693 }
2694
2695 /* helper to check if there is any shared block in the path */
2696 static int check_path_shared(struct btrfs_root *root,
2697                              struct btrfs_path *path)
2698 {
2699         struct extent_buffer *eb;
2700         int level;
2701         u64 refs = 1;
2702         int uninitialized_var(ret);
2703
2704         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2705                 if (!path->nodes[level])
2706                         break;
2707                 eb = path->nodes[level];
2708                 if (!btrfs_block_can_be_shared(root, eb))
2709                         continue;
2710                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2711                                                &refs, NULL);
2712                 if (refs > 1)
2713                         return 1;
2714         }
2715         return ret; /* XXX callers? */
2716 }
2717
2718 /*
2719  * helper to start transaction for unlink and rmdir.
2720  *
2721  * unlink and rmdir are special in btrfs, they do not always free space.
2722  * so in enospc case, we should make sure they will free space before
2723  * allowing them to use the global metadata reservation.
2724  */
2725 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2726                                                        struct dentry *dentry)
2727 {
2728         struct btrfs_trans_handle *trans;
2729         struct btrfs_root *root = BTRFS_I(dir)->root;
2730         struct btrfs_path *path;
2731         struct btrfs_inode_ref *ref;
2732         struct btrfs_dir_item *di;
2733         struct inode *inode = dentry->d_inode;
2734         u64 index;
2735         int check_link = 1;
2736         int err = -ENOSPC;
2737         int ret;
2738
2739         trans = btrfs_start_transaction(root, 10);
2740         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2741                 return trans;
2742
2743         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2744                 return ERR_PTR(-ENOSPC);
2745
2746         /* check if there is someone else holds reference */
2747         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2748                 return ERR_PTR(-ENOSPC);
2749
2750         if (atomic_read(&inode->i_count) > 2)
2751                 return ERR_PTR(-ENOSPC);
2752
2753         if (xchg(&root->fs_info->enospc_unlink, 1))
2754                 return ERR_PTR(-ENOSPC);
2755
2756         path = btrfs_alloc_path();
2757         if (!path) {
2758                 root->fs_info->enospc_unlink = 0;
2759                 return ERR_PTR(-ENOMEM);
2760         }
2761
2762         trans = btrfs_start_transaction(root, 0);
2763         if (IS_ERR(trans)) {
2764                 btrfs_free_path(path);
2765                 root->fs_info->enospc_unlink = 0;
2766                 return trans;
2767         }
2768
2769         path->skip_locking = 1;
2770         path->search_commit_root = 1;
2771
2772         ret = btrfs_lookup_inode(trans, root, path,
2773                                 &BTRFS_I(dir)->location, 0);
2774         if (ret < 0) {
2775                 err = ret;
2776                 goto out;
2777         }
2778         if (ret == 0) {
2779                 if (check_path_shared(root, path))
2780                         goto out;
2781         } else {
2782                 check_link = 0;
2783         }
2784         btrfs_release_path(root, path);
2785
2786         ret = btrfs_lookup_inode(trans, root, path,
2787                                 &BTRFS_I(inode)->location, 0);
2788         if (ret < 0) {
2789                 err = ret;
2790                 goto out;
2791         }
2792         if (ret == 0) {
2793                 if (check_path_shared(root, path))
2794                         goto out;
2795         } else {
2796                 check_link = 0;
2797         }
2798         btrfs_release_path(root, path);
2799
2800         if (ret == 0 && S_ISREG(inode->i_mode)) {
2801                 ret = btrfs_lookup_file_extent(trans, root, path,
2802                                                inode->i_ino, (u64)-1, 0);
2803                 if (ret < 0) {
2804                         err = ret;
2805                         goto out;
2806                 }
2807                 BUG_ON(ret == 0);
2808                 if (check_path_shared(root, path))
2809                         goto out;
2810                 btrfs_release_path(root, path);
2811         }
2812
2813         if (!check_link) {
2814                 err = 0;
2815                 goto out;
2816         }
2817
2818         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2819                                 dentry->d_name.name, dentry->d_name.len, 0);
2820         if (IS_ERR(di)) {
2821                 err = PTR_ERR(di);
2822                 goto out;
2823         }
2824         if (di) {
2825                 if (check_path_shared(root, path))
2826                         goto out;
2827         } else {
2828                 err = 0;
2829                 goto out;
2830         }
2831         btrfs_release_path(root, path);
2832
2833         ref = btrfs_lookup_inode_ref(trans, root, path,
2834                                 dentry->d_name.name, dentry->d_name.len,
2835                                 inode->i_ino, dir->i_ino, 0);
2836         if (IS_ERR(ref)) {
2837                 err = PTR_ERR(ref);
2838                 goto out;
2839         }
2840         BUG_ON(!ref);
2841         if (check_path_shared(root, path))
2842                 goto out;
2843         index = btrfs_inode_ref_index(path->nodes[0], ref);
2844         btrfs_release_path(root, path);
2845
2846         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2847                                 dentry->d_name.name, dentry->d_name.len, 0);
2848         if (IS_ERR(di)) {
2849                 err = PTR_ERR(di);
2850                 goto out;
2851         }
2852         BUG_ON(ret == -ENOENT);
2853         if (check_path_shared(root, path))
2854                 goto out;
2855
2856         err = 0;
2857 out:
2858         btrfs_free_path(path);
2859         if (err) {
2860                 btrfs_end_transaction(trans, root);
2861                 root->fs_info->enospc_unlink = 0;
2862                 return ERR_PTR(err);
2863         }
2864
2865         trans->block_rsv = &root->fs_info->global_block_rsv;
2866         return trans;
2867 }
2868
2869 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2870                                struct btrfs_root *root)
2871 {
2872         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2873                 BUG_ON(!root->fs_info->enospc_unlink);
2874                 root->fs_info->enospc_unlink = 0;
2875         }
2876         btrfs_end_transaction_throttle(trans, root);
2877 }
2878
2879 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2880 {
2881         struct btrfs_root *root = BTRFS_I(dir)->root;
2882         struct btrfs_trans_handle *trans;
2883         struct inode *inode = dentry->d_inode;
2884         int ret;
2885         unsigned long nr = 0;
2886
2887         trans = __unlink_start_trans(dir, dentry);
2888         if (IS_ERR(trans))
2889                 return PTR_ERR(trans);
2890
2891         btrfs_set_trans_block_group(trans, dir);
2892
2893         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2894
2895         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2896                                  dentry->d_name.name, dentry->d_name.len);
2897         BUG_ON(ret);
2898
2899         if (inode->i_nlink == 0) {
2900                 ret = btrfs_orphan_add(trans, inode);
2901                 BUG_ON(ret);
2902         }
2903
2904         nr = trans->blocks_used;
2905         __unlink_end_trans(trans, root);
2906         btrfs_btree_balance_dirty(root, nr);
2907         return ret;
2908 }
2909
2910 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2911                         struct btrfs_root *root,
2912                         struct inode *dir, u64 objectid,
2913                         const char *name, int name_len)
2914 {
2915         struct btrfs_path *path;
2916         struct extent_buffer *leaf;
2917         struct btrfs_dir_item *di;
2918         struct btrfs_key key;
2919         u64 index;
2920         int ret;
2921
2922         path = btrfs_alloc_path();
2923         if (!path)
2924                 return -ENOMEM;
2925
2926         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2927                                    name, name_len, -1);
2928         BUG_ON(!di || IS_ERR(di));
2929
2930         leaf = path->nodes[0];
2931         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2932         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2933         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2934         BUG_ON(ret);
2935         btrfs_release_path(root, path);
2936
2937         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2938                                  objectid, root->root_key.objectid,
2939                                  dir->i_ino, &index, name, name_len);
2940         if (ret < 0) {
2941                 BUG_ON(ret != -ENOENT);
2942                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
2943                                                  name, name_len);
2944                 BUG_ON(!di || IS_ERR(di));
2945
2946                 leaf = path->nodes[0];
2947                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2948                 btrfs_release_path(root, path);
2949                 index = key.offset;
2950         }
2951
2952         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2953                                          index, name, name_len, -1);
2954         BUG_ON(!di || IS_ERR(di));
2955
2956         leaf = path->nodes[0];
2957         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2958         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2959         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2960         BUG_ON(ret);
2961         btrfs_release_path(root, path);
2962
2963         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2964         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2965         ret = btrfs_update_inode(trans, root, dir);
2966         BUG_ON(ret);
2967
2968         btrfs_free_path(path);
2969         return 0;
2970 }
2971
2972 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
2973 {
2974         struct inode *inode = dentry->d_inode;
2975         int err = 0;
2976         struct btrfs_root *root = BTRFS_I(dir)->root;
2977         struct btrfs_trans_handle *trans;
2978         unsigned long nr = 0;
2979
2980         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
2981             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
2982                 return -ENOTEMPTY;
2983
2984         trans = __unlink_start_trans(dir, dentry);
2985         if (IS_ERR(trans))
2986                 return PTR_ERR(trans);
2987
2988         btrfs_set_trans_block_group(trans, dir);
2989
2990         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
2991                 err = btrfs_unlink_subvol(trans, root, dir,
2992                                           BTRFS_I(inode)->location.objectid,
2993                                           dentry->d_name.name,
2994                                           dentry->d_name.len);
2995                 goto out;
2996         }
2997
2998         err = btrfs_orphan_add(trans, inode);
2999         if (err)
3000                 goto out;
3001
3002         /* now the directory is empty */
3003         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3004                                  dentry->d_name.name, dentry->d_name.len);
3005         if (!err)
3006                 btrfs_i_size_write(inode, 0);
3007 out:
3008         nr = trans->blocks_used;
3009         __unlink_end_trans(trans, root);
3010         btrfs_btree_balance_dirty(root, nr);
3011
3012         return err;
3013 }
3014
3015 #if 0
3016 /*
3017  * when truncating bytes in a file, it is possible to avoid reading
3018  * the leaves that contain only checksum items.  This can be the
3019  * majority of the IO required to delete a large file, but it must
3020  * be done carefully.
3021  *
3022  * The keys in the level just above the leaves are checked to make sure
3023  * the lowest key in a given leaf is a csum key, and starts at an offset
3024  * after the new  size.
3025  *
3026  * Then the key for the next leaf is checked to make sure it also has
3027  * a checksum item for the same file.  If it does, we know our target leaf
3028  * contains only checksum items, and it can be safely freed without reading
3029  * it.
3030  *
3031  * This is just an optimization targeted at large files.  It may do
3032  * nothing.  It will return 0 unless things went badly.
3033  */
3034 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3035                                      struct btrfs_root *root,
3036                                      struct btrfs_path *path,
3037                                      struct inode *inode, u64 new_size)
3038 {
3039         struct btrfs_key key;
3040         int ret;
3041         int nritems;
3042         struct btrfs_key found_key;
3043         struct btrfs_key other_key;
3044         struct btrfs_leaf_ref *ref;
3045         u64 leaf_gen;
3046         u64 leaf_start;
3047
3048         path->lowest_level = 1;
3049         key.objectid = inode->i_ino;
3050         key.type = BTRFS_CSUM_ITEM_KEY;
3051         key.offset = new_size;
3052 again:
3053         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3054         if (ret < 0)
3055                 goto out;
3056
3057         if (path->nodes[1] == NULL) {
3058                 ret = 0;
3059                 goto out;
3060         }
3061         ret = 0;
3062         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3063         nritems = btrfs_header_nritems(path->nodes[1]);
3064
3065         if (!nritems)
3066                 goto out;
3067
3068         if (path->slots[1] >= nritems)
3069                 goto next_node;
3070
3071         /* did we find a key greater than anything we want to delete? */
3072         if (found_key.objectid > inode->i_ino ||
3073            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3074                 goto out;
3075
3076         /* we check the next key in the node to make sure the leave contains
3077          * only checksum items.  This comparison doesn't work if our
3078          * leaf is the last one in the node
3079          */
3080         if (path->slots[1] + 1 >= nritems) {
3081 next_node:
3082                 /* search forward from the last key in the node, this
3083                  * will bring us into the next node in the tree
3084                  */
3085                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3086
3087                 /* unlikely, but we inc below, so check to be safe */
3088                 if (found_key.offset == (u64)-1)
3089                         goto out;
3090
3091                 /* search_forward needs a path with locks held, do the
3092                  * search again for the original key.  It is possible
3093                  * this will race with a balance and return a path that
3094                  * we could modify, but this drop is just an optimization
3095                  * and is allowed to miss some leaves.
3096                  */
3097                 btrfs_release_path(root, path);
3098                 found_key.offset++;
3099
3100                 /* setup a max key for search_forward */
3101                 other_key.offset = (u64)-1;
3102                 other_key.type = key.type;
3103                 other_key.objectid = key.objectid;
3104
3105                 path->keep_locks = 1;
3106                 ret = btrfs_search_forward(root, &found_key, &other_key,
3107                                            path, 0, 0);
3108                 path->keep_locks = 0;
3109                 if (ret || found_key.objectid != key.objectid ||
3110                     found_key.type != key.type) {
3111                         ret = 0;
3112                         goto out;
3113                 }
3114
3115                 key.offset = found_key.offset;
3116                 btrfs_release_path(root, path);
3117                 cond_resched();
3118                 goto again;
3119         }
3120
3121         /* we know there's one more slot after us in the tree,
3122          * read that key so we can verify it is also a checksum item
3123          */
3124         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3125
3126         if (found_key.objectid < inode->i_ino)
3127                 goto next_key;
3128
3129         if (found_key.type != key.type || found_key.offset < new_size)
3130                 goto next_key;
3131
3132         /*
3133          * if the key for the next leaf isn't a csum key from this objectid,
3134          * we can't be sure there aren't good items inside this leaf.
3135          * Bail out
3136          */
3137         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3138                 goto out;
3139
3140         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3141         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3142         /*
3143          * it is safe to delete this leaf, it contains only
3144          * csum items from this inode at an offset >= new_size
3145          */
3146         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3147         BUG_ON(ret);
3148
3149         if (root->ref_cows && leaf_gen < trans->transid) {
3150                 ref = btrfs_alloc_leaf_ref(root, 0);
3151                 if (ref) {
3152                         ref->root_gen = root->root_key.offset;
3153                         ref->bytenr = leaf_start;
3154                         ref->owner = 0;
3155                         ref->generation = leaf_gen;
3156                         ref->nritems = 0;
3157
3158                         btrfs_sort_leaf_ref(ref);
3159
3160                         ret = btrfs_add_leaf_ref(root, ref, 0);
3161                         WARN_ON(ret);
3162                         btrfs_free_leaf_ref(root, ref);
3163                 } else {
3164                         WARN_ON(1);
3165                 }
3166         }
3167 next_key:
3168         btrfs_release_path(root, path);
3169
3170         if (other_key.objectid == inode->i_ino &&
3171             other_key.type == key.type && other_key.offset > key.offset) {
3172                 key.offset = other_key.offset;
3173                 cond_resched();
3174                 goto again;
3175         }
3176         ret = 0;
3177 out:
3178         /* fixup any changes we've made to the path */
3179         path->lowest_level = 0;
3180         path->keep_locks = 0;
3181         btrfs_release_path(root, path);
3182         return ret;
3183 }
3184
3185 #endif
3186
3187 /*
3188  * this can truncate away extent items, csum items and directory items.
3189  * It starts at a high offset and removes keys until it can't find
3190  * any higher than new_size
3191  *
3192  * csum items that cross the new i_size are truncated to the new size
3193  * as well.
3194  *
3195  * min_type is the minimum key type to truncate down to.  If set to 0, this
3196  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3197  */
3198 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3199                                struct btrfs_root *root,
3200                                struct inode *inode,
3201                                u64 new_size, u32 min_type)
3202 {
3203         struct btrfs_path *path;
3204         struct extent_buffer *leaf;
3205         struct btrfs_file_extent_item *fi;
3206         struct btrfs_key key;
3207         struct btrfs_key found_key;
3208         u64 extent_start = 0;
3209         u64 extent_num_bytes = 0;
3210         u64 extent_offset = 0;
3211         u64 item_end = 0;
3212         u64 mask = root->sectorsize - 1;
3213         u32 found_type = (u8)-1;
3214         int found_extent;
3215         int del_item;
3216         int pending_del_nr = 0;
3217         int pending_del_slot = 0;
3218         int extent_type = -1;
3219         int encoding;
3220         int ret;
3221         int err = 0;
3222
3223         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3224
3225         if (root->ref_cows || root == root->fs_info->tree_root)
3226                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3227
3228         path = btrfs_alloc_path();
3229         BUG_ON(!path);
3230         path->reada = -1;
3231
3232         key.objectid = inode->i_ino;
3233         key.offset = (u64)-1;
3234         key.type = (u8)-1;
3235
3236 search_again:
3237         path->leave_spinning = 1;
3238         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3239         if (ret < 0) {
3240                 err = ret;
3241                 goto out;
3242         }
3243
3244         if (ret > 0) {
3245                 /* there are no items in the tree for us to truncate, we're
3246                  * done
3247                  */
3248                 if (path->slots[0] == 0)
3249                         goto out;
3250                 path->slots[0]--;
3251         }
3252
3253         while (1) {
3254                 fi = NULL;
3255                 leaf = path->nodes[0];
3256                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3257                 found_type = btrfs_key_type(&found_key);
3258                 encoding = 0;
3259
3260                 if (found_key.objectid != inode->i_ino)
3261                         break;
3262
3263                 if (found_type < min_type)
3264                         break;
3265
3266                 item_end = found_key.offset;
3267                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3268                         fi = btrfs_item_ptr(leaf, path->slots[0],
3269                                             struct btrfs_file_extent_item);
3270                         extent_type = btrfs_file_extent_type(leaf, fi);
3271                         encoding = btrfs_file_extent_compression(leaf, fi);
3272                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3273                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3274
3275                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3276                                 item_end +=
3277                                     btrfs_file_extent_num_bytes(leaf, fi);
3278                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3279                                 item_end += btrfs_file_extent_inline_len(leaf,
3280                                                                          fi);
3281                         }
3282                         item_end--;
3283                 }
3284                 if (found_type > min_type) {
3285                         del_item = 1;
3286                 } else {
3287                         if (item_end < new_size)
3288                                 break;
3289                         if (found_key.offset >= new_size)
3290                                 del_item = 1;
3291                         else
3292                                 del_item = 0;
3293                 }
3294                 found_extent = 0;
3295                 /* FIXME, shrink the extent if the ref count is only 1 */
3296                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3297                         goto delete;
3298
3299                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3300                         u64 num_dec;
3301                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3302                         if (!del_item && !encoding) {
3303                                 u64 orig_num_bytes =
3304                                         btrfs_file_extent_num_bytes(leaf, fi);
3305                                 extent_num_bytes = new_size -
3306                                         found_key.offset + root->sectorsize - 1;
3307                                 extent_num_bytes = extent_num_bytes &
3308                                         ~((u64)root->sectorsize - 1);
3309                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3310                                                          extent_num_bytes);
3311                                 num_dec = (orig_num_bytes -
3312                                            extent_num_bytes);
3313                                 if (root->ref_cows && extent_start != 0)
3314                                         inode_sub_bytes(inode, num_dec);
3315                                 btrfs_mark_buffer_dirty(leaf);
3316                         } else {
3317                                 extent_num_bytes =
3318                                         btrfs_file_extent_disk_num_bytes(leaf,
3319                                                                          fi);
3320                                 extent_offset = found_key.offset -
3321                                         btrfs_file_extent_offset(leaf, fi);
3322
3323                                 /* FIXME blocksize != 4096 */
3324                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3325                                 if (extent_start != 0) {
3326                                         found_extent = 1;
3327                                         if (root->ref_cows)
3328                                                 inode_sub_bytes(inode, num_dec);
3329                                 }
3330                         }
3331                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3332                         /*
3333                          * we can't truncate inline items that have had
3334                          * special encodings
3335                          */
3336                         if (!del_item &&
3337                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3338                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3339                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3340                                 u32 size = new_size - found_key.offset;
3341
3342                                 if (root->ref_cows) {
3343                                         inode_sub_bytes(inode, item_end + 1 -
3344                                                         new_size);
3345                                 }
3346                                 size =
3347                                     btrfs_file_extent_calc_inline_size(size);
3348                                 ret = btrfs_truncate_item(trans, root, path,
3349                                                           size, 1);
3350                                 BUG_ON(ret);
3351                         } else if (root->ref_cows) {
3352                                 inode_sub_bytes(inode, item_end + 1 -
3353                                                 found_key.offset);
3354                         }
3355                 }
3356 delete:
3357                 if (del_item) {
3358                         if (!pending_del_nr) {
3359                                 /* no pending yet, add ourselves */
3360                                 pending_del_slot = path->slots[0];
3361                                 pending_del_nr = 1;
3362                         } else if (pending_del_nr &&
3363                                    path->slots[0] + 1 == pending_del_slot) {
3364                                 /* hop on the pending chunk */
3365                                 pending_del_nr++;
3366                                 pending_del_slot = path->slots[0];
3367                         } else {
3368                                 BUG();
3369                         }
3370                 } else {
3371                         break;
3372                 }
3373                 if (found_extent && (root->ref_cows ||
3374                                      root == root->fs_info->tree_root)) {
3375                         btrfs_set_path_blocking(path);
3376                         ret = btrfs_free_extent(trans, root, extent_start,
3377                                                 extent_num_bytes, 0,
3378                                                 btrfs_header_owner(leaf),
3379                                                 inode->i_ino, extent_offset);
3380                         BUG_ON(ret);
3381                 }
3382
3383                 if (found_type == BTRFS_INODE_ITEM_KEY)
3384                         break;
3385
3386                 if (path->slots[0] == 0 ||
3387                     path->slots[0] != pending_del_slot) {
3388                         if (root->ref_cows) {
3389                                 err = -EAGAIN;
3390                                 goto out;
3391                         }
3392                         if (pending_del_nr) {
3393                                 ret = btrfs_del_items(trans, root, path,
3394                                                 pending_del_slot,
3395                                                 pending_del_nr);
3396                                 BUG_ON(ret);
3397                                 pending_del_nr = 0;
3398                         }
3399                         btrfs_release_path(root, path);
3400                         goto search_again;
3401                 } else {
3402                         path->slots[0]--;
3403                 }
3404         }
3405 out:
3406         if (pending_del_nr) {
3407                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3408                                       pending_del_nr);
3409                 BUG_ON(ret);
3410         }
3411         btrfs_free_path(path);
3412         return err;
3413 }
3414
3415 /*
3416  * taken from block_truncate_page, but does cow as it zeros out
3417  * any bytes left in the last page in the file.
3418  */
3419 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3420 {
3421         struct inode *inode = mapping->host;
3422         struct btrfs_root *root = BTRFS_I(inode)->root;
3423         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3424         struct btrfs_ordered_extent *ordered;
3425         struct extent_state *cached_state = NULL;
3426         char *kaddr;
3427         u32 blocksize = root->sectorsize;
3428         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3429         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3430         struct page *page;
3431         int ret = 0;
3432         u64 page_start;
3433         u64 page_end;
3434
3435         if ((offset & (blocksize - 1)) == 0)
3436                 goto out;
3437         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3438         if (ret)
3439                 goto out;
3440
3441         ret = -ENOMEM;
3442 again:
3443         page = grab_cache_page(mapping, index);
3444         if (!page) {
3445                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3446                 goto out;
3447         }
3448
3449         page_start = page_offset(page);
3450         page_end = page_start + PAGE_CACHE_SIZE - 1;
3451
3452         if (!PageUptodate(page)) {
3453                 ret = btrfs_readpage(NULL, page);
3454                 lock_page(page);
3455                 if (page->mapping != mapping) {
3456                         unlock_page(page);
3457                         page_cache_release(page);
3458                         goto again;
3459                 }
3460                 if (!PageUptodate(page)) {
3461                         ret = -EIO;
3462                         goto out_unlock;
3463                 }
3464         }
3465         wait_on_page_writeback(page);
3466
3467         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3468                          GFP_NOFS);
3469         set_page_extent_mapped(page);
3470
3471         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3472         if (ordered) {
3473                 unlock_extent_cached(io_tree, page_start, page_end,
3474                                      &cached_state, GFP_NOFS);
3475                 unlock_page(page);
3476                 page_cache_release(page);
3477                 btrfs_start_ordered_extent(inode, ordered, 1);
3478                 btrfs_put_ordered_extent(ordered);
3479                 goto again;
3480         }
3481
3482         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3483                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3484                           0, 0, &cached_state, GFP_NOFS);
3485
3486         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3487                                         &cached_state);
3488         if (ret) {
3489                 unlock_extent_cached(io_tree, page_start, page_end,
3490                                      &cached_state, GFP_NOFS);
3491                 goto out_unlock;
3492         }
3493
3494         ret = 0;
3495         if (offset != PAGE_CACHE_SIZE) {
3496                 kaddr = kmap(page);
3497                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3498                 flush_dcache_page(page);
3499                 kunmap(page);
3500         }
3501         ClearPageChecked(page);
3502         set_page_dirty(page);
3503         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3504                              GFP_NOFS);
3505
3506 out_unlock:
3507         if (ret)
3508                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3509         unlock_page(page);
3510         page_cache_release(page);
3511 out:
3512         return ret;
3513 }
3514
3515 int btrfs_cont_expand(struct inode *inode, loff_t size)
3516 {
3517         struct btrfs_trans_handle *trans;
3518         struct btrfs_root *root = BTRFS_I(inode)->root;
3519         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3520         struct extent_map *em = NULL;
3521         struct extent_state *cached_state = NULL;
3522         u64 mask = root->sectorsize - 1;
3523         u64 hole_start = (inode->i_size + mask) & ~mask;
3524         u64 block_end = (size + mask) & ~mask;
3525         u64 last_byte;
3526         u64 cur_offset;
3527         u64 hole_size;
3528         int err = 0;
3529
3530         if (size <= hole_start)
3531                 return 0;
3532
3533         while (1) {
3534                 struct btrfs_ordered_extent *ordered;
3535                 btrfs_wait_ordered_range(inode, hole_start,
3536                                          block_end - hole_start);
3537                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3538                                  &cached_state, GFP_NOFS);
3539                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3540                 if (!ordered)
3541                         break;
3542                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3543                                      &cached_state, GFP_NOFS);
3544                 btrfs_put_ordered_extent(ordered);
3545         }
3546
3547         cur_offset = hole_start;
3548         while (1) {
3549                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3550                                 block_end - cur_offset, 0);
3551                 BUG_ON(IS_ERR(em) || !em);
3552                 last_byte = min(extent_map_end(em), block_end);
3553                 last_byte = (last_byte + mask) & ~mask;
3554                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3555                         u64 hint_byte = 0;
3556                         hole_size = last_byte - cur_offset;
3557
3558                         trans = btrfs_start_transaction(root, 2);
3559                         if (IS_ERR(trans)) {
3560                                 err = PTR_ERR(trans);
3561                                 break;
3562                         }
3563                         btrfs_set_trans_block_group(trans, inode);
3564
3565                         err = btrfs_drop_extents(trans, inode, cur_offset,
3566                                                  cur_offset + hole_size,
3567                                                  &hint_byte, 1);
3568                         BUG_ON(err);
3569
3570                         err = btrfs_insert_file_extent(trans, root,
3571                                         inode->i_ino, cur_offset, 0,
3572                                         0, hole_size, 0, hole_size,
3573                                         0, 0, 0);
3574                         BUG_ON(err);
3575
3576                         btrfs_drop_extent_cache(inode, hole_start,
3577                                         last_byte - 1, 0);
3578
3579                         btrfs_end_transaction(trans, root);
3580                 }
3581                 free_extent_map(em);
3582                 em = NULL;
3583                 cur_offset = last_byte;
3584                 if (cur_offset >= block_end)
3585                         break;
3586         }
3587
3588         free_extent_map(em);
3589         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3590                              GFP_NOFS);
3591         return err;
3592 }
3593
3594 static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
3595 {
3596         struct btrfs_root *root = BTRFS_I(inode)->root;
3597         struct btrfs_trans_handle *trans;
3598         unsigned long nr;
3599         int ret;
3600
3601         if (attr->ia_size == inode->i_size)
3602                 return 0;
3603
3604         if (attr->ia_size > inode->i_size) {
3605                 unsigned long limit;
3606                 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
3607                 if (attr->ia_size > inode->i_sb->s_maxbytes)
3608                         return -EFBIG;
3609                 if (limit != RLIM_INFINITY && attr->ia_size > limit) {
3610                         send_sig(SIGXFSZ, current, 0);
3611                         return -EFBIG;
3612                 }
3613         }
3614
3615         trans = btrfs_start_transaction(root, 5);
3616         if (IS_ERR(trans))
3617                 return PTR_ERR(trans);
3618
3619         btrfs_set_trans_block_group(trans, inode);
3620
3621         ret = btrfs_orphan_add(trans, inode);
3622         BUG_ON(ret);
3623
3624         nr = trans->blocks_used;
3625         btrfs_end_transaction(trans, root);
3626         btrfs_btree_balance_dirty(root, nr);
3627
3628         if (attr->ia_size > inode->i_size) {
3629                 ret = btrfs_cont_expand(inode, attr->ia_size);
3630                 if (ret) {
3631                         btrfs_truncate(inode);
3632                         return ret;
3633                 }
3634
3635                 i_size_write(inode, attr->ia_size);
3636                 btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
3637
3638                 trans = btrfs_start_transaction(root, 0);
3639                 BUG_ON(IS_ERR(trans));
3640                 btrfs_set_trans_block_group(trans, inode);
3641                 trans->block_rsv = root->orphan_block_rsv;
3642                 BUG_ON(!trans->block_rsv);
3643
3644                 ret = btrfs_update_inode(trans, root, inode);
3645                 BUG_ON(ret);
3646                 if (inode->i_nlink > 0) {
3647                         ret = btrfs_orphan_del(trans, inode);
3648                         BUG_ON(ret);
3649                 }
3650                 nr = trans->blocks_used;
3651                 btrfs_end_transaction(trans, root);
3652                 btrfs_btree_balance_dirty(root, nr);
3653                 return 0;
3654         }
3655
3656         /*
3657          * We're truncating a file that used to have good data down to
3658          * zero. Make sure it gets into the ordered flush list so that
3659          * any new writes get down to disk quickly.
3660          */
3661         if (attr->ia_size == 0)
3662                 BTRFS_I(inode)->ordered_data_close = 1;
3663
3664         /* we don't support swapfiles, so vmtruncate shouldn't fail */
3665         ret = vmtruncate(inode, attr->ia_size);
3666         BUG_ON(ret);
3667
3668         return 0;
3669 }
3670
3671 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3672 {
3673         struct inode *inode = dentry->d_inode;
3674         struct btrfs_root *root = BTRFS_I(inode)->root;
3675         int err;
3676
3677         if (btrfs_root_readonly(root))
3678                 return -EROFS;
3679
3680         err = inode_change_ok(inode, attr);
3681         if (err)
3682                 return err;
3683
3684         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3685                 err = btrfs_setattr_size(inode, attr);
3686                 if (err)
3687                         return err;
3688         }
3689
3690         if (attr->ia_valid) {
3691                 setattr_copy(inode, attr);
3692                 mark_inode_dirty(inode);
3693
3694                 if (attr->ia_valid & ATTR_MODE)
3695                         err = btrfs_acl_chmod(inode);
3696         }
3697
3698         return err;
3699 }
3700
3701 void btrfs_evict_inode(struct inode *inode)
3702 {
3703         struct btrfs_trans_handle *trans;
3704         struct btrfs_root *root = BTRFS_I(inode)->root;
3705         unsigned long nr;
3706         int ret;
3707
3708         truncate_inode_pages(&inode->i_data, 0);
3709         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3710                                root == root->fs_info->tree_root))
3711                 goto no_delete;
3712
3713         if (is_bad_inode(inode)) {
3714                 btrfs_orphan_del(NULL, inode);
3715                 goto no_delete;
3716         }
3717         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3718         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3719
3720         if (root->fs_info->log_root_recovering) {
3721                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3722                 goto no_delete;
3723         }
3724
3725         if (inode->i_nlink > 0) {
3726                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3727                 goto no_delete;
3728         }
3729
3730         btrfs_i_size_write(inode, 0);
3731
3732         while (1) {
3733                 trans = btrfs_start_transaction(root, 0);
3734                 BUG_ON(IS_ERR(trans));
3735                 btrfs_set_trans_block_group(trans, inode);
3736                 trans->block_rsv = root->orphan_block_rsv;
3737
3738                 ret = btrfs_block_rsv_check(trans, root,
3739                                             root->orphan_block_rsv, 0, 5);
3740                 if (ret) {
3741                         BUG_ON(ret != -EAGAIN);
3742                         ret = btrfs_commit_transaction(trans, root);
3743                         BUG_ON(ret);
3744                         continue;
3745                 }
3746
3747                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3748                 if (ret != -EAGAIN)
3749                         break;
3750
3751                 nr = trans->blocks_used;
3752                 btrfs_end_transaction(trans, root);
3753                 trans = NULL;
3754                 btrfs_btree_balance_dirty(root, nr);
3755
3756         }
3757
3758         if (ret == 0) {
3759                 ret = btrfs_orphan_del(trans, inode);
3760                 BUG_ON(ret);
3761         }
3762
3763         nr = trans->blocks_used;
3764         btrfs_end_transaction(trans, root);
3765         btrfs_btree_balance_dirty(root, nr);
3766 no_delete:
3767         end_writeback(inode);
3768         return;
3769 }
3770
3771 /*
3772  * this returns the key found in the dir entry in the location pointer.
3773  * If no dir entries were found, location->objectid is 0.
3774  */
3775 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3776                                struct btrfs_key *location)
3777 {
3778         const char *name = dentry->d_name.name;
3779         int namelen = dentry->d_name.len;
3780         struct btrfs_dir_item *di;
3781         struct btrfs_path *path;
3782         struct btrfs_root *root = BTRFS_I(dir)->root;
3783         int ret = 0;
3784
3785         path = btrfs_alloc_path();
3786         BUG_ON(!path);
3787
3788         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3789                                     namelen, 0);
3790         if (IS_ERR(di))
3791                 ret = PTR_ERR(di);
3792
3793         if (!di || IS_ERR(di))
3794                 goto out_err;
3795
3796         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3797 out:
3798         btrfs_free_path(path);
3799         return ret;
3800 out_err:
3801         location->objectid = 0;
3802         goto out;
3803 }
3804
3805 /*
3806  * when we hit a tree root in a directory, the btrfs part of the inode
3807  * needs to be changed to reflect the root directory of the tree root.  This
3808  * is kind of like crossing a mount point.
3809  */
3810 static int fixup_tree_root_location(struct btrfs_root *root,
3811                                     struct inode *dir,
3812                                     struct dentry *dentry,
3813                                     struct btrfs_key *location,
3814                                     struct btrfs_root **sub_root)
3815 {
3816         struct btrfs_path *path;
3817         struct btrfs_root *new_root;
3818         struct btrfs_root_ref *ref;
3819         struct extent_buffer *leaf;
3820         int ret;
3821         int err = 0;
3822
3823         path = btrfs_alloc_path();
3824         if (!path) {
3825                 err = -ENOMEM;
3826                 goto out;
3827         }
3828
3829         err = -ENOENT;
3830         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3831                                   BTRFS_I(dir)->root->root_key.objectid,
3832                                   location->objectid);
3833         if (ret) {
3834                 if (ret < 0)
3835                         err = ret;
3836                 goto out;
3837         }
3838
3839         leaf = path->nodes[0];
3840         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3841         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3842             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3843                 goto out;
3844
3845         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3846                                    (unsigned long)(ref + 1),
3847                                    dentry->d_name.len);
3848         if (ret)
3849                 goto out;
3850
3851         btrfs_release_path(root->fs_info->tree_root, path);
3852
3853         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3854         if (IS_ERR(new_root)) {
3855                 err = PTR_ERR(new_root);
3856                 goto out;
3857         }
3858
3859         if (btrfs_root_refs(&new_root->root_item) == 0) {
3860                 err = -ENOENT;
3861                 goto out;
3862         }
3863
3864         *sub_root = new_root;
3865         location->objectid = btrfs_root_dirid(&new_root->root_item);
3866         location->type = BTRFS_INODE_ITEM_KEY;
3867         location->offset = 0;
3868         err = 0;
3869 out:
3870         btrfs_free_path(path);
3871         return err;
3872 }
3873
3874 static void inode_tree_add(struct inode *inode)
3875 {
3876         struct btrfs_root *root = BTRFS_I(inode)->root;
3877         struct btrfs_inode *entry;
3878         struct rb_node **p;
3879         struct rb_node *parent;
3880 again:
3881         p = &root->inode_tree.rb_node;
3882         parent = NULL;
3883
3884         if (hlist_unhashed(&inode->i_hash))
3885                 return;
3886
3887         spin_lock(&root->inode_lock);
3888         while (*p) {
3889                 parent = *p;
3890                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3891
3892                 if (inode->i_ino < entry->vfs_inode.i_ino)
3893                         p = &parent->rb_left;
3894                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3895                         p = &parent->rb_right;
3896                 else {
3897                         WARN_ON(!(entry->vfs_inode.i_state &
3898                                   (I_WILL_FREE | I_FREEING)));
3899                         rb_erase(parent, &root->inode_tree);
3900                         RB_CLEAR_NODE(parent);
3901                         spin_unlock(&root->inode_lock);
3902                         goto again;
3903                 }
3904         }
3905         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3906         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3907         spin_unlock(&root->inode_lock);
3908 }
3909
3910 static void inode_tree_del(struct inode *inode)
3911 {
3912         struct btrfs_root *root = BTRFS_I(inode)->root;
3913         int empty = 0;
3914
3915         spin_lock(&root->inode_lock);
3916         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3917                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3918                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3919                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3920         }
3921         spin_unlock(&root->inode_lock);
3922
3923         /*
3924          * Free space cache has inodes in the tree root, but the tree root has a
3925          * root_refs of 0, so this could end up dropping the tree root as a
3926          * snapshot, so we need the extra !root->fs_info->tree_root check to
3927          * make sure we don't drop it.
3928          */
3929         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3930             root != root->fs_info->tree_root) {
3931                 synchronize_srcu(&root->fs_info->subvol_srcu);
3932                 spin_lock(&root->inode_lock);
3933                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3934                 spin_unlock(&root->inode_lock);
3935                 if (empty)
3936                         btrfs_add_dead_root(root);
3937         }
3938 }
3939
3940 int btrfs_invalidate_inodes(struct btrfs_root *root)
3941 {
3942         struct rb_node *node;
3943         struct rb_node *prev;
3944         struct btrfs_inode *entry;
3945         struct inode *inode;
3946         u64 objectid = 0;
3947
3948         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3949
3950         spin_lock(&root->inode_lock);
3951 again:
3952         node = root->inode_tree.rb_node;
3953         prev = NULL;
3954         while (node) {
3955                 prev = node;
3956                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3957
3958                 if (objectid < entry->vfs_inode.i_ino)
3959                         node = node->rb_left;
3960                 else if (objectid > entry->vfs_inode.i_ino)
3961                         node = node->rb_right;
3962                 else
3963                         break;
3964         }
3965         if (!node) {
3966                 while (prev) {
3967                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3968                         if (objectid <= entry->vfs_inode.i_ino) {
3969                                 node = prev;
3970                                 break;
3971                         }
3972                         prev = rb_next(prev);
3973                 }
3974         }
3975         while (node) {
3976                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3977                 objectid = entry->vfs_inode.i_ino + 1;
3978                 inode = igrab(&entry->vfs_inode);
3979                 if (inode) {
3980                         spin_unlock(&root->inode_lock);
3981                         if (atomic_read(&inode->i_count) > 1)
3982                                 d_prune_aliases(inode);
3983                         /*
3984                          * btrfs_drop_inode will have it removed from
3985                          * the inode cache when its usage count
3986                          * hits zero.
3987                          */
3988                         iput(inode);
3989                         cond_resched();
3990                         spin_lock(&root->inode_lock);
3991                         goto again;
3992                 }
3993
3994                 if (cond_resched_lock(&root->inode_lock))
3995                         goto again;
3996
3997                 node = rb_next(node);
3998         }
3999         spin_unlock(&root->inode_lock);
4000         return 0;
4001 }
4002
4003 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4004 {
4005         struct btrfs_iget_args *args = p;
4006         inode->i_ino = args->ino;
4007         BTRFS_I(inode)->root = args->root;
4008         btrfs_set_inode_space_info(args->root, inode);
4009         return 0;
4010 }
4011
4012 static int btrfs_find_actor(struct inode *inode, void *opaque)
4013 {
4014         struct btrfs_iget_args *args = opaque;
4015         return args->ino == inode->i_ino &&
4016                 args->root == BTRFS_I(inode)->root;
4017 }
4018
4019 static struct inode *btrfs_iget_locked(struct super_block *s,
4020                                        u64 objectid,
4021                                        struct btrfs_root *root)
4022 {
4023         struct inode *inode;
4024         struct btrfs_iget_args args;
4025         args.ino = objectid;
4026         args.root = root;
4027
4028         inode = iget5_locked(s, objectid, btrfs_find_actor,
4029                              btrfs_init_locked_inode,
4030                              (void *)&args);
4031         return inode;
4032 }
4033
4034 /* Get an inode object given its location and corresponding root.
4035  * Returns in *is_new if the inode was read from disk
4036  */
4037 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4038                          struct btrfs_root *root, int *new)
4039 {
4040         struct inode *inode;
4041
4042         inode = btrfs_iget_locked(s, location->objectid, root);
4043         if (!inode)
4044                 return ERR_PTR(-ENOMEM);
4045
4046         if (inode->i_state & I_NEW) {
4047                 BTRFS_I(inode)->root = root;
4048                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4049                 btrfs_read_locked_inode(inode);
4050
4051                 inode_tree_add(inode);
4052                 unlock_new_inode(inode);
4053                 if (new)
4054                         *new = 1;
4055         }
4056
4057         return inode;
4058 }
4059
4060 static struct inode *new_simple_dir(struct super_block *s,
4061                                     struct btrfs_key *key,
4062                                     struct btrfs_root *root)
4063 {
4064         struct inode *inode = new_inode(s);
4065
4066         if (!inode)
4067                 return ERR_PTR(-ENOMEM);
4068
4069         BTRFS_I(inode)->root = root;
4070         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4071         BTRFS_I(inode)->dummy_inode = 1;
4072
4073         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4074         inode->i_op = &simple_dir_inode_operations;
4075         inode->i_fop = &simple_dir_operations;
4076         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4077         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4078
4079         return inode;
4080 }
4081
4082 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4083 {
4084         struct inode *inode;
4085         struct btrfs_root *root = BTRFS_I(dir)->root;
4086         struct btrfs_root *sub_root = root;
4087         struct btrfs_key location;
4088         int index;
4089         int ret;
4090
4091         dentry->d_op = &btrfs_dentry_operations;
4092
4093         if (dentry->d_name.len > BTRFS_NAME_LEN)
4094                 return ERR_PTR(-ENAMETOOLONG);
4095
4096         ret = btrfs_inode_by_name(dir, dentry, &location);
4097
4098         if (ret < 0)
4099                 return ERR_PTR(ret);
4100
4101         if (location.objectid == 0)
4102                 return NULL;
4103
4104         if (location.type == BTRFS_INODE_ITEM_KEY) {
4105                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4106                 return inode;
4107         }
4108
4109         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4110
4111         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4112         ret = fixup_tree_root_location(root, dir, dentry,
4113                                        &location, &sub_root);
4114         if (ret < 0) {
4115                 if (ret != -ENOENT)
4116                         inode = ERR_PTR(ret);
4117                 else
4118                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4119         } else {
4120                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4121         }
4122         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4123
4124         if (root != sub_root) {
4125                 down_read(&root->fs_info->cleanup_work_sem);
4126                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4127                         btrfs_orphan_cleanup(sub_root);
4128                 up_read(&root->fs_info->cleanup_work_sem);
4129         }
4130
4131         return inode;
4132 }
4133
4134 static int btrfs_dentry_delete(struct dentry *dentry)
4135 {
4136         struct btrfs_root *root;
4137
4138         if (!dentry->d_inode && !IS_ROOT(dentry))
4139                 dentry = dentry->d_parent;
4140
4141         if (dentry->d_inode) {
4142                 root = BTRFS_I(dentry->d_inode)->root;
4143                 if (btrfs_root_refs(&root->root_item) == 0)
4144                         return 1;
4145         }
4146         return 0;
4147 }
4148
4149 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4150                                    struct nameidata *nd)
4151 {
4152         struct inode *inode;
4153
4154         inode = btrfs_lookup_dentry(dir, dentry);
4155         if (IS_ERR(inode))
4156                 return ERR_CAST(inode);
4157
4158         return d_splice_alias(inode, dentry);
4159 }
4160
4161 static unsigned char btrfs_filetype_table[] = {
4162         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4163 };
4164
4165 static int btrfs_real_readdir(struct file *filp, void *dirent,
4166                               filldir_t filldir)
4167 {
4168         struct inode *inode = filp->f_dentry->d_inode;
4169         struct btrfs_root *root = BTRFS_I(inode)->root;
4170         struct btrfs_item *item;
4171         struct btrfs_dir_item *di;
4172         struct btrfs_key key;
4173         struct btrfs_key found_key;
4174         struct btrfs_path *path;
4175         int ret;
4176         u32 nritems;
4177         struct extent_buffer *leaf;
4178         int slot;
4179         int advance;
4180         unsigned char d_type;
4181         int over = 0;
4182         u32 di_cur;
4183         u32 di_total;
4184         u32 di_len;
4185         int key_type = BTRFS_DIR_INDEX_KEY;
4186         char tmp_name[32];
4187         char *name_ptr;
4188         int name_len;
4189
4190         /* FIXME, use a real flag for deciding about the key type */
4191         if (root->fs_info->tree_root == root)
4192                 key_type = BTRFS_DIR_ITEM_KEY;
4193
4194         /* special case for "." */
4195         if (filp->f_pos == 0) {
4196                 over = filldir(dirent, ".", 1,
4197                                1, inode->i_ino,
4198                                DT_DIR);
4199                 if (over)
4200                         return 0;
4201                 filp->f_pos = 1;
4202         }
4203         /* special case for .., just use the back ref */
4204         if (filp->f_pos == 1) {
4205                 u64 pino = parent_ino(filp->f_path.dentry);
4206                 over = filldir(dirent, "..", 2,
4207                                2, pino, DT_DIR);
4208                 if (over)
4209                         return 0;
4210                 filp->f_pos = 2;
4211         }
4212         path = btrfs_alloc_path();
4213         path->reada = 2;
4214
4215         btrfs_set_key_type(&key, key_type);
4216         key.offset = filp->f_pos;
4217         key.objectid = inode->i_ino;
4218
4219         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4220         if (ret < 0)
4221                 goto err;
4222         advance = 0;
4223
4224         while (1) {
4225                 leaf = path->nodes[0];
4226                 nritems = btrfs_header_nritems(leaf);
4227                 slot = path->slots[0];
4228                 if (advance || slot >= nritems) {
4229                         if (slot >= nritems - 1) {
4230                                 ret = btrfs_next_leaf(root, path);
4231                                 if (ret)
4232                                         break;
4233                                 leaf = path->nodes[0];
4234                                 nritems = btrfs_header_nritems(leaf);
4235                                 slot = path->slots[0];
4236                         } else {
4237                                 slot++;
4238                                 path->slots[0]++;
4239                         }
4240                 }
4241
4242                 advance = 1;
4243                 item = btrfs_item_nr(leaf, slot);
4244                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4245
4246                 if (found_key.objectid != key.objectid)
4247                         break;
4248                 if (btrfs_key_type(&found_key) != key_type)
4249                         break;
4250                 if (found_key.offset < filp->f_pos)
4251                         continue;
4252
4253                 filp->f_pos = found_key.offset;
4254
4255                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4256                 di_cur = 0;
4257                 di_total = btrfs_item_size(leaf, item);
4258
4259                 while (di_cur < di_total) {
4260                         struct btrfs_key location;
4261
4262                         name_len = btrfs_dir_name_len(leaf, di);
4263                         if (name_len <= sizeof(tmp_name)) {
4264                                 name_ptr = tmp_name;
4265                         } else {
4266                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4267                                 if (!name_ptr) {
4268                                         ret = -ENOMEM;
4269                                         goto err;
4270                                 }
4271                         }
4272                         read_extent_buffer(leaf, name_ptr,
4273                                            (unsigned long)(di + 1), name_len);
4274
4275                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4276                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4277
4278                         /* is this a reference to our own snapshot? If so
4279                          * skip it
4280                          */
4281                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4282                             location.objectid == root->root_key.objectid) {
4283                                 over = 0;
4284                                 goto skip;
4285                         }
4286                         over = filldir(dirent, name_ptr, name_len,
4287                                        found_key.offset, location.objectid,
4288                                        d_type);
4289
4290 skip:
4291                         if (name_ptr != tmp_name)
4292                                 kfree(name_ptr);
4293
4294                         if (over)
4295                                 goto nopos;
4296                         di_len = btrfs_dir_name_len(leaf, di) +
4297                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4298                         di_cur += di_len;
4299                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4300                 }
4301         }
4302
4303         /* Reached end of directory/root. Bump pos past the last item. */
4304         if (key_type == BTRFS_DIR_INDEX_KEY)
4305                 /*
4306                  * 32-bit glibc will use getdents64, but then strtol -
4307                  * so the last number we can serve is this.
4308                  */
4309                 filp->f_pos = 0x7fffffff;
4310         else
4311                 filp->f_pos++;
4312 nopos:
4313         ret = 0;
4314 err:
4315         btrfs_free_path(path);
4316         return ret;
4317 }
4318
4319 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4320 {
4321         struct btrfs_root *root = BTRFS_I(inode)->root;
4322         struct btrfs_trans_handle *trans;
4323         int ret = 0;
4324         bool nolock = false;
4325
4326         if (BTRFS_I(inode)->dummy_inode)
4327                 return 0;
4328
4329         smp_mb();
4330         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4331
4332         if (wbc->sync_mode == WB_SYNC_ALL) {
4333                 if (nolock)
4334                         trans = btrfs_join_transaction_nolock(root, 1);
4335                 else
4336                         trans = btrfs_join_transaction(root, 1);
4337                 btrfs_set_trans_block_group(trans, inode);
4338                 if (nolock)
4339                         ret = btrfs_end_transaction_nolock(trans, root);
4340                 else
4341                         ret = btrfs_commit_transaction(trans, root);
4342         }
4343         return ret;
4344 }
4345
4346 /*
4347  * This is somewhat expensive, updating the tree every time the
4348  * inode changes.  But, it is most likely to find the inode in cache.
4349  * FIXME, needs more benchmarking...there are no reasons other than performance
4350  * to keep or drop this code.
4351  */
4352 void btrfs_dirty_inode(struct inode *inode)
4353 {
4354         struct btrfs_root *root = BTRFS_I(inode)->root;
4355         struct btrfs_trans_handle *trans;
4356         int ret;
4357
4358         if (BTRFS_I(inode)->dummy_inode)
4359                 return;
4360
4361         trans = btrfs_join_transaction(root, 1);
4362         btrfs_set_trans_block_group(trans, inode);
4363
4364         ret = btrfs_update_inode(trans, root, inode);
4365         if (ret && ret == -ENOSPC) {
4366                 /* whoops, lets try again with the full transaction */
4367                 btrfs_end_transaction(trans, root);
4368                 trans = btrfs_start_transaction(root, 1);
4369                 if (IS_ERR(trans)) {
4370                         if (printk_ratelimit()) {
4371                                 printk(KERN_ERR "btrfs: fail to "
4372                                        "dirty  inode %lu error %ld\n",
4373                                        inode->i_ino, PTR_ERR(trans));
4374                         }
4375                         return;
4376                 }
4377                 btrfs_set_trans_block_group(trans, inode);
4378
4379                 ret = btrfs_update_inode(trans, root, inode);
4380                 if (ret) {
4381                         if (printk_ratelimit()) {
4382                                 printk(KERN_ERR "btrfs: fail to "
4383                                        "dirty  inode %lu error %d\n",
4384                                        inode->i_ino, ret);
4385                         }
4386                 }
4387         }
4388         btrfs_end_transaction(trans, root);
4389 }
4390
4391 /*
4392  * find the highest existing sequence number in a directory
4393  * and then set the in-memory index_cnt variable to reflect
4394  * free sequence numbers
4395  */
4396 static int btrfs_set_inode_index_count(struct inode *inode)
4397 {
4398         struct btrfs_root *root = BTRFS_I(inode)->root;
4399         struct btrfs_key key, found_key;
4400         struct btrfs_path *path;
4401         struct extent_buffer *leaf;
4402         int ret;
4403
4404         key.objectid = inode->i_ino;
4405         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4406         key.offset = (u64)-1;
4407
4408         path = btrfs_alloc_path();
4409         if (!path)
4410                 return -ENOMEM;
4411
4412         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4413         if (ret < 0)
4414                 goto out;
4415         /* FIXME: we should be able to handle this */
4416         if (ret == 0)
4417                 goto out;
4418         ret = 0;
4419
4420         /*
4421          * MAGIC NUMBER EXPLANATION:
4422          * since we search a directory based on f_pos we have to start at 2
4423          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4424          * else has to start at 2
4425          */
4426         if (path->slots[0] == 0) {
4427                 BTRFS_I(inode)->index_cnt = 2;
4428                 goto out;
4429         }
4430
4431         path->slots[0]--;
4432
4433         leaf = path->nodes[0];
4434         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4435
4436         if (found_key.objectid != inode->i_ino ||
4437             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4438                 BTRFS_I(inode)->index_cnt = 2;
4439                 goto out;
4440         }
4441
4442         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4443 out:
4444         btrfs_free_path(path);
4445         return ret;
4446 }
4447
4448 /*
4449  * helper to find a free sequence number in a given directory.  This current
4450  * code is very simple, later versions will do smarter things in the btree
4451  */
4452 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4453 {
4454         int ret = 0;
4455
4456         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4457                 ret = btrfs_set_inode_index_count(dir);
4458                 if (ret)
4459                         return ret;
4460         }
4461
4462         *index = BTRFS_I(dir)->index_cnt;
4463         BTRFS_I(dir)->index_cnt++;
4464
4465         return ret;
4466 }
4467
4468 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4469                                      struct btrfs_root *root,
4470                                      struct inode *dir,
4471                                      const char *name, int name_len,
4472                                      u64 ref_objectid, u64 objectid,
4473                                      u64 alloc_hint, int mode, u64 *index)
4474 {
4475         struct inode *inode;
4476         struct btrfs_inode_item *inode_item;
4477         struct btrfs_key *location;
4478         struct btrfs_path *path;
4479         struct btrfs_inode_ref *ref;
4480         struct btrfs_key key[2];
4481         u32 sizes[2];
4482         unsigned long ptr;
4483         int ret;
4484         int owner;
4485
4486         path = btrfs_alloc_path();
4487         BUG_ON(!path);
4488
4489         inode = new_inode(root->fs_info->sb);
4490         if (!inode)
4491                 return ERR_PTR(-ENOMEM);
4492
4493         if (dir) {
4494                 ret = btrfs_set_inode_index(dir, index);
4495                 if (ret) {
4496                         iput(inode);
4497                         return ERR_PTR(ret);
4498                 }
4499         }
4500         /*
4501          * index_cnt is ignored for everything but a dir,
4502          * btrfs_get_inode_index_count has an explanation for the magic
4503          * number
4504          */
4505         BTRFS_I(inode)->index_cnt = 2;
4506         BTRFS_I(inode)->root = root;
4507         BTRFS_I(inode)->generation = trans->transid;
4508         inode->i_generation = BTRFS_I(inode)->generation;
4509         btrfs_set_inode_space_info(root, inode);
4510
4511         if (mode & S_IFDIR)
4512                 owner = 0;
4513         else
4514                 owner = 1;
4515         BTRFS_I(inode)->block_group =
4516                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4517
4518         key[0].objectid = objectid;
4519         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4520         key[0].offset = 0;
4521
4522         key[1].objectid = objectid;
4523         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4524         key[1].offset = ref_objectid;
4525
4526         sizes[0] = sizeof(struct btrfs_inode_item);
4527         sizes[1] = name_len + sizeof(*ref);
4528
4529         path->leave_spinning = 1;
4530         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4531         if (ret != 0)
4532                 goto fail;
4533
4534         inode_init_owner(inode, dir, mode);
4535         inode->i_ino = objectid;
4536         inode_set_bytes(inode, 0);
4537         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4538         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4539                                   struct btrfs_inode_item);
4540         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4541
4542         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4543                              struct btrfs_inode_ref);
4544         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4545         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4546         ptr = (unsigned long)(ref + 1);
4547         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4548
4549         btrfs_mark_buffer_dirty(path->nodes[0]);
4550         btrfs_free_path(path);
4551
4552         location = &BTRFS_I(inode)->location;
4553         location->objectid = objectid;
4554         location->offset = 0;
4555         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4556
4557         btrfs_inherit_iflags(inode, dir);
4558
4559         if ((mode & S_IFREG)) {
4560                 if (btrfs_test_opt(root, NODATASUM))
4561                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4562                 if (btrfs_test_opt(root, NODATACOW))
4563                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4564         }
4565
4566         insert_inode_hash(inode);
4567         inode_tree_add(inode);
4568         return inode;
4569 fail:
4570         if (dir)
4571                 BTRFS_I(dir)->index_cnt--;
4572         btrfs_free_path(path);
4573         iput(inode);
4574         return ERR_PTR(ret);
4575 }
4576
4577 static inline u8 btrfs_inode_type(struct inode *inode)
4578 {
4579         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4580 }
4581
4582 /*
4583  * utility function to add 'inode' into 'parent_inode' with
4584  * a give name and a given sequence number.
4585  * if 'add_backref' is true, also insert a backref from the
4586  * inode to the parent directory.
4587  */
4588 int btrfs_add_link(struct btrfs_trans_handle *trans,
4589                    struct inode *parent_inode, struct inode *inode,
4590                    const char *name, int name_len, int add_backref, u64 index)
4591 {
4592         int ret = 0;
4593         struct btrfs_key key;
4594         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4595
4596         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4597                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4598         } else {
4599                 key.objectid = inode->i_ino;
4600                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4601                 key.offset = 0;
4602         }
4603
4604         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4605                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4606                                          key.objectid, root->root_key.objectid,
4607                                          parent_inode->i_ino,
4608                                          index, name, name_len);
4609         } else if (add_backref) {
4610                 ret = btrfs_insert_inode_ref(trans, root,
4611                                              name, name_len, inode->i_ino,
4612                                              parent_inode->i_ino, index);
4613         }
4614
4615         if (ret == 0) {
4616                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4617                                             parent_inode->i_ino, &key,
4618                                             btrfs_inode_type(inode), index);
4619                 BUG_ON(ret);
4620
4621                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4622                                    name_len * 2);
4623                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4624                 ret = btrfs_update_inode(trans, root, parent_inode);
4625         }
4626         return ret;
4627 }
4628
4629 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4630                             struct inode *dir, struct dentry *dentry,
4631                             struct inode *inode, int backref, u64 index)
4632 {
4633         int err = btrfs_add_link(trans, dir, inode,
4634                                  dentry->d_name.name, dentry->d_name.len,
4635                                  backref, index);
4636         if (!err) {
4637                 d_instantiate(dentry, inode);
4638                 return 0;
4639         }
4640         if (err > 0)
4641                 err = -EEXIST;
4642         return err;
4643 }
4644
4645 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4646                         int mode, dev_t rdev)
4647 {
4648         struct btrfs_trans_handle *trans;
4649         struct btrfs_root *root = BTRFS_I(dir)->root;
4650         struct inode *inode = NULL;
4651         int err;
4652         int drop_inode = 0;
4653         u64 objectid;
4654         unsigned long nr = 0;
4655         u64 index = 0;
4656
4657         if (!new_valid_dev(rdev))
4658                 return -EINVAL;
4659
4660         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4661         if (err)
4662                 return err;
4663
4664         /*
4665          * 2 for inode item and ref
4666          * 2 for dir items
4667          * 1 for xattr if selinux is on
4668          */
4669         trans = btrfs_start_transaction(root, 5);
4670         if (IS_ERR(trans))
4671                 return PTR_ERR(trans);
4672
4673         btrfs_set_trans_block_group(trans, dir);
4674
4675         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4676                                 dentry->d_name.len, dir->i_ino, objectid,
4677                                 BTRFS_I(dir)->block_group, mode, &index);
4678         err = PTR_ERR(inode);
4679         if (IS_ERR(inode))
4680                 goto out_unlock;
4681
4682         err = btrfs_init_inode_security(trans, inode, dir);
4683         if (err) {
4684                 drop_inode = 1;
4685                 goto out_unlock;
4686         }
4687
4688         btrfs_set_trans_block_group(trans, inode);
4689         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4690         if (err)
4691                 drop_inode = 1;
4692         else {
4693                 inode->i_op = &btrfs_special_inode_operations;
4694                 init_special_inode(inode, inode->i_mode, rdev);
4695                 btrfs_update_inode(trans, root, inode);
4696         }
4697         btrfs_update_inode_block_group(trans, inode);
4698         btrfs_update_inode_block_group(trans, dir);
4699 out_unlock:
4700         nr = trans->blocks_used;
4701         btrfs_end_transaction_throttle(trans, root);
4702         btrfs_btree_balance_dirty(root, nr);
4703         if (drop_inode) {
4704                 inode_dec_link_count(inode);
4705                 iput(inode);
4706         }
4707         return err;
4708 }
4709
4710 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4711                         int mode, struct nameidata *nd)
4712 {
4713         struct btrfs_trans_handle *trans;
4714         struct btrfs_root *root = BTRFS_I(dir)->root;
4715         struct inode *inode = NULL;
4716         int drop_inode = 0;
4717         int err;
4718         unsigned long nr = 0;
4719         u64 objectid;
4720         u64 index = 0;
4721
4722         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4723         if (err)
4724                 return err;
4725         /*
4726          * 2 for inode item and ref
4727          * 2 for dir items
4728          * 1 for xattr if selinux is on
4729          */
4730         trans = btrfs_start_transaction(root, 5);
4731         if (IS_ERR(trans))
4732                 return PTR_ERR(trans);
4733
4734         btrfs_set_trans_block_group(trans, dir);
4735
4736         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4737                                 dentry->d_name.len, dir->i_ino, objectid,
4738                                 BTRFS_I(dir)->block_group, mode, &index);
4739         err = PTR_ERR(inode);
4740         if (IS_ERR(inode))
4741                 goto out_unlock;
4742
4743         err = btrfs_init_inode_security(trans, inode, dir);
4744         if (err) {
4745                 drop_inode = 1;
4746                 goto out_unlock;
4747         }
4748
4749         btrfs_set_trans_block_group(trans, inode);
4750         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4751         if (err)
4752                 drop_inode = 1;
4753         else {
4754                 inode->i_mapping->a_ops = &btrfs_aops;
4755                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4756                 inode->i_fop = &btrfs_file_operations;
4757                 inode->i_op = &btrfs_file_inode_operations;
4758                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4759         }
4760         btrfs_update_inode_block_group(trans, inode);
4761         btrfs_update_inode_block_group(trans, dir);
4762 out_unlock:
4763         nr = trans->blocks_used;
4764         btrfs_end_transaction_throttle(trans, root);
4765         if (drop_inode) {
4766                 inode_dec_link_count(inode);
4767                 iput(inode);
4768         }
4769         btrfs_btree_balance_dirty(root, nr);
4770         return err;
4771 }
4772
4773 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4774                       struct dentry *dentry)
4775 {
4776         struct btrfs_trans_handle *trans;
4777         struct btrfs_root *root = BTRFS_I(dir)->root;
4778         struct inode *inode = old_dentry->d_inode;
4779         u64 index;
4780         unsigned long nr = 0;
4781         int err;
4782         int drop_inode = 0;
4783
4784         if (inode->i_nlink == 0)
4785                 return -ENOENT;
4786
4787         /* do not allow sys_link's with other subvols of the same device */
4788         if (root->objectid != BTRFS_I(inode)->root->objectid)
4789                 return -EPERM;
4790
4791         btrfs_inc_nlink(inode);
4792         inode->i_ctime = CURRENT_TIME;
4793
4794         err = btrfs_set_inode_index(dir, &index);
4795         if (err)
4796                 goto fail;
4797
4798         /*
4799          * 1 item for inode ref
4800          * 2 items for dir items
4801          */
4802         trans = btrfs_start_transaction(root, 3);
4803         if (IS_ERR(trans)) {
4804                 err = PTR_ERR(trans);
4805                 goto fail;
4806         }
4807
4808         btrfs_set_trans_block_group(trans, dir);
4809         atomic_inc(&inode->i_count);
4810
4811         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4812
4813         if (err) {
4814                 drop_inode = 1;
4815         } else {
4816                 struct dentry *parent = dget_parent(dentry);
4817                 btrfs_update_inode_block_group(trans, dir);
4818                 err = btrfs_update_inode(trans, root, inode);
4819                 BUG_ON(err);
4820                 btrfs_log_new_name(trans, inode, NULL, parent);
4821                 dput(parent);
4822         }
4823
4824         nr = trans->blocks_used;
4825         btrfs_end_transaction_throttle(trans, root);
4826 fail:
4827         if (drop_inode) {
4828                 inode_dec_link_count(inode);
4829                 iput(inode);
4830         }
4831         btrfs_btree_balance_dirty(root, nr);
4832         return err;
4833 }
4834
4835 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4836 {
4837         struct inode *inode = NULL;
4838         struct btrfs_trans_handle *trans;
4839         struct btrfs_root *root = BTRFS_I(dir)->root;
4840         int err = 0;
4841         int drop_on_err = 0;
4842         u64 objectid = 0;
4843         u64 index = 0;
4844         unsigned long nr = 1;
4845
4846         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4847         if (err)
4848                 return err;
4849
4850         /*
4851          * 2 items for inode and ref
4852          * 2 items for dir items
4853          * 1 for xattr if selinux is on
4854          */
4855         trans = btrfs_start_transaction(root, 5);
4856         if (IS_ERR(trans))
4857                 return PTR_ERR(trans);
4858         btrfs_set_trans_block_group(trans, dir);
4859
4860         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4861                                 dentry->d_name.len, dir->i_ino, objectid,
4862                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4863                                 &index);
4864         if (IS_ERR(inode)) {
4865                 err = PTR_ERR(inode);
4866                 goto out_fail;
4867         }
4868
4869         drop_on_err = 1;
4870
4871         err = btrfs_init_inode_security(trans, inode, dir);
4872         if (err)
4873                 goto out_fail;
4874
4875         inode->i_op = &btrfs_dir_inode_operations;
4876         inode->i_fop = &btrfs_dir_file_operations;
4877         btrfs_set_trans_block_group(trans, inode);
4878
4879         btrfs_i_size_write(inode, 0);
4880         err = btrfs_update_inode(trans, root, inode);
4881         if (err)
4882                 goto out_fail;
4883
4884         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4885                              dentry->d_name.len, 0, index);
4886         if (err)
4887                 goto out_fail;
4888
4889         d_instantiate(dentry, inode);
4890         drop_on_err = 0;
4891         btrfs_update_inode_block_group(trans, inode);
4892         btrfs_update_inode_block_group(trans, dir);
4893
4894 out_fail:
4895         nr = trans->blocks_used;
4896         btrfs_end_transaction_throttle(trans, root);
4897         if (drop_on_err)
4898                 iput(inode);
4899         btrfs_btree_balance_dirty(root, nr);
4900         return err;
4901 }
4902
4903 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4904  * and an extent that you want to insert, deal with overlap and insert
4905  * the new extent into the tree.
4906  */
4907 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4908                                 struct extent_map *existing,
4909                                 struct extent_map *em,
4910                                 u64 map_start, u64 map_len)
4911 {
4912         u64 start_diff;
4913
4914         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4915         start_diff = map_start - em->start;
4916         em->start = map_start;
4917         em->len = map_len;
4918         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4919             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4920                 em->block_start += start_diff;
4921                 em->block_len -= start_diff;
4922         }
4923         return add_extent_mapping(em_tree, em);
4924 }
4925
4926 static noinline int uncompress_inline(struct btrfs_path *path,
4927                                       struct inode *inode, struct page *page,
4928                                       size_t pg_offset, u64 extent_offset,
4929                                       struct btrfs_file_extent_item *item)
4930 {
4931         int ret;
4932         struct extent_buffer *leaf = path->nodes[0];
4933         char *tmp;
4934         size_t max_size;
4935         unsigned long inline_size;
4936         unsigned long ptr;
4937
4938         WARN_ON(pg_offset != 0);
4939         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4940         inline_size = btrfs_file_extent_inline_item_len(leaf,
4941                                         btrfs_item_nr(leaf, path->slots[0]));
4942         tmp = kmalloc(inline_size, GFP_NOFS);
4943         ptr = btrfs_file_extent_inline_start(item);
4944
4945         read_extent_buffer(leaf, tmp, ptr, inline_size);
4946
4947         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4948         ret = btrfs_zlib_decompress(tmp, page, extent_offset,
4949                                     inline_size, max_size);
4950         if (ret) {
4951                 char *kaddr = kmap_atomic(page, KM_USER0);
4952                 unsigned long copy_size = min_t(u64,
4953                                   PAGE_CACHE_SIZE - pg_offset,
4954                                   max_size - extent_offset);
4955                 memset(kaddr + pg_offset, 0, copy_size);
4956                 kunmap_atomic(kaddr, KM_USER0);
4957         }
4958         kfree(tmp);
4959         return 0;
4960 }
4961
4962 /*
4963  * a bit scary, this does extent mapping from logical file offset to the disk.
4964  * the ugly parts come from merging extents from the disk with the in-ram
4965  * representation.  This gets more complex because of the data=ordered code,
4966  * where the in-ram extents might be locked pending data=ordered completion.
4967  *
4968  * This also copies inline extents directly into the page.
4969  */
4970
4971 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4972                                     size_t pg_offset, u64 start, u64 len,
4973                                     int create)
4974 {
4975         int ret;
4976         int err = 0;
4977         u64 bytenr;
4978         u64 extent_start = 0;
4979         u64 extent_end = 0;
4980         u64 objectid = inode->i_ino;
4981         u32 found_type;
4982         struct btrfs_path *path = NULL;
4983         struct btrfs_root *root = BTRFS_I(inode)->root;
4984         struct btrfs_file_extent_item *item;
4985         struct extent_buffer *leaf;
4986         struct btrfs_key found_key;
4987         struct extent_map *em = NULL;
4988         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4989         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4990         struct btrfs_trans_handle *trans = NULL;
4991         int compressed;
4992
4993 again:
4994         read_lock(&em_tree->lock);
4995         em = lookup_extent_mapping(em_tree, start, len);
4996         if (em)
4997                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4998         read_unlock(&em_tree->lock);
4999
5000         if (em) {
5001                 if (em->start > start || em->start + em->len <= start)
5002                         free_extent_map(em);
5003                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5004                         free_extent_map(em);
5005                 else
5006                         goto out;
5007         }
5008         em = alloc_extent_map(GFP_NOFS);
5009         if (!em) {
5010                 err = -ENOMEM;
5011                 goto out;
5012         }
5013         em->bdev = root->fs_info->fs_devices->latest_bdev;
5014         em->start = EXTENT_MAP_HOLE;
5015         em->orig_start = EXTENT_MAP_HOLE;
5016         em->len = (u64)-1;
5017         em->block_len = (u64)-1;
5018
5019         if (!path) {
5020                 path = btrfs_alloc_path();
5021                 BUG_ON(!path);
5022         }
5023
5024         ret = btrfs_lookup_file_extent(trans, root, path,
5025                                        objectid, start, trans != NULL);
5026         if (ret < 0) {
5027                 err = ret;
5028                 goto out;
5029         }
5030
5031         if (ret != 0) {
5032                 if (path->slots[0] == 0)
5033                         goto not_found;
5034                 path->slots[0]--;
5035         }
5036
5037         leaf = path->nodes[0];
5038         item = btrfs_item_ptr(leaf, path->slots[0],
5039                               struct btrfs_file_extent_item);
5040         /* are we inside the extent that was found? */
5041         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5042         found_type = btrfs_key_type(&found_key);
5043         if (found_key.objectid != objectid ||
5044             found_type != BTRFS_EXTENT_DATA_KEY) {
5045                 goto not_found;
5046         }
5047
5048         found_type = btrfs_file_extent_type(leaf, item);
5049         extent_start = found_key.offset;
5050         compressed = btrfs_file_extent_compression(leaf, item);
5051         if (found_type == BTRFS_FILE_EXTENT_REG ||
5052             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5053                 extent_end = extent_start +
5054                        btrfs_file_extent_num_bytes(leaf, item);
5055         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5056                 size_t size;
5057                 size = btrfs_file_extent_inline_len(leaf, item);
5058                 extent_end = (extent_start + size + root->sectorsize - 1) &
5059                         ~((u64)root->sectorsize - 1);
5060         }
5061
5062         if (start >= extent_end) {
5063                 path->slots[0]++;
5064                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5065                         ret = btrfs_next_leaf(root, path);
5066                         if (ret < 0) {
5067                                 err = ret;
5068                                 goto out;
5069                         }
5070                         if (ret > 0)
5071                                 goto not_found;
5072                         leaf = path->nodes[0];
5073                 }
5074                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5075                 if (found_key.objectid != objectid ||
5076                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5077                         goto not_found;
5078                 if (start + len <= found_key.offset)
5079                         goto not_found;
5080                 em->start = start;
5081                 em->len = found_key.offset - start;
5082                 goto not_found_em;
5083         }
5084
5085         if (found_type == BTRFS_FILE_EXTENT_REG ||
5086             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5087                 em->start = extent_start;
5088                 em->len = extent_end - extent_start;
5089                 em->orig_start = extent_start -
5090                                  btrfs_file_extent_offset(leaf, item);
5091                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5092                 if (bytenr == 0) {
5093                         em->block_start = EXTENT_MAP_HOLE;
5094                         goto insert;
5095                 }
5096                 if (compressed) {
5097                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5098                         em->block_start = bytenr;
5099                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5100                                                                          item);
5101                 } else {
5102                         bytenr += btrfs_file_extent_offset(leaf, item);
5103                         em->block_start = bytenr;
5104                         em->block_len = em->len;
5105                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5106                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5107                 }
5108                 goto insert;
5109         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5110                 unsigned long ptr;
5111                 char *map;
5112                 size_t size;
5113                 size_t extent_offset;
5114                 size_t copy_size;
5115
5116                 em->block_start = EXTENT_MAP_INLINE;
5117                 if (!page || create) {
5118                         em->start = extent_start;
5119                         em->len = extent_end - extent_start;
5120                         goto out;
5121                 }
5122
5123                 size = btrfs_file_extent_inline_len(leaf, item);
5124                 extent_offset = page_offset(page) + pg_offset - extent_start;
5125                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5126                                 size - extent_offset);
5127                 em->start = extent_start + extent_offset;
5128                 em->len = (copy_size + root->sectorsize - 1) &
5129                         ~((u64)root->sectorsize - 1);
5130                 em->orig_start = EXTENT_MAP_INLINE;
5131                 if (compressed)
5132                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5133                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5134                 if (create == 0 && !PageUptodate(page)) {
5135                         if (btrfs_file_extent_compression(leaf, item) ==
5136                             BTRFS_COMPRESS_ZLIB) {
5137                                 ret = uncompress_inline(path, inode, page,
5138                                                         pg_offset,
5139                                                         extent_offset, item);
5140                                 BUG_ON(ret);
5141                         } else {
5142                                 map = kmap(page);
5143                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5144                                                    copy_size);
5145                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5146                                         memset(map + pg_offset + copy_size, 0,
5147                                                PAGE_CACHE_SIZE - pg_offset -
5148                                                copy_size);
5149                                 }
5150                                 kunmap(page);
5151                         }
5152                         flush_dcache_page(page);
5153                 } else if (create && PageUptodate(page)) {
5154                         WARN_ON(1);
5155                         if (!trans) {
5156                                 kunmap(page);
5157                                 free_extent_map(em);
5158                                 em = NULL;
5159                                 btrfs_release_path(root, path);
5160                                 trans = btrfs_join_transaction(root, 1);
5161                                 goto again;
5162                         }
5163                         map = kmap(page);
5164                         write_extent_buffer(leaf, map + pg_offset, ptr,
5165                                             copy_size);
5166                         kunmap(page);
5167                         btrfs_mark_buffer_dirty(leaf);
5168                 }
5169                 set_extent_uptodate(io_tree, em->start,
5170                                     extent_map_end(em) - 1, GFP_NOFS);
5171                 goto insert;
5172         } else {
5173                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5174                 WARN_ON(1);
5175         }
5176 not_found:
5177         em->start = start;
5178         em->len = len;
5179 not_found_em:
5180         em->block_start = EXTENT_MAP_HOLE;
5181         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5182 insert:
5183         btrfs_release_path(root, path);
5184         if (em->start > start || extent_map_end(em) <= start) {
5185                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5186                        "[%llu %llu]\n", (unsigned long long)em->start,
5187                        (unsigned long long)em->len,
5188                        (unsigned long long)start,
5189                        (unsigned long long)len);
5190                 err = -EIO;
5191                 goto out;
5192         }
5193
5194         err = 0;
5195         write_lock(&em_tree->lock);
5196         ret = add_extent_mapping(em_tree, em);
5197         /* it is possible that someone inserted the extent into the tree
5198          * while we had the lock dropped.  It is also possible that
5199          * an overlapping map exists in the tree
5200          */
5201         if (ret == -EEXIST) {
5202                 struct extent_map *existing;
5203
5204                 ret = 0;
5205
5206                 existing = lookup_extent_mapping(em_tree, start, len);
5207                 if (existing && (existing->start > start ||
5208                     existing->start + existing->len <= start)) {
5209                         free_extent_map(existing);
5210                         existing = NULL;
5211                 }
5212                 if (!existing) {
5213                         existing = lookup_extent_mapping(em_tree, em->start,
5214                                                          em->len);
5215                         if (existing) {
5216                                 err = merge_extent_mapping(em_tree, existing,
5217                                                            em, start,
5218                                                            root->sectorsize);
5219                                 free_extent_map(existing);
5220                                 if (err) {
5221                                         free_extent_map(em);
5222                                         em = NULL;
5223                                 }
5224                         } else {
5225                                 err = -EIO;
5226                                 free_extent_map(em);
5227                                 em = NULL;
5228                         }
5229                 } else {
5230                         free_extent_map(em);
5231                         em = existing;
5232                         err = 0;
5233                 }
5234         }
5235         write_unlock(&em_tree->lock);
5236 out:
5237         if (path)
5238                 btrfs_free_path(path);
5239         if (trans) {
5240                 ret = btrfs_end_transaction(trans, root);
5241                 if (!err)
5242                         err = ret;
5243         }
5244         if (err) {
5245                 free_extent_map(em);
5246                 return ERR_PTR(err);
5247         }
5248         return em;
5249 }
5250
5251 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5252                                                   u64 start, u64 len)
5253 {
5254         struct btrfs_root *root = BTRFS_I(inode)->root;
5255         struct btrfs_trans_handle *trans;
5256         struct extent_map *em;
5257         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5258         struct btrfs_key ins;
5259         u64 alloc_hint;
5260         int ret;
5261
5262         btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5263
5264         trans = btrfs_join_transaction(root, 0);
5265         if (!trans)
5266                 return ERR_PTR(-ENOMEM);
5267
5268         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5269
5270         alloc_hint = get_extent_allocation_hint(inode, start, len);
5271         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5272                                    alloc_hint, (u64)-1, &ins, 1);
5273         if (ret) {
5274                 em = ERR_PTR(ret);
5275                 goto out;
5276         }
5277
5278         em = alloc_extent_map(GFP_NOFS);
5279         if (!em) {
5280                 em = ERR_PTR(-ENOMEM);
5281                 goto out;
5282         }
5283
5284         em->start = start;
5285         em->orig_start = em->start;
5286         em->len = ins.offset;
5287
5288         em->block_start = ins.objectid;
5289         em->block_len = ins.offset;
5290         em->bdev = root->fs_info->fs_devices->latest_bdev;
5291         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5292
5293         while (1) {
5294                 write_lock(&em_tree->lock);
5295                 ret = add_extent_mapping(em_tree, em);
5296                 write_unlock(&em_tree->lock);
5297                 if (ret != -EEXIST)
5298                         break;
5299                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5300         }
5301
5302         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5303                                            ins.offset, ins.offset, 0);
5304         if (ret) {
5305                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5306                 em = ERR_PTR(ret);
5307         }
5308 out:
5309         btrfs_end_transaction(trans, root);
5310         return em;
5311 }
5312
5313 /*
5314  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5315  * block must be cow'd
5316  */
5317 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5318                                       struct inode *inode, u64 offset, u64 len)
5319 {
5320         struct btrfs_path *path;
5321         int ret;
5322         struct extent_buffer *leaf;
5323         struct btrfs_root *root = BTRFS_I(inode)->root;
5324         struct btrfs_file_extent_item *fi;
5325         struct btrfs_key key;
5326         u64 disk_bytenr;
5327         u64 backref_offset;
5328         u64 extent_end;
5329         u64 num_bytes;
5330         int slot;
5331         int found_type;
5332
5333         path = btrfs_alloc_path();
5334         if (!path)
5335                 return -ENOMEM;
5336
5337         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5338                                        offset, 0);
5339         if (ret < 0)
5340                 goto out;
5341
5342         slot = path->slots[0];
5343         if (ret == 1) {
5344                 if (slot == 0) {
5345                         /* can't find the item, must cow */
5346                         ret = 0;
5347                         goto out;
5348                 }
5349                 slot--;
5350         }
5351         ret = 0;
5352         leaf = path->nodes[0];
5353         btrfs_item_key_to_cpu(leaf, &key, slot);
5354         if (key.objectid != inode->i_ino ||
5355             key.type != BTRFS_EXTENT_DATA_KEY) {
5356                 /* not our file or wrong item type, must cow */
5357                 goto out;
5358         }
5359
5360         if (key.offset > offset) {
5361                 /* Wrong offset, must cow */
5362                 goto out;
5363         }
5364
5365         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5366         found_type = btrfs_file_extent_type(leaf, fi);
5367         if (found_type != BTRFS_FILE_EXTENT_REG &&
5368             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5369                 /* not a regular extent, must cow */
5370                 goto out;
5371         }
5372         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5373         backref_offset = btrfs_file_extent_offset(leaf, fi);
5374
5375         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5376         if (extent_end < offset + len) {
5377                 /* extent doesn't include our full range, must cow */
5378                 goto out;
5379         }
5380
5381         if (btrfs_extent_readonly(root, disk_bytenr))
5382                 goto out;
5383
5384         /*
5385          * look for other files referencing this extent, if we
5386          * find any we must cow
5387          */
5388         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5389                                   key.offset - backref_offset, disk_bytenr))
5390                 goto out;
5391
5392         /*
5393          * adjust disk_bytenr and num_bytes to cover just the bytes
5394          * in this extent we are about to write.  If there
5395          * are any csums in that range we have to cow in order
5396          * to keep the csums correct
5397          */
5398         disk_bytenr += backref_offset;
5399         disk_bytenr += offset - key.offset;
5400         num_bytes = min(offset + len, extent_end) - offset;
5401         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5402                                 goto out;
5403         /*
5404          * all of the above have passed, it is safe to overwrite this extent
5405          * without cow
5406          */
5407         ret = 1;
5408 out:
5409         btrfs_free_path(path);
5410         return ret;
5411 }
5412
5413 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5414                                    struct buffer_head *bh_result, int create)
5415 {
5416         struct extent_map *em;
5417         struct btrfs_root *root = BTRFS_I(inode)->root;
5418         u64 start = iblock << inode->i_blkbits;
5419         u64 len = bh_result->b_size;
5420         struct btrfs_trans_handle *trans;
5421
5422         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5423         if (IS_ERR(em))
5424                 return PTR_ERR(em);
5425
5426         /*
5427          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5428          * io.  INLINE is special, and we could probably kludge it in here, but
5429          * it's still buffered so for safety lets just fall back to the generic
5430          * buffered path.
5431          *
5432          * For COMPRESSED we _have_ to read the entire extent in so we can
5433          * decompress it, so there will be buffering required no matter what we
5434          * do, so go ahead and fallback to buffered.
5435          *
5436          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5437          * to buffered IO.  Don't blame me, this is the price we pay for using
5438          * the generic code.
5439          */
5440         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5441             em->block_start == EXTENT_MAP_INLINE) {
5442                 free_extent_map(em);
5443                 return -ENOTBLK;
5444         }
5445
5446         /* Just a good old fashioned hole, return */
5447         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5448                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5449                 free_extent_map(em);
5450                 /* DIO will do one hole at a time, so just unlock a sector */
5451                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5452                               start + root->sectorsize - 1, GFP_NOFS);
5453                 return 0;
5454         }
5455
5456         /*
5457          * We don't allocate a new extent in the following cases
5458          *
5459          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5460          * existing extent.
5461          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5462          * just use the extent.
5463          *
5464          */
5465         if (!create) {
5466                 len = em->len - (start - em->start);
5467                 goto map;
5468         }
5469
5470         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5471             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5472              em->block_start != EXTENT_MAP_HOLE)) {
5473                 int type;
5474                 int ret;
5475                 u64 block_start;
5476
5477                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5478                         type = BTRFS_ORDERED_PREALLOC;
5479                 else
5480                         type = BTRFS_ORDERED_NOCOW;
5481                 len = min(len, em->len - (start - em->start));
5482                 block_start = em->block_start + (start - em->start);
5483
5484                 /*
5485                  * we're not going to log anything, but we do need
5486                  * to make sure the current transaction stays open
5487                  * while we look for nocow cross refs
5488                  */
5489                 trans = btrfs_join_transaction(root, 0);
5490                 if (!trans)
5491                         goto must_cow;
5492
5493                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5494                         ret = btrfs_add_ordered_extent_dio(inode, start,
5495                                            block_start, len, len, type);
5496                         btrfs_end_transaction(trans, root);
5497                         if (ret) {
5498                                 free_extent_map(em);
5499                                 return ret;
5500                         }
5501                         goto unlock;
5502                 }
5503                 btrfs_end_transaction(trans, root);
5504         }
5505 must_cow:
5506         /*
5507          * this will cow the extent, reset the len in case we changed
5508          * it above
5509          */
5510         len = bh_result->b_size;
5511         free_extent_map(em);
5512         em = btrfs_new_extent_direct(inode, start, len);
5513         if (IS_ERR(em))
5514                 return PTR_ERR(em);
5515         len = min(len, em->len - (start - em->start));
5516 unlock:
5517         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5518                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5519                           0, NULL, GFP_NOFS);
5520 map:
5521         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5522                 inode->i_blkbits;
5523         bh_result->b_size = len;
5524         bh_result->b_bdev = em->bdev;
5525         set_buffer_mapped(bh_result);
5526         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5527                 set_buffer_new(bh_result);
5528
5529         free_extent_map(em);
5530
5531         return 0;
5532 }
5533
5534 struct btrfs_dio_private {
5535         struct inode *inode;
5536         u64 logical_offset;
5537         u64 disk_bytenr;
5538         u64 bytes;
5539         u32 *csums;
5540         void *private;
5541
5542         /* number of bios pending for this dio */
5543         atomic_t pending_bios;
5544
5545         /* IO errors */
5546         int errors;
5547
5548         struct bio *orig_bio;
5549 };
5550
5551 static void btrfs_endio_direct_read(struct bio *bio, int err)
5552 {
5553         struct btrfs_dio_private *dip = bio->bi_private;
5554         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5555         struct bio_vec *bvec = bio->bi_io_vec;
5556         struct inode *inode = dip->inode;
5557         struct btrfs_root *root = BTRFS_I(inode)->root;
5558         u64 start;
5559         u32 *private = dip->csums;
5560
5561         start = dip->logical_offset;
5562         do {
5563                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5564                         struct page *page = bvec->bv_page;
5565                         char *kaddr;
5566                         u32 csum = ~(u32)0;
5567                         unsigned long flags;
5568
5569                         local_irq_save(flags);
5570                         kaddr = kmap_atomic(page, KM_IRQ0);
5571                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5572                                                csum, bvec->bv_len);
5573                         btrfs_csum_final(csum, (char *)&csum);
5574                         kunmap_atomic(kaddr, KM_IRQ0);
5575                         local_irq_restore(flags);
5576
5577                         flush_dcache_page(bvec->bv_page);
5578                         if (csum != *private) {
5579                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5580                                       " %llu csum %u private %u\n",
5581                                       inode->i_ino, (unsigned long long)start,
5582                                       csum, *private);
5583                                 err = -EIO;
5584                         }
5585                 }
5586
5587                 start += bvec->bv_len;
5588                 private++;
5589                 bvec++;
5590         } while (bvec <= bvec_end);
5591
5592         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5593                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5594         bio->bi_private = dip->private;
5595
5596         kfree(dip->csums);
5597         kfree(dip);
5598         dio_end_io(bio, err);
5599 }
5600
5601 static void btrfs_endio_direct_write(struct bio *bio, int err)
5602 {
5603         struct btrfs_dio_private *dip = bio->bi_private;
5604         struct inode *inode = dip->inode;
5605         struct btrfs_root *root = BTRFS_I(inode)->root;
5606         struct btrfs_trans_handle *trans;
5607         struct btrfs_ordered_extent *ordered = NULL;
5608         struct extent_state *cached_state = NULL;
5609         u64 ordered_offset = dip->logical_offset;
5610         u64 ordered_bytes = dip->bytes;
5611         int ret;
5612
5613         if (err)
5614                 goto out_done;
5615 again:
5616         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5617                                                    &ordered_offset,
5618                                                    ordered_bytes);
5619         if (!ret)
5620                 goto out_test;
5621
5622         BUG_ON(!ordered);
5623
5624         trans = btrfs_join_transaction(root, 1);
5625         if (!trans) {
5626                 err = -ENOMEM;
5627                 goto out;
5628         }
5629         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5630
5631         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5632                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5633                 if (!ret)
5634                         ret = btrfs_update_inode(trans, root, inode);
5635                 err = ret;
5636                 goto out;
5637         }
5638
5639         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5640                          ordered->file_offset + ordered->len - 1, 0,
5641                          &cached_state, GFP_NOFS);
5642
5643         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5644                 ret = btrfs_mark_extent_written(trans, inode,
5645                                                 ordered->file_offset,
5646                                                 ordered->file_offset +
5647                                                 ordered->len);
5648                 if (ret) {
5649                         err = ret;
5650                         goto out_unlock;
5651                 }
5652         } else {
5653                 ret = insert_reserved_file_extent(trans, inode,
5654                                                   ordered->file_offset,
5655                                                   ordered->start,
5656                                                   ordered->disk_len,
5657                                                   ordered->len,
5658                                                   ordered->len,
5659                                                   0, 0, 0,
5660                                                   BTRFS_FILE_EXTENT_REG);
5661                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5662                                    ordered->file_offset, ordered->len);
5663                 if (ret) {
5664                         err = ret;
5665                         WARN_ON(1);
5666                         goto out_unlock;
5667                 }
5668         }
5669
5670         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5671         btrfs_ordered_update_i_size(inode, 0, ordered);
5672         btrfs_update_inode(trans, root, inode);
5673 out_unlock:
5674         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5675                              ordered->file_offset + ordered->len - 1,
5676                              &cached_state, GFP_NOFS);
5677 out:
5678         btrfs_delalloc_release_metadata(inode, ordered->len);
5679         btrfs_end_transaction(trans, root);
5680         ordered_offset = ordered->file_offset + ordered->len;
5681         btrfs_put_ordered_extent(ordered);
5682         btrfs_put_ordered_extent(ordered);
5683
5684 out_test:
5685         /*
5686          * our bio might span multiple ordered extents.  If we haven't
5687          * completed the accounting for the whole dio, go back and try again
5688          */
5689         if (ordered_offset < dip->logical_offset + dip->bytes) {
5690                 ordered_bytes = dip->logical_offset + dip->bytes -
5691                         ordered_offset;
5692                 goto again;
5693         }
5694 out_done:
5695         bio->bi_private = dip->private;
5696
5697         kfree(dip->csums);
5698         kfree(dip);
5699         dio_end_io(bio, err);
5700 }
5701
5702 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5703                                     struct bio *bio, int mirror_num,
5704                                     unsigned long bio_flags, u64 offset)
5705 {
5706         int ret;
5707         struct btrfs_root *root = BTRFS_I(inode)->root;
5708         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5709         BUG_ON(ret);
5710         return 0;
5711 }
5712
5713 static void btrfs_end_dio_bio(struct bio *bio, int err)
5714 {
5715         struct btrfs_dio_private *dip = bio->bi_private;
5716
5717         if (err) {
5718                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5719                       "sector %#Lx len %u err no %d\n",
5720                       dip->inode->i_ino, bio->bi_rw,
5721                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5722                 dip->errors = 1;
5723
5724                 /*
5725                  * before atomic variable goto zero, we must make sure
5726                  * dip->errors is perceived to be set.
5727                  */
5728                 smp_mb__before_atomic_dec();
5729         }
5730
5731         /* if there are more bios still pending for this dio, just exit */
5732         if (!atomic_dec_and_test(&dip->pending_bios))
5733                 goto out;
5734
5735         if (dip->errors)
5736                 bio_io_error(dip->orig_bio);
5737         else {
5738                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5739                 bio_endio(dip->orig_bio, 0);
5740         }
5741 out:
5742         bio_put(bio);
5743 }
5744
5745 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5746                                        u64 first_sector, gfp_t gfp_flags)
5747 {
5748         int nr_vecs = bio_get_nr_vecs(bdev);
5749         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5750 }
5751
5752 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5753                                          int rw, u64 file_offset, int skip_sum,
5754                                          u32 *csums)
5755 {
5756         int write = rw & REQ_WRITE;
5757         struct btrfs_root *root = BTRFS_I(inode)->root;
5758         int ret;
5759
5760         bio_get(bio);
5761         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5762         if (ret)
5763                 goto err;
5764
5765         if (write && !skip_sum) {
5766                 ret = btrfs_wq_submit_bio(root->fs_info,
5767                                    inode, rw, bio, 0, 0,
5768                                    file_offset,
5769                                    __btrfs_submit_bio_start_direct_io,
5770                                    __btrfs_submit_bio_done);
5771                 goto err;
5772         } else if (!skip_sum)
5773                 btrfs_lookup_bio_sums_dio(root, inode, bio,
5774                                           file_offset, csums);
5775
5776         ret = btrfs_map_bio(root, rw, bio, 0, 1);
5777 err:
5778         bio_put(bio);
5779         return ret;
5780 }
5781
5782 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5783                                     int skip_sum)
5784 {
5785         struct inode *inode = dip->inode;
5786         struct btrfs_root *root = BTRFS_I(inode)->root;
5787         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5788         struct bio *bio;
5789         struct bio *orig_bio = dip->orig_bio;
5790         struct bio_vec *bvec = orig_bio->bi_io_vec;
5791         u64 start_sector = orig_bio->bi_sector;
5792         u64 file_offset = dip->logical_offset;
5793         u64 submit_len = 0;
5794         u64 map_length;
5795         int nr_pages = 0;
5796         u32 *csums = dip->csums;
5797         int ret = 0;
5798
5799         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5800         if (!bio)
5801                 return -ENOMEM;
5802         bio->bi_private = dip;
5803         bio->bi_end_io = btrfs_end_dio_bio;
5804         atomic_inc(&dip->pending_bios);
5805
5806         map_length = orig_bio->bi_size;
5807         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5808                               &map_length, NULL, 0);
5809         if (ret) {
5810                 bio_put(bio);
5811                 return -EIO;
5812         }
5813
5814         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5815                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5816                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5817                                  bvec->bv_offset) < bvec->bv_len)) {
5818                         /*
5819                          * inc the count before we submit the bio so
5820                          * we know the end IO handler won't happen before
5821                          * we inc the count. Otherwise, the dip might get freed
5822                          * before we're done setting it up
5823                          */
5824                         atomic_inc(&dip->pending_bios);
5825                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
5826                                                      file_offset, skip_sum,
5827                                                      csums);
5828                         if (ret) {
5829                                 bio_put(bio);
5830                                 atomic_dec(&dip->pending_bios);
5831                                 goto out_err;
5832                         }
5833
5834                         if (!skip_sum)
5835                                 csums = csums + nr_pages;
5836                         start_sector += submit_len >> 9;
5837                         file_offset += submit_len;
5838
5839                         submit_len = 0;
5840                         nr_pages = 0;
5841
5842                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
5843                                                   start_sector, GFP_NOFS);
5844                         if (!bio)
5845                                 goto out_err;
5846                         bio->bi_private = dip;
5847                         bio->bi_end_io = btrfs_end_dio_bio;
5848
5849                         map_length = orig_bio->bi_size;
5850                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5851                                               &map_length, NULL, 0);
5852                         if (ret) {
5853                                 bio_put(bio);
5854                                 goto out_err;
5855                         }
5856                 } else {
5857                         submit_len += bvec->bv_len;
5858                         nr_pages ++;
5859                         bvec++;
5860                 }
5861         }
5862
5863         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
5864                                      csums);
5865         if (!ret)
5866                 return 0;
5867
5868         bio_put(bio);
5869 out_err:
5870         dip->errors = 1;
5871         /*
5872          * before atomic variable goto zero, we must
5873          * make sure dip->errors is perceived to be set.
5874          */
5875         smp_mb__before_atomic_dec();
5876         if (atomic_dec_and_test(&dip->pending_bios))
5877                 bio_io_error(dip->orig_bio);
5878
5879         /* bio_end_io() will handle error, so we needn't return it */
5880         return 0;
5881 }
5882
5883 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
5884                                 loff_t file_offset)
5885 {
5886         struct btrfs_root *root = BTRFS_I(inode)->root;
5887         struct btrfs_dio_private *dip;
5888         struct bio_vec *bvec = bio->bi_io_vec;
5889         int skip_sum;
5890         int write = rw & REQ_WRITE;
5891         int ret = 0;
5892
5893         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
5894
5895         dip = kmalloc(sizeof(*dip), GFP_NOFS);
5896         if (!dip) {
5897                 ret = -ENOMEM;
5898                 goto free_ordered;
5899         }
5900         dip->csums = NULL;
5901
5902         if (!skip_sum) {
5903                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
5904                 if (!dip->csums) {
5905                         ret = -ENOMEM;
5906                         goto free_ordered;
5907                 }
5908         }
5909
5910         dip->private = bio->bi_private;
5911         dip->inode = inode;
5912         dip->logical_offset = file_offset;
5913
5914         dip->bytes = 0;
5915         do {
5916                 dip->bytes += bvec->bv_len;
5917                 bvec++;
5918         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
5919
5920         dip->disk_bytenr = (u64)bio->bi_sector << 9;
5921         bio->bi_private = dip;
5922         dip->errors = 0;
5923         dip->orig_bio = bio;
5924         atomic_set(&dip->pending_bios, 0);
5925
5926         if (write)
5927                 bio->bi_end_io = btrfs_endio_direct_write;
5928         else
5929                 bio->bi_end_io = btrfs_endio_direct_read;
5930
5931         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
5932         if (!ret)
5933                 return;
5934 free_ordered:
5935         /*
5936          * If this is a write, we need to clean up the reserved space and kill
5937          * the ordered extent.
5938          */
5939         if (write) {
5940                 struct btrfs_ordered_extent *ordered;
5941                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
5942                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
5943                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
5944                         btrfs_free_reserved_extent(root, ordered->start,
5945                                                    ordered->disk_len);
5946                 btrfs_put_ordered_extent(ordered);
5947                 btrfs_put_ordered_extent(ordered);
5948         }
5949         bio_endio(bio, ret);
5950 }
5951
5952 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
5953                         const struct iovec *iov, loff_t offset,
5954                         unsigned long nr_segs)
5955 {
5956         int seg;
5957         size_t size;
5958         unsigned long addr;
5959         unsigned blocksize_mask = root->sectorsize - 1;
5960         ssize_t retval = -EINVAL;
5961         loff_t end = offset;
5962
5963         if (offset & blocksize_mask)
5964                 goto out;
5965
5966         /* Check the memory alignment.  Blocks cannot straddle pages */
5967         for (seg = 0; seg < nr_segs; seg++) {
5968                 addr = (unsigned long)iov[seg].iov_base;
5969                 size = iov[seg].iov_len;
5970                 end += size;
5971                 if ((addr & blocksize_mask) || (size & blocksize_mask))
5972                         goto out;
5973         }
5974         retval = 0;
5975 out:
5976         return retval;
5977 }
5978 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
5979                         const struct iovec *iov, loff_t offset,
5980                         unsigned long nr_segs)
5981 {
5982         struct file *file = iocb->ki_filp;
5983         struct inode *inode = file->f_mapping->host;
5984         struct btrfs_ordered_extent *ordered;
5985         struct extent_state *cached_state = NULL;
5986         u64 lockstart, lockend;
5987         ssize_t ret;
5988         int writing = rw & WRITE;
5989         int write_bits = 0;
5990         size_t count = iov_length(iov, nr_segs);
5991
5992         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
5993                             offset, nr_segs)) {
5994                 return 0;
5995         }
5996
5997         lockstart = offset;
5998         lockend = offset + count - 1;
5999
6000         if (writing) {
6001                 ret = btrfs_delalloc_reserve_space(inode, count);
6002                 if (ret)
6003                         goto out;
6004         }
6005
6006         while (1) {
6007                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6008                                  0, &cached_state, GFP_NOFS);
6009                 /*
6010                  * We're concerned with the entire range that we're going to be
6011                  * doing DIO to, so we need to make sure theres no ordered
6012                  * extents in this range.
6013                  */
6014                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6015                                                      lockend - lockstart + 1);
6016                 if (!ordered)
6017                         break;
6018                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6019                                      &cached_state, GFP_NOFS);
6020                 btrfs_start_ordered_extent(inode, ordered, 1);
6021                 btrfs_put_ordered_extent(ordered);
6022                 cond_resched();
6023         }
6024
6025         /*
6026          * we don't use btrfs_set_extent_delalloc because we don't want
6027          * the dirty or uptodate bits
6028          */
6029         if (writing) {
6030                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6031                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6032                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6033                                      GFP_NOFS);
6034                 if (ret) {
6035                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6036                                          lockend, EXTENT_LOCKED | write_bits,
6037                                          1, 0, &cached_state, GFP_NOFS);
6038                         goto out;
6039                 }
6040         }
6041
6042         free_extent_state(cached_state);
6043         cached_state = NULL;
6044
6045         ret = __blockdev_direct_IO(rw, iocb, inode,
6046                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6047                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6048                    btrfs_submit_direct, 0);
6049
6050         if (ret < 0 && ret != -EIOCBQUEUED) {
6051                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6052                               offset + iov_length(iov, nr_segs) - 1,
6053                               EXTENT_LOCKED | write_bits, 1, 0,
6054                               &cached_state, GFP_NOFS);
6055         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6056                 /*
6057                  * We're falling back to buffered, unlock the section we didn't
6058                  * do IO on.
6059                  */
6060                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6061                               offset + iov_length(iov, nr_segs) - 1,
6062                               EXTENT_LOCKED | write_bits, 1, 0,
6063                               &cached_state, GFP_NOFS);
6064         }
6065 out:
6066         free_extent_state(cached_state);
6067         return ret;
6068 }
6069
6070 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6071                 __u64 start, __u64 len)
6072 {
6073         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent);
6074 }
6075
6076 int btrfs_readpage(struct file *file, struct page *page)
6077 {
6078         struct extent_io_tree *tree;
6079         tree = &BTRFS_I(page->mapping->host)->io_tree;
6080         return extent_read_full_page(tree, page, btrfs_get_extent);
6081 }
6082
6083 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6084 {
6085         struct extent_io_tree *tree;
6086
6087
6088         if (current->flags & PF_MEMALLOC) {
6089                 redirty_page_for_writepage(wbc, page);
6090                 unlock_page(page);
6091                 return 0;
6092         }
6093         tree = &BTRFS_I(page->mapping->host)->io_tree;
6094         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6095 }
6096
6097 int btrfs_writepages(struct address_space *mapping,
6098                      struct writeback_control *wbc)
6099 {
6100         struct extent_io_tree *tree;
6101
6102         tree = &BTRFS_I(mapping->host)->io_tree;
6103         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6104 }
6105
6106 static int
6107 btrfs_readpages(struct file *file, struct address_space *mapping,
6108                 struct list_head *pages, unsigned nr_pages)
6109 {
6110         struct extent_io_tree *tree;
6111         tree = &BTRFS_I(mapping->host)->io_tree;
6112         return extent_readpages(tree, mapping, pages, nr_pages,
6113                                 btrfs_get_extent);
6114 }
6115 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6116 {
6117         struct extent_io_tree *tree;
6118         struct extent_map_tree *map;
6119         int ret;
6120
6121         tree = &BTRFS_I(page->mapping->host)->io_tree;
6122         map = &BTRFS_I(page->mapping->host)->extent_tree;
6123         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6124         if (ret == 1) {
6125                 ClearPagePrivate(page);
6126                 set_page_private(page, 0);
6127                 page_cache_release(page);
6128         }
6129         return ret;
6130 }
6131
6132 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6133 {
6134         if (PageWriteback(page) || PageDirty(page))
6135                 return 0;
6136         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6137 }
6138
6139 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6140 {
6141         struct extent_io_tree *tree;
6142         struct btrfs_ordered_extent *ordered;
6143         struct extent_state *cached_state = NULL;
6144         u64 page_start = page_offset(page);
6145         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6146
6147
6148         /*
6149          * we have the page locked, so new writeback can't start,
6150          * and the dirty bit won't be cleared while we are here.
6151          *
6152          * Wait for IO on this page so that we can safely clear
6153          * the PagePrivate2 bit and do ordered accounting
6154          */
6155         wait_on_page_writeback(page);
6156
6157         tree = &BTRFS_I(page->mapping->host)->io_tree;
6158         if (offset) {
6159                 btrfs_releasepage(page, GFP_NOFS);
6160                 return;
6161         }
6162         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6163                          GFP_NOFS);
6164         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6165                                            page_offset(page));
6166         if (ordered) {
6167                 /*
6168                  * IO on this page will never be started, so we need
6169                  * to account for any ordered extents now
6170                  */
6171                 clear_extent_bit(tree, page_start, page_end,
6172                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6173                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6174                                  &cached_state, GFP_NOFS);
6175                 /*
6176                  * whoever cleared the private bit is responsible
6177                  * for the finish_ordered_io
6178                  */
6179                 if (TestClearPagePrivate2(page)) {
6180                         btrfs_finish_ordered_io(page->mapping->host,
6181                                                 page_start, page_end);
6182                 }
6183                 btrfs_put_ordered_extent(ordered);
6184                 cached_state = NULL;
6185                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6186                                  GFP_NOFS);
6187         }
6188         clear_extent_bit(tree, page_start, page_end,
6189                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6190                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6191         __btrfs_releasepage(page, GFP_NOFS);
6192
6193         ClearPageChecked(page);
6194         if (PagePrivate(page)) {
6195                 ClearPagePrivate(page);
6196                 set_page_private(page, 0);
6197                 page_cache_release(page);
6198         }
6199 }
6200
6201 /*
6202  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6203  * called from a page fault handler when a page is first dirtied. Hence we must
6204  * be careful to check for EOF conditions here. We set the page up correctly
6205  * for a written page which means we get ENOSPC checking when writing into
6206  * holes and correct delalloc and unwritten extent mapping on filesystems that
6207  * support these features.
6208  *
6209  * We are not allowed to take the i_mutex here so we have to play games to
6210  * protect against truncate races as the page could now be beyond EOF.  Because
6211  * vmtruncate() writes the inode size before removing pages, once we have the
6212  * page lock we can determine safely if the page is beyond EOF. If it is not
6213  * beyond EOF, then the page is guaranteed safe against truncation until we
6214  * unlock the page.
6215  */
6216 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6217 {
6218         struct page *page = vmf->page;
6219         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6220         struct btrfs_root *root = BTRFS_I(inode)->root;
6221         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6222         struct btrfs_ordered_extent *ordered;
6223         struct extent_state *cached_state = NULL;
6224         char *kaddr;
6225         unsigned long zero_start;
6226         loff_t size;
6227         int ret;
6228         u64 page_start;
6229         u64 page_end;
6230
6231         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6232         if (ret) {
6233                 if (ret == -ENOMEM)
6234                         ret = VM_FAULT_OOM;
6235                 else /* -ENOSPC, -EIO, etc */
6236                         ret = VM_FAULT_SIGBUS;
6237                 goto out;
6238         }
6239
6240         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6241 again:
6242         lock_page(page);
6243         size = i_size_read(inode);
6244         page_start = page_offset(page);
6245         page_end = page_start + PAGE_CACHE_SIZE - 1;
6246
6247         if ((page->mapping != inode->i_mapping) ||
6248             (page_start >= size)) {
6249                 /* page got truncated out from underneath us */
6250                 goto out_unlock;
6251         }
6252         wait_on_page_writeback(page);
6253
6254         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6255                          GFP_NOFS);
6256         set_page_extent_mapped(page);
6257
6258         /*
6259          * we can't set the delalloc bits if there are pending ordered
6260          * extents.  Drop our locks and wait for them to finish
6261          */
6262         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6263         if (ordered) {
6264                 unlock_extent_cached(io_tree, page_start, page_end,
6265                                      &cached_state, GFP_NOFS);
6266                 unlock_page(page);
6267                 btrfs_start_ordered_extent(inode, ordered, 1);
6268                 btrfs_put_ordered_extent(ordered);
6269                 goto again;
6270         }
6271
6272         /*
6273          * XXX - page_mkwrite gets called every time the page is dirtied, even
6274          * if it was already dirty, so for space accounting reasons we need to
6275          * clear any delalloc bits for the range we are fixing to save.  There
6276          * is probably a better way to do this, but for now keep consistent with
6277          * prepare_pages in the normal write path.
6278          */
6279         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6280                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6281                           0, 0, &cached_state, GFP_NOFS);
6282
6283         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6284                                         &cached_state);
6285         if (ret) {
6286                 unlock_extent_cached(io_tree, page_start, page_end,
6287                                      &cached_state, GFP_NOFS);
6288                 ret = VM_FAULT_SIGBUS;
6289                 goto out_unlock;
6290         }
6291         ret = 0;
6292
6293         /* page is wholly or partially inside EOF */
6294         if (page_start + PAGE_CACHE_SIZE > size)
6295                 zero_start = size & ~PAGE_CACHE_MASK;
6296         else
6297                 zero_start = PAGE_CACHE_SIZE;
6298
6299         if (zero_start != PAGE_CACHE_SIZE) {
6300                 kaddr = kmap(page);
6301                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6302                 flush_dcache_page(page);
6303                 kunmap(page);
6304         }
6305         ClearPageChecked(page);
6306         set_page_dirty(page);
6307         SetPageUptodate(page);
6308
6309         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6310         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6311
6312         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6313
6314 out_unlock:
6315         if (!ret)
6316                 return VM_FAULT_LOCKED;
6317         unlock_page(page);
6318         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6319 out:
6320         return ret;
6321 }
6322
6323 static void btrfs_truncate(struct inode *inode)
6324 {
6325         struct btrfs_root *root = BTRFS_I(inode)->root;
6326         int ret;
6327         struct btrfs_trans_handle *trans;
6328         unsigned long nr;
6329         u64 mask = root->sectorsize - 1;
6330
6331         if (!S_ISREG(inode->i_mode)) {
6332                 WARN_ON(1);
6333                 return;
6334         }
6335
6336         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6337         if (ret)
6338                 return;
6339
6340         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6341         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6342
6343         trans = btrfs_start_transaction(root, 0);
6344         BUG_ON(IS_ERR(trans));
6345         btrfs_set_trans_block_group(trans, inode);
6346         trans->block_rsv = root->orphan_block_rsv;
6347
6348         /*
6349          * setattr is responsible for setting the ordered_data_close flag,
6350          * but that is only tested during the last file release.  That
6351          * could happen well after the next commit, leaving a great big
6352          * window where new writes may get lost if someone chooses to write
6353          * to this file after truncating to zero
6354          *
6355          * The inode doesn't have any dirty data here, and so if we commit
6356          * this is a noop.  If someone immediately starts writing to the inode
6357          * it is very likely we'll catch some of their writes in this
6358          * transaction, and the commit will find this file on the ordered
6359          * data list with good things to send down.
6360          *
6361          * This is a best effort solution, there is still a window where
6362          * using truncate to replace the contents of the file will
6363          * end up with a zero length file after a crash.
6364          */
6365         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6366                 btrfs_add_ordered_operation(trans, root, inode);
6367
6368         while (1) {
6369                 if (!trans) {
6370                         trans = btrfs_start_transaction(root, 0);
6371                         BUG_ON(IS_ERR(trans));
6372                         btrfs_set_trans_block_group(trans, inode);
6373                         trans->block_rsv = root->orphan_block_rsv;
6374                 }
6375
6376                 ret = btrfs_block_rsv_check(trans, root,
6377                                             root->orphan_block_rsv, 0, 5);
6378                 if (ret) {
6379                         BUG_ON(ret != -EAGAIN);
6380                         ret = btrfs_commit_transaction(trans, root);
6381                         BUG_ON(ret);
6382                         trans = NULL;
6383                         continue;
6384                 }
6385
6386                 ret = btrfs_truncate_inode_items(trans, root, inode,
6387                                                  inode->i_size,
6388                                                  BTRFS_EXTENT_DATA_KEY);
6389                 if (ret != -EAGAIN)
6390                         break;
6391
6392                 ret = btrfs_update_inode(trans, root, inode);
6393                 BUG_ON(ret);
6394
6395                 nr = trans->blocks_used;
6396                 btrfs_end_transaction(trans, root);
6397                 trans = NULL;
6398                 btrfs_btree_balance_dirty(root, nr);
6399         }
6400
6401         if (ret == 0 && inode->i_nlink > 0) {
6402                 ret = btrfs_orphan_del(trans, inode);
6403                 BUG_ON(ret);
6404         }
6405
6406         ret = btrfs_update_inode(trans, root, inode);
6407         BUG_ON(ret);
6408
6409         nr = trans->blocks_used;
6410         ret = btrfs_end_transaction_throttle(trans, root);
6411         BUG_ON(ret);
6412         btrfs_btree_balance_dirty(root, nr);
6413 }
6414
6415 /*
6416  * create a new subvolume directory/inode (helper for the ioctl).
6417  */
6418 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6419                              struct btrfs_root *new_root,
6420                              u64 new_dirid, u64 alloc_hint)
6421 {
6422         struct inode *inode;
6423         int err;
6424         u64 index = 0;
6425
6426         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6427                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6428         if (IS_ERR(inode))
6429                 return PTR_ERR(inode);
6430         inode->i_op = &btrfs_dir_inode_operations;
6431         inode->i_fop = &btrfs_dir_file_operations;
6432
6433         inode->i_nlink = 1;
6434         btrfs_i_size_write(inode, 0);
6435
6436         err = btrfs_update_inode(trans, new_root, inode);
6437         BUG_ON(err);
6438
6439         iput(inode);
6440         return 0;
6441 }
6442
6443 /* helper function for file defrag and space balancing.  This
6444  * forces readahead on a given range of bytes in an inode
6445  */
6446 unsigned long btrfs_force_ra(struct address_space *mapping,
6447                               struct file_ra_state *ra, struct file *file,
6448                               pgoff_t offset, pgoff_t last_index)
6449 {
6450         pgoff_t req_size = last_index - offset + 1;
6451
6452         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6453         return offset + req_size;
6454 }
6455
6456 struct inode *btrfs_alloc_inode(struct super_block *sb)
6457 {
6458         struct btrfs_inode *ei;
6459         struct inode *inode;
6460
6461         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6462         if (!ei)
6463                 return NULL;
6464
6465         ei->root = NULL;
6466         ei->space_info = NULL;
6467         ei->generation = 0;
6468         ei->sequence = 0;
6469         ei->last_trans = 0;
6470         ei->last_sub_trans = 0;
6471         ei->logged_trans = 0;
6472         ei->delalloc_bytes = 0;
6473         ei->reserved_bytes = 0;
6474         ei->disk_i_size = 0;
6475         ei->flags = 0;
6476         ei->index_cnt = (u64)-1;
6477         ei->last_unlink_trans = 0;
6478
6479         spin_lock_init(&ei->accounting_lock);
6480         atomic_set(&ei->outstanding_extents, 0);
6481         ei->reserved_extents = 0;
6482
6483         ei->ordered_data_close = 0;
6484         ei->orphan_meta_reserved = 0;
6485         ei->dummy_inode = 0;
6486         ei->force_compress = 0;
6487
6488         inode = &ei->vfs_inode;
6489         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6490         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6491         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6492         mutex_init(&ei->log_mutex);
6493         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6494         INIT_LIST_HEAD(&ei->i_orphan);
6495         INIT_LIST_HEAD(&ei->delalloc_inodes);
6496         INIT_LIST_HEAD(&ei->ordered_operations);
6497         RB_CLEAR_NODE(&ei->rb_node);
6498
6499         return inode;
6500 }
6501
6502 void btrfs_destroy_inode(struct inode *inode)
6503 {
6504         struct btrfs_ordered_extent *ordered;
6505         struct btrfs_root *root = BTRFS_I(inode)->root;
6506
6507         WARN_ON(!list_empty(&inode->i_dentry));
6508         WARN_ON(inode->i_data.nrpages);
6509         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6510         WARN_ON(BTRFS_I(inode)->reserved_extents);
6511
6512         /*
6513          * This can happen where we create an inode, but somebody else also
6514          * created the same inode and we need to destroy the one we already
6515          * created.
6516          */
6517         if (!root)
6518                 goto free;
6519
6520         /*
6521          * Make sure we're properly removed from the ordered operation
6522          * lists.
6523          */
6524         smp_mb();
6525         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6526                 spin_lock(&root->fs_info->ordered_extent_lock);
6527                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6528                 spin_unlock(&root->fs_info->ordered_extent_lock);
6529         }
6530
6531         if (root == root->fs_info->tree_root) {
6532                 struct btrfs_block_group_cache *block_group;
6533
6534                 block_group = btrfs_lookup_block_group(root->fs_info,
6535                                                 BTRFS_I(inode)->block_group);
6536                 if (block_group && block_group->inode == inode) {
6537                         spin_lock(&block_group->lock);
6538                         block_group->inode = NULL;
6539                         spin_unlock(&block_group->lock);
6540                         btrfs_put_block_group(block_group);
6541                 } else if (block_group) {
6542                         btrfs_put_block_group(block_group);
6543                 }
6544         }
6545
6546         spin_lock(&root->orphan_lock);
6547         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6548                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6549                        inode->i_ino);
6550                 list_del_init(&BTRFS_I(inode)->i_orphan);
6551         }
6552         spin_unlock(&root->orphan_lock);
6553
6554         while (1) {
6555                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6556                 if (!ordered)
6557                         break;
6558                 else {
6559                         printk(KERN_ERR "btrfs found ordered "
6560                                "extent %llu %llu on inode cleanup\n",
6561                                (unsigned long long)ordered->file_offset,
6562                                (unsigned long long)ordered->len);
6563                         btrfs_remove_ordered_extent(inode, ordered);
6564                         btrfs_put_ordered_extent(ordered);
6565                         btrfs_put_ordered_extent(ordered);
6566                 }
6567         }
6568         inode_tree_del(inode);
6569         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6570 free:
6571         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6572 }
6573
6574 int btrfs_drop_inode(struct inode *inode)
6575 {
6576         struct btrfs_root *root = BTRFS_I(inode)->root;
6577
6578         if (btrfs_root_refs(&root->root_item) == 0 &&
6579             root != root->fs_info->tree_root)
6580                 return 1;
6581         else
6582                 return generic_drop_inode(inode);
6583 }
6584
6585 static void init_once(void *foo)
6586 {
6587         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6588
6589         inode_init_once(&ei->vfs_inode);
6590 }
6591
6592 void btrfs_destroy_cachep(void)
6593 {
6594         if (btrfs_inode_cachep)
6595                 kmem_cache_destroy(btrfs_inode_cachep);
6596         if (btrfs_trans_handle_cachep)
6597                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6598         if (btrfs_transaction_cachep)
6599                 kmem_cache_destroy(btrfs_transaction_cachep);
6600         if (btrfs_path_cachep)
6601                 kmem_cache_destroy(btrfs_path_cachep);
6602 }
6603
6604 int btrfs_init_cachep(void)
6605 {
6606         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6607                         sizeof(struct btrfs_inode), 0,
6608                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6609         if (!btrfs_inode_cachep)
6610                 goto fail;
6611
6612         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6613                         sizeof(struct btrfs_trans_handle), 0,
6614                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6615         if (!btrfs_trans_handle_cachep)
6616                 goto fail;
6617
6618         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6619                         sizeof(struct btrfs_transaction), 0,
6620                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6621         if (!btrfs_transaction_cachep)
6622                 goto fail;
6623
6624         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6625                         sizeof(struct btrfs_path), 0,
6626                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6627         if (!btrfs_path_cachep)
6628                 goto fail;
6629
6630         return 0;
6631 fail:
6632         btrfs_destroy_cachep();
6633         return -ENOMEM;
6634 }
6635
6636 static int btrfs_getattr(struct vfsmount *mnt,
6637                          struct dentry *dentry, struct kstat *stat)
6638 {
6639         struct inode *inode = dentry->d_inode;
6640         generic_fillattr(inode, stat);
6641         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6642         stat->blksize = PAGE_CACHE_SIZE;
6643         stat->blocks = (inode_get_bytes(inode) +
6644                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6645         return 0;
6646 }
6647
6648 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6649                            struct inode *new_dir, struct dentry *new_dentry)
6650 {
6651         struct btrfs_trans_handle *trans;
6652         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6653         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6654         struct inode *new_inode = new_dentry->d_inode;
6655         struct inode *old_inode = old_dentry->d_inode;
6656         struct timespec ctime = CURRENT_TIME;
6657         u64 index = 0;
6658         u64 root_objectid;
6659         int ret;
6660
6661         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6662                 return -EPERM;
6663
6664         /* we only allow rename subvolume link between subvolumes */
6665         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6666                 return -EXDEV;
6667
6668         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6669             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6670                 return -ENOTEMPTY;
6671
6672         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6673             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6674                 return -ENOTEMPTY;
6675         /*
6676          * we're using rename to replace one file with another.
6677          * and the replacement file is large.  Start IO on it now so
6678          * we don't add too much work to the end of the transaction
6679          */
6680         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6681             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6682                 filemap_flush(old_inode->i_mapping);
6683
6684         /* close the racy window with snapshot create/destroy ioctl */
6685         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6686                 down_read(&root->fs_info->subvol_sem);
6687         /*
6688          * We want to reserve the absolute worst case amount of items.  So if
6689          * both inodes are subvols and we need to unlink them then that would
6690          * require 4 item modifications, but if they are both normal inodes it
6691          * would require 5 item modifications, so we'll assume their normal
6692          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6693          * should cover the worst case number of items we'll modify.
6694          */
6695         trans = btrfs_start_transaction(root, 20);
6696         if (IS_ERR(trans))
6697                 return PTR_ERR(trans);
6698
6699         btrfs_set_trans_block_group(trans, new_dir);
6700
6701         if (dest != root)
6702                 btrfs_record_root_in_trans(trans, dest);
6703
6704         ret = btrfs_set_inode_index(new_dir, &index);
6705         if (ret)
6706                 goto out_fail;
6707
6708         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6709                 /* force full log commit if subvolume involved. */
6710                 root->fs_info->last_trans_log_full_commit = trans->transid;
6711         } else {
6712                 ret = btrfs_insert_inode_ref(trans, dest,
6713                                              new_dentry->d_name.name,
6714                                              new_dentry->d_name.len,
6715                                              old_inode->i_ino,
6716                                              new_dir->i_ino, index);
6717                 if (ret)
6718                         goto out_fail;
6719                 /*
6720                  * this is an ugly little race, but the rename is required
6721                  * to make sure that if we crash, the inode is either at the
6722                  * old name or the new one.  pinning the log transaction lets
6723                  * us make sure we don't allow a log commit to come in after
6724                  * we unlink the name but before we add the new name back in.
6725                  */
6726                 btrfs_pin_log_trans(root);
6727         }
6728         /*
6729          * make sure the inode gets flushed if it is replacing
6730          * something.
6731          */
6732         if (new_inode && new_inode->i_size &&
6733             old_inode && S_ISREG(old_inode->i_mode)) {
6734                 btrfs_add_ordered_operation(trans, root, old_inode);
6735         }
6736
6737         old_dir->i_ctime = old_dir->i_mtime = ctime;
6738         new_dir->i_ctime = new_dir->i_mtime = ctime;
6739         old_inode->i_ctime = ctime;
6740
6741         if (old_dentry->d_parent != new_dentry->d_parent)
6742                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
6743
6744         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
6745                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
6746                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
6747                                         old_dentry->d_name.name,
6748                                         old_dentry->d_name.len);
6749         } else {
6750                 btrfs_inc_nlink(old_dentry->d_inode);
6751                 ret = btrfs_unlink_inode(trans, root, old_dir,
6752                                          old_dentry->d_inode,
6753                                          old_dentry->d_name.name,
6754                                          old_dentry->d_name.len);
6755         }
6756         BUG_ON(ret);
6757
6758         if (new_inode) {
6759                 new_inode->i_ctime = CURRENT_TIME;
6760                 if (unlikely(new_inode->i_ino ==
6761                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
6762                         root_objectid = BTRFS_I(new_inode)->location.objectid;
6763                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
6764                                                 root_objectid,
6765                                                 new_dentry->d_name.name,
6766                                                 new_dentry->d_name.len);
6767                         BUG_ON(new_inode->i_nlink == 0);
6768                 } else {
6769                         ret = btrfs_unlink_inode(trans, dest, new_dir,
6770                                                  new_dentry->d_inode,
6771                                                  new_dentry->d_name.name,
6772                                                  new_dentry->d_name.len);
6773                 }
6774                 BUG_ON(ret);
6775                 if (new_inode->i_nlink == 0) {
6776                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
6777                         BUG_ON(ret);
6778                 }
6779         }
6780
6781         ret = btrfs_add_link(trans, new_dir, old_inode,
6782                              new_dentry->d_name.name,
6783                              new_dentry->d_name.len, 0, index);
6784         BUG_ON(ret);
6785
6786         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
6787                 struct dentry *parent = dget_parent(new_dentry);
6788                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
6789                 dput(parent);
6790                 btrfs_end_log_trans(root);
6791         }
6792 out_fail:
6793         btrfs_end_transaction_throttle(trans, root);
6794
6795         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
6796                 up_read(&root->fs_info->subvol_sem);
6797
6798         return ret;
6799 }
6800
6801 /*
6802  * some fairly slow code that needs optimization. This walks the list
6803  * of all the inodes with pending delalloc and forces them to disk.
6804  */
6805 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
6806 {
6807         struct list_head *head = &root->fs_info->delalloc_inodes;
6808         struct btrfs_inode *binode;
6809         struct inode *inode;
6810
6811         if (root->fs_info->sb->s_flags & MS_RDONLY)
6812                 return -EROFS;
6813
6814         spin_lock(&root->fs_info->delalloc_lock);
6815         while (!list_empty(head)) {
6816                 binode = list_entry(head->next, struct btrfs_inode,
6817                                     delalloc_inodes);
6818                 inode = igrab(&binode->vfs_inode);
6819                 if (!inode)
6820                         list_del_init(&binode->delalloc_inodes);
6821                 spin_unlock(&root->fs_info->delalloc_lock);
6822                 if (inode) {
6823                         filemap_flush(inode->i_mapping);
6824                         if (delay_iput)
6825                                 btrfs_add_delayed_iput(inode);
6826                         else
6827                                 iput(inode);
6828                 }
6829                 cond_resched();
6830                 spin_lock(&root->fs_info->delalloc_lock);
6831         }
6832         spin_unlock(&root->fs_info->delalloc_lock);
6833
6834         /* the filemap_flush will queue IO into the worker threads, but
6835          * we have to make sure the IO is actually started and that
6836          * ordered extents get created before we return
6837          */
6838         atomic_inc(&root->fs_info->async_submit_draining);
6839         while (atomic_read(&root->fs_info->nr_async_submits) ||
6840               atomic_read(&root->fs_info->async_delalloc_pages)) {
6841                 wait_event(root->fs_info->async_submit_wait,
6842                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
6843                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
6844         }
6845         atomic_dec(&root->fs_info->async_submit_draining);
6846         return 0;
6847 }
6848
6849 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
6850                                    int sync)
6851 {
6852         struct btrfs_inode *binode;
6853         struct inode *inode = NULL;
6854
6855         spin_lock(&root->fs_info->delalloc_lock);
6856         while (!list_empty(&root->fs_info->delalloc_inodes)) {
6857                 binode = list_entry(root->fs_info->delalloc_inodes.next,
6858                                     struct btrfs_inode, delalloc_inodes);
6859                 inode = igrab(&binode->vfs_inode);
6860                 if (inode) {
6861                         list_move_tail(&binode->delalloc_inodes,
6862                                        &root->fs_info->delalloc_inodes);
6863                         break;
6864                 }
6865
6866                 list_del_init(&binode->delalloc_inodes);
6867                 cond_resched_lock(&root->fs_info->delalloc_lock);
6868         }
6869         spin_unlock(&root->fs_info->delalloc_lock);
6870
6871         if (inode) {
6872                 if (sync) {
6873                         filemap_write_and_wait(inode->i_mapping);
6874                         /*
6875                          * We have to do this because compression doesn't
6876                          * actually set PG_writeback until it submits the pages
6877                          * for IO, which happens in an async thread, so we could
6878                          * race and not actually wait for any writeback pages
6879                          * because they've not been submitted yet.  Technically
6880                          * this could still be the case for the ordered stuff
6881                          * since the async thread may not have started to do its
6882                          * work yet.  If this becomes the case then we need to
6883                          * figure out a way to make sure that in writepage we
6884                          * wait for any async pages to be submitted before
6885                          * returning so that fdatawait does what its supposed to
6886                          * do.
6887                          */
6888                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
6889                 } else {
6890                         filemap_flush(inode->i_mapping);
6891                 }
6892                 if (delay_iput)
6893                         btrfs_add_delayed_iput(inode);
6894                 else
6895                         iput(inode);
6896                 return 1;
6897         }
6898         return 0;
6899 }
6900
6901 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
6902                          const char *symname)
6903 {
6904         struct btrfs_trans_handle *trans;
6905         struct btrfs_root *root = BTRFS_I(dir)->root;
6906         struct btrfs_path *path;
6907         struct btrfs_key key;
6908         struct inode *inode = NULL;
6909         int err;
6910         int drop_inode = 0;
6911         u64 objectid;
6912         u64 index = 0 ;
6913         int name_len;
6914         int datasize;
6915         unsigned long ptr;
6916         struct btrfs_file_extent_item *ei;
6917         struct extent_buffer *leaf;
6918         unsigned long nr = 0;
6919
6920         name_len = strlen(symname) + 1;
6921         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
6922                 return -ENAMETOOLONG;
6923
6924         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
6925         if (err)
6926                 return err;
6927         /*
6928          * 2 items for inode item and ref
6929          * 2 items for dir items
6930          * 1 item for xattr if selinux is on
6931          */
6932         trans = btrfs_start_transaction(root, 5);
6933         if (IS_ERR(trans))
6934                 return PTR_ERR(trans);
6935
6936         btrfs_set_trans_block_group(trans, dir);
6937
6938         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
6939                                 dentry->d_name.len, dir->i_ino, objectid,
6940                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
6941                                 &index);
6942         err = PTR_ERR(inode);
6943         if (IS_ERR(inode))
6944                 goto out_unlock;
6945
6946         err = btrfs_init_inode_security(trans, inode, dir);
6947         if (err) {
6948                 drop_inode = 1;
6949                 goto out_unlock;
6950         }
6951
6952         btrfs_set_trans_block_group(trans, inode);
6953         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
6954         if (err)
6955                 drop_inode = 1;
6956         else {
6957                 inode->i_mapping->a_ops = &btrfs_aops;
6958                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6959                 inode->i_fop = &btrfs_file_operations;
6960                 inode->i_op = &btrfs_file_inode_operations;
6961                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
6962         }
6963         btrfs_update_inode_block_group(trans, inode);
6964         btrfs_update_inode_block_group(trans, dir);
6965         if (drop_inode)
6966                 goto out_unlock;
6967
6968         path = btrfs_alloc_path();
6969         BUG_ON(!path);
6970         key.objectid = inode->i_ino;
6971         key.offset = 0;
6972         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
6973         datasize = btrfs_file_extent_calc_inline_size(name_len);
6974         err = btrfs_insert_empty_item(trans, root, path, &key,
6975                                       datasize);
6976         if (err) {
6977                 drop_inode = 1;
6978                 goto out_unlock;
6979         }
6980         leaf = path->nodes[0];
6981         ei = btrfs_item_ptr(leaf, path->slots[0],
6982                             struct btrfs_file_extent_item);
6983         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
6984         btrfs_set_file_extent_type(leaf, ei,
6985                                    BTRFS_FILE_EXTENT_INLINE);
6986         btrfs_set_file_extent_encryption(leaf, ei, 0);
6987         btrfs_set_file_extent_compression(leaf, ei, 0);
6988         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
6989         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
6990
6991         ptr = btrfs_file_extent_inline_start(ei);
6992         write_extent_buffer(leaf, symname, ptr, name_len);
6993         btrfs_mark_buffer_dirty(leaf);
6994         btrfs_free_path(path);
6995
6996         inode->i_op = &btrfs_symlink_inode_operations;
6997         inode->i_mapping->a_ops = &btrfs_symlink_aops;
6998         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
6999         inode_set_bytes(inode, name_len);
7000         btrfs_i_size_write(inode, name_len - 1);
7001         err = btrfs_update_inode(trans, root, inode);
7002         if (err)
7003                 drop_inode = 1;
7004
7005 out_unlock:
7006         nr = trans->blocks_used;
7007         btrfs_end_transaction_throttle(trans, root);
7008         if (drop_inode) {
7009                 inode_dec_link_count(inode);
7010                 iput(inode);
7011         }
7012         btrfs_btree_balance_dirty(root, nr);
7013         return err;
7014 }
7015
7016 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7017                                        u64 start, u64 num_bytes, u64 min_size,
7018                                        loff_t actual_len, u64 *alloc_hint,
7019                                        struct btrfs_trans_handle *trans)
7020 {
7021         struct btrfs_root *root = BTRFS_I(inode)->root;
7022         struct btrfs_key ins;
7023         u64 cur_offset = start;
7024         u64 i_size;
7025         int ret = 0;
7026         bool own_trans = true;
7027
7028         if (trans)
7029                 own_trans = false;
7030         while (num_bytes > 0) {
7031                 if (own_trans) {
7032                         trans = btrfs_start_transaction(root, 3);
7033                         if (IS_ERR(trans)) {
7034                                 ret = PTR_ERR(trans);
7035                                 break;
7036                         }
7037                 }
7038
7039                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7040                                            0, *alloc_hint, (u64)-1, &ins, 1);
7041                 if (ret) {
7042                         if (own_trans)
7043                                 btrfs_end_transaction(trans, root);
7044                         break;
7045                 }
7046
7047                 ret = insert_reserved_file_extent(trans, inode,
7048                                                   cur_offset, ins.objectid,
7049                                                   ins.offset, ins.offset,
7050                                                   ins.offset, 0, 0, 0,
7051                                                   BTRFS_FILE_EXTENT_PREALLOC);
7052                 BUG_ON(ret);
7053                 btrfs_drop_extent_cache(inode, cur_offset,
7054                                         cur_offset + ins.offset -1, 0);
7055
7056                 num_bytes -= ins.offset;
7057                 cur_offset += ins.offset;
7058                 *alloc_hint = ins.objectid + ins.offset;
7059
7060                 inode->i_ctime = CURRENT_TIME;
7061                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7062                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7063                     (actual_len > inode->i_size) &&
7064                     (cur_offset > inode->i_size)) {
7065                         if (cur_offset > actual_len)
7066                                 i_size = actual_len;
7067                         else
7068                                 i_size = cur_offset;
7069                         i_size_write(inode, i_size);
7070                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7071                 }
7072
7073                 ret = btrfs_update_inode(trans, root, inode);
7074                 BUG_ON(ret);
7075
7076                 if (own_trans)
7077                         btrfs_end_transaction(trans, root);
7078         }
7079         return ret;
7080 }
7081
7082 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7083                               u64 start, u64 num_bytes, u64 min_size,
7084                               loff_t actual_len, u64 *alloc_hint)
7085 {
7086         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7087                                            min_size, actual_len, alloc_hint,
7088                                            NULL);
7089 }
7090
7091 int btrfs_prealloc_file_range_trans(struct inode *inode,
7092                                     struct btrfs_trans_handle *trans, int mode,
7093                                     u64 start, u64 num_bytes, u64 min_size,
7094                                     loff_t actual_len, u64 *alloc_hint)
7095 {
7096         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7097                                            min_size, actual_len, alloc_hint, trans);
7098 }
7099
7100 static long btrfs_fallocate(struct inode *inode, int mode,
7101                             loff_t offset, loff_t len)
7102 {
7103         struct extent_state *cached_state = NULL;
7104         u64 cur_offset;
7105         u64 last_byte;
7106         u64 alloc_start;
7107         u64 alloc_end;
7108         u64 alloc_hint = 0;
7109         u64 locked_end;
7110         u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
7111         struct extent_map *em;
7112         int ret;
7113
7114         alloc_start = offset & ~mask;
7115         alloc_end =  (offset + len + mask) & ~mask;
7116
7117         /*
7118          * wait for ordered IO before we have any locks.  We'll loop again
7119          * below with the locks held.
7120          */
7121         btrfs_wait_ordered_range(inode, alloc_start, alloc_end - alloc_start);
7122
7123         mutex_lock(&inode->i_mutex);
7124         ret = inode_newsize_ok(inode, alloc_end);
7125         if (ret)
7126                 goto out;
7127
7128         if (alloc_start > inode->i_size) {
7129                 ret = btrfs_cont_expand(inode, alloc_start);
7130                 if (ret)
7131                         goto out;
7132         }
7133
7134         ret = btrfs_check_data_free_space(inode, alloc_end - alloc_start);
7135         if (ret)
7136                 goto out;
7137
7138         locked_end = alloc_end - 1;
7139         while (1) {
7140                 struct btrfs_ordered_extent *ordered;
7141
7142                 /* the extent lock is ordered inside the running
7143                  * transaction
7144                  */
7145                 lock_extent_bits(&BTRFS_I(inode)->io_tree, alloc_start,
7146                                  locked_end, 0, &cached_state, GFP_NOFS);
7147                 ordered = btrfs_lookup_first_ordered_extent(inode,
7148                                                             alloc_end - 1);
7149                 if (ordered &&
7150                     ordered->file_offset + ordered->len > alloc_start &&
7151                     ordered->file_offset < alloc_end) {
7152                         btrfs_put_ordered_extent(ordered);
7153                         unlock_extent_cached(&BTRFS_I(inode)->io_tree,
7154                                              alloc_start, locked_end,
7155                                              &cached_state, GFP_NOFS);
7156                         /*
7157                          * we can't wait on the range with the transaction
7158                          * running or with the extent lock held
7159                          */
7160                         btrfs_wait_ordered_range(inode, alloc_start,
7161                                                  alloc_end - alloc_start);
7162                 } else {
7163                         if (ordered)
7164                                 btrfs_put_ordered_extent(ordered);
7165                         break;
7166                 }
7167         }
7168
7169         cur_offset = alloc_start;
7170         while (1) {
7171                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
7172                                       alloc_end - cur_offset, 0);
7173                 BUG_ON(IS_ERR(em) || !em);
7174                 last_byte = min(extent_map_end(em), alloc_end);
7175                 last_byte = (last_byte + mask) & ~mask;
7176                 if (em->block_start == EXTENT_MAP_HOLE ||
7177                     (cur_offset >= inode->i_size &&
7178                      !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
7179                         ret = btrfs_prealloc_file_range(inode, mode, cur_offset,
7180                                                         last_byte - cur_offset,
7181                                                         1 << inode->i_blkbits,
7182                                                         offset + len,
7183                                                         &alloc_hint);
7184                         if (ret < 0) {
7185                                 free_extent_map(em);
7186                                 break;
7187                         }
7188                 }
7189                 free_extent_map(em);
7190
7191                 cur_offset = last_byte;
7192                 if (cur_offset >= alloc_end) {
7193                         ret = 0;
7194                         break;
7195                 }
7196         }
7197         unlock_extent_cached(&BTRFS_I(inode)->io_tree, alloc_start, locked_end,
7198                              &cached_state, GFP_NOFS);
7199
7200         btrfs_free_reserved_data_space(inode, alloc_end - alloc_start);
7201 out:
7202         mutex_unlock(&inode->i_mutex);
7203         return ret;
7204 }
7205
7206 static int btrfs_set_page_dirty(struct page *page)
7207 {
7208         return __set_page_dirty_nobuffers(page);
7209 }
7210
7211 static int btrfs_permission(struct inode *inode, int mask)
7212 {
7213         struct btrfs_root *root = BTRFS_I(inode)->root;
7214
7215         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7216                 return -EROFS;
7217         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7218                 return -EACCES;
7219         return generic_permission(inode, mask, btrfs_check_acl);
7220 }
7221
7222 static const struct inode_operations btrfs_dir_inode_operations = {
7223         .getattr        = btrfs_getattr,
7224         .lookup         = btrfs_lookup,
7225         .create         = btrfs_create,
7226         .unlink         = btrfs_unlink,
7227         .link           = btrfs_link,
7228         .mkdir          = btrfs_mkdir,
7229         .rmdir          = btrfs_rmdir,
7230         .rename         = btrfs_rename,
7231         .symlink        = btrfs_symlink,
7232         .setattr        = btrfs_setattr,
7233         .mknod          = btrfs_mknod,
7234         .setxattr       = btrfs_setxattr,
7235         .getxattr       = btrfs_getxattr,
7236         .listxattr      = btrfs_listxattr,
7237         .removexattr    = btrfs_removexattr,
7238         .permission     = btrfs_permission,
7239 };
7240 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7241         .lookup         = btrfs_lookup,
7242         .permission     = btrfs_permission,
7243 };
7244
7245 static const struct file_operations btrfs_dir_file_operations = {
7246         .llseek         = generic_file_llseek,
7247         .read           = generic_read_dir,
7248         .readdir        = btrfs_real_readdir,
7249         .unlocked_ioctl = btrfs_ioctl,
7250 #ifdef CONFIG_COMPAT
7251         .compat_ioctl   = btrfs_ioctl,
7252 #endif
7253         .release        = btrfs_release_file,
7254         .fsync          = btrfs_sync_file,
7255 };
7256
7257 static struct extent_io_ops btrfs_extent_io_ops = {
7258         .fill_delalloc = run_delalloc_range,
7259         .submit_bio_hook = btrfs_submit_bio_hook,
7260         .merge_bio_hook = btrfs_merge_bio_hook,
7261         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7262         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7263         .writepage_start_hook = btrfs_writepage_start_hook,
7264         .readpage_io_failed_hook = btrfs_io_failed_hook,
7265         .set_bit_hook = btrfs_set_bit_hook,
7266         .clear_bit_hook = btrfs_clear_bit_hook,
7267         .merge_extent_hook = btrfs_merge_extent_hook,
7268         .split_extent_hook = btrfs_split_extent_hook,
7269 };
7270
7271 /*
7272  * btrfs doesn't support the bmap operation because swapfiles
7273  * use bmap to make a mapping of extents in the file.  They assume
7274  * these extents won't change over the life of the file and they
7275  * use the bmap result to do IO directly to the drive.
7276  *
7277  * the btrfs bmap call would return logical addresses that aren't
7278  * suitable for IO and they also will change frequently as COW
7279  * operations happen.  So, swapfile + btrfs == corruption.
7280  *
7281  * For now we're avoiding this by dropping bmap.
7282  */
7283 static const struct address_space_operations btrfs_aops = {
7284         .readpage       = btrfs_readpage,
7285         .writepage      = btrfs_writepage,
7286         .writepages     = btrfs_writepages,
7287         .readpages      = btrfs_readpages,
7288         .sync_page      = block_sync_page,
7289         .direct_IO      = btrfs_direct_IO,
7290         .invalidatepage = btrfs_invalidatepage,
7291         .releasepage    = btrfs_releasepage,
7292         .set_page_dirty = btrfs_set_page_dirty,
7293         .error_remove_page = generic_error_remove_page,
7294 };
7295
7296 static const struct address_space_operations btrfs_symlink_aops = {
7297         .readpage       = btrfs_readpage,
7298         .writepage      = btrfs_writepage,
7299         .invalidatepage = btrfs_invalidatepage,
7300         .releasepage    = btrfs_releasepage,
7301 };
7302
7303 static const struct inode_operations btrfs_file_inode_operations = {
7304         .truncate       = btrfs_truncate,
7305         .getattr        = btrfs_getattr,
7306         .setattr        = btrfs_setattr,
7307         .setxattr       = btrfs_setxattr,
7308         .getxattr       = btrfs_getxattr,
7309         .listxattr      = btrfs_listxattr,
7310         .removexattr    = btrfs_removexattr,
7311         .permission     = btrfs_permission,
7312         .fallocate      = btrfs_fallocate,
7313         .fiemap         = btrfs_fiemap,
7314 };
7315 static const struct inode_operations btrfs_special_inode_operations = {
7316         .getattr        = btrfs_getattr,
7317         .setattr        = btrfs_setattr,
7318         .permission     = btrfs_permission,
7319         .setxattr       = btrfs_setxattr,
7320         .getxattr       = btrfs_getxattr,
7321         .listxattr      = btrfs_listxattr,
7322         .removexattr    = btrfs_removexattr,
7323 };
7324 static const struct inode_operations btrfs_symlink_inode_operations = {
7325         .readlink       = generic_readlink,
7326         .follow_link    = page_follow_link_light,
7327         .put_link       = page_put_link,
7328         .getattr        = btrfs_getattr,
7329         .permission     = btrfs_permission,
7330         .setxattr       = btrfs_setxattr,
7331         .getxattr       = btrfs_getxattr,
7332         .listxattr      = btrfs_listxattr,
7333         .removexattr    = btrfs_removexattr,
7334 };
7335
7336 const struct dentry_operations btrfs_dentry_operations = {
7337         .d_delete       = btrfs_dentry_delete,
7338 };