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 #include "free-space-cache.h"
  54
  55 struct btrfs_iget_args {
  56         u64 ino;
  57         struct btrfs_root *root;
  58 };
  59
  60 static const struct inode_operations btrfs_dir_inode_operations;
  61 static const struct inode_operations btrfs_symlink_inode_operations;
  62 static const struct inode_operations btrfs_dir_ro_inode_operations;
  63 static const struct inode_operations btrfs_special_inode_operations;
  64 static const struct inode_operations btrfs_file_inode_operations;
  65 static const struct address_space_operations btrfs_aops;
  66 static const struct address_space_operations btrfs_symlink_aops;
  67 static const struct file_operations btrfs_dir_file_operations;
  68 static struct extent_io_ops btrfs_extent_io_ops;
  69
  70 static struct kmem_cache *btrfs_inode_cachep;
  71 struct kmem_cache *btrfs_trans_handle_cachep;
  72 struct kmem_cache *btrfs_transaction_cachep;
  73 struct kmem_cache *btrfs_path_cachep;
  74 struct kmem_cache *btrfs_free_space_cachep;
  75
  76 #define S_SHIFT 12
  77 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  78         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  79         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  80         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  81         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  82         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  83         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  84         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  85 };
  86
  87 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  88 static int btrfs_truncate(struct inode *inode);
  89 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  90 static noinline int cow_file_range(struct inode *inode,
  91                                    struct page *locked_page,
  92                                    u64 start, u64 end, int *page_started,
  93                                    unsigned long *nr_written, int unlock);
  94
  95 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  96                                      struct inode *inode,  struct inode *dir)
  97 {
  98         int err;
  99
 100         err = btrfs_init_acl(trans, inode, dir);
 101         if (!err)
 102                 err = btrfs_xattr_security_init(trans, inode, dir);
 103         return err;
 104 }
 105
 106 /*
 107  * this does all the hard work for inserting an inline extent into
 108  * the btree.  The caller should have done a btrfs_drop_extents so that
 109  * no overlapping inline items exist in the btree
 110  */
 111 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 112                                 struct btrfs_root *root, struct inode *inode,
 113                                 u64 start, size_t size, size_t compressed_size,
 114                                 int compress_type,
 115                                 struct page **compressed_pages)
 116 {
 117         struct btrfs_key key;
 118         struct btrfs_path *path;
 119         struct extent_buffer *leaf;
 120         struct page *page = NULL;
 121         char *kaddr;
 122         unsigned long ptr;
 123         struct btrfs_file_extent_item *ei;
 124         int err = 0;
 125         int ret;
 126         size_t cur_size = size;
 127         size_t datasize;
 128         unsigned long offset;
 129
 130         if (compressed_size && compressed_pages)
 131                 cur_size = compressed_size;
 132
 133         path = btrfs_alloc_path();
 134         if (!path)
 135                 return -ENOMEM;
 136
 137         path->leave_spinning = 1;
 138         btrfs_set_trans_block_group(trans, inode);
 139
 140         key.objectid = inode->i_ino;
 141         key.offset = start;
 142         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 143         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 144
 145         inode_add_bytes(inode, size);
 146         ret = btrfs_insert_empty_item(trans, root, path, &key,
 147                                       datasize);
 148         BUG_ON(ret);
 149         if (ret) {
 150                 err = ret;
 151                 goto fail;
 152         }
 153         leaf = path->nodes[0];
 154         ei = btrfs_item_ptr(leaf, path->slots[0],
 155                             struct btrfs_file_extent_item);
 156         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 157         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 158         btrfs_set_file_extent_encryption(leaf, ei, 0);
 159         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 160         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 161         ptr = btrfs_file_extent_inline_start(ei);
 162
 163         if (compress_type != BTRFS_COMPRESS_NONE) {
 164                 struct page *cpage;
 165                 int i = 0;
 166                 while (compressed_size > 0) {
 167                         cpage = compressed_pages[i];
 168                         cur_size = min_t(unsigned long, compressed_size,
 169                                        PAGE_CACHE_SIZE);
 170
 171                         kaddr = kmap_atomic(cpage, KM_USER0);
 172                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 173                         kunmap_atomic(kaddr, KM_USER0);
 174
 175                         i++;
 176                         ptr += cur_size;
 177                         compressed_size -= cur_size;
 178                 }
 179                 btrfs_set_file_extent_compression(leaf, ei,
 180                                                   compress_type);
 181         } else {
 182                 page = find_get_page(inode->i_mapping,
 183                                      start >> PAGE_CACHE_SHIFT);
 184                 btrfs_set_file_extent_compression(leaf, ei, 0);
 185                 kaddr = kmap_atomic(page, KM_USER0);
 186                 offset = start & (PAGE_CACHE_SIZE - 1);
 187                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 188                 kunmap_atomic(kaddr, KM_USER0);
 189                 page_cache_release(page);
 190         }
 191         btrfs_mark_buffer_dirty(leaf);
 192         btrfs_free_path(path);
 193
 194         /*
 195          * we're an inline extent, so nobody can
 196          * extend the file past i_size without locking
 197          * a page we already have locked.
 198          *
 199          * We must do any isize and inode updates
 200          * before we unlock the pages.  Otherwise we
 201          * could end up racing with unlink.
 202          */
 203         BTRFS_I(inode)->disk_i_size = inode->i_size;
 204         btrfs_update_inode(trans, root, inode);
 205
 206         return 0;
 207 fail:
 208         btrfs_free_path(path);
 209         return err;
 210 }
 211
 212
 213 /*
 214  * conditionally insert an inline extent into the file.  This
 215  * does the checks required to make sure the data is small enough
 216  * to fit as an inline extent.
 217  */
 218 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 219                                  struct btrfs_root *root,
 220                                  struct inode *inode, u64 start, u64 end,
 221                                  size_t compressed_size, int compress_type,
 222                                  struct page **compressed_pages)
 223 {
 224         u64 isize = i_size_read(inode);
 225         u64 actual_end = min(end + 1, isize);
 226         u64 inline_len = actual_end - start;
 227         u64 aligned_end = (end + root->sectorsize - 1) &
 228                         ~((u64)root->sectorsize - 1);
 229         u64 hint_byte;
 230         u64 data_len = inline_len;
 231         int ret;
 232
 233         if (compressed_size)
 234                 data_len = compressed_size;
 235
 236         if (start > 0 ||
 237             actual_end >= PAGE_CACHE_SIZE ||
 238             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 239             (!compressed_size &&
 240             (actual_end & (root->sectorsize - 1)) == 0) ||
 241             end + 1 < isize ||
 242             data_len > root->fs_info->max_inline) {
 243                 return 1;
 244         }
 245
 246         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 247                                  &hint_byte, 1);
 248         BUG_ON(ret);
 249
 250         if (isize > actual_end)
 251                 inline_len = min_t(u64, isize, actual_end);
 252         ret = insert_inline_extent(trans, root, inode, start,
 253                                    inline_len, compressed_size,
 254                                    compress_type, compressed_pages);
 255         BUG_ON(ret);
 256         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 257         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 258         return 0;
 259 }
 260
 261 struct async_extent {
 262         u64 start;
 263         u64 ram_size;
 264         u64 compressed_size;
 265         struct page **pages;
 266         unsigned long nr_pages;
 267         int compress_type;
 268         struct list_head list;
 269 };
 270
 271 struct async_cow {
 272         struct inode *inode;
 273         struct btrfs_root *root;
 274         struct page *locked_page;
 275         u64 start;
 276         u64 end;
 277         struct list_head extents;
 278         struct btrfs_work work;
 279 };
 280
 281 static noinline int add_async_extent(struct async_cow *cow,
 282                                      u64 start, u64 ram_size,
 283                                      u64 compressed_size,
 284                                      struct page **pages,
 285                                      unsigned long nr_pages,
 286                                      int compress_type)
 287 {
 288         struct async_extent *async_extent;
 289
 290         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 291         BUG_ON(!async_extent);
 292         async_extent->start = start;
 293         async_extent->ram_size = ram_size;
 294         async_extent->compressed_size = compressed_size;
 295         async_extent->pages = pages;
 296         async_extent->nr_pages = nr_pages;
 297         async_extent->compress_type = compress_type;
 298         list_add_tail(&async_extent->list, &cow->extents);
 299         return 0;
 300 }
 301
 302 /*
 303  * we create compressed extents in two phases.  The first
 304  * phase compresses a range of pages that have already been
 305  * locked (both pages and state bits are locked).
 306  *
 307  * This is done inside an ordered work queue, and the compression
 308  * is spread across many cpus.  The actual IO submission is step
 309  * two, and the ordered work queue takes care of making sure that
 310  * happens in the same order things were put onto the queue by
 311  * writepages and friends.
 312  *
 313  * If this code finds it can't get good compression, it puts an
 314  * entry onto the work queue to write the uncompressed bytes.  This
 315  * makes sure that both compressed inodes and uncompressed inodes
 316  * are written in the same order that pdflush sent them down.
 317  */
 318 static noinline int compress_file_range(struct inode *inode,
 319                                         struct page *locked_page,
 320                                         u64 start, u64 end,
 321                                         struct async_cow *async_cow,
 322                                         int *num_added)
 323 {
 324         struct btrfs_root *root = BTRFS_I(inode)->root;
 325         struct btrfs_trans_handle *trans;
 326         u64 num_bytes;
 327         u64 blocksize = root->sectorsize;
 328         u64 actual_end;
 329         u64 isize = i_size_read(inode);
 330         int ret = 0;
 331         struct page **pages = NULL;
 332         unsigned long nr_pages;
 333         unsigned long nr_pages_ret = 0;
 334         unsigned long total_compressed = 0;
 335         unsigned long total_in = 0;
 336         unsigned long max_compressed = 128 * 1024;
 337         unsigned long max_uncompressed = 128 * 1024;
 338         int i;
 339         int will_compress;
 340         int compress_type = root->fs_info->compress_type;
 341
 342         actual_end = min_t(u64, isize, end + 1);
 343 again:
 344         will_compress = 0;
 345         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 346         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 347
 348         /*
 349          * we don't want to send crud past the end of i_size through
 350          * compression, that's just a waste of CPU time.  So, if the
 351          * end of the file is before the start of our current
 352          * requested range of bytes, we bail out to the uncompressed
 353          * cleanup code that can deal with all of this.
 354          *
 355          * It isn't really the fastest way to fix things, but this is a
 356          * very uncommon corner.
 357          */
 358         if (actual_end <= start)
 359                 goto cleanup_and_bail_uncompressed;
 360
 361         total_compressed = actual_end - start;
 362
 363         /* we want to make sure that amount of ram required to uncompress
 364          * an extent is reasonable, so we limit the total size in ram
 365          * of a compressed extent to 128k.  This is a crucial number
 366          * because it also controls how easily we can spread reads across
 367          * cpus for decompression.
 368          *
 369          * We also want to make sure the amount of IO required to do
 370          * a random read is reasonably small, so we limit the size of
 371          * a compressed extent to 128k.
 372          */
 373         total_compressed = min(total_compressed, max_uncompressed);
 374         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 375         num_bytes = max(blocksize,  num_bytes);
 376         total_in = 0;
 377         ret = 0;
 378
 379         /*
 380          * we do compression for mount -o compress and when the
 381          * inode has not been flagged as nocompress.  This flag can
 382          * change at any time if we discover bad compression ratios.
 383          */
 384         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 385             (btrfs_test_opt(root, COMPRESS) ||
 386              (BTRFS_I(inode)->force_compress) ||
 387              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 388                 WARN_ON(pages);
 389                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 390                 BUG_ON(!pages);
 391
 392                 if (BTRFS_I(inode)->force_compress)
 393                         compress_type = BTRFS_I(inode)->force_compress;
 394
 395                 ret = btrfs_compress_pages(compress_type,
 396                                            inode->i_mapping, start,
 397                                            total_compressed, pages,
 398                                            nr_pages, &nr_pages_ret,
 399                                            &total_in,
 400                                            &total_compressed,
 401                                            max_compressed);
 402
 403                 if (!ret) {
 404                         unsigned long offset = total_compressed &
 405                                 (PAGE_CACHE_SIZE - 1);
 406                         struct page *page = pages[nr_pages_ret - 1];
 407                         char *kaddr;
 408
 409                         /* zero the tail end of the last page, we might be
 410                          * sending it down to disk
 411                          */
 412                         if (offset) {
 413                                 kaddr = kmap_atomic(page, KM_USER0);
 414                                 memset(kaddr + offset, 0,
 415                                        PAGE_CACHE_SIZE - offset);
 416                                 kunmap_atomic(kaddr, KM_USER0);
 417                         }
 418                         will_compress = 1;
 419                 }
 420         }
 421         if (start == 0) {
 422                 trans = btrfs_join_transaction(root, 1);
 423                 BUG_ON(IS_ERR(trans));
 424                 btrfs_set_trans_block_group(trans, inode);
 425                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 426
 427                 /* lets try to make an inline extent */
 428                 if (ret || total_in < (actual_end - start)) {
 429                         /* we didn't compress the entire range, try
 430                          * to make an uncompressed inline extent.
 431                          */
 432                         ret = cow_file_range_inline(trans, root, inode,
 433                                                     start, end, 0, 0, NULL);
 434                 } else {
 435                         /* try making a compressed inline extent */
 436                         ret = cow_file_range_inline(trans, root, inode,
 437                                                     start, end,
 438                                                     total_compressed,
 439                                                     compress_type, pages);
 440                 }
 441                 if (ret == 0) {
 442                         /*
 443                          * inline extent creation worked, we don't need
 444                          * to create any more async work items.  Unlock
 445                          * and free up our temp pages.
 446                          */
 447                         extent_clear_unlock_delalloc(inode,
 448                              &BTRFS_I(inode)->io_tree,
 449                              start, end, NULL,
 450                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 451                              EXTENT_CLEAR_DELALLOC |
 452                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 453
 454                         btrfs_end_transaction(trans, root);
 455                         goto free_pages_out;
 456                 }
 457                 btrfs_end_transaction(trans, root);
 458         }
 459
 460         if (will_compress) {
 461                 /*
 462                  * we aren't doing an inline extent round the compressed size
 463                  * up to a block size boundary so the allocator does sane
 464                  * things
 465                  */
 466                 total_compressed = (total_compressed + blocksize - 1) &
 467                         ~(blocksize - 1);
 468
 469                 /*
 470                  * one last check to make sure the compression is really a
 471                  * win, compare the page count read with the blocks on disk
 472                  */
 473                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 474                         ~(PAGE_CACHE_SIZE - 1);
 475                 if (total_compressed >= total_in) {
 476                         will_compress = 0;
 477                 } else {
 478                         num_bytes = total_in;
 479                 }
 480         }
 481         if (!will_compress && pages) {
 482                 /*
 483                  * the compression code ran but failed to make things smaller,
 484                  * free any pages it allocated and our page pointer array
 485                  */
 486                 for (i = 0; i < nr_pages_ret; i++) {
 487                         WARN_ON(pages[i]->mapping);
 488                         page_cache_release(pages[i]);
 489                 }
 490                 kfree(pages);
 491                 pages = NULL;
 492                 total_compressed = 0;
 493                 nr_pages_ret = 0;
 494
 495                 /* flag the file so we don't compress in the future */
 496                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 497                     !(BTRFS_I(inode)->force_compress)) {
 498                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 499                 }
 500         }
 501         if (will_compress) {
 502                 *num_added += 1;
 503
 504                 /* the async work queues will take care of doing actual
 505                  * allocation on disk for these compressed pages,
 506                  * and will submit them to the elevator.
 507                  */
 508                 add_async_extent(async_cow, start, num_bytes,
 509                                  total_compressed, pages, nr_pages_ret,
 510                                  compress_type);
 511
 512                 if (start + num_bytes < end) {
 513                         start += num_bytes;
 514                         pages = NULL;
 515                         cond_resched();
 516                         goto again;
 517                 }
 518         } else {
 519 cleanup_and_bail_uncompressed:
 520                 /*
 521                  * No compression, but we still need to write the pages in
 522                  * the file we've been given so far.  redirty the locked
 523                  * page if it corresponds to our extent and set things up
 524                  * for the async work queue to run cow_file_range to do
 525                  * the normal delalloc dance
 526                  */
 527                 if (page_offset(locked_page) >= start &&
 528                     page_offset(locked_page) <= end) {
 529                         __set_page_dirty_nobuffers(locked_page);
 530                         /* unlocked later on in the async handlers */
 531                 }
 532                 add_async_extent(async_cow, start, end - start + 1,
 533                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 534                 *num_added += 1;
 535         }
 536
 537 out:
 538         return 0;
 539
 540 free_pages_out:
 541         for (i = 0; i < nr_pages_ret; i++) {
 542                 WARN_ON(pages[i]->mapping);
 543                 page_cache_release(pages[i]);
 544         }
 545         kfree(pages);
 546
 547         goto out;
 548 }
 549
 550 /*
 551  * phase two of compressed writeback.  This is the ordered portion
 552  * of the code, which only gets called in the order the work was
 553  * queued.  We walk all the async extents created by compress_file_range
 554  * and send them down to the disk.
 555  */
 556 static noinline int submit_compressed_extents(struct inode *inode,
 557                                               struct async_cow *async_cow)
 558 {
 559         struct async_extent *async_extent;
 560         u64 alloc_hint = 0;
 561         struct btrfs_trans_handle *trans;
 562         struct btrfs_key ins;
 563         struct extent_map *em;
 564         struct btrfs_root *root = BTRFS_I(inode)->root;
 565         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 566         struct extent_io_tree *io_tree;
 567         int ret = 0;
 568
 569         if (list_empty(&async_cow->extents))
 570                 return 0;
 571
 572
 573         while (!list_empty(&async_cow->extents)) {
 574                 async_extent = list_entry(async_cow->extents.next,
 575                                           struct async_extent, list);
 576                 list_del(&async_extent->list);
 577
 578                 io_tree = &BTRFS_I(inode)->io_tree;
 579
 580 retry:
 581                 /* did the compression code fall back to uncompressed IO? */
 582                 if (!async_extent->pages) {
 583                         int page_started = 0;
 584                         unsigned long nr_written = 0;
 585
 586                         lock_extent(io_tree, async_extent->start,
 587                                          async_extent->start +
 588                                          async_extent->ram_size - 1, GFP_NOFS);
 589
 590                         /* allocate blocks */
 591                         ret = cow_file_range(inode, async_cow->locked_page,
 592                                              async_extent->start,
 593                                              async_extent->start +
 594                                              async_extent->ram_size - 1,
 595                                              &page_started, &nr_written, 0);
 596
 597                         /*
 598                          * if page_started, cow_file_range inserted an
 599                          * inline extent and took care of all the unlocking
 600                          * and IO for us.  Otherwise, we need to submit
 601                          * all those pages down to the drive.
 602                          */
 603                         if (!page_started && !ret)
 604                                 extent_write_locked_range(io_tree,
 605                                                   inode, async_extent->start,
 606                                                   async_extent->start +
 607                                                   async_extent->ram_size - 1,
 608                                                   btrfs_get_extent,
 609                                                   WB_SYNC_ALL);
 610                         kfree(async_extent);
 611                         cond_resched();
 612                         continue;
 613                 }
 614
 615                 lock_extent(io_tree, async_extent->start,
 616                             async_extent->start + async_extent->ram_size - 1,
 617                             GFP_NOFS);
 618
 619                 trans = btrfs_join_transaction(root, 1);
 620                 BUG_ON(IS_ERR(trans));
 621                 ret = btrfs_reserve_extent(trans, root,
 622                                            async_extent->compressed_size,
 623                                            async_extent->compressed_size,
 624                                            0, alloc_hint,
 625                                            (u64)-1, &ins, 1);
 626                 btrfs_end_transaction(trans, root);
 627
 628                 if (ret) {
 629                         int i;
 630                         for (i = 0; i < async_extent->nr_pages; i++) {
 631                                 WARN_ON(async_extent->pages[i]->mapping);
 632                                 page_cache_release(async_extent->pages[i]);
 633                         }
 634                         kfree(async_extent->pages);
 635                         async_extent->nr_pages = 0;
 636                         async_extent->pages = NULL;
 637                         unlock_extent(io_tree, async_extent->start,
 638                                       async_extent->start +
 639                                       async_extent->ram_size - 1, GFP_NOFS);
 640                         goto retry;
 641                 }
 642
 643                 /*
 644                  * here we're doing allocation and writeback of the
 645                  * compressed pages
 646                  */
 647                 btrfs_drop_extent_cache(inode, async_extent->start,
 648                                         async_extent->start +
 649                                         async_extent->ram_size - 1, 0);
 650
 651                 em = alloc_extent_map(GFP_NOFS);
 652                 BUG_ON(!em);
 653                 em->start = async_extent->start;
 654                 em->len = async_extent->ram_size;
 655                 em->orig_start = em->start;
 656
 657                 em->block_start = ins.objectid;
 658                 em->block_len = ins.offset;
 659                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 660                 em->compress_type = async_extent->compress_type;
 661                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 662                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 663
 664                 while (1) {
 665                         write_lock(&em_tree->lock);
 666                         ret = add_extent_mapping(em_tree, em);
 667                         write_unlock(&em_tree->lock);
 668                         if (ret != -EEXIST) {
 669                                 free_extent_map(em);
 670                                 break;
 671                         }
 672                         btrfs_drop_extent_cache(inode, async_extent->start,
 673                                                 async_extent->start +
 674                                                 async_extent->ram_size - 1, 0);
 675                 }
 676
 677                 ret = btrfs_add_ordered_extent_compress(inode,
 678                                                 async_extent->start,
 679                                                 ins.objectid,
 680                                                 async_extent->ram_size,
 681                                                 ins.offset,
 682                                                 BTRFS_ORDERED_COMPRESSED,
 683                                                 async_extent->compress_type);
 684                 BUG_ON(ret);
 685
 686                 /*
 687                  * clear dirty, set writeback and unlock the pages.
 688                  */
 689                 extent_clear_unlock_delalloc(inode,
 690                                 &BTRFS_I(inode)->io_tree,
 691                                 async_extent->start,
 692                                 async_extent->start +
 693                                 async_extent->ram_size - 1,
 694                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 695                                 EXTENT_CLEAR_UNLOCK |
 696                                 EXTENT_CLEAR_DELALLOC |
 697                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 698
 699                 ret = btrfs_submit_compressed_write(inode,
 700                                     async_extent->start,
 701                                     async_extent->ram_size,
 702                                     ins.objectid,
 703                                     ins.offset, async_extent->pages,
 704                                     async_extent->nr_pages);
 705
 706                 BUG_ON(ret);
 707                 alloc_hint = ins.objectid + ins.offset;
 708                 kfree(async_extent);
 709                 cond_resched();
 710         }
 711
 712         return 0;
 713 }
 714
 715 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 716                                       u64 num_bytes)
 717 {
 718         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 719         struct extent_map *em;
 720         u64 alloc_hint = 0;
 721
 722         read_lock(&em_tree->lock);
 723         em = search_extent_mapping(em_tree, start, num_bytes);
 724         if (em) {
 725                 /*
 726                  * if block start isn't an actual block number then find the
 727                  * first block in this inode and use that as a hint.  If that
 728                  * block is also bogus then just don't worry about it.
 729                  */
 730                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 731                         free_extent_map(em);
 732                         em = search_extent_mapping(em_tree, 0, 0);
 733                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 734                                 alloc_hint = em->block_start;
 735                         if (em)
 736                                 free_extent_map(em);
 737                 } else {
 738                         alloc_hint = em->block_start;
 739                         free_extent_map(em);
 740                 }
 741         }
 742         read_unlock(&em_tree->lock);
 743
 744         return alloc_hint;
 745 }
 746
 747 /*
 748  * when extent_io.c finds a delayed allocation range in the file,
 749  * the call backs end up in this code.  The basic idea is to
 750  * allocate extents on disk for the range, and create ordered data structs
 751  * in ram to track those extents.
 752  *
 753  * locked_page is the page that writepage had locked already.  We use
 754  * it to make sure we don't do extra locks or unlocks.
 755  *
 756  * *page_started is set to one if we unlock locked_page and do everything
 757  * required to start IO on it.  It may be clean and already done with
 758  * IO when we return.
 759  */
 760 static noinline int cow_file_range(struct inode *inode,
 761                                    struct page *locked_page,
 762                                    u64 start, u64 end, int *page_started,
 763                                    unsigned long *nr_written,
 764                                    int unlock)
 765 {
 766         struct btrfs_root *root = BTRFS_I(inode)->root;
 767         struct btrfs_trans_handle *trans;
 768         u64 alloc_hint = 0;
 769         u64 num_bytes;
 770         unsigned long ram_size;
 771         u64 disk_num_bytes;
 772         u64 cur_alloc_size;
 773         u64 blocksize = root->sectorsize;
 774         struct btrfs_key ins;
 775         struct extent_map *em;
 776         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 777         int ret = 0;
 778
 779         BUG_ON(root == root->fs_info->tree_root);
 780         trans = btrfs_join_transaction(root, 1);
 781         BUG_ON(IS_ERR(trans));
 782         btrfs_set_trans_block_group(trans, inode);
 783         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 784
 785         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 786         num_bytes = max(blocksize,  num_bytes);
 787         disk_num_bytes = num_bytes;
 788         ret = 0;
 789
 790         if (start == 0) {
 791                 /* lets try to make an inline extent */
 792                 ret = cow_file_range_inline(trans, root, inode,
 793                                             start, end, 0, 0, NULL);
 794                 if (ret == 0) {
 795                         extent_clear_unlock_delalloc(inode,
 796                                      &BTRFS_I(inode)->io_tree,
 797                                      start, end, NULL,
 798                                      EXTENT_CLEAR_UNLOCK_PAGE |
 799                                      EXTENT_CLEAR_UNLOCK |
 800                                      EXTENT_CLEAR_DELALLOC |
 801                                      EXTENT_CLEAR_DIRTY |
 802                                      EXTENT_SET_WRITEBACK |
 803                                      EXTENT_END_WRITEBACK);
 804
 805                         *nr_written = *nr_written +
 806                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 807                         *page_started = 1;
 808                         ret = 0;
 809                         goto out;
 810                 }
 811         }
 812
 813         BUG_ON(disk_num_bytes >
 814                btrfs_super_total_bytes(&root->fs_info->super_copy));
 815
 816         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 817         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 818
 819         while (disk_num_bytes > 0) {
 820                 unsigned long op;
 821
 822                 cur_alloc_size = disk_num_bytes;
 823                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 824                                            root->sectorsize, 0, alloc_hint,
 825                                            (u64)-1, &ins, 1);
 826                 BUG_ON(ret);
 827
 828                 em = alloc_extent_map(GFP_NOFS);
 829                 BUG_ON(!em);
 830                 em->start = start;
 831                 em->orig_start = em->start;
 832                 ram_size = ins.offset;
 833                 em->len = ins.offset;
 834
 835                 em->block_start = ins.objectid;
 836                 em->block_len = ins.offset;
 837                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 838                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 839
 840                 while (1) {
 841                         write_lock(&em_tree->lock);
 842                         ret = add_extent_mapping(em_tree, em);
 843                         write_unlock(&em_tree->lock);
 844                         if (ret != -EEXIST) {
 845                                 free_extent_map(em);
 846                                 break;
 847                         }
 848                         btrfs_drop_extent_cache(inode, start,
 849                                                 start + ram_size - 1, 0);
 850                 }
 851
 852                 cur_alloc_size = ins.offset;
 853                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 854                                                ram_size, cur_alloc_size, 0);
 855                 BUG_ON(ret);
 856
 857                 if (root->root_key.objectid ==
 858                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 859                         ret = btrfs_reloc_clone_csums(inode, start,
 860                                                       cur_alloc_size);
 861                         BUG_ON(ret);
 862                 }
 863
 864                 if (disk_num_bytes < cur_alloc_size)
 865                         break;
 866
 867                 /* we're not doing compressed IO, don't unlock the first
 868                  * page (which the caller expects to stay locked), don't
 869                  * clear any dirty bits and don't set any writeback bits
 870                  *
 871                  * Do set the Private2 bit so we know this page was properly
 872                  * setup for writepage
 873                  */
 874                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 875                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 876                         EXTENT_SET_PRIVATE2;
 877
 878                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 879                                              start, start + ram_size - 1,
 880                                              locked_page, op);
 881                 disk_num_bytes -= cur_alloc_size;
 882                 num_bytes -= cur_alloc_size;
 883                 alloc_hint = ins.objectid + ins.offset;
 884                 start += cur_alloc_size;
 885         }
 886 out:
 887         ret = 0;
 888         btrfs_end_transaction(trans, root);
 889
 890         return ret;
 891 }
 892
 893 /*
 894  * work queue call back to started compression on a file and pages
 895  */
 896 static noinline void async_cow_start(struct btrfs_work *work)
 897 {
 898         struct async_cow *async_cow;
 899         int num_added = 0;
 900         async_cow = container_of(work, struct async_cow, work);
 901
 902         compress_file_range(async_cow->inode, async_cow->locked_page,
 903                             async_cow->start, async_cow->end, async_cow,
 904                             &num_added);
 905         if (num_added == 0)
 906                 async_cow->inode = NULL;
 907 }
 908
 909 /*
 910  * work queue call back to submit previously compressed pages
 911  */
 912 static noinline void async_cow_submit(struct btrfs_work *work)
 913 {
 914         struct async_cow *async_cow;
 915         struct btrfs_root *root;
 916         unsigned long nr_pages;
 917
 918         async_cow = container_of(work, struct async_cow, work);
 919
 920         root = async_cow->root;
 921         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 922                 PAGE_CACHE_SHIFT;
 923
 924         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 925
 926         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 927             5 * 1042 * 1024 &&
 928             waitqueue_active(&root->fs_info->async_submit_wait))
 929                 wake_up(&root->fs_info->async_submit_wait);
 930
 931         if (async_cow->inode)
 932                 submit_compressed_extents(async_cow->inode, async_cow);
 933 }
 934
 935 static noinline void async_cow_free(struct btrfs_work *work)
 936 {
 937         struct async_cow *async_cow;
 938         async_cow = container_of(work, struct async_cow, work);
 939         kfree(async_cow);
 940 }
 941
 942 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 943                                 u64 start, u64 end, int *page_started,
 944                                 unsigned long *nr_written)
 945 {
 946         struct async_cow *async_cow;
 947         struct btrfs_root *root = BTRFS_I(inode)->root;
 948         unsigned long nr_pages;
 949         u64 cur_end;
 950         int limit = 10 * 1024 * 1042;
 951
 952         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 953                          1, 0, NULL, GFP_NOFS);
 954         while (start < end) {
 955                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 956                 async_cow->inode = inode;
 957                 async_cow->root = root;
 958                 async_cow->locked_page = locked_page;
 959                 async_cow->start = start;
 960
 961                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 962                         cur_end = end;
 963                 else
 964                         cur_end = min(end, start + 512 * 1024 - 1);
 965
 966                 async_cow->end = cur_end;
 967                 INIT_LIST_HEAD(&async_cow->extents);
 968
 969                 async_cow->work.func = async_cow_start;
 970                 async_cow->work.ordered_func = async_cow_submit;
 971                 async_cow->work.ordered_free = async_cow_free;
 972                 async_cow->work.flags = 0;
 973
 974                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 975                         PAGE_CACHE_SHIFT;
 976                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 977
 978                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 979                                    &async_cow->work);
 980
 981                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 982                         wait_event(root->fs_info->async_submit_wait,
 983                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 984                             limit));
 985                 }
 986
 987                 while (atomic_read(&root->fs_info->async_submit_draining) &&
 988                       atomic_read(&root->fs_info->async_delalloc_pages)) {
 989                         wait_event(root->fs_info->async_submit_wait,
 990                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
 991                            0));
 992                 }
 993
 994                 *nr_written += nr_pages;
 995                 start = cur_end + 1;
 996         }
 997         *page_started = 1;
 998         return 0;
 999 }
1000
1001 static noinline int csum_exist_in_range(struct btrfs_root *root,
1002                                         u64 bytenr, u64 num_bytes)
1003 {
1004         int ret;
1005         struct btrfs_ordered_sum *sums;
1006         LIST_HEAD(list);
1007
1008         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1009                                        bytenr + num_bytes - 1, &list);
1010         if (ret == 0 && list_empty(&list))
1011                 return 0;
1012
1013         while (!list_empty(&list)) {
1014                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1015                 list_del(&sums->list);
1016                 kfree(sums);
1017         }
1018         return 1;
1019 }
1020
1021 /*
1022  * when nowcow writeback call back.  This checks for snapshots or COW copies
1023  * of the extents that exist in the file, and COWs the file as required.
1024  *
1025  * If no cow copies or snapshots exist, we write directly to the existing
1026  * blocks on disk
1027  */
1028 static noinline int run_delalloc_nocow(struct inode *inode,
1029                                        struct page *locked_page,
1030                               u64 start, u64 end, int *page_started, int force,
1031                               unsigned long *nr_written)
1032 {
1033         struct btrfs_root *root = BTRFS_I(inode)->root;
1034         struct btrfs_trans_handle *trans;
1035         struct extent_buffer *leaf;
1036         struct btrfs_path *path;
1037         struct btrfs_file_extent_item *fi;
1038         struct btrfs_key found_key;
1039         u64 cow_start;
1040         u64 cur_offset;
1041         u64 extent_end;
1042         u64 extent_offset;
1043         u64 disk_bytenr;
1044         u64 num_bytes;
1045         int extent_type;
1046         int ret;
1047         int type;
1048         int nocow;
1049         int check_prev = 1;
1050         bool nolock = false;
1051
1052         path = btrfs_alloc_path();
1053         BUG_ON(!path);
1054         if (root == root->fs_info->tree_root) {
1055                 nolock = true;
1056                 trans = btrfs_join_transaction_nolock(root, 1);
1057         } else {
1058                 trans = btrfs_join_transaction(root, 1);
1059         }
1060         BUG_ON(IS_ERR(trans));
1061
1062         cow_start = (u64)-1;
1063         cur_offset = start;
1064         while (1) {
1065                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
1066                                                cur_offset, 0);
1067                 BUG_ON(ret < 0);
1068                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1069                         leaf = path->nodes[0];
1070                         btrfs_item_key_to_cpu(leaf, &found_key,
1071                                               path->slots[0] - 1);
1072                         if (found_key.objectid == inode->i_ino &&
1073                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1074                                 path->slots[0]--;
1075                 }
1076                 check_prev = 0;
1077 next_slot:
1078                 leaf = path->nodes[0];
1079                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1080                         ret = btrfs_next_leaf(root, path);
1081                         if (ret < 0)
1082                                 BUG_ON(1);
1083                         if (ret > 0)
1084                                 break;
1085                         leaf = path->nodes[0];
1086                 }
1087
1088                 nocow = 0;
1089                 disk_bytenr = 0;
1090                 num_bytes = 0;
1091                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1092
1093                 if (found_key.objectid > inode->i_ino ||
1094                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1095                     found_key.offset > end)
1096                         break;
1097
1098                 if (found_key.offset > cur_offset) {
1099                         extent_end = found_key.offset;
1100                         extent_type = 0;
1101                         goto out_check;
1102                 }
1103
1104                 fi = btrfs_item_ptr(leaf, path->slots[0],
1105                                     struct btrfs_file_extent_item);
1106                 extent_type = btrfs_file_extent_type(leaf, fi);
1107
1108                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1109                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1110                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1111                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1112                         extent_end = found_key.offset +
1113                                 btrfs_file_extent_num_bytes(leaf, fi);
1114                         if (extent_end <= start) {
1115                                 path->slots[0]++;
1116                                 goto next_slot;
1117                         }
1118                         if (disk_bytenr == 0)
1119                                 goto out_check;
1120                         if (btrfs_file_extent_compression(leaf, fi) ||
1121                             btrfs_file_extent_encryption(leaf, fi) ||
1122                             btrfs_file_extent_other_encoding(leaf, fi))
1123                                 goto out_check;
1124                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1125                                 goto out_check;
1126                         if (btrfs_extent_readonly(root, disk_bytenr))
1127                                 goto out_check;
1128                         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
1129                                                   found_key.offset -
1130                                                   extent_offset, disk_bytenr))
1131                                 goto out_check;
1132                         disk_bytenr += extent_offset;
1133                         disk_bytenr += cur_offset - found_key.offset;
1134                         num_bytes = min(end + 1, extent_end) - cur_offset;
1135                         /*
1136                          * force cow if csum exists in the range.
1137                          * this ensure that csum for a given extent are
1138                          * either valid or do not exist.
1139                          */
1140                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1141                                 goto out_check;
1142                         nocow = 1;
1143                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1144                         extent_end = found_key.offset +
1145                                 btrfs_file_extent_inline_len(leaf, fi);
1146                         extent_end = ALIGN(extent_end, root->sectorsize);
1147                 } else {
1148                         BUG_ON(1);
1149                 }
1150 out_check:
1151                 if (extent_end <= start) {
1152                         path->slots[0]++;
1153                         goto next_slot;
1154                 }
1155                 if (!nocow) {
1156                         if (cow_start == (u64)-1)
1157                                 cow_start = cur_offset;
1158                         cur_offset = extent_end;
1159                         if (cur_offset > end)
1160                                 break;
1161                         path->slots[0]++;
1162                         goto next_slot;
1163                 }
1164
1165                 btrfs_release_path(root, path);
1166                 if (cow_start != (u64)-1) {
1167                         ret = cow_file_range(inode, locked_page, cow_start,
1168                                         found_key.offset - 1, page_started,
1169                                         nr_written, 1);
1170                         BUG_ON(ret);
1171                         cow_start = (u64)-1;
1172                 }
1173
1174                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1175                         struct extent_map *em;
1176                         struct extent_map_tree *em_tree;
1177                         em_tree = &BTRFS_I(inode)->extent_tree;
1178                         em = alloc_extent_map(GFP_NOFS);
1179                         BUG_ON(!em);
1180                         em->start = cur_offset;
1181                         em->orig_start = em->start;
1182                         em->len = num_bytes;
1183                         em->block_len = num_bytes;
1184                         em->block_start = disk_bytenr;
1185                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1186                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1187                         while (1) {
1188                                 write_lock(&em_tree->lock);
1189                                 ret = add_extent_mapping(em_tree, em);
1190                                 write_unlock(&em_tree->lock);
1191                                 if (ret != -EEXIST) {
1192                                         free_extent_map(em);
1193                                         break;
1194                                 }
1195                                 btrfs_drop_extent_cache(inode, em->start,
1196                                                 em->start + em->len - 1, 0);
1197                         }
1198                         type = BTRFS_ORDERED_PREALLOC;
1199                 } else {
1200                         type = BTRFS_ORDERED_NOCOW;
1201                 }
1202
1203                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1204                                                num_bytes, num_bytes, type);
1205                 BUG_ON(ret);
1206
1207                 if (root->root_key.objectid ==
1208                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1209                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1210                                                       num_bytes);
1211                         BUG_ON(ret);
1212                 }
1213
1214                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1215                                 cur_offset, cur_offset + num_bytes - 1,
1216                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1217                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1218                                 EXTENT_SET_PRIVATE2);
1219                 cur_offset = extent_end;
1220                 if (cur_offset > end)
1221                         break;
1222         }
1223         btrfs_release_path(root, path);
1224
1225         if (cur_offset <= end && cow_start == (u64)-1)
1226                 cow_start = cur_offset;
1227         if (cow_start != (u64)-1) {
1228                 ret = cow_file_range(inode, locked_page, cow_start, end,
1229                                      page_started, nr_written, 1);
1230                 BUG_ON(ret);
1231         }
1232
1233         if (nolock) {
1234                 ret = btrfs_end_transaction_nolock(trans, root);
1235                 BUG_ON(ret);
1236         } else {
1237                 ret = btrfs_end_transaction(trans, root);
1238                 BUG_ON(ret);
1239         }
1240         btrfs_free_path(path);
1241         return 0;
1242 }
1243
1244 /*
1245  * extent_io.c call back to do delayed allocation processing
1246  */
1247 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1248                               u64 start, u64 end, int *page_started,
1249                               unsigned long *nr_written)
1250 {
1251         int ret;
1252         struct btrfs_root *root = BTRFS_I(inode)->root;
1253
1254         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1255                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1256                                          page_started, 1, nr_written);
1257         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1258                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1259                                          page_started, 0, nr_written);
1260         else if (!btrfs_test_opt(root, COMPRESS) &&
1261                  !(BTRFS_I(inode)->force_compress) &&
1262                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1263                 ret = cow_file_range(inode, locked_page, start, end,
1264                                       page_started, nr_written, 1);
1265         else
1266                 ret = cow_file_range_async(inode, locked_page, start, end,
1267                                            page_started, nr_written);
1268         return ret;
1269 }
1270
1271 static int btrfs_split_extent_hook(struct inode *inode,
1272                                    struct extent_state *orig, u64 split)
1273 {
1274         /* not delalloc, ignore it */
1275         if (!(orig->state & EXTENT_DELALLOC))
1276                 return 0;
1277
1278         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1279         return 0;
1280 }
1281
1282 /*
1283  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1284  * extents so we can keep track of new extents that are just merged onto old
1285  * extents, such as when we are doing sequential writes, so we can properly
1286  * account for the metadata space we'll need.
1287  */
1288 static int btrfs_merge_extent_hook(struct inode *inode,
1289                                    struct extent_state *new,
1290                                    struct extent_state *other)
1291 {
1292         /* not delalloc, ignore it */
1293         if (!(other->state & EXTENT_DELALLOC))
1294                 return 0;
1295
1296         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1297         return 0;
1298 }
1299
1300 /*
1301  * extent_io.c set_bit_hook, used to track delayed allocation
1302  * bytes in this file, and to maintain the list of inodes that
1303  * have pending delalloc work to be done.
1304  */
1305 static int btrfs_set_bit_hook(struct inode *inode,
1306                               struct extent_state *state, int *bits)
1307 {
1308
1309         /*
1310          * set_bit and clear bit hooks normally require _irqsave/restore
1311          * but in this case, we are only testeing for the DELALLOC
1312          * bit, which is only set or cleared with irqs on
1313          */
1314         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1315                 struct btrfs_root *root = BTRFS_I(inode)->root;
1316                 u64 len = state->end + 1 - state->start;
1317                 int do_list = (root->root_key.objectid !=
1318                                BTRFS_ROOT_TREE_OBJECTID);
1319
1320                 if (*bits & EXTENT_FIRST_DELALLOC)
1321                         *bits &= ~EXTENT_FIRST_DELALLOC;
1322                 else
1323                         atomic_inc(&BTRFS_I(inode)->outstanding_extents);
1324
1325                 spin_lock(&root->fs_info->delalloc_lock);
1326                 BTRFS_I(inode)->delalloc_bytes += len;
1327                 root->fs_info->delalloc_bytes += len;
1328                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1329                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1330                                       &root->fs_info->delalloc_inodes);
1331                 }
1332                 spin_unlock(&root->fs_info->delalloc_lock);
1333         }
1334         return 0;
1335 }
1336
1337 /*
1338  * extent_io.c clear_bit_hook, see set_bit_hook for why
1339  */
1340 static int btrfs_clear_bit_hook(struct inode *inode,
1341                                 struct extent_state *state, int *bits)
1342 {
1343         /*
1344          * set_bit and clear bit hooks normally require _irqsave/restore
1345          * but in this case, we are only testeing for the DELALLOC
1346          * bit, which is only set or cleared with irqs on
1347          */
1348         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1349                 struct btrfs_root *root = BTRFS_I(inode)->root;
1350                 u64 len = state->end + 1 - state->start;
1351                 int do_list = (root->root_key.objectid !=
1352                                BTRFS_ROOT_TREE_OBJECTID);
1353
1354                 if (*bits & EXTENT_FIRST_DELALLOC)
1355                         *bits &= ~EXTENT_FIRST_DELALLOC;
1356                 else if (!(*bits & EXTENT_DO_ACCOUNTING))
1357                         atomic_dec(&BTRFS_I(inode)->outstanding_extents);
1358
1359                 if (*bits & EXTENT_DO_ACCOUNTING)
1360                         btrfs_delalloc_release_metadata(inode, len);
1361
1362                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1363                     && do_list)
1364                         btrfs_free_reserved_data_space(inode, len);
1365
1366                 spin_lock(&root->fs_info->delalloc_lock);
1367                 root->fs_info->delalloc_bytes -= len;
1368                 BTRFS_I(inode)->delalloc_bytes -= len;
1369
1370                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1371                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1372                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1373                 }
1374                 spin_unlock(&root->fs_info->delalloc_lock);
1375         }
1376         return 0;
1377 }
1378
1379 /*
1380  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1381  * we don't create bios that span stripes or chunks
1382  */
1383 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1384                          size_t size, struct bio *bio,
1385                          unsigned long bio_flags)
1386 {
1387         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1388         struct btrfs_mapping_tree *map_tree;
1389         u64 logical = (u64)bio->bi_sector << 9;
1390         u64 length = 0;
1391         u64 map_length;
1392         int ret;
1393
1394         if (bio_flags & EXTENT_BIO_COMPRESSED)
1395                 return 0;
1396
1397         length = bio->bi_size;
1398         map_tree = &root->fs_info->mapping_tree;
1399         map_length = length;
1400         ret = btrfs_map_block(map_tree, READ, logical,
1401                               &map_length, NULL, 0);
1402
1403         if (map_length < length + size)
1404                 return 1;
1405         return ret;
1406 }
1407
1408 /*
1409  * in order to insert checksums into the metadata in large chunks,
1410  * we wait until bio submission time.   All the pages in the bio are
1411  * checksummed and sums are attached onto the ordered extent record.
1412  *
1413  * At IO completion time the cums attached on the ordered extent record
1414  * are inserted into the btree
1415  */
1416 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1417                                     struct bio *bio, int mirror_num,
1418                                     unsigned long bio_flags,
1419                                     u64 bio_offset)
1420 {
1421         struct btrfs_root *root = BTRFS_I(inode)->root;
1422         int ret = 0;
1423
1424         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1425         BUG_ON(ret);
1426         return 0;
1427 }
1428
1429 /*
1430  * in order to insert checksums into the metadata in large chunks,
1431  * we wait until bio submission time.   All the pages in the bio are
1432  * checksummed and sums are attached onto the ordered extent record.
1433  *
1434  * At IO completion time the cums attached on the ordered extent record
1435  * are inserted into the btree
1436  */
1437 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1438                           int mirror_num, unsigned long bio_flags,
1439                           u64 bio_offset)
1440 {
1441         struct btrfs_root *root = BTRFS_I(inode)->root;
1442         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1443 }
1444
1445 /*
1446  * extent_io.c submission hook. This does the right thing for csum calculation
1447  * on write, or reading the csums from the tree before a read
1448  */
1449 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1450                           int mirror_num, unsigned long bio_flags,
1451                           u64 bio_offset)
1452 {
1453         struct btrfs_root *root = BTRFS_I(inode)->root;
1454         int ret = 0;
1455         int skip_sum;
1456
1457         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1458
1459         if (root == root->fs_info->tree_root)
1460                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1461         else
1462                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1463         BUG_ON(ret);
1464
1465         if (!(rw & REQ_WRITE)) {
1466                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1467                         return btrfs_submit_compressed_read(inode, bio,
1468                                                     mirror_num, bio_flags);
1469                 } else if (!skip_sum) {
1470                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1471                         if (ret)
1472                                 return ret;
1473                 }
1474                 goto mapit;
1475         } else if (!skip_sum) {
1476                 /* csum items have already been cloned */
1477                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1478                         goto mapit;
1479                 /* we're doing a write, do the async checksumming */
1480                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1481                                    inode, rw, bio, mirror_num,
1482                                    bio_flags, bio_offset,
1483                                    __btrfs_submit_bio_start,
1484                                    __btrfs_submit_bio_done);
1485         }
1486
1487 mapit:
1488         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1489 }
1490
1491 /*
1492  * given a list of ordered sums record them in the inode.  This happens
1493  * at IO completion time based on sums calculated at bio submission time.
1494  */
1495 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1496                              struct inode *inode, u64 file_offset,
1497                              struct list_head *list)
1498 {
1499         struct btrfs_ordered_sum *sum;
1500
1501         btrfs_set_trans_block_group(trans, inode);
1502
1503         list_for_each_entry(sum, list, list) {
1504                 btrfs_csum_file_blocks(trans,
1505                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1506         }
1507         return 0;
1508 }
1509
1510 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1511                               struct extent_state **cached_state)
1512 {
1513         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1514                 WARN_ON(1);
1515         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1516                                    cached_state, GFP_NOFS);
1517 }
1518
1519 /* see btrfs_writepage_start_hook for details on why this is required */
1520 struct btrfs_writepage_fixup {
1521         struct page *page;
1522         struct btrfs_work work;
1523 };
1524
1525 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1526 {
1527         struct btrfs_writepage_fixup *fixup;
1528         struct btrfs_ordered_extent *ordered;
1529         struct extent_state *cached_state = NULL;
1530         struct page *page;
1531         struct inode *inode;
1532         u64 page_start;
1533         u64 page_end;
1534
1535         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1536         page = fixup->page;
1537 again:
1538         lock_page(page);
1539         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1540                 ClearPageChecked(page);
1541                 goto out_page;
1542         }
1543
1544         inode = page->mapping->host;
1545         page_start = page_offset(page);
1546         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1547
1548         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1549                          &cached_state, GFP_NOFS);
1550
1551         /* already ordered? We're done */
1552         if (PagePrivate2(page))
1553                 goto out;
1554
1555         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1556         if (ordered) {
1557                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1558                                      page_end, &cached_state, GFP_NOFS);
1559                 unlock_page(page);
1560                 btrfs_start_ordered_extent(inode, ordered, 1);
1561                 goto again;
1562         }
1563
1564         BUG();
1565         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1566         ClearPageChecked(page);
1567 out:
1568         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1569                              &cached_state, GFP_NOFS);
1570 out_page:
1571         unlock_page(page);
1572         page_cache_release(page);
1573         kfree(fixup);
1574 }
1575
1576 /*
1577  * There are a few paths in the higher layers of the kernel that directly
1578  * set the page dirty bit without asking the filesystem if it is a
1579  * good idea.  This causes problems because we want to make sure COW
1580  * properly happens and the data=ordered rules are followed.
1581  *
1582  * In our case any range that doesn't have the ORDERED bit set
1583  * hasn't been properly setup for IO.  We kick off an async process
1584  * to fix it up.  The async helper will wait for ordered extents, set
1585  * the delalloc bit and make it safe to write the page.
1586  */
1587 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1588 {
1589         struct inode *inode = page->mapping->host;
1590         struct btrfs_writepage_fixup *fixup;
1591         struct btrfs_root *root = BTRFS_I(inode)->root;
1592
1593         /* this page is properly in the ordered list */
1594         if (TestClearPagePrivate2(page))
1595                 return 0;
1596
1597         if (PageChecked(page))
1598                 return -EAGAIN;
1599
1600         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1601         if (!fixup)
1602                 return -EAGAIN;
1603
1604         SetPageChecked(page);
1605         page_cache_get(page);
1606         fixup->work.func = btrfs_writepage_fixup_worker;
1607         fixup->page = page;
1608         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1609         return -EAGAIN;
1610 }
1611
1612 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1613                                        struct inode *inode, u64 file_pos,
1614                                        u64 disk_bytenr, u64 disk_num_bytes,
1615                                        u64 num_bytes, u64 ram_bytes,
1616                                        u8 compression, u8 encryption,
1617                                        u16 other_encoding, int extent_type)
1618 {
1619         struct btrfs_root *root = BTRFS_I(inode)->root;
1620         struct btrfs_file_extent_item *fi;
1621         struct btrfs_path *path;
1622         struct extent_buffer *leaf;
1623         struct btrfs_key ins;
1624         u64 hint;
1625         int ret;
1626
1627         path = btrfs_alloc_path();
1628         BUG_ON(!path);
1629
1630         path->leave_spinning = 1;
1631
1632         /*
1633          * we may be replacing one extent in the tree with another.
1634          * The new extent is pinned in the extent map, and we don't want
1635          * to drop it from the cache until it is completely in the btree.
1636          *
1637          * So, tell btrfs_drop_extents to leave this extent in the cache.
1638          * the caller is expected to unpin it and allow it to be merged
1639          * with the others.
1640          */
1641         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1642                                  &hint, 0);
1643         BUG_ON(ret);
1644
1645         ins.objectid = inode->i_ino;
1646         ins.offset = file_pos;
1647         ins.type = BTRFS_EXTENT_DATA_KEY;
1648         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1649         BUG_ON(ret);
1650         leaf = path->nodes[0];
1651         fi = btrfs_item_ptr(leaf, path->slots[0],
1652                             struct btrfs_file_extent_item);
1653         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1654         btrfs_set_file_extent_type(leaf, fi, extent_type);
1655         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1656         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1657         btrfs_set_file_extent_offset(leaf, fi, 0);
1658         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1659         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1660         btrfs_set_file_extent_compression(leaf, fi, compression);
1661         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1662         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1663
1664         btrfs_unlock_up_safe(path, 1);
1665         btrfs_set_lock_blocking(leaf);
1666
1667         btrfs_mark_buffer_dirty(leaf);
1668
1669         inode_add_bytes(inode, num_bytes);
1670
1671         ins.objectid = disk_bytenr;
1672         ins.offset = disk_num_bytes;
1673         ins.type = BTRFS_EXTENT_ITEM_KEY;
1674         ret = btrfs_alloc_reserved_file_extent(trans, root,
1675                                         root->root_key.objectid,
1676                                         inode->i_ino, file_pos, &ins);
1677         BUG_ON(ret);
1678         btrfs_free_path(path);
1679
1680         return 0;
1681 }
1682
1683 /*
1684  * helper function for btrfs_finish_ordered_io, this
1685  * just reads in some of the csum leaves to prime them into ram
1686  * before we start the transaction.  It limits the amount of btree
1687  * reads required while inside the transaction.
1688  */
1689 /* as ordered data IO finishes, this gets called so we can finish
1690  * an ordered extent if the range of bytes in the file it covers are
1691  * fully written.
1692  */
1693 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1694 {
1695         struct btrfs_root *root = BTRFS_I(inode)->root;
1696         struct btrfs_trans_handle *trans = NULL;
1697         struct btrfs_ordered_extent *ordered_extent = NULL;
1698         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1699         struct extent_state *cached_state = NULL;
1700         int compress_type = 0;
1701         int ret;
1702         bool nolock = false;
1703
1704         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1705                                              end - start + 1);
1706         if (!ret)
1707                 return 0;
1708         BUG_ON(!ordered_extent);
1709
1710         nolock = (root == root->fs_info->tree_root);
1711
1712         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1713                 BUG_ON(!list_empty(&ordered_extent->list));
1714                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1715                 if (!ret) {
1716                         if (nolock)
1717                                 trans = btrfs_join_transaction_nolock(root, 1);
1718                         else
1719                                 trans = btrfs_join_transaction(root, 1);
1720                         BUG_ON(IS_ERR(trans));
1721                         btrfs_set_trans_block_group(trans, inode);
1722                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1723                         ret = btrfs_update_inode(trans, root, inode);
1724                         BUG_ON(ret);
1725                 }
1726                 goto out;
1727         }
1728
1729         lock_extent_bits(io_tree, ordered_extent->file_offset,
1730                          ordered_extent->file_offset + ordered_extent->len - 1,
1731                          0, &cached_state, GFP_NOFS);
1732
1733         if (nolock)
1734                 trans = btrfs_join_transaction_nolock(root, 1);
1735         else
1736                 trans = btrfs_join_transaction(root, 1);
1737         BUG_ON(IS_ERR(trans));
1738         btrfs_set_trans_block_group(trans, inode);
1739         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1740
1741         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1742                 compress_type = ordered_extent->compress_type;
1743         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1744                 BUG_ON(compress_type);
1745                 ret = btrfs_mark_extent_written(trans, inode,
1746                                                 ordered_extent->file_offset,
1747                                                 ordered_extent->file_offset +
1748                                                 ordered_extent->len);
1749                 BUG_ON(ret);
1750         } else {
1751                 BUG_ON(root == root->fs_info->tree_root);
1752                 ret = insert_reserved_file_extent(trans, inode,
1753                                                 ordered_extent->file_offset,
1754                                                 ordered_extent->start,
1755                                                 ordered_extent->disk_len,
1756                                                 ordered_extent->len,
1757                                                 ordered_extent->len,
1758                                                 compress_type, 0, 0,
1759                                                 BTRFS_FILE_EXTENT_REG);
1760                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1761                                    ordered_extent->file_offset,
1762                                    ordered_extent->len);
1763                 BUG_ON(ret);
1764         }
1765         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1766                              ordered_extent->file_offset +
1767                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1768
1769         add_pending_csums(trans, inode, ordered_extent->file_offset,
1770                           &ordered_extent->list);
1771
1772         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1773         if (!ret) {
1774                 ret = btrfs_update_inode(trans, root, inode);
1775                 BUG_ON(ret);
1776         }
1777         ret = 0;
1778 out:
1779         if (nolock) {
1780                 if (trans)
1781                         btrfs_end_transaction_nolock(trans, root);
1782         } else {
1783                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1784                 if (trans)
1785                         btrfs_end_transaction(trans, root);
1786         }
1787
1788         /* once for us */
1789         btrfs_put_ordered_extent(ordered_extent);
1790         /* once for the tree */
1791         btrfs_put_ordered_extent(ordered_extent);
1792
1793         return 0;
1794 }
1795
1796 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1797                                 struct extent_state *state, int uptodate)
1798 {
1799         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1800
1801         ClearPagePrivate2(page);
1802         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1803 }
1804
1805 /*
1806  * When IO fails, either with EIO or csum verification fails, we
1807  * try other mirrors that might have a good copy of the data.  This
1808  * io_failure_record is used to record state as we go through all the
1809  * mirrors.  If another mirror has good data, the page is set up to date
1810  * and things continue.  If a good mirror can't be found, the original
1811  * bio end_io callback is called to indicate things have failed.
1812  */
1813 struct io_failure_record {
1814         struct page *page;
1815         u64 start;
1816         u64 len;
1817         u64 logical;
1818         unsigned long bio_flags;
1819         int last_mirror;
1820 };
1821
1822 static int btrfs_io_failed_hook(struct bio *failed_bio,
1823                          struct page *page, u64 start, u64 end,
1824                          struct extent_state *state)
1825 {
1826         struct io_failure_record *failrec = NULL;
1827         u64 private;
1828         struct extent_map *em;
1829         struct inode *inode = page->mapping->host;
1830         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1831         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1832         struct bio *bio;
1833         int num_copies;
1834         int ret;
1835         int rw;
1836         u64 logical;
1837
1838         ret = get_state_private(failure_tree, start, &private);
1839         if (ret) {
1840                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1841                 if (!failrec)
1842                         return -ENOMEM;
1843                 failrec->start = start;
1844                 failrec->len = end - start + 1;
1845                 failrec->last_mirror = 0;
1846                 failrec->bio_flags = 0;
1847
1848                 read_lock(&em_tree->lock);
1849                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1850                 if (em->start > start || em->start + em->len < start) {
1851                         free_extent_map(em);
1852                         em = NULL;
1853                 }
1854                 read_unlock(&em_tree->lock);
1855
1856                 if (!em || IS_ERR(em)) {
1857                         kfree(failrec);
1858                         return -EIO;
1859                 }
1860                 logical = start - em->start;
1861                 logical = em->block_start + logical;
1862                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1863                         logical = em->block_start;
1864                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1865                         extent_set_compress_type(&failrec->bio_flags,
1866                                                  em->compress_type);
1867                 }
1868                 failrec->logical = logical;
1869                 free_extent_map(em);
1870                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1871                                 EXTENT_DIRTY, GFP_NOFS);
1872                 set_state_private(failure_tree, start,
1873                                  (u64)(unsigned long)failrec);
1874         } else {
1875                 failrec = (struct io_failure_record *)(unsigned long)private;
1876         }
1877         num_copies = btrfs_num_copies(
1878                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1879                               failrec->logical, failrec->len);
1880         failrec->last_mirror++;
1881         if (!state) {
1882                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1883                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1884                                                     failrec->start,
1885                                                     EXTENT_LOCKED);
1886                 if (state && state->start != failrec->start)
1887                         state = NULL;
1888                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1889         }
1890         if (!state || failrec->last_mirror > num_copies) {
1891                 set_state_private(failure_tree, failrec->start, 0);
1892                 clear_extent_bits(failure_tree, failrec->start,
1893                                   failrec->start + failrec->len - 1,
1894                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1895                 kfree(failrec);
1896                 return -EIO;
1897         }
1898         bio = bio_alloc(GFP_NOFS, 1);
1899         bio->bi_private = state;
1900         bio->bi_end_io = failed_bio->bi_end_io;
1901         bio->bi_sector = failrec->logical >> 9;
1902         bio->bi_bdev = failed_bio->bi_bdev;
1903         bio->bi_size = 0;
1904
1905         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1906         if (failed_bio->bi_rw & REQ_WRITE)
1907                 rw = WRITE;
1908         else
1909                 rw = READ;
1910
1911         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1912                                                       failrec->last_mirror,
1913                                                       failrec->bio_flags, 0);
1914         return ret;
1915 }
1916
1917 /*
1918  * each time an IO finishes, we do a fast check in the IO failure tree
1919  * to see if we need to process or clean up an io_failure_record
1920  */
1921 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1922 {
1923         u64 private;
1924         u64 private_failure;
1925         struct io_failure_record *failure;
1926         int ret;
1927
1928         private = 0;
1929         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1930                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1931                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1932                                         start, &private_failure);
1933                 if (ret == 0) {
1934                         failure = (struct io_failure_record *)(unsigned long)
1935                                    private_failure;
1936                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1937                                           failure->start, 0);
1938                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1939                                           failure->start,
1940                                           failure->start + failure->len - 1,
1941                                           EXTENT_DIRTY | EXTENT_LOCKED,
1942                                           GFP_NOFS);
1943                         kfree(failure);
1944                 }
1945         }
1946         return 0;
1947 }
1948
1949 /*
1950  * when reads are done, we need to check csums to verify the data is correct
1951  * if there's a match, we allow the bio to finish.  If not, we go through
1952  * the io_failure_record routines to find good copies
1953  */
1954 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1955                                struct extent_state *state)
1956 {
1957         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1958         struct inode *inode = page->mapping->host;
1959         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1960         char *kaddr;
1961         u64 private = ~(u32)0;
1962         int ret;
1963         struct btrfs_root *root = BTRFS_I(inode)->root;
1964         u32 csum = ~(u32)0;
1965
1966         if (PageChecked(page)) {
1967                 ClearPageChecked(page);
1968                 goto good;
1969         }
1970
1971         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1972                 return 0;
1973
1974         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1975             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1976                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1977                                   GFP_NOFS);
1978                 return 0;
1979         }
1980
1981         if (state && state->start == start) {
1982                 private = state->private;
1983                 ret = 0;
1984         } else {
1985                 ret = get_state_private(io_tree, start, &private);
1986         }
1987         kaddr = kmap_atomic(page, KM_USER0);
1988         if (ret)
1989                 goto zeroit;
1990
1991         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1992         btrfs_csum_final(csum, (char *)&csum);
1993         if (csum != private)
1994                 goto zeroit;
1995
1996         kunmap_atomic(kaddr, KM_USER0);
1997 good:
1998         /* if the io failure tree for this inode is non-empty,
1999          * check to see if we've recovered from a failed IO
2000          */
2001         btrfs_clean_io_failures(inode, start);
2002         return 0;
2003
2004 zeroit:
2005         if (printk_ratelimit()) {
2006                 printk(KERN_INFO "btrfs csum failed ino %lu off %llu csum %u "
2007                        "private %llu\n", page->mapping->host->i_ino,
2008                        (unsigned long long)start, csum,
2009                        (unsigned long long)private);
2010         }
2011         memset(kaddr + offset, 1, end - start + 1);
2012         flush_dcache_page(page);
2013         kunmap_atomic(kaddr, KM_USER0);
2014         if (private == 0)
2015                 return 0;
2016         return -EIO;
2017 }
2018
2019 struct delayed_iput {
2020         struct list_head list;
2021         struct inode *inode;
2022 };
2023
2024 void btrfs_add_delayed_iput(struct inode *inode)
2025 {
2026         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2027         struct delayed_iput *delayed;
2028
2029         if (atomic_add_unless(&inode->i_count, -1, 1))
2030                 return;
2031
2032         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2033         delayed->inode = inode;
2034
2035         spin_lock(&fs_info->delayed_iput_lock);
2036         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2037         spin_unlock(&fs_info->delayed_iput_lock);
2038 }
2039
2040 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2041 {
2042         LIST_HEAD(list);
2043         struct btrfs_fs_info *fs_info = root->fs_info;
2044         struct delayed_iput *delayed;
2045         int empty;
2046
2047         spin_lock(&fs_info->delayed_iput_lock);
2048         empty = list_empty(&fs_info->delayed_iputs);
2049         spin_unlock(&fs_info->delayed_iput_lock);
2050         if (empty)
2051                 return;
2052
2053         down_read(&root->fs_info->cleanup_work_sem);
2054         spin_lock(&fs_info->delayed_iput_lock);
2055         list_splice_init(&fs_info->delayed_iputs, &list);
2056         spin_unlock(&fs_info->delayed_iput_lock);
2057
2058         while (!list_empty(&list)) {
2059                 delayed = list_entry(list.next, struct delayed_iput, list);
2060                 list_del(&delayed->list);
2061                 iput(delayed->inode);
2062                 kfree(delayed);
2063         }
2064         up_read(&root->fs_info->cleanup_work_sem);
2065 }
2066
2067 /*
2068  * calculate extra metadata reservation when snapshotting a subvolume
2069  * contains orphan files.
2070  */
2071 void btrfs_orphan_pre_snapshot(struct btrfs_trans_handle *trans,
2072                                 struct btrfs_pending_snapshot *pending,
2073                                 u64 *bytes_to_reserve)
2074 {
2075         struct btrfs_root *root;
2076         struct btrfs_block_rsv *block_rsv;
2077         u64 num_bytes;
2078         int index;
2079
2080         root = pending->root;
2081         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2082                 return;
2083
2084         block_rsv = root->orphan_block_rsv;
2085
2086         /* orphan block reservation for the snapshot */
2087         num_bytes = block_rsv->size;
2088
2089         /*
2090          * after the snapshot is created, COWing tree blocks may use more
2091          * space than it frees. So we should make sure there is enough
2092          * reserved space.
2093          */
2094         index = trans->transid & 0x1;
2095         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2096                 num_bytes += block_rsv->size -
2097                              (block_rsv->reserved + block_rsv->freed[index]);
2098         }
2099
2100         *bytes_to_reserve += num_bytes;
2101 }
2102
2103 void btrfs_orphan_post_snapshot(struct btrfs_trans_handle *trans,
2104                                 struct btrfs_pending_snapshot *pending)
2105 {
2106         struct btrfs_root *root = pending->root;
2107         struct btrfs_root *snap = pending->snap;
2108         struct btrfs_block_rsv *block_rsv;
2109         u64 num_bytes;
2110         int index;
2111         int ret;
2112
2113         if (!root->orphan_block_rsv || list_empty(&root->orphan_list))
2114                 return;
2115
2116         /* refill source subvolume's orphan block reservation */
2117         block_rsv = root->orphan_block_rsv;
2118         index = trans->transid & 0x1;
2119         if (block_rsv->reserved + block_rsv->freed[index] < block_rsv->size) {
2120                 num_bytes = block_rsv->size -
2121                             (block_rsv->reserved + block_rsv->freed[index]);
2122                 ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2123                                               root->orphan_block_rsv,
2124                                               num_bytes);
2125                 BUG_ON(ret);
2126         }
2127
2128         /* setup orphan block reservation for the snapshot */
2129         block_rsv = btrfs_alloc_block_rsv(snap);
2130         BUG_ON(!block_rsv);
2131
2132         btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2133         snap->orphan_block_rsv = block_rsv;
2134
2135         num_bytes = root->orphan_block_rsv->size;
2136         ret = btrfs_block_rsv_migrate(&pending->block_rsv,
2137                                       block_rsv, num_bytes);
2138         BUG_ON(ret);
2139
2140 #if 0
2141         /* insert orphan item for the snapshot */
2142         WARN_ON(!root->orphan_item_inserted);
2143         ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2144                                        snap->root_key.objectid);
2145         BUG_ON(ret);
2146         snap->orphan_item_inserted = 1;
2147 #endif
2148 }
2149
2150 enum btrfs_orphan_cleanup_state {
2151         ORPHAN_CLEANUP_STARTED  = 1,
2152         ORPHAN_CLEANUP_DONE     = 2,
2153 };
2154
2155 /*
2156  * This is called in transaction commmit time. If there are no orphan
2157  * files in the subvolume, it removes orphan item and frees block_rsv
2158  * structure.
2159  */
2160 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2161                               struct btrfs_root *root)
2162 {
2163         int ret;
2164
2165         if (!list_empty(&root->orphan_list) ||
2166             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2167                 return;
2168
2169         if (root->orphan_item_inserted &&
2170             btrfs_root_refs(&root->root_item) > 0) {
2171                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2172                                             root->root_key.objectid);
2173                 BUG_ON(ret);
2174                 root->orphan_item_inserted = 0;
2175         }
2176
2177         if (root->orphan_block_rsv) {
2178                 WARN_ON(root->orphan_block_rsv->size > 0);
2179                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2180                 root->orphan_block_rsv = NULL;
2181         }
2182 }
2183
2184 /*
2185  * This creates an orphan entry for the given inode in case something goes
2186  * wrong in the middle of an unlink/truncate.
2187  *
2188  * NOTE: caller of this function should reserve 5 units of metadata for
2189  *       this function.
2190  */
2191 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2192 {
2193         struct btrfs_root *root = BTRFS_I(inode)->root;
2194         struct btrfs_block_rsv *block_rsv = NULL;
2195         int reserve = 0;
2196         int insert = 0;
2197         int ret;
2198
2199         if (!root->orphan_block_rsv) {
2200                 block_rsv = btrfs_alloc_block_rsv(root);
2201                 BUG_ON(!block_rsv);
2202         }
2203
2204         spin_lock(&root->orphan_lock);
2205         if (!root->orphan_block_rsv) {
2206                 root->orphan_block_rsv = block_rsv;
2207         } else if (block_rsv) {
2208                 btrfs_free_block_rsv(root, block_rsv);
2209                 block_rsv = NULL;
2210         }
2211
2212         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2213                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2214 #if 0
2215                 /*
2216                  * For proper ENOSPC handling, we should do orphan
2217                  * cleanup when mounting. But this introduces backward
2218                  * compatibility issue.
2219                  */
2220                 if (!xchg(&root->orphan_item_inserted, 1))
2221                         insert = 2;
2222                 else
2223                         insert = 1;
2224 #endif
2225                 insert = 1;
2226         }
2227
2228         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2229                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2230                 reserve = 1;
2231         }
2232         spin_unlock(&root->orphan_lock);
2233
2234         if (block_rsv)
2235                 btrfs_add_durable_block_rsv(root->fs_info, block_rsv);
2236
2237         /* grab metadata reservation from transaction handle */
2238         if (reserve) {
2239                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2240                 BUG_ON(ret);
2241         }
2242
2243         /* insert an orphan item to track this unlinked/truncated file */
2244         if (insert >= 1) {
2245                 ret = btrfs_insert_orphan_item(trans, root, inode->i_ino);
2246                 BUG_ON(ret);
2247         }
2248
2249         /* insert an orphan item to track subvolume contains orphan files */
2250         if (insert >= 2) {
2251                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2252                                                root->root_key.objectid);
2253                 BUG_ON(ret);
2254         }
2255         return 0;
2256 }
2257
2258 /*
2259  * We have done the truncate/delete so we can go ahead and remove the orphan
2260  * item for this particular inode.
2261  */
2262 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2263 {
2264         struct btrfs_root *root = BTRFS_I(inode)->root;
2265         int delete_item = 0;
2266         int release_rsv = 0;
2267         int ret = 0;
2268
2269         spin_lock(&root->orphan_lock);
2270         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2271                 list_del_init(&BTRFS_I(inode)->i_orphan);
2272                 delete_item = 1;
2273         }
2274
2275         if (BTRFS_I(inode)->orphan_meta_reserved) {
2276                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2277                 release_rsv = 1;
2278         }
2279         spin_unlock(&root->orphan_lock);
2280
2281         if (trans && delete_item) {
2282                 ret = btrfs_del_orphan_item(trans, root, inode->i_ino);
2283                 BUG_ON(ret);
2284         }
2285
2286         if (release_rsv)
2287                 btrfs_orphan_release_metadata(inode);
2288
2289         return 0;
2290 }
2291
2292 /*
2293  * this cleans up any orphans that may be left on the list from the last use
2294  * of this root.
2295  */
2296 int btrfs_orphan_cleanup(struct btrfs_root *root)
2297 {
2298         struct btrfs_path *path;
2299         struct extent_buffer *leaf;
2300         struct btrfs_key key, found_key;
2301         struct btrfs_trans_handle *trans;
2302         struct inode *inode;
2303         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2304
2305         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2306                 return 0;
2307
2308         path = btrfs_alloc_path();
2309         if (!path) {
2310                 ret = -ENOMEM;
2311                 goto out;
2312         }
2313         path->reada = -1;
2314
2315         key.objectid = BTRFS_ORPHAN_OBJECTID;
2316         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2317         key.offset = (u64)-1;
2318
2319         while (1) {
2320                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2321                 if (ret < 0)
2322                         goto out;
2323
2324                 /*
2325                  * if ret == 0 means we found what we were searching for, which
2326                  * is weird, but possible, so only screw with path if we didnt
2327                  * find the key and see if we have stuff that matches
2328                  */
2329                 if (ret > 0) {
2330                         ret = 0;
2331                         if (path->slots[0] == 0)
2332                                 break;
2333                         path->slots[0]--;
2334                 }
2335
2336                 /* pull out the item */
2337                 leaf = path->nodes[0];
2338                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2339
2340                 /* make sure the item matches what we want */
2341                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2342                         break;
2343                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2344                         break;
2345
2346                 /* release the path since we're done with it */
2347                 btrfs_release_path(root, path);
2348
2349                 /*
2350                  * this is where we are basically btrfs_lookup, without the
2351                  * crossing root thing.  we store the inode number in the
2352                  * offset of the orphan item.
2353                  */
2354                 found_key.objectid = found_key.offset;
2355                 found_key.type = BTRFS_INODE_ITEM_KEY;
2356                 found_key.offset = 0;
2357                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2358                 if (IS_ERR(inode)) {
2359                         ret = PTR_ERR(inode);
2360                         goto out;
2361                 }
2362
2363                 /*
2364                  * add this inode to the orphan list so btrfs_orphan_del does
2365                  * the proper thing when we hit it
2366                  */
2367                 spin_lock(&root->orphan_lock);
2368                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2369                 spin_unlock(&root->orphan_lock);
2370
2371                 /*
2372                  * if this is a bad inode, means we actually succeeded in
2373                  * removing the inode, but not the orphan record, which means
2374                  * we need to manually delete the orphan since iput will just
2375                  * do a destroy_inode
2376                  */
2377                 if (is_bad_inode(inode)) {
2378                         trans = btrfs_start_transaction(root, 0);
2379                         if (IS_ERR(trans)) {
2380                                 ret = PTR_ERR(trans);
2381                                 goto out;
2382                         }
2383                         btrfs_orphan_del(trans, inode);
2384                         btrfs_end_transaction(trans, root);
2385                         iput(inode);
2386                         continue;
2387                 }
2388
2389                 /* if we have links, this was a truncate, lets do that */
2390                 if (inode->i_nlink) {
2391                         if (!S_ISREG(inode->i_mode)) {
2392                                 WARN_ON(1);
2393                                 iput(inode);
2394                                 continue;
2395                         }
2396                         nr_truncate++;
2397                         ret = btrfs_truncate(inode);
2398                 } else {
2399                         nr_unlink++;
2400                 }
2401
2402                 /* this will do delete_inode and everything for us */
2403                 iput(inode);
2404                 if (ret)
2405                         goto out;
2406         }
2407         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2408
2409         if (root->orphan_block_rsv)
2410                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2411                                         (u64)-1);
2412
2413         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2414                 trans = btrfs_join_transaction(root, 1);
2415                 if (!IS_ERR(trans))
2416                         btrfs_end_transaction(trans, root);
2417         }
2418
2419         if (nr_unlink)
2420                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2421         if (nr_truncate)
2422                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2423
2424 out:
2425         if (ret)
2426                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2427         btrfs_free_path(path);
2428         return ret;
2429 }
2430
2431 /*
2432  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2433  * don't find any xattrs, we know there can't be any acls.
2434  *
2435  * slot is the slot the inode is in, objectid is the objectid of the inode
2436  */
2437 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2438                                           int slot, u64 objectid)
2439 {
2440         u32 nritems = btrfs_header_nritems(leaf);
2441         struct btrfs_key found_key;
2442         int scanned = 0;
2443
2444         slot++;
2445         while (slot < nritems) {
2446                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2447
2448                 /* we found a different objectid, there must not be acls */
2449                 if (found_key.objectid != objectid)
2450                         return 0;
2451
2452                 /* we found an xattr, assume we've got an acl */
2453                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2454                         return 1;
2455
2456                 /*
2457                  * we found a key greater than an xattr key, there can't
2458                  * be any acls later on
2459                  */
2460                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2461                         return 0;
2462
2463                 slot++;
2464                 scanned++;
2465
2466                 /*
2467                  * it goes inode, inode backrefs, xattrs, extents,
2468                  * so if there are a ton of hard links to an inode there can
2469                  * be a lot of backrefs.  Don't waste time searching too hard,
2470                  * this is just an optimization
2471                  */
2472                 if (scanned >= 8)
2473                         break;
2474         }
2475         /* we hit the end of the leaf before we found an xattr or
2476          * something larger than an xattr.  We have to assume the inode
2477          * has acls
2478          */
2479         return 1;
2480 }
2481
2482 /*
2483  * read an inode from the btree into the in-memory inode
2484  */
2485 static void btrfs_read_locked_inode(struct inode *inode)
2486 {
2487         struct btrfs_path *path;
2488         struct extent_buffer *leaf;
2489         struct btrfs_inode_item *inode_item;
2490         struct btrfs_timespec *tspec;
2491         struct btrfs_root *root = BTRFS_I(inode)->root;
2492         struct btrfs_key location;
2493         int maybe_acls;
2494         u64 alloc_group_block;
2495         u32 rdev;
2496         int ret;
2497
2498         path = btrfs_alloc_path();
2499         BUG_ON(!path);
2500         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2501
2502         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2503         if (ret)
2504                 goto make_bad;
2505
2506         leaf = path->nodes[0];
2507         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2508                                     struct btrfs_inode_item);
2509
2510         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2511         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2512         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2513         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2514         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2515
2516         tspec = btrfs_inode_atime(inode_item);
2517         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2518         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2519
2520         tspec = btrfs_inode_mtime(inode_item);
2521         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2522         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2523
2524         tspec = btrfs_inode_ctime(inode_item);
2525         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2526         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2527
2528         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2529         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2530         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2531         inode->i_generation = BTRFS_I(inode)->generation;
2532         inode->i_rdev = 0;
2533         rdev = btrfs_inode_rdev(leaf, inode_item);
2534
2535         BTRFS_I(inode)->index_cnt = (u64)-1;
2536         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2537
2538         alloc_group_block = btrfs_inode_block_group(leaf, inode_item);
2539
2540         /*
2541          * try to precache a NULL acl entry for files that don't have
2542          * any xattrs or acls
2543          */
2544         maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
2545         if (!maybe_acls)
2546                 cache_no_acl(inode);
2547
2548         BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
2549                                                 alloc_group_block, 0);
2550         btrfs_free_path(path);
2551         inode_item = NULL;
2552
2553         switch (inode->i_mode & S_IFMT) {
2554         case S_IFREG:
2555                 inode->i_mapping->a_ops = &btrfs_aops;
2556                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2557                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2558                 inode->i_fop = &btrfs_file_operations;
2559                 inode->i_op = &btrfs_file_inode_operations;
2560                 break;
2561         case S_IFDIR:
2562                 inode->i_fop = &btrfs_dir_file_operations;
2563                 if (root == root->fs_info->tree_root)
2564                         inode->i_op = &btrfs_dir_ro_inode_operations;
2565                 else
2566                         inode->i_op = &btrfs_dir_inode_operations;
2567                 break;
2568         case S_IFLNK:
2569                 inode->i_op = &btrfs_symlink_inode_operations;
2570                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2571                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2572                 break;
2573         default:
2574                 inode->i_op = &btrfs_special_inode_operations;
2575                 init_special_inode(inode, inode->i_mode, rdev);
2576                 break;
2577         }
2578
2579         btrfs_update_iflags(inode);
2580         return;
2581
2582 make_bad:
2583         btrfs_free_path(path);
2584         make_bad_inode(inode);
2585 }
2586
2587 /*
2588  * given a leaf and an inode, copy the inode fields into the leaf
2589  */
2590 static void fill_inode_item(struct btrfs_trans_handle *trans,
2591                             struct extent_buffer *leaf,
2592                             struct btrfs_inode_item *item,
2593                             struct inode *inode)
2594 {
2595         if (!leaf->map_token)
2596                 map_private_extent_buffer(leaf, (unsigned long)item,
2597                                           sizeof(struct btrfs_inode_item),
2598                                           &leaf->map_token, &leaf->kaddr,
2599                                           &leaf->map_start, &leaf->map_len,
2600                                           KM_USER1);
2601
2602         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2603         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2604         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2605         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2606         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2607
2608         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2609                                inode->i_atime.tv_sec);
2610         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2611                                 inode->i_atime.tv_nsec);
2612
2613         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2614                                inode->i_mtime.tv_sec);
2615         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2616                                 inode->i_mtime.tv_nsec);
2617
2618         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2619                                inode->i_ctime.tv_sec);
2620         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2621                                 inode->i_ctime.tv_nsec);
2622
2623         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2624         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2625         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2626         btrfs_set_inode_transid(leaf, item, trans->transid);
2627         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2628         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2629         btrfs_set_inode_block_group(leaf, item, BTRFS_I(inode)->block_group);
2630
2631         if (leaf->map_token) {
2632                 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
2633                 leaf->map_token = NULL;
2634         }
2635 }
2636
2637 /*
2638  * copy everything in the in-memory inode into the btree.
2639  */
2640 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2641                                 struct btrfs_root *root, struct inode *inode)
2642 {
2643         struct btrfs_inode_item *inode_item;
2644         struct btrfs_path *path;
2645         struct extent_buffer *leaf;
2646         int ret;
2647
2648         path = btrfs_alloc_path();
2649         BUG_ON(!path);
2650         path->leave_spinning = 1;
2651         ret = btrfs_lookup_inode(trans, root, path,
2652                                  &BTRFS_I(inode)->location, 1);
2653         if (ret) {
2654                 if (ret > 0)
2655                         ret = -ENOENT;
2656                 goto failed;
2657         }
2658
2659         btrfs_unlock_up_safe(path, 1);
2660         leaf = path->nodes[0];
2661         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2662                                   struct btrfs_inode_item);
2663
2664         fill_inode_item(trans, leaf, inode_item, inode);
2665         btrfs_mark_buffer_dirty(leaf);
2666         btrfs_set_inode_last_trans(trans, inode);
2667         ret = 0;
2668 failed:
2669         btrfs_free_path(path);
2670         return ret;
2671 }
2672
2673
2674 /*
2675  * unlink helper that gets used here in inode.c and in the tree logging
2676  * recovery code.  It remove a link in a directory with a given name, and
2677  * also drops the back refs in the inode to the directory
2678  */
2679 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2680                                 struct btrfs_root *root,
2681                                 struct inode *dir, struct inode *inode,
2682                                 const char *name, int name_len)
2683 {
2684         struct btrfs_path *path;
2685         int ret = 0;
2686         struct extent_buffer *leaf;
2687         struct btrfs_dir_item *di;
2688         struct btrfs_key key;
2689         u64 index;
2690
2691         path = btrfs_alloc_path();
2692         if (!path) {
2693                 ret = -ENOMEM;
2694                 goto out;
2695         }
2696
2697         path->leave_spinning = 1;
2698         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2699                                     name, name_len, -1);
2700         if (IS_ERR(di)) {
2701                 ret = PTR_ERR(di);
2702                 goto err;
2703         }
2704         if (!di) {
2705                 ret = -ENOENT;
2706                 goto err;
2707         }
2708         leaf = path->nodes[0];
2709         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2710         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2711         if (ret)
2712                 goto err;
2713         btrfs_release_path(root, path);
2714
2715         ret = btrfs_del_inode_ref(trans, root, name, name_len,
2716                                   inode->i_ino,
2717                                   dir->i_ino, &index);
2718         if (ret) {
2719                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2720                        "inode %lu parent %lu\n", name_len, name,
2721                        inode->i_ino, dir->i_ino);
2722                 goto err;
2723         }
2724
2725         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
2726                                          index, name, name_len, -1);
2727         if (IS_ERR(di)) {
2728                 ret = PTR_ERR(di);
2729                 goto err;
2730         }
2731         if (!di) {
2732                 ret = -ENOENT;
2733                 goto err;
2734         }
2735         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2736         btrfs_release_path(root, path);
2737
2738         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2739                                          inode, dir->i_ino);
2740         BUG_ON(ret != 0 && ret != -ENOENT);
2741
2742         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2743                                            dir, index);
2744         if (ret == -ENOENT)
2745                 ret = 0;
2746 err:
2747         btrfs_free_path(path);
2748         if (ret)
2749                 goto out;
2750
2751         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2752         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2753         btrfs_update_inode(trans, root, dir);
2754 out:
2755         return ret;
2756 }
2757
2758 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2759                        struct btrfs_root *root,
2760                        struct inode *dir, struct inode *inode,
2761                        const char *name, int name_len)
2762 {
2763         int ret;
2764         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2765         if (!ret) {
2766                 btrfs_drop_nlink(inode);
2767                 ret = btrfs_update_inode(trans, root, inode);
2768         }
2769         return ret;
2770 }
2771
2772
2773 /* helper to check if there is any shared block in the path */
2774 static int check_path_shared(struct btrfs_root *root,
2775                              struct btrfs_path *path)
2776 {
2777         struct extent_buffer *eb;
2778         int level;
2779         u64 refs = 1;
2780
2781         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2782                 int ret;
2783
2784                 if (!path->nodes[level])
2785                         break;
2786                 eb = path->nodes[level];
2787                 if (!btrfs_block_can_be_shared(root, eb))
2788                         continue;
2789                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2790                                                &refs, NULL);
2791                 if (refs > 1)
2792                         return 1;
2793         }
2794         return 0;
2795 }
2796
2797 /*
2798  * helper to start transaction for unlink and rmdir.
2799  *
2800  * unlink and rmdir are special in btrfs, they do not always free space.
2801  * so in enospc case, we should make sure they will free space before
2802  * allowing them to use the global metadata reservation.
2803  */
2804 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2805                                                        struct dentry *dentry)
2806 {
2807         struct btrfs_trans_handle *trans;
2808         struct btrfs_root *root = BTRFS_I(dir)->root;
2809         struct btrfs_path *path;
2810         struct btrfs_inode_ref *ref;
2811         struct btrfs_dir_item *di;
2812         struct inode *inode = dentry->d_inode;
2813         u64 index;
2814         int check_link = 1;
2815         int err = -ENOSPC;
2816         int ret;
2817
2818         trans = btrfs_start_transaction(root, 10);
2819         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2820                 return trans;
2821
2822         if (inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2823                 return ERR_PTR(-ENOSPC);
2824
2825         /* check if there is someone else holds reference */
2826         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2827                 return ERR_PTR(-ENOSPC);
2828
2829         if (atomic_read(&inode->i_count) > 2)
2830                 return ERR_PTR(-ENOSPC);
2831
2832         if (xchg(&root->fs_info->enospc_unlink, 1))
2833                 return ERR_PTR(-ENOSPC);
2834
2835         path = btrfs_alloc_path();
2836         if (!path) {
2837                 root->fs_info->enospc_unlink = 0;
2838                 return ERR_PTR(-ENOMEM);
2839         }
2840
2841         trans = btrfs_start_transaction(root, 0);
2842         if (IS_ERR(trans)) {
2843                 btrfs_free_path(path);
2844                 root->fs_info->enospc_unlink = 0;
2845                 return trans;
2846         }
2847
2848         path->skip_locking = 1;
2849         path->search_commit_root = 1;
2850
2851         ret = btrfs_lookup_inode(trans, root, path,
2852                                 &BTRFS_I(dir)->location, 0);
2853         if (ret < 0) {
2854                 err = ret;
2855                 goto out;
2856         }
2857         if (ret == 0) {
2858                 if (check_path_shared(root, path))
2859                         goto out;
2860         } else {
2861                 check_link = 0;
2862         }
2863         btrfs_release_path(root, path);
2864
2865         ret = btrfs_lookup_inode(trans, root, path,
2866                                 &BTRFS_I(inode)->location, 0);
2867         if (ret < 0) {
2868                 err = ret;
2869                 goto out;
2870         }
2871         if (ret == 0) {
2872                 if (check_path_shared(root, path))
2873                         goto out;
2874         } else {
2875                 check_link = 0;
2876         }
2877         btrfs_release_path(root, path);
2878
2879         if (ret == 0 && S_ISREG(inode->i_mode)) {
2880                 ret = btrfs_lookup_file_extent(trans, root, path,
2881                                                inode->i_ino, (u64)-1, 0);
2882                 if (ret < 0) {
2883                         err = ret;
2884                         goto out;
2885                 }
2886                 BUG_ON(ret == 0);
2887                 if (check_path_shared(root, path))
2888                         goto out;
2889                 btrfs_release_path(root, path);
2890         }
2891
2892         if (!check_link) {
2893                 err = 0;
2894                 goto out;
2895         }
2896
2897         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
2898                                 dentry->d_name.name, dentry->d_name.len, 0);
2899         if (IS_ERR(di)) {
2900                 err = PTR_ERR(di);
2901                 goto out;
2902         }
2903         if (di) {
2904                 if (check_path_shared(root, path))
2905                         goto out;
2906         } else {
2907                 err = 0;
2908                 goto out;
2909         }
2910         btrfs_release_path(root, path);
2911
2912         ref = btrfs_lookup_inode_ref(trans, root, path,
2913                                 dentry->d_name.name, dentry->d_name.len,
2914                                 inode->i_ino, dir->i_ino, 0);
2915         if (IS_ERR(ref)) {
2916                 err = PTR_ERR(ref);
2917                 goto out;
2918         }
2919         BUG_ON(!ref);
2920         if (check_path_shared(root, path))
2921                 goto out;
2922         index = btrfs_inode_ref_index(path->nodes[0], ref);
2923         btrfs_release_path(root, path);
2924
2925         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino, index,
2926                                 dentry->d_name.name, dentry->d_name.len, 0);
2927         if (IS_ERR(di)) {
2928                 err = PTR_ERR(di);
2929                 goto out;
2930         }
2931         BUG_ON(ret == -ENOENT);
2932         if (check_path_shared(root, path))
2933                 goto out;
2934
2935         err = 0;
2936 out:
2937         btrfs_free_path(path);
2938         if (err) {
2939                 btrfs_end_transaction(trans, root);
2940                 root->fs_info->enospc_unlink = 0;
2941                 return ERR_PTR(err);
2942         }
2943
2944         trans->block_rsv = &root->fs_info->global_block_rsv;
2945         return trans;
2946 }
2947
2948 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2949                                struct btrfs_root *root)
2950 {
2951         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2952                 BUG_ON(!root->fs_info->enospc_unlink);
2953                 root->fs_info->enospc_unlink = 0;
2954         }
2955         btrfs_end_transaction_throttle(trans, root);
2956 }
2957
2958 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2959 {
2960         struct btrfs_root *root = BTRFS_I(dir)->root;
2961         struct btrfs_trans_handle *trans;
2962         struct inode *inode = dentry->d_inode;
2963         int ret;
2964         unsigned long nr = 0;
2965
2966         trans = __unlink_start_trans(dir, dentry);
2967         if (IS_ERR(trans))
2968                 return PTR_ERR(trans);
2969
2970         btrfs_set_trans_block_group(trans, dir);
2971
2972         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2973
2974         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2975                                  dentry->d_name.name, dentry->d_name.len);
2976         BUG_ON(ret);
2977
2978         if (inode->i_nlink == 0) {
2979                 ret = btrfs_orphan_add(trans, inode);
2980                 BUG_ON(ret);
2981         }
2982
2983         nr = trans->blocks_used;
2984         __unlink_end_trans(trans, root);
2985         btrfs_btree_balance_dirty(root, nr);
2986         return ret;
2987 }
2988
2989 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2990                         struct btrfs_root *root,
2991                         struct inode *dir, u64 objectid,
2992                         const char *name, int name_len)
2993 {
2994         struct btrfs_path *path;
2995         struct extent_buffer *leaf;
2996         struct btrfs_dir_item *di;
2997         struct btrfs_key key;
2998         u64 index;
2999         int ret;
3000
3001         path = btrfs_alloc_path();
3002         if (!path)
3003                 return -ENOMEM;
3004
3005         di = btrfs_lookup_dir_item(trans, root, path, dir->i_ino,
3006                                    name, name_len, -1);
3007         BUG_ON(!di || IS_ERR(di));
3008
3009         leaf = path->nodes[0];
3010         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3011         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3012         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3013         BUG_ON(ret);
3014         btrfs_release_path(root, path);
3015
3016         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3017                                  objectid, root->root_key.objectid,
3018                                  dir->i_ino, &index, name, name_len);
3019         if (ret < 0) {
3020                 BUG_ON(ret != -ENOENT);
3021                 di = btrfs_search_dir_index_item(root, path, dir->i_ino,
3022                                                  name, name_len);
3023                 BUG_ON(!di || IS_ERR(di));
3024
3025                 leaf = path->nodes[0];
3026                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3027                 btrfs_release_path(root, path);
3028                 index = key.offset;
3029         }
3030
3031         di = btrfs_lookup_dir_index_item(trans, root, path, dir->i_ino,
3032                                          index, name, name_len, -1);
3033         BUG_ON(!di || IS_ERR(di));
3034
3035         leaf = path->nodes[0];
3036         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3037         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3038         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3039         BUG_ON(ret);
3040         btrfs_release_path(root, path);
3041
3042         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3043         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3044         ret = btrfs_update_inode(trans, root, dir);
3045         BUG_ON(ret);
3046
3047         btrfs_free_path(path);
3048         return 0;
3049 }
3050
3051 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3052 {
3053         struct inode *inode = dentry->d_inode;
3054         int err = 0;
3055         struct btrfs_root *root = BTRFS_I(dir)->root;
3056         struct btrfs_trans_handle *trans;
3057         unsigned long nr = 0;
3058
3059         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3060             inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
3061                 return -ENOTEMPTY;
3062
3063         trans = __unlink_start_trans(dir, dentry);
3064         if (IS_ERR(trans))
3065                 return PTR_ERR(trans);
3066
3067         btrfs_set_trans_block_group(trans, dir);
3068
3069         if (unlikely(inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3070                 err = btrfs_unlink_subvol(trans, root, dir,
3071                                           BTRFS_I(inode)->location.objectid,
3072                                           dentry->d_name.name,
3073                                           dentry->d_name.len);
3074                 goto out;
3075         }
3076
3077         err = btrfs_orphan_add(trans, inode);
3078         if (err)
3079                 goto out;
3080
3081         /* now the directory is empty */
3082         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3083                                  dentry->d_name.name, dentry->d_name.len);
3084         if (!err)
3085                 btrfs_i_size_write(inode, 0);
3086 out:
3087         nr = trans->blocks_used;
3088         __unlink_end_trans(trans, root);
3089         btrfs_btree_balance_dirty(root, nr);
3090
3091         return err;
3092 }
3093
3094 #if 0
3095 /*
3096  * when truncating bytes in a file, it is possible to avoid reading
3097  * the leaves that contain only checksum items.  This can be the
3098  * majority of the IO required to delete a large file, but it must
3099  * be done carefully.
3100  *
3101  * The keys in the level just above the leaves are checked to make sure
3102  * the lowest key in a given leaf is a csum key, and starts at an offset
3103  * after the new  size.
3104  *
3105  * Then the key for the next leaf is checked to make sure it also has
3106  * a checksum item for the same file.  If it does, we know our target leaf
3107  * contains only checksum items, and it can be safely freed without reading
3108  * it.
3109  *
3110  * This is just an optimization targeted at large files.  It may do
3111  * nothing.  It will return 0 unless things went badly.
3112  */
3113 static noinline int drop_csum_leaves(struct btrfs_trans_handle *trans,
3114                                      struct btrfs_root *root,
3115                                      struct btrfs_path *path,
3116                                      struct inode *inode, u64 new_size)
3117 {
3118         struct btrfs_key key;
3119         int ret;
3120         int nritems;
3121         struct btrfs_key found_key;
3122         struct btrfs_key other_key;
3123         struct btrfs_leaf_ref *ref;
3124         u64 leaf_gen;
3125         u64 leaf_start;
3126
3127         path->lowest_level = 1;
3128         key.objectid = inode->i_ino;
3129         key.type = BTRFS_CSUM_ITEM_KEY;
3130         key.offset = new_size;
3131 again:
3132         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3133         if (ret < 0)
3134                 goto out;
3135
3136         if (path->nodes[1] == NULL) {
3137                 ret = 0;
3138                 goto out;
3139         }
3140         ret = 0;
3141         btrfs_node_key_to_cpu(path->nodes[1], &found_key, path->slots[1]);
3142         nritems = btrfs_header_nritems(path->nodes[1]);
3143
3144         if (!nritems)
3145                 goto out;
3146
3147         if (path->slots[1] >= nritems)
3148                 goto next_node;
3149
3150         /* did we find a key greater than anything we want to delete? */
3151         if (found_key.objectid > inode->i_ino ||
3152            (found_key.objectid == inode->i_ino && found_key.type > key.type))
3153                 goto out;
3154
3155         /* we check the next key in the node to make sure the leave contains
3156          * only checksum items.  This comparison doesn't work if our
3157          * leaf is the last one in the node
3158          */
3159         if (path->slots[1] + 1 >= nritems) {
3160 next_node:
3161                 /* search forward from the last key in the node, this
3162                  * will bring us into the next node in the tree
3163                  */
3164                 btrfs_node_key_to_cpu(path->nodes[1], &found_key, nritems - 1);
3165
3166                 /* unlikely, but we inc below, so check to be safe */
3167                 if (found_key.offset == (u64)-1)
3168                         goto out;
3169
3170                 /* search_forward needs a path with locks held, do the
3171                  * search again for the original key.  It is possible
3172                  * this will race with a balance and return a path that
3173                  * we could modify, but this drop is just an optimization
3174                  * and is allowed to miss some leaves.
3175                  */
3176                 btrfs_release_path(root, path);
3177                 found_key.offset++;
3178
3179                 /* setup a max key for search_forward */
3180                 other_key.offset = (u64)-1;
3181                 other_key.type = key.type;
3182                 other_key.objectid = key.objectid;
3183
3184                 path->keep_locks = 1;
3185                 ret = btrfs_search_forward(root, &found_key, &other_key,
3186                                            path, 0, 0);
3187                 path->keep_locks = 0;
3188                 if (ret || found_key.objectid != key.objectid ||
3189                     found_key.type != key.type) {
3190                         ret = 0;
3191                         goto out;
3192                 }
3193
3194                 key.offset = found_key.offset;
3195                 btrfs_release_path(root, path);
3196                 cond_resched();
3197                 goto again;
3198         }
3199
3200         /* we know there's one more slot after us in the tree,
3201          * read that key so we can verify it is also a checksum item
3202          */
3203         btrfs_node_key_to_cpu(path->nodes[1], &other_key, path->slots[1] + 1);
3204
3205         if (found_key.objectid < inode->i_ino)
3206                 goto next_key;
3207
3208         if (found_key.type != key.type || found_key.offset < new_size)
3209                 goto next_key;
3210
3211         /*
3212          * if the key for the next leaf isn't a csum key from this objectid,
3213          * we can't be sure there aren't good items inside this leaf.
3214          * Bail out
3215          */
3216         if (other_key.objectid != inode->i_ino || other_key.type != key.type)
3217                 goto out;
3218
3219         leaf_start = btrfs_node_blockptr(path->nodes[1], path->slots[1]);
3220         leaf_gen = btrfs_node_ptr_generation(path->nodes[1], path->slots[1]);
3221         /*
3222          * it is safe to delete this leaf, it contains only
3223          * csum items from this inode at an offset >= new_size
3224          */
3225         ret = btrfs_del_leaf(trans, root, path, leaf_start);
3226         BUG_ON(ret);
3227
3228         if (root->ref_cows && leaf_gen < trans->transid) {
3229                 ref = btrfs_alloc_leaf_ref(root, 0);
3230                 if (ref) {
3231                         ref->root_gen = root->root_key.offset;
3232                         ref->bytenr = leaf_start;
3233                         ref->owner = 0;
3234                         ref->generation = leaf_gen;
3235                         ref->nritems = 0;
3236
3237                         btrfs_sort_leaf_ref(ref);
3238
3239                         ret = btrfs_add_leaf_ref(root, ref, 0);
3240                         WARN_ON(ret);
3241                         btrfs_free_leaf_ref(root, ref);
3242                 } else {
3243                         WARN_ON(1);
3244                 }
3245         }
3246 next_key:
3247         btrfs_release_path(root, path);
3248
3249         if (other_key.objectid == inode->i_ino &&
3250             other_key.type == key.type && other_key.offset > key.offset) {
3251                 key.offset = other_key.offset;
3252                 cond_resched();
3253                 goto again;
3254         }
3255         ret = 0;
3256 out:
3257         /* fixup any changes we've made to the path */
3258         path->lowest_level = 0;
3259         path->keep_locks = 0;
3260         btrfs_release_path(root, path);
3261         return ret;
3262 }
3263
3264 #endif
3265
3266 /*
3267  * this can truncate away extent items, csum items and directory items.
3268  * It starts at a high offset and removes keys until it can't find
3269  * any higher than new_size
3270  *
3271  * csum items that cross the new i_size are truncated to the new size
3272  * as well.
3273  *
3274  * min_type is the minimum key type to truncate down to.  If set to 0, this
3275  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3276  */
3277 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3278                                struct btrfs_root *root,
3279                                struct inode *inode,
3280                                u64 new_size, u32 min_type)
3281 {
3282         struct btrfs_path *path;
3283         struct extent_buffer *leaf;
3284         struct btrfs_file_extent_item *fi;
3285         struct btrfs_key key;
3286         struct btrfs_key found_key;
3287         u64 extent_start = 0;
3288         u64 extent_num_bytes = 0;
3289         u64 extent_offset = 0;
3290         u64 item_end = 0;
3291         u64 mask = root->sectorsize - 1;
3292         u32 found_type = (u8)-1;
3293         int found_extent;
3294         int del_item;
3295         int pending_del_nr = 0;
3296         int pending_del_slot = 0;
3297         int extent_type = -1;
3298         int encoding;
3299         int ret;
3300         int err = 0;
3301
3302         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3303
3304         if (root->ref_cows || root == root->fs_info->tree_root)
3305                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3306
3307         path = btrfs_alloc_path();
3308         BUG_ON(!path);
3309         path->reada = -1;
3310
3311         key.objectid = inode->i_ino;
3312         key.offset = (u64)-1;
3313         key.type = (u8)-1;
3314
3315 search_again:
3316         path->leave_spinning = 1;
3317         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3318         if (ret < 0) {
3319                 err = ret;
3320                 goto out;
3321         }
3322
3323         if (ret > 0) {
3324                 /* there are no items in the tree for us to truncate, we're
3325                  * done
3326                  */
3327                 if (path->slots[0] == 0)
3328                         goto out;
3329                 path->slots[0]--;
3330         }
3331
3332         while (1) {
3333                 fi = NULL;
3334                 leaf = path->nodes[0];
3335                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3336                 found_type = btrfs_key_type(&found_key);
3337                 encoding = 0;
3338
3339                 if (found_key.objectid != inode->i_ino)
3340                         break;
3341
3342                 if (found_type < min_type)
3343                         break;
3344
3345                 item_end = found_key.offset;
3346                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3347                         fi = btrfs_item_ptr(leaf, path->slots[0],
3348                                             struct btrfs_file_extent_item);
3349                         extent_type = btrfs_file_extent_type(leaf, fi);
3350                         encoding = btrfs_file_extent_compression(leaf, fi);
3351                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3352                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3353
3354                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3355                                 item_end +=
3356                                     btrfs_file_extent_num_bytes(leaf, fi);
3357                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3358                                 item_end += btrfs_file_extent_inline_len(leaf,
3359                                                                          fi);
3360                         }
3361                         item_end--;
3362                 }
3363                 if (found_type > min_type) {
3364                         del_item = 1;
3365                 } else {
3366                         if (item_end < new_size)
3367                                 break;
3368                         if (found_key.offset >= new_size)
3369                                 del_item = 1;
3370                         else
3371                                 del_item = 0;
3372                 }
3373                 found_extent = 0;
3374                 /* FIXME, shrink the extent if the ref count is only 1 */
3375                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3376                         goto delete;
3377
3378                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3379                         u64 num_dec;
3380                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3381                         if (!del_item && !encoding) {
3382                                 u64 orig_num_bytes =
3383                                         btrfs_file_extent_num_bytes(leaf, fi);
3384                                 extent_num_bytes = new_size -
3385                                         found_key.offset + root->sectorsize - 1;
3386                                 extent_num_bytes = extent_num_bytes &
3387                                         ~((u64)root->sectorsize - 1);
3388                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3389                                                          extent_num_bytes);
3390                                 num_dec = (orig_num_bytes -
3391                                            extent_num_bytes);
3392                                 if (root->ref_cows && extent_start != 0)
3393                                         inode_sub_bytes(inode, num_dec);
3394                                 btrfs_mark_buffer_dirty(leaf);
3395                         } else {
3396                                 extent_num_bytes =
3397                                         btrfs_file_extent_disk_num_bytes(leaf,
3398                                                                          fi);
3399                                 extent_offset = found_key.offset -
3400                                         btrfs_file_extent_offset(leaf, fi);
3401
3402                                 /* FIXME blocksize != 4096 */
3403                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3404                                 if (extent_start != 0) {
3405                                         found_extent = 1;
3406                                         if (root->ref_cows)
3407                                                 inode_sub_bytes(inode, num_dec);
3408                                 }
3409                         }
3410                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3411                         /*
3412                          * we can't truncate inline items that have had
3413                          * special encodings
3414                          */
3415                         if (!del_item &&
3416                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3417                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3418                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3419                                 u32 size = new_size - found_key.offset;
3420
3421                                 if (root->ref_cows) {
3422                                         inode_sub_bytes(inode, item_end + 1 -
3423                                                         new_size);
3424                                 }
3425                                 size =
3426                                     btrfs_file_extent_calc_inline_size(size);
3427                                 ret = btrfs_truncate_item(trans, root, path,
3428                                                           size, 1);
3429                                 BUG_ON(ret);
3430                         } else if (root->ref_cows) {
3431                                 inode_sub_bytes(inode, item_end + 1 -
3432                                                 found_key.offset);
3433                         }
3434                 }
3435 delete:
3436                 if (del_item) {
3437                         if (!pending_del_nr) {
3438                                 /* no pending yet, add ourselves */
3439                                 pending_del_slot = path->slots[0];
3440                                 pending_del_nr = 1;
3441                         } else if (pending_del_nr &&
3442                                    path->slots[0] + 1 == pending_del_slot) {
3443                                 /* hop on the pending chunk */
3444                                 pending_del_nr++;
3445                                 pending_del_slot = path->slots[0];
3446                         } else {
3447                                 BUG();
3448                         }
3449                 } else {
3450                         break;
3451                 }
3452                 if (found_extent && (root->ref_cows ||
3453                                      root == root->fs_info->tree_root)) {
3454                         btrfs_set_path_blocking(path);
3455                         ret = btrfs_free_extent(trans, root, extent_start,
3456                                                 extent_num_bytes, 0,
3457                                                 btrfs_header_owner(leaf),
3458                                                 inode->i_ino, extent_offset);
3459                         BUG_ON(ret);
3460                 }
3461
3462                 if (found_type == BTRFS_INODE_ITEM_KEY)
3463                         break;
3464
3465                 if (path->slots[0] == 0 ||
3466                     path->slots[0] != pending_del_slot) {
3467                         if (root->ref_cows) {
3468                                 err = -EAGAIN;
3469                                 goto out;
3470                         }
3471                         if (pending_del_nr) {
3472                                 ret = btrfs_del_items(trans, root, path,
3473                                                 pending_del_slot,
3474                                                 pending_del_nr);
3475                                 BUG_ON(ret);
3476                                 pending_del_nr = 0;
3477                         }
3478                         btrfs_release_path(root, path);
3479                         goto search_again;
3480                 } else {
3481                         path->slots[0]--;
3482                 }
3483         }
3484 out:
3485         if (pending_del_nr) {
3486                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3487                                       pending_del_nr);
3488                 BUG_ON(ret);
3489         }
3490         btrfs_free_path(path);
3491         return err;
3492 }
3493
3494 /*
3495  * taken from block_truncate_page, but does cow as it zeros out
3496  * any bytes left in the last page in the file.
3497  */
3498 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3499 {
3500         struct inode *inode = mapping->host;
3501         struct btrfs_root *root = BTRFS_I(inode)->root;
3502         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3503         struct btrfs_ordered_extent *ordered;
3504         struct extent_state *cached_state = NULL;
3505         char *kaddr;
3506         u32 blocksize = root->sectorsize;
3507         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3508         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3509         struct page *page;
3510         int ret = 0;
3511         u64 page_start;
3512         u64 page_end;
3513
3514         if ((offset & (blocksize - 1)) == 0)
3515                 goto out;
3516         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3517         if (ret)
3518                 goto out;
3519
3520         ret = -ENOMEM;
3521 again:
3522         page = grab_cache_page(mapping, index);
3523         if (!page) {
3524                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3525                 goto out;
3526         }
3527
3528         page_start = page_offset(page);
3529         page_end = page_start + PAGE_CACHE_SIZE - 1;
3530
3531         if (!PageUptodate(page)) {
3532                 ret = btrfs_readpage(NULL, page);
3533                 lock_page(page);
3534                 if (page->mapping != mapping) {
3535                         unlock_page(page);
3536                         page_cache_release(page);
3537                         goto again;
3538                 }
3539                 if (!PageUptodate(page)) {
3540                         ret = -EIO;
3541                         goto out_unlock;
3542                 }
3543         }
3544         wait_on_page_writeback(page);
3545
3546         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3547                          GFP_NOFS);
3548         set_page_extent_mapped(page);
3549
3550         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3551         if (ordered) {
3552                 unlock_extent_cached(io_tree, page_start, page_end,
3553                                      &cached_state, GFP_NOFS);
3554                 unlock_page(page);
3555                 page_cache_release(page);
3556                 btrfs_start_ordered_extent(inode, ordered, 1);
3557                 btrfs_put_ordered_extent(ordered);
3558                 goto again;
3559         }
3560
3561         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3562                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3563                           0, 0, &cached_state, GFP_NOFS);
3564
3565         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3566                                         &cached_state);
3567         if (ret) {
3568                 unlock_extent_cached(io_tree, page_start, page_end,
3569                                      &cached_state, GFP_NOFS);
3570                 goto out_unlock;
3571         }
3572
3573         ret = 0;
3574         if (offset != PAGE_CACHE_SIZE) {
3575                 kaddr = kmap(page);
3576                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3577                 flush_dcache_page(page);
3578                 kunmap(page);
3579         }
3580         ClearPageChecked(page);
3581         set_page_dirty(page);
3582         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3583                              GFP_NOFS);
3584
3585 out_unlock:
3586         if (ret)
3587                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3588         unlock_page(page);
3589         page_cache_release(page);
3590 out:
3591         return ret;
3592 }
3593
3594 /*
3595  * This function puts in dummy file extents for the area we're creating a hole
3596  * for.  So if we are truncating this file to a larger size we need to insert
3597  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3598  * the range between oldsize and size
3599  */
3600 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3601 {
3602         struct btrfs_trans_handle *trans;
3603         struct btrfs_root *root = BTRFS_I(inode)->root;
3604         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3605         struct extent_map *em = NULL;
3606         struct extent_state *cached_state = NULL;
3607         u64 mask = root->sectorsize - 1;
3608         u64 hole_start = (oldsize + mask) & ~mask;
3609         u64 block_end = (size + mask) & ~mask;
3610         u64 last_byte;
3611         u64 cur_offset;
3612         u64 hole_size;
3613         int err = 0;
3614
3615         if (size <= hole_start)
3616                 return 0;
3617
3618         while (1) {
3619                 struct btrfs_ordered_extent *ordered;
3620                 btrfs_wait_ordered_range(inode, hole_start,
3621                                          block_end - hole_start);
3622                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3623                                  &cached_state, GFP_NOFS);
3624                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3625                 if (!ordered)
3626                         break;
3627                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3628                                      &cached_state, GFP_NOFS);
3629                 btrfs_put_ordered_extent(ordered);
3630         }
3631
3632         cur_offset = hole_start;
3633         while (1) {
3634                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3635                                 block_end - cur_offset, 0);
3636                 BUG_ON(IS_ERR(em) || !em);
3637                 last_byte = min(extent_map_end(em), block_end);
3638                 last_byte = (last_byte + mask) & ~mask;
3639                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3640                         u64 hint_byte = 0;
3641                         hole_size = last_byte - cur_offset;
3642
3643                         trans = btrfs_start_transaction(root, 2);
3644                         if (IS_ERR(trans)) {
3645                                 err = PTR_ERR(trans);
3646                                 break;
3647                         }
3648                         btrfs_set_trans_block_group(trans, inode);
3649
3650                         err = btrfs_drop_extents(trans, inode, cur_offset,
3651                                                  cur_offset + hole_size,
3652                                                  &hint_byte, 1);
3653                         if (err)
3654                                 break;
3655
3656                         err = btrfs_insert_file_extent(trans, root,
3657                                         inode->i_ino, cur_offset, 0,
3658                                         0, hole_size, 0, hole_size,
3659                                         0, 0, 0);
3660                         if (err)
3661                                 break;
3662
3663                         btrfs_drop_extent_cache(inode, hole_start,
3664                                         last_byte - 1, 0);
3665
3666                         btrfs_end_transaction(trans, root);
3667                 }
3668                 free_extent_map(em);
3669                 em = NULL;
3670                 cur_offset = last_byte;
3671                 if (cur_offset >= block_end)
3672                         break;
3673         }
3674
3675         free_extent_map(em);
3676         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3677                              GFP_NOFS);
3678         return err;
3679 }
3680
3681 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3682 {
3683         loff_t oldsize = i_size_read(inode);
3684         int ret;
3685
3686         if (newsize == oldsize)
3687                 return 0;
3688
3689         if (newsize > oldsize) {
3690                 i_size_write(inode, newsize);
3691                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3692                 truncate_pagecache(inode, oldsize, newsize);
3693                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3694                 if (ret) {
3695                         btrfs_setsize(inode, oldsize);
3696                         return ret;
3697                 }
3698
3699                 mark_inode_dirty(inode);
3700         } else {
3701
3702                 /*
3703                  * We're truncating a file that used to have good data down to
3704                  * zero. Make sure it gets into the ordered flush list so that
3705                  * any new writes get down to disk quickly.
3706                  */
3707                 if (newsize == 0)
3708                         BTRFS_I(inode)->ordered_data_close = 1;
3709
3710                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3711                 truncate_setsize(inode, newsize);
3712                 ret = btrfs_truncate(inode);
3713         }
3714
3715         return ret;
3716 }
3717
3718 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3719 {
3720         struct inode *inode = dentry->d_inode;
3721         struct btrfs_root *root = BTRFS_I(inode)->root;
3722         int err;
3723
3724         if (btrfs_root_readonly(root))
3725                 return -EROFS;
3726
3727         err = inode_change_ok(inode, attr);
3728         if (err)
3729                 return err;
3730
3731         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3732                 err = btrfs_setsize(inode, attr->ia_size);
3733                 if (err)
3734                         return err;
3735         }
3736
3737         if (attr->ia_valid) {
3738                 setattr_copy(inode, attr);
3739                 mark_inode_dirty(inode);
3740
3741                 if (attr->ia_valid & ATTR_MODE)
3742                         err = btrfs_acl_chmod(inode);
3743         }
3744
3745         return err;
3746 }
3747
3748 void btrfs_evict_inode(struct inode *inode)
3749 {
3750         struct btrfs_trans_handle *trans;
3751         struct btrfs_root *root = BTRFS_I(inode)->root;
3752         unsigned long nr;
3753         int ret;
3754
3755         trace_btrfs_inode_evict(inode);
3756
3757         truncate_inode_pages(&inode->i_data, 0);
3758         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3759                                root == root->fs_info->tree_root))
3760                 goto no_delete;
3761
3762         if (is_bad_inode(inode)) {
3763                 btrfs_orphan_del(NULL, inode);
3764                 goto no_delete;
3765         }
3766         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3767         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3768
3769         if (root->fs_info->log_root_recovering) {
3770                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3771                 goto no_delete;
3772         }
3773
3774         if (inode->i_nlink > 0) {
3775                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3776                 goto no_delete;
3777         }
3778
3779         btrfs_i_size_write(inode, 0);
3780
3781         while (1) {
3782                 trans = btrfs_start_transaction(root, 0);
3783                 BUG_ON(IS_ERR(trans));
3784                 btrfs_set_trans_block_group(trans, inode);
3785                 trans->block_rsv = root->orphan_block_rsv;
3786
3787                 ret = btrfs_block_rsv_check(trans, root,
3788                                             root->orphan_block_rsv, 0, 5);
3789                 if (ret) {
3790                         BUG_ON(ret != -EAGAIN);
3791                         ret = btrfs_commit_transaction(trans, root);
3792                         BUG_ON(ret);
3793                         continue;
3794                 }
3795
3796                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3797                 if (ret != -EAGAIN)
3798                         break;
3799
3800                 nr = trans->blocks_used;
3801                 btrfs_end_transaction(trans, root);
3802                 trans = NULL;
3803                 btrfs_btree_balance_dirty(root, nr);
3804
3805         }
3806
3807         if (ret == 0) {
3808                 ret = btrfs_orphan_del(trans, inode);
3809                 BUG_ON(ret);
3810         }
3811
3812         nr = trans->blocks_used;
3813         btrfs_end_transaction(trans, root);
3814         btrfs_btree_balance_dirty(root, nr);
3815 no_delete:
3816         end_writeback(inode);
3817         return;
3818 }
3819
3820 /*
3821  * this returns the key found in the dir entry in the location pointer.
3822  * If no dir entries were found, location->objectid is 0.
3823  */
3824 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3825                                struct btrfs_key *location)
3826 {
3827         const char *name = dentry->d_name.name;
3828         int namelen = dentry->d_name.len;
3829         struct btrfs_dir_item *di;
3830         struct btrfs_path *path;
3831         struct btrfs_root *root = BTRFS_I(dir)->root;
3832         int ret = 0;
3833
3834         path = btrfs_alloc_path();
3835         BUG_ON(!path);
3836
3837         di = btrfs_lookup_dir_item(NULL, root, path, dir->i_ino, name,
3838                                     namelen, 0);
3839         if (IS_ERR(di))
3840                 ret = PTR_ERR(di);
3841
3842         if (!di || IS_ERR(di))
3843                 goto out_err;
3844
3845         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3846 out:
3847         btrfs_free_path(path);
3848         return ret;
3849 out_err:
3850         location->objectid = 0;
3851         goto out;
3852 }
3853
3854 /*
3855  * when we hit a tree root in a directory, the btrfs part of the inode
3856  * needs to be changed to reflect the root directory of the tree root.  This
3857  * is kind of like crossing a mount point.
3858  */
3859 static int fixup_tree_root_location(struct btrfs_root *root,
3860                                     struct inode *dir,
3861                                     struct dentry *dentry,
3862                                     struct btrfs_key *location,
3863                                     struct btrfs_root **sub_root)
3864 {
3865         struct btrfs_path *path;
3866         struct btrfs_root *new_root;
3867         struct btrfs_root_ref *ref;
3868         struct extent_buffer *leaf;
3869         int ret;
3870         int err = 0;
3871
3872         path = btrfs_alloc_path();
3873         if (!path) {
3874                 err = -ENOMEM;
3875                 goto out;
3876         }
3877
3878         err = -ENOENT;
3879         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3880                                   BTRFS_I(dir)->root->root_key.objectid,
3881                                   location->objectid);
3882         if (ret) {
3883                 if (ret < 0)
3884                         err = ret;
3885                 goto out;
3886         }
3887
3888         leaf = path->nodes[0];
3889         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3890         if (btrfs_root_ref_dirid(leaf, ref) != dir->i_ino ||
3891             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3892                 goto out;
3893
3894         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3895                                    (unsigned long)(ref + 1),
3896                                    dentry->d_name.len);
3897         if (ret)
3898                 goto out;
3899
3900         btrfs_release_path(root->fs_info->tree_root, path);
3901
3902         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3903         if (IS_ERR(new_root)) {
3904                 err = PTR_ERR(new_root);
3905                 goto out;
3906         }
3907
3908         if (btrfs_root_refs(&new_root->root_item) == 0) {
3909                 err = -ENOENT;
3910                 goto out;
3911         }
3912
3913         *sub_root = new_root;
3914         location->objectid = btrfs_root_dirid(&new_root->root_item);
3915         location->type = BTRFS_INODE_ITEM_KEY;
3916         location->offset = 0;
3917         err = 0;
3918 out:
3919         btrfs_free_path(path);
3920         return err;
3921 }
3922
3923 static void inode_tree_add(struct inode *inode)
3924 {
3925         struct btrfs_root *root = BTRFS_I(inode)->root;
3926         struct btrfs_inode *entry;
3927         struct rb_node **p;
3928         struct rb_node *parent;
3929 again:
3930         p = &root->inode_tree.rb_node;
3931         parent = NULL;
3932
3933         if (inode_unhashed(inode))
3934                 return;
3935
3936         spin_lock(&root->inode_lock);
3937         while (*p) {
3938                 parent = *p;
3939                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3940
3941                 if (inode->i_ino < entry->vfs_inode.i_ino)
3942                         p = &parent->rb_left;
3943                 else if (inode->i_ino > entry->vfs_inode.i_ino)
3944                         p = &parent->rb_right;
3945                 else {
3946                         WARN_ON(!(entry->vfs_inode.i_state &
3947                                   (I_WILL_FREE | I_FREEING)));
3948                         rb_erase(parent, &root->inode_tree);
3949                         RB_CLEAR_NODE(parent);
3950                         spin_unlock(&root->inode_lock);
3951                         goto again;
3952                 }
3953         }
3954         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3955         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3956         spin_unlock(&root->inode_lock);
3957 }
3958
3959 static void inode_tree_del(struct inode *inode)
3960 {
3961         struct btrfs_root *root = BTRFS_I(inode)->root;
3962         int empty = 0;
3963
3964         spin_lock(&root->inode_lock);
3965         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3966                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3967                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3968                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3969         }
3970         spin_unlock(&root->inode_lock);
3971
3972         /*
3973          * Free space cache has inodes in the tree root, but the tree root has a
3974          * root_refs of 0, so this could end up dropping the tree root as a
3975          * snapshot, so we need the extra !root->fs_info->tree_root check to
3976          * make sure we don't drop it.
3977          */
3978         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3979             root != root->fs_info->tree_root) {
3980                 synchronize_srcu(&root->fs_info->subvol_srcu);
3981                 spin_lock(&root->inode_lock);
3982                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3983                 spin_unlock(&root->inode_lock);
3984                 if (empty)
3985                         btrfs_add_dead_root(root);
3986         }
3987 }
3988
3989 int btrfs_invalidate_inodes(struct btrfs_root *root)
3990 {
3991         struct rb_node *node;
3992         struct rb_node *prev;
3993         struct btrfs_inode *entry;
3994         struct inode *inode;
3995         u64 objectid = 0;
3996
3997         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3998
3999         spin_lock(&root->inode_lock);
4000 again:
4001         node = root->inode_tree.rb_node;
4002         prev = NULL;
4003         while (node) {
4004                 prev = node;
4005                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4006
4007                 if (objectid < entry->vfs_inode.i_ino)
4008                         node = node->rb_left;
4009                 else if (objectid > entry->vfs_inode.i_ino)
4010                         node = node->rb_right;
4011                 else
4012                         break;
4013         }
4014         if (!node) {
4015                 while (prev) {
4016                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4017                         if (objectid <= entry->vfs_inode.i_ino) {
4018                                 node = prev;
4019                                 break;
4020                         }
4021                         prev = rb_next(prev);
4022                 }
4023         }
4024         while (node) {
4025                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4026                 objectid = entry->vfs_inode.i_ino + 1;
4027                 inode = igrab(&entry->vfs_inode);
4028                 if (inode) {
4029                         spin_unlock(&root->inode_lock);
4030                         if (atomic_read(&inode->i_count) > 1)
4031                                 d_prune_aliases(inode);
4032                         /*
4033                          * btrfs_drop_inode will have it removed from
4034                          * the inode cache when its usage count
4035                          * hits zero.
4036                          */
4037                         iput(inode);
4038                         cond_resched();
4039                         spin_lock(&root->inode_lock);
4040                         goto again;
4041                 }
4042
4043                 if (cond_resched_lock(&root->inode_lock))
4044                         goto again;
4045
4046                 node = rb_next(node);
4047         }
4048         spin_unlock(&root->inode_lock);
4049         return 0;
4050 }
4051
4052 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4053 {
4054         struct btrfs_iget_args *args = p;
4055         inode->i_ino = args->ino;
4056         BTRFS_I(inode)->root = args->root;
4057         btrfs_set_inode_space_info(args->root, inode);
4058         return 0;
4059 }
4060
4061 static int btrfs_find_actor(struct inode *inode, void *opaque)
4062 {
4063         struct btrfs_iget_args *args = opaque;
4064         return args->ino == inode->i_ino &&
4065                 args->root == BTRFS_I(inode)->root;
4066 }
4067
4068 static struct inode *btrfs_iget_locked(struct super_block *s,
4069                                        u64 objectid,
4070                                        struct btrfs_root *root)
4071 {
4072         struct inode *inode;
4073         struct btrfs_iget_args args;
4074         args.ino = objectid;
4075         args.root = root;
4076
4077         inode = iget5_locked(s, objectid, btrfs_find_actor,
4078                              btrfs_init_locked_inode,
4079                              (void *)&args);
4080         return inode;
4081 }
4082
4083 /* Get an inode object given its location and corresponding root.
4084  * Returns in *is_new if the inode was read from disk
4085  */
4086 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4087                          struct btrfs_root *root, int *new)
4088 {
4089         struct inode *inode;
4090
4091         inode = btrfs_iget_locked(s, location->objectid, root);
4092         if (!inode)
4093                 return ERR_PTR(-ENOMEM);
4094
4095         if (inode->i_state & I_NEW) {
4096                 BTRFS_I(inode)->root = root;
4097                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4098                 btrfs_read_locked_inode(inode);
4099                 inode_tree_add(inode);
4100                 unlock_new_inode(inode);
4101                 if (new)
4102                         *new = 1;
4103         }
4104
4105         return inode;
4106 }
4107
4108 static struct inode *new_simple_dir(struct super_block *s,
4109                                     struct btrfs_key *key,
4110                                     struct btrfs_root *root)
4111 {
4112         struct inode *inode = new_inode(s);
4113
4114         if (!inode)
4115                 return ERR_PTR(-ENOMEM);
4116
4117         BTRFS_I(inode)->root = root;
4118         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4119         BTRFS_I(inode)->dummy_inode = 1;
4120
4121         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4122         inode->i_op = &simple_dir_inode_operations;
4123         inode->i_fop = &simple_dir_operations;
4124         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4125         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4126
4127         return inode;
4128 }
4129
4130 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4131 {
4132         struct inode *inode;
4133         struct btrfs_root *root = BTRFS_I(dir)->root;
4134         struct btrfs_root *sub_root = root;
4135         struct btrfs_key location;
4136         int index;
4137         int ret;
4138
4139         if (dentry->d_name.len > BTRFS_NAME_LEN)
4140                 return ERR_PTR(-ENAMETOOLONG);
4141
4142         ret = btrfs_inode_by_name(dir, dentry, &location);
4143
4144         if (ret < 0)
4145                 return ERR_PTR(ret);
4146
4147         if (location.objectid == 0)
4148                 return NULL;
4149
4150         if (location.type == BTRFS_INODE_ITEM_KEY) {
4151                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4152                 return inode;
4153         }
4154
4155         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4156
4157         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4158         ret = fixup_tree_root_location(root, dir, dentry,
4159                                        &location, &sub_root);
4160         if (ret < 0) {
4161                 if (ret != -ENOENT)
4162                         inode = ERR_PTR(ret);
4163                 else
4164                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4165         } else {
4166                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4167         }
4168         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4169
4170         if (!IS_ERR(inode) && root != sub_root) {
4171                 down_read(&root->fs_info->cleanup_work_sem);
4172                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4173                         ret = btrfs_orphan_cleanup(sub_root);
4174                 up_read(&root->fs_info->cleanup_work_sem);
4175                 if (ret)
4176                         inode = ERR_PTR(ret);
4177         }
4178
4179         return inode;
4180 }
4181
4182 static int btrfs_dentry_delete(const struct dentry *dentry)
4183 {
4184         struct btrfs_root *root;
4185
4186         if (!dentry->d_inode && !IS_ROOT(dentry))
4187                 dentry = dentry->d_parent;
4188
4189         if (dentry->d_inode) {
4190                 root = BTRFS_I(dentry->d_inode)->root;
4191                 if (btrfs_root_refs(&root->root_item) == 0)
4192                         return 1;
4193         }
4194         return 0;
4195 }
4196
4197 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4198                                    struct nameidata *nd)
4199 {
4200         struct inode *inode;
4201
4202         inode = btrfs_lookup_dentry(dir, dentry);
4203         if (IS_ERR(inode))
4204                 return ERR_CAST(inode);
4205
4206         return d_splice_alias(inode, dentry);
4207 }
4208
4209 static unsigned char btrfs_filetype_table[] = {
4210         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4211 };
4212
4213 static int btrfs_real_readdir(struct file *filp, void *dirent,
4214                               filldir_t filldir)
4215 {
4216         struct inode *inode = filp->f_dentry->d_inode;
4217         struct btrfs_root *root = BTRFS_I(inode)->root;
4218         struct btrfs_item *item;
4219         struct btrfs_dir_item *di;
4220         struct btrfs_key key;
4221         struct btrfs_key found_key;
4222         struct btrfs_path *path;
4223         int ret;
4224         u32 nritems;
4225         struct extent_buffer *leaf;
4226         int slot;
4227         int advance;
4228         unsigned char d_type;
4229         int over = 0;
4230         u32 di_cur;
4231         u32 di_total;
4232         u32 di_len;
4233         int key_type = BTRFS_DIR_INDEX_KEY;
4234         char tmp_name[32];
4235         char *name_ptr;
4236         int name_len;
4237
4238         /* FIXME, use a real flag for deciding about the key type */
4239         if (root->fs_info->tree_root == root)
4240                 key_type = BTRFS_DIR_ITEM_KEY;
4241
4242         /* special case for "." */
4243         if (filp->f_pos == 0) {
4244                 over = filldir(dirent, ".", 1,
4245                                1, inode->i_ino,
4246                                DT_DIR);
4247                 if (over)
4248                         return 0;
4249                 filp->f_pos = 1;
4250         }
4251         /* special case for .., just use the back ref */
4252         if (filp->f_pos == 1) {
4253                 u64 pino = parent_ino(filp->f_path.dentry);
4254                 over = filldir(dirent, "..", 2,
4255                                2, pino, DT_DIR);
4256                 if (over)
4257                         return 0;
4258                 filp->f_pos = 2;
4259         }
4260         path = btrfs_alloc_path();
4261         path->reada = 2;
4262
4263         btrfs_set_key_type(&key, key_type);
4264         key.offset = filp->f_pos;
4265         key.objectid = inode->i_ino;
4266
4267         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4268         if (ret < 0)
4269                 goto err;
4270         advance = 0;
4271
4272         while (1) {
4273                 leaf = path->nodes[0];
4274                 nritems = btrfs_header_nritems(leaf);
4275                 slot = path->slots[0];
4276                 if (advance || slot >= nritems) {
4277                         if (slot >= nritems - 1) {
4278                                 ret = btrfs_next_leaf(root, path);
4279                                 if (ret)
4280                                         break;
4281                                 leaf = path->nodes[0];
4282                                 nritems = btrfs_header_nritems(leaf);
4283                                 slot = path->slots[0];
4284                         } else {
4285                                 slot++;
4286                                 path->slots[0]++;
4287                         }
4288                 }
4289
4290                 advance = 1;
4291                 item = btrfs_item_nr(leaf, slot);
4292                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4293
4294                 if (found_key.objectid != key.objectid)
4295                         break;
4296                 if (btrfs_key_type(&found_key) != key_type)
4297                         break;
4298                 if (found_key.offset < filp->f_pos)
4299                         continue;
4300
4301                 filp->f_pos = found_key.offset;
4302
4303                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4304                 di_cur = 0;
4305                 di_total = btrfs_item_size(leaf, item);
4306
4307                 while (di_cur < di_total) {
4308                         struct btrfs_key location;
4309
4310                         if (verify_dir_item(root, leaf, di))
4311                                 break;
4312
4313                         name_len = btrfs_dir_name_len(leaf, di);
4314                         if (name_len <= sizeof(tmp_name)) {
4315                                 name_ptr = tmp_name;
4316                         } else {
4317                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4318                                 if (!name_ptr) {
4319                                         ret = -ENOMEM;
4320                                         goto err;
4321                                 }
4322                         }
4323                         read_extent_buffer(leaf, name_ptr,
4324                                            (unsigned long)(di + 1), name_len);
4325
4326                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4327                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4328
4329                         /* is this a reference to our own snapshot? If so
4330                          * skip it
4331                          */
4332                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4333                             location.objectid == root->root_key.objectid) {
4334                                 over = 0;
4335                                 goto skip;
4336                         }
4337                         over = filldir(dirent, name_ptr, name_len,
4338                                        found_key.offset, location.objectid,
4339                                        d_type);
4340
4341 skip:
4342                         if (name_ptr != tmp_name)
4343                                 kfree(name_ptr);
4344
4345                         if (over)
4346                                 goto nopos;
4347                         di_len = btrfs_dir_name_len(leaf, di) +
4348                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4349                         di_cur += di_len;
4350                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4351                 }
4352         }
4353
4354         /* Reached end of directory/root. Bump pos past the last item. */
4355         if (key_type == BTRFS_DIR_INDEX_KEY)
4356                 /*
4357                  * 32-bit glibc will use getdents64, but then strtol -
4358                  * so the last number we can serve is this.
4359                  */
4360                 filp->f_pos = 0x7fffffff;
4361         else
4362                 filp->f_pos++;
4363 nopos:
4364         ret = 0;
4365 err:
4366         btrfs_free_path(path);
4367         return ret;
4368 }
4369
4370 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4371 {
4372         struct btrfs_root *root = BTRFS_I(inode)->root;
4373         struct btrfs_trans_handle *trans;
4374         int ret = 0;
4375         bool nolock = false;
4376
4377         if (BTRFS_I(inode)->dummy_inode)
4378                 return 0;
4379
4380         smp_mb();
4381         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4382
4383         if (wbc->sync_mode == WB_SYNC_ALL) {
4384                 if (nolock)
4385                         trans = btrfs_join_transaction_nolock(root, 1);
4386                 else
4387                         trans = btrfs_join_transaction(root, 1);
4388                 if (IS_ERR(trans))
4389                         return PTR_ERR(trans);
4390                 btrfs_set_trans_block_group(trans, inode);
4391                 if (nolock)
4392                         ret = btrfs_end_transaction_nolock(trans, root);
4393                 else
4394                         ret = btrfs_commit_transaction(trans, root);
4395         }
4396         return ret;
4397 }
4398
4399 /*
4400  * This is somewhat expensive, updating the tree every time the
4401  * inode changes.  But, it is most likely to find the inode in cache.
4402  * FIXME, needs more benchmarking...there are no reasons other than performance
4403  * to keep or drop this code.
4404  */
4405 void btrfs_dirty_inode(struct inode *inode)
4406 {
4407         struct btrfs_root *root = BTRFS_I(inode)->root;
4408         struct btrfs_trans_handle *trans;
4409         int ret;
4410
4411         if (BTRFS_I(inode)->dummy_inode)
4412                 return;
4413
4414         trans = btrfs_join_transaction(root, 1);
4415         BUG_ON(IS_ERR(trans));
4416         btrfs_set_trans_block_group(trans, inode);
4417
4418         ret = btrfs_update_inode(trans, root, inode);
4419         if (ret && ret == -ENOSPC) {
4420                 /* whoops, lets try again with the full transaction */
4421                 btrfs_end_transaction(trans, root);
4422                 trans = btrfs_start_transaction(root, 1);
4423                 if (IS_ERR(trans)) {
4424                         if (printk_ratelimit()) {
4425                                 printk(KERN_ERR "btrfs: fail to "
4426                                        "dirty  inode %lu error %ld\n",
4427                                        inode->i_ino, PTR_ERR(trans));
4428                         }
4429                         return;
4430                 }
4431                 btrfs_set_trans_block_group(trans, inode);
4432
4433                 ret = btrfs_update_inode(trans, root, inode);
4434                 if (ret) {
4435                         if (printk_ratelimit()) {
4436                                 printk(KERN_ERR "btrfs: fail to "
4437                                        "dirty  inode %lu error %d\n",
4438                                        inode->i_ino, ret);
4439                         }
4440                 }
4441         }
4442         btrfs_end_transaction(trans, root);
4443 }
4444
4445 /*
4446  * find the highest existing sequence number in a directory
4447  * and then set the in-memory index_cnt variable to reflect
4448  * free sequence numbers
4449  */
4450 static int btrfs_set_inode_index_count(struct inode *inode)
4451 {
4452         struct btrfs_root *root = BTRFS_I(inode)->root;
4453         struct btrfs_key key, found_key;
4454         struct btrfs_path *path;
4455         struct extent_buffer *leaf;
4456         int ret;
4457
4458         key.objectid = inode->i_ino;
4459         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4460         key.offset = (u64)-1;
4461
4462         path = btrfs_alloc_path();
4463         if (!path)
4464                 return -ENOMEM;
4465
4466         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4467         if (ret < 0)
4468                 goto out;
4469         /* FIXME: we should be able to handle this */
4470         if (ret == 0)
4471                 goto out;
4472         ret = 0;
4473
4474         /*
4475          * MAGIC NUMBER EXPLANATION:
4476          * since we search a directory based on f_pos we have to start at 2
4477          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4478          * else has to start at 2
4479          */
4480         if (path->slots[0] == 0) {
4481                 BTRFS_I(inode)->index_cnt = 2;
4482                 goto out;
4483         }
4484
4485         path->slots[0]--;
4486
4487         leaf = path->nodes[0];
4488         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4489
4490         if (found_key.objectid != inode->i_ino ||
4491             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4492                 BTRFS_I(inode)->index_cnt = 2;
4493                 goto out;
4494         }
4495
4496         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4497 out:
4498         btrfs_free_path(path);
4499         return ret;
4500 }
4501
4502 /*
4503  * helper to find a free sequence number in a given directory.  This current
4504  * code is very simple, later versions will do smarter things in the btree
4505  */
4506 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4507 {
4508         int ret = 0;
4509
4510         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4511                 ret = btrfs_set_inode_index_count(dir);
4512                 if (ret)
4513                         return ret;
4514         }
4515
4516         *index = BTRFS_I(dir)->index_cnt;
4517         BTRFS_I(dir)->index_cnt++;
4518
4519         return ret;
4520 }
4521
4522 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4523                                      struct btrfs_root *root,
4524                                      struct inode *dir,
4525                                      const char *name, int name_len,
4526                                      u64 ref_objectid, u64 objectid,
4527                                      u64 alloc_hint, int mode, u64 *index)
4528 {
4529         struct inode *inode;
4530         struct btrfs_inode_item *inode_item;
4531         struct btrfs_key *location;
4532         struct btrfs_path *path;
4533         struct btrfs_inode_ref *ref;
4534         struct btrfs_key key[2];
4535         u32 sizes[2];
4536         unsigned long ptr;
4537         int ret;
4538         int owner;
4539
4540         path = btrfs_alloc_path();
4541         BUG_ON(!path);
4542
4543         inode = new_inode(root->fs_info->sb);
4544         if (!inode)
4545                 return ERR_PTR(-ENOMEM);
4546
4547         if (dir) {
4548                 trace_btrfs_inode_request(dir);
4549
4550                 ret = btrfs_set_inode_index(dir, index);
4551                 if (ret) {
4552                         iput(inode);
4553                         return ERR_PTR(ret);
4554                 }
4555         }
4556         /*
4557          * index_cnt is ignored for everything but a dir,
4558          * btrfs_get_inode_index_count has an explanation for the magic
4559          * number
4560          */
4561         BTRFS_I(inode)->index_cnt = 2;
4562         BTRFS_I(inode)->root = root;
4563         BTRFS_I(inode)->generation = trans->transid;
4564         inode->i_generation = BTRFS_I(inode)->generation;
4565         btrfs_set_inode_space_info(root, inode);
4566
4567         if (mode & S_IFDIR)
4568                 owner = 0;
4569         else
4570                 owner = 1;
4571         BTRFS_I(inode)->block_group =
4572                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4573
4574         key[0].objectid = objectid;
4575         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4576         key[0].offset = 0;
4577
4578         key[1].objectid = objectid;
4579         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4580         key[1].offset = ref_objectid;
4581
4582         sizes[0] = sizeof(struct btrfs_inode_item);
4583         sizes[1] = name_len + sizeof(*ref);
4584
4585         path->leave_spinning = 1;
4586         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4587         if (ret != 0)
4588                 goto fail;
4589
4590         inode_init_owner(inode, dir, mode);
4591         inode->i_ino = objectid;
4592         inode_set_bytes(inode, 0);
4593         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4594         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4595                                   struct btrfs_inode_item);
4596         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4597
4598         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4599                              struct btrfs_inode_ref);
4600         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4601         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4602         ptr = (unsigned long)(ref + 1);
4603         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4604
4605         btrfs_mark_buffer_dirty(path->nodes[0]);
4606         btrfs_free_path(path);
4607
4608         location = &BTRFS_I(inode)->location;
4609         location->objectid = objectid;
4610         location->offset = 0;
4611         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4612
4613         btrfs_inherit_iflags(inode, dir);
4614
4615         if ((mode & S_IFREG)) {
4616                 if (btrfs_test_opt(root, NODATASUM))
4617                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4618                 if (btrfs_test_opt(root, NODATACOW) ||
4619                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4620                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4621         }
4622
4623         insert_inode_hash(inode);
4624         inode_tree_add(inode);
4625
4626         trace_btrfs_inode_new(inode);
4627
4628         return inode;
4629 fail:
4630         if (dir)
4631                 BTRFS_I(dir)->index_cnt--;
4632         btrfs_free_path(path);
4633         iput(inode);
4634         return ERR_PTR(ret);
4635 }
4636
4637 static inline u8 btrfs_inode_type(struct inode *inode)
4638 {
4639         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4640 }
4641
4642 /*
4643  * utility function to add 'inode' into 'parent_inode' with
4644  * a give name and a given sequence number.
4645  * if 'add_backref' is true, also insert a backref from the
4646  * inode to the parent directory.
4647  */
4648 int btrfs_add_link(struct btrfs_trans_handle *trans,
4649                    struct inode *parent_inode, struct inode *inode,
4650                    const char *name, int name_len, int add_backref, u64 index)
4651 {
4652         int ret = 0;
4653         struct btrfs_key key;
4654         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4655
4656         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4657                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4658         } else {
4659                 key.objectid = inode->i_ino;
4660                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4661                 key.offset = 0;
4662         }
4663
4664         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4665                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4666                                          key.objectid, root->root_key.objectid,
4667                                          parent_inode->i_ino,
4668                                          index, name, name_len);
4669         } else if (add_backref) {
4670                 ret = btrfs_insert_inode_ref(trans, root,
4671                                              name, name_len, inode->i_ino,
4672                                              parent_inode->i_ino, index);
4673         }
4674
4675         if (ret == 0) {
4676                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4677                                             parent_inode->i_ino, &key,
4678                                             btrfs_inode_type(inode), index);
4679                 BUG_ON(ret);
4680
4681                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4682                                    name_len * 2);
4683                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4684                 ret = btrfs_update_inode(trans, root, parent_inode);
4685         }
4686         return ret;
4687 }
4688
4689 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4690                             struct inode *dir, struct dentry *dentry,
4691                             struct inode *inode, int backref, u64 index)
4692 {
4693         int err = btrfs_add_link(trans, dir, inode,
4694                                  dentry->d_name.name, dentry->d_name.len,
4695                                  backref, index);
4696         if (!err) {
4697                 d_instantiate(dentry, inode);
4698                 return 0;
4699         }
4700         if (err > 0)
4701                 err = -EEXIST;
4702         return err;
4703 }
4704
4705 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4706                         int mode, dev_t rdev)
4707 {
4708         struct btrfs_trans_handle *trans;
4709         struct btrfs_root *root = BTRFS_I(dir)->root;
4710         struct inode *inode = NULL;
4711         int err;
4712         int drop_inode = 0;
4713         u64 objectid;
4714         unsigned long nr = 0;
4715         u64 index = 0;
4716
4717         if (!new_valid_dev(rdev))
4718                 return -EINVAL;
4719
4720         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4721         if (err)
4722                 return err;
4723
4724         /*
4725          * 2 for inode item and ref
4726          * 2 for dir items
4727          * 1 for xattr if selinux is on
4728          */
4729         trans = btrfs_start_transaction(root, 5);
4730         if (IS_ERR(trans))
4731                 return PTR_ERR(trans);
4732
4733         btrfs_set_trans_block_group(trans, dir);
4734
4735         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4736                                 dentry->d_name.len, dir->i_ino, objectid,
4737                                 BTRFS_I(dir)->block_group, mode, &index);
4738         err = PTR_ERR(inode);
4739         if (IS_ERR(inode))
4740                 goto out_unlock;
4741
4742         err = btrfs_init_inode_security(trans, inode, dir);
4743         if (err) {
4744                 drop_inode = 1;
4745                 goto out_unlock;
4746         }
4747
4748         btrfs_set_trans_block_group(trans, inode);
4749         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4750         if (err)
4751                 drop_inode = 1;
4752         else {
4753                 inode->i_op = &btrfs_special_inode_operations;
4754                 init_special_inode(inode, inode->i_mode, rdev);
4755                 btrfs_update_inode(trans, root, inode);
4756         }
4757         btrfs_update_inode_block_group(trans, inode);
4758         btrfs_update_inode_block_group(trans, dir);
4759 out_unlock:
4760         nr = trans->blocks_used;
4761         btrfs_end_transaction_throttle(trans, root);
4762         btrfs_btree_balance_dirty(root, nr);
4763         if (drop_inode) {
4764                 inode_dec_link_count(inode);
4765                 iput(inode);
4766         }
4767         return err;
4768 }
4769
4770 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4771                         int mode, struct nameidata *nd)
4772 {
4773         struct btrfs_trans_handle *trans;
4774         struct btrfs_root *root = BTRFS_I(dir)->root;
4775         struct inode *inode = NULL;
4776         int drop_inode = 0;
4777         int err;
4778         unsigned long nr = 0;
4779         u64 objectid;
4780         u64 index = 0;
4781
4782         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4783         if (err)
4784                 return err;
4785         /*
4786          * 2 for inode item and ref
4787          * 2 for dir items
4788          * 1 for xattr if selinux is on
4789          */
4790         trans = btrfs_start_transaction(root, 5);
4791         if (IS_ERR(trans))
4792                 return PTR_ERR(trans);
4793
4794         btrfs_set_trans_block_group(trans, dir);
4795
4796         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4797                                 dentry->d_name.len, dir->i_ino, objectid,
4798                                 BTRFS_I(dir)->block_group, mode, &index);
4799         err = PTR_ERR(inode);
4800         if (IS_ERR(inode))
4801                 goto out_unlock;
4802
4803         err = btrfs_init_inode_security(trans, inode, dir);
4804         if (err) {
4805                 drop_inode = 1;
4806                 goto out_unlock;
4807         }
4808
4809         btrfs_set_trans_block_group(trans, inode);
4810         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4811         if (err)
4812                 drop_inode = 1;
4813         else {
4814                 inode->i_mapping->a_ops = &btrfs_aops;
4815                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4816                 inode->i_fop = &btrfs_file_operations;
4817                 inode->i_op = &btrfs_file_inode_operations;
4818                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4819         }
4820         btrfs_update_inode_block_group(trans, inode);
4821         btrfs_update_inode_block_group(trans, dir);
4822 out_unlock:
4823         nr = trans->blocks_used;
4824         btrfs_end_transaction_throttle(trans, root);
4825         if (drop_inode) {
4826                 inode_dec_link_count(inode);
4827                 iput(inode);
4828         }
4829         btrfs_btree_balance_dirty(root, nr);
4830         return err;
4831 }
4832
4833 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4834                       struct dentry *dentry)
4835 {
4836         struct btrfs_trans_handle *trans;
4837         struct btrfs_root *root = BTRFS_I(dir)->root;
4838         struct inode *inode = old_dentry->d_inode;
4839         u64 index;
4840         unsigned long nr = 0;
4841         int err;
4842         int drop_inode = 0;
4843
4844         if (inode->i_nlink == 0)
4845                 return -ENOENT;
4846
4847         /* do not allow sys_link's with other subvols of the same device */
4848         if (root->objectid != BTRFS_I(inode)->root->objectid)
4849                 return -EXDEV;
4850
4851         if (inode->i_nlink == ~0U)
4852                 return -EMLINK;
4853
4854         btrfs_inc_nlink(inode);
4855         inode->i_ctime = CURRENT_TIME;
4856
4857         err = btrfs_set_inode_index(dir, &index);
4858         if (err)
4859                 goto fail;
4860
4861         /*
4862          * 2 items for inode and inode ref
4863          * 2 items for dir items
4864          * 1 item for parent inode
4865          */
4866         trans = btrfs_start_transaction(root, 5);
4867         if (IS_ERR(trans)) {
4868                 err = PTR_ERR(trans);
4869                 goto fail;
4870         }
4871
4872         btrfs_set_trans_block_group(trans, dir);
4873         ihold(inode);
4874
4875         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4876
4877         if (err) {
4878                 drop_inode = 1;
4879         } else {
4880                 struct dentry *parent = dget_parent(dentry);
4881                 btrfs_update_inode_block_group(trans, dir);
4882                 err = btrfs_update_inode(trans, root, inode);
4883                 BUG_ON(err);
4884                 btrfs_log_new_name(trans, inode, NULL, parent);
4885                 dput(parent);
4886         }
4887
4888         nr = trans->blocks_used;
4889         btrfs_end_transaction_throttle(trans, root);
4890 fail:
4891         if (drop_inode) {
4892                 inode_dec_link_count(inode);
4893                 iput(inode);
4894         }
4895         btrfs_btree_balance_dirty(root, nr);
4896         return err;
4897 }
4898
4899 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4900 {
4901         struct inode *inode = NULL;
4902         struct btrfs_trans_handle *trans;
4903         struct btrfs_root *root = BTRFS_I(dir)->root;
4904         int err = 0;
4905         int drop_on_err = 0;
4906         u64 objectid = 0;
4907         u64 index = 0;
4908         unsigned long nr = 1;
4909
4910         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4911         if (err)
4912                 return err;
4913
4914         /*
4915          * 2 items for inode and ref
4916          * 2 items for dir items
4917          * 1 for xattr if selinux is on
4918          */
4919         trans = btrfs_start_transaction(root, 5);
4920         if (IS_ERR(trans))
4921                 return PTR_ERR(trans);
4922         btrfs_set_trans_block_group(trans, dir);
4923
4924         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4925                                 dentry->d_name.len, dir->i_ino, objectid,
4926                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4927                                 &index);
4928         if (IS_ERR(inode)) {
4929                 err = PTR_ERR(inode);
4930                 goto out_fail;
4931         }
4932
4933         drop_on_err = 1;
4934
4935         err = btrfs_init_inode_security(trans, inode, dir);
4936         if (err)
4937                 goto out_fail;
4938
4939         inode->i_op = &btrfs_dir_inode_operations;
4940         inode->i_fop = &btrfs_dir_file_operations;
4941         btrfs_set_trans_block_group(trans, inode);
4942
4943         btrfs_i_size_write(inode, 0);
4944         err = btrfs_update_inode(trans, root, inode);
4945         if (err)
4946                 goto out_fail;
4947
4948         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4949                              dentry->d_name.len, 0, index);
4950         if (err)
4951                 goto out_fail;
4952
4953         d_instantiate(dentry, inode);
4954         drop_on_err = 0;
4955         btrfs_update_inode_block_group(trans, inode);
4956         btrfs_update_inode_block_group(trans, dir);
4957
4958 out_fail:
4959         nr = trans->blocks_used;
4960         btrfs_end_transaction_throttle(trans, root);
4961         if (drop_on_err)
4962                 iput(inode);
4963         btrfs_btree_balance_dirty(root, nr);
4964         return err;
4965 }
4966
4967 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4968  * and an extent that you want to insert, deal with overlap and insert
4969  * the new extent into the tree.
4970  */
4971 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4972                                 struct extent_map *existing,
4973                                 struct extent_map *em,
4974                                 u64 map_start, u64 map_len)
4975 {
4976         u64 start_diff;
4977
4978         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4979         start_diff = map_start - em->start;
4980         em->start = map_start;
4981         em->len = map_len;
4982         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4983             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4984                 em->block_start += start_diff;
4985                 em->block_len -= start_diff;
4986         }
4987         return add_extent_mapping(em_tree, em);
4988 }
4989
4990 static noinline int uncompress_inline(struct btrfs_path *path,
4991                                       struct inode *inode, struct page *page,
4992                                       size_t pg_offset, u64 extent_offset,
4993                                       struct btrfs_file_extent_item *item)
4994 {
4995         int ret;
4996         struct extent_buffer *leaf = path->nodes[0];
4997         char *tmp;
4998         size_t max_size;
4999         unsigned long inline_size;
5000         unsigned long ptr;
5001         int compress_type;
5002
5003         WARN_ON(pg_offset != 0);
5004         compress_type = btrfs_file_extent_compression(leaf, item);
5005         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5006         inline_size = btrfs_file_extent_inline_item_len(leaf,
5007                                         btrfs_item_nr(leaf, path->slots[0]));
5008         tmp = kmalloc(inline_size, GFP_NOFS);
5009         ptr = btrfs_file_extent_inline_start(item);
5010
5011         read_extent_buffer(leaf, tmp, ptr, inline_size);
5012
5013         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5014         ret = btrfs_decompress(compress_type, tmp, page,
5015                                extent_offset, inline_size, max_size);
5016         if (ret) {
5017                 char *kaddr = kmap_atomic(page, KM_USER0);
5018                 unsigned long copy_size = min_t(u64,
5019                                   PAGE_CACHE_SIZE - pg_offset,
5020                                   max_size - extent_offset);
5021                 memset(kaddr + pg_offset, 0, copy_size);
5022                 kunmap_atomic(kaddr, KM_USER0);
5023         }
5024         kfree(tmp);
5025         return 0;
5026 }
5027
5028 /*
5029  * a bit scary, this does extent mapping from logical file offset to the disk.
5030  * the ugly parts come from merging extents from the disk with the in-ram
5031  * representation.  This gets more complex because of the data=ordered code,
5032  * where the in-ram extents might be locked pending data=ordered completion.
5033  *
5034  * This also copies inline extents directly into the page.
5035  */
5036
5037 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5038                                     size_t pg_offset, u64 start, u64 len,
5039                                     int create)
5040 {
5041         int ret;
5042         int err = 0;
5043         u64 bytenr;
5044         u64 extent_start = 0;
5045         u64 extent_end = 0;
5046         u64 objectid = inode->i_ino;
5047         u32 found_type;
5048         struct btrfs_path *path = NULL;
5049         struct btrfs_root *root = BTRFS_I(inode)->root;
5050         struct btrfs_file_extent_item *item;
5051         struct extent_buffer *leaf;
5052         struct btrfs_key found_key;
5053         struct extent_map *em = NULL;
5054         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5055         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5056         struct btrfs_trans_handle *trans = NULL;
5057         int compress_type;
5058
5059 again:
5060         read_lock(&em_tree->lock);
5061         em = lookup_extent_mapping(em_tree, start, len);
5062         if (em)
5063                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5064         read_unlock(&em_tree->lock);
5065
5066         if (em) {
5067                 if (em->start > start || em->start + em->len <= start)
5068                         free_extent_map(em);
5069                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5070                         free_extent_map(em);
5071                 else
5072                         goto out;
5073         }
5074         em = alloc_extent_map(GFP_NOFS);
5075         if (!em) {
5076                 err = -ENOMEM;
5077                 goto out;
5078         }
5079         em->bdev = root->fs_info->fs_devices->latest_bdev;
5080         em->start = EXTENT_MAP_HOLE;
5081         em->orig_start = EXTENT_MAP_HOLE;
5082         em->len = (u64)-1;
5083         em->block_len = (u64)-1;
5084
5085         if (!path) {
5086                 path = btrfs_alloc_path();
5087                 BUG_ON(!path);
5088         }
5089
5090         ret = btrfs_lookup_file_extent(trans, root, path,
5091                                        objectid, start, trans != NULL);
5092         if (ret < 0) {
5093                 err = ret;
5094                 goto out;
5095         }
5096
5097         if (ret != 0) {
5098                 if (path->slots[0] == 0)
5099                         goto not_found;
5100                 path->slots[0]--;
5101         }
5102
5103         leaf = path->nodes[0];
5104         item = btrfs_item_ptr(leaf, path->slots[0],
5105                               struct btrfs_file_extent_item);
5106         /* are we inside the extent that was found? */
5107         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5108         found_type = btrfs_key_type(&found_key);
5109         if (found_key.objectid != objectid ||
5110             found_type != BTRFS_EXTENT_DATA_KEY) {
5111                 goto not_found;
5112         }
5113
5114         found_type = btrfs_file_extent_type(leaf, item);
5115         extent_start = found_key.offset;
5116         compress_type = btrfs_file_extent_compression(leaf, item);
5117         if (found_type == BTRFS_FILE_EXTENT_REG ||
5118             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5119                 extent_end = extent_start +
5120                        btrfs_file_extent_num_bytes(leaf, item);
5121         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5122                 size_t size;
5123                 size = btrfs_file_extent_inline_len(leaf, item);
5124                 extent_end = (extent_start + size + root->sectorsize - 1) &
5125                         ~((u64)root->sectorsize - 1);
5126         }
5127
5128         if (start >= extent_end) {
5129                 path->slots[0]++;
5130                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5131                         ret = btrfs_next_leaf(root, path);
5132                         if (ret < 0) {
5133                                 err = ret;
5134                                 goto out;
5135                         }
5136                         if (ret > 0)
5137                                 goto not_found;
5138                         leaf = path->nodes[0];
5139                 }
5140                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5141                 if (found_key.objectid != objectid ||
5142                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5143                         goto not_found;
5144                 if (start + len <= found_key.offset)
5145                         goto not_found;
5146                 em->start = start;
5147                 em->len = found_key.offset - start;
5148                 goto not_found_em;
5149         }
5150
5151         if (found_type == BTRFS_FILE_EXTENT_REG ||
5152             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5153                 em->start = extent_start;
5154                 em->len = extent_end - extent_start;
5155                 em->orig_start = extent_start -
5156                                  btrfs_file_extent_offset(leaf, item);
5157                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5158                 if (bytenr == 0) {
5159                         em->block_start = EXTENT_MAP_HOLE;
5160                         goto insert;
5161                 }
5162                 if (compress_type != BTRFS_COMPRESS_NONE) {
5163                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5164                         em->compress_type = compress_type;
5165                         em->block_start = bytenr;
5166                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5167                                                                          item);
5168                 } else {
5169                         bytenr += btrfs_file_extent_offset(leaf, item);
5170                         em->block_start = bytenr;
5171                         em->block_len = em->len;
5172                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5173                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5174                 }
5175                 goto insert;
5176         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5177                 unsigned long ptr;
5178                 char *map;
5179                 size_t size;
5180                 size_t extent_offset;
5181                 size_t copy_size;
5182
5183                 em->block_start = EXTENT_MAP_INLINE;
5184                 if (!page || create) {
5185                         em->start = extent_start;
5186                         em->len = extent_end - extent_start;
5187                         goto out;
5188                 }
5189
5190                 size = btrfs_file_extent_inline_len(leaf, item);
5191                 extent_offset = page_offset(page) + pg_offset - extent_start;
5192                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5193                                 size - extent_offset);
5194                 em->start = extent_start + extent_offset;
5195                 em->len = (copy_size + root->sectorsize - 1) &
5196                         ~((u64)root->sectorsize - 1);
5197                 em->orig_start = EXTENT_MAP_INLINE;
5198                 if (compress_type) {
5199                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5200                         em->compress_type = compress_type;
5201                 }
5202                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5203                 if (create == 0 && !PageUptodate(page)) {
5204                         if (btrfs_file_extent_compression(leaf, item) !=
5205                             BTRFS_COMPRESS_NONE) {
5206                                 ret = uncompress_inline(path, inode, page,
5207                                                         pg_offset,
5208                                                         extent_offset, item);
5209                                 BUG_ON(ret);
5210                         } else {
5211                                 map = kmap(page);
5212                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5213                                                    copy_size);
5214                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5215                                         memset(map + pg_offset + copy_size, 0,
5216                                                PAGE_CACHE_SIZE - pg_offset -
5217                                                copy_size);
5218                                 }
5219                                 kunmap(page);
5220                         }
5221                         flush_dcache_page(page);
5222                 } else if (create && PageUptodate(page)) {
5223                         WARN_ON(1);
5224                         if (!trans) {
5225                                 kunmap(page);
5226                                 free_extent_map(em);
5227                                 em = NULL;
5228                                 btrfs_release_path(root, path);
5229                                 trans = btrfs_join_transaction(root, 1);
5230                                 if (IS_ERR(trans))
5231                                         return ERR_CAST(trans);
5232                                 goto again;
5233                         }
5234                         map = kmap(page);
5235                         write_extent_buffer(leaf, map + pg_offset, ptr,
5236                                             copy_size);
5237                         kunmap(page);
5238                         btrfs_mark_buffer_dirty(leaf);
5239                 }
5240                 set_extent_uptodate(io_tree, em->start,
5241                                     extent_map_end(em) - 1, GFP_NOFS);
5242                 goto insert;
5243         } else {
5244                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5245                 WARN_ON(1);
5246         }
5247 not_found:
5248         em->start = start;
5249         em->len = len;
5250 not_found_em:
5251         em->block_start = EXTENT_MAP_HOLE;
5252         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5253 insert:
5254         btrfs_release_path(root, path);
5255         if (em->start > start || extent_map_end(em) <= start) {
5256                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5257                        "[%llu %llu]\n", (unsigned long long)em->start,
5258                        (unsigned long long)em->len,
5259                        (unsigned long long)start,
5260                        (unsigned long long)len);
5261                 err = -EIO;
5262                 goto out;
5263         }
5264
5265         err = 0;
5266         write_lock(&em_tree->lock);
5267         ret = add_extent_mapping(em_tree, em);
5268         /* it is possible that someone inserted the extent into the tree
5269          * while we had the lock dropped.  It is also possible that
5270          * an overlapping map exists in the tree
5271          */
5272         if (ret == -EEXIST) {
5273                 struct extent_map *existing;
5274
5275                 ret = 0;
5276
5277                 existing = lookup_extent_mapping(em_tree, start, len);
5278                 if (existing && (existing->start > start ||
5279                     existing->start + existing->len <= start)) {
5280                         free_extent_map(existing);
5281                         existing = NULL;
5282                 }
5283                 if (!existing) {
5284                         existing = lookup_extent_mapping(em_tree, em->start,
5285                                                          em->len);
5286                         if (existing) {
5287                                 err = merge_extent_mapping(em_tree, existing,
5288                                                            em, start,
5289                                                            root->sectorsize);
5290                                 free_extent_map(existing);
5291                                 if (err) {
5292                                         free_extent_map(em);
5293                                         em = NULL;
5294                                 }
5295                         } else {
5296                                 err = -EIO;
5297                                 free_extent_map(em);
5298                                 em = NULL;
5299                         }
5300                 } else {
5301                         free_extent_map(em);
5302                         em = existing;
5303                         err = 0;
5304                 }
5305         }
5306         write_unlock(&em_tree->lock);
5307 out:
5308
5309         trace_btrfs_get_extent(root, em);
5310
5311         if (path)
5312                 btrfs_free_path(path);
5313         if (trans) {
5314                 ret = btrfs_end_transaction(trans, root);
5315                 if (!err)
5316                         err = ret;
5317         }
5318         if (err) {
5319                 free_extent_map(em);
5320                 return ERR_PTR(err);
5321         }
5322         return em;
5323 }
5324
5325 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5326                                            size_t pg_offset, u64 start, u64 len,
5327                                            int create)
5328 {
5329         struct extent_map *em;
5330         struct extent_map *hole_em = NULL;
5331         u64 range_start = start;
5332         u64 end;
5333         u64 found;
5334         u64 found_end;
5335         int err = 0;
5336
5337         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5338         if (IS_ERR(em))
5339                 return em;
5340         if (em) {
5341                 /*
5342                  * if our em maps to a hole, there might
5343                  * actually be delalloc bytes behind it
5344                  */
5345                 if (em->block_start != EXTENT_MAP_HOLE)
5346                         return em;
5347                 else
5348                         hole_em = em;
5349         }
5350
5351         /* check to see if we've wrapped (len == -1 or similar) */
5352         end = start + len;
5353         if (end < start)
5354                 end = (u64)-1;
5355         else
5356                 end -= 1;
5357
5358         em = NULL;
5359
5360         /* ok, we didn't find anything, lets look for delalloc */
5361         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5362                                  end, len, EXTENT_DELALLOC, 1);
5363         found_end = range_start + found;
5364         if (found_end < range_start)
5365                 found_end = (u64)-1;
5366
5367         /*
5368          * we didn't find anything useful, return
5369          * the original results from get_extent()
5370          */
5371         if (range_start > end || found_end <= start) {
5372                 em = hole_em;
5373                 hole_em = NULL;
5374                 goto out;
5375         }
5376
5377         /* adjust the range_start to make sure it doesn't
5378          * go backwards from the start they passed in
5379          */
5380         range_start = max(start,range_start);
5381         found = found_end - range_start;
5382
5383         if (found > 0) {
5384                 u64 hole_start = start;
5385                 u64 hole_len = len;
5386
5387                 em = alloc_extent_map(GFP_NOFS);
5388                 if (!em) {
5389                         err = -ENOMEM;
5390                         goto out;
5391                 }
5392                 /*
5393                  * when btrfs_get_extent can't find anything it
5394                  * returns one huge hole
5395                  *
5396                  * make sure what it found really fits our range, and
5397                  * adjust to make sure it is based on the start from
5398                  * the caller
5399                  */
5400                 if (hole_em) {
5401                         u64 calc_end = extent_map_end(hole_em);
5402
5403                         if (calc_end <= start || (hole_em->start > end)) {
5404                                 free_extent_map(hole_em);
5405                                 hole_em = NULL;
5406                         } else {
5407                                 hole_start = max(hole_em->start, start);
5408                                 hole_len = calc_end - hole_start;
5409                         }
5410                 }
5411                 em->bdev = NULL;
5412                 if (hole_em && range_start > hole_start) {
5413                         /* our hole starts before our delalloc, so we
5414                          * have to return just the parts of the hole
5415                          * that go until  the delalloc starts
5416                          */
5417                         em->len = min(hole_len,
5418                                       range_start - hole_start);
5419                         em->start = hole_start;
5420                         em->orig_start = hole_start;
5421                         /*
5422                          * don't adjust block start at all,
5423                          * it is fixed at EXTENT_MAP_HOLE
5424                          */
5425                         em->block_start = hole_em->block_start;
5426                         em->block_len = hole_len;
5427                 } else {
5428                         em->start = range_start;
5429                         em->len = found;
5430                         em->orig_start = range_start;
5431                         em->block_start = EXTENT_MAP_DELALLOC;
5432                         em->block_len = found;
5433                 }
5434         } else if (hole_em) {
5435                 return hole_em;
5436         }
5437 out:
5438
5439         free_extent_map(hole_em);
5440         if (err) {
5441                 free_extent_map(em);
5442                 return ERR_PTR(err);
5443         }
5444         return em;
5445 }
5446
5447 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5448                                                   struct extent_map *em,
5449                                                   u64 start, u64 len)
5450 {
5451         struct btrfs_root *root = BTRFS_I(inode)->root;
5452         struct btrfs_trans_handle *trans;
5453         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5454         struct btrfs_key ins;
5455         u64 alloc_hint;
5456         int ret;
5457         bool insert = false;
5458
5459         /*
5460          * Ok if the extent map we looked up is a hole and is for the exact
5461          * range we want, there is no reason to allocate a new one, however if
5462          * it is not right then we need to free this one and drop the cache for
5463          * our range.
5464          */
5465         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5466             em->len != len) {
5467                 free_extent_map(em);
5468                 em = NULL;
5469                 insert = true;
5470                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5471         }
5472
5473         trans = btrfs_join_transaction(root, 0);
5474         if (IS_ERR(trans))
5475                 return ERR_CAST(trans);
5476
5477         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5478
5479         alloc_hint = get_extent_allocation_hint(inode, start, len);
5480         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5481                                    alloc_hint, (u64)-1, &ins, 1);
5482         if (ret) {
5483                 em = ERR_PTR(ret);
5484                 goto out;
5485         }
5486
5487         if (!em) {
5488                 em = alloc_extent_map(GFP_NOFS);
5489                 if (!em) {
5490                         em = ERR_PTR(-ENOMEM);
5491                         goto out;
5492                 }
5493         }
5494
5495         em->start = start;
5496         em->orig_start = em->start;
5497         em->len = ins.offset;
5498
5499         em->block_start = ins.objectid;
5500         em->block_len = ins.offset;
5501         em->bdev = root->fs_info->fs_devices->latest_bdev;
5502
5503         /*
5504          * We need to do this because if we're using the original em we searched
5505          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5506          */
5507         em->flags = 0;
5508         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5509
5510         while (insert) {
5511                 write_lock(&em_tree->lock);
5512                 ret = add_extent_mapping(em_tree, em);
5513                 write_unlock(&em_tree->lock);
5514                 if (ret != -EEXIST)
5515                         break;
5516                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5517         }
5518
5519         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5520                                            ins.offset, ins.offset, 0);
5521         if (ret) {
5522                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5523                 em = ERR_PTR(ret);
5524         }
5525 out:
5526         btrfs_end_transaction(trans, root);
5527         return em;
5528 }
5529
5530 /*
5531  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5532  * block must be cow'd
5533  */
5534 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5535                                       struct inode *inode, u64 offset, u64 len)
5536 {
5537         struct btrfs_path *path;
5538         int ret;
5539         struct extent_buffer *leaf;
5540         struct btrfs_root *root = BTRFS_I(inode)->root;
5541         struct btrfs_file_extent_item *fi;
5542         struct btrfs_key key;
5543         u64 disk_bytenr;
5544         u64 backref_offset;
5545         u64 extent_end;
5546         u64 num_bytes;
5547         int slot;
5548         int found_type;
5549
5550         path = btrfs_alloc_path();
5551         if (!path)
5552                 return -ENOMEM;
5553
5554         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5555                                        offset, 0);
5556         if (ret < 0)
5557                 goto out;
5558
5559         slot = path->slots[0];
5560         if (ret == 1) {
5561                 if (slot == 0) {
5562                         /* can't find the item, must cow */
5563                         ret = 0;
5564                         goto out;
5565                 }
5566                 slot--;
5567         }
5568         ret = 0;
5569         leaf = path->nodes[0];
5570         btrfs_item_key_to_cpu(leaf, &key, slot);
5571         if (key.objectid != inode->i_ino ||
5572             key.type != BTRFS_EXTENT_DATA_KEY) {
5573                 /* not our file or wrong item type, must cow */
5574                 goto out;
5575         }
5576
5577         if (key.offset > offset) {
5578                 /* Wrong offset, must cow */
5579                 goto out;
5580         }
5581
5582         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5583         found_type = btrfs_file_extent_type(leaf, fi);
5584         if (found_type != BTRFS_FILE_EXTENT_REG &&
5585             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5586                 /* not a regular extent, must cow */
5587                 goto out;
5588         }
5589         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5590         backref_offset = btrfs_file_extent_offset(leaf, fi);
5591
5592         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5593         if (extent_end < offset + len) {
5594                 /* extent doesn't include our full range, must cow */
5595                 goto out;
5596         }
5597
5598         if (btrfs_extent_readonly(root, disk_bytenr))
5599                 goto out;
5600
5601         /*
5602          * look for other files referencing this extent, if we
5603          * find any we must cow
5604          */
5605         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5606                                   key.offset - backref_offset, disk_bytenr))
5607                 goto out;
5608
5609         /*
5610          * adjust disk_bytenr and num_bytes to cover just the bytes
5611          * in this extent we are about to write.  If there
5612          * are any csums in that range we have to cow in order
5613          * to keep the csums correct
5614          */
5615         disk_bytenr += backref_offset;
5616         disk_bytenr += offset - key.offset;
5617         num_bytes = min(offset + len, extent_end) - offset;
5618         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5619                                 goto out;
5620         /*
5621          * all of the above have passed, it is safe to overwrite this extent
5622          * without cow
5623          */
5624         ret = 1;
5625 out:
5626         btrfs_free_path(path);
5627         return ret;
5628 }
5629
5630 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5631                                    struct buffer_head *bh_result, int create)
5632 {
5633         struct extent_map *em;
5634         struct btrfs_root *root = BTRFS_I(inode)->root;
5635         u64 start = iblock << inode->i_blkbits;
5636         u64 len = bh_result->b_size;
5637         struct btrfs_trans_handle *trans;
5638
5639         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5640         if (IS_ERR(em))
5641                 return PTR_ERR(em);
5642
5643         /*
5644          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5645          * io.  INLINE is special, and we could probably kludge it in here, but
5646          * it's still buffered so for safety lets just fall back to the generic
5647          * buffered path.
5648          *
5649          * For COMPRESSED we _have_ to read the entire extent in so we can
5650          * decompress it, so there will be buffering required no matter what we
5651          * do, so go ahead and fallback to buffered.
5652          *
5653          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5654          * to buffered IO.  Don't blame me, this is the price we pay for using
5655          * the generic code.
5656          */
5657         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5658             em->block_start == EXTENT_MAP_INLINE) {
5659                 free_extent_map(em);
5660                 return -ENOTBLK;
5661         }
5662
5663         /* Just a good old fashioned hole, return */
5664         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5665                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5666                 free_extent_map(em);
5667                 /* DIO will do one hole at a time, so just unlock a sector */
5668                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5669                               start + root->sectorsize - 1, GFP_NOFS);
5670                 return 0;
5671         }
5672
5673         /*
5674          * We don't allocate a new extent in the following cases
5675          *
5676          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5677          * existing extent.
5678          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5679          * just use the extent.
5680          *
5681          */
5682         if (!create) {
5683                 len = em->len - (start - em->start);
5684                 goto map;
5685         }
5686
5687         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5688             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5689              em->block_start != EXTENT_MAP_HOLE)) {
5690                 int type;
5691                 int ret;
5692                 u64 block_start;
5693
5694                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5695                         type = BTRFS_ORDERED_PREALLOC;
5696                 else
5697                         type = BTRFS_ORDERED_NOCOW;
5698                 len = min(len, em->len - (start - em->start));
5699                 block_start = em->block_start + (start - em->start);
5700
5701                 /*
5702                  * we're not going to log anything, but we do need
5703                  * to make sure the current transaction stays open
5704                  * while we look for nocow cross refs
5705                  */
5706                 trans = btrfs_join_transaction(root, 0);
5707                 if (IS_ERR(trans))
5708                         goto must_cow;
5709
5710                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5711                         ret = btrfs_add_ordered_extent_dio(inode, start,
5712                                            block_start, len, len, type);
5713                         btrfs_end_transaction(trans, root);
5714                         if (ret) {
5715                                 free_extent_map(em);
5716                                 return ret;
5717                         }
5718                         goto unlock;
5719                 }
5720                 btrfs_end_transaction(trans, root);
5721         }
5722 must_cow:
5723         /*
5724          * this will cow the extent, reset the len in case we changed
5725          * it above
5726          */
5727         len = bh_result->b_size;
5728         em = btrfs_new_extent_direct(inode, em, start, len);
5729         if (IS_ERR(em))
5730                 return PTR_ERR(em);
5731         len = min(len, em->len - (start - em->start));
5732 unlock:
5733         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5734                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5735                           0, NULL, GFP_NOFS);
5736 map:
5737         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5738                 inode->i_blkbits;
5739         bh_result->b_size = len;
5740         bh_result->b_bdev = em->bdev;
5741         set_buffer_mapped(bh_result);
5742         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5743                 set_buffer_new(bh_result);
5744
5745         free_extent_map(em);
5746
5747         return 0;
5748 }
5749
5750 struct btrfs_dio_private {
5751         struct inode *inode;
5752         u64 logical_offset;
5753         u64 disk_bytenr;
5754         u64 bytes;
5755         u32 *csums;
5756         void *private;
5757
5758         /* number of bios pending for this dio */
5759         atomic_t pending_bios;
5760
5761         /* IO errors */
5762         int errors;
5763
5764         struct bio *orig_bio;
5765 };
5766
5767 static void btrfs_endio_direct_read(struct bio *bio, int err)
5768 {
5769         struct btrfs_dio_private *dip = bio->bi_private;
5770         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5771         struct bio_vec *bvec = bio->bi_io_vec;
5772         struct inode *inode = dip->inode;
5773         struct btrfs_root *root = BTRFS_I(inode)->root;
5774         u64 start;
5775         u32 *private = dip->csums;
5776
5777         start = dip->logical_offset;
5778         do {
5779                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5780                         struct page *page = bvec->bv_page;
5781                         char *kaddr;
5782                         u32 csum = ~(u32)0;
5783                         unsigned long flags;
5784
5785                         local_irq_save(flags);
5786                         kaddr = kmap_atomic(page, KM_IRQ0);
5787                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5788                                                csum, bvec->bv_len);
5789                         btrfs_csum_final(csum, (char *)&csum);
5790                         kunmap_atomic(kaddr, KM_IRQ0);
5791                         local_irq_restore(flags);
5792
5793                         flush_dcache_page(bvec->bv_page);
5794                         if (csum != *private) {
5795                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5796                                       " %llu csum %u private %u\n",
5797                                       inode->i_ino, (unsigned long long)start,
5798                                       csum, *private);
5799                                 err = -EIO;
5800                         }
5801                 }
5802
5803                 start += bvec->bv_len;
5804                 private++;
5805                 bvec++;
5806         } while (bvec <= bvec_end);
5807
5808         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5809                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5810         bio->bi_private = dip->private;
5811
5812         kfree(dip->csums);
5813         kfree(dip);
5814
5815         /* If we had a csum failure make sure to clear the uptodate flag */
5816         if (err)
5817                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5818         dio_end_io(bio, err);
5819 }
5820
5821 static void btrfs_endio_direct_write(struct bio *bio, int err)
5822 {
5823         struct btrfs_dio_private *dip = bio->bi_private;
5824         struct inode *inode = dip->inode;
5825         struct btrfs_root *root = BTRFS_I(inode)->root;
5826         struct btrfs_trans_handle *trans;
5827         struct btrfs_ordered_extent *ordered = NULL;
5828         struct extent_state *cached_state = NULL;
5829         u64 ordered_offset = dip->logical_offset;
5830         u64 ordered_bytes = dip->bytes;
5831         int ret;
5832
5833         if (err)
5834                 goto out_done;
5835 again:
5836         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5837                                                    &ordered_offset,
5838                                                    ordered_bytes);
5839         if (!ret)
5840                 goto out_test;
5841
5842         BUG_ON(!ordered);
5843
5844         trans = btrfs_join_transaction(root, 1);
5845         if (IS_ERR(trans)) {
5846                 err = -ENOMEM;
5847                 goto out;
5848         }
5849         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5850
5851         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5852                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5853                 if (!ret)
5854                         ret = btrfs_update_inode(trans, root, inode);
5855                 err = ret;
5856                 goto out;
5857         }
5858
5859         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5860                          ordered->file_offset + ordered->len - 1, 0,
5861                          &cached_state, GFP_NOFS);
5862
5863         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5864                 ret = btrfs_mark_extent_written(trans, inode,
5865                                                 ordered->file_offset,
5866                                                 ordered->file_offset +
5867                                                 ordered->len);
5868                 if (ret) {
5869                         err = ret;
5870                         goto out_unlock;
5871                 }
5872         } else {
5873                 ret = insert_reserved_file_extent(trans, inode,
5874                                                   ordered->file_offset,
5875                                                   ordered->start,
5876                                                   ordered->disk_len,
5877                                                   ordered->len,
5878                                                   ordered->len,
5879                                                   0, 0, 0,
5880                                                   BTRFS_FILE_EXTENT_REG);
5881                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5882                                    ordered->file_offset, ordered->len);
5883                 if (ret) {
5884                         err = ret;
5885                         WARN_ON(1);
5886                         goto out_unlock;
5887                 }
5888         }
5889
5890         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5891         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5892         if (!ret)
5893                 btrfs_update_inode(trans, root, inode);
5894         ret = 0;
5895 out_unlock:
5896         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5897                              ordered->file_offset + ordered->len - 1,
5898                              &cached_state, GFP_NOFS);
5899 out:
5900         btrfs_delalloc_release_metadata(inode, ordered->len);
5901         btrfs_end_transaction(trans, root);
5902         ordered_offset = ordered->file_offset + ordered->len;
5903         btrfs_put_ordered_extent(ordered);
5904         btrfs_put_ordered_extent(ordered);
5905
5906 out_test:
5907         /*
5908          * our bio might span multiple ordered extents.  If we haven't
5909          * completed the accounting for the whole dio, go back and try again
5910          */
5911         if (ordered_offset < dip->logical_offset + dip->bytes) {
5912                 ordered_bytes = dip->logical_offset + dip->bytes -
5913                         ordered_offset;
5914                 goto again;
5915         }
5916 out_done:
5917         bio->bi_private = dip->private;
5918
5919         kfree(dip->csums);
5920         kfree(dip);
5921
5922         /* If we had an error make sure to clear the uptodate flag */
5923         if (err)
5924                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5925         dio_end_io(bio, err);
5926 }
5927
5928 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5929                                     struct bio *bio, int mirror_num,
5930                                     unsigned long bio_flags, u64 offset)
5931 {
5932         int ret;
5933         struct btrfs_root *root = BTRFS_I(inode)->root;
5934         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5935         BUG_ON(ret);
5936         return 0;
5937 }
5938
5939 static void btrfs_end_dio_bio(struct bio *bio, int err)
5940 {
5941         struct btrfs_dio_private *dip = bio->bi_private;
5942
5943         if (err) {
5944                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5945                       "sector %#Lx len %u err no %d\n",
5946                       dip->inode->i_ino, bio->bi_rw,
5947                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5948                 dip->errors = 1;
5949
5950                 /*
5951                  * before atomic variable goto zero, we must make sure
5952                  * dip->errors is perceived to be set.
5953                  */
5954                 smp_mb__before_atomic_dec();
5955         }
5956
5957         /* if there are more bios still pending for this dio, just exit */
5958         if (!atomic_dec_and_test(&dip->pending_bios))
5959                 goto out;
5960
5961         if (dip->errors)
5962                 bio_io_error(dip->orig_bio);
5963         else {
5964                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5965                 bio_endio(dip->orig_bio, 0);
5966         }
5967 out:
5968         bio_put(bio);
5969 }
5970
5971 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5972                                        u64 first_sector, gfp_t gfp_flags)
5973 {
5974         int nr_vecs = bio_get_nr_vecs(bdev);
5975         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5976 }
5977
5978 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5979                                          int rw, u64 file_offset, int skip_sum,
5980                                          u32 *csums, int async_submit)
5981 {
5982         int write = rw & REQ_WRITE;
5983         struct btrfs_root *root = BTRFS_I(inode)->root;
5984         int ret;
5985
5986         bio_get(bio);
5987         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5988         if (ret)
5989                 goto err;
5990
5991         if (skip_sum)
5992                 goto map;
5993
5994         if (write && async_submit) {
5995                 ret = btrfs_wq_submit_bio(root->fs_info,
5996                                    inode, rw, bio, 0, 0,
5997                                    file_offset,
5998                                    __btrfs_submit_bio_start_direct_io,
5999                                    __btrfs_submit_bio_done);
6000                 goto err;
6001         } else if (write) {
6002                 /*
6003                  * If we aren't doing async submit, calculate the csum of the
6004                  * bio now.
6005                  */
6006                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6007                 if (ret)
6008                         goto err;
6009         } else if (!skip_sum) {
6010                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6011                                           file_offset, csums);
6012                 if (ret)
6013                         goto err;
6014         }
6015
6016 map:
6017         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6018 err:
6019         bio_put(bio);
6020         return ret;
6021 }
6022
6023 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6024                                     int skip_sum)
6025 {
6026         struct inode *inode = dip->inode;
6027         struct btrfs_root *root = BTRFS_I(inode)->root;
6028         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6029         struct bio *bio;
6030         struct bio *orig_bio = dip->orig_bio;
6031         struct bio_vec *bvec = orig_bio->bi_io_vec;
6032         u64 start_sector = orig_bio->bi_sector;
6033         u64 file_offset = dip->logical_offset;
6034         u64 submit_len = 0;
6035         u64 map_length;
6036         int nr_pages = 0;
6037         u32 *csums = dip->csums;
6038         int ret = 0;
6039         int async_submit = 0;
6040         int write = rw & REQ_WRITE;
6041
6042         map_length = orig_bio->bi_size;
6043         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6044                               &map_length, NULL, 0);
6045         if (ret) {
6046                 bio_put(bio);
6047                 return -EIO;
6048         }
6049
6050         if (map_length >= orig_bio->bi_size) {
6051                 bio = orig_bio;
6052                 goto submit;
6053         }
6054
6055         async_submit = 1;
6056         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6057         if (!bio)
6058                 return -ENOMEM;
6059         bio->bi_private = dip;
6060         bio->bi_end_io = btrfs_end_dio_bio;
6061         atomic_inc(&dip->pending_bios);
6062
6063         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6064                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6065                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6066                                  bvec->bv_offset) < bvec->bv_len)) {
6067                         /*
6068                          * inc the count before we submit the bio so
6069                          * we know the end IO handler won't happen before
6070                          * we inc the count. Otherwise, the dip might get freed
6071                          * before we're done setting it up
6072                          */
6073                         atomic_inc(&dip->pending_bios);
6074                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6075                                                      file_offset, skip_sum,
6076                                                      csums, async_submit);
6077                         if (ret) {
6078                                 bio_put(bio);
6079                                 atomic_dec(&dip->pending_bios);
6080                                 goto out_err;
6081                         }
6082
6083                         /* Write's use the ordered csums */
6084                         if (!write && !skip_sum)
6085                                 csums = csums + nr_pages;
6086                         start_sector += submit_len >> 9;
6087                         file_offset += submit_len;
6088
6089                         submit_len = 0;
6090                         nr_pages = 0;
6091
6092                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6093                                                   start_sector, GFP_NOFS);
6094                         if (!bio)
6095                                 goto out_err;
6096                         bio->bi_private = dip;
6097                         bio->bi_end_io = btrfs_end_dio_bio;
6098
6099                         map_length = orig_bio->bi_size;
6100                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6101                                               &map_length, NULL, 0);
6102                         if (ret) {
6103                                 bio_put(bio);
6104                                 goto out_err;
6105                         }
6106                 } else {
6107                         submit_len += bvec->bv_len;
6108                         nr_pages ++;
6109                         bvec++;
6110                 }
6111         }
6112
6113 submit:
6114         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6115                                      csums, async_submit);
6116         if (!ret)
6117                 return 0;
6118
6119         bio_put(bio);
6120 out_err:
6121         dip->errors = 1;
6122         /*
6123          * before atomic variable goto zero, we must
6124          * make sure dip->errors is perceived to be set.
6125          */
6126         smp_mb__before_atomic_dec();
6127         if (atomic_dec_and_test(&dip->pending_bios))
6128                 bio_io_error(dip->orig_bio);
6129
6130         /* bio_end_io() will handle error, so we needn't return it */
6131         return 0;
6132 }
6133
6134 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6135                                 loff_t file_offset)
6136 {
6137         struct btrfs_root *root = BTRFS_I(inode)->root;
6138         struct btrfs_dio_private *dip;
6139         struct bio_vec *bvec = bio->bi_io_vec;
6140         int skip_sum;
6141         int write = rw & REQ_WRITE;
6142         int ret = 0;
6143
6144         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6145
6146         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6147         if (!dip) {
6148                 ret = -ENOMEM;
6149                 goto free_ordered;
6150         }
6151         dip->csums = NULL;
6152
6153         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6154         if (!write && !skip_sum) {
6155                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6156                 if (!dip->csums) {
6157                         kfree(dip);
6158                         ret = -ENOMEM;
6159                         goto free_ordered;
6160                 }
6161         }
6162
6163         dip->private = bio->bi_private;
6164         dip->inode = inode;
6165         dip->logical_offset = file_offset;
6166
6167         dip->bytes = 0;
6168         do {
6169                 dip->bytes += bvec->bv_len;
6170                 bvec++;
6171         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6172
6173         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6174         bio->bi_private = dip;
6175         dip->errors = 0;
6176         dip->orig_bio = bio;
6177         atomic_set(&dip->pending_bios, 0);
6178
6179         if (write)
6180                 bio->bi_end_io = btrfs_endio_direct_write;
6181         else
6182                 bio->bi_end_io = btrfs_endio_direct_read;
6183
6184         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6185         if (!ret)
6186                 return;
6187 free_ordered:
6188         /*
6189          * If this is a write, we need to clean up the reserved space and kill
6190          * the ordered extent.
6191          */
6192         if (write) {
6193                 struct btrfs_ordered_extent *ordered;
6194                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6195                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6196                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6197                         btrfs_free_reserved_extent(root, ordered->start,
6198                                                    ordered->disk_len);
6199                 btrfs_put_ordered_extent(ordered);
6200                 btrfs_put_ordered_extent(ordered);
6201         }
6202         bio_endio(bio, ret);
6203 }
6204
6205 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6206                         const struct iovec *iov, loff_t offset,
6207                         unsigned long nr_segs)
6208 {
6209         int seg;
6210         size_t size;
6211         unsigned long addr;
6212         unsigned blocksize_mask = root->sectorsize - 1;
6213         ssize_t retval = -EINVAL;
6214         loff_t end = offset;
6215
6216         if (offset & blocksize_mask)
6217                 goto out;
6218
6219         /* Check the memory alignment.  Blocks cannot straddle pages */
6220         for (seg = 0; seg < nr_segs; seg++) {
6221                 addr = (unsigned long)iov[seg].iov_base;
6222                 size = iov[seg].iov_len;
6223                 end += size;
6224                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6225                         goto out;
6226         }
6227         retval = 0;
6228 out:
6229         return retval;
6230 }
6231 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6232                         const struct iovec *iov, loff_t offset,
6233                         unsigned long nr_segs)
6234 {
6235         struct file *file = iocb->ki_filp;
6236         struct inode *inode = file->f_mapping->host;
6237         struct btrfs_ordered_extent *ordered;
6238         struct extent_state *cached_state = NULL;
6239         u64 lockstart, lockend;
6240         ssize_t ret;
6241         int writing = rw & WRITE;
6242         int write_bits = 0;
6243         size_t count = iov_length(iov, nr_segs);
6244
6245         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6246                             offset, nr_segs)) {
6247                 return 0;
6248         }
6249
6250         lockstart = offset;
6251         lockend = offset + count - 1;
6252
6253         if (writing) {
6254                 ret = btrfs_delalloc_reserve_space(inode, count);
6255                 if (ret)
6256                         goto out;
6257         }
6258
6259         while (1) {
6260                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6261                                  0, &cached_state, GFP_NOFS);
6262                 /*
6263                  * We're concerned with the entire range that we're going to be
6264                  * doing DIO to, so we need to make sure theres no ordered
6265                  * extents in this range.
6266                  */
6267                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6268                                                      lockend - lockstart + 1);
6269                 if (!ordered)
6270                         break;
6271                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6272                                      &cached_state, GFP_NOFS);
6273                 btrfs_start_ordered_extent(inode, ordered, 1);
6274                 btrfs_put_ordered_extent(ordered);
6275                 cond_resched();
6276         }
6277
6278         /*
6279          * we don't use btrfs_set_extent_delalloc because we don't want
6280          * the dirty or uptodate bits
6281          */
6282         if (writing) {
6283                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6284                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6285                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6286                                      GFP_NOFS);
6287                 if (ret) {
6288                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6289                                          lockend, EXTENT_LOCKED | write_bits,
6290                                          1, 0, &cached_state, GFP_NOFS);
6291                         goto out;
6292                 }
6293         }
6294
6295         free_extent_state(cached_state);
6296         cached_state = NULL;
6297
6298         ret = __blockdev_direct_IO(rw, iocb, inode,
6299                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6300                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6301                    btrfs_submit_direct, 0);
6302
6303         if (ret < 0 && ret != -EIOCBQUEUED) {
6304                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6305                               offset + iov_length(iov, nr_segs) - 1,
6306                               EXTENT_LOCKED | write_bits, 1, 0,
6307                               &cached_state, GFP_NOFS);
6308         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6309                 /*
6310                  * We're falling back to buffered, unlock the section we didn't
6311                  * do IO on.
6312                  */
6313                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6314                               offset + iov_length(iov, nr_segs) - 1,
6315                               EXTENT_LOCKED | write_bits, 1, 0,
6316                               &cached_state, GFP_NOFS);
6317         }
6318 out:
6319         free_extent_state(cached_state);
6320         return ret;
6321 }
6322
6323 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6324                 __u64 start, __u64 len)
6325 {
6326         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6327 }
6328
6329 int btrfs_readpage(struct file *file, struct page *page)
6330 {
6331         struct extent_io_tree *tree;
6332         tree = &BTRFS_I(page->mapping->host)->io_tree;
6333         return extent_read_full_page(tree, page, btrfs_get_extent);
6334 }
6335
6336 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6337 {
6338         struct extent_io_tree *tree;
6339
6340
6341         if (current->flags & PF_MEMALLOC) {
6342                 redirty_page_for_writepage(wbc, page);
6343                 unlock_page(page);
6344                 return 0;
6345         }
6346         tree = &BTRFS_I(page->mapping->host)->io_tree;
6347         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6348 }
6349
6350 int btrfs_writepages(struct address_space *mapping,
6351                      struct writeback_control *wbc)
6352 {
6353         struct extent_io_tree *tree;
6354
6355         tree = &BTRFS_I(mapping->host)->io_tree;
6356         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6357 }
6358
6359 static int
6360 btrfs_readpages(struct file *file, struct address_space *mapping,
6361                 struct list_head *pages, unsigned nr_pages)
6362 {
6363         struct extent_io_tree *tree;
6364         tree = &BTRFS_I(mapping->host)->io_tree;
6365         return extent_readpages(tree, mapping, pages, nr_pages,
6366                                 btrfs_get_extent);
6367 }
6368 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6369 {
6370         struct extent_io_tree *tree;
6371         struct extent_map_tree *map;
6372         int ret;
6373
6374         tree = &BTRFS_I(page->mapping->host)->io_tree;
6375         map = &BTRFS_I(page->mapping->host)->extent_tree;
6376         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6377         if (ret == 1) {
6378                 ClearPagePrivate(page);
6379                 set_page_private(page, 0);
6380                 page_cache_release(page);
6381         }
6382         return ret;
6383 }
6384
6385 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6386 {
6387         if (PageWriteback(page) || PageDirty(page))
6388                 return 0;
6389         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6390 }
6391
6392 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6393 {
6394         struct extent_io_tree *tree;
6395         struct btrfs_ordered_extent *ordered;
6396         struct extent_state *cached_state = NULL;
6397         u64 page_start = page_offset(page);
6398         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6399
6400
6401         /*
6402          * we have the page locked, so new writeback can't start,
6403          * and the dirty bit won't be cleared while we are here.
6404          *
6405          * Wait for IO on this page so that we can safely clear
6406          * the PagePrivate2 bit and do ordered accounting
6407          */
6408         wait_on_page_writeback(page);
6409
6410         tree = &BTRFS_I(page->mapping->host)->io_tree;
6411         if (offset) {
6412                 btrfs_releasepage(page, GFP_NOFS);
6413                 return;
6414         }
6415         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6416                          GFP_NOFS);
6417         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6418                                            page_offset(page));
6419         if (ordered) {
6420                 /*
6421                  * IO on this page will never be started, so we need
6422                  * to account for any ordered extents now
6423                  */
6424                 clear_extent_bit(tree, page_start, page_end,
6425                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6426                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6427                                  &cached_state, GFP_NOFS);
6428                 /*
6429                  * whoever cleared the private bit is responsible
6430                  * for the finish_ordered_io
6431                  */
6432                 if (TestClearPagePrivate2(page)) {
6433                         btrfs_finish_ordered_io(page->mapping->host,
6434                                                 page_start, page_end);
6435                 }
6436                 btrfs_put_ordered_extent(ordered);
6437                 cached_state = NULL;
6438                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6439                                  GFP_NOFS);
6440         }
6441         clear_extent_bit(tree, page_start, page_end,
6442                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6443                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6444         __btrfs_releasepage(page, GFP_NOFS);
6445
6446         ClearPageChecked(page);
6447         if (PagePrivate(page)) {
6448                 ClearPagePrivate(page);
6449                 set_page_private(page, 0);
6450                 page_cache_release(page);
6451         }
6452 }
6453
6454 /*
6455  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6456  * called from a page fault handler when a page is first dirtied. Hence we must
6457  * be careful to check for EOF conditions here. We set the page up correctly
6458  * for a written page which means we get ENOSPC checking when writing into
6459  * holes and correct delalloc and unwritten extent mapping on filesystems that
6460  * support these features.
6461  *
6462  * We are not allowed to take the i_mutex here so we have to play games to
6463  * protect against truncate races as the page could now be beyond EOF.  Because
6464  * vmtruncate() writes the inode size before removing pages, once we have the
6465  * page lock we can determine safely if the page is beyond EOF. If it is not
6466  * beyond EOF, then the page is guaranteed safe against truncation until we
6467  * unlock the page.
6468  */
6469 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6470 {
6471         struct page *page = vmf->page;
6472         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6473         struct btrfs_root *root = BTRFS_I(inode)->root;
6474         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6475         struct btrfs_ordered_extent *ordered;
6476         struct extent_state *cached_state = NULL;
6477         char *kaddr;
6478         unsigned long zero_start;
6479         loff_t size;
6480         int ret;
6481         u64 page_start;
6482         u64 page_end;
6483
6484         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6485         if (ret) {
6486                 if (ret == -ENOMEM)
6487                         ret = VM_FAULT_OOM;
6488                 else /* -ENOSPC, -EIO, etc */
6489                         ret = VM_FAULT_SIGBUS;
6490                 goto out;
6491         }
6492
6493         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6494 again:
6495         lock_page(page);
6496         size = i_size_read(inode);
6497         page_start = page_offset(page);
6498         page_end = page_start + PAGE_CACHE_SIZE - 1;
6499
6500         if ((page->mapping != inode->i_mapping) ||
6501             (page_start >= size)) {
6502                 /* page got truncated out from underneath us */
6503                 goto out_unlock;
6504         }
6505         wait_on_page_writeback(page);
6506
6507         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6508                          GFP_NOFS);
6509         set_page_extent_mapped(page);
6510
6511         /*
6512          * we can't set the delalloc bits if there are pending ordered
6513          * extents.  Drop our locks and wait for them to finish
6514          */
6515         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6516         if (ordered) {
6517                 unlock_extent_cached(io_tree, page_start, page_end,
6518                                      &cached_state, GFP_NOFS);
6519                 unlock_page(page);
6520                 btrfs_start_ordered_extent(inode, ordered, 1);
6521                 btrfs_put_ordered_extent(ordered);
6522                 goto again;
6523         }
6524
6525         /*
6526          * XXX - page_mkwrite gets called every time the page is dirtied, even
6527          * if it was already dirty, so for space accounting reasons we need to
6528          * clear any delalloc bits for the range we are fixing to save.  There
6529          * is probably a better way to do this, but for now keep consistent with
6530          * prepare_pages in the normal write path.
6531          */
6532         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6533                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6534                           0, 0, &cached_state, GFP_NOFS);
6535
6536         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6537                                         &cached_state);
6538         if (ret) {
6539                 unlock_extent_cached(io_tree, page_start, page_end,
6540                                      &cached_state, GFP_NOFS);
6541                 ret = VM_FAULT_SIGBUS;
6542                 goto out_unlock;
6543         }
6544         ret = 0;
6545
6546         /* page is wholly or partially inside EOF */
6547         if (page_start + PAGE_CACHE_SIZE > size)
6548                 zero_start = size & ~PAGE_CACHE_MASK;
6549         else
6550                 zero_start = PAGE_CACHE_SIZE;
6551
6552         if (zero_start != PAGE_CACHE_SIZE) {
6553                 kaddr = kmap(page);
6554                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6555                 flush_dcache_page(page);
6556                 kunmap(page);
6557         }
6558         ClearPageChecked(page);
6559         set_page_dirty(page);
6560         SetPageUptodate(page);
6561
6562         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6563         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6564
6565         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6566
6567 out_unlock:
6568         if (!ret)
6569                 return VM_FAULT_LOCKED;
6570         unlock_page(page);
6571         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6572 out:
6573         return ret;
6574 }
6575
6576 static int btrfs_truncate(struct inode *inode)
6577 {
6578         struct btrfs_root *root = BTRFS_I(inode)->root;
6579         int ret;
6580         int err = 0;
6581         struct btrfs_trans_handle *trans;
6582         unsigned long nr;
6583         u64 mask = root->sectorsize - 1;
6584
6585         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6586         if (ret)
6587                 return ret;
6588
6589         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6590         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6591
6592         trans = btrfs_start_transaction(root, 5);
6593         if (IS_ERR(trans))
6594                 return PTR_ERR(trans);
6595
6596         btrfs_set_trans_block_group(trans, inode);
6597
6598         ret = btrfs_orphan_add(trans, inode);
6599         if (ret) {
6600                 btrfs_end_transaction(trans, root);
6601                 return ret;
6602         }
6603
6604         nr = trans->blocks_used;
6605         btrfs_end_transaction(trans, root);
6606         btrfs_btree_balance_dirty(root, nr);
6607
6608         /* Now start a transaction for the truncate */
6609         trans = btrfs_start_transaction(root, 0);
6610         if (IS_ERR(trans))
6611                 return PTR_ERR(trans);
6612         btrfs_set_trans_block_group(trans, inode);
6613         trans->block_rsv = root->orphan_block_rsv;
6614
6615         /*
6616          * setattr is responsible for setting the ordered_data_close flag,
6617          * but that is only tested during the last file release.  That
6618          * could happen well after the next commit, leaving a great big
6619          * window where new writes may get lost if someone chooses to write
6620          * to this file after truncating to zero
6621          *
6622          * The inode doesn't have any dirty data here, and so if we commit
6623          * this is a noop.  If someone immediately starts writing to the inode
6624          * it is very likely we'll catch some of their writes in this
6625          * transaction, and the commit will find this file on the ordered
6626          * data list with good things to send down.
6627          *
6628          * This is a best effort solution, there is still a window where
6629          * using truncate to replace the contents of the file will
6630          * end up with a zero length file after a crash.
6631          */
6632         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6633                 btrfs_add_ordered_operation(trans, root, inode);
6634
6635         while (1) {
6636                 if (!trans) {
6637                         trans = btrfs_start_transaction(root, 0);
6638                         if (IS_ERR(trans))
6639                                 return PTR_ERR(trans);
6640                         btrfs_set_trans_block_group(trans, inode);
6641                         trans->block_rsv = root->orphan_block_rsv;
6642                 }
6643
6644                 ret = btrfs_block_rsv_check(trans, root,
6645                                             root->orphan_block_rsv, 0, 5);
6646                 if (ret == -EAGAIN) {
6647                         ret = btrfs_commit_transaction(trans, root);
6648                         if (ret)
6649                                 return ret;
6650                         trans = NULL;
6651                         continue;
6652                 } else if (ret) {
6653                         err = ret;
6654                         break;
6655                 }
6656
6657                 ret = btrfs_truncate_inode_items(trans, root, inode,
6658                                                  inode->i_size,
6659                                                  BTRFS_EXTENT_DATA_KEY);
6660                 if (ret != -EAGAIN) {
6661                         err = ret;
6662                         break;
6663                 }
6664
6665                 ret = btrfs_update_inode(trans, root, inode);
6666                 if (ret) {
6667                         err = ret;
6668                         break;
6669                 }
6670
6671                 nr = trans->blocks_used;
6672                 btrfs_end_transaction(trans, root);
6673                 trans = NULL;
6674                 btrfs_btree_balance_dirty(root, nr);
6675         }
6676
6677         if (ret == 0 && inode->i_nlink > 0) {
6678                 ret = btrfs_orphan_del(trans, inode);
6679                 if (ret)
6680                         err = ret;
6681         } else if (ret && inode->i_nlink > 0) {
6682                 /*
6683                  * Failed to do the truncate, remove us from the in memory
6684                  * orphan list.
6685                  */
6686                 ret = btrfs_orphan_del(NULL, inode);
6687         }
6688
6689         ret = btrfs_update_inode(trans, root, inode);
6690         if (ret && !err)
6691                 err = ret;
6692
6693         nr = trans->blocks_used;
6694         ret = btrfs_end_transaction_throttle(trans, root);
6695         if (ret && !err)
6696                 err = ret;
6697         btrfs_btree_balance_dirty(root, nr);
6698
6699         return err;
6700 }
6701
6702 /*
6703  * create a new subvolume directory/inode (helper for the ioctl).
6704  */
6705 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6706                              struct btrfs_root *new_root,
6707                              u64 new_dirid, u64 alloc_hint)
6708 {
6709         struct inode *inode;
6710         int err;
6711         u64 index = 0;
6712
6713         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6714                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6715         if (IS_ERR(inode))
6716                 return PTR_ERR(inode);
6717         inode->i_op = &btrfs_dir_inode_operations;
6718         inode->i_fop = &btrfs_dir_file_operations;
6719
6720         inode->i_nlink = 1;
6721         btrfs_i_size_write(inode, 0);
6722
6723         err = btrfs_update_inode(trans, new_root, inode);
6724         BUG_ON(err);
6725
6726         iput(inode);
6727         return 0;
6728 }
6729
6730 /* helper function for file defrag and space balancing.  This
6731  * forces readahead on a given range of bytes in an inode
6732  */
6733 unsigned long btrfs_force_ra(struct address_space *mapping,
6734                               struct file_ra_state *ra, struct file *file,
6735                               pgoff_t offset, pgoff_t last_index)
6736 {
6737         pgoff_t req_size = last_index - offset + 1;
6738
6739         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6740         return offset + req_size;
6741 }
6742
6743 struct inode *btrfs_alloc_inode(struct super_block *sb)
6744 {
6745         struct btrfs_inode *ei;
6746         struct inode *inode;
6747
6748         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6749         if (!ei)
6750                 return NULL;
6751
6752         ei->root = NULL;
6753         ei->space_info = NULL;
6754         ei->generation = 0;
6755         ei->sequence = 0;
6756         ei->last_trans = 0;
6757         ei->last_sub_trans = 0;
6758         ei->logged_trans = 0;
6759         ei->delalloc_bytes = 0;
6760         ei->reserved_bytes = 0;
6761         ei->disk_i_size = 0;
6762         ei->flags = 0;
6763         ei->index_cnt = (u64)-1;
6764         ei->last_unlink_trans = 0;
6765
6766         atomic_set(&ei->outstanding_extents, 0);
6767         atomic_set(&ei->reserved_extents, 0);
6768
6769         ei->ordered_data_close = 0;
6770         ei->orphan_meta_reserved = 0;
6771         ei->dummy_inode = 0;
6772         ei->force_compress = BTRFS_COMPRESS_NONE;
6773
6774         inode = &ei->vfs_inode;
6775         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6776         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6777         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6778         mutex_init(&ei->log_mutex);
6779         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6780         INIT_LIST_HEAD(&ei->i_orphan);
6781         INIT_LIST_HEAD(&ei->delalloc_inodes);
6782         INIT_LIST_HEAD(&ei->ordered_operations);
6783         RB_CLEAR_NODE(&ei->rb_node);
6784
6785         return inode;
6786 }
6787
6788 static void btrfs_i_callback(struct rcu_head *head)
6789 {
6790         struct inode *inode = container_of(head, struct inode, i_rcu);
6791         INIT_LIST_HEAD(&inode->i_dentry);
6792         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6793 }
6794
6795 void btrfs_destroy_inode(struct inode *inode)
6796 {
6797         struct btrfs_ordered_extent *ordered;
6798         struct btrfs_root *root = BTRFS_I(inode)->root;
6799
6800         WARN_ON(!list_empty(&inode->i_dentry));
6801         WARN_ON(inode->i_data.nrpages);
6802         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6803         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6804
6805         /*
6806          * This can happen where we create an inode, but somebody else also
6807          * created the same inode and we need to destroy the one we already
6808          * created.
6809          */
6810         if (!root)
6811                 goto free;
6812
6813         /*
6814          * Make sure we're properly removed from the ordered operation
6815          * lists.
6816          */
6817         smp_mb();
6818         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6819                 spin_lock(&root->fs_info->ordered_extent_lock);
6820                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6821                 spin_unlock(&root->fs_info->ordered_extent_lock);
6822         }
6823
6824         if (root == root->fs_info->tree_root) {
6825                 struct btrfs_block_group_cache *block_group;
6826
6827                 block_group = btrfs_lookup_block_group(root->fs_info,
6828                                                 BTRFS_I(inode)->block_group);
6829                 if (block_group && block_group->inode == inode) {
6830                         spin_lock(&block_group->lock);
6831                         block_group->inode = NULL;
6832                         spin_unlock(&block_group->lock);
6833                         btrfs_put_block_group(block_group);
6834                 } else if (block_group) {
6835                         btrfs_put_block_group(block_group);
6836                 }
6837         }
6838
6839         spin_lock(&root->orphan_lock);
6840         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6841                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6842                        inode->i_ino);
6843                 list_del_init(&BTRFS_I(inode)->i_orphan);
6844         }
6845         spin_unlock(&root->orphan_lock);
6846
6847         while (1) {
6848                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6849                 if (!ordered)
6850                         break;
6851                 else {
6852                         printk(KERN_ERR "btrfs found ordered "
6853                                "extent %llu %llu on inode cleanup\n",
6854                                (unsigned long long)ordered->file_offset,
6855                                (unsigned long long)ordered->len);
6856                         btrfs_remove_ordered_extent(inode, ordered);
6857                         btrfs_put_ordered_extent(ordered);
6858                         btrfs_put_ordered_extent(ordered);
6859                 }
6860         }
6861         inode_tree_del(inode);
6862         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6863 free:
6864         call_rcu(&inode->i_rcu, btrfs_i_callback);
6865 }
6866
6867 int btrfs_drop_inode(struct inode *inode)
6868 {
6869         struct btrfs_root *root = BTRFS_I(inode)->root;
6870
6871         if (btrfs_root_refs(&root->root_item) == 0 &&
6872             root != root->fs_info->tree_root)
6873                 return 1;
6874         else
6875                 return generic_drop_inode(inode);
6876 }
6877
6878 static void init_once(void *foo)
6879 {
6880         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6881
6882         inode_init_once(&ei->vfs_inode);
6883 }
6884
6885 void btrfs_destroy_cachep(void)
6886 {
6887         if (btrfs_inode_cachep)
6888                 kmem_cache_destroy(btrfs_inode_cachep);
6889         if (btrfs_trans_handle_cachep)
6890                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6891         if (btrfs_transaction_cachep)
6892                 kmem_cache_destroy(btrfs_transaction_cachep);
6893         if (btrfs_path_cachep)
6894                 kmem_cache_destroy(btrfs_path_cachep);
6895         if (btrfs_free_space_cachep)
6896                 kmem_cache_destroy(btrfs_free_space_cachep);
6897 }
6898
6899 int btrfs_init_cachep(void)
6900 {
6901         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6902                         sizeof(struct btrfs_inode), 0,
6903                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6904         if (!btrfs_inode_cachep)
6905                 goto fail;
6906
6907         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6908                         sizeof(struct btrfs_trans_handle), 0,
6909                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6910         if (!btrfs_trans_handle_cachep)
6911                 goto fail;
6912
6913         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6914                         sizeof(struct btrfs_transaction), 0,
6915                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6916         if (!btrfs_transaction_cachep)
6917                 goto fail;
6918
6919         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6920                         sizeof(struct btrfs_path), 0,
6921                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6922         if (!btrfs_path_cachep)
6923                 goto fail;
6924
6925         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6926                         sizeof(struct btrfs_free_space), 0,
6927                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6928         if (!btrfs_free_space_cachep)
6929                 goto fail;
6930
6931         return 0;
6932 fail:
6933         btrfs_destroy_cachep();
6934         return -ENOMEM;
6935 }
6936
6937 static int btrfs_getattr(struct vfsmount *mnt,
6938                          struct dentry *dentry, struct kstat *stat)
6939 {
6940         struct inode *inode = dentry->d_inode;
6941         generic_fillattr(inode, stat);
6942         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6943         stat->blksize = PAGE_CACHE_SIZE;
6944         stat->blocks = (inode_get_bytes(inode) +
6945                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6946         return 0;
6947 }
6948
6949 /*
6950  * If a file is moved, it will inherit the cow and compression flags of the new
6951  * directory.
6952  */
6953 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6954 {
6955         struct btrfs_inode *b_dir = BTRFS_I(dir);
6956         struct btrfs_inode *b_inode = BTRFS_I(inode);
6957
6958         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6959                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6960         else
6961                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6962
6963         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6964                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6965         else
6966                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6967 }
6968
6969 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6970                            struct inode *new_dir, struct dentry *new_dentry)
6971 {
6972         struct btrfs_trans_handle *trans;
6973         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6974         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6975         struct inode *new_inode = new_dentry->d_inode;
6976         struct inode *old_inode = old_dentry->d_inode;
6977         struct timespec ctime = CURRENT_TIME;
6978         u64 index = 0;
6979         u64 root_objectid;
6980         int ret;
6981
6982         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6983                 return -EPERM;
6984
6985         /* we only allow rename subvolume link between subvolumes */
6986         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6987                 return -EXDEV;
6988
6989         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6990             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
6991                 return -ENOTEMPTY;
6992
6993         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6994             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6995                 return -ENOTEMPTY;
6996         /*
6997          * we're using rename to replace one file with another.
6998          * and the replacement file is large.  Start IO on it now so
6999          * we don't add too much work to the end of the transaction
7000          */
7001         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7002             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7003                 filemap_flush(old_inode->i_mapping);
7004
7005         /* close the racy window with snapshot create/destroy ioctl */
7006         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7007                 down_read(&root->fs_info->subvol_sem);
7008         /*
7009          * We want to reserve the absolute worst case amount of items.  So if
7010          * both inodes are subvols and we need to unlink them then that would
7011          * require 4 item modifications, but if they are both normal inodes it
7012          * would require 5 item modifications, so we'll assume their normal
7013          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7014          * should cover the worst case number of items we'll modify.
7015          */
7016         trans = btrfs_start_transaction(root, 20);
7017         if (IS_ERR(trans)) {
7018                 ret = PTR_ERR(trans);
7019                 goto out_notrans;
7020         }
7021
7022         btrfs_set_trans_block_group(trans, new_dir);
7023
7024         if (dest != root)
7025                 btrfs_record_root_in_trans(trans, dest);
7026
7027         ret = btrfs_set_inode_index(new_dir, &index);
7028         if (ret)
7029                 goto out_fail;
7030
7031         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7032                 /* force full log commit if subvolume involved. */
7033                 root->fs_info->last_trans_log_full_commit = trans->transid;
7034         } else {
7035                 ret = btrfs_insert_inode_ref(trans, dest,
7036                                              new_dentry->d_name.name,
7037                                              new_dentry->d_name.len,
7038                                              old_inode->i_ino,
7039                                              new_dir->i_ino, index);
7040                 if (ret)
7041                         goto out_fail;
7042                 /*
7043                  * this is an ugly little race, but the rename is required
7044                  * to make sure that if we crash, the inode is either at the
7045                  * old name or the new one.  pinning the log transaction lets
7046                  * us make sure we don't allow a log commit to come in after
7047                  * we unlink the name but before we add the new name back in.
7048                  */
7049                 btrfs_pin_log_trans(root);
7050         }
7051         /*
7052          * make sure the inode gets flushed if it is replacing
7053          * something.
7054          */
7055         if (new_inode && new_inode->i_size &&
7056             old_inode && S_ISREG(old_inode->i_mode)) {
7057                 btrfs_add_ordered_operation(trans, root, old_inode);
7058         }
7059
7060         old_dir->i_ctime = old_dir->i_mtime = ctime;
7061         new_dir->i_ctime = new_dir->i_mtime = ctime;
7062         old_inode->i_ctime = ctime;
7063
7064         if (old_dentry->d_parent != new_dentry->d_parent)
7065                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7066
7067         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7068                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7069                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7070                                         old_dentry->d_name.name,
7071                                         old_dentry->d_name.len);
7072         } else {
7073                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7074                                         old_dentry->d_inode,
7075                                         old_dentry->d_name.name,
7076                                         old_dentry->d_name.len);
7077                 if (!ret)
7078                         ret = btrfs_update_inode(trans, root, old_inode);
7079         }
7080         BUG_ON(ret);
7081
7082         if (new_inode) {
7083                 new_inode->i_ctime = CURRENT_TIME;
7084                 if (unlikely(new_inode->i_ino ==
7085                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7086                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7087                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7088                                                 root_objectid,
7089                                                 new_dentry->d_name.name,
7090                                                 new_dentry->d_name.len);
7091                         BUG_ON(new_inode->i_nlink == 0);
7092                 } else {
7093                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7094                                                  new_dentry->d_inode,
7095                                                  new_dentry->d_name.name,
7096                                                  new_dentry->d_name.len);
7097                 }
7098                 BUG_ON(ret);
7099                 if (new_inode->i_nlink == 0) {
7100                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7101                         BUG_ON(ret);
7102                 }
7103         }
7104
7105         fixup_inode_flags(new_dir, old_inode);
7106
7107         ret = btrfs_add_link(trans, new_dir, old_inode,
7108                              new_dentry->d_name.name,
7109                              new_dentry->d_name.len, 0, index);
7110         BUG_ON(ret);
7111
7112         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7113                 struct dentry *parent = dget_parent(new_dentry);
7114                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7115                 dput(parent);
7116                 btrfs_end_log_trans(root);
7117         }
7118 out_fail:
7119         btrfs_end_transaction_throttle(trans, root);
7120 out_notrans:
7121         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7122                 up_read(&root->fs_info->subvol_sem);
7123
7124         return ret;
7125 }
7126
7127 /*
7128  * some fairly slow code that needs optimization. This walks the list
7129  * of all the inodes with pending delalloc and forces them to disk.
7130  */
7131 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7132 {
7133         struct list_head *head = &root->fs_info->delalloc_inodes;
7134         struct btrfs_inode *binode;
7135         struct inode *inode;
7136
7137         if (root->fs_info->sb->s_flags & MS_RDONLY)
7138                 return -EROFS;
7139
7140         spin_lock(&root->fs_info->delalloc_lock);
7141         while (!list_empty(head)) {
7142                 binode = list_entry(head->next, struct btrfs_inode,
7143                                     delalloc_inodes);
7144                 inode = igrab(&binode->vfs_inode);
7145                 if (!inode)
7146                         list_del_init(&binode->delalloc_inodes);
7147                 spin_unlock(&root->fs_info->delalloc_lock);
7148                 if (inode) {
7149                         filemap_flush(inode->i_mapping);
7150                         if (delay_iput)
7151                                 btrfs_add_delayed_iput(inode);
7152                         else
7153                                 iput(inode);
7154                 }
7155                 cond_resched();
7156                 spin_lock(&root->fs_info->delalloc_lock);
7157         }
7158         spin_unlock(&root->fs_info->delalloc_lock);
7159
7160         /* the filemap_flush will queue IO into the worker threads, but
7161          * we have to make sure the IO is actually started and that
7162          * ordered extents get created before we return
7163          */
7164         atomic_inc(&root->fs_info->async_submit_draining);
7165         while (atomic_read(&root->fs_info->nr_async_submits) ||
7166               atomic_read(&root->fs_info->async_delalloc_pages)) {
7167                 wait_event(root->fs_info->async_submit_wait,
7168                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7169                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7170         }
7171         atomic_dec(&root->fs_info->async_submit_draining);
7172         return 0;
7173 }
7174
7175 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7176                                    int sync)
7177 {
7178         struct btrfs_inode *binode;
7179         struct inode *inode = NULL;
7180
7181         spin_lock(&root->fs_info->delalloc_lock);
7182         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7183                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7184                                     struct btrfs_inode, delalloc_inodes);
7185                 inode = igrab(&binode->vfs_inode);
7186                 if (inode) {
7187                         list_move_tail(&binode->delalloc_inodes,
7188                                        &root->fs_info->delalloc_inodes);
7189                         break;
7190                 }
7191
7192                 list_del_init(&binode->delalloc_inodes);
7193                 cond_resched_lock(&root->fs_info->delalloc_lock);
7194         }
7195         spin_unlock(&root->fs_info->delalloc_lock);
7196
7197         if (inode) {
7198                 if (sync) {
7199                         filemap_write_and_wait(inode->i_mapping);
7200                         /*
7201                          * We have to do this because compression doesn't
7202                          * actually set PG_writeback until it submits the pages
7203                          * for IO, which happens in an async thread, so we could
7204                          * race and not actually wait for any writeback pages
7205                          * because they've not been submitted yet.  Technically
7206                          * this could still be the case for the ordered stuff
7207                          * since the async thread may not have started to do its
7208                          * work yet.  If this becomes the case then we need to
7209                          * figure out a way to make sure that in writepage we
7210                          * wait for any async pages to be submitted before
7211                          * returning so that fdatawait does what its supposed to
7212                          * do.
7213                          */
7214                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7215                 } else {
7216                         filemap_flush(inode->i_mapping);
7217                 }
7218                 if (delay_iput)
7219                         btrfs_add_delayed_iput(inode);
7220                 else
7221                         iput(inode);
7222                 return 1;
7223         }
7224         return 0;
7225 }
7226
7227 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7228                          const char *symname)
7229 {
7230         struct btrfs_trans_handle *trans;
7231         struct btrfs_root *root = BTRFS_I(dir)->root;
7232         struct btrfs_path *path;
7233         struct btrfs_key key;
7234         struct inode *inode = NULL;
7235         int err;
7236         int drop_inode = 0;
7237         u64 objectid;
7238         u64 index = 0 ;
7239         int name_len;
7240         int datasize;
7241         unsigned long ptr;
7242         struct btrfs_file_extent_item *ei;
7243         struct extent_buffer *leaf;
7244         unsigned long nr = 0;
7245
7246         name_len = strlen(symname) + 1;
7247         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7248                 return -ENAMETOOLONG;
7249
7250         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7251         if (err)
7252                 return err;
7253         /*
7254          * 2 items for inode item and ref
7255          * 2 items for dir items
7256          * 1 item for xattr if selinux is on
7257          */
7258         trans = btrfs_start_transaction(root, 5);
7259         if (IS_ERR(trans))
7260                 return PTR_ERR(trans);
7261
7262         btrfs_set_trans_block_group(trans, dir);
7263
7264         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7265                                 dentry->d_name.len, dir->i_ino, objectid,
7266                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7267                                 &index);
7268         err = PTR_ERR(inode);
7269         if (IS_ERR(inode))
7270                 goto out_unlock;
7271
7272         err = btrfs_init_inode_security(trans, inode, dir);
7273         if (err) {
7274                 drop_inode = 1;
7275                 goto out_unlock;
7276         }
7277
7278         btrfs_set_trans_block_group(trans, inode);
7279         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7280         if (err)
7281                 drop_inode = 1;
7282         else {
7283                 inode->i_mapping->a_ops = &btrfs_aops;
7284                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7285                 inode->i_fop = &btrfs_file_operations;
7286                 inode->i_op = &btrfs_file_inode_operations;
7287                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7288         }
7289         btrfs_update_inode_block_group(trans, inode);
7290         btrfs_update_inode_block_group(trans, dir);
7291         if (drop_inode)
7292                 goto out_unlock;
7293
7294         path = btrfs_alloc_path();
7295         BUG_ON(!path);
7296         key.objectid = inode->i_ino;
7297         key.offset = 0;
7298         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7299         datasize = btrfs_file_extent_calc_inline_size(name_len);
7300         err = btrfs_insert_empty_item(trans, root, path, &key,
7301                                       datasize);
7302         if (err) {
7303                 drop_inode = 1;
7304                 goto out_unlock;
7305         }
7306         leaf = path->nodes[0];
7307         ei = btrfs_item_ptr(leaf, path->slots[0],
7308                             struct btrfs_file_extent_item);
7309         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7310         btrfs_set_file_extent_type(leaf, ei,
7311                                    BTRFS_FILE_EXTENT_INLINE);
7312         btrfs_set_file_extent_encryption(leaf, ei, 0);
7313         btrfs_set_file_extent_compression(leaf, ei, 0);
7314         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7315         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7316
7317         ptr = btrfs_file_extent_inline_start(ei);
7318         write_extent_buffer(leaf, symname, ptr, name_len);
7319         btrfs_mark_buffer_dirty(leaf);
7320         btrfs_free_path(path);
7321
7322         inode->i_op = &btrfs_symlink_inode_operations;
7323         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7324         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7325         inode_set_bytes(inode, name_len);
7326         btrfs_i_size_write(inode, name_len - 1);
7327         err = btrfs_update_inode(trans, root, inode);
7328         if (err)
7329                 drop_inode = 1;
7330
7331 out_unlock:
7332         nr = trans->blocks_used;
7333         btrfs_end_transaction_throttle(trans, root);
7334         if (drop_inode) {
7335                 inode_dec_link_count(inode);
7336                 iput(inode);
7337         }
7338         btrfs_btree_balance_dirty(root, nr);
7339         return err;
7340 }
7341
7342 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7343                                        u64 start, u64 num_bytes, u64 min_size,
7344                                        loff_t actual_len, u64 *alloc_hint,
7345                                        struct btrfs_trans_handle *trans)
7346 {
7347         struct btrfs_root *root = BTRFS_I(inode)->root;
7348         struct btrfs_key ins;
7349         u64 cur_offset = start;
7350         u64 i_size;
7351         int ret = 0;
7352         bool own_trans = true;
7353
7354         if (trans)
7355                 own_trans = false;
7356         while (num_bytes > 0) {
7357                 if (own_trans) {
7358                         trans = btrfs_start_transaction(root, 3);
7359                         if (IS_ERR(trans)) {
7360                                 ret = PTR_ERR(trans);
7361                                 break;
7362                         }
7363                 }
7364
7365                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7366                                            0, *alloc_hint, (u64)-1, &ins, 1);
7367                 if (ret) {
7368                         if (own_trans)
7369                                 btrfs_end_transaction(trans, root);
7370                         break;
7371                 }
7372
7373                 ret = insert_reserved_file_extent(trans, inode,
7374                                                   cur_offset, ins.objectid,
7375                                                   ins.offset, ins.offset,
7376                                                   ins.offset, 0, 0, 0,
7377                                                   BTRFS_FILE_EXTENT_PREALLOC);
7378                 BUG_ON(ret);
7379                 btrfs_drop_extent_cache(inode, cur_offset,
7380                                         cur_offset + ins.offset -1, 0);
7381
7382                 num_bytes -= ins.offset;
7383                 cur_offset += ins.offset;
7384                 *alloc_hint = ins.objectid + ins.offset;
7385
7386                 inode->i_ctime = CURRENT_TIME;
7387                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7388                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7389                     (actual_len > inode->i_size) &&
7390                     (cur_offset > inode->i_size)) {
7391                         if (cur_offset > actual_len)
7392                                 i_size = actual_len;
7393                         else
7394                                 i_size = cur_offset;
7395                         i_size_write(inode, i_size);
7396                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7397                 }
7398
7399                 ret = btrfs_update_inode(trans, root, inode);
7400                 BUG_ON(ret);
7401
7402                 if (own_trans)
7403                         btrfs_end_transaction(trans, root);
7404         }
7405         return ret;
7406 }
7407
7408 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7409                               u64 start, u64 num_bytes, u64 min_size,
7410                               loff_t actual_len, u64 *alloc_hint)
7411 {
7412         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7413                                            min_size, actual_len, alloc_hint,
7414                                            NULL);
7415 }
7416
7417 int btrfs_prealloc_file_range_trans(struct inode *inode,
7418                                     struct btrfs_trans_handle *trans, int mode,
7419                                     u64 start, u64 num_bytes, u64 min_size,
7420                                     loff_t actual_len, u64 *alloc_hint)
7421 {
7422         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7423                                            min_size, actual_len, alloc_hint, trans);
7424 }
7425
7426 static int btrfs_set_page_dirty(struct page *page)
7427 {
7428         return __set_page_dirty_nobuffers(page);
7429 }
7430
7431 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7432 {
7433         struct btrfs_root *root = BTRFS_I(inode)->root;
7434
7435         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7436                 return -EROFS;
7437         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7438                 return -EACCES;
7439         return generic_permission(inode, mask, flags, btrfs_check_acl);
7440 }
7441
7442 static const struct inode_operations btrfs_dir_inode_operations = {
7443         .getattr        = btrfs_getattr,
7444         .lookup         = btrfs_lookup,
7445         .create         = btrfs_create,
7446         .unlink         = btrfs_unlink,
7447         .link           = btrfs_link,
7448         .mkdir          = btrfs_mkdir,
7449         .rmdir          = btrfs_rmdir,
7450         .rename         = btrfs_rename,
7451         .symlink        = btrfs_symlink,
7452         .setattr        = btrfs_setattr,
7453         .mknod          = btrfs_mknod,
7454         .setxattr       = btrfs_setxattr,
7455         .getxattr       = btrfs_getxattr,
7456         .listxattr      = btrfs_listxattr,
7457         .removexattr    = btrfs_removexattr,
7458         .permission     = btrfs_permission,
7459 };
7460 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7461         .lookup         = btrfs_lookup,
7462         .permission     = btrfs_permission,
7463 };
7464
7465 static const struct file_operations btrfs_dir_file_operations = {
7466         .llseek         = generic_file_llseek,
7467         .read           = generic_read_dir,
7468         .readdir        = btrfs_real_readdir,
7469         .unlocked_ioctl = btrfs_ioctl,
7470 #ifdef CONFIG_COMPAT
7471         .compat_ioctl   = btrfs_ioctl,
7472 #endif
7473         .release        = btrfs_release_file,
7474         .fsync          = btrfs_sync_file,
7475 };
7476
7477 static struct extent_io_ops btrfs_extent_io_ops = {
7478         .fill_delalloc = run_delalloc_range,
7479         .submit_bio_hook = btrfs_submit_bio_hook,
7480         .merge_bio_hook = btrfs_merge_bio_hook,
7481         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7482         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7483         .writepage_start_hook = btrfs_writepage_start_hook,
7484         .readpage_io_failed_hook = btrfs_io_failed_hook,
7485         .set_bit_hook = btrfs_set_bit_hook,
7486         .clear_bit_hook = btrfs_clear_bit_hook,
7487         .merge_extent_hook = btrfs_merge_extent_hook,
7488         .split_extent_hook = btrfs_split_extent_hook,
7489 };
7490
7491 /*
7492  * btrfs doesn't support the bmap operation because swapfiles
7493  * use bmap to make a mapping of extents in the file.  They assume
7494  * these extents won't change over the life of the file and they
7495  * use the bmap result to do IO directly to the drive.
7496  *
7497  * the btrfs bmap call would return logical addresses that aren't
7498  * suitable for IO and they also will change frequently as COW
7499  * operations happen.  So, swapfile + btrfs == corruption.
7500  *
7501  * For now we're avoiding this by dropping bmap.
7502  */
7503 static const struct address_space_operations btrfs_aops = {
7504         .readpage       = btrfs_readpage,
7505         .writepage      = btrfs_writepage,
7506         .writepages     = btrfs_writepages,
7507         .readpages      = btrfs_readpages,
7508         .sync_page      = block_sync_page,
7509         .direct_IO      = btrfs_direct_IO,
7510         .invalidatepage = btrfs_invalidatepage,
7511         .releasepage    = btrfs_releasepage,
7512         .set_page_dirty = btrfs_set_page_dirty,
7513         .error_remove_page = generic_error_remove_page,
7514 };
7515
7516 static const struct address_space_operations btrfs_symlink_aops = {
7517         .readpage       = btrfs_readpage,
7518         .writepage      = btrfs_writepage,
7519         .invalidatepage = btrfs_invalidatepage,
7520         .releasepage    = btrfs_releasepage,
7521 };
7522
7523 static const struct inode_operations btrfs_file_inode_operations = {
7524         .getattr        = btrfs_getattr,
7525         .setattr        = btrfs_setattr,
7526         .setxattr       = btrfs_setxattr,
7527         .getxattr       = btrfs_getxattr,
7528         .listxattr      = btrfs_listxattr,
7529         .removexattr    = btrfs_removexattr,
7530         .permission     = btrfs_permission,
7531         .fiemap         = btrfs_fiemap,
7532 };
7533 static const struct inode_operations btrfs_special_inode_operations = {
7534         .getattr        = btrfs_getattr,
7535         .setattr        = btrfs_setattr,
7536         .permission     = btrfs_permission,
7537         .setxattr       = btrfs_setxattr,
7538         .getxattr       = btrfs_getxattr,
7539         .listxattr      = btrfs_listxattr,
7540         .removexattr    = btrfs_removexattr,
7541 };
7542 static const struct inode_operations btrfs_symlink_inode_operations = {
7543         .readlink       = generic_readlink,
7544         .follow_link    = page_follow_link_light,
7545         .put_link       = page_put_link,
7546         .getattr        = btrfs_getattr,
7547         .permission     = btrfs_permission,
7548         .setxattr       = btrfs_setxattr,
7549         .getxattr       = btrfs_getxattr,
7550         .listxattr      = btrfs_listxattr,
7551         .removexattr    = btrfs_removexattr,
7552 };
7553
7554 const struct dentry_operations btrfs_dentry_operations = {
7555         .d_delete       = btrfs_dentry_delete,
7556 };