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         struct extent_buffer *leaf;
4225         int slot;
4226         unsigned char d_type;
4227         int over = 0;
4228         u32 di_cur;
4229         u32 di_total;
4230         u32 di_len;
4231         int key_type = BTRFS_DIR_INDEX_KEY;
4232         char tmp_name[32];
4233         char *name_ptr;
4234         int name_len;
4235
4236         /* FIXME, use a real flag for deciding about the key type */
4237         if (root->fs_info->tree_root == root)
4238                 key_type = BTRFS_DIR_ITEM_KEY;
4239
4240         /* special case for "." */
4241         if (filp->f_pos == 0) {
4242                 over = filldir(dirent, ".", 1,
4243                                1, inode->i_ino,
4244                                DT_DIR);
4245                 if (over)
4246                         return 0;
4247                 filp->f_pos = 1;
4248         }
4249         /* special case for .., just use the back ref */
4250         if (filp->f_pos == 1) {
4251                 u64 pino = parent_ino(filp->f_path.dentry);
4252                 over = filldir(dirent, "..", 2,
4253                                2, pino, DT_DIR);
4254                 if (over)
4255                         return 0;
4256                 filp->f_pos = 2;
4257         }
4258         path = btrfs_alloc_path();
4259         path->reada = 2;
4260
4261         btrfs_set_key_type(&key, key_type);
4262         key.offset = filp->f_pos;
4263         key.objectid = inode->i_ino;
4264
4265         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4266         if (ret < 0)
4267                 goto err;
4268
4269         while (1) {
4270                 leaf = path->nodes[0];
4271                 slot = path->slots[0];
4272                 if (slot >= btrfs_header_nritems(leaf)) {
4273                         ret = btrfs_next_leaf(root, path);
4274                         if (ret < 0)
4275                                 goto err;
4276                         else if (ret > 0)
4277                                 break;
4278                         continue;
4279                 }
4280
4281                 item = btrfs_item_nr(leaf, slot);
4282                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4283
4284                 if (found_key.objectid != key.objectid)
4285                         break;
4286                 if (btrfs_key_type(&found_key) != key_type)
4287                         break;
4288                 if (found_key.offset < filp->f_pos)
4289                         goto next;
4290
4291                 filp->f_pos = found_key.offset;
4292
4293                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4294                 di_cur = 0;
4295                 di_total = btrfs_item_size(leaf, item);
4296
4297                 while (di_cur < di_total) {
4298                         struct btrfs_key location;
4299
4300                         if (verify_dir_item(root, leaf, di))
4301                                 break;
4302
4303                         name_len = btrfs_dir_name_len(leaf, di);
4304                         if (name_len <= sizeof(tmp_name)) {
4305                                 name_ptr = tmp_name;
4306                         } else {
4307                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4308                                 if (!name_ptr) {
4309                                         ret = -ENOMEM;
4310                                         goto err;
4311                                 }
4312                         }
4313                         read_extent_buffer(leaf, name_ptr,
4314                                            (unsigned long)(di + 1), name_len);
4315
4316                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4317                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4318
4319                         /* is this a reference to our own snapshot? If so
4320                          * skip it
4321                          */
4322                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4323                             location.objectid == root->root_key.objectid) {
4324                                 over = 0;
4325                                 goto skip;
4326                         }
4327                         over = filldir(dirent, name_ptr, name_len,
4328                                        found_key.offset, location.objectid,
4329                                        d_type);
4330
4331 skip:
4332                         if (name_ptr != tmp_name)
4333                                 kfree(name_ptr);
4334
4335                         if (over)
4336                                 goto nopos;
4337                         di_len = btrfs_dir_name_len(leaf, di) +
4338                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4339                         di_cur += di_len;
4340                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4341                 }
4342 next:
4343                 path->slots[0]++;
4344         }
4345
4346         /* Reached end of directory/root. Bump pos past the last item. */
4347         if (key_type == BTRFS_DIR_INDEX_KEY)
4348                 /*
4349                  * 32-bit glibc will use getdents64, but then strtol -
4350                  * so the last number we can serve is this.
4351                  */
4352                 filp->f_pos = 0x7fffffff;
4353         else
4354                 filp->f_pos++;
4355 nopos:
4356         ret = 0;
4357 err:
4358         btrfs_free_path(path);
4359         return ret;
4360 }
4361
4362 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4363 {
4364         struct btrfs_root *root = BTRFS_I(inode)->root;
4365         struct btrfs_trans_handle *trans;
4366         int ret = 0;
4367         bool nolock = false;
4368
4369         if (BTRFS_I(inode)->dummy_inode)
4370                 return 0;
4371
4372         smp_mb();
4373         nolock = (root->fs_info->closing && root == root->fs_info->tree_root);
4374
4375         if (wbc->sync_mode == WB_SYNC_ALL) {
4376                 if (nolock)
4377                         trans = btrfs_join_transaction_nolock(root, 1);
4378                 else
4379                         trans = btrfs_join_transaction(root, 1);
4380                 if (IS_ERR(trans))
4381                         return PTR_ERR(trans);
4382                 btrfs_set_trans_block_group(trans, inode);
4383                 if (nolock)
4384                         ret = btrfs_end_transaction_nolock(trans, root);
4385                 else
4386                         ret = btrfs_commit_transaction(trans, root);
4387         }
4388         return ret;
4389 }
4390
4391 /*
4392  * This is somewhat expensive, updating the tree every time the
4393  * inode changes.  But, it is most likely to find the inode in cache.
4394  * FIXME, needs more benchmarking...there are no reasons other than performance
4395  * to keep or drop this code.
4396  */
4397 void btrfs_dirty_inode(struct inode *inode)
4398 {
4399         struct btrfs_root *root = BTRFS_I(inode)->root;
4400         struct btrfs_trans_handle *trans;
4401         int ret;
4402
4403         if (BTRFS_I(inode)->dummy_inode)
4404                 return;
4405
4406         trans = btrfs_join_transaction(root, 1);
4407         BUG_ON(IS_ERR(trans));
4408         btrfs_set_trans_block_group(trans, inode);
4409
4410         ret = btrfs_update_inode(trans, root, inode);
4411         if (ret && ret == -ENOSPC) {
4412                 /* whoops, lets try again with the full transaction */
4413                 btrfs_end_transaction(trans, root);
4414                 trans = btrfs_start_transaction(root, 1);
4415                 if (IS_ERR(trans)) {
4416                         if (printk_ratelimit()) {
4417                                 printk(KERN_ERR "btrfs: fail to "
4418                                        "dirty  inode %lu error %ld\n",
4419                                        inode->i_ino, PTR_ERR(trans));
4420                         }
4421                         return;
4422                 }
4423                 btrfs_set_trans_block_group(trans, inode);
4424
4425                 ret = btrfs_update_inode(trans, root, inode);
4426                 if (ret) {
4427                         if (printk_ratelimit()) {
4428                                 printk(KERN_ERR "btrfs: fail to "
4429                                        "dirty  inode %lu error %d\n",
4430                                        inode->i_ino, ret);
4431                         }
4432                 }
4433         }
4434         btrfs_end_transaction(trans, root);
4435 }
4436
4437 /*
4438  * find the highest existing sequence number in a directory
4439  * and then set the in-memory index_cnt variable to reflect
4440  * free sequence numbers
4441  */
4442 static int btrfs_set_inode_index_count(struct inode *inode)
4443 {
4444         struct btrfs_root *root = BTRFS_I(inode)->root;
4445         struct btrfs_key key, found_key;
4446         struct btrfs_path *path;
4447         struct extent_buffer *leaf;
4448         int ret;
4449
4450         key.objectid = inode->i_ino;
4451         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4452         key.offset = (u64)-1;
4453
4454         path = btrfs_alloc_path();
4455         if (!path)
4456                 return -ENOMEM;
4457
4458         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4459         if (ret < 0)
4460                 goto out;
4461         /* FIXME: we should be able to handle this */
4462         if (ret == 0)
4463                 goto out;
4464         ret = 0;
4465
4466         /*
4467          * MAGIC NUMBER EXPLANATION:
4468          * since we search a directory based on f_pos we have to start at 2
4469          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4470          * else has to start at 2
4471          */
4472         if (path->slots[0] == 0) {
4473                 BTRFS_I(inode)->index_cnt = 2;
4474                 goto out;
4475         }
4476
4477         path->slots[0]--;
4478
4479         leaf = path->nodes[0];
4480         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4481
4482         if (found_key.objectid != inode->i_ino ||
4483             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4484                 BTRFS_I(inode)->index_cnt = 2;
4485                 goto out;
4486         }
4487
4488         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4489 out:
4490         btrfs_free_path(path);
4491         return ret;
4492 }
4493
4494 /*
4495  * helper to find a free sequence number in a given directory.  This current
4496  * code is very simple, later versions will do smarter things in the btree
4497  */
4498 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4499 {
4500         int ret = 0;
4501
4502         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4503                 ret = btrfs_set_inode_index_count(dir);
4504                 if (ret)
4505                         return ret;
4506         }
4507
4508         *index = BTRFS_I(dir)->index_cnt;
4509         BTRFS_I(dir)->index_cnt++;
4510
4511         return ret;
4512 }
4513
4514 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4515                                      struct btrfs_root *root,
4516                                      struct inode *dir,
4517                                      const char *name, int name_len,
4518                                      u64 ref_objectid, u64 objectid,
4519                                      u64 alloc_hint, int mode, u64 *index)
4520 {
4521         struct inode *inode;
4522         struct btrfs_inode_item *inode_item;
4523         struct btrfs_key *location;
4524         struct btrfs_path *path;
4525         struct btrfs_inode_ref *ref;
4526         struct btrfs_key key[2];
4527         u32 sizes[2];
4528         unsigned long ptr;
4529         int ret;
4530         int owner;
4531
4532         path = btrfs_alloc_path();
4533         BUG_ON(!path);
4534
4535         inode = new_inode(root->fs_info->sb);
4536         if (!inode) {
4537                 btrfs_free_path(path);
4538                 return ERR_PTR(-ENOMEM);
4539         }
4540
4541         if (dir) {
4542                 trace_btrfs_inode_request(dir);
4543
4544                 ret = btrfs_set_inode_index(dir, index);
4545                 if (ret) {
4546                         btrfs_free_path(path);
4547                         iput(inode);
4548                         return ERR_PTR(ret);
4549                 }
4550         }
4551         /*
4552          * index_cnt is ignored for everything but a dir,
4553          * btrfs_get_inode_index_count has an explanation for the magic
4554          * number
4555          */
4556         BTRFS_I(inode)->index_cnt = 2;
4557         BTRFS_I(inode)->root = root;
4558         BTRFS_I(inode)->generation = trans->transid;
4559         inode->i_generation = BTRFS_I(inode)->generation;
4560         btrfs_set_inode_space_info(root, inode);
4561
4562         if (mode & S_IFDIR)
4563                 owner = 0;
4564         else
4565                 owner = 1;
4566         BTRFS_I(inode)->block_group =
4567                         btrfs_find_block_group(root, 0, alloc_hint, owner);
4568
4569         key[0].objectid = objectid;
4570         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4571         key[0].offset = 0;
4572
4573         key[1].objectid = objectid;
4574         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4575         key[1].offset = ref_objectid;
4576
4577         sizes[0] = sizeof(struct btrfs_inode_item);
4578         sizes[1] = name_len + sizeof(*ref);
4579
4580         path->leave_spinning = 1;
4581         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4582         if (ret != 0)
4583                 goto fail;
4584
4585         inode_init_owner(inode, dir, mode);
4586         inode->i_ino = objectid;
4587         inode_set_bytes(inode, 0);
4588         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4589         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4590                                   struct btrfs_inode_item);
4591         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4592
4593         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4594                              struct btrfs_inode_ref);
4595         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4596         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4597         ptr = (unsigned long)(ref + 1);
4598         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4599
4600         btrfs_mark_buffer_dirty(path->nodes[0]);
4601         btrfs_free_path(path);
4602
4603         location = &BTRFS_I(inode)->location;
4604         location->objectid = objectid;
4605         location->offset = 0;
4606         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4607
4608         btrfs_inherit_iflags(inode, dir);
4609
4610         if ((mode & S_IFREG)) {
4611                 if (btrfs_test_opt(root, NODATASUM))
4612                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4613                 if (btrfs_test_opt(root, NODATACOW) ||
4614                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4615                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4616         }
4617
4618         insert_inode_hash(inode);
4619         inode_tree_add(inode);
4620
4621         trace_btrfs_inode_new(inode);
4622
4623         return inode;
4624 fail:
4625         if (dir)
4626                 BTRFS_I(dir)->index_cnt--;
4627         btrfs_free_path(path);
4628         iput(inode);
4629         return ERR_PTR(ret);
4630 }
4631
4632 static inline u8 btrfs_inode_type(struct inode *inode)
4633 {
4634         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4635 }
4636
4637 /*
4638  * utility function to add 'inode' into 'parent_inode' with
4639  * a give name and a given sequence number.
4640  * if 'add_backref' is true, also insert a backref from the
4641  * inode to the parent directory.
4642  */
4643 int btrfs_add_link(struct btrfs_trans_handle *trans,
4644                    struct inode *parent_inode, struct inode *inode,
4645                    const char *name, int name_len, int add_backref, u64 index)
4646 {
4647         int ret = 0;
4648         struct btrfs_key key;
4649         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4650
4651         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4652                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4653         } else {
4654                 key.objectid = inode->i_ino;
4655                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4656                 key.offset = 0;
4657         }
4658
4659         if (unlikely(inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
4660                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4661                                          key.objectid, root->root_key.objectid,
4662                                          parent_inode->i_ino,
4663                                          index, name, name_len);
4664         } else if (add_backref) {
4665                 ret = btrfs_insert_inode_ref(trans, root,
4666                                              name, name_len, inode->i_ino,
4667                                              parent_inode->i_ino, index);
4668         }
4669
4670         if (ret == 0) {
4671                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4672                                             parent_inode->i_ino, &key,
4673                                             btrfs_inode_type(inode), index);
4674                 BUG_ON(ret);
4675
4676                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4677                                    name_len * 2);
4678                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4679                 ret = btrfs_update_inode(trans, root, parent_inode);
4680         }
4681         return ret;
4682 }
4683
4684 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4685                             struct inode *dir, struct dentry *dentry,
4686                             struct inode *inode, int backref, u64 index)
4687 {
4688         int err = btrfs_add_link(trans, dir, inode,
4689                                  dentry->d_name.name, dentry->d_name.len,
4690                                  backref, index);
4691         if (!err) {
4692                 d_instantiate(dentry, inode);
4693                 return 0;
4694         }
4695         if (err > 0)
4696                 err = -EEXIST;
4697         return err;
4698 }
4699
4700 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4701                         int mode, dev_t rdev)
4702 {
4703         struct btrfs_trans_handle *trans;
4704         struct btrfs_root *root = BTRFS_I(dir)->root;
4705         struct inode *inode = NULL;
4706         int err;
4707         int drop_inode = 0;
4708         u64 objectid;
4709         unsigned long nr = 0;
4710         u64 index = 0;
4711
4712         if (!new_valid_dev(rdev))
4713                 return -EINVAL;
4714
4715         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4716         if (err)
4717                 return err;
4718
4719         /*
4720          * 2 for inode item and ref
4721          * 2 for dir items
4722          * 1 for xattr if selinux is on
4723          */
4724         trans = btrfs_start_transaction(root, 5);
4725         if (IS_ERR(trans))
4726                 return PTR_ERR(trans);
4727
4728         btrfs_set_trans_block_group(trans, dir);
4729
4730         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4731                                 dentry->d_name.len, dir->i_ino, objectid,
4732                                 BTRFS_I(dir)->block_group, mode, &index);
4733         err = PTR_ERR(inode);
4734         if (IS_ERR(inode))
4735                 goto out_unlock;
4736
4737         err = btrfs_init_inode_security(trans, inode, dir);
4738         if (err) {
4739                 drop_inode = 1;
4740                 goto out_unlock;
4741         }
4742
4743         btrfs_set_trans_block_group(trans, inode);
4744         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4745         if (err)
4746                 drop_inode = 1;
4747         else {
4748                 inode->i_op = &btrfs_special_inode_operations;
4749                 init_special_inode(inode, inode->i_mode, rdev);
4750                 btrfs_update_inode(trans, root, inode);
4751         }
4752         btrfs_update_inode_block_group(trans, inode);
4753         btrfs_update_inode_block_group(trans, dir);
4754 out_unlock:
4755         nr = trans->blocks_used;
4756         btrfs_end_transaction_throttle(trans, root);
4757         btrfs_btree_balance_dirty(root, nr);
4758         if (drop_inode) {
4759                 inode_dec_link_count(inode);
4760                 iput(inode);
4761         }
4762         return err;
4763 }
4764
4765 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4766                         int mode, struct nameidata *nd)
4767 {
4768         struct btrfs_trans_handle *trans;
4769         struct btrfs_root *root = BTRFS_I(dir)->root;
4770         struct inode *inode = NULL;
4771         int drop_inode = 0;
4772         int err;
4773         unsigned long nr = 0;
4774         u64 objectid;
4775         u64 index = 0;
4776
4777         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4778         if (err)
4779                 return err;
4780         /*
4781          * 2 for inode item and ref
4782          * 2 for dir items
4783          * 1 for xattr if selinux is on
4784          */
4785         trans = btrfs_start_transaction(root, 5);
4786         if (IS_ERR(trans))
4787                 return PTR_ERR(trans);
4788
4789         btrfs_set_trans_block_group(trans, dir);
4790
4791         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4792                                 dentry->d_name.len, dir->i_ino, objectid,
4793                                 BTRFS_I(dir)->block_group, mode, &index);
4794         err = PTR_ERR(inode);
4795         if (IS_ERR(inode))
4796                 goto out_unlock;
4797
4798         err = btrfs_init_inode_security(trans, inode, dir);
4799         if (err) {
4800                 drop_inode = 1;
4801                 goto out_unlock;
4802         }
4803
4804         btrfs_set_trans_block_group(trans, inode);
4805         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4806         if (err)
4807                 drop_inode = 1;
4808         else {
4809                 inode->i_mapping->a_ops = &btrfs_aops;
4810                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4811                 inode->i_fop = &btrfs_file_operations;
4812                 inode->i_op = &btrfs_file_inode_operations;
4813                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4814         }
4815         btrfs_update_inode_block_group(trans, inode);
4816         btrfs_update_inode_block_group(trans, dir);
4817 out_unlock:
4818         nr = trans->blocks_used;
4819         btrfs_end_transaction_throttle(trans, root);
4820         if (drop_inode) {
4821                 inode_dec_link_count(inode);
4822                 iput(inode);
4823         }
4824         btrfs_btree_balance_dirty(root, nr);
4825         return err;
4826 }
4827
4828 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4829                       struct dentry *dentry)
4830 {
4831         struct btrfs_trans_handle *trans;
4832         struct btrfs_root *root = BTRFS_I(dir)->root;
4833         struct inode *inode = old_dentry->d_inode;
4834         u64 index;
4835         unsigned long nr = 0;
4836         int err;
4837         int drop_inode = 0;
4838
4839         if (inode->i_nlink == 0)
4840                 return -ENOENT;
4841
4842         /* do not allow sys_link's with other subvols of the same device */
4843         if (root->objectid != BTRFS_I(inode)->root->objectid)
4844                 return -EXDEV;
4845
4846         if (inode->i_nlink == ~0U)
4847                 return -EMLINK;
4848
4849         err = btrfs_set_inode_index(dir, &index);
4850         if (err)
4851                 goto fail;
4852
4853         /*
4854          * 2 items for inode and inode ref
4855          * 2 items for dir items
4856          * 1 item for parent inode
4857          */
4858         trans = btrfs_start_transaction(root, 5);
4859         if (IS_ERR(trans)) {
4860                 err = PTR_ERR(trans);
4861                 goto fail;
4862         }
4863
4864         btrfs_inc_nlink(inode);
4865         inode->i_ctime = CURRENT_TIME;
4866
4867         btrfs_set_trans_block_group(trans, dir);
4868         ihold(inode);
4869
4870         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4871
4872         if (err) {
4873                 drop_inode = 1;
4874         } else {
4875                 struct dentry *parent = dget_parent(dentry);
4876                 btrfs_update_inode_block_group(trans, dir);
4877                 err = btrfs_update_inode(trans, root, inode);
4878                 BUG_ON(err);
4879                 btrfs_log_new_name(trans, inode, NULL, parent);
4880                 dput(parent);
4881         }
4882
4883         nr = trans->blocks_used;
4884         btrfs_end_transaction_throttle(trans, root);
4885 fail:
4886         if (drop_inode) {
4887                 inode_dec_link_count(inode);
4888                 iput(inode);
4889         }
4890         btrfs_btree_balance_dirty(root, nr);
4891         return err;
4892 }
4893
4894 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4895 {
4896         struct inode *inode = NULL;
4897         struct btrfs_trans_handle *trans;
4898         struct btrfs_root *root = BTRFS_I(dir)->root;
4899         int err = 0;
4900         int drop_on_err = 0;
4901         u64 objectid = 0;
4902         u64 index = 0;
4903         unsigned long nr = 1;
4904
4905         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
4906         if (err)
4907                 return err;
4908
4909         /*
4910          * 2 items for inode and ref
4911          * 2 items for dir items
4912          * 1 for xattr if selinux is on
4913          */
4914         trans = btrfs_start_transaction(root, 5);
4915         if (IS_ERR(trans))
4916                 return PTR_ERR(trans);
4917         btrfs_set_trans_block_group(trans, dir);
4918
4919         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4920                                 dentry->d_name.len, dir->i_ino, objectid,
4921                                 BTRFS_I(dir)->block_group, S_IFDIR | mode,
4922                                 &index);
4923         if (IS_ERR(inode)) {
4924                 err = PTR_ERR(inode);
4925                 goto out_fail;
4926         }
4927
4928         drop_on_err = 1;
4929
4930         err = btrfs_init_inode_security(trans, inode, dir);
4931         if (err)
4932                 goto out_fail;
4933
4934         inode->i_op = &btrfs_dir_inode_operations;
4935         inode->i_fop = &btrfs_dir_file_operations;
4936         btrfs_set_trans_block_group(trans, inode);
4937
4938         btrfs_i_size_write(inode, 0);
4939         err = btrfs_update_inode(trans, root, inode);
4940         if (err)
4941                 goto out_fail;
4942
4943         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4944                              dentry->d_name.len, 0, index);
4945         if (err)
4946                 goto out_fail;
4947
4948         d_instantiate(dentry, inode);
4949         drop_on_err = 0;
4950         btrfs_update_inode_block_group(trans, inode);
4951         btrfs_update_inode_block_group(trans, dir);
4952
4953 out_fail:
4954         nr = trans->blocks_used;
4955         btrfs_end_transaction_throttle(trans, root);
4956         if (drop_on_err)
4957                 iput(inode);
4958         btrfs_btree_balance_dirty(root, nr);
4959         return err;
4960 }
4961
4962 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4963  * and an extent that you want to insert, deal with overlap and insert
4964  * the new extent into the tree.
4965  */
4966 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4967                                 struct extent_map *existing,
4968                                 struct extent_map *em,
4969                                 u64 map_start, u64 map_len)
4970 {
4971         u64 start_diff;
4972
4973         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4974         start_diff = map_start - em->start;
4975         em->start = map_start;
4976         em->len = map_len;
4977         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4978             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4979                 em->block_start += start_diff;
4980                 em->block_len -= start_diff;
4981         }
4982         return add_extent_mapping(em_tree, em);
4983 }
4984
4985 static noinline int uncompress_inline(struct btrfs_path *path,
4986                                       struct inode *inode, struct page *page,
4987                                       size_t pg_offset, u64 extent_offset,
4988                                       struct btrfs_file_extent_item *item)
4989 {
4990         int ret;
4991         struct extent_buffer *leaf = path->nodes[0];
4992         char *tmp;
4993         size_t max_size;
4994         unsigned long inline_size;
4995         unsigned long ptr;
4996         int compress_type;
4997
4998         WARN_ON(pg_offset != 0);
4999         compress_type = btrfs_file_extent_compression(leaf, item);
5000         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5001         inline_size = btrfs_file_extent_inline_item_len(leaf,
5002                                         btrfs_item_nr(leaf, path->slots[0]));
5003         tmp = kmalloc(inline_size, GFP_NOFS);
5004         ptr = btrfs_file_extent_inline_start(item);
5005
5006         read_extent_buffer(leaf, tmp, ptr, inline_size);
5007
5008         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5009         ret = btrfs_decompress(compress_type, tmp, page,
5010                                extent_offset, inline_size, max_size);
5011         if (ret) {
5012                 char *kaddr = kmap_atomic(page, KM_USER0);
5013                 unsigned long copy_size = min_t(u64,
5014                                   PAGE_CACHE_SIZE - pg_offset,
5015                                   max_size - extent_offset);
5016                 memset(kaddr + pg_offset, 0, copy_size);
5017                 kunmap_atomic(kaddr, KM_USER0);
5018         }
5019         kfree(tmp);
5020         return 0;
5021 }
5022
5023 /*
5024  * a bit scary, this does extent mapping from logical file offset to the disk.
5025  * the ugly parts come from merging extents from the disk with the in-ram
5026  * representation.  This gets more complex because of the data=ordered code,
5027  * where the in-ram extents might be locked pending data=ordered completion.
5028  *
5029  * This also copies inline extents directly into the page.
5030  */
5031
5032 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5033                                     size_t pg_offset, u64 start, u64 len,
5034                                     int create)
5035 {
5036         int ret;
5037         int err = 0;
5038         u64 bytenr;
5039         u64 extent_start = 0;
5040         u64 extent_end = 0;
5041         u64 objectid = inode->i_ino;
5042         u32 found_type;
5043         struct btrfs_path *path = NULL;
5044         struct btrfs_root *root = BTRFS_I(inode)->root;
5045         struct btrfs_file_extent_item *item;
5046         struct extent_buffer *leaf;
5047         struct btrfs_key found_key;
5048         struct extent_map *em = NULL;
5049         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5050         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5051         struct btrfs_trans_handle *trans = NULL;
5052         int compress_type;
5053
5054 again:
5055         read_lock(&em_tree->lock);
5056         em = lookup_extent_mapping(em_tree, start, len);
5057         if (em)
5058                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5059         read_unlock(&em_tree->lock);
5060
5061         if (em) {
5062                 if (em->start > start || em->start + em->len <= start)
5063                         free_extent_map(em);
5064                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5065                         free_extent_map(em);
5066                 else
5067                         goto out;
5068         }
5069         em = alloc_extent_map(GFP_NOFS);
5070         if (!em) {
5071                 err = -ENOMEM;
5072                 goto out;
5073         }
5074         em->bdev = root->fs_info->fs_devices->latest_bdev;
5075         em->start = EXTENT_MAP_HOLE;
5076         em->orig_start = EXTENT_MAP_HOLE;
5077         em->len = (u64)-1;
5078         em->block_len = (u64)-1;
5079
5080         if (!path) {
5081                 path = btrfs_alloc_path();
5082                 BUG_ON(!path);
5083         }
5084
5085         ret = btrfs_lookup_file_extent(trans, root, path,
5086                                        objectid, start, trans != NULL);
5087         if (ret < 0) {
5088                 err = ret;
5089                 goto out;
5090         }
5091
5092         if (ret != 0) {
5093                 if (path->slots[0] == 0)
5094                         goto not_found;
5095                 path->slots[0]--;
5096         }
5097
5098         leaf = path->nodes[0];
5099         item = btrfs_item_ptr(leaf, path->slots[0],
5100                               struct btrfs_file_extent_item);
5101         /* are we inside the extent that was found? */
5102         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5103         found_type = btrfs_key_type(&found_key);
5104         if (found_key.objectid != objectid ||
5105             found_type != BTRFS_EXTENT_DATA_KEY) {
5106                 goto not_found;
5107         }
5108
5109         found_type = btrfs_file_extent_type(leaf, item);
5110         extent_start = found_key.offset;
5111         compress_type = btrfs_file_extent_compression(leaf, item);
5112         if (found_type == BTRFS_FILE_EXTENT_REG ||
5113             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5114                 extent_end = extent_start +
5115                        btrfs_file_extent_num_bytes(leaf, item);
5116         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5117                 size_t size;
5118                 size = btrfs_file_extent_inline_len(leaf, item);
5119                 extent_end = (extent_start + size + root->sectorsize - 1) &
5120                         ~((u64)root->sectorsize - 1);
5121         }
5122
5123         if (start >= extent_end) {
5124                 path->slots[0]++;
5125                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5126                         ret = btrfs_next_leaf(root, path);
5127                         if (ret < 0) {
5128                                 err = ret;
5129                                 goto out;
5130                         }
5131                         if (ret > 0)
5132                                 goto not_found;
5133                         leaf = path->nodes[0];
5134                 }
5135                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5136                 if (found_key.objectid != objectid ||
5137                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5138                         goto not_found;
5139                 if (start + len <= found_key.offset)
5140                         goto not_found;
5141                 em->start = start;
5142                 em->len = found_key.offset - start;
5143                 goto not_found_em;
5144         }
5145
5146         if (found_type == BTRFS_FILE_EXTENT_REG ||
5147             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5148                 em->start = extent_start;
5149                 em->len = extent_end - extent_start;
5150                 em->orig_start = extent_start -
5151                                  btrfs_file_extent_offset(leaf, item);
5152                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5153                 if (bytenr == 0) {
5154                         em->block_start = EXTENT_MAP_HOLE;
5155                         goto insert;
5156                 }
5157                 if (compress_type != BTRFS_COMPRESS_NONE) {
5158                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5159                         em->compress_type = compress_type;
5160                         em->block_start = bytenr;
5161                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5162                                                                          item);
5163                 } else {
5164                         bytenr += btrfs_file_extent_offset(leaf, item);
5165                         em->block_start = bytenr;
5166                         em->block_len = em->len;
5167                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5168                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5169                 }
5170                 goto insert;
5171         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5172                 unsigned long ptr;
5173                 char *map;
5174                 size_t size;
5175                 size_t extent_offset;
5176                 size_t copy_size;
5177
5178                 em->block_start = EXTENT_MAP_INLINE;
5179                 if (!page || create) {
5180                         em->start = extent_start;
5181                         em->len = extent_end - extent_start;
5182                         goto out;
5183                 }
5184
5185                 size = btrfs_file_extent_inline_len(leaf, item);
5186                 extent_offset = page_offset(page) + pg_offset - extent_start;
5187                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5188                                 size - extent_offset);
5189                 em->start = extent_start + extent_offset;
5190                 em->len = (copy_size + root->sectorsize - 1) &
5191                         ~((u64)root->sectorsize - 1);
5192                 em->orig_start = EXTENT_MAP_INLINE;
5193                 if (compress_type) {
5194                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5195                         em->compress_type = compress_type;
5196                 }
5197                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5198                 if (create == 0 && !PageUptodate(page)) {
5199                         if (btrfs_file_extent_compression(leaf, item) !=
5200                             BTRFS_COMPRESS_NONE) {
5201                                 ret = uncompress_inline(path, inode, page,
5202                                                         pg_offset,
5203                                                         extent_offset, item);
5204                                 BUG_ON(ret);
5205                         } else {
5206                                 map = kmap(page);
5207                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5208                                                    copy_size);
5209                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5210                                         memset(map + pg_offset + copy_size, 0,
5211                                                PAGE_CACHE_SIZE - pg_offset -
5212                                                copy_size);
5213                                 }
5214                                 kunmap(page);
5215                         }
5216                         flush_dcache_page(page);
5217                 } else if (create && PageUptodate(page)) {
5218                         WARN_ON(1);
5219                         if (!trans) {
5220                                 kunmap(page);
5221                                 free_extent_map(em);
5222                                 em = NULL;
5223                                 btrfs_release_path(root, path);
5224                                 trans = btrfs_join_transaction(root, 1);
5225                                 if (IS_ERR(trans))
5226                                         return ERR_CAST(trans);
5227                                 goto again;
5228                         }
5229                         map = kmap(page);
5230                         write_extent_buffer(leaf, map + pg_offset, ptr,
5231                                             copy_size);
5232                         kunmap(page);
5233                         btrfs_mark_buffer_dirty(leaf);
5234                 }
5235                 set_extent_uptodate(io_tree, em->start,
5236                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5237                 goto insert;
5238         } else {
5239                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5240                 WARN_ON(1);
5241         }
5242 not_found:
5243         em->start = start;
5244         em->len = len;
5245 not_found_em:
5246         em->block_start = EXTENT_MAP_HOLE;
5247         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5248 insert:
5249         btrfs_release_path(root, path);
5250         if (em->start > start || extent_map_end(em) <= start) {
5251                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5252                        "[%llu %llu]\n", (unsigned long long)em->start,
5253                        (unsigned long long)em->len,
5254                        (unsigned long long)start,
5255                        (unsigned long long)len);
5256                 err = -EIO;
5257                 goto out;
5258         }
5259
5260         err = 0;
5261         write_lock(&em_tree->lock);
5262         ret = add_extent_mapping(em_tree, em);
5263         /* it is possible that someone inserted the extent into the tree
5264          * while we had the lock dropped.  It is also possible that
5265          * an overlapping map exists in the tree
5266          */
5267         if (ret == -EEXIST) {
5268                 struct extent_map *existing;
5269
5270                 ret = 0;
5271
5272                 existing = lookup_extent_mapping(em_tree, start, len);
5273                 if (existing && (existing->start > start ||
5274                     existing->start + existing->len <= start)) {
5275                         free_extent_map(existing);
5276                         existing = NULL;
5277                 }
5278                 if (!existing) {
5279                         existing = lookup_extent_mapping(em_tree, em->start,
5280                                                          em->len);
5281                         if (existing) {
5282                                 err = merge_extent_mapping(em_tree, existing,
5283                                                            em, start,
5284                                                            root->sectorsize);
5285                                 free_extent_map(existing);
5286                                 if (err) {
5287                                         free_extent_map(em);
5288                                         em = NULL;
5289                                 }
5290                         } else {
5291                                 err = -EIO;
5292                                 free_extent_map(em);
5293                                 em = NULL;
5294                         }
5295                 } else {
5296                         free_extent_map(em);
5297                         em = existing;
5298                         err = 0;
5299                 }
5300         }
5301         write_unlock(&em_tree->lock);
5302 out:
5303
5304         trace_btrfs_get_extent(root, em);
5305
5306         if (path)
5307                 btrfs_free_path(path);
5308         if (trans) {
5309                 ret = btrfs_end_transaction(trans, root);
5310                 if (!err)
5311                         err = ret;
5312         }
5313         if (err) {
5314                 free_extent_map(em);
5315                 return ERR_PTR(err);
5316         }
5317         return em;
5318 }
5319
5320 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5321                                            size_t pg_offset, u64 start, u64 len,
5322                                            int create)
5323 {
5324         struct extent_map *em;
5325         struct extent_map *hole_em = NULL;
5326         u64 range_start = start;
5327         u64 end;
5328         u64 found;
5329         u64 found_end;
5330         int err = 0;
5331
5332         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5333         if (IS_ERR(em))
5334                 return em;
5335         if (em) {
5336                 /*
5337                  * if our em maps to a hole, there might
5338                  * actually be delalloc bytes behind it
5339                  */
5340                 if (em->block_start != EXTENT_MAP_HOLE)
5341                         return em;
5342                 else
5343                         hole_em = em;
5344         }
5345
5346         /* check to see if we've wrapped (len == -1 or similar) */
5347         end = start + len;
5348         if (end < start)
5349                 end = (u64)-1;
5350         else
5351                 end -= 1;
5352
5353         em = NULL;
5354
5355         /* ok, we didn't find anything, lets look for delalloc */
5356         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5357                                  end, len, EXTENT_DELALLOC, 1);
5358         found_end = range_start + found;
5359         if (found_end < range_start)
5360                 found_end = (u64)-1;
5361
5362         /*
5363          * we didn't find anything useful, return
5364          * the original results from get_extent()
5365          */
5366         if (range_start > end || found_end <= start) {
5367                 em = hole_em;
5368                 hole_em = NULL;
5369                 goto out;
5370         }
5371
5372         /* adjust the range_start to make sure it doesn't
5373          * go backwards from the start they passed in
5374          */
5375         range_start = max(start,range_start);
5376         found = found_end - range_start;
5377
5378         if (found > 0) {
5379                 u64 hole_start = start;
5380                 u64 hole_len = len;
5381
5382                 em = alloc_extent_map(GFP_NOFS);
5383                 if (!em) {
5384                         err = -ENOMEM;
5385                         goto out;
5386                 }
5387                 /*
5388                  * when btrfs_get_extent can't find anything it
5389                  * returns one huge hole
5390                  *
5391                  * make sure what it found really fits our range, and
5392                  * adjust to make sure it is based on the start from
5393                  * the caller
5394                  */
5395                 if (hole_em) {
5396                         u64 calc_end = extent_map_end(hole_em);
5397
5398                         if (calc_end <= start || (hole_em->start > end)) {
5399                                 free_extent_map(hole_em);
5400                                 hole_em = NULL;
5401                         } else {
5402                                 hole_start = max(hole_em->start, start);
5403                                 hole_len = calc_end - hole_start;
5404                         }
5405                 }
5406                 em->bdev = NULL;
5407                 if (hole_em && range_start > hole_start) {
5408                         /* our hole starts before our delalloc, so we
5409                          * have to return just the parts of the hole
5410                          * that go until  the delalloc starts
5411                          */
5412                         em->len = min(hole_len,
5413                                       range_start - hole_start);
5414                         em->start = hole_start;
5415                         em->orig_start = hole_start;
5416                         /*
5417                          * don't adjust block start at all,
5418                          * it is fixed at EXTENT_MAP_HOLE
5419                          */
5420                         em->block_start = hole_em->block_start;
5421                         em->block_len = hole_len;
5422                 } else {
5423                         em->start = range_start;
5424                         em->len = found;
5425                         em->orig_start = range_start;
5426                         em->block_start = EXTENT_MAP_DELALLOC;
5427                         em->block_len = found;
5428                 }
5429         } else if (hole_em) {
5430                 return hole_em;
5431         }
5432 out:
5433
5434         free_extent_map(hole_em);
5435         if (err) {
5436                 free_extent_map(em);
5437                 return ERR_PTR(err);
5438         }
5439         return em;
5440 }
5441
5442 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5443                                                   struct extent_map *em,
5444                                                   u64 start, u64 len)
5445 {
5446         struct btrfs_root *root = BTRFS_I(inode)->root;
5447         struct btrfs_trans_handle *trans;
5448         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5449         struct btrfs_key ins;
5450         u64 alloc_hint;
5451         int ret;
5452         bool insert = false;
5453
5454         /*
5455          * Ok if the extent map we looked up is a hole and is for the exact
5456          * range we want, there is no reason to allocate a new one, however if
5457          * it is not right then we need to free this one and drop the cache for
5458          * our range.
5459          */
5460         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5461             em->len != len) {
5462                 free_extent_map(em);
5463                 em = NULL;
5464                 insert = true;
5465                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5466         }
5467
5468         trans = btrfs_join_transaction(root, 0);
5469         if (IS_ERR(trans))
5470                 return ERR_CAST(trans);
5471
5472         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5473
5474         alloc_hint = get_extent_allocation_hint(inode, start, len);
5475         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5476                                    alloc_hint, (u64)-1, &ins, 1);
5477         if (ret) {
5478                 em = ERR_PTR(ret);
5479                 goto out;
5480         }
5481
5482         if (!em) {
5483                 em = alloc_extent_map(GFP_NOFS);
5484                 if (!em) {
5485                         em = ERR_PTR(-ENOMEM);
5486                         goto out;
5487                 }
5488         }
5489
5490         em->start = start;
5491         em->orig_start = em->start;
5492         em->len = ins.offset;
5493
5494         em->block_start = ins.objectid;
5495         em->block_len = ins.offset;
5496         em->bdev = root->fs_info->fs_devices->latest_bdev;
5497
5498         /*
5499          * We need to do this because if we're using the original em we searched
5500          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5501          */
5502         em->flags = 0;
5503         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5504
5505         while (insert) {
5506                 write_lock(&em_tree->lock);
5507                 ret = add_extent_mapping(em_tree, em);
5508                 write_unlock(&em_tree->lock);
5509                 if (ret != -EEXIST)
5510                         break;
5511                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5512         }
5513
5514         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5515                                            ins.offset, ins.offset, 0);
5516         if (ret) {
5517                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5518                 em = ERR_PTR(ret);
5519         }
5520 out:
5521         btrfs_end_transaction(trans, root);
5522         return em;
5523 }
5524
5525 /*
5526  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5527  * block must be cow'd
5528  */
5529 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5530                                       struct inode *inode, u64 offset, u64 len)
5531 {
5532         struct btrfs_path *path;
5533         int ret;
5534         struct extent_buffer *leaf;
5535         struct btrfs_root *root = BTRFS_I(inode)->root;
5536         struct btrfs_file_extent_item *fi;
5537         struct btrfs_key key;
5538         u64 disk_bytenr;
5539         u64 backref_offset;
5540         u64 extent_end;
5541         u64 num_bytes;
5542         int slot;
5543         int found_type;
5544
5545         path = btrfs_alloc_path();
5546         if (!path)
5547                 return -ENOMEM;
5548
5549         ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
5550                                        offset, 0);
5551         if (ret < 0)
5552                 goto out;
5553
5554         slot = path->slots[0];
5555         if (ret == 1) {
5556                 if (slot == 0) {
5557                         /* can't find the item, must cow */
5558                         ret = 0;
5559                         goto out;
5560                 }
5561                 slot--;
5562         }
5563         ret = 0;
5564         leaf = path->nodes[0];
5565         btrfs_item_key_to_cpu(leaf, &key, slot);
5566         if (key.objectid != inode->i_ino ||
5567             key.type != BTRFS_EXTENT_DATA_KEY) {
5568                 /* not our file or wrong item type, must cow */
5569                 goto out;
5570         }
5571
5572         if (key.offset > offset) {
5573                 /* Wrong offset, must cow */
5574                 goto out;
5575         }
5576
5577         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5578         found_type = btrfs_file_extent_type(leaf, fi);
5579         if (found_type != BTRFS_FILE_EXTENT_REG &&
5580             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5581                 /* not a regular extent, must cow */
5582                 goto out;
5583         }
5584         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5585         backref_offset = btrfs_file_extent_offset(leaf, fi);
5586
5587         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5588         if (extent_end < offset + len) {
5589                 /* extent doesn't include our full range, must cow */
5590                 goto out;
5591         }
5592
5593         if (btrfs_extent_readonly(root, disk_bytenr))
5594                 goto out;
5595
5596         /*
5597          * look for other files referencing this extent, if we
5598          * find any we must cow
5599          */
5600         if (btrfs_cross_ref_exist(trans, root, inode->i_ino,
5601                                   key.offset - backref_offset, disk_bytenr))
5602                 goto out;
5603
5604         /*
5605          * adjust disk_bytenr and num_bytes to cover just the bytes
5606          * in this extent we are about to write.  If there
5607          * are any csums in that range we have to cow in order
5608          * to keep the csums correct
5609          */
5610         disk_bytenr += backref_offset;
5611         disk_bytenr += offset - key.offset;
5612         num_bytes = min(offset + len, extent_end) - offset;
5613         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5614                                 goto out;
5615         /*
5616          * all of the above have passed, it is safe to overwrite this extent
5617          * without cow
5618          */
5619         ret = 1;
5620 out:
5621         btrfs_free_path(path);
5622         return ret;
5623 }
5624
5625 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5626                                    struct buffer_head *bh_result, int create)
5627 {
5628         struct extent_map *em;
5629         struct btrfs_root *root = BTRFS_I(inode)->root;
5630         u64 start = iblock << inode->i_blkbits;
5631         u64 len = bh_result->b_size;
5632         struct btrfs_trans_handle *trans;
5633
5634         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5635         if (IS_ERR(em))
5636                 return PTR_ERR(em);
5637
5638         /*
5639          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5640          * io.  INLINE is special, and we could probably kludge it in here, but
5641          * it's still buffered so for safety lets just fall back to the generic
5642          * buffered path.
5643          *
5644          * For COMPRESSED we _have_ to read the entire extent in so we can
5645          * decompress it, so there will be buffering required no matter what we
5646          * do, so go ahead and fallback to buffered.
5647          *
5648          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5649          * to buffered IO.  Don't blame me, this is the price we pay for using
5650          * the generic code.
5651          */
5652         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5653             em->block_start == EXTENT_MAP_INLINE) {
5654                 free_extent_map(em);
5655                 return -ENOTBLK;
5656         }
5657
5658         /* Just a good old fashioned hole, return */
5659         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5660                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5661                 free_extent_map(em);
5662                 /* DIO will do one hole at a time, so just unlock a sector */
5663                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5664                               start + root->sectorsize - 1, GFP_NOFS);
5665                 return 0;
5666         }
5667
5668         /*
5669          * We don't allocate a new extent in the following cases
5670          *
5671          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5672          * existing extent.
5673          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5674          * just use the extent.
5675          *
5676          */
5677         if (!create) {
5678                 len = em->len - (start - em->start);
5679                 goto map;
5680         }
5681
5682         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5683             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5684              em->block_start != EXTENT_MAP_HOLE)) {
5685                 int type;
5686                 int ret;
5687                 u64 block_start;
5688
5689                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5690                         type = BTRFS_ORDERED_PREALLOC;
5691                 else
5692                         type = BTRFS_ORDERED_NOCOW;
5693                 len = min(len, em->len - (start - em->start));
5694                 block_start = em->block_start + (start - em->start);
5695
5696                 /*
5697                  * we're not going to log anything, but we do need
5698                  * to make sure the current transaction stays open
5699                  * while we look for nocow cross refs
5700                  */
5701                 trans = btrfs_join_transaction(root, 0);
5702                 if (IS_ERR(trans))
5703                         goto must_cow;
5704
5705                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5706                         ret = btrfs_add_ordered_extent_dio(inode, start,
5707                                            block_start, len, len, type);
5708                         btrfs_end_transaction(trans, root);
5709                         if (ret) {
5710                                 free_extent_map(em);
5711                                 return ret;
5712                         }
5713                         goto unlock;
5714                 }
5715                 btrfs_end_transaction(trans, root);
5716         }
5717 must_cow:
5718         /*
5719          * this will cow the extent, reset the len in case we changed
5720          * it above
5721          */
5722         len = bh_result->b_size;
5723         em = btrfs_new_extent_direct(inode, em, start, len);
5724         if (IS_ERR(em))
5725                 return PTR_ERR(em);
5726         len = min(len, em->len - (start - em->start));
5727 unlock:
5728         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5729                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5730                           0, NULL, GFP_NOFS);
5731 map:
5732         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5733                 inode->i_blkbits;
5734         bh_result->b_size = len;
5735         bh_result->b_bdev = em->bdev;
5736         set_buffer_mapped(bh_result);
5737         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5738                 set_buffer_new(bh_result);
5739
5740         free_extent_map(em);
5741
5742         return 0;
5743 }
5744
5745 struct btrfs_dio_private {
5746         struct inode *inode;
5747         u64 logical_offset;
5748         u64 disk_bytenr;
5749         u64 bytes;
5750         u32 *csums;
5751         void *private;
5752
5753         /* number of bios pending for this dio */
5754         atomic_t pending_bios;
5755
5756         /* IO errors */
5757         int errors;
5758
5759         struct bio *orig_bio;
5760 };
5761
5762 static void btrfs_endio_direct_read(struct bio *bio, int err)
5763 {
5764         struct btrfs_dio_private *dip = bio->bi_private;
5765         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5766         struct bio_vec *bvec = bio->bi_io_vec;
5767         struct inode *inode = dip->inode;
5768         struct btrfs_root *root = BTRFS_I(inode)->root;
5769         u64 start;
5770         u32 *private = dip->csums;
5771
5772         start = dip->logical_offset;
5773         do {
5774                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5775                         struct page *page = bvec->bv_page;
5776                         char *kaddr;
5777                         u32 csum = ~(u32)0;
5778                         unsigned long flags;
5779
5780                         local_irq_save(flags);
5781                         kaddr = kmap_atomic(page, KM_IRQ0);
5782                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5783                                                csum, bvec->bv_len);
5784                         btrfs_csum_final(csum, (char *)&csum);
5785                         kunmap_atomic(kaddr, KM_IRQ0);
5786                         local_irq_restore(flags);
5787
5788                         flush_dcache_page(bvec->bv_page);
5789                         if (csum != *private) {
5790                                 printk(KERN_ERR "btrfs csum failed ino %lu off"
5791                                       " %llu csum %u private %u\n",
5792                                       inode->i_ino, (unsigned long long)start,
5793                                       csum, *private);
5794                                 err = -EIO;
5795                         }
5796                 }
5797
5798                 start += bvec->bv_len;
5799                 private++;
5800                 bvec++;
5801         } while (bvec <= bvec_end);
5802
5803         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5804                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5805         bio->bi_private = dip->private;
5806
5807         kfree(dip->csums);
5808         kfree(dip);
5809
5810         /* If we had a csum failure make sure to clear the uptodate flag */
5811         if (err)
5812                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5813         dio_end_io(bio, err);
5814 }
5815
5816 static void btrfs_endio_direct_write(struct bio *bio, int err)
5817 {
5818         struct btrfs_dio_private *dip = bio->bi_private;
5819         struct inode *inode = dip->inode;
5820         struct btrfs_root *root = BTRFS_I(inode)->root;
5821         struct btrfs_trans_handle *trans;
5822         struct btrfs_ordered_extent *ordered = NULL;
5823         struct extent_state *cached_state = NULL;
5824         u64 ordered_offset = dip->logical_offset;
5825         u64 ordered_bytes = dip->bytes;
5826         int ret;
5827
5828         if (err)
5829                 goto out_done;
5830 again:
5831         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5832                                                    &ordered_offset,
5833                                                    ordered_bytes);
5834         if (!ret)
5835                 goto out_test;
5836
5837         BUG_ON(!ordered);
5838
5839         trans = btrfs_join_transaction(root, 1);
5840         if (IS_ERR(trans)) {
5841                 err = -ENOMEM;
5842                 goto out;
5843         }
5844         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5845
5846         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5847                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5848                 if (!ret)
5849                         ret = btrfs_update_inode(trans, root, inode);
5850                 err = ret;
5851                 goto out;
5852         }
5853
5854         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5855                          ordered->file_offset + ordered->len - 1, 0,
5856                          &cached_state, GFP_NOFS);
5857
5858         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5859                 ret = btrfs_mark_extent_written(trans, inode,
5860                                                 ordered->file_offset,
5861                                                 ordered->file_offset +
5862                                                 ordered->len);
5863                 if (ret) {
5864                         err = ret;
5865                         goto out_unlock;
5866                 }
5867         } else {
5868                 ret = insert_reserved_file_extent(trans, inode,
5869                                                   ordered->file_offset,
5870                                                   ordered->start,
5871                                                   ordered->disk_len,
5872                                                   ordered->len,
5873                                                   ordered->len,
5874                                                   0, 0, 0,
5875                                                   BTRFS_FILE_EXTENT_REG);
5876                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5877                                    ordered->file_offset, ordered->len);
5878                 if (ret) {
5879                         err = ret;
5880                         WARN_ON(1);
5881                         goto out_unlock;
5882                 }
5883         }
5884
5885         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5886         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5887         if (!ret)
5888                 btrfs_update_inode(trans, root, inode);
5889         ret = 0;
5890 out_unlock:
5891         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5892                              ordered->file_offset + ordered->len - 1,
5893                              &cached_state, GFP_NOFS);
5894 out:
5895         btrfs_delalloc_release_metadata(inode, ordered->len);
5896         btrfs_end_transaction(trans, root);
5897         ordered_offset = ordered->file_offset + ordered->len;
5898         btrfs_put_ordered_extent(ordered);
5899         btrfs_put_ordered_extent(ordered);
5900
5901 out_test:
5902         /*
5903          * our bio might span multiple ordered extents.  If we haven't
5904          * completed the accounting for the whole dio, go back and try again
5905          */
5906         if (ordered_offset < dip->logical_offset + dip->bytes) {
5907                 ordered_bytes = dip->logical_offset + dip->bytes -
5908                         ordered_offset;
5909                 goto again;
5910         }
5911 out_done:
5912         bio->bi_private = dip->private;
5913
5914         kfree(dip->csums);
5915         kfree(dip);
5916
5917         /* If we had an error make sure to clear the uptodate flag */
5918         if (err)
5919                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5920         dio_end_io(bio, err);
5921 }
5922
5923 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5924                                     struct bio *bio, int mirror_num,
5925                                     unsigned long bio_flags, u64 offset)
5926 {
5927         int ret;
5928         struct btrfs_root *root = BTRFS_I(inode)->root;
5929         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5930         BUG_ON(ret);
5931         return 0;
5932 }
5933
5934 static void btrfs_end_dio_bio(struct bio *bio, int err)
5935 {
5936         struct btrfs_dio_private *dip = bio->bi_private;
5937
5938         if (err) {
5939                 printk(KERN_ERR "btrfs direct IO failed ino %lu rw %lu "
5940                       "sector %#Lx len %u err no %d\n",
5941                       dip->inode->i_ino, bio->bi_rw,
5942                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5943                 dip->errors = 1;
5944
5945                 /*
5946                  * before atomic variable goto zero, we must make sure
5947                  * dip->errors is perceived to be set.
5948                  */
5949                 smp_mb__before_atomic_dec();
5950         }
5951
5952         /* if there are more bios still pending for this dio, just exit */
5953         if (!atomic_dec_and_test(&dip->pending_bios))
5954                 goto out;
5955
5956         if (dip->errors)
5957                 bio_io_error(dip->orig_bio);
5958         else {
5959                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5960                 bio_endio(dip->orig_bio, 0);
5961         }
5962 out:
5963         bio_put(bio);
5964 }
5965
5966 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5967                                        u64 first_sector, gfp_t gfp_flags)
5968 {
5969         int nr_vecs = bio_get_nr_vecs(bdev);
5970         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5971 }
5972
5973 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5974                                          int rw, u64 file_offset, int skip_sum,
5975                                          u32 *csums, int async_submit)
5976 {
5977         int write = rw & REQ_WRITE;
5978         struct btrfs_root *root = BTRFS_I(inode)->root;
5979         int ret;
5980
5981         bio_get(bio);
5982         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5983         if (ret)
5984                 goto err;
5985
5986         if (skip_sum)
5987                 goto map;
5988
5989         if (write && async_submit) {
5990                 ret = btrfs_wq_submit_bio(root->fs_info,
5991                                    inode, rw, bio, 0, 0,
5992                                    file_offset,
5993                                    __btrfs_submit_bio_start_direct_io,
5994                                    __btrfs_submit_bio_done);
5995                 goto err;
5996         } else if (write) {
5997                 /*
5998                  * If we aren't doing async submit, calculate the csum of the
5999                  * bio now.
6000                  */
6001                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6002                 if (ret)
6003                         goto err;
6004         } else if (!skip_sum) {
6005                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6006                                           file_offset, csums);
6007                 if (ret)
6008                         goto err;
6009         }
6010
6011 map:
6012         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6013 err:
6014         bio_put(bio);
6015         return ret;
6016 }
6017
6018 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6019                                     int skip_sum)
6020 {
6021         struct inode *inode = dip->inode;
6022         struct btrfs_root *root = BTRFS_I(inode)->root;
6023         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6024         struct bio *bio;
6025         struct bio *orig_bio = dip->orig_bio;
6026         struct bio_vec *bvec = orig_bio->bi_io_vec;
6027         u64 start_sector = orig_bio->bi_sector;
6028         u64 file_offset = dip->logical_offset;
6029         u64 submit_len = 0;
6030         u64 map_length;
6031         int nr_pages = 0;
6032         u32 *csums = dip->csums;
6033         int ret = 0;
6034         int async_submit = 0;
6035         int write = rw & REQ_WRITE;
6036
6037         map_length = orig_bio->bi_size;
6038         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6039                               &map_length, NULL, 0);
6040         if (ret) {
6041                 bio_put(bio);
6042                 return -EIO;
6043         }
6044
6045         if (map_length >= orig_bio->bi_size) {
6046                 bio = orig_bio;
6047                 goto submit;
6048         }
6049
6050         async_submit = 1;
6051         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6052         if (!bio)
6053                 return -ENOMEM;
6054         bio->bi_private = dip;
6055         bio->bi_end_io = btrfs_end_dio_bio;
6056         atomic_inc(&dip->pending_bios);
6057
6058         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6059                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6060                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6061                                  bvec->bv_offset) < bvec->bv_len)) {
6062                         /*
6063                          * inc the count before we submit the bio so
6064                          * we know the end IO handler won't happen before
6065                          * we inc the count. Otherwise, the dip might get freed
6066                          * before we're done setting it up
6067                          */
6068                         atomic_inc(&dip->pending_bios);
6069                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6070                                                      file_offset, skip_sum,
6071                                                      csums, async_submit);
6072                         if (ret) {
6073                                 bio_put(bio);
6074                                 atomic_dec(&dip->pending_bios);
6075                                 goto out_err;
6076                         }
6077
6078                         /* Write's use the ordered csums */
6079                         if (!write && !skip_sum)
6080                                 csums = csums + nr_pages;
6081                         start_sector += submit_len >> 9;
6082                         file_offset += submit_len;
6083
6084                         submit_len = 0;
6085                         nr_pages = 0;
6086
6087                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6088                                                   start_sector, GFP_NOFS);
6089                         if (!bio)
6090                                 goto out_err;
6091                         bio->bi_private = dip;
6092                         bio->bi_end_io = btrfs_end_dio_bio;
6093
6094                         map_length = orig_bio->bi_size;
6095                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6096                                               &map_length, NULL, 0);
6097                         if (ret) {
6098                                 bio_put(bio);
6099                                 goto out_err;
6100                         }
6101                 } else {
6102                         submit_len += bvec->bv_len;
6103                         nr_pages ++;
6104                         bvec++;
6105                 }
6106         }
6107
6108 submit:
6109         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6110                                      csums, async_submit);
6111         if (!ret)
6112                 return 0;
6113
6114         bio_put(bio);
6115 out_err:
6116         dip->errors = 1;
6117         /*
6118          * before atomic variable goto zero, we must
6119          * make sure dip->errors is perceived to be set.
6120          */
6121         smp_mb__before_atomic_dec();
6122         if (atomic_dec_and_test(&dip->pending_bios))
6123                 bio_io_error(dip->orig_bio);
6124
6125         /* bio_end_io() will handle error, so we needn't return it */
6126         return 0;
6127 }
6128
6129 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6130                                 loff_t file_offset)
6131 {
6132         struct btrfs_root *root = BTRFS_I(inode)->root;
6133         struct btrfs_dio_private *dip;
6134         struct bio_vec *bvec = bio->bi_io_vec;
6135         int skip_sum;
6136         int write = rw & REQ_WRITE;
6137         int ret = 0;
6138
6139         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6140
6141         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6142         if (!dip) {
6143                 ret = -ENOMEM;
6144                 goto free_ordered;
6145         }
6146         dip->csums = NULL;
6147
6148         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6149         if (!write && !skip_sum) {
6150                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6151                 if (!dip->csums) {
6152                         kfree(dip);
6153                         ret = -ENOMEM;
6154                         goto free_ordered;
6155                 }
6156         }
6157
6158         dip->private = bio->bi_private;
6159         dip->inode = inode;
6160         dip->logical_offset = file_offset;
6161
6162         dip->bytes = 0;
6163         do {
6164                 dip->bytes += bvec->bv_len;
6165                 bvec++;
6166         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6167
6168         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6169         bio->bi_private = dip;
6170         dip->errors = 0;
6171         dip->orig_bio = bio;
6172         atomic_set(&dip->pending_bios, 0);
6173
6174         if (write)
6175                 bio->bi_end_io = btrfs_endio_direct_write;
6176         else
6177                 bio->bi_end_io = btrfs_endio_direct_read;
6178
6179         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6180         if (!ret)
6181                 return;
6182 free_ordered:
6183         /*
6184          * If this is a write, we need to clean up the reserved space and kill
6185          * the ordered extent.
6186          */
6187         if (write) {
6188                 struct btrfs_ordered_extent *ordered;
6189                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6190                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6191                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6192                         btrfs_free_reserved_extent(root, ordered->start,
6193                                                    ordered->disk_len);
6194                 btrfs_put_ordered_extent(ordered);
6195                 btrfs_put_ordered_extent(ordered);
6196         }
6197         bio_endio(bio, ret);
6198 }
6199
6200 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6201                         const struct iovec *iov, loff_t offset,
6202                         unsigned long nr_segs)
6203 {
6204         int seg;
6205         int i;
6206         size_t size;
6207         unsigned long addr;
6208         unsigned blocksize_mask = root->sectorsize - 1;
6209         ssize_t retval = -EINVAL;
6210         loff_t end = offset;
6211
6212         if (offset & blocksize_mask)
6213                 goto out;
6214
6215         /* Check the memory alignment.  Blocks cannot straddle pages */
6216         for (seg = 0; seg < nr_segs; seg++) {
6217                 addr = (unsigned long)iov[seg].iov_base;
6218                 size = iov[seg].iov_len;
6219                 end += size;
6220                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6221                         goto out;
6222
6223                 /* If this is a write we don't need to check anymore */
6224                 if (rw & WRITE)
6225                         continue;
6226
6227                 /*
6228                  * Check to make sure we don't have duplicate iov_base's in this
6229                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6230                  * when reading back.
6231                  */
6232                 for (i = seg + 1; i < nr_segs; i++) {
6233                         if (iov[seg].iov_base == iov[i].iov_base)
6234                                 goto out;
6235                 }
6236         }
6237         retval = 0;
6238 out:
6239         return retval;
6240 }
6241 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6242                         const struct iovec *iov, loff_t offset,
6243                         unsigned long nr_segs)
6244 {
6245         struct file *file = iocb->ki_filp;
6246         struct inode *inode = file->f_mapping->host;
6247         struct btrfs_ordered_extent *ordered;
6248         struct extent_state *cached_state = NULL;
6249         u64 lockstart, lockend;
6250         ssize_t ret;
6251         int writing = rw & WRITE;
6252         int write_bits = 0;
6253         size_t count = iov_length(iov, nr_segs);
6254
6255         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6256                             offset, nr_segs)) {
6257                 return 0;
6258         }
6259
6260         lockstart = offset;
6261         lockend = offset + count - 1;
6262
6263         if (writing) {
6264                 ret = btrfs_delalloc_reserve_space(inode, count);
6265                 if (ret)
6266                         goto out;
6267         }
6268
6269         while (1) {
6270                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6271                                  0, &cached_state, GFP_NOFS);
6272                 /*
6273                  * We're concerned with the entire range that we're going to be
6274                  * doing DIO to, so we need to make sure theres no ordered
6275                  * extents in this range.
6276                  */
6277                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6278                                                      lockend - lockstart + 1);
6279                 if (!ordered)
6280                         break;
6281                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6282                                      &cached_state, GFP_NOFS);
6283                 btrfs_start_ordered_extent(inode, ordered, 1);
6284                 btrfs_put_ordered_extent(ordered);
6285                 cond_resched();
6286         }
6287
6288         /*
6289          * we don't use btrfs_set_extent_delalloc because we don't want
6290          * the dirty or uptodate bits
6291          */
6292         if (writing) {
6293                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6294                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6295                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6296                                      GFP_NOFS);
6297                 if (ret) {
6298                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6299                                          lockend, EXTENT_LOCKED | write_bits,
6300                                          1, 0, &cached_state, GFP_NOFS);
6301                         goto out;
6302                 }
6303         }
6304
6305         free_extent_state(cached_state);
6306         cached_state = NULL;
6307
6308         ret = __blockdev_direct_IO(rw, iocb, inode,
6309                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6310                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6311                    btrfs_submit_direct, 0);
6312
6313         if (ret < 0 && ret != -EIOCBQUEUED) {
6314                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6315                               offset + iov_length(iov, nr_segs) - 1,
6316                               EXTENT_LOCKED | write_bits, 1, 0,
6317                               &cached_state, GFP_NOFS);
6318         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6319                 /*
6320                  * We're falling back to buffered, unlock the section we didn't
6321                  * do IO on.
6322                  */
6323                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6324                               offset + iov_length(iov, nr_segs) - 1,
6325                               EXTENT_LOCKED | write_bits, 1, 0,
6326                               &cached_state, GFP_NOFS);
6327         }
6328 out:
6329         free_extent_state(cached_state);
6330         return ret;
6331 }
6332
6333 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6334                 __u64 start, __u64 len)
6335 {
6336         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6337 }
6338
6339 int btrfs_readpage(struct file *file, struct page *page)
6340 {
6341         struct extent_io_tree *tree;
6342         tree = &BTRFS_I(page->mapping->host)->io_tree;
6343         return extent_read_full_page(tree, page, btrfs_get_extent);
6344 }
6345
6346 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6347 {
6348         struct extent_io_tree *tree;
6349
6350
6351         if (current->flags & PF_MEMALLOC) {
6352                 redirty_page_for_writepage(wbc, page);
6353                 unlock_page(page);
6354                 return 0;
6355         }
6356         tree = &BTRFS_I(page->mapping->host)->io_tree;
6357         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6358 }
6359
6360 int btrfs_writepages(struct address_space *mapping,
6361                      struct writeback_control *wbc)
6362 {
6363         struct extent_io_tree *tree;
6364
6365         tree = &BTRFS_I(mapping->host)->io_tree;
6366         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6367 }
6368
6369 static int
6370 btrfs_readpages(struct file *file, struct address_space *mapping,
6371                 struct list_head *pages, unsigned nr_pages)
6372 {
6373         struct extent_io_tree *tree;
6374         tree = &BTRFS_I(mapping->host)->io_tree;
6375         return extent_readpages(tree, mapping, pages, nr_pages,
6376                                 btrfs_get_extent);
6377 }
6378 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6379 {
6380         struct extent_io_tree *tree;
6381         struct extent_map_tree *map;
6382         int ret;
6383
6384         tree = &BTRFS_I(page->mapping->host)->io_tree;
6385         map = &BTRFS_I(page->mapping->host)->extent_tree;
6386         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6387         if (ret == 1) {
6388                 ClearPagePrivate(page);
6389                 set_page_private(page, 0);
6390                 page_cache_release(page);
6391         }
6392         return ret;
6393 }
6394
6395 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6396 {
6397         if (PageWriteback(page) || PageDirty(page))
6398                 return 0;
6399         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6400 }
6401
6402 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6403 {
6404         struct extent_io_tree *tree;
6405         struct btrfs_ordered_extent *ordered;
6406         struct extent_state *cached_state = NULL;
6407         u64 page_start = page_offset(page);
6408         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6409
6410
6411         /*
6412          * we have the page locked, so new writeback can't start,
6413          * and the dirty bit won't be cleared while we are here.
6414          *
6415          * Wait for IO on this page so that we can safely clear
6416          * the PagePrivate2 bit and do ordered accounting
6417          */
6418         wait_on_page_writeback(page);
6419
6420         tree = &BTRFS_I(page->mapping->host)->io_tree;
6421         if (offset) {
6422                 btrfs_releasepage(page, GFP_NOFS);
6423                 return;
6424         }
6425         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6426                          GFP_NOFS);
6427         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6428                                            page_offset(page));
6429         if (ordered) {
6430                 /*
6431                  * IO on this page will never be started, so we need
6432                  * to account for any ordered extents now
6433                  */
6434                 clear_extent_bit(tree, page_start, page_end,
6435                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6436                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6437                                  &cached_state, GFP_NOFS);
6438                 /*
6439                  * whoever cleared the private bit is responsible
6440                  * for the finish_ordered_io
6441                  */
6442                 if (TestClearPagePrivate2(page)) {
6443                         btrfs_finish_ordered_io(page->mapping->host,
6444                                                 page_start, page_end);
6445                 }
6446                 btrfs_put_ordered_extent(ordered);
6447                 cached_state = NULL;
6448                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6449                                  GFP_NOFS);
6450         }
6451         clear_extent_bit(tree, page_start, page_end,
6452                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6453                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6454         __btrfs_releasepage(page, GFP_NOFS);
6455
6456         ClearPageChecked(page);
6457         if (PagePrivate(page)) {
6458                 ClearPagePrivate(page);
6459                 set_page_private(page, 0);
6460                 page_cache_release(page);
6461         }
6462 }
6463
6464 /*
6465  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6466  * called from a page fault handler when a page is first dirtied. Hence we must
6467  * be careful to check for EOF conditions here. We set the page up correctly
6468  * for a written page which means we get ENOSPC checking when writing into
6469  * holes and correct delalloc and unwritten extent mapping on filesystems that
6470  * support these features.
6471  *
6472  * We are not allowed to take the i_mutex here so we have to play games to
6473  * protect against truncate races as the page could now be beyond EOF.  Because
6474  * vmtruncate() writes the inode size before removing pages, once we have the
6475  * page lock we can determine safely if the page is beyond EOF. If it is not
6476  * beyond EOF, then the page is guaranteed safe against truncation until we
6477  * unlock the page.
6478  */
6479 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6480 {
6481         struct page *page = vmf->page;
6482         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6483         struct btrfs_root *root = BTRFS_I(inode)->root;
6484         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6485         struct btrfs_ordered_extent *ordered;
6486         struct extent_state *cached_state = NULL;
6487         char *kaddr;
6488         unsigned long zero_start;
6489         loff_t size;
6490         int ret;
6491         u64 page_start;
6492         u64 page_end;
6493
6494         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6495         if (ret) {
6496                 if (ret == -ENOMEM)
6497                         ret = VM_FAULT_OOM;
6498                 else /* -ENOSPC, -EIO, etc */
6499                         ret = VM_FAULT_SIGBUS;
6500                 goto out;
6501         }
6502
6503         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6504 again:
6505         lock_page(page);
6506         size = i_size_read(inode);
6507         page_start = page_offset(page);
6508         page_end = page_start + PAGE_CACHE_SIZE - 1;
6509
6510         if ((page->mapping != inode->i_mapping) ||
6511             (page_start >= size)) {
6512                 /* page got truncated out from underneath us */
6513                 goto out_unlock;
6514         }
6515         wait_on_page_writeback(page);
6516
6517         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6518                          GFP_NOFS);
6519         set_page_extent_mapped(page);
6520
6521         /*
6522          * we can't set the delalloc bits if there are pending ordered
6523          * extents.  Drop our locks and wait for them to finish
6524          */
6525         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6526         if (ordered) {
6527                 unlock_extent_cached(io_tree, page_start, page_end,
6528                                      &cached_state, GFP_NOFS);
6529                 unlock_page(page);
6530                 btrfs_start_ordered_extent(inode, ordered, 1);
6531                 btrfs_put_ordered_extent(ordered);
6532                 goto again;
6533         }
6534
6535         /*
6536          * XXX - page_mkwrite gets called every time the page is dirtied, even
6537          * if it was already dirty, so for space accounting reasons we need to
6538          * clear any delalloc bits for the range we are fixing to save.  There
6539          * is probably a better way to do this, but for now keep consistent with
6540          * prepare_pages in the normal write path.
6541          */
6542         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6543                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6544                           0, 0, &cached_state, GFP_NOFS);
6545
6546         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6547                                         &cached_state);
6548         if (ret) {
6549                 unlock_extent_cached(io_tree, page_start, page_end,
6550                                      &cached_state, GFP_NOFS);
6551                 ret = VM_FAULT_SIGBUS;
6552                 goto out_unlock;
6553         }
6554         ret = 0;
6555
6556         /* page is wholly or partially inside EOF */
6557         if (page_start + PAGE_CACHE_SIZE > size)
6558                 zero_start = size & ~PAGE_CACHE_MASK;
6559         else
6560                 zero_start = PAGE_CACHE_SIZE;
6561
6562         if (zero_start != PAGE_CACHE_SIZE) {
6563                 kaddr = kmap(page);
6564                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6565                 flush_dcache_page(page);
6566                 kunmap(page);
6567         }
6568         ClearPageChecked(page);
6569         set_page_dirty(page);
6570         SetPageUptodate(page);
6571
6572         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6573         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6574
6575         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6576
6577 out_unlock:
6578         if (!ret)
6579                 return VM_FAULT_LOCKED;
6580         unlock_page(page);
6581         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6582 out:
6583         return ret;
6584 }
6585
6586 static int btrfs_truncate(struct inode *inode)
6587 {
6588         struct btrfs_root *root = BTRFS_I(inode)->root;
6589         int ret;
6590         int err = 0;
6591         struct btrfs_trans_handle *trans;
6592         unsigned long nr;
6593         u64 mask = root->sectorsize - 1;
6594
6595         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6596         if (ret)
6597                 return ret;
6598
6599         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6600         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6601
6602         trans = btrfs_start_transaction(root, 5);
6603         if (IS_ERR(trans))
6604                 return PTR_ERR(trans);
6605
6606         btrfs_set_trans_block_group(trans, inode);
6607
6608         ret = btrfs_orphan_add(trans, inode);
6609         if (ret) {
6610                 btrfs_end_transaction(trans, root);
6611                 return ret;
6612         }
6613
6614         nr = trans->blocks_used;
6615         btrfs_end_transaction(trans, root);
6616         btrfs_btree_balance_dirty(root, nr);
6617
6618         /* Now start a transaction for the truncate */
6619         trans = btrfs_start_transaction(root, 0);
6620         if (IS_ERR(trans))
6621                 return PTR_ERR(trans);
6622         btrfs_set_trans_block_group(trans, inode);
6623         trans->block_rsv = root->orphan_block_rsv;
6624
6625         /*
6626          * setattr is responsible for setting the ordered_data_close flag,
6627          * but that is only tested during the last file release.  That
6628          * could happen well after the next commit, leaving a great big
6629          * window where new writes may get lost if someone chooses to write
6630          * to this file after truncating to zero
6631          *
6632          * The inode doesn't have any dirty data here, and so if we commit
6633          * this is a noop.  If someone immediately starts writing to the inode
6634          * it is very likely we'll catch some of their writes in this
6635          * transaction, and the commit will find this file on the ordered
6636          * data list with good things to send down.
6637          *
6638          * This is a best effort solution, there is still a window where
6639          * using truncate to replace the contents of the file will
6640          * end up with a zero length file after a crash.
6641          */
6642         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6643                 btrfs_add_ordered_operation(trans, root, inode);
6644
6645         while (1) {
6646                 if (!trans) {
6647                         trans = btrfs_start_transaction(root, 0);
6648                         if (IS_ERR(trans))
6649                                 return PTR_ERR(trans);
6650                         btrfs_set_trans_block_group(trans, inode);
6651                         trans->block_rsv = root->orphan_block_rsv;
6652                 }
6653
6654                 ret = btrfs_block_rsv_check(trans, root,
6655                                             root->orphan_block_rsv, 0, 5);
6656                 if (ret == -EAGAIN) {
6657                         ret = btrfs_commit_transaction(trans, root);
6658                         if (ret)
6659                                 return ret;
6660                         trans = NULL;
6661                         continue;
6662                 } else if (ret) {
6663                         err = ret;
6664                         break;
6665                 }
6666
6667                 ret = btrfs_truncate_inode_items(trans, root, inode,
6668                                                  inode->i_size,
6669                                                  BTRFS_EXTENT_DATA_KEY);
6670                 if (ret != -EAGAIN) {
6671                         err = ret;
6672                         break;
6673                 }
6674
6675                 ret = btrfs_update_inode(trans, root, inode);
6676                 if (ret) {
6677                         err = ret;
6678                         break;
6679                 }
6680
6681                 nr = trans->blocks_used;
6682                 btrfs_end_transaction(trans, root);
6683                 trans = NULL;
6684                 btrfs_btree_balance_dirty(root, nr);
6685         }
6686
6687         if (ret == 0 && inode->i_nlink > 0) {
6688                 ret = btrfs_orphan_del(trans, inode);
6689                 if (ret)
6690                         err = ret;
6691         } else if (ret && inode->i_nlink > 0) {
6692                 /*
6693                  * Failed to do the truncate, remove us from the in memory
6694                  * orphan list.
6695                  */
6696                 ret = btrfs_orphan_del(NULL, inode);
6697         }
6698
6699         ret = btrfs_update_inode(trans, root, inode);
6700         if (ret && !err)
6701                 err = ret;
6702
6703         nr = trans->blocks_used;
6704         ret = btrfs_end_transaction_throttle(trans, root);
6705         if (ret && !err)
6706                 err = ret;
6707         btrfs_btree_balance_dirty(root, nr);
6708
6709         return err;
6710 }
6711
6712 /*
6713  * create a new subvolume directory/inode (helper for the ioctl).
6714  */
6715 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6716                              struct btrfs_root *new_root,
6717                              u64 new_dirid, u64 alloc_hint)
6718 {
6719         struct inode *inode;
6720         int err;
6721         u64 index = 0;
6722
6723         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6724                                 new_dirid, alloc_hint, S_IFDIR | 0700, &index);
6725         if (IS_ERR(inode))
6726                 return PTR_ERR(inode);
6727         inode->i_op = &btrfs_dir_inode_operations;
6728         inode->i_fop = &btrfs_dir_file_operations;
6729
6730         inode->i_nlink = 1;
6731         btrfs_i_size_write(inode, 0);
6732
6733         err = btrfs_update_inode(trans, new_root, inode);
6734         BUG_ON(err);
6735
6736         iput(inode);
6737         return 0;
6738 }
6739
6740 /* helper function for file defrag and space balancing.  This
6741  * forces readahead on a given range of bytes in an inode
6742  */
6743 unsigned long btrfs_force_ra(struct address_space *mapping,
6744                               struct file_ra_state *ra, struct file *file,
6745                               pgoff_t offset, pgoff_t last_index)
6746 {
6747         pgoff_t req_size = last_index - offset + 1;
6748
6749         page_cache_sync_readahead(mapping, ra, file, offset, req_size);
6750         return offset + req_size;
6751 }
6752
6753 struct inode *btrfs_alloc_inode(struct super_block *sb)
6754 {
6755         struct btrfs_inode *ei;
6756         struct inode *inode;
6757
6758         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6759         if (!ei)
6760                 return NULL;
6761
6762         ei->root = NULL;
6763         ei->space_info = NULL;
6764         ei->generation = 0;
6765         ei->sequence = 0;
6766         ei->last_trans = 0;
6767         ei->last_sub_trans = 0;
6768         ei->logged_trans = 0;
6769         ei->delalloc_bytes = 0;
6770         ei->reserved_bytes = 0;
6771         ei->disk_i_size = 0;
6772         ei->flags = 0;
6773         ei->index_cnt = (u64)-1;
6774         ei->last_unlink_trans = 0;
6775
6776         atomic_set(&ei->outstanding_extents, 0);
6777         atomic_set(&ei->reserved_extents, 0);
6778
6779         ei->ordered_data_close = 0;
6780         ei->orphan_meta_reserved = 0;
6781         ei->dummy_inode = 0;
6782         ei->force_compress = BTRFS_COMPRESS_NONE;
6783
6784         inode = &ei->vfs_inode;
6785         extent_map_tree_init(&ei->extent_tree, GFP_NOFS);
6786         extent_io_tree_init(&ei->io_tree, &inode->i_data, GFP_NOFS);
6787         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data, GFP_NOFS);
6788         mutex_init(&ei->log_mutex);
6789         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6790         INIT_LIST_HEAD(&ei->i_orphan);
6791         INIT_LIST_HEAD(&ei->delalloc_inodes);
6792         INIT_LIST_HEAD(&ei->ordered_operations);
6793         RB_CLEAR_NODE(&ei->rb_node);
6794
6795         return inode;
6796 }
6797
6798 static void btrfs_i_callback(struct rcu_head *head)
6799 {
6800         struct inode *inode = container_of(head, struct inode, i_rcu);
6801         INIT_LIST_HEAD(&inode->i_dentry);
6802         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6803 }
6804
6805 void btrfs_destroy_inode(struct inode *inode)
6806 {
6807         struct btrfs_ordered_extent *ordered;
6808         struct btrfs_root *root = BTRFS_I(inode)->root;
6809
6810         WARN_ON(!list_empty(&inode->i_dentry));
6811         WARN_ON(inode->i_data.nrpages);
6812         WARN_ON(atomic_read(&BTRFS_I(inode)->outstanding_extents));
6813         WARN_ON(atomic_read(&BTRFS_I(inode)->reserved_extents));
6814
6815         /*
6816          * This can happen where we create an inode, but somebody else also
6817          * created the same inode and we need to destroy the one we already
6818          * created.
6819          */
6820         if (!root)
6821                 goto free;
6822
6823         /*
6824          * Make sure we're properly removed from the ordered operation
6825          * lists.
6826          */
6827         smp_mb();
6828         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6829                 spin_lock(&root->fs_info->ordered_extent_lock);
6830                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6831                 spin_unlock(&root->fs_info->ordered_extent_lock);
6832         }
6833
6834         if (root == root->fs_info->tree_root) {
6835                 struct btrfs_block_group_cache *block_group;
6836
6837                 block_group = btrfs_lookup_block_group(root->fs_info,
6838                                                 BTRFS_I(inode)->block_group);
6839                 if (block_group && block_group->inode == inode) {
6840                         spin_lock(&block_group->lock);
6841                         block_group->inode = NULL;
6842                         spin_unlock(&block_group->lock);
6843                         btrfs_put_block_group(block_group);
6844                 } else if (block_group) {
6845                         btrfs_put_block_group(block_group);
6846                 }
6847         }
6848
6849         spin_lock(&root->orphan_lock);
6850         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6851                 printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
6852                        inode->i_ino);
6853                 list_del_init(&BTRFS_I(inode)->i_orphan);
6854         }
6855         spin_unlock(&root->orphan_lock);
6856
6857         while (1) {
6858                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6859                 if (!ordered)
6860                         break;
6861                 else {
6862                         printk(KERN_ERR "btrfs found ordered "
6863                                "extent %llu %llu on inode cleanup\n",
6864                                (unsigned long long)ordered->file_offset,
6865                                (unsigned long long)ordered->len);
6866                         btrfs_remove_ordered_extent(inode, ordered);
6867                         btrfs_put_ordered_extent(ordered);
6868                         btrfs_put_ordered_extent(ordered);
6869                 }
6870         }
6871         inode_tree_del(inode);
6872         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6873 free:
6874         call_rcu(&inode->i_rcu, btrfs_i_callback);
6875 }
6876
6877 int btrfs_drop_inode(struct inode *inode)
6878 {
6879         struct btrfs_root *root = BTRFS_I(inode)->root;
6880
6881         if (btrfs_root_refs(&root->root_item) == 0 &&
6882             root != root->fs_info->tree_root)
6883                 return 1;
6884         else
6885                 return generic_drop_inode(inode);
6886 }
6887
6888 static void init_once(void *foo)
6889 {
6890         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6891
6892         inode_init_once(&ei->vfs_inode);
6893 }
6894
6895 void btrfs_destroy_cachep(void)
6896 {
6897         if (btrfs_inode_cachep)
6898                 kmem_cache_destroy(btrfs_inode_cachep);
6899         if (btrfs_trans_handle_cachep)
6900                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6901         if (btrfs_transaction_cachep)
6902                 kmem_cache_destroy(btrfs_transaction_cachep);
6903         if (btrfs_path_cachep)
6904                 kmem_cache_destroy(btrfs_path_cachep);
6905         if (btrfs_free_space_cachep)
6906                 kmem_cache_destroy(btrfs_free_space_cachep);
6907 }
6908
6909 int btrfs_init_cachep(void)
6910 {
6911         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6912                         sizeof(struct btrfs_inode), 0,
6913                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6914         if (!btrfs_inode_cachep)
6915                 goto fail;
6916
6917         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6918                         sizeof(struct btrfs_trans_handle), 0,
6919                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6920         if (!btrfs_trans_handle_cachep)
6921                 goto fail;
6922
6923         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6924                         sizeof(struct btrfs_transaction), 0,
6925                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6926         if (!btrfs_transaction_cachep)
6927                 goto fail;
6928
6929         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6930                         sizeof(struct btrfs_path), 0,
6931                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6932         if (!btrfs_path_cachep)
6933                 goto fail;
6934
6935         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6936                         sizeof(struct btrfs_free_space), 0,
6937                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6938         if (!btrfs_free_space_cachep)
6939                 goto fail;
6940
6941         return 0;
6942 fail:
6943         btrfs_destroy_cachep();
6944         return -ENOMEM;
6945 }
6946
6947 static int btrfs_getattr(struct vfsmount *mnt,
6948                          struct dentry *dentry, struct kstat *stat)
6949 {
6950         struct inode *inode = dentry->d_inode;
6951         generic_fillattr(inode, stat);
6952         stat->dev = BTRFS_I(inode)->root->anon_super.s_dev;
6953         stat->blksize = PAGE_CACHE_SIZE;
6954         stat->blocks = (inode_get_bytes(inode) +
6955                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6956         return 0;
6957 }
6958
6959 /*
6960  * If a file is moved, it will inherit the cow and compression flags of the new
6961  * directory.
6962  */
6963 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6964 {
6965         struct btrfs_inode *b_dir = BTRFS_I(dir);
6966         struct btrfs_inode *b_inode = BTRFS_I(inode);
6967
6968         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6969                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6970         else
6971                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6972
6973         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6974                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6975         else
6976                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6977 }
6978
6979 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6980                            struct inode *new_dir, struct dentry *new_dentry)
6981 {
6982         struct btrfs_trans_handle *trans;
6983         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6984         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6985         struct inode *new_inode = new_dentry->d_inode;
6986         struct inode *old_inode = old_dentry->d_inode;
6987         struct timespec ctime = CURRENT_TIME;
6988         u64 index = 0;
6989         u64 root_objectid;
6990         int ret;
6991
6992         if (new_dir->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6993                 return -EPERM;
6994
6995         /* we only allow rename subvolume link between subvolumes */
6996         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6997                 return -EXDEV;
6998
6999         if (old_inode->i_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7000             (new_inode && new_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID))
7001                 return -ENOTEMPTY;
7002
7003         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7004             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7005                 return -ENOTEMPTY;
7006         /*
7007          * we're using rename to replace one file with another.
7008          * and the replacement file is large.  Start IO on it now so
7009          * we don't add too much work to the end of the transaction
7010          */
7011         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7012             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7013                 filemap_flush(old_inode->i_mapping);
7014
7015         /* close the racy window with snapshot create/destroy ioctl */
7016         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7017                 down_read(&root->fs_info->subvol_sem);
7018         /*
7019          * We want to reserve the absolute worst case amount of items.  So if
7020          * both inodes are subvols and we need to unlink them then that would
7021          * require 4 item modifications, but if they are both normal inodes it
7022          * would require 5 item modifications, so we'll assume their normal
7023          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7024          * should cover the worst case number of items we'll modify.
7025          */
7026         trans = btrfs_start_transaction(root, 20);
7027         if (IS_ERR(trans)) {
7028                 ret = PTR_ERR(trans);
7029                 goto out_notrans;
7030         }
7031
7032         btrfs_set_trans_block_group(trans, new_dir);
7033
7034         if (dest != root)
7035                 btrfs_record_root_in_trans(trans, dest);
7036
7037         ret = btrfs_set_inode_index(new_dir, &index);
7038         if (ret)
7039                 goto out_fail;
7040
7041         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7042                 /* force full log commit if subvolume involved. */
7043                 root->fs_info->last_trans_log_full_commit = trans->transid;
7044         } else {
7045                 ret = btrfs_insert_inode_ref(trans, dest,
7046                                              new_dentry->d_name.name,
7047                                              new_dentry->d_name.len,
7048                                              old_inode->i_ino,
7049                                              new_dir->i_ino, index);
7050                 if (ret)
7051                         goto out_fail;
7052                 /*
7053                  * this is an ugly little race, but the rename is required
7054                  * to make sure that if we crash, the inode is either at the
7055                  * old name or the new one.  pinning the log transaction lets
7056                  * us make sure we don't allow a log commit to come in after
7057                  * we unlink the name but before we add the new name back in.
7058                  */
7059                 btrfs_pin_log_trans(root);
7060         }
7061         /*
7062          * make sure the inode gets flushed if it is replacing
7063          * something.
7064          */
7065         if (new_inode && new_inode->i_size &&
7066             old_inode && S_ISREG(old_inode->i_mode)) {
7067                 btrfs_add_ordered_operation(trans, root, old_inode);
7068         }
7069
7070         old_dir->i_ctime = old_dir->i_mtime = ctime;
7071         new_dir->i_ctime = new_dir->i_mtime = ctime;
7072         old_inode->i_ctime = ctime;
7073
7074         if (old_dentry->d_parent != new_dentry->d_parent)
7075                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7076
7077         if (unlikely(old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7078                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7079                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7080                                         old_dentry->d_name.name,
7081                                         old_dentry->d_name.len);
7082         } else {
7083                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7084                                         old_dentry->d_inode,
7085                                         old_dentry->d_name.name,
7086                                         old_dentry->d_name.len);
7087                 if (!ret)
7088                         ret = btrfs_update_inode(trans, root, old_inode);
7089         }
7090         BUG_ON(ret);
7091
7092         if (new_inode) {
7093                 new_inode->i_ctime = CURRENT_TIME;
7094                 if (unlikely(new_inode->i_ino ==
7095                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7096                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7097                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7098                                                 root_objectid,
7099                                                 new_dentry->d_name.name,
7100                                                 new_dentry->d_name.len);
7101                         BUG_ON(new_inode->i_nlink == 0);
7102                 } else {
7103                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7104                                                  new_dentry->d_inode,
7105                                                  new_dentry->d_name.name,
7106                                                  new_dentry->d_name.len);
7107                 }
7108                 BUG_ON(ret);
7109                 if (new_inode->i_nlink == 0) {
7110                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7111                         BUG_ON(ret);
7112                 }
7113         }
7114
7115         fixup_inode_flags(new_dir, old_inode);
7116
7117         ret = btrfs_add_link(trans, new_dir, old_inode,
7118                              new_dentry->d_name.name,
7119                              new_dentry->d_name.len, 0, index);
7120         BUG_ON(ret);
7121
7122         if (old_inode->i_ino != BTRFS_FIRST_FREE_OBJECTID) {
7123                 struct dentry *parent = dget_parent(new_dentry);
7124                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7125                 dput(parent);
7126                 btrfs_end_log_trans(root);
7127         }
7128 out_fail:
7129         btrfs_end_transaction_throttle(trans, root);
7130 out_notrans:
7131         if (old_inode->i_ino == BTRFS_FIRST_FREE_OBJECTID)
7132                 up_read(&root->fs_info->subvol_sem);
7133
7134         return ret;
7135 }
7136
7137 /*
7138  * some fairly slow code that needs optimization. This walks the list
7139  * of all the inodes with pending delalloc and forces them to disk.
7140  */
7141 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7142 {
7143         struct list_head *head = &root->fs_info->delalloc_inodes;
7144         struct btrfs_inode *binode;
7145         struct inode *inode;
7146
7147         if (root->fs_info->sb->s_flags & MS_RDONLY)
7148                 return -EROFS;
7149
7150         spin_lock(&root->fs_info->delalloc_lock);
7151         while (!list_empty(head)) {
7152                 binode = list_entry(head->next, struct btrfs_inode,
7153                                     delalloc_inodes);
7154                 inode = igrab(&binode->vfs_inode);
7155                 if (!inode)
7156                         list_del_init(&binode->delalloc_inodes);
7157                 spin_unlock(&root->fs_info->delalloc_lock);
7158                 if (inode) {
7159                         filemap_flush(inode->i_mapping);
7160                         if (delay_iput)
7161                                 btrfs_add_delayed_iput(inode);
7162                         else
7163                                 iput(inode);
7164                 }
7165                 cond_resched();
7166                 spin_lock(&root->fs_info->delalloc_lock);
7167         }
7168         spin_unlock(&root->fs_info->delalloc_lock);
7169
7170         /* the filemap_flush will queue IO into the worker threads, but
7171          * we have to make sure the IO is actually started and that
7172          * ordered extents get created before we return
7173          */
7174         atomic_inc(&root->fs_info->async_submit_draining);
7175         while (atomic_read(&root->fs_info->nr_async_submits) ||
7176               atomic_read(&root->fs_info->async_delalloc_pages)) {
7177                 wait_event(root->fs_info->async_submit_wait,
7178                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7179                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7180         }
7181         atomic_dec(&root->fs_info->async_submit_draining);
7182         return 0;
7183 }
7184
7185 int btrfs_start_one_delalloc_inode(struct btrfs_root *root, int delay_iput,
7186                                    int sync)
7187 {
7188         struct btrfs_inode *binode;
7189         struct inode *inode = NULL;
7190
7191         spin_lock(&root->fs_info->delalloc_lock);
7192         while (!list_empty(&root->fs_info->delalloc_inodes)) {
7193                 binode = list_entry(root->fs_info->delalloc_inodes.next,
7194                                     struct btrfs_inode, delalloc_inodes);
7195                 inode = igrab(&binode->vfs_inode);
7196                 if (inode) {
7197                         list_move_tail(&binode->delalloc_inodes,
7198                                        &root->fs_info->delalloc_inodes);
7199                         break;
7200                 }
7201
7202                 list_del_init(&binode->delalloc_inodes);
7203                 cond_resched_lock(&root->fs_info->delalloc_lock);
7204         }
7205         spin_unlock(&root->fs_info->delalloc_lock);
7206
7207         if (inode) {
7208                 if (sync) {
7209                         filemap_write_and_wait(inode->i_mapping);
7210                         /*
7211                          * We have to do this because compression doesn't
7212                          * actually set PG_writeback until it submits the pages
7213                          * for IO, which happens in an async thread, so we could
7214                          * race and not actually wait for any writeback pages
7215                          * because they've not been submitted yet.  Technically
7216                          * this could still be the case for the ordered stuff
7217                          * since the async thread may not have started to do its
7218                          * work yet.  If this becomes the case then we need to
7219                          * figure out a way to make sure that in writepage we
7220                          * wait for any async pages to be submitted before
7221                          * returning so that fdatawait does what its supposed to
7222                          * do.
7223                          */
7224                         btrfs_wait_ordered_range(inode, 0, (u64)-1);
7225                 } else {
7226                         filemap_flush(inode->i_mapping);
7227                 }
7228                 if (delay_iput)
7229                         btrfs_add_delayed_iput(inode);
7230                 else
7231                         iput(inode);
7232                 return 1;
7233         }
7234         return 0;
7235 }
7236
7237 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7238                          const char *symname)
7239 {
7240         struct btrfs_trans_handle *trans;
7241         struct btrfs_root *root = BTRFS_I(dir)->root;
7242         struct btrfs_path *path;
7243         struct btrfs_key key;
7244         struct inode *inode = NULL;
7245         int err;
7246         int drop_inode = 0;
7247         u64 objectid;
7248         u64 index = 0 ;
7249         int name_len;
7250         int datasize;
7251         unsigned long ptr;
7252         struct btrfs_file_extent_item *ei;
7253         struct extent_buffer *leaf;
7254         unsigned long nr = 0;
7255
7256         name_len = strlen(symname) + 1;
7257         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7258                 return -ENAMETOOLONG;
7259
7260         err = btrfs_find_free_objectid(NULL, root, dir->i_ino, &objectid);
7261         if (err)
7262                 return err;
7263         /*
7264          * 2 items for inode item and ref
7265          * 2 items for dir items
7266          * 1 item for xattr if selinux is on
7267          */
7268         trans = btrfs_start_transaction(root, 5);
7269         if (IS_ERR(trans))
7270                 return PTR_ERR(trans);
7271
7272         btrfs_set_trans_block_group(trans, dir);
7273
7274         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7275                                 dentry->d_name.len, dir->i_ino, objectid,
7276                                 BTRFS_I(dir)->block_group, S_IFLNK|S_IRWXUGO,
7277                                 &index);
7278         err = PTR_ERR(inode);
7279         if (IS_ERR(inode))
7280                 goto out_unlock;
7281
7282         err = btrfs_init_inode_security(trans, inode, dir);
7283         if (err) {
7284                 drop_inode = 1;
7285                 goto out_unlock;
7286         }
7287
7288         btrfs_set_trans_block_group(trans, inode);
7289         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7290         if (err)
7291                 drop_inode = 1;
7292         else {
7293                 inode->i_mapping->a_ops = &btrfs_aops;
7294                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7295                 inode->i_fop = &btrfs_file_operations;
7296                 inode->i_op = &btrfs_file_inode_operations;
7297                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7298         }
7299         btrfs_update_inode_block_group(trans, inode);
7300         btrfs_update_inode_block_group(trans, dir);
7301         if (drop_inode)
7302                 goto out_unlock;
7303
7304         path = btrfs_alloc_path();
7305         BUG_ON(!path);
7306         key.objectid = inode->i_ino;
7307         key.offset = 0;
7308         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7309         datasize = btrfs_file_extent_calc_inline_size(name_len);
7310         err = btrfs_insert_empty_item(trans, root, path, &key,
7311                                       datasize);
7312         if (err) {
7313                 drop_inode = 1;
7314                 goto out_unlock;
7315         }
7316         leaf = path->nodes[0];
7317         ei = btrfs_item_ptr(leaf, path->slots[0],
7318                             struct btrfs_file_extent_item);
7319         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7320         btrfs_set_file_extent_type(leaf, ei,
7321                                    BTRFS_FILE_EXTENT_INLINE);
7322         btrfs_set_file_extent_encryption(leaf, ei, 0);
7323         btrfs_set_file_extent_compression(leaf, ei, 0);
7324         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7325         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7326
7327         ptr = btrfs_file_extent_inline_start(ei);
7328         write_extent_buffer(leaf, symname, ptr, name_len);
7329         btrfs_mark_buffer_dirty(leaf);
7330         btrfs_free_path(path);
7331
7332         inode->i_op = &btrfs_symlink_inode_operations;
7333         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7334         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7335         inode_set_bytes(inode, name_len);
7336         btrfs_i_size_write(inode, name_len - 1);
7337         err = btrfs_update_inode(trans, root, inode);
7338         if (err)
7339                 drop_inode = 1;
7340
7341 out_unlock:
7342         nr = trans->blocks_used;
7343         btrfs_end_transaction_throttle(trans, root);
7344         if (drop_inode) {
7345                 inode_dec_link_count(inode);
7346                 iput(inode);
7347         }
7348         btrfs_btree_balance_dirty(root, nr);
7349         return err;
7350 }
7351
7352 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7353                                        u64 start, u64 num_bytes, u64 min_size,
7354                                        loff_t actual_len, u64 *alloc_hint,
7355                                        struct btrfs_trans_handle *trans)
7356 {
7357         struct btrfs_root *root = BTRFS_I(inode)->root;
7358         struct btrfs_key ins;
7359         u64 cur_offset = start;
7360         u64 i_size;
7361         int ret = 0;
7362         bool own_trans = true;
7363
7364         if (trans)
7365                 own_trans = false;
7366         while (num_bytes > 0) {
7367                 if (own_trans) {
7368                         trans = btrfs_start_transaction(root, 3);
7369                         if (IS_ERR(trans)) {
7370                                 ret = PTR_ERR(trans);
7371                                 break;
7372                         }
7373                 }
7374
7375                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7376                                            0, *alloc_hint, (u64)-1, &ins, 1);
7377                 if (ret) {
7378                         if (own_trans)
7379                                 btrfs_end_transaction(trans, root);
7380                         break;
7381                 }
7382
7383                 ret = insert_reserved_file_extent(trans, inode,
7384                                                   cur_offset, ins.objectid,
7385                                                   ins.offset, ins.offset,
7386                                                   ins.offset, 0, 0, 0,
7387                                                   BTRFS_FILE_EXTENT_PREALLOC);
7388                 BUG_ON(ret);
7389                 btrfs_drop_extent_cache(inode, cur_offset,
7390                                         cur_offset + ins.offset -1, 0);
7391
7392                 num_bytes -= ins.offset;
7393                 cur_offset += ins.offset;
7394                 *alloc_hint = ins.objectid + ins.offset;
7395
7396                 inode->i_ctime = CURRENT_TIME;
7397                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7398                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7399                     (actual_len > inode->i_size) &&
7400                     (cur_offset > inode->i_size)) {
7401                         if (cur_offset > actual_len)
7402                                 i_size = actual_len;
7403                         else
7404                                 i_size = cur_offset;
7405                         i_size_write(inode, i_size);
7406                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7407                 }
7408
7409                 ret = btrfs_update_inode(trans, root, inode);
7410                 BUG_ON(ret);
7411
7412                 if (own_trans)
7413                         btrfs_end_transaction(trans, root);
7414         }
7415         return ret;
7416 }
7417
7418 int btrfs_prealloc_file_range(struct inode *inode, 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,
7424                                            NULL);
7425 }
7426
7427 int btrfs_prealloc_file_range_trans(struct inode *inode,
7428                                     struct btrfs_trans_handle *trans, int mode,
7429                                     u64 start, u64 num_bytes, u64 min_size,
7430                                     loff_t actual_len, u64 *alloc_hint)
7431 {
7432         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7433                                            min_size, actual_len, alloc_hint, trans);
7434 }
7435
7436 static int btrfs_set_page_dirty(struct page *page)
7437 {
7438         return __set_page_dirty_nobuffers(page);
7439 }
7440
7441 static int btrfs_permission(struct inode *inode, int mask, unsigned int flags)
7442 {
7443         struct btrfs_root *root = BTRFS_I(inode)->root;
7444
7445         if (btrfs_root_readonly(root) && (mask & MAY_WRITE))
7446                 return -EROFS;
7447         if ((BTRFS_I(inode)->flags & BTRFS_INODE_READONLY) && (mask & MAY_WRITE))
7448                 return -EACCES;
7449         return generic_permission(inode, mask, flags, btrfs_check_acl);
7450 }
7451
7452 static const struct inode_operations btrfs_dir_inode_operations = {
7453         .getattr        = btrfs_getattr,
7454         .lookup         = btrfs_lookup,
7455         .create         = btrfs_create,
7456         .unlink         = btrfs_unlink,
7457         .link           = btrfs_link,
7458         .mkdir          = btrfs_mkdir,
7459         .rmdir          = btrfs_rmdir,
7460         .rename         = btrfs_rename,
7461         .symlink        = btrfs_symlink,
7462         .setattr        = btrfs_setattr,
7463         .mknod          = btrfs_mknod,
7464         .setxattr       = btrfs_setxattr,
7465         .getxattr       = btrfs_getxattr,
7466         .listxattr      = btrfs_listxattr,
7467         .removexattr    = btrfs_removexattr,
7468         .permission     = btrfs_permission,
7469 };
7470 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7471         .lookup         = btrfs_lookup,
7472         .permission     = btrfs_permission,
7473 };
7474
7475 static const struct file_operations btrfs_dir_file_operations = {
7476         .llseek         = generic_file_llseek,
7477         .read           = generic_read_dir,
7478         .readdir        = btrfs_real_readdir,
7479         .unlocked_ioctl = btrfs_ioctl,
7480 #ifdef CONFIG_COMPAT
7481         .compat_ioctl   = btrfs_ioctl,
7482 #endif
7483         .release        = btrfs_release_file,
7484         .fsync          = btrfs_sync_file,
7485 };
7486
7487 static struct extent_io_ops btrfs_extent_io_ops = {
7488         .fill_delalloc = run_delalloc_range,
7489         .submit_bio_hook = btrfs_submit_bio_hook,
7490         .merge_bio_hook = btrfs_merge_bio_hook,
7491         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7492         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7493         .writepage_start_hook = btrfs_writepage_start_hook,
7494         .readpage_io_failed_hook = btrfs_io_failed_hook,
7495         .set_bit_hook = btrfs_set_bit_hook,
7496         .clear_bit_hook = btrfs_clear_bit_hook,
7497         .merge_extent_hook = btrfs_merge_extent_hook,
7498         .split_extent_hook = btrfs_split_extent_hook,
7499 };
7500
7501 /*
7502  * btrfs doesn't support the bmap operation because swapfiles
7503  * use bmap to make a mapping of extents in the file.  They assume
7504  * these extents won't change over the life of the file and they
7505  * use the bmap result to do IO directly to the drive.
7506  *
7507  * the btrfs bmap call would return logical addresses that aren't
7508  * suitable for IO and they also will change frequently as COW
7509  * operations happen.  So, swapfile + btrfs == corruption.
7510  *
7511  * For now we're avoiding this by dropping bmap.
7512  */
7513 static const struct address_space_operations btrfs_aops = {
7514         .readpage       = btrfs_readpage,
7515         .writepage      = btrfs_writepage,
7516         .writepages     = btrfs_writepages,
7517         .readpages      = btrfs_readpages,
7518         .sync_page      = block_sync_page,
7519         .direct_IO      = btrfs_direct_IO,
7520         .invalidatepage = btrfs_invalidatepage,
7521         .releasepage    = btrfs_releasepage,
7522         .set_page_dirty = btrfs_set_page_dirty,
7523         .error_remove_page = generic_error_remove_page,
7524 };
7525
7526 static const struct address_space_operations btrfs_symlink_aops = {
7527         .readpage       = btrfs_readpage,
7528         .writepage      = btrfs_writepage,
7529         .invalidatepage = btrfs_invalidatepage,
7530         .releasepage    = btrfs_releasepage,
7531 };
7532
7533 static const struct inode_operations btrfs_file_inode_operations = {
7534         .getattr        = btrfs_getattr,
7535         .setattr        = btrfs_setattr,
7536         .setxattr       = btrfs_setxattr,
7537         .getxattr       = btrfs_getxattr,
7538         .listxattr      = btrfs_listxattr,
7539         .removexattr    = btrfs_removexattr,
7540         .permission     = btrfs_permission,
7541         .fiemap         = btrfs_fiemap,
7542 };
7543 static const struct inode_operations btrfs_special_inode_operations = {
7544         .getattr        = btrfs_getattr,
7545         .setattr        = btrfs_setattr,
7546         .permission     = btrfs_permission,
7547         .setxattr       = btrfs_setxattr,
7548         .getxattr       = btrfs_getxattr,
7549         .listxattr      = btrfs_listxattr,
7550         .removexattr    = btrfs_removexattr,
7551 };
7552 static const struct inode_operations btrfs_symlink_inode_operations = {
7553         .readlink       = generic_readlink,
7554         .follow_link    = page_follow_link_light,
7555         .put_link       = page_put_link,
7556         .getattr        = btrfs_getattr,
7557         .permission     = btrfs_permission,
7558         .setxattr       = btrfs_setxattr,
7559         .getxattr       = btrfs_getxattr,
7560         .listxattr      = btrfs_listxattr,
7561         .removexattr    = btrfs_removexattr,
7562 };
7563
7564 const struct dentry_operations btrfs_dentry_operations = {
7565         .d_delete       = btrfs_dentry_delete,
7566 };