fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 struct kmem_cache *btrfs_trans_handle_cachep;
  75 struct kmem_cache *btrfs_transaction_cachep;
  76 struct kmem_cache *btrfs_path_cachep;
  77 struct kmem_cache *btrfs_free_space_cachep;
  78
  79 #define S_SHIFT 12
  80 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  81         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  82         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  83         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  84         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  85         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  86         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  87         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  88 };
  89
  90 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  91 static int btrfs_truncate(struct inode *inode);
  92 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  93 static noinline int cow_file_range(struct inode *inode,
  94                                    struct page *locked_page,
  95                                    u64 start, u64 end, int *page_started,
  96                                    unsigned long *nr_written, int unlock);
  97 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
  98                                 struct btrfs_root *root, struct inode *inode);
  99
 100 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 101                                      struct inode *inode,  struct inode *dir,
 102                                      const struct qstr *qstr)
 103 {
 104         int err;
 105
 106         err = btrfs_init_acl(trans, inode, dir);
 107         if (!err)
 108                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 109         return err;
 110 }
 111
 112 /*
 113  * this does all the hard work for inserting an inline extent into
 114  * the btree.  The caller should have done a btrfs_drop_extents so that
 115  * no overlapping inline items exist in the btree
 116  */
 117 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 118                                 struct btrfs_root *root, struct inode *inode,
 119                                 u64 start, size_t size, size_t compressed_size,
 120                                 int compress_type,
 121                                 struct page **compressed_pages)
 122 {
 123         struct btrfs_key key;
 124         struct btrfs_path *path;
 125         struct extent_buffer *leaf;
 126         struct page *page = NULL;
 127         char *kaddr;
 128         unsigned long ptr;
 129         struct btrfs_file_extent_item *ei;
 130         int err = 0;
 131         int ret;
 132         size_t cur_size = size;
 133         size_t datasize;
 134         unsigned long offset;
 135
 136         if (compressed_size && compressed_pages)
 137                 cur_size = compressed_size;
 138
 139         path = btrfs_alloc_path();
 140         if (!path)
 141                 return -ENOMEM;
 142
 143         path->leave_spinning = 1;
 144
 145         key.objectid = btrfs_ino(inode);
 146         key.offset = start;
 147         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 148         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 149
 150         inode_add_bytes(inode, size);
 151         ret = btrfs_insert_empty_item(trans, root, path, &key,
 152                                       datasize);
 153         if (ret) {
 154                 err = ret;
 155                 goto fail;
 156         }
 157         leaf = path->nodes[0];
 158         ei = btrfs_item_ptr(leaf, path->slots[0],
 159                             struct btrfs_file_extent_item);
 160         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 161         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 162         btrfs_set_file_extent_encryption(leaf, ei, 0);
 163         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 164         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 165         ptr = btrfs_file_extent_inline_start(ei);
 166
 167         if (compress_type != BTRFS_COMPRESS_NONE) {
 168                 struct page *cpage;
 169                 int i = 0;
 170                 while (compressed_size > 0) {
 171                         cpage = compressed_pages[i];
 172                         cur_size = min_t(unsigned long, compressed_size,
 173                                        PAGE_CACHE_SIZE);
 174
 175                         kaddr = kmap_atomic(cpage);
 176                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 177                         kunmap_atomic(kaddr);
 178
 179                         i++;
 180                         ptr += cur_size;
 181                         compressed_size -= cur_size;
 182                 }
 183                 btrfs_set_file_extent_compression(leaf, ei,
 184                                                   compress_type);
 185         } else {
 186                 page = find_get_page(inode->i_mapping,
 187                                      start >> PAGE_CACHE_SHIFT);
 188                 btrfs_set_file_extent_compression(leaf, ei, 0);
 189                 kaddr = kmap_atomic(page);
 190                 offset = start & (PAGE_CACHE_SIZE - 1);
 191                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 192                 kunmap_atomic(kaddr);
 193                 page_cache_release(page);
 194         }
 195         btrfs_mark_buffer_dirty(leaf);
 196         btrfs_free_path(path);
 197
 198         /*
 199          * we're an inline extent, so nobody can
 200          * extend the file past i_size without locking
 201          * a page we already have locked.
 202          *
 203          * We must do any isize and inode updates
 204          * before we unlock the pages.  Otherwise we
 205          * could end up racing with unlink.
 206          */
 207         BTRFS_I(inode)->disk_i_size = inode->i_size;
 208         ret = btrfs_update_inode(trans, root, inode);
 209
 210         return ret;
 211 fail:
 212         btrfs_free_path(path);
 213         return err;
 214 }
 215
 216
 217 /*
 218  * conditionally insert an inline extent into the file.  This
 219  * does the checks required to make sure the data is small enough
 220  * to fit as an inline extent.
 221  */
 222 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 223                                  struct btrfs_root *root,
 224                                  struct inode *inode, u64 start, u64 end,
 225                                  size_t compressed_size, int compress_type,
 226                                  struct page **compressed_pages)
 227 {
 228         u64 isize = i_size_read(inode);
 229         u64 actual_end = min(end + 1, isize);
 230         u64 inline_len = actual_end - start;
 231         u64 aligned_end = (end + root->sectorsize - 1) &
 232                         ~((u64)root->sectorsize - 1);
 233         u64 hint_byte;
 234         u64 data_len = inline_len;
 235         int ret;
 236
 237         if (compressed_size)
 238                 data_len = compressed_size;
 239
 240         if (start > 0 ||
 241             actual_end >= PAGE_CACHE_SIZE ||
 242             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 243             (!compressed_size &&
 244             (actual_end & (root->sectorsize - 1)) == 0) ||
 245             end + 1 < isize ||
 246             data_len > root->fs_info->max_inline) {
 247                 return 1;
 248         }
 249
 250         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 251                                  &hint_byte, 1);
 252         if (ret)
 253                 return ret;
 254
 255         if (isize > actual_end)
 256                 inline_len = min_t(u64, isize, actual_end);
 257         ret = insert_inline_extent(trans, root, inode, start,
 258                                    inline_len, compressed_size,
 259                                    compress_type, compressed_pages);
 260         if (ret) {
 261                 btrfs_abort_transaction(trans, root, ret);
 262                 return ret;
 263         }
 264         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 265         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         int compress_type;
 276         struct list_head list;
 277 };
 278
 279 struct async_cow {
 280         struct inode *inode;
 281         struct btrfs_root *root;
 282         struct page *locked_page;
 283         u64 start;
 284         u64 end;
 285         struct list_head extents;
 286         struct btrfs_work work;
 287 };
 288
 289 static noinline int add_async_extent(struct async_cow *cow,
 290                                      u64 start, u64 ram_size,
 291                                      u64 compressed_size,
 292                                      struct page **pages,
 293                                      unsigned long nr_pages,
 294                                      int compress_type)
 295 {
 296         struct async_extent *async_extent;
 297
 298         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 299         BUG_ON(!async_extent); /* -ENOMEM */
 300         async_extent->start = start;
 301         async_extent->ram_size = ram_size;
 302         async_extent->compressed_size = compressed_size;
 303         async_extent->pages = pages;
 304         async_extent->nr_pages = nr_pages;
 305         async_extent->compress_type = compress_type;
 306         list_add_tail(&async_extent->list, &cow->extents);
 307         return 0;
 308 }
 309
 310 /*
 311  * we create compressed extents in two phases.  The first
 312  * phase compresses a range of pages that have already been
 313  * locked (both pages and state bits are locked).
 314  *
 315  * This is done inside an ordered work queue, and the compression
 316  * is spread across many cpus.  The actual IO submission is step
 317  * two, and the ordered work queue takes care of making sure that
 318  * happens in the same order things were put onto the queue by
 319  * writepages and friends.
 320  *
 321  * If this code finds it can't get good compression, it puts an
 322  * entry onto the work queue to write the uncompressed bytes.  This
 323  * makes sure that both compressed inodes and uncompressed inodes
 324  * are written in the same order that pdflush sent them down.
 325  */
 326 static noinline int compress_file_range(struct inode *inode,
 327                                         struct page *locked_page,
 328                                         u64 start, u64 end,
 329                                         struct async_cow *async_cow,
 330                                         int *num_added)
 331 {
 332         struct btrfs_root *root = BTRFS_I(inode)->root;
 333         struct btrfs_trans_handle *trans;
 334         u64 num_bytes;
 335         u64 blocksize = root->sectorsize;
 336         u64 actual_end;
 337         u64 isize = i_size_read(inode);
 338         int ret = 0;
 339         struct page **pages = NULL;
 340         unsigned long nr_pages;
 341         unsigned long nr_pages_ret = 0;
 342         unsigned long total_compressed = 0;
 343         unsigned long total_in = 0;
 344         unsigned long max_compressed = 128 * 1024;
 345         unsigned long max_uncompressed = 128 * 1024;
 346         int i;
 347         int will_compress;
 348         int compress_type = root->fs_info->compress_type;
 349
 350         /* if this is a small write inside eof, kick off a defrag */
 351         if ((end - start + 1) < 16 * 1024 &&
 352             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 353                 btrfs_add_inode_defrag(NULL, inode);
 354
 355         actual_end = min_t(u64, isize, end + 1);
 356 again:
 357         will_compress = 0;
 358         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 359         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 360
 361         /*
 362          * we don't want to send crud past the end of i_size through
 363          * compression, that's just a waste of CPU time.  So, if the
 364          * end of the file is before the start of our current
 365          * requested range of bytes, we bail out to the uncompressed
 366          * cleanup code that can deal with all of this.
 367          *
 368          * It isn't really the fastest way to fix things, but this is a
 369          * very uncommon corner.
 370          */
 371         if (actual_end <= start)
 372                 goto cleanup_and_bail_uncompressed;
 373
 374         total_compressed = actual_end - start;
 375
 376         /* we want to make sure that amount of ram required to uncompress
 377          * an extent is reasonable, so we limit the total size in ram
 378          * of a compressed extent to 128k.  This is a crucial number
 379          * because it also controls how easily we can spread reads across
 380          * cpus for decompression.
 381          *
 382          * We also want to make sure the amount of IO required to do
 383          * a random read is reasonably small, so we limit the size of
 384          * a compressed extent to 128k.
 385          */
 386         total_compressed = min(total_compressed, max_uncompressed);
 387         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 388         num_bytes = max(blocksize,  num_bytes);
 389         total_in = 0;
 390         ret = 0;
 391
 392         /*
 393          * we do compression for mount -o compress and when the
 394          * inode has not been flagged as nocompress.  This flag can
 395          * change at any time if we discover bad compression ratios.
 396          */
 397         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 398             (btrfs_test_opt(root, COMPRESS) ||
 399              (BTRFS_I(inode)->force_compress) ||
 400              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 401                 WARN_ON(pages);
 402                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 403                 if (!pages) {
 404                         /* just bail out to the uncompressed code */
 405                         goto cont;
 406                 }
 407
 408                 if (BTRFS_I(inode)->force_compress)
 409                         compress_type = BTRFS_I(inode)->force_compress;
 410
 411                 ret = btrfs_compress_pages(compress_type,
 412                                            inode->i_mapping, start,
 413                                            total_compressed, pages,
 414                                            nr_pages, &nr_pages_ret,
 415                                            &total_in,
 416                                            &total_compressed,
 417                                            max_compressed);
 418
 419                 if (!ret) {
 420                         unsigned long offset = total_compressed &
 421                                 (PAGE_CACHE_SIZE - 1);
 422                         struct page *page = pages[nr_pages_ret - 1];
 423                         char *kaddr;
 424
 425                         /* zero the tail end of the last page, we might be
 426                          * sending it down to disk
 427                          */
 428                         if (offset) {
 429                                 kaddr = kmap_atomic(page);
 430                                 memset(kaddr + offset, 0,
 431                                        PAGE_CACHE_SIZE - offset);
 432                                 kunmap_atomic(kaddr);
 433                         }
 434                         will_compress = 1;
 435                 }
 436         }
 437 cont:
 438         if (start == 0) {
 439                 trans = btrfs_join_transaction(root);
 440                 if (IS_ERR(trans)) {
 441                         ret = PTR_ERR(trans);
 442                         trans = NULL;
 443                         goto cleanup_and_out;
 444                 }
 445                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 446
 447                 /* lets try to make an inline extent */
 448                 if (ret || total_in < (actual_end - start)) {
 449                         /* we didn't compress the entire range, try
 450                          * to make an uncompressed inline extent.
 451                          */
 452                         ret = cow_file_range_inline(trans, root, inode,
 453                                                     start, end, 0, 0, NULL);
 454                 } else {
 455                         /* try making a compressed inline extent */
 456                         ret = cow_file_range_inline(trans, root, inode,
 457                                                     start, end,
 458                                                     total_compressed,
 459                                                     compress_type, pages);
 460                 }
 461                 if (ret <= 0) {
 462                         /*
 463                          * inline extent creation worked or returned error,
 464                          * we don't need to create any more async work items.
 465                          * Unlock and free up our temp pages.
 466                          */
 467                         extent_clear_unlock_delalloc(inode,
 468                              &BTRFS_I(inode)->io_tree,
 469                              start, end, NULL,
 470                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 471                              EXTENT_CLEAR_DELALLOC |
 472                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 473
 474                         btrfs_end_transaction(trans, root);
 475                         goto free_pages_out;
 476                 }
 477                 btrfs_end_transaction(trans, root);
 478         }
 479
 480         if (will_compress) {
 481                 /*
 482                  * we aren't doing an inline extent round the compressed size
 483                  * up to a block size boundary so the allocator does sane
 484                  * things
 485                  */
 486                 total_compressed = (total_compressed + blocksize - 1) &
 487                         ~(blocksize - 1);
 488
 489                 /*
 490                  * one last check to make sure the compression is really a
 491                  * win, compare the page count read with the blocks on disk
 492                  */
 493                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 494                         ~(PAGE_CACHE_SIZE - 1);
 495                 if (total_compressed >= total_in) {
 496                         will_compress = 0;
 497                 } else {
 498                         num_bytes = total_in;
 499                 }
 500         }
 501         if (!will_compress && pages) {
 502                 /*
 503                  * the compression code ran but failed to make things smaller,
 504                  * free any pages it allocated and our page pointer array
 505                  */
 506                 for (i = 0; i < nr_pages_ret; i++) {
 507                         WARN_ON(pages[i]->mapping);
 508                         page_cache_release(pages[i]);
 509                 }
 510                 kfree(pages);
 511                 pages = NULL;
 512                 total_compressed = 0;
 513                 nr_pages_ret = 0;
 514
 515                 /* flag the file so we don't compress in the future */
 516                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 517                     !(BTRFS_I(inode)->force_compress)) {
 518                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 519                 }
 520         }
 521         if (will_compress) {
 522                 *num_added += 1;
 523
 524                 /* the async work queues will take care of doing actual
 525                  * allocation on disk for these compressed pages,
 526                  * and will submit them to the elevator.
 527                  */
 528                 add_async_extent(async_cow, start, num_bytes,
 529                                  total_compressed, pages, nr_pages_ret,
 530                                  compress_type);
 531
 532                 if (start + num_bytes < end) {
 533                         start += num_bytes;
 534                         pages = NULL;
 535                         cond_resched();
 536                         goto again;
 537                 }
 538         } else {
 539 cleanup_and_bail_uncompressed:
 540                 /*
 541                  * No compression, but we still need to write the pages in
 542                  * the file we've been given so far.  redirty the locked
 543                  * page if it corresponds to our extent and set things up
 544                  * for the async work queue to run cow_file_range to do
 545                  * the normal delalloc dance
 546                  */
 547                 if (page_offset(locked_page) >= start &&
 548                     page_offset(locked_page) <= end) {
 549                         __set_page_dirty_nobuffers(locked_page);
 550                         /* unlocked later on in the async handlers */
 551                 }
 552                 add_async_extent(async_cow, start, end - start + 1,
 553                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 554                 *num_added += 1;
 555         }
 556
 557 out:
 558         return ret;
 559
 560 free_pages_out:
 561         for (i = 0; i < nr_pages_ret; i++) {
 562                 WARN_ON(pages[i]->mapping);
 563                 page_cache_release(pages[i]);
 564         }
 565         kfree(pages);
 566
 567         goto out;
 568
 569 cleanup_and_out:
 570         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 571                                      start, end, NULL,
 572                                      EXTENT_CLEAR_UNLOCK_PAGE |
 573                                      EXTENT_CLEAR_DIRTY |
 574                                      EXTENT_CLEAR_DELALLOC |
 575                                      EXTENT_SET_WRITEBACK |
 576                                      EXTENT_END_WRITEBACK);
 577         if (!trans || IS_ERR(trans))
 578                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 579         else
 580                 btrfs_abort_transaction(trans, root, ret);
 581         goto free_pages_out;
 582 }
 583
 584 /*
 585  * phase two of compressed writeback.  This is the ordered portion
 586  * of the code, which only gets called in the order the work was
 587  * queued.  We walk all the async extents created by compress_file_range
 588  * and send them down to the disk.
 589  */
 590 static noinline int submit_compressed_extents(struct inode *inode,
 591                                               struct async_cow *async_cow)
 592 {
 593         struct async_extent *async_extent;
 594         u64 alloc_hint = 0;
 595         struct btrfs_trans_handle *trans;
 596         struct btrfs_key ins;
 597         struct extent_map *em;
 598         struct btrfs_root *root = BTRFS_I(inode)->root;
 599         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 600         struct extent_io_tree *io_tree;
 601         int ret = 0;
 602
 603         if (list_empty(&async_cow->extents))
 604                 return 0;
 605
 606
 607         while (!list_empty(&async_cow->extents)) {
 608                 async_extent = list_entry(async_cow->extents.next,
 609                                           struct async_extent, list);
 610                 list_del(&async_extent->list);
 611
 612                 io_tree = &BTRFS_I(inode)->io_tree;
 613
 614 retry:
 615                 /* did the compression code fall back to uncompressed IO? */
 616                 if (!async_extent->pages) {
 617                         int page_started = 0;
 618                         unsigned long nr_written = 0;
 619
 620                         lock_extent(io_tree, async_extent->start,
 621                                          async_extent->start +
 622                                          async_extent->ram_size - 1);
 623
 624                         /* allocate blocks */
 625                         ret = cow_file_range(inode, async_cow->locked_page,
 626                                              async_extent->start,
 627                                              async_extent->start +
 628                                              async_extent->ram_size - 1,
 629                                              &page_started, &nr_written, 0);
 630
 631                         /* JDM XXX */
 632
 633                         /*
 634                          * if page_started, cow_file_range inserted an
 635                          * inline extent and took care of all the unlocking
 636                          * and IO for us.  Otherwise, we need to submit
 637                          * all those pages down to the drive.
 638                          */
 639                         if (!page_started && !ret)
 640                                 extent_write_locked_range(io_tree,
 641                                                   inode, async_extent->start,
 642                                                   async_extent->start +
 643                                                   async_extent->ram_size - 1,
 644                                                   btrfs_get_extent,
 645                                                   WB_SYNC_ALL);
 646                         kfree(async_extent);
 647                         cond_resched();
 648                         continue;
 649                 }
 650
 651                 lock_extent(io_tree, async_extent->start,
 652                             async_extent->start + async_extent->ram_size - 1);
 653
 654                 trans = btrfs_join_transaction(root);
 655                 if (IS_ERR(trans)) {
 656                         ret = PTR_ERR(trans);
 657                 } else {
 658                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 659                         ret = btrfs_reserve_extent(trans, root,
 660                                            async_extent->compressed_size,
 661                                            async_extent->compressed_size,
 662                                            0, alloc_hint, &ins, 1);
 663                         if (ret)
 664                                 btrfs_abort_transaction(trans, root, ret);
 665                         btrfs_end_transaction(trans, root);
 666                 }
 667
 668                 if (ret) {
 669                         int i;
 670                         for (i = 0; i < async_extent->nr_pages; i++) {
 671                                 WARN_ON(async_extent->pages[i]->mapping);
 672                                 page_cache_release(async_extent->pages[i]);
 673                         }
 674                         kfree(async_extent->pages);
 675                         async_extent->nr_pages = 0;
 676                         async_extent->pages = NULL;
 677                         unlock_extent(io_tree, async_extent->start,
 678                                       async_extent->start +
 679                                       async_extent->ram_size - 1);
 680                         if (ret == -ENOSPC)
 681                                 goto retry;
 682                         goto out_free; /* JDM: Requeue? */
 683                 }
 684
 685                 /*
 686                  * here we're doing allocation and writeback of the
 687                  * compressed pages
 688                  */
 689                 btrfs_drop_extent_cache(inode, async_extent->start,
 690                                         async_extent->start +
 691                                         async_extent->ram_size - 1, 0);
 692
 693                 em = alloc_extent_map();
 694                 BUG_ON(!em); /* -ENOMEM */
 695                 em->start = async_extent->start;
 696                 em->len = async_extent->ram_size;
 697                 em->orig_start = em->start;
 698
 699                 em->block_start = ins.objectid;
 700                 em->block_len = ins.offset;
 701                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 702                 em->compress_type = async_extent->compress_type;
 703                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 704                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 705
 706                 while (1) {
 707                         write_lock(&em_tree->lock);
 708                         ret = add_extent_mapping(em_tree, em);
 709                         write_unlock(&em_tree->lock);
 710                         if (ret != -EEXIST) {
 711                                 free_extent_map(em);
 712                                 break;
 713                         }
 714                         btrfs_drop_extent_cache(inode, async_extent->start,
 715                                                 async_extent->start +
 716                                                 async_extent->ram_size - 1, 0);
 717                 }
 718
 719                 ret = btrfs_add_ordered_extent_compress(inode,
 720                                                 async_extent->start,
 721                                                 ins.objectid,
 722                                                 async_extent->ram_size,
 723                                                 ins.offset,
 724                                                 BTRFS_ORDERED_COMPRESSED,
 725                                                 async_extent->compress_type);
 726                 BUG_ON(ret); /* -ENOMEM */
 727
 728                 /*
 729                  * clear dirty, set writeback and unlock the pages.
 730                  */
 731                 extent_clear_unlock_delalloc(inode,
 732                                 &BTRFS_I(inode)->io_tree,
 733                                 async_extent->start,
 734                                 async_extent->start +
 735                                 async_extent->ram_size - 1,
 736                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 737                                 EXTENT_CLEAR_UNLOCK |
 738                                 EXTENT_CLEAR_DELALLOC |
 739                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 740
 741                 ret = btrfs_submit_compressed_write(inode,
 742                                     async_extent->start,
 743                                     async_extent->ram_size,
 744                                     ins.objectid,
 745                                     ins.offset, async_extent->pages,
 746                                     async_extent->nr_pages);
 747
 748                 BUG_ON(ret); /* -ENOMEM */
 749                 alloc_hint = ins.objectid + ins.offset;
 750                 kfree(async_extent);
 751                 cond_resched();
 752         }
 753         ret = 0;
 754 out:
 755         return ret;
 756 out_free:
 757         kfree(async_extent);
 758         goto out;
 759 }
 760
 761 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 762                                       u64 num_bytes)
 763 {
 764         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 765         struct extent_map *em;
 766         u64 alloc_hint = 0;
 767
 768         read_lock(&em_tree->lock);
 769         em = search_extent_mapping(em_tree, start, num_bytes);
 770         if (em) {
 771                 /*
 772                  * if block start isn't an actual block number then find the
 773                  * first block in this inode and use that as a hint.  If that
 774                  * block is also bogus then just don't worry about it.
 775                  */
 776                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 777                         free_extent_map(em);
 778                         em = search_extent_mapping(em_tree, 0, 0);
 779                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 780                                 alloc_hint = em->block_start;
 781                         if (em)
 782                                 free_extent_map(em);
 783                 } else {
 784                         alloc_hint = em->block_start;
 785                         free_extent_map(em);
 786                 }
 787         }
 788         read_unlock(&em_tree->lock);
 789
 790         return alloc_hint;
 791 }
 792
 793 /*
 794  * when extent_io.c finds a delayed allocation range in the file,
 795  * the call backs end up in this code.  The basic idea is to
 796  * allocate extents on disk for the range, and create ordered data structs
 797  * in ram to track those extents.
 798  *
 799  * locked_page is the page that writepage had locked already.  We use
 800  * it to make sure we don't do extra locks or unlocks.
 801  *
 802  * *page_started is set to one if we unlock locked_page and do everything
 803  * required to start IO on it.  It may be clean and already done with
 804  * IO when we return.
 805  */
 806 static noinline int cow_file_range(struct inode *inode,
 807                                    struct page *locked_page,
 808                                    u64 start, u64 end, int *page_started,
 809                                    unsigned long *nr_written,
 810                                    int unlock)
 811 {
 812         struct btrfs_root *root = BTRFS_I(inode)->root;
 813         struct btrfs_trans_handle *trans;
 814         u64 alloc_hint = 0;
 815         u64 num_bytes;
 816         unsigned long ram_size;
 817         u64 disk_num_bytes;
 818         u64 cur_alloc_size;
 819         u64 blocksize = root->sectorsize;
 820         struct btrfs_key ins;
 821         struct extent_map *em;
 822         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 823         int ret = 0;
 824
 825         BUG_ON(btrfs_is_free_space_inode(root, inode));
 826         trans = btrfs_join_transaction(root);
 827         if (IS_ERR(trans)) {
 828                 extent_clear_unlock_delalloc(inode,
 829                              &BTRFS_I(inode)->io_tree,
 830                              start, end, NULL,
 831                              EXTENT_CLEAR_UNLOCK_PAGE |
 832                              EXTENT_CLEAR_UNLOCK |
 833                              EXTENT_CLEAR_DELALLOC |
 834                              EXTENT_CLEAR_DIRTY |
 835                              EXTENT_SET_WRITEBACK |
 836                              EXTENT_END_WRITEBACK);
 837                 return PTR_ERR(trans);
 838         }
 839         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 840
 841         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 842         num_bytes = max(blocksize,  num_bytes);
 843         disk_num_bytes = num_bytes;
 844         ret = 0;
 845
 846         /* if this is a small write inside eof, kick off defrag */
 847         if (num_bytes < 64 * 1024 &&
 848             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 849                 btrfs_add_inode_defrag(trans, inode);
 850
 851         if (start == 0) {
 852                 /* lets try to make an inline extent */
 853                 ret = cow_file_range_inline(trans, root, inode,
 854                                             start, end, 0, 0, NULL);
 855                 if (ret == 0) {
 856                         extent_clear_unlock_delalloc(inode,
 857                                      &BTRFS_I(inode)->io_tree,
 858                                      start, end, NULL,
 859                                      EXTENT_CLEAR_UNLOCK_PAGE |
 860                                      EXTENT_CLEAR_UNLOCK |
 861                                      EXTENT_CLEAR_DELALLOC |
 862                                      EXTENT_CLEAR_DIRTY |
 863                                      EXTENT_SET_WRITEBACK |
 864                                      EXTENT_END_WRITEBACK);
 865
 866                         *nr_written = *nr_written +
 867                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 868                         *page_started = 1;
 869                         goto out;
 870                 } else if (ret < 0) {
 871                         btrfs_abort_transaction(trans, root, ret);
 872                         goto out_unlock;
 873                 }
 874         }
 875
 876         BUG_ON(disk_num_bytes >
 877                btrfs_super_total_bytes(root->fs_info->super_copy));
 878
 879         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 880         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 881
 882         while (disk_num_bytes > 0) {
 883                 unsigned long op;
 884
 885                 cur_alloc_size = disk_num_bytes;
 886                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 887                                            root->sectorsize, 0, alloc_hint,
 888                                            &ins, 1);
 889                 if (ret < 0) {
 890                         btrfs_abort_transaction(trans, root, ret);
 891                         goto out_unlock;
 892                 }
 893
 894                 em = alloc_extent_map();
 895                 BUG_ON(!em); /* -ENOMEM */
 896                 em->start = start;
 897                 em->orig_start = em->start;
 898                 ram_size = ins.offset;
 899                 em->len = ins.offset;
 900
 901                 em->block_start = ins.objectid;
 902                 em->block_len = ins.offset;
 903                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 904                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 905
 906                 while (1) {
 907                         write_lock(&em_tree->lock);
 908                         ret = add_extent_mapping(em_tree, em);
 909                         write_unlock(&em_tree->lock);
 910                         if (ret != -EEXIST) {
 911                                 free_extent_map(em);
 912                                 break;
 913                         }
 914                         btrfs_drop_extent_cache(inode, start,
 915                                                 start + ram_size - 1, 0);
 916                 }
 917
 918                 cur_alloc_size = ins.offset;
 919                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 920                                                ram_size, cur_alloc_size, 0);
 921                 BUG_ON(ret); /* -ENOMEM */
 922
 923                 if (root->root_key.objectid ==
 924                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 925                         ret = btrfs_reloc_clone_csums(inode, start,
 926                                                       cur_alloc_size);
 927                         if (ret) {
 928                                 btrfs_abort_transaction(trans, root, ret);
 929                                 goto out_unlock;
 930                         }
 931                 }
 932
 933                 if (disk_num_bytes < cur_alloc_size)
 934                         break;
 935
 936                 /* we're not doing compressed IO, don't unlock the first
 937                  * page (which the caller expects to stay locked), don't
 938                  * clear any dirty bits and don't set any writeback bits
 939                  *
 940                  * Do set the Private2 bit so we know this page was properly
 941                  * setup for writepage
 942                  */
 943                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 944                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 945                         EXTENT_SET_PRIVATE2;
 946
 947                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 948                                              start, start + ram_size - 1,
 949                                              locked_page, op);
 950                 disk_num_bytes -= cur_alloc_size;
 951                 num_bytes -= cur_alloc_size;
 952                 alloc_hint = ins.objectid + ins.offset;
 953                 start += cur_alloc_size;
 954         }
 955         ret = 0;
 956 out:
 957         btrfs_end_transaction(trans, root);
 958
 959         return ret;
 960 out_unlock:
 961         extent_clear_unlock_delalloc(inode,
 962                      &BTRFS_I(inode)->io_tree,
 963                      start, end, NULL,
 964                      EXTENT_CLEAR_UNLOCK_PAGE |
 965                      EXTENT_CLEAR_UNLOCK |
 966                      EXTENT_CLEAR_DELALLOC |
 967                      EXTENT_CLEAR_DIRTY |
 968                      EXTENT_SET_WRITEBACK |
 969                      EXTENT_END_WRITEBACK);
 970
 971         goto out;
 972 }
 973
 974 /*
 975  * work queue call back to started compression on a file and pages
 976  */
 977 static noinline void async_cow_start(struct btrfs_work *work)
 978 {
 979         struct async_cow *async_cow;
 980         int num_added = 0;
 981         async_cow = container_of(work, struct async_cow, work);
 982
 983         compress_file_range(async_cow->inode, async_cow->locked_page,
 984                             async_cow->start, async_cow->end, async_cow,
 985                             &num_added);
 986         if (num_added == 0)
 987                 async_cow->inode = NULL;
 988 }
 989
 990 /*
 991  * work queue call back to submit previously compressed pages
 992  */
 993 static noinline void async_cow_submit(struct btrfs_work *work)
 994 {
 995         struct async_cow *async_cow;
 996         struct btrfs_root *root;
 997         unsigned long nr_pages;
 998
 999         async_cow = container_of(work, struct async_cow, work);
1000
1001         root = async_cow->root;
1002         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1003                 PAGE_CACHE_SHIFT;
1004
1005         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
1006
1007         if (atomic_read(&root->fs_info->async_delalloc_pages) <
1008             5 * 1042 * 1024 &&
1009             waitqueue_active(&root->fs_info->async_submit_wait))
1010                 wake_up(&root->fs_info->async_submit_wait);
1011
1012         if (async_cow->inode)
1013                 submit_compressed_extents(async_cow->inode, async_cow);
1014 }
1015
1016 static noinline void async_cow_free(struct btrfs_work *work)
1017 {
1018         struct async_cow *async_cow;
1019         async_cow = container_of(work, struct async_cow, work);
1020         kfree(async_cow);
1021 }
1022
1023 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1024                                 u64 start, u64 end, int *page_started,
1025                                 unsigned long *nr_written)
1026 {
1027         struct async_cow *async_cow;
1028         struct btrfs_root *root = BTRFS_I(inode)->root;
1029         unsigned long nr_pages;
1030         u64 cur_end;
1031         int limit = 10 * 1024 * 1042;
1032
1033         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1034                          1, 0, NULL, GFP_NOFS);
1035         while (start < end) {
1036                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1037                 BUG_ON(!async_cow); /* -ENOMEM */
1038                 async_cow->inode = inode;
1039                 async_cow->root = root;
1040                 async_cow->locked_page = locked_page;
1041                 async_cow->start = start;
1042
1043                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1044                         cur_end = end;
1045                 else
1046                         cur_end = min(end, start + 512 * 1024 - 1);
1047
1048                 async_cow->end = cur_end;
1049                 INIT_LIST_HEAD(&async_cow->extents);
1050
1051                 async_cow->work.func = async_cow_start;
1052                 async_cow->work.ordered_func = async_cow_submit;
1053                 async_cow->work.ordered_free = async_cow_free;
1054                 async_cow->work.flags = 0;
1055
1056                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1057                         PAGE_CACHE_SHIFT;
1058                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1059
1060                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1061                                    &async_cow->work);
1062
1063                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1064                         wait_event(root->fs_info->async_submit_wait,
1065                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1066                             limit));
1067                 }
1068
1069                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1070                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1071                         wait_event(root->fs_info->async_submit_wait,
1072                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1073                            0));
1074                 }
1075
1076                 *nr_written += nr_pages;
1077                 start = cur_end + 1;
1078         }
1079         *page_started = 1;
1080         return 0;
1081 }
1082
1083 static noinline int csum_exist_in_range(struct btrfs_root *root,
1084                                         u64 bytenr, u64 num_bytes)
1085 {
1086         int ret;
1087         struct btrfs_ordered_sum *sums;
1088         LIST_HEAD(list);
1089
1090         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1091                                        bytenr + num_bytes - 1, &list, 0);
1092         if (ret == 0 && list_empty(&list))
1093                 return 0;
1094
1095         while (!list_empty(&list)) {
1096                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1097                 list_del(&sums->list);
1098                 kfree(sums);
1099         }
1100         return 1;
1101 }
1102
1103 /*
1104  * when nowcow writeback call back.  This checks for snapshots or COW copies
1105  * of the extents that exist in the file, and COWs the file as required.
1106  *
1107  * If no cow copies or snapshots exist, we write directly to the existing
1108  * blocks on disk
1109  */
1110 static noinline int run_delalloc_nocow(struct inode *inode,
1111                                        struct page *locked_page,
1112                               u64 start, u64 end, int *page_started, int force,
1113                               unsigned long *nr_written)
1114 {
1115         struct btrfs_root *root = BTRFS_I(inode)->root;
1116         struct btrfs_trans_handle *trans;
1117         struct extent_buffer *leaf;
1118         struct btrfs_path *path;
1119         struct btrfs_file_extent_item *fi;
1120         struct btrfs_key found_key;
1121         u64 cow_start;
1122         u64 cur_offset;
1123         u64 extent_end;
1124         u64 extent_offset;
1125         u64 disk_bytenr;
1126         u64 num_bytes;
1127         int extent_type;
1128         int ret, err;
1129         int type;
1130         int nocow;
1131         int check_prev = 1;
1132         bool nolock;
1133         u64 ino = btrfs_ino(inode);
1134
1135         path = btrfs_alloc_path();
1136         if (!path)
1137                 return -ENOMEM;
1138
1139         nolock = btrfs_is_free_space_inode(root, inode);
1140
1141         if (nolock)
1142                 trans = btrfs_join_transaction_nolock(root);
1143         else
1144                 trans = btrfs_join_transaction(root);
1145
1146         if (IS_ERR(trans)) {
1147                 btrfs_free_path(path);
1148                 return PTR_ERR(trans);
1149         }
1150
1151         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1152
1153         cow_start = (u64)-1;
1154         cur_offset = start;
1155         while (1) {
1156                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1157                                                cur_offset, 0);
1158                 if (ret < 0) {
1159                         btrfs_abort_transaction(trans, root, ret);
1160                         goto error;
1161                 }
1162                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1163                         leaf = path->nodes[0];
1164                         btrfs_item_key_to_cpu(leaf, &found_key,
1165                                               path->slots[0] - 1);
1166                         if (found_key.objectid == ino &&
1167                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1168                                 path->slots[0]--;
1169                 }
1170                 check_prev = 0;
1171 next_slot:
1172                 leaf = path->nodes[0];
1173                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1174                         ret = btrfs_next_leaf(root, path);
1175                         if (ret < 0) {
1176                                 btrfs_abort_transaction(trans, root, ret);
1177                                 goto error;
1178                         }
1179                         if (ret > 0)
1180                                 break;
1181                         leaf = path->nodes[0];
1182                 }
1183
1184                 nocow = 0;
1185                 disk_bytenr = 0;
1186                 num_bytes = 0;
1187                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1188
1189                 if (found_key.objectid > ino ||
1190                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1191                     found_key.offset > end)
1192                         break;
1193
1194                 if (found_key.offset > cur_offset) {
1195                         extent_end = found_key.offset;
1196                         extent_type = 0;
1197                         goto out_check;
1198                 }
1199
1200                 fi = btrfs_item_ptr(leaf, path->slots[0],
1201                                     struct btrfs_file_extent_item);
1202                 extent_type = btrfs_file_extent_type(leaf, fi);
1203
1204                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1205                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1206                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1207                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1208                         extent_end = found_key.offset +
1209                                 btrfs_file_extent_num_bytes(leaf, fi);
1210                         if (extent_end <= start) {
1211                                 path->slots[0]++;
1212                                 goto next_slot;
1213                         }
1214                         if (disk_bytenr == 0)
1215                                 goto out_check;
1216                         if (btrfs_file_extent_compression(leaf, fi) ||
1217                             btrfs_file_extent_encryption(leaf, fi) ||
1218                             btrfs_file_extent_other_encoding(leaf, fi))
1219                                 goto out_check;
1220                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1221                                 goto out_check;
1222                         if (btrfs_extent_readonly(root, disk_bytenr))
1223                                 goto out_check;
1224                         if (btrfs_cross_ref_exist(trans, root, ino,
1225                                                   found_key.offset -
1226                                                   extent_offset, disk_bytenr))
1227                                 goto out_check;
1228                         disk_bytenr += extent_offset;
1229                         disk_bytenr += cur_offset - found_key.offset;
1230                         num_bytes = min(end + 1, extent_end) - cur_offset;
1231                         /*
1232                          * force cow if csum exists in the range.
1233                          * this ensure that csum for a given extent are
1234                          * either valid or do not exist.
1235                          */
1236                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1237                                 goto out_check;
1238                         nocow = 1;
1239                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1240                         extent_end = found_key.offset +
1241                                 btrfs_file_extent_inline_len(leaf, fi);
1242                         extent_end = ALIGN(extent_end, root->sectorsize);
1243                 } else {
1244                         BUG_ON(1);
1245                 }
1246 out_check:
1247                 if (extent_end <= start) {
1248                         path->slots[0]++;
1249                         goto next_slot;
1250                 }
1251                 if (!nocow) {
1252                         if (cow_start == (u64)-1)
1253                                 cow_start = cur_offset;
1254                         cur_offset = extent_end;
1255                         if (cur_offset > end)
1256                                 break;
1257                         path->slots[0]++;
1258                         goto next_slot;
1259                 }
1260
1261                 btrfs_release_path(path);
1262                 if (cow_start != (u64)-1) {
1263                         ret = cow_file_range(inode, locked_page, cow_start,
1264                                         found_key.offset - 1, page_started,
1265                                         nr_written, 1);
1266                         if (ret) {
1267                                 btrfs_abort_transaction(trans, root, ret);
1268                                 goto error;
1269                         }
1270                         cow_start = (u64)-1;
1271                 }
1272
1273                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1274                         struct extent_map *em;
1275                         struct extent_map_tree *em_tree;
1276                         em_tree = &BTRFS_I(inode)->extent_tree;
1277                         em = alloc_extent_map();
1278                         BUG_ON(!em); /* -ENOMEM */
1279                         em->start = cur_offset;
1280                         em->orig_start = em->start;
1281                         em->len = num_bytes;
1282                         em->block_len = num_bytes;
1283                         em->block_start = disk_bytenr;
1284                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1285                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1286                         while (1) {
1287                                 write_lock(&em_tree->lock);
1288                                 ret = add_extent_mapping(em_tree, em);
1289                                 write_unlock(&em_tree->lock);
1290                                 if (ret != -EEXIST) {
1291                                         free_extent_map(em);
1292                                         break;
1293                                 }
1294                                 btrfs_drop_extent_cache(inode, em->start,
1295                                                 em->start + em->len - 1, 0);
1296                         }
1297                         type = BTRFS_ORDERED_PREALLOC;
1298                 } else {
1299                         type = BTRFS_ORDERED_NOCOW;
1300                 }
1301
1302                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1303                                                num_bytes, num_bytes, type);
1304                 BUG_ON(ret); /* -ENOMEM */
1305
1306                 if (root->root_key.objectid ==
1307                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1308                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1309                                                       num_bytes);
1310                         if (ret) {
1311                                 btrfs_abort_transaction(trans, root, ret);
1312                                 goto error;
1313                         }
1314                 }
1315
1316                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1317                                 cur_offset, cur_offset + num_bytes - 1,
1318                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1319                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1320                                 EXTENT_SET_PRIVATE2);
1321                 cur_offset = extent_end;
1322                 if (cur_offset > end)
1323                         break;
1324         }
1325         btrfs_release_path(path);
1326
1327         if (cur_offset <= end && cow_start == (u64)-1)
1328                 cow_start = cur_offset;
1329         if (cow_start != (u64)-1) {
1330                 ret = cow_file_range(inode, locked_page, cow_start, end,
1331                                      page_started, nr_written, 1);
1332                 if (ret) {
1333                         btrfs_abort_transaction(trans, root, ret);
1334                         goto error;
1335                 }
1336         }
1337
1338 error:
1339         if (nolock) {
1340                 err = btrfs_end_transaction_nolock(trans, root);
1341         } else {
1342                 err = btrfs_end_transaction(trans, root);
1343         }
1344         if (!ret)
1345                 ret = err;
1346
1347         btrfs_free_path(path);
1348         return ret;
1349 }
1350
1351 /*
1352  * extent_io.c call back to do delayed allocation processing
1353  */
1354 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1355                               u64 start, u64 end, int *page_started,
1356                               unsigned long *nr_written)
1357 {
1358         int ret;
1359         struct btrfs_root *root = BTRFS_I(inode)->root;
1360
1361         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1362                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1363                                          page_started, 1, nr_written);
1364         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1365                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1366                                          page_started, 0, nr_written);
1367         else if (!btrfs_test_opt(root, COMPRESS) &&
1368                  !(BTRFS_I(inode)->force_compress) &&
1369                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1370                 ret = cow_file_range(inode, locked_page, start, end,
1371                                       page_started, nr_written, 1);
1372         else
1373                 ret = cow_file_range_async(inode, locked_page, start, end,
1374                                            page_started, nr_written);
1375         return ret;
1376 }
1377
1378 static void btrfs_split_extent_hook(struct inode *inode,
1379                                     struct extent_state *orig, u64 split)
1380 {
1381         /* not delalloc, ignore it */
1382         if (!(orig->state & EXTENT_DELALLOC))
1383                 return;
1384
1385         spin_lock(&BTRFS_I(inode)->lock);
1386         BTRFS_I(inode)->outstanding_extents++;
1387         spin_unlock(&BTRFS_I(inode)->lock);
1388 }
1389
1390 /*
1391  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1392  * extents so we can keep track of new extents that are just merged onto old
1393  * extents, such as when we are doing sequential writes, so we can properly
1394  * account for the metadata space we'll need.
1395  */
1396 static void btrfs_merge_extent_hook(struct inode *inode,
1397                                     struct extent_state *new,
1398                                     struct extent_state *other)
1399 {
1400         /* not delalloc, ignore it */
1401         if (!(other->state & EXTENT_DELALLOC))
1402                 return;
1403
1404         spin_lock(&BTRFS_I(inode)->lock);
1405         BTRFS_I(inode)->outstanding_extents--;
1406         spin_unlock(&BTRFS_I(inode)->lock);
1407 }
1408
1409 /*
1410  * extent_io.c set_bit_hook, used to track delayed allocation
1411  * bytes in this file, and to maintain the list of inodes that
1412  * have pending delalloc work to be done.
1413  */
1414 static void btrfs_set_bit_hook(struct inode *inode,
1415                                struct extent_state *state, int *bits)
1416 {
1417
1418         /*
1419          * set_bit and clear bit hooks normally require _irqsave/restore
1420          * but in this case, we are only testing for the DELALLOC
1421          * bit, which is only set or cleared with irqs on
1422          */
1423         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1424                 struct btrfs_root *root = BTRFS_I(inode)->root;
1425                 u64 len = state->end + 1 - state->start;
1426                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1427
1428                 if (*bits & EXTENT_FIRST_DELALLOC) {
1429                         *bits &= ~EXTENT_FIRST_DELALLOC;
1430                 } else {
1431                         spin_lock(&BTRFS_I(inode)->lock);
1432                         BTRFS_I(inode)->outstanding_extents++;
1433                         spin_unlock(&BTRFS_I(inode)->lock);
1434                 }
1435
1436                 spin_lock(&root->fs_info->delalloc_lock);
1437                 BTRFS_I(inode)->delalloc_bytes += len;
1438                 root->fs_info->delalloc_bytes += len;
1439                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1440                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1441                                       &root->fs_info->delalloc_inodes);
1442                 }
1443                 spin_unlock(&root->fs_info->delalloc_lock);
1444         }
1445 }
1446
1447 /*
1448  * extent_io.c clear_bit_hook, see set_bit_hook for why
1449  */
1450 static void btrfs_clear_bit_hook(struct inode *inode,
1451                                  struct extent_state *state, int *bits)
1452 {
1453         /*
1454          * set_bit and clear bit hooks normally require _irqsave/restore
1455          * but in this case, we are only testing for the DELALLOC
1456          * bit, which is only set or cleared with irqs on
1457          */
1458         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1459                 struct btrfs_root *root = BTRFS_I(inode)->root;
1460                 u64 len = state->end + 1 - state->start;
1461                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1462
1463                 if (*bits & EXTENT_FIRST_DELALLOC) {
1464                         *bits &= ~EXTENT_FIRST_DELALLOC;
1465                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1466                         spin_lock(&BTRFS_I(inode)->lock);
1467                         BTRFS_I(inode)->outstanding_extents--;
1468                         spin_unlock(&BTRFS_I(inode)->lock);
1469                 }
1470
1471                 if (*bits & EXTENT_DO_ACCOUNTING)
1472                         btrfs_delalloc_release_metadata(inode, len);
1473
1474                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1475                     && do_list)
1476                         btrfs_free_reserved_data_space(inode, len);
1477
1478                 spin_lock(&root->fs_info->delalloc_lock);
1479                 root->fs_info->delalloc_bytes -= len;
1480                 BTRFS_I(inode)->delalloc_bytes -= len;
1481
1482                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1483                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1484                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1485                 }
1486                 spin_unlock(&root->fs_info->delalloc_lock);
1487         }
1488 }
1489
1490 /*
1491  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1492  * we don't create bios that span stripes or chunks
1493  */
1494 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1495                          size_t size, struct bio *bio,
1496                          unsigned long bio_flags)
1497 {
1498         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1499         struct btrfs_mapping_tree *map_tree;
1500         u64 logical = (u64)bio->bi_sector << 9;
1501         u64 length = 0;
1502         u64 map_length;
1503         int ret;
1504
1505         if (bio_flags & EXTENT_BIO_COMPRESSED)
1506                 return 0;
1507
1508         length = bio->bi_size;
1509         map_tree = &root->fs_info->mapping_tree;
1510         map_length = length;
1511         ret = btrfs_map_block(map_tree, READ, logical,
1512                               &map_length, NULL, 0);
1513         /* Will always return 0 or 1 with map_multi == NULL */
1514         BUG_ON(ret < 0);
1515         if (map_length < length + size)
1516                 return 1;
1517         return 0;
1518 }
1519
1520 /*
1521  * in order to insert checksums into the metadata in large chunks,
1522  * we wait until bio submission time.   All the pages in the bio are
1523  * checksummed and sums are attached onto the ordered extent record.
1524  *
1525  * At IO completion time the cums attached on the ordered extent record
1526  * are inserted into the btree
1527  */
1528 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1529                                     struct bio *bio, int mirror_num,
1530                                     unsigned long bio_flags,
1531                                     u64 bio_offset)
1532 {
1533         struct btrfs_root *root = BTRFS_I(inode)->root;
1534         int ret = 0;
1535
1536         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1537         BUG_ON(ret); /* -ENOMEM */
1538         return 0;
1539 }
1540
1541 /*
1542  * in order to insert checksums into the metadata in large chunks,
1543  * we wait until bio submission time.   All the pages in the bio are
1544  * checksummed and sums are attached onto the ordered extent record.
1545  *
1546  * At IO completion time the cums attached on the ordered extent record
1547  * are inserted into the btree
1548  */
1549 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1550                           int mirror_num, unsigned long bio_flags,
1551                           u64 bio_offset)
1552 {
1553         struct btrfs_root *root = BTRFS_I(inode)->root;
1554         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1555 }
1556
1557 /*
1558  * extent_io.c submission hook. This does the right thing for csum calculation
1559  * on write, or reading the csums from the tree before a read
1560  */
1561 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1562                           int mirror_num, unsigned long bio_flags,
1563                           u64 bio_offset)
1564 {
1565         struct btrfs_root *root = BTRFS_I(inode)->root;
1566         int ret = 0;
1567         int skip_sum;
1568         int metadata = 0;
1569
1570         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1571
1572         if (btrfs_is_free_space_inode(root, inode))
1573                 metadata = 2;
1574
1575         ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1576         if (ret)
1577                 return ret;
1578
1579         if (!(rw & REQ_WRITE)) {
1580                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1581                         return btrfs_submit_compressed_read(inode, bio,
1582                                                     mirror_num, bio_flags);
1583                 } else if (!skip_sum) {
1584                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1585                         if (ret)
1586                                 return ret;
1587                 }
1588                 goto mapit;
1589         } else if (!skip_sum) {
1590                 /* csum items have already been cloned */
1591                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1592                         goto mapit;
1593                 /* we're doing a write, do the async checksumming */
1594                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1595                                    inode, rw, bio, mirror_num,
1596                                    bio_flags, bio_offset,
1597                                    __btrfs_submit_bio_start,
1598                                    __btrfs_submit_bio_done);
1599         }
1600
1601 mapit:
1602         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1603 }
1604
1605 /*
1606  * given a list of ordered sums record them in the inode.  This happens
1607  * at IO completion time based on sums calculated at bio submission time.
1608  */
1609 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1610                              struct inode *inode, u64 file_offset,
1611                              struct list_head *list)
1612 {
1613         struct btrfs_ordered_sum *sum;
1614
1615         list_for_each_entry(sum, list, list) {
1616                 btrfs_csum_file_blocks(trans,
1617                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1618         }
1619         return 0;
1620 }
1621
1622 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1623                               struct extent_state **cached_state)
1624 {
1625         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1626                 WARN_ON(1);
1627         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1628                                    cached_state, GFP_NOFS);
1629 }
1630
1631 /* see btrfs_writepage_start_hook for details on why this is required */
1632 struct btrfs_writepage_fixup {
1633         struct page *page;
1634         struct btrfs_work work;
1635 };
1636
1637 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1638 {
1639         struct btrfs_writepage_fixup *fixup;
1640         struct btrfs_ordered_extent *ordered;
1641         struct extent_state *cached_state = NULL;
1642         struct page *page;
1643         struct inode *inode;
1644         u64 page_start;
1645         u64 page_end;
1646         int ret;
1647
1648         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1649         page = fixup->page;
1650 again:
1651         lock_page(page);
1652         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1653                 ClearPageChecked(page);
1654                 goto out_page;
1655         }
1656
1657         inode = page->mapping->host;
1658         page_start = page_offset(page);
1659         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1660
1661         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1662                          &cached_state);
1663
1664         /* already ordered? We're done */
1665         if (PagePrivate2(page))
1666                 goto out;
1667
1668         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1669         if (ordered) {
1670                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1671                                      page_end, &cached_state, GFP_NOFS);
1672                 unlock_page(page);
1673                 btrfs_start_ordered_extent(inode, ordered, 1);
1674                 btrfs_put_ordered_extent(ordered);
1675                 goto again;
1676         }
1677
1678         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1679         if (ret) {
1680                 mapping_set_error(page->mapping, ret);
1681                 end_extent_writepage(page, ret, page_start, page_end);
1682                 ClearPageChecked(page);
1683                 goto out;
1684          }
1685
1686         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1687         ClearPageChecked(page);
1688         set_page_dirty(page);
1689 out:
1690         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1691                              &cached_state, GFP_NOFS);
1692 out_page:
1693         unlock_page(page);
1694         page_cache_release(page);
1695         kfree(fixup);
1696 }
1697
1698 /*
1699  * There are a few paths in the higher layers of the kernel that directly
1700  * set the page dirty bit without asking the filesystem if it is a
1701  * good idea.  This causes problems because we want to make sure COW
1702  * properly happens and the data=ordered rules are followed.
1703  *
1704  * In our case any range that doesn't have the ORDERED bit set
1705  * hasn't been properly setup for IO.  We kick off an async process
1706  * to fix it up.  The async helper will wait for ordered extents, set
1707  * the delalloc bit and make it safe to write the page.
1708  */
1709 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1710 {
1711         struct inode *inode = page->mapping->host;
1712         struct btrfs_writepage_fixup *fixup;
1713         struct btrfs_root *root = BTRFS_I(inode)->root;
1714
1715         /* this page is properly in the ordered list */
1716         if (TestClearPagePrivate2(page))
1717                 return 0;
1718
1719         if (PageChecked(page))
1720                 return -EAGAIN;
1721
1722         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1723         if (!fixup)
1724                 return -EAGAIN;
1725
1726         SetPageChecked(page);
1727         page_cache_get(page);
1728         fixup->work.func = btrfs_writepage_fixup_worker;
1729         fixup->page = page;
1730         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1731         return -EBUSY;
1732 }
1733
1734 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1735                                        struct inode *inode, u64 file_pos,
1736                                        u64 disk_bytenr, u64 disk_num_bytes,
1737                                        u64 num_bytes, u64 ram_bytes,
1738                                        u8 compression, u8 encryption,
1739                                        u16 other_encoding, int extent_type)
1740 {
1741         struct btrfs_root *root = BTRFS_I(inode)->root;
1742         struct btrfs_file_extent_item *fi;
1743         struct btrfs_path *path;
1744         struct extent_buffer *leaf;
1745         struct btrfs_key ins;
1746         u64 hint;
1747         int ret;
1748
1749         path = btrfs_alloc_path();
1750         if (!path)
1751                 return -ENOMEM;
1752
1753         path->leave_spinning = 1;
1754
1755         /*
1756          * we may be replacing one extent in the tree with another.
1757          * The new extent is pinned in the extent map, and we don't want
1758          * to drop it from the cache until it is completely in the btree.
1759          *
1760          * So, tell btrfs_drop_extents to leave this extent in the cache.
1761          * the caller is expected to unpin it and allow it to be merged
1762          * with the others.
1763          */
1764         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1765                                  &hint, 0);
1766         if (ret)
1767                 goto out;
1768
1769         ins.objectid = btrfs_ino(inode);
1770         ins.offset = file_pos;
1771         ins.type = BTRFS_EXTENT_DATA_KEY;
1772         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1773         if (ret)
1774                 goto out;
1775         leaf = path->nodes[0];
1776         fi = btrfs_item_ptr(leaf, path->slots[0],
1777                             struct btrfs_file_extent_item);
1778         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1779         btrfs_set_file_extent_type(leaf, fi, extent_type);
1780         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1781         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1782         btrfs_set_file_extent_offset(leaf, fi, 0);
1783         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1784         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1785         btrfs_set_file_extent_compression(leaf, fi, compression);
1786         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1787         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1788
1789         btrfs_unlock_up_safe(path, 1);
1790         btrfs_set_lock_blocking(leaf);
1791
1792         btrfs_mark_buffer_dirty(leaf);
1793
1794         inode_add_bytes(inode, num_bytes);
1795
1796         ins.objectid = disk_bytenr;
1797         ins.offset = disk_num_bytes;
1798         ins.type = BTRFS_EXTENT_ITEM_KEY;
1799         ret = btrfs_alloc_reserved_file_extent(trans, root,
1800                                         root->root_key.objectid,
1801                                         btrfs_ino(inode), file_pos, &ins);
1802 out:
1803         btrfs_free_path(path);
1804
1805         return ret;
1806 }
1807
1808 /*
1809  * helper function for btrfs_finish_ordered_io, this
1810  * just reads in some of the csum leaves to prime them into ram
1811  * before we start the transaction.  It limits the amount of btree
1812  * reads required while inside the transaction.
1813  */
1814 /* as ordered data IO finishes, this gets called so we can finish
1815  * an ordered extent if the range of bytes in the file it covers are
1816  * fully written.
1817  */
1818 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1819 {
1820         struct btrfs_root *root = BTRFS_I(inode)->root;
1821         struct btrfs_trans_handle *trans = NULL;
1822         struct btrfs_ordered_extent *ordered_extent = NULL;
1823         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1824         struct extent_state *cached_state = NULL;
1825         int compress_type = 0;
1826         int ret;
1827         bool nolock;
1828
1829         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1830                                              end - start + 1);
1831         if (!ret)
1832                 return 0;
1833         BUG_ON(!ordered_extent); /* Logic error */
1834
1835         nolock = btrfs_is_free_space_inode(root, inode);
1836
1837         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1838                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1839                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1840                 if (!ret) {
1841                         if (nolock)
1842                                 trans = btrfs_join_transaction_nolock(root);
1843                         else
1844                                 trans = btrfs_join_transaction(root);
1845                         if (IS_ERR(trans))
1846                                 return PTR_ERR(trans);
1847                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1848                         ret = btrfs_update_inode_fallback(trans, root, inode);
1849                         if (ret) /* -ENOMEM or corruption */
1850                                 btrfs_abort_transaction(trans, root, ret);
1851                 }
1852                 goto out;
1853         }
1854
1855         lock_extent_bits(io_tree, ordered_extent->file_offset,
1856                          ordered_extent->file_offset + ordered_extent->len - 1,
1857                          0, &cached_state);
1858
1859         if (nolock)
1860                 trans = btrfs_join_transaction_nolock(root);
1861         else
1862                 trans = btrfs_join_transaction(root);
1863         if (IS_ERR(trans)) {
1864                 ret = PTR_ERR(trans);
1865                 trans = NULL;
1866                 goto out_unlock;
1867         }
1868         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1869
1870         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1871                 compress_type = ordered_extent->compress_type;
1872         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1873                 BUG_ON(compress_type);
1874                 ret = btrfs_mark_extent_written(trans, inode,
1875                                                 ordered_extent->file_offset,
1876                                                 ordered_extent->file_offset +
1877                                                 ordered_extent->len);
1878         } else {
1879                 BUG_ON(root == root->fs_info->tree_root);
1880                 ret = insert_reserved_file_extent(trans, inode,
1881                                                 ordered_extent->file_offset,
1882                                                 ordered_extent->start,
1883                                                 ordered_extent->disk_len,
1884                                                 ordered_extent->len,
1885                                                 ordered_extent->len,
1886                                                 compress_type, 0, 0,
1887                                                 BTRFS_FILE_EXTENT_REG);
1888                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1889                                    ordered_extent->file_offset,
1890                                    ordered_extent->len);
1891         }
1892         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1893                              ordered_extent->file_offset +
1894                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1895         if (ret < 0) {
1896                 btrfs_abort_transaction(trans, root, ret);
1897                 goto out;
1898         }
1899
1900         add_pending_csums(trans, inode, ordered_extent->file_offset,
1901                           &ordered_extent->list);
1902
1903         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1904         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1905                 ret = btrfs_update_inode_fallback(trans, root, inode);
1906                 if (ret) { /* -ENOMEM or corruption */
1907                         btrfs_abort_transaction(trans, root, ret);
1908                         goto out;
1909                 }
1910         }
1911         ret = 0;
1912 out:
1913         if (root != root->fs_info->tree_root)
1914                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1915         if (trans) {
1916                 if (nolock)
1917                         btrfs_end_transaction_nolock(trans, root);
1918                 else
1919                         btrfs_end_transaction(trans, root);
1920         }
1921
1922         /* once for us */
1923         btrfs_put_ordered_extent(ordered_extent);
1924         /* once for the tree */
1925         btrfs_put_ordered_extent(ordered_extent);
1926
1927         return 0;
1928 out_unlock:
1929         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1930                              ordered_extent->file_offset +
1931                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1932         goto out;
1933 }
1934
1935 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1936                                 struct extent_state *state, int uptodate)
1937 {
1938         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1939
1940         ClearPagePrivate2(page);
1941         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1942 }
1943
1944 /*
1945  * when reads are done, we need to check csums to verify the data is correct
1946  * if there's a match, we allow the bio to finish.  If not, the code in
1947  * extent_io.c will try to find good copies for us.
1948  */
1949 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1950                                struct extent_state *state, int mirror)
1951 {
1952         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1953         struct inode *inode = page->mapping->host;
1954         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1955         char *kaddr;
1956         u64 private = ~(u32)0;
1957         int ret;
1958         struct btrfs_root *root = BTRFS_I(inode)->root;
1959         u32 csum = ~(u32)0;
1960
1961         if (PageChecked(page)) {
1962                 ClearPageChecked(page);
1963                 goto good;
1964         }
1965
1966         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1967                 goto good;
1968
1969         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1970             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1971                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1972                                   GFP_NOFS);
1973                 return 0;
1974         }
1975
1976         if (state && state->start == start) {
1977                 private = state->private;
1978                 ret = 0;
1979         } else {
1980                 ret = get_state_private(io_tree, start, &private);
1981         }
1982         kaddr = kmap_atomic(page);
1983         if (ret)
1984                 goto zeroit;
1985
1986         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
1987         btrfs_csum_final(csum, (char *)&csum);
1988         if (csum != private)
1989                 goto zeroit;
1990
1991         kunmap_atomic(kaddr);
1992 good:
1993         return 0;
1994
1995 zeroit:
1996         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
1997                        "private %llu\n",
1998                        (unsigned long long)btrfs_ino(page->mapping->host),
1999                        (unsigned long long)start, csum,
2000                        (unsigned long long)private);
2001         memset(kaddr + offset, 1, end - start + 1);
2002         flush_dcache_page(page);
2003         kunmap_atomic(kaddr);
2004         if (private == 0)
2005                 return 0;
2006         return -EIO;
2007 }
2008
2009 struct delayed_iput {
2010         struct list_head list;
2011         struct inode *inode;
2012 };
2013
2014 /* JDM: If this is fs-wide, why can't we add a pointer to
2015  * btrfs_inode instead and avoid the allocation? */
2016 void btrfs_add_delayed_iput(struct inode *inode)
2017 {
2018         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2019         struct delayed_iput *delayed;
2020
2021         if (atomic_add_unless(&inode->i_count, -1, 1))
2022                 return;
2023
2024         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2025         delayed->inode = inode;
2026
2027         spin_lock(&fs_info->delayed_iput_lock);
2028         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2029         spin_unlock(&fs_info->delayed_iput_lock);
2030 }
2031
2032 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2033 {
2034         LIST_HEAD(list);
2035         struct btrfs_fs_info *fs_info = root->fs_info;
2036         struct delayed_iput *delayed;
2037         int empty;
2038
2039         spin_lock(&fs_info->delayed_iput_lock);
2040         empty = list_empty(&fs_info->delayed_iputs);
2041         spin_unlock(&fs_info->delayed_iput_lock);
2042         if (empty)
2043                 return;
2044
2045         down_read(&root->fs_info->cleanup_work_sem);
2046         spin_lock(&fs_info->delayed_iput_lock);
2047         list_splice_init(&fs_info->delayed_iputs, &list);
2048         spin_unlock(&fs_info->delayed_iput_lock);
2049
2050         while (!list_empty(&list)) {
2051                 delayed = list_entry(list.next, struct delayed_iput, list);
2052                 list_del(&delayed->list);
2053                 iput(delayed->inode);
2054                 kfree(delayed);
2055         }
2056         up_read(&root->fs_info->cleanup_work_sem);
2057 }
2058
2059 enum btrfs_orphan_cleanup_state {
2060         ORPHAN_CLEANUP_STARTED  = 1,
2061         ORPHAN_CLEANUP_DONE     = 2,
2062 };
2063
2064 /*
2065  * This is called in transaction commit time. If there are no orphan
2066  * files in the subvolume, it removes orphan item and frees block_rsv
2067  * structure.
2068  */
2069 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2070                               struct btrfs_root *root)
2071 {
2072         struct btrfs_block_rsv *block_rsv;
2073         int ret;
2074
2075         if (!list_empty(&root->orphan_list) ||
2076             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2077                 return;
2078
2079         spin_lock(&root->orphan_lock);
2080         if (!list_empty(&root->orphan_list)) {
2081                 spin_unlock(&root->orphan_lock);
2082                 return;
2083         }
2084
2085         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2086                 spin_unlock(&root->orphan_lock);
2087                 return;
2088         }
2089
2090         block_rsv = root->orphan_block_rsv;
2091         root->orphan_block_rsv = NULL;
2092         spin_unlock(&root->orphan_lock);
2093
2094         if (root->orphan_item_inserted &&
2095             btrfs_root_refs(&root->root_item) > 0) {
2096                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2097                                             root->root_key.objectid);
2098                 BUG_ON(ret);
2099                 root->orphan_item_inserted = 0;
2100         }
2101
2102         if (block_rsv) {
2103                 WARN_ON(block_rsv->size > 0);
2104                 btrfs_free_block_rsv(root, block_rsv);
2105         }
2106 }
2107
2108 /*
2109  * This creates an orphan entry for the given inode in case something goes
2110  * wrong in the middle of an unlink/truncate.
2111  *
2112  * NOTE: caller of this function should reserve 5 units of metadata for
2113  *       this function.
2114  */
2115 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2116 {
2117         struct btrfs_root *root = BTRFS_I(inode)->root;
2118         struct btrfs_block_rsv *block_rsv = NULL;
2119         int reserve = 0;
2120         int insert = 0;
2121         int ret;
2122
2123         if (!root->orphan_block_rsv) {
2124                 block_rsv = btrfs_alloc_block_rsv(root);
2125                 if (!block_rsv)
2126                         return -ENOMEM;
2127         }
2128
2129         spin_lock(&root->orphan_lock);
2130         if (!root->orphan_block_rsv) {
2131                 root->orphan_block_rsv = block_rsv;
2132         } else if (block_rsv) {
2133                 btrfs_free_block_rsv(root, block_rsv);
2134                 block_rsv = NULL;
2135         }
2136
2137         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2138                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2139 #if 0
2140                 /*
2141                  * For proper ENOSPC handling, we should do orphan
2142                  * cleanup when mounting. But this introduces backward
2143                  * compatibility issue.
2144                  */
2145                 if (!xchg(&root->orphan_item_inserted, 1))
2146                         insert = 2;
2147                 else
2148                         insert = 1;
2149 #endif
2150                 insert = 1;
2151         }
2152
2153         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2154                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2155                 reserve = 1;
2156         }
2157         spin_unlock(&root->orphan_lock);
2158
2159         /* grab metadata reservation from transaction handle */
2160         if (reserve) {
2161                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2162                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2163         }
2164
2165         /* insert an orphan item to track this unlinked/truncated file */
2166         if (insert >= 1) {
2167                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2168                 if (ret && ret != -EEXIST) {
2169                         btrfs_abort_transaction(trans, root, ret);
2170                         return ret;
2171                 }
2172                 ret = 0;
2173         }
2174
2175         /* insert an orphan item to track subvolume contains orphan files */
2176         if (insert >= 2) {
2177                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2178                                                root->root_key.objectid);
2179                 if (ret && ret != -EEXIST) {
2180                         btrfs_abort_transaction(trans, root, ret);
2181                         return ret;
2182                 }
2183         }
2184         return 0;
2185 }
2186
2187 /*
2188  * We have done the truncate/delete so we can go ahead and remove the orphan
2189  * item for this particular inode.
2190  */
2191 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2192 {
2193         struct btrfs_root *root = BTRFS_I(inode)->root;
2194         int delete_item = 0;
2195         int release_rsv = 0;
2196         int ret = 0;
2197
2198         spin_lock(&root->orphan_lock);
2199         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2200                 list_del_init(&BTRFS_I(inode)->i_orphan);
2201                 delete_item = 1;
2202         }
2203
2204         if (BTRFS_I(inode)->orphan_meta_reserved) {
2205                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2206                 release_rsv = 1;
2207         }
2208         spin_unlock(&root->orphan_lock);
2209
2210         if (trans && delete_item) {
2211                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2212                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2213         }
2214
2215         if (release_rsv)
2216                 btrfs_orphan_release_metadata(inode);
2217
2218         return 0;
2219 }
2220
2221 /*
2222  * this cleans up any orphans that may be left on the list from the last use
2223  * of this root.
2224  */
2225 int btrfs_orphan_cleanup(struct btrfs_root *root)
2226 {
2227         struct btrfs_path *path;
2228         struct extent_buffer *leaf;
2229         struct btrfs_key key, found_key;
2230         struct btrfs_trans_handle *trans;
2231         struct inode *inode;
2232         u64 last_objectid = 0;
2233         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2234
2235         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2236                 return 0;
2237
2238         path = btrfs_alloc_path();
2239         if (!path) {
2240                 ret = -ENOMEM;
2241                 goto out;
2242         }
2243         path->reada = -1;
2244
2245         key.objectid = BTRFS_ORPHAN_OBJECTID;
2246         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2247         key.offset = (u64)-1;
2248
2249         while (1) {
2250                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2251                 if (ret < 0)
2252                         goto out;
2253
2254                 /*
2255                  * if ret == 0 means we found what we were searching for, which
2256                  * is weird, but possible, so only screw with path if we didn't
2257                  * find the key and see if we have stuff that matches
2258                  */
2259                 if (ret > 0) {
2260                         ret = 0;
2261                         if (path->slots[0] == 0)
2262                                 break;
2263                         path->slots[0]--;
2264                 }
2265
2266                 /* pull out the item */
2267                 leaf = path->nodes[0];
2268                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2269
2270                 /* make sure the item matches what we want */
2271                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2272                         break;
2273                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2274                         break;
2275
2276                 /* release the path since we're done with it */
2277                 btrfs_release_path(path);
2278
2279                 /*
2280                  * this is where we are basically btrfs_lookup, without the
2281                  * crossing root thing.  we store the inode number in the
2282                  * offset of the orphan item.
2283                  */
2284
2285                 if (found_key.offset == last_objectid) {
2286                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2287                                "stopping orphan cleanup\n");
2288                         ret = -EINVAL;
2289                         goto out;
2290                 }
2291
2292                 last_objectid = found_key.offset;
2293
2294                 found_key.objectid = found_key.offset;
2295                 found_key.type = BTRFS_INODE_ITEM_KEY;
2296                 found_key.offset = 0;
2297                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2298                 ret = PTR_RET(inode);
2299                 if (ret && ret != -ESTALE)
2300                         goto out;
2301
2302                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2303                         struct btrfs_root *dead_root;
2304                         struct btrfs_fs_info *fs_info = root->fs_info;
2305                         int is_dead_root = 0;
2306
2307                         /*
2308                          * this is an orphan in the tree root. Currently these
2309                          * could come from 2 sources:
2310                          *  a) a snapshot deletion in progress
2311                          *  b) a free space cache inode
2312                          * We need to distinguish those two, as the snapshot
2313                          * orphan must not get deleted.
2314                          * find_dead_roots already ran before us, so if this
2315                          * is a snapshot deletion, we should find the root
2316                          * in the dead_roots list
2317                          */
2318                         spin_lock(&fs_info->trans_lock);
2319                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2320                                             root_list) {
2321                                 if (dead_root->root_key.objectid ==
2322                                     found_key.objectid) {
2323                                         is_dead_root = 1;
2324                                         break;
2325                                 }
2326                         }
2327                         spin_unlock(&fs_info->trans_lock);
2328                         if (is_dead_root) {
2329                                 /* prevent this orphan from being found again */
2330                                 key.offset = found_key.objectid - 1;
2331                                 continue;
2332                         }
2333                 }
2334                 /*
2335                  * Inode is already gone but the orphan item is still there,
2336                  * kill the orphan item.
2337                  */
2338                 if (ret == -ESTALE) {
2339                         trans = btrfs_start_transaction(root, 1);
2340                         if (IS_ERR(trans)) {
2341                                 ret = PTR_ERR(trans);
2342                                 goto out;
2343                         }
2344                         ret = btrfs_del_orphan_item(trans, root,
2345                                                     found_key.objectid);
2346                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2347                         btrfs_end_transaction(trans, root);
2348                         continue;
2349                 }
2350
2351                 /*
2352                  * add this inode to the orphan list so btrfs_orphan_del does
2353                  * the proper thing when we hit it
2354                  */
2355                 spin_lock(&root->orphan_lock);
2356                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2357                 spin_unlock(&root->orphan_lock);
2358
2359                 /* if we have links, this was a truncate, lets do that */
2360                 if (inode->i_nlink) {
2361                         if (!S_ISREG(inode->i_mode)) {
2362                                 WARN_ON(1);
2363                                 iput(inode);
2364                                 continue;
2365                         }
2366                         nr_truncate++;
2367                         ret = btrfs_truncate(inode);
2368                 } else {
2369                         nr_unlink++;
2370                 }
2371
2372                 /* this will do delete_inode and everything for us */
2373                 iput(inode);
2374                 if (ret)
2375                         goto out;
2376         }
2377         /* release the path since we're done with it */
2378         btrfs_release_path(path);
2379
2380         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2381
2382         if (root->orphan_block_rsv)
2383                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2384                                         (u64)-1);
2385
2386         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2387                 trans = btrfs_join_transaction(root);
2388                 if (!IS_ERR(trans))
2389                         btrfs_end_transaction(trans, root);
2390         }
2391
2392         if (nr_unlink)
2393                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2394         if (nr_truncate)
2395                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2396
2397 out:
2398         if (ret)
2399                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2400         btrfs_free_path(path);
2401         return ret;
2402 }
2403
2404 /*
2405  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2406  * don't find any xattrs, we know there can't be any acls.
2407  *
2408  * slot is the slot the inode is in, objectid is the objectid of the inode
2409  */
2410 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2411                                           int slot, u64 objectid)
2412 {
2413         u32 nritems = btrfs_header_nritems(leaf);
2414         struct btrfs_key found_key;
2415         int scanned = 0;
2416
2417         slot++;
2418         while (slot < nritems) {
2419                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2420
2421                 /* we found a different objectid, there must not be acls */
2422                 if (found_key.objectid != objectid)
2423                         return 0;
2424
2425                 /* we found an xattr, assume we've got an acl */
2426                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2427                         return 1;
2428
2429                 /*
2430                  * we found a key greater than an xattr key, there can't
2431                  * be any acls later on
2432                  */
2433                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2434                         return 0;
2435
2436                 slot++;
2437                 scanned++;
2438
2439                 /*
2440                  * it goes inode, inode backrefs, xattrs, extents,
2441                  * so if there are a ton of hard links to an inode there can
2442                  * be a lot of backrefs.  Don't waste time searching too hard,
2443                  * this is just an optimization
2444                  */
2445                 if (scanned >= 8)
2446                         break;
2447         }
2448         /* we hit the end of the leaf before we found an xattr or
2449          * something larger than an xattr.  We have to assume the inode
2450          * has acls
2451          */
2452         return 1;
2453 }
2454
2455 /*
2456  * read an inode from the btree into the in-memory inode
2457  */
2458 static void btrfs_read_locked_inode(struct inode *inode)
2459 {
2460         struct btrfs_path *path;
2461         struct extent_buffer *leaf;
2462         struct btrfs_inode_item *inode_item;
2463         struct btrfs_timespec *tspec;
2464         struct btrfs_root *root = BTRFS_I(inode)->root;
2465         struct btrfs_key location;
2466         int maybe_acls;
2467         u32 rdev;
2468         int ret;
2469         bool filled = false;
2470
2471         ret = btrfs_fill_inode(inode, &rdev);
2472         if (!ret)
2473                 filled = true;
2474
2475         path = btrfs_alloc_path();
2476         if (!path)
2477                 goto make_bad;
2478
2479         path->leave_spinning = 1;
2480         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2481
2482         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2483         if (ret)
2484                 goto make_bad;
2485
2486         leaf = path->nodes[0];
2487
2488         if (filled)
2489                 goto cache_acl;
2490
2491         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2492                                     struct btrfs_inode_item);
2493         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2494         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2495         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2496         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2497         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2498
2499         tspec = btrfs_inode_atime(inode_item);
2500         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2501         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2502
2503         tspec = btrfs_inode_mtime(inode_item);
2504         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2505         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2506
2507         tspec = btrfs_inode_ctime(inode_item);
2508         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2509         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2510
2511         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2512         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2513         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2514         inode->i_generation = BTRFS_I(inode)->generation;
2515         inode->i_rdev = 0;
2516         rdev = btrfs_inode_rdev(leaf, inode_item);
2517
2518         BTRFS_I(inode)->index_cnt = (u64)-1;
2519         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2520 cache_acl:
2521         /*
2522          * try to precache a NULL acl entry for files that don't have
2523          * any xattrs or acls
2524          */
2525         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2526                                            btrfs_ino(inode));
2527         if (!maybe_acls)
2528                 cache_no_acl(inode);
2529
2530         btrfs_free_path(path);
2531
2532         switch (inode->i_mode & S_IFMT) {
2533         case S_IFREG:
2534                 inode->i_mapping->a_ops = &btrfs_aops;
2535                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2536                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2537                 inode->i_fop = &btrfs_file_operations;
2538                 inode->i_op = &btrfs_file_inode_operations;
2539                 break;
2540         case S_IFDIR:
2541                 inode->i_fop = &btrfs_dir_file_operations;
2542                 if (root == root->fs_info->tree_root)
2543                         inode->i_op = &btrfs_dir_ro_inode_operations;
2544                 else
2545                         inode->i_op = &btrfs_dir_inode_operations;
2546                 break;
2547         case S_IFLNK:
2548                 inode->i_op = &btrfs_symlink_inode_operations;
2549                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2550                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2551                 break;
2552         default:
2553                 inode->i_op = &btrfs_special_inode_operations;
2554                 init_special_inode(inode, inode->i_mode, rdev);
2555                 break;
2556         }
2557
2558         btrfs_update_iflags(inode);
2559         return;
2560
2561 make_bad:
2562         btrfs_free_path(path);
2563         make_bad_inode(inode);
2564 }
2565
2566 /*
2567  * given a leaf and an inode, copy the inode fields into the leaf
2568  */
2569 static void fill_inode_item(struct btrfs_trans_handle *trans,
2570                             struct extent_buffer *leaf,
2571                             struct btrfs_inode_item *item,
2572                             struct inode *inode)
2573 {
2574         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2575         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2576         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2577         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2578         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2579
2580         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2581                                inode->i_atime.tv_sec);
2582         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2583                                 inode->i_atime.tv_nsec);
2584
2585         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2586                                inode->i_mtime.tv_sec);
2587         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2588                                 inode->i_mtime.tv_nsec);
2589
2590         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2591                                inode->i_ctime.tv_sec);
2592         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2593                                 inode->i_ctime.tv_nsec);
2594
2595         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2596         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2597         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2598         btrfs_set_inode_transid(leaf, item, trans->transid);
2599         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2600         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2601         btrfs_set_inode_block_group(leaf, item, 0);
2602 }
2603
2604 /*
2605  * copy everything in the in-memory inode into the btree.
2606  */
2607 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2608                                 struct btrfs_root *root, struct inode *inode)
2609 {
2610         struct btrfs_inode_item *inode_item;
2611         struct btrfs_path *path;
2612         struct extent_buffer *leaf;
2613         int ret;
2614
2615         path = btrfs_alloc_path();
2616         if (!path)
2617                 return -ENOMEM;
2618
2619         path->leave_spinning = 1;
2620         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2621                                  1);
2622         if (ret) {
2623                 if (ret > 0)
2624                         ret = -ENOENT;
2625                 goto failed;
2626         }
2627
2628         btrfs_unlock_up_safe(path, 1);
2629         leaf = path->nodes[0];
2630         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2631                                     struct btrfs_inode_item);
2632
2633         fill_inode_item(trans, leaf, inode_item, inode);
2634         btrfs_mark_buffer_dirty(leaf);
2635         btrfs_set_inode_last_trans(trans, inode);
2636         ret = 0;
2637 failed:
2638         btrfs_free_path(path);
2639         return ret;
2640 }
2641
2642 /*
2643  * copy everything in the in-memory inode into the btree.
2644  */
2645 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2646                                 struct btrfs_root *root, struct inode *inode)
2647 {
2648         int ret;
2649
2650         /*
2651          * If the inode is a free space inode, we can deadlock during commit
2652          * if we put it into the delayed code.
2653          *
2654          * The data relocation inode should also be directly updated
2655          * without delay
2656          */
2657         if (!btrfs_is_free_space_inode(root, inode)
2658             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2659                 ret = btrfs_delayed_update_inode(trans, root, inode);
2660                 if (!ret)
2661                         btrfs_set_inode_last_trans(trans, inode);
2662                 return ret;
2663         }
2664
2665         return btrfs_update_inode_item(trans, root, inode);
2666 }
2667
2668 static noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2669                                 struct btrfs_root *root, struct inode *inode)
2670 {
2671         int ret;
2672
2673         ret = btrfs_update_inode(trans, root, inode);
2674         if (ret == -ENOSPC)
2675                 return btrfs_update_inode_item(trans, root, inode);
2676         return ret;
2677 }
2678
2679 /*
2680  * unlink helper that gets used here in inode.c and in the tree logging
2681  * recovery code.  It remove a link in a directory with a given name, and
2682  * also drops the back refs in the inode to the directory
2683  */
2684 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2685                                 struct btrfs_root *root,
2686                                 struct inode *dir, struct inode *inode,
2687                                 const char *name, int name_len)
2688 {
2689         struct btrfs_path *path;
2690         int ret = 0;
2691         struct extent_buffer *leaf;
2692         struct btrfs_dir_item *di;
2693         struct btrfs_key key;
2694         u64 index;
2695         u64 ino = btrfs_ino(inode);
2696         u64 dir_ino = btrfs_ino(dir);
2697
2698         path = btrfs_alloc_path();
2699         if (!path) {
2700                 ret = -ENOMEM;
2701                 goto out;
2702         }
2703
2704         path->leave_spinning = 1;
2705         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2706                                     name, name_len, -1);
2707         if (IS_ERR(di)) {
2708                 ret = PTR_ERR(di);
2709                 goto err;
2710         }
2711         if (!di) {
2712                 ret = -ENOENT;
2713                 goto err;
2714         }
2715         leaf = path->nodes[0];
2716         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2717         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2718         if (ret)
2719                 goto err;
2720         btrfs_release_path(path);
2721
2722         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2723                                   dir_ino, &index);
2724         if (ret) {
2725                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2726                        "inode %llu parent %llu\n", name_len, name,
2727                        (unsigned long long)ino, (unsigned long long)dir_ino);
2728                 btrfs_abort_transaction(trans, root, ret);
2729                 goto err;
2730         }
2731
2732         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2733         if (ret) {
2734                 btrfs_abort_transaction(trans, root, ret);
2735                 goto err;
2736         }
2737
2738         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2739                                          inode, dir_ino);
2740         if (ret != 0 && ret != -ENOENT) {
2741                 btrfs_abort_transaction(trans, root, ret);
2742                 goto err;
2743         }
2744
2745         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2746                                            dir, index);
2747         if (ret == -ENOENT)
2748                 ret = 0;
2749 err:
2750         btrfs_free_path(path);
2751         if (ret)
2752                 goto out;
2753
2754         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2755         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2756         btrfs_update_inode(trans, root, dir);
2757 out:
2758         return ret;
2759 }
2760
2761 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2762                        struct btrfs_root *root,
2763                        struct inode *dir, struct inode *inode,
2764                        const char *name, int name_len)
2765 {
2766         int ret;
2767         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2768         if (!ret) {
2769                 btrfs_drop_nlink(inode);
2770                 ret = btrfs_update_inode(trans, root, inode);
2771         }
2772         return ret;
2773 }
2774
2775
2776 /* helper to check if there is any shared block in the path */
2777 static int check_path_shared(struct btrfs_root *root,
2778                              struct btrfs_path *path)
2779 {
2780         struct extent_buffer *eb;
2781         int level;
2782         u64 refs = 1;
2783
2784         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2785                 int ret;
2786
2787                 if (!path->nodes[level])
2788                         break;
2789                 eb = path->nodes[level];
2790                 if (!btrfs_block_can_be_shared(root, eb))
2791                         continue;
2792                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2793                                                &refs, NULL);
2794                 if (refs > 1)
2795                         return 1;
2796         }
2797         return 0;
2798 }
2799
2800 /*
2801  * helper to start transaction for unlink and rmdir.
2802  *
2803  * unlink and rmdir are special in btrfs, they do not always free space.
2804  * so in enospc case, we should make sure they will free space before
2805  * allowing them to use the global metadata reservation.
2806  */
2807 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2808                                                        struct dentry *dentry)
2809 {
2810         struct btrfs_trans_handle *trans;
2811         struct btrfs_root *root = BTRFS_I(dir)->root;
2812         struct btrfs_path *path;
2813         struct btrfs_inode_ref *ref;
2814         struct btrfs_dir_item *di;
2815         struct inode *inode = dentry->d_inode;
2816         u64 index;
2817         int check_link = 1;
2818         int err = -ENOSPC;
2819         int ret;
2820         u64 ino = btrfs_ino(inode);
2821         u64 dir_ino = btrfs_ino(dir);
2822
2823         /*
2824          * 1 for the possible orphan item
2825          * 1 for the dir item
2826          * 1 for the dir index
2827          * 1 for the inode ref
2828          * 1 for the inode ref in the tree log
2829          * 2 for the dir entries in the log
2830          * 1 for the inode
2831          */
2832         trans = btrfs_start_transaction(root, 8);
2833         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2834                 return trans;
2835
2836         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2837                 return ERR_PTR(-ENOSPC);
2838
2839         /* check if there is someone else holds reference */
2840         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2841                 return ERR_PTR(-ENOSPC);
2842
2843         if (atomic_read(&inode->i_count) > 2)
2844                 return ERR_PTR(-ENOSPC);
2845
2846         if (xchg(&root->fs_info->enospc_unlink, 1))
2847                 return ERR_PTR(-ENOSPC);
2848
2849         path = btrfs_alloc_path();
2850         if (!path) {
2851                 root->fs_info->enospc_unlink = 0;
2852                 return ERR_PTR(-ENOMEM);
2853         }
2854
2855         /* 1 for the orphan item */
2856         trans = btrfs_start_transaction(root, 1);
2857         if (IS_ERR(trans)) {
2858                 btrfs_free_path(path);
2859                 root->fs_info->enospc_unlink = 0;
2860                 return trans;
2861         }
2862
2863         path->skip_locking = 1;
2864         path->search_commit_root = 1;
2865
2866         ret = btrfs_lookup_inode(trans, root, path,
2867                                 &BTRFS_I(dir)->location, 0);
2868         if (ret < 0) {
2869                 err = ret;
2870                 goto out;
2871         }
2872         if (ret == 0) {
2873                 if (check_path_shared(root, path))
2874                         goto out;
2875         } else {
2876                 check_link = 0;
2877         }
2878         btrfs_release_path(path);
2879
2880         ret = btrfs_lookup_inode(trans, root, path,
2881                                 &BTRFS_I(inode)->location, 0);
2882         if (ret < 0) {
2883                 err = ret;
2884                 goto out;
2885         }
2886         if (ret == 0) {
2887                 if (check_path_shared(root, path))
2888                         goto out;
2889         } else {
2890                 check_link = 0;
2891         }
2892         btrfs_release_path(path);
2893
2894         if (ret == 0 && S_ISREG(inode->i_mode)) {
2895                 ret = btrfs_lookup_file_extent(trans, root, path,
2896                                                ino, (u64)-1, 0);
2897                 if (ret < 0) {
2898                         err = ret;
2899                         goto out;
2900                 }
2901                 BUG_ON(ret == 0); /* Corruption */
2902                 if (check_path_shared(root, path))
2903                         goto out;
2904                 btrfs_release_path(path);
2905         }
2906
2907         if (!check_link) {
2908                 err = 0;
2909                 goto out;
2910         }
2911
2912         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2913                                 dentry->d_name.name, dentry->d_name.len, 0);
2914         if (IS_ERR(di)) {
2915                 err = PTR_ERR(di);
2916                 goto out;
2917         }
2918         if (di) {
2919                 if (check_path_shared(root, path))
2920                         goto out;
2921         } else {
2922                 err = 0;
2923                 goto out;
2924         }
2925         btrfs_release_path(path);
2926
2927         ref = btrfs_lookup_inode_ref(trans, root, path,
2928                                 dentry->d_name.name, dentry->d_name.len,
2929                                 ino, dir_ino, 0);
2930         if (IS_ERR(ref)) {
2931                 err = PTR_ERR(ref);
2932                 goto out;
2933         }
2934         BUG_ON(!ref); /* Logic error */
2935         if (check_path_shared(root, path))
2936                 goto out;
2937         index = btrfs_inode_ref_index(path->nodes[0], ref);
2938         btrfs_release_path(path);
2939
2940         /*
2941          * This is a commit root search, if we can lookup inode item and other
2942          * relative items in the commit root, it means the transaction of
2943          * dir/file creation has been committed, and the dir index item that we
2944          * delay to insert has also been inserted into the commit root. So
2945          * we needn't worry about the delayed insertion of the dir index item
2946          * here.
2947          */
2948         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2949                                 dentry->d_name.name, dentry->d_name.len, 0);
2950         if (IS_ERR(di)) {
2951                 err = PTR_ERR(di);
2952                 goto out;
2953         }
2954         BUG_ON(ret == -ENOENT);
2955         if (check_path_shared(root, path))
2956                 goto out;
2957
2958         err = 0;
2959 out:
2960         btrfs_free_path(path);
2961         /* Migrate the orphan reservation over */
2962         if (!err)
2963                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2964                                 &root->fs_info->global_block_rsv,
2965                                 trans->bytes_reserved);
2966
2967         if (err) {
2968                 btrfs_end_transaction(trans, root);
2969                 root->fs_info->enospc_unlink = 0;
2970                 return ERR_PTR(err);
2971         }
2972
2973         trans->block_rsv = &root->fs_info->global_block_rsv;
2974         return trans;
2975 }
2976
2977 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2978                                struct btrfs_root *root)
2979 {
2980         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2981                 btrfs_block_rsv_release(root, trans->block_rsv,
2982                                         trans->bytes_reserved);
2983                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2984                 BUG_ON(!root->fs_info->enospc_unlink);
2985                 root->fs_info->enospc_unlink = 0;
2986         }
2987         btrfs_end_transaction(trans, root);
2988 }
2989
2990 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2991 {
2992         struct btrfs_root *root = BTRFS_I(dir)->root;
2993         struct btrfs_trans_handle *trans;
2994         struct inode *inode = dentry->d_inode;
2995         int ret;
2996         unsigned long nr = 0;
2997
2998         trans = __unlink_start_trans(dir, dentry);
2999         if (IS_ERR(trans))
3000                 return PTR_ERR(trans);
3001
3002         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3003
3004         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3005                                  dentry->d_name.name, dentry->d_name.len);
3006         if (ret)
3007                 goto out;
3008
3009         if (inode->i_nlink == 0) {
3010                 ret = btrfs_orphan_add(trans, inode);
3011                 if (ret)
3012                         goto out;
3013         }
3014
3015 out:
3016         nr = trans->blocks_used;
3017         __unlink_end_trans(trans, root);
3018         btrfs_btree_balance_dirty(root, nr);
3019         return ret;
3020 }
3021
3022 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3023                         struct btrfs_root *root,
3024                         struct inode *dir, u64 objectid,
3025                         const char *name, int name_len)
3026 {
3027         struct btrfs_path *path;
3028         struct extent_buffer *leaf;
3029         struct btrfs_dir_item *di;
3030         struct btrfs_key key;
3031         u64 index;
3032         int ret;
3033         u64 dir_ino = btrfs_ino(dir);
3034
3035         path = btrfs_alloc_path();
3036         if (!path)
3037                 return -ENOMEM;
3038
3039         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3040                                    name, name_len, -1);
3041         if (IS_ERR_OR_NULL(di)) {
3042                 if (!di)
3043                         ret = -ENOENT;
3044                 else
3045                         ret = PTR_ERR(di);
3046                 goto out;
3047         }
3048
3049         leaf = path->nodes[0];
3050         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3051         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3052         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3053         if (ret) {
3054                 btrfs_abort_transaction(trans, root, ret);
3055                 goto out;
3056         }
3057         btrfs_release_path(path);
3058
3059         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3060                                  objectid, root->root_key.objectid,
3061                                  dir_ino, &index, name, name_len);
3062         if (ret < 0) {
3063                 if (ret != -ENOENT) {
3064                         btrfs_abort_transaction(trans, root, ret);
3065                         goto out;
3066                 }
3067                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3068                                                  name, name_len);
3069                 if (IS_ERR_OR_NULL(di)) {
3070                         if (!di)
3071                                 ret = -ENOENT;
3072                         else
3073                                 ret = PTR_ERR(di);
3074                         btrfs_abort_transaction(trans, root, ret);
3075                         goto out;
3076                 }
3077
3078                 leaf = path->nodes[0];
3079                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3080                 btrfs_release_path(path);
3081                 index = key.offset;
3082         }
3083         btrfs_release_path(path);
3084
3085         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3086         if (ret) {
3087                 btrfs_abort_transaction(trans, root, ret);
3088                 goto out;
3089         }
3090
3091         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3092         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3093         ret = btrfs_update_inode(trans, root, dir);
3094         if (ret)
3095                 btrfs_abort_transaction(trans, root, ret);
3096 out:
3097         btrfs_free_path(path);
3098         return ret;
3099 }
3100
3101 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3102 {
3103         struct inode *inode = dentry->d_inode;
3104         int err = 0;
3105         struct btrfs_root *root = BTRFS_I(dir)->root;
3106         struct btrfs_trans_handle *trans;
3107         unsigned long nr = 0;
3108
3109         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3110             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3111                 return -ENOTEMPTY;
3112
3113         trans = __unlink_start_trans(dir, dentry);
3114         if (IS_ERR(trans))
3115                 return PTR_ERR(trans);
3116
3117         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3118                 err = btrfs_unlink_subvol(trans, root, dir,
3119                                           BTRFS_I(inode)->location.objectid,
3120                                           dentry->d_name.name,
3121                                           dentry->d_name.len);
3122                 goto out;
3123         }
3124
3125         err = btrfs_orphan_add(trans, inode);
3126         if (err)
3127                 goto out;
3128
3129         /* now the directory is empty */
3130         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3131                                  dentry->d_name.name, dentry->d_name.len);
3132         if (!err)
3133                 btrfs_i_size_write(inode, 0);
3134 out:
3135         nr = trans->blocks_used;
3136         __unlink_end_trans(trans, root);
3137         btrfs_btree_balance_dirty(root, nr);
3138
3139         return err;
3140 }
3141
3142 /*
3143  * this can truncate away extent items, csum items and directory items.
3144  * It starts at a high offset and removes keys until it can't find
3145  * any higher than new_size
3146  *
3147  * csum items that cross the new i_size are truncated to the new size
3148  * as well.
3149  *
3150  * min_type is the minimum key type to truncate down to.  If set to 0, this
3151  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3152  */
3153 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3154                                struct btrfs_root *root,
3155                                struct inode *inode,
3156                                u64 new_size, u32 min_type)
3157 {
3158         struct btrfs_path *path;
3159         struct extent_buffer *leaf;
3160         struct btrfs_file_extent_item *fi;
3161         struct btrfs_key key;
3162         struct btrfs_key found_key;
3163         u64 extent_start = 0;
3164         u64 extent_num_bytes = 0;
3165         u64 extent_offset = 0;
3166         u64 item_end = 0;
3167         u64 mask = root->sectorsize - 1;
3168         u32 found_type = (u8)-1;
3169         int found_extent;
3170         int del_item;
3171         int pending_del_nr = 0;
3172         int pending_del_slot = 0;
3173         int extent_type = -1;
3174         int ret;
3175         int err = 0;
3176         u64 ino = btrfs_ino(inode);
3177
3178         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3179
3180         path = btrfs_alloc_path();
3181         if (!path)
3182                 return -ENOMEM;
3183         path->reada = -1;
3184
3185         if (root->ref_cows || root == root->fs_info->tree_root)
3186                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3187
3188         /*
3189          * This function is also used to drop the items in the log tree before
3190          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3191          * it is used to drop the loged items. So we shouldn't kill the delayed
3192          * items.
3193          */
3194         if (min_type == 0 && root == BTRFS_I(inode)->root)
3195                 btrfs_kill_delayed_inode_items(inode);
3196
3197         key.objectid = ino;
3198         key.offset = (u64)-1;
3199         key.type = (u8)-1;
3200
3201 search_again:
3202         path->leave_spinning = 1;
3203         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3204         if (ret < 0) {
3205                 err = ret;
3206                 goto out;
3207         }
3208
3209         if (ret > 0) {
3210                 /* there are no items in the tree for us to truncate, we're
3211                  * done
3212                  */
3213                 if (path->slots[0] == 0)
3214                         goto out;
3215                 path->slots[0]--;
3216         }
3217
3218         while (1) {
3219                 fi = NULL;
3220                 leaf = path->nodes[0];
3221                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3222                 found_type = btrfs_key_type(&found_key);
3223
3224                 if (found_key.objectid != ino)
3225                         break;
3226
3227                 if (found_type < min_type)
3228                         break;
3229
3230                 item_end = found_key.offset;
3231                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3232                         fi = btrfs_item_ptr(leaf, path->slots[0],
3233                                             struct btrfs_file_extent_item);
3234                         extent_type = btrfs_file_extent_type(leaf, fi);
3235                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3236                                 item_end +=
3237                                     btrfs_file_extent_num_bytes(leaf, fi);
3238                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3239                                 item_end += btrfs_file_extent_inline_len(leaf,
3240                                                                          fi);
3241                         }
3242                         item_end--;
3243                 }
3244                 if (found_type > min_type) {
3245                         del_item = 1;
3246                 } else {
3247                         if (item_end < new_size)
3248                                 break;
3249                         if (found_key.offset >= new_size)
3250                                 del_item = 1;
3251                         else
3252                                 del_item = 0;
3253                 }
3254                 found_extent = 0;
3255                 /* FIXME, shrink the extent if the ref count is only 1 */
3256                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3257                         goto delete;
3258
3259                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3260                         u64 num_dec;
3261                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3262                         if (!del_item) {
3263                                 u64 orig_num_bytes =
3264                                         btrfs_file_extent_num_bytes(leaf, fi);
3265                                 extent_num_bytes = new_size -
3266                                         found_key.offset + root->sectorsize - 1;
3267                                 extent_num_bytes = extent_num_bytes &
3268                                         ~((u64)root->sectorsize - 1);
3269                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3270                                                          extent_num_bytes);
3271                                 num_dec = (orig_num_bytes -
3272                                            extent_num_bytes);
3273                                 if (root->ref_cows && extent_start != 0)
3274                                         inode_sub_bytes(inode, num_dec);
3275                                 btrfs_mark_buffer_dirty(leaf);
3276                         } else {
3277                                 extent_num_bytes =
3278                                         btrfs_file_extent_disk_num_bytes(leaf,
3279                                                                          fi);
3280                                 extent_offset = found_key.offset -
3281                                         btrfs_file_extent_offset(leaf, fi);
3282
3283                                 /* FIXME blocksize != 4096 */
3284                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3285                                 if (extent_start != 0) {
3286                                         found_extent = 1;
3287                                         if (root->ref_cows)
3288                                                 inode_sub_bytes(inode, num_dec);
3289                                 }
3290                         }
3291                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3292                         /*
3293                          * we can't truncate inline items that have had
3294                          * special encodings
3295                          */
3296                         if (!del_item &&
3297                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3298                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3299                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3300                                 u32 size = new_size - found_key.offset;
3301
3302                                 if (root->ref_cows) {
3303                                         inode_sub_bytes(inode, item_end + 1 -
3304                                                         new_size);
3305                                 }
3306                                 size =
3307                                     btrfs_file_extent_calc_inline_size(size);
3308                                 btrfs_truncate_item(trans, root, path,
3309                                                     size, 1);
3310                         } else if (root->ref_cows) {
3311                                 inode_sub_bytes(inode, item_end + 1 -
3312                                                 found_key.offset);
3313                         }
3314                 }
3315 delete:
3316                 if (del_item) {
3317                         if (!pending_del_nr) {
3318                                 /* no pending yet, add ourselves */
3319                                 pending_del_slot = path->slots[0];
3320                                 pending_del_nr = 1;
3321                         } else if (pending_del_nr &&
3322                                    path->slots[0] + 1 == pending_del_slot) {
3323                                 /* hop on the pending chunk */
3324                                 pending_del_nr++;
3325                                 pending_del_slot = path->slots[0];
3326                         } else {
3327                                 BUG();
3328                         }
3329                 } else {
3330                         break;
3331                 }
3332                 if (found_extent && (root->ref_cows ||
3333                                      root == root->fs_info->tree_root)) {
3334                         btrfs_set_path_blocking(path);
3335                         ret = btrfs_free_extent(trans, root, extent_start,
3336                                                 extent_num_bytes, 0,
3337                                                 btrfs_header_owner(leaf),
3338                                                 ino, extent_offset, 0);
3339                         BUG_ON(ret);
3340                 }
3341
3342                 if (found_type == BTRFS_INODE_ITEM_KEY)
3343                         break;
3344
3345                 if (path->slots[0] == 0 ||
3346                     path->slots[0] != pending_del_slot) {
3347                         if (root->ref_cows &&
3348                             BTRFS_I(inode)->location.objectid !=
3349                                                 BTRFS_FREE_INO_OBJECTID) {
3350                                 err = -EAGAIN;
3351                                 goto out;
3352                         }
3353                         if (pending_del_nr) {
3354                                 ret = btrfs_del_items(trans, root, path,
3355                                                 pending_del_slot,
3356                                                 pending_del_nr);
3357                                 if (ret) {
3358                                         btrfs_abort_transaction(trans,
3359                                                                 root, ret);
3360                                         goto error;
3361                                 }
3362                                 pending_del_nr = 0;
3363                         }
3364                         btrfs_release_path(path);
3365                         goto search_again;
3366                 } else {
3367                         path->slots[0]--;
3368                 }
3369         }
3370 out:
3371         if (pending_del_nr) {
3372                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3373                                       pending_del_nr);
3374                 if (ret)
3375                         btrfs_abort_transaction(trans, root, ret);
3376         }
3377 error:
3378         btrfs_free_path(path);
3379         return err;
3380 }
3381
3382 /*
3383  * taken from block_truncate_page, but does cow as it zeros out
3384  * any bytes left in the last page in the file.
3385  */
3386 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3387 {
3388         struct inode *inode = mapping->host;
3389         struct btrfs_root *root = BTRFS_I(inode)->root;
3390         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3391         struct btrfs_ordered_extent *ordered;
3392         struct extent_state *cached_state = NULL;
3393         char *kaddr;
3394         u32 blocksize = root->sectorsize;
3395         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3396         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3397         struct page *page;
3398         gfp_t mask = btrfs_alloc_write_mask(mapping);
3399         int ret = 0;
3400         u64 page_start;
3401         u64 page_end;
3402
3403         if ((offset & (blocksize - 1)) == 0)
3404                 goto out;
3405         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3406         if (ret)
3407                 goto out;
3408
3409         ret = -ENOMEM;
3410 again:
3411         page = find_or_create_page(mapping, index, mask);
3412         if (!page) {
3413                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3414                 goto out;
3415         }
3416
3417         page_start = page_offset(page);
3418         page_end = page_start + PAGE_CACHE_SIZE - 1;
3419
3420         if (!PageUptodate(page)) {
3421                 ret = btrfs_readpage(NULL, page);
3422                 lock_page(page);
3423                 if (page->mapping != mapping) {
3424                         unlock_page(page);
3425                         page_cache_release(page);
3426                         goto again;
3427                 }
3428                 if (!PageUptodate(page)) {
3429                         ret = -EIO;
3430                         goto out_unlock;
3431                 }
3432         }
3433         wait_on_page_writeback(page);
3434
3435         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3436         set_page_extent_mapped(page);
3437
3438         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3439         if (ordered) {
3440                 unlock_extent_cached(io_tree, page_start, page_end,
3441                                      &cached_state, GFP_NOFS);
3442                 unlock_page(page);
3443                 page_cache_release(page);
3444                 btrfs_start_ordered_extent(inode, ordered, 1);
3445                 btrfs_put_ordered_extent(ordered);
3446                 goto again;
3447         }
3448
3449         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3450                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3451                           0, 0, &cached_state, GFP_NOFS);
3452
3453         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3454                                         &cached_state);
3455         if (ret) {
3456                 unlock_extent_cached(io_tree, page_start, page_end,
3457                                      &cached_state, GFP_NOFS);
3458                 goto out_unlock;
3459         }
3460
3461         ret = 0;
3462         if (offset != PAGE_CACHE_SIZE) {
3463                 kaddr = kmap(page);
3464                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3465                 flush_dcache_page(page);
3466                 kunmap(page);
3467         }
3468         ClearPageChecked(page);
3469         set_page_dirty(page);
3470         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3471                              GFP_NOFS);
3472
3473 out_unlock:
3474         if (ret)
3475                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3476         unlock_page(page);
3477         page_cache_release(page);
3478 out:
3479         return ret;
3480 }
3481
3482 /*
3483  * This function puts in dummy file extents for the area we're creating a hole
3484  * for.  So if we are truncating this file to a larger size we need to insert
3485  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3486  * the range between oldsize and size
3487  */
3488 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3489 {
3490         struct btrfs_trans_handle *trans;
3491         struct btrfs_root *root = BTRFS_I(inode)->root;
3492         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3493         struct extent_map *em = NULL;
3494         struct extent_state *cached_state = NULL;
3495         u64 mask = root->sectorsize - 1;
3496         u64 hole_start = (oldsize + mask) & ~mask;
3497         u64 block_end = (size + mask) & ~mask;
3498         u64 last_byte;
3499         u64 cur_offset;
3500         u64 hole_size;
3501         int err = 0;
3502
3503         if (size <= hole_start)
3504                 return 0;
3505
3506         while (1) {
3507                 struct btrfs_ordered_extent *ordered;
3508                 btrfs_wait_ordered_range(inode, hole_start,
3509                                          block_end - hole_start);
3510                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3511                                  &cached_state);
3512                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3513                 if (!ordered)
3514                         break;
3515                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3516                                      &cached_state, GFP_NOFS);
3517                 btrfs_put_ordered_extent(ordered);
3518         }
3519
3520         cur_offset = hole_start;
3521         while (1) {
3522                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3523                                 block_end - cur_offset, 0);
3524                 if (IS_ERR(em)) {
3525                         err = PTR_ERR(em);
3526                         break;
3527                 }
3528                 last_byte = min(extent_map_end(em), block_end);
3529                 last_byte = (last_byte + mask) & ~mask;
3530                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3531                         u64 hint_byte = 0;
3532                         hole_size = last_byte - cur_offset;
3533
3534                         trans = btrfs_start_transaction(root, 3);
3535                         if (IS_ERR(trans)) {
3536                                 err = PTR_ERR(trans);
3537                                 break;
3538                         }
3539
3540                         err = btrfs_drop_extents(trans, inode, cur_offset,
3541                                                  cur_offset + hole_size,
3542                                                  &hint_byte, 1);
3543                         if (err) {
3544                                 btrfs_abort_transaction(trans, root, err);
3545                                 btrfs_end_transaction(trans, root);
3546                                 break;
3547                         }
3548
3549                         err = btrfs_insert_file_extent(trans, root,
3550                                         btrfs_ino(inode), cur_offset, 0,
3551                                         0, hole_size, 0, hole_size,
3552                                         0, 0, 0);
3553                         if (err) {
3554                                 btrfs_abort_transaction(trans, root, err);
3555                                 btrfs_end_transaction(trans, root);
3556                                 break;
3557                         }
3558
3559                         btrfs_drop_extent_cache(inode, hole_start,
3560                                         last_byte - 1, 0);
3561
3562                         btrfs_update_inode(trans, root, inode);
3563                         btrfs_end_transaction(trans, root);
3564                 }
3565                 free_extent_map(em);
3566                 em = NULL;
3567                 cur_offset = last_byte;
3568                 if (cur_offset >= block_end)
3569                         break;
3570         }
3571
3572         free_extent_map(em);
3573         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3574                              GFP_NOFS);
3575         return err;
3576 }
3577
3578 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3579 {
3580         struct btrfs_root *root = BTRFS_I(inode)->root;
3581         struct btrfs_trans_handle *trans;
3582         loff_t oldsize = i_size_read(inode);
3583         int ret;
3584
3585         if (newsize == oldsize)
3586                 return 0;
3587
3588         if (newsize > oldsize) {
3589                 truncate_pagecache(inode, oldsize, newsize);
3590                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3591                 if (ret)
3592                         return ret;
3593
3594                 trans = btrfs_start_transaction(root, 1);
3595                 if (IS_ERR(trans))
3596                         return PTR_ERR(trans);
3597
3598                 i_size_write(inode, newsize);
3599                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3600                 ret = btrfs_update_inode(trans, root, inode);
3601                 btrfs_end_transaction(trans, root);
3602         } else {
3603
3604                 /*
3605                  * We're truncating a file that used to have good data down to
3606                  * zero. Make sure it gets into the ordered flush list so that
3607                  * any new writes get down to disk quickly.
3608                  */
3609                 if (newsize == 0)
3610                         BTRFS_I(inode)->ordered_data_close = 1;
3611
3612                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3613                 truncate_setsize(inode, newsize);
3614                 ret = btrfs_truncate(inode);
3615         }
3616
3617         return ret;
3618 }
3619
3620 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3621 {
3622         struct inode *inode = dentry->d_inode;
3623         struct btrfs_root *root = BTRFS_I(inode)->root;
3624         int err;
3625
3626         if (btrfs_root_readonly(root))
3627                 return -EROFS;
3628
3629         err = inode_change_ok(inode, attr);
3630         if (err)
3631                 return err;
3632
3633         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3634                 err = btrfs_setsize(inode, attr->ia_size);
3635                 if (err)
3636                         return err;
3637         }
3638
3639         if (attr->ia_valid) {
3640                 setattr_copy(inode, attr);
3641                 err = btrfs_dirty_inode(inode);
3642
3643                 if (!err && attr->ia_valid & ATTR_MODE)
3644                         err = btrfs_acl_chmod(inode);
3645         }
3646
3647         return err;
3648 }
3649
3650 void btrfs_evict_inode(struct inode *inode)
3651 {
3652         struct btrfs_trans_handle *trans;
3653         struct btrfs_root *root = BTRFS_I(inode)->root;
3654         struct btrfs_block_rsv *rsv, *global_rsv;
3655         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3656         unsigned long nr;
3657         int ret;
3658
3659         trace_btrfs_inode_evict(inode);
3660
3661         truncate_inode_pages(&inode->i_data, 0);
3662         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3663                                btrfs_is_free_space_inode(root, inode)))
3664                 goto no_delete;
3665
3666         if (is_bad_inode(inode)) {
3667                 btrfs_orphan_del(NULL, inode);
3668                 goto no_delete;
3669         }
3670         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3671         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3672
3673         if (root->fs_info->log_root_recovering) {
3674                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3675                 goto no_delete;
3676         }
3677
3678         if (inode->i_nlink > 0) {
3679                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3680                 goto no_delete;
3681         }
3682
3683         rsv = btrfs_alloc_block_rsv(root);
3684         if (!rsv) {
3685                 btrfs_orphan_del(NULL, inode);
3686                 goto no_delete;
3687         }
3688         rsv->size = min_size;
3689         global_rsv = &root->fs_info->global_block_rsv;
3690
3691         btrfs_i_size_write(inode, 0);
3692
3693         /*
3694          * This is a bit simpler than btrfs_truncate since
3695          *
3696          * 1) We've already reserved our space for our orphan item in the
3697          *    unlink.
3698          * 2) We're going to delete the inode item, so we don't need to update
3699          *    it at all.
3700          *
3701          * So we just need to reserve some slack space in case we add bytes when
3702          * doing the truncate.
3703          */
3704         while (1) {
3705                 ret = btrfs_block_rsv_refill_noflush(root, rsv, min_size);
3706
3707                 /*
3708                  * Try and steal from the global reserve since we will
3709                  * likely not use this space anyway, we want to try as
3710                  * hard as possible to get this to work.
3711                  */
3712                 if (ret)
3713                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3714
3715                 if (ret) {
3716                         printk(KERN_WARNING "Could not get space for a "
3717                                "delete, will truncate on mount %d\n", ret);
3718                         btrfs_orphan_del(NULL, inode);
3719                         btrfs_free_block_rsv(root, rsv);
3720                         goto no_delete;
3721                 }
3722
3723                 trans = btrfs_start_transaction(root, 0);
3724                 if (IS_ERR(trans)) {
3725                         btrfs_orphan_del(NULL, inode);
3726                         btrfs_free_block_rsv(root, rsv);
3727                         goto no_delete;
3728                 }
3729
3730                 trans->block_rsv = rsv;
3731
3732                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3733                 if (ret != -EAGAIN)
3734                         break;
3735
3736                 nr = trans->blocks_used;
3737                 btrfs_end_transaction(trans, root);
3738                 trans = NULL;
3739                 btrfs_btree_balance_dirty(root, nr);
3740         }
3741
3742         btrfs_free_block_rsv(root, rsv);
3743
3744         if (ret == 0) {
3745                 trans->block_rsv = root->orphan_block_rsv;
3746                 ret = btrfs_orphan_del(trans, inode);
3747                 BUG_ON(ret);
3748         }
3749
3750         trans->block_rsv = &root->fs_info->trans_block_rsv;
3751         if (!(root == root->fs_info->tree_root ||
3752               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3753                 btrfs_return_ino(root, btrfs_ino(inode));
3754
3755         nr = trans->blocks_used;
3756         btrfs_end_transaction(trans, root);
3757         btrfs_btree_balance_dirty(root, nr);
3758 no_delete:
3759         clear_inode(inode);
3760         return;
3761 }
3762
3763 /*
3764  * this returns the key found in the dir entry in the location pointer.
3765  * If no dir entries were found, location->objectid is 0.
3766  */
3767 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3768                                struct btrfs_key *location)
3769 {
3770         const char *name = dentry->d_name.name;
3771         int namelen = dentry->d_name.len;
3772         struct btrfs_dir_item *di;
3773         struct btrfs_path *path;
3774         struct btrfs_root *root = BTRFS_I(dir)->root;
3775         int ret = 0;
3776
3777         path = btrfs_alloc_path();
3778         if (!path)
3779                 return -ENOMEM;
3780
3781         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3782                                     namelen, 0);
3783         if (IS_ERR(di))
3784                 ret = PTR_ERR(di);
3785
3786         if (IS_ERR_OR_NULL(di))
3787                 goto out_err;
3788
3789         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3790 out:
3791         btrfs_free_path(path);
3792         return ret;
3793 out_err:
3794         location->objectid = 0;
3795         goto out;
3796 }
3797
3798 /*
3799  * when we hit a tree root in a directory, the btrfs part of the inode
3800  * needs to be changed to reflect the root directory of the tree root.  This
3801  * is kind of like crossing a mount point.
3802  */
3803 static int fixup_tree_root_location(struct btrfs_root *root,
3804                                     struct inode *dir,
3805                                     struct dentry *dentry,
3806                                     struct btrfs_key *location,
3807                                     struct btrfs_root **sub_root)
3808 {
3809         struct btrfs_path *path;
3810         struct btrfs_root *new_root;
3811         struct btrfs_root_ref *ref;
3812         struct extent_buffer *leaf;
3813         int ret;
3814         int err = 0;
3815
3816         path = btrfs_alloc_path();
3817         if (!path) {
3818                 err = -ENOMEM;
3819                 goto out;
3820         }
3821
3822         err = -ENOENT;
3823         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3824                                   BTRFS_I(dir)->root->root_key.objectid,
3825                                   location->objectid);
3826         if (ret) {
3827                 if (ret < 0)
3828                         err = ret;
3829                 goto out;
3830         }
3831
3832         leaf = path->nodes[0];
3833         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3834         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3835             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3836                 goto out;
3837
3838         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3839                                    (unsigned long)(ref + 1),
3840                                    dentry->d_name.len);
3841         if (ret)
3842                 goto out;
3843
3844         btrfs_release_path(path);
3845
3846         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3847         if (IS_ERR(new_root)) {
3848                 err = PTR_ERR(new_root);
3849                 goto out;
3850         }
3851
3852         if (btrfs_root_refs(&new_root->root_item) == 0) {
3853                 err = -ENOENT;
3854                 goto out;
3855         }
3856
3857         *sub_root = new_root;
3858         location->objectid = btrfs_root_dirid(&new_root->root_item);
3859         location->type = BTRFS_INODE_ITEM_KEY;
3860         location->offset = 0;
3861         err = 0;
3862 out:
3863         btrfs_free_path(path);
3864         return err;
3865 }
3866
3867 static void inode_tree_add(struct inode *inode)
3868 {
3869         struct btrfs_root *root = BTRFS_I(inode)->root;
3870         struct btrfs_inode *entry;
3871         struct rb_node **p;
3872         struct rb_node *parent;
3873         u64 ino = btrfs_ino(inode);
3874 again:
3875         p = &root->inode_tree.rb_node;
3876         parent = NULL;
3877
3878         if (inode_unhashed(inode))
3879                 return;
3880
3881         spin_lock(&root->inode_lock);
3882         while (*p) {
3883                 parent = *p;
3884                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3885
3886                 if (ino < btrfs_ino(&entry->vfs_inode))
3887                         p = &parent->rb_left;
3888                 else if (ino > btrfs_ino(&entry->vfs_inode))
3889                         p = &parent->rb_right;
3890                 else {
3891                         WARN_ON(!(entry->vfs_inode.i_state &
3892                                   (I_WILL_FREE | I_FREEING)));
3893                         rb_erase(parent, &root->inode_tree);
3894                         RB_CLEAR_NODE(parent);
3895                         spin_unlock(&root->inode_lock);
3896                         goto again;
3897                 }
3898         }
3899         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3900         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3901         spin_unlock(&root->inode_lock);
3902 }
3903
3904 static void inode_tree_del(struct inode *inode)
3905 {
3906         struct btrfs_root *root = BTRFS_I(inode)->root;
3907         int empty = 0;
3908
3909         spin_lock(&root->inode_lock);
3910         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3911                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3912                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3913                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3914         }
3915         spin_unlock(&root->inode_lock);
3916
3917         /*
3918          * Free space cache has inodes in the tree root, but the tree root has a
3919          * root_refs of 0, so this could end up dropping the tree root as a
3920          * snapshot, so we need the extra !root->fs_info->tree_root check to
3921          * make sure we don't drop it.
3922          */
3923         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3924             root != root->fs_info->tree_root) {
3925                 synchronize_srcu(&root->fs_info->subvol_srcu);
3926                 spin_lock(&root->inode_lock);
3927                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3928                 spin_unlock(&root->inode_lock);
3929                 if (empty)
3930                         btrfs_add_dead_root(root);
3931         }
3932 }
3933
3934 void btrfs_invalidate_inodes(struct btrfs_root *root)
3935 {
3936         struct rb_node *node;
3937         struct rb_node *prev;
3938         struct btrfs_inode *entry;
3939         struct inode *inode;
3940         u64 objectid = 0;
3941
3942         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3943
3944         spin_lock(&root->inode_lock);
3945 again:
3946         node = root->inode_tree.rb_node;
3947         prev = NULL;
3948         while (node) {
3949                 prev = node;
3950                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3951
3952                 if (objectid < btrfs_ino(&entry->vfs_inode))
3953                         node = node->rb_left;
3954                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3955                         node = node->rb_right;
3956                 else
3957                         break;
3958         }
3959         if (!node) {
3960                 while (prev) {
3961                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3962                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3963                                 node = prev;
3964                                 break;
3965                         }
3966                         prev = rb_next(prev);
3967                 }
3968         }
3969         while (node) {
3970                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3971                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3972                 inode = igrab(&entry->vfs_inode);
3973                 if (inode) {
3974                         spin_unlock(&root->inode_lock);
3975                         if (atomic_read(&inode->i_count) > 1)
3976                                 d_prune_aliases(inode);
3977                         /*
3978                          * btrfs_drop_inode will have it removed from
3979                          * the inode cache when its usage count
3980                          * hits zero.
3981                          */
3982                         iput(inode);
3983                         cond_resched();
3984                         spin_lock(&root->inode_lock);
3985                         goto again;
3986                 }
3987
3988                 if (cond_resched_lock(&root->inode_lock))
3989                         goto again;
3990
3991                 node = rb_next(node);
3992         }
3993         spin_unlock(&root->inode_lock);
3994 }
3995
3996 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3997 {
3998         struct btrfs_iget_args *args = p;
3999         inode->i_ino = args->ino;
4000         BTRFS_I(inode)->root = args->root;
4001         btrfs_set_inode_space_info(args->root, inode);
4002         return 0;
4003 }
4004
4005 static int btrfs_find_actor(struct inode *inode, void *opaque)
4006 {
4007         struct btrfs_iget_args *args = opaque;
4008         return args->ino == btrfs_ino(inode) &&
4009                 args->root == BTRFS_I(inode)->root;
4010 }
4011
4012 static struct inode *btrfs_iget_locked(struct super_block *s,
4013                                        u64 objectid,
4014                                        struct btrfs_root *root)
4015 {
4016         struct inode *inode;
4017         struct btrfs_iget_args args;
4018         args.ino = objectid;
4019         args.root = root;
4020
4021         inode = iget5_locked(s, objectid, btrfs_find_actor,
4022                              btrfs_init_locked_inode,
4023                              (void *)&args);
4024         return inode;
4025 }
4026
4027 /* Get an inode object given its location and corresponding root.
4028  * Returns in *is_new if the inode was read from disk
4029  */
4030 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4031                          struct btrfs_root *root, int *new)
4032 {
4033         struct inode *inode;
4034
4035         inode = btrfs_iget_locked(s, location->objectid, root);
4036         if (!inode)
4037                 return ERR_PTR(-ENOMEM);
4038
4039         if (inode->i_state & I_NEW) {
4040                 BTRFS_I(inode)->root = root;
4041                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4042                 btrfs_read_locked_inode(inode);
4043                 if (!is_bad_inode(inode)) {
4044                         inode_tree_add(inode);
4045                         unlock_new_inode(inode);
4046                         if (new)
4047                                 *new = 1;
4048                 } else {
4049                         unlock_new_inode(inode);
4050                         iput(inode);
4051                         inode = ERR_PTR(-ESTALE);
4052                 }
4053         }
4054
4055         return inode;
4056 }
4057
4058 static struct inode *new_simple_dir(struct super_block *s,
4059                                     struct btrfs_key *key,
4060                                     struct btrfs_root *root)
4061 {
4062         struct inode *inode = new_inode(s);
4063
4064         if (!inode)
4065                 return ERR_PTR(-ENOMEM);
4066
4067         BTRFS_I(inode)->root = root;
4068         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4069         BTRFS_I(inode)->dummy_inode = 1;
4070
4071         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4072         inode->i_op = &btrfs_dir_ro_inode_operations;
4073         inode->i_fop = &simple_dir_operations;
4074         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4075         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4076
4077         return inode;
4078 }
4079
4080 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4081 {
4082         struct inode *inode;
4083         struct btrfs_root *root = BTRFS_I(dir)->root;
4084         struct btrfs_root *sub_root = root;
4085         struct btrfs_key location;
4086         int index;
4087         int ret = 0;
4088
4089         if (dentry->d_name.len > BTRFS_NAME_LEN)
4090                 return ERR_PTR(-ENAMETOOLONG);
4091
4092         if (unlikely(d_need_lookup(dentry))) {
4093                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4094                 kfree(dentry->d_fsdata);
4095                 dentry->d_fsdata = NULL;
4096                 /* This thing is hashed, drop it for now */
4097                 d_drop(dentry);
4098         } else {
4099                 ret = btrfs_inode_by_name(dir, dentry, &location);
4100         }
4101
4102         if (ret < 0)
4103                 return ERR_PTR(ret);
4104
4105         if (location.objectid == 0)
4106                 return NULL;
4107
4108         if (location.type == BTRFS_INODE_ITEM_KEY) {
4109                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4110                 return inode;
4111         }
4112
4113         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4114
4115         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4116         ret = fixup_tree_root_location(root, dir, dentry,
4117                                        &location, &sub_root);
4118         if (ret < 0) {
4119                 if (ret != -ENOENT)
4120                         inode = ERR_PTR(ret);
4121                 else
4122                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4123         } else {
4124                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4125         }
4126         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4127
4128         if (!IS_ERR(inode) && root != sub_root) {
4129                 down_read(&root->fs_info->cleanup_work_sem);
4130                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4131                         ret = btrfs_orphan_cleanup(sub_root);
4132                 up_read(&root->fs_info->cleanup_work_sem);
4133                 if (ret)
4134                         inode = ERR_PTR(ret);
4135         }
4136
4137         return inode;
4138 }
4139
4140 static int btrfs_dentry_delete(const struct dentry *dentry)
4141 {
4142         struct btrfs_root *root;
4143         struct inode *inode = dentry->d_inode;
4144
4145         if (!inode && !IS_ROOT(dentry))
4146                 inode = dentry->d_parent->d_inode;
4147
4148         if (inode) {
4149                 root = BTRFS_I(inode)->root;
4150                 if (btrfs_root_refs(&root->root_item) == 0)
4151                         return 1;
4152
4153                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4154                         return 1;
4155         }
4156         return 0;
4157 }
4158
4159 static void btrfs_dentry_release(struct dentry *dentry)
4160 {
4161         if (dentry->d_fsdata)
4162                 kfree(dentry->d_fsdata);
4163 }
4164
4165 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4166                                    struct nameidata *nd)
4167 {
4168         struct dentry *ret;
4169
4170         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4171         if (unlikely(d_need_lookup(dentry))) {
4172                 spin_lock(&dentry->d_lock);
4173                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4174                 spin_unlock(&dentry->d_lock);
4175         }
4176         return ret;
4177 }
4178
4179 unsigned char btrfs_filetype_table[] = {
4180         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4181 };
4182
4183 static int btrfs_real_readdir(struct file *filp, void *dirent,
4184                               filldir_t filldir)
4185 {
4186         struct inode *inode = filp->f_dentry->d_inode;
4187         struct btrfs_root *root = BTRFS_I(inode)->root;
4188         struct btrfs_item *item;
4189         struct btrfs_dir_item *di;
4190         struct btrfs_key key;
4191         struct btrfs_key found_key;
4192         struct btrfs_path *path;
4193         struct list_head ins_list;
4194         struct list_head del_list;
4195         int ret;
4196         struct extent_buffer *leaf;
4197         int slot;
4198         unsigned char d_type;
4199         int over = 0;
4200         u32 di_cur;
4201         u32 di_total;
4202         u32 di_len;
4203         int key_type = BTRFS_DIR_INDEX_KEY;
4204         char tmp_name[32];
4205         char *name_ptr;
4206         int name_len;
4207         int is_curr = 0;        /* filp->f_pos points to the current index? */
4208
4209         /* FIXME, use a real flag for deciding about the key type */
4210         if (root->fs_info->tree_root == root)
4211                 key_type = BTRFS_DIR_ITEM_KEY;
4212
4213         /* special case for "." */
4214         if (filp->f_pos == 0) {
4215                 over = filldir(dirent, ".", 1,
4216                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4217                 if (over)
4218                         return 0;
4219                 filp->f_pos = 1;
4220         }
4221         /* special case for .., just use the back ref */
4222         if (filp->f_pos == 1) {
4223                 u64 pino = parent_ino(filp->f_path.dentry);
4224                 over = filldir(dirent, "..", 2,
4225                                filp->f_pos, pino, DT_DIR);
4226                 if (over)
4227                         return 0;
4228                 filp->f_pos = 2;
4229         }
4230         path = btrfs_alloc_path();
4231         if (!path)
4232                 return -ENOMEM;
4233
4234         path->reada = 1;
4235
4236         if (key_type == BTRFS_DIR_INDEX_KEY) {
4237                 INIT_LIST_HEAD(&ins_list);
4238                 INIT_LIST_HEAD(&del_list);
4239                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4240         }
4241
4242         btrfs_set_key_type(&key, key_type);
4243         key.offset = filp->f_pos;
4244         key.objectid = btrfs_ino(inode);
4245
4246         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4247         if (ret < 0)
4248                 goto err;
4249
4250         while (1) {
4251                 leaf = path->nodes[0];
4252                 slot = path->slots[0];
4253                 if (slot >= btrfs_header_nritems(leaf)) {
4254                         ret = btrfs_next_leaf(root, path);
4255                         if (ret < 0)
4256                                 goto err;
4257                         else if (ret > 0)
4258                                 break;
4259                         continue;
4260                 }
4261
4262                 item = btrfs_item_nr(leaf, slot);
4263                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4264
4265                 if (found_key.objectid != key.objectid)
4266                         break;
4267                 if (btrfs_key_type(&found_key) != key_type)
4268                         break;
4269                 if (found_key.offset < filp->f_pos)
4270                         goto next;
4271                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4272                     btrfs_should_delete_dir_index(&del_list,
4273                                                   found_key.offset))
4274                         goto next;
4275
4276                 filp->f_pos = found_key.offset;
4277                 is_curr = 1;
4278
4279                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4280                 di_cur = 0;
4281                 di_total = btrfs_item_size(leaf, item);
4282
4283                 while (di_cur < di_total) {
4284                         struct btrfs_key location;
4285
4286                         if (verify_dir_item(root, leaf, di))
4287                                 break;
4288
4289                         name_len = btrfs_dir_name_len(leaf, di);
4290                         if (name_len <= sizeof(tmp_name)) {
4291                                 name_ptr = tmp_name;
4292                         } else {
4293                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4294                                 if (!name_ptr) {
4295                                         ret = -ENOMEM;
4296                                         goto err;
4297                                 }
4298                         }
4299                         read_extent_buffer(leaf, name_ptr,
4300                                            (unsigned long)(di + 1), name_len);
4301
4302                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4303                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4304
4305
4306                         /* is this a reference to our own snapshot? If so
4307                          * skip it.
4308                          *
4309                          * In contrast to old kernels, we insert the snapshot's
4310                          * dir item and dir index after it has been created, so
4311                          * we won't find a reference to our own snapshot. We
4312                          * still keep the following code for backward
4313                          * compatibility.
4314                          */
4315                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4316                             location.objectid == root->root_key.objectid) {
4317                                 over = 0;
4318                                 goto skip;
4319                         }
4320                         over = filldir(dirent, name_ptr, name_len,
4321                                        found_key.offset, location.objectid,
4322                                        d_type);
4323
4324 skip:
4325                         if (name_ptr != tmp_name)
4326                                 kfree(name_ptr);
4327
4328                         if (over)
4329                                 goto nopos;
4330                         di_len = btrfs_dir_name_len(leaf, di) +
4331                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4332                         di_cur += di_len;
4333                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4334                 }
4335 next:
4336                 path->slots[0]++;
4337         }
4338
4339         if (key_type == BTRFS_DIR_INDEX_KEY) {
4340                 if (is_curr)
4341                         filp->f_pos++;
4342                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4343                                                       &ins_list);
4344                 if (ret)
4345                         goto nopos;
4346         }
4347
4348         /* Reached end of directory/root. Bump pos past the last item. */
4349         if (key_type == BTRFS_DIR_INDEX_KEY)
4350                 /*
4351                  * 32-bit glibc will use getdents64, but then strtol -
4352                  * so the last number we can serve is this.
4353                  */
4354                 filp->f_pos = 0x7fffffff;
4355         else
4356                 filp->f_pos++;
4357 nopos:
4358         ret = 0;
4359 err:
4360         if (key_type == BTRFS_DIR_INDEX_KEY)
4361                 btrfs_put_delayed_items(&ins_list, &del_list);
4362         btrfs_free_path(path);
4363         return ret;
4364 }
4365
4366 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4367 {
4368         struct btrfs_root *root = BTRFS_I(inode)->root;
4369         struct btrfs_trans_handle *trans;
4370         int ret = 0;
4371         bool nolock = false;
4372
4373         if (BTRFS_I(inode)->dummy_inode)
4374                 return 0;
4375
4376         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4377                 nolock = true;
4378
4379         if (wbc->sync_mode == WB_SYNC_ALL) {
4380                 if (nolock)
4381                         trans = btrfs_join_transaction_nolock(root);
4382                 else
4383                         trans = btrfs_join_transaction(root);
4384                 if (IS_ERR(trans))
4385                         return PTR_ERR(trans);
4386                 if (nolock)
4387                         ret = btrfs_end_transaction_nolock(trans, root);
4388                 else
4389                         ret = btrfs_commit_transaction(trans, root);
4390         }
4391         return ret;
4392 }
4393
4394 /*
4395  * This is somewhat expensive, updating the tree every time the
4396  * inode changes.  But, it is most likely to find the inode in cache.
4397  * FIXME, needs more benchmarking...there are no reasons other than performance
4398  * to keep or drop this code.
4399  */
4400 int btrfs_dirty_inode(struct inode *inode)
4401 {
4402         struct btrfs_root *root = BTRFS_I(inode)->root;
4403         struct btrfs_trans_handle *trans;
4404         int ret;
4405
4406         if (BTRFS_I(inode)->dummy_inode)
4407                 return 0;
4408
4409         trans = btrfs_join_transaction(root);
4410         if (IS_ERR(trans))
4411                 return PTR_ERR(trans);
4412
4413         ret = btrfs_update_inode(trans, root, inode);
4414         if (ret && ret == -ENOSPC) {
4415                 /* whoops, lets try again with the full transaction */
4416                 btrfs_end_transaction(trans, root);
4417                 trans = btrfs_start_transaction(root, 1);
4418                 if (IS_ERR(trans))
4419                         return PTR_ERR(trans);
4420
4421                 ret = btrfs_update_inode(trans, root, inode);
4422         }
4423         btrfs_end_transaction(trans, root);
4424         if (BTRFS_I(inode)->delayed_node)
4425                 btrfs_balance_delayed_items(root);
4426
4427         return ret;
4428 }
4429
4430 /*
4431  * This is a copy of file_update_time.  We need this so we can return error on
4432  * ENOSPC for updating the inode in the case of file write and mmap writes.
4433  */
4434 int btrfs_update_time(struct file *file)
4435 {
4436         struct inode *inode = file->f_path.dentry->d_inode;
4437         struct timespec now;
4438         int ret;
4439         enum { S_MTIME = 1, S_CTIME = 2, S_VERSION = 4 } sync_it = 0;
4440
4441         /* First try to exhaust all avenues to not sync */
4442         if (IS_NOCMTIME(inode))
4443                 return 0;
4444
4445         now = current_fs_time(inode->i_sb);
4446         if (!timespec_equal(&inode->i_mtime, &now))
4447                 sync_it = S_MTIME;
4448
4449         if (!timespec_equal(&inode->i_ctime, &now))
4450                 sync_it |= S_CTIME;
4451
4452         if (IS_I_VERSION(inode))
4453                 sync_it |= S_VERSION;
4454
4455         if (!sync_it)
4456                 return 0;
4457
4458         /* Finally allowed to write? Takes lock. */
4459         if (mnt_want_write_file(file))
4460                 return 0;
4461
4462         /* Only change inode inside the lock region */
4463         if (sync_it & S_VERSION)
4464                 inode_inc_iversion(inode);
4465         if (sync_it & S_CTIME)
4466                 inode->i_ctime = now;
4467         if (sync_it & S_MTIME)
4468                 inode->i_mtime = now;
4469         ret = btrfs_dirty_inode(inode);
4470         if (!ret)
4471                 mark_inode_dirty_sync(inode);
4472         mnt_drop_write(file->f_path.mnt);
4473         return ret;
4474 }
4475
4476 /*
4477  * find the highest existing sequence number in a directory
4478  * and then set the in-memory index_cnt variable to reflect
4479  * free sequence numbers
4480  */
4481 static int btrfs_set_inode_index_count(struct inode *inode)
4482 {
4483         struct btrfs_root *root = BTRFS_I(inode)->root;
4484         struct btrfs_key key, found_key;
4485         struct btrfs_path *path;
4486         struct extent_buffer *leaf;
4487         int ret;
4488
4489         key.objectid = btrfs_ino(inode);
4490         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4491         key.offset = (u64)-1;
4492
4493         path = btrfs_alloc_path();
4494         if (!path)
4495                 return -ENOMEM;
4496
4497         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4498         if (ret < 0)
4499                 goto out;
4500         /* FIXME: we should be able to handle this */
4501         if (ret == 0)
4502                 goto out;
4503         ret = 0;
4504
4505         /*
4506          * MAGIC NUMBER EXPLANATION:
4507          * since we search a directory based on f_pos we have to start at 2
4508          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4509          * else has to start at 2
4510          */
4511         if (path->slots[0] == 0) {
4512                 BTRFS_I(inode)->index_cnt = 2;
4513                 goto out;
4514         }
4515
4516         path->slots[0]--;
4517
4518         leaf = path->nodes[0];
4519         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4520
4521         if (found_key.objectid != btrfs_ino(inode) ||
4522             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4523                 BTRFS_I(inode)->index_cnt = 2;
4524                 goto out;
4525         }
4526
4527         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4528 out:
4529         btrfs_free_path(path);
4530         return ret;
4531 }
4532
4533 /*
4534  * helper to find a free sequence number in a given directory.  This current
4535  * code is very simple, later versions will do smarter things in the btree
4536  */
4537 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4538 {
4539         int ret = 0;
4540
4541         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4542                 ret = btrfs_inode_delayed_dir_index_count(dir);
4543                 if (ret) {
4544                         ret = btrfs_set_inode_index_count(dir);
4545                         if (ret)
4546                                 return ret;
4547                 }
4548         }
4549
4550         *index = BTRFS_I(dir)->index_cnt;
4551         BTRFS_I(dir)->index_cnt++;
4552
4553         return ret;
4554 }
4555
4556 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4557                                      struct btrfs_root *root,
4558                                      struct inode *dir,
4559                                      const char *name, int name_len,
4560                                      u64 ref_objectid, u64 objectid,
4561                                      umode_t mode, u64 *index)
4562 {
4563         struct inode *inode;
4564         struct btrfs_inode_item *inode_item;
4565         struct btrfs_key *location;
4566         struct btrfs_path *path;
4567         struct btrfs_inode_ref *ref;
4568         struct btrfs_key key[2];
4569         u32 sizes[2];
4570         unsigned long ptr;
4571         int ret;
4572         int owner;
4573
4574         path = btrfs_alloc_path();
4575         if (!path)
4576                 return ERR_PTR(-ENOMEM);
4577
4578         inode = new_inode(root->fs_info->sb);
4579         if (!inode) {
4580                 btrfs_free_path(path);
4581                 return ERR_PTR(-ENOMEM);
4582         }
4583
4584         /*
4585          * we have to initialize this early, so we can reclaim the inode
4586          * number if we fail afterwards in this function.
4587          */
4588         inode->i_ino = objectid;
4589
4590         if (dir) {
4591                 trace_btrfs_inode_request(dir);
4592
4593                 ret = btrfs_set_inode_index(dir, index);
4594                 if (ret) {
4595                         btrfs_free_path(path);
4596                         iput(inode);
4597                         return ERR_PTR(ret);
4598                 }
4599         }
4600         /*
4601          * index_cnt is ignored for everything but a dir,
4602          * btrfs_get_inode_index_count has an explanation for the magic
4603          * number
4604          */
4605         BTRFS_I(inode)->index_cnt = 2;
4606         BTRFS_I(inode)->root = root;
4607         BTRFS_I(inode)->generation = trans->transid;
4608         inode->i_generation = BTRFS_I(inode)->generation;
4609         btrfs_set_inode_space_info(root, inode);
4610
4611         if (S_ISDIR(mode))
4612                 owner = 0;
4613         else
4614                 owner = 1;
4615
4616         key[0].objectid = objectid;
4617         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4618         key[0].offset = 0;
4619
4620         key[1].objectid = objectid;
4621         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4622         key[1].offset = ref_objectid;
4623
4624         sizes[0] = sizeof(struct btrfs_inode_item);
4625         sizes[1] = name_len + sizeof(*ref);
4626
4627         path->leave_spinning = 1;
4628         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4629         if (ret != 0)
4630                 goto fail;
4631
4632         inode_init_owner(inode, dir, mode);
4633         inode_set_bytes(inode, 0);
4634         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4635         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4636                                   struct btrfs_inode_item);
4637         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4638
4639         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4640                              struct btrfs_inode_ref);
4641         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4642         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4643         ptr = (unsigned long)(ref + 1);
4644         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4645
4646         btrfs_mark_buffer_dirty(path->nodes[0]);
4647         btrfs_free_path(path);
4648
4649         location = &BTRFS_I(inode)->location;
4650         location->objectid = objectid;
4651         location->offset = 0;
4652         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4653
4654         btrfs_inherit_iflags(inode, dir);
4655
4656         if (S_ISREG(mode)) {
4657                 if (btrfs_test_opt(root, NODATASUM))
4658                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4659                 if (btrfs_test_opt(root, NODATACOW) ||
4660                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4661                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4662         }
4663
4664         insert_inode_hash(inode);
4665         inode_tree_add(inode);
4666
4667         trace_btrfs_inode_new(inode);
4668         btrfs_set_inode_last_trans(trans, inode);
4669
4670         return inode;
4671 fail:
4672         if (dir)
4673                 BTRFS_I(dir)->index_cnt--;
4674         btrfs_free_path(path);
4675         iput(inode);
4676         return ERR_PTR(ret);
4677 }
4678
4679 static inline u8 btrfs_inode_type(struct inode *inode)
4680 {
4681         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4682 }
4683
4684 /*
4685  * utility function to add 'inode' into 'parent_inode' with
4686  * a give name and a given sequence number.
4687  * if 'add_backref' is true, also insert a backref from the
4688  * inode to the parent directory.
4689  */
4690 int btrfs_add_link(struct btrfs_trans_handle *trans,
4691                    struct inode *parent_inode, struct inode *inode,
4692                    const char *name, int name_len, int add_backref, u64 index)
4693 {
4694         int ret = 0;
4695         struct btrfs_key key;
4696         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4697         u64 ino = btrfs_ino(inode);
4698         u64 parent_ino = btrfs_ino(parent_inode);
4699
4700         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4701                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4702         } else {
4703                 key.objectid = ino;
4704                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4705                 key.offset = 0;
4706         }
4707
4708         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4709                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4710                                          key.objectid, root->root_key.objectid,
4711                                          parent_ino, index, name, name_len);
4712         } else if (add_backref) {
4713                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4714                                              parent_ino, index);
4715         }
4716
4717         /* Nothing to clean up yet */
4718         if (ret)
4719                 return ret;
4720
4721         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4722                                     parent_inode, &key,
4723                                     btrfs_inode_type(inode), index);
4724         if (ret == -EEXIST)
4725                 goto fail_dir_item;
4726         else if (ret) {
4727                 btrfs_abort_transaction(trans, root, ret);
4728                 return ret;
4729         }
4730
4731         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4732                            name_len * 2);
4733         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4734         ret = btrfs_update_inode(trans, root, parent_inode);
4735         if (ret)
4736                 btrfs_abort_transaction(trans, root, ret);
4737         return ret;
4738
4739 fail_dir_item:
4740         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4741                 u64 local_index;
4742                 int err;
4743                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4744                                  key.objectid, root->root_key.objectid,
4745                                  parent_ino, &local_index, name, name_len);
4746
4747         } else if (add_backref) {
4748                 u64 local_index;
4749                 int err;
4750
4751                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4752                                           ino, parent_ino, &local_index);
4753         }
4754         return ret;
4755 }
4756
4757 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4758                             struct inode *dir, struct dentry *dentry,
4759                             struct inode *inode, int backref, u64 index)
4760 {
4761         int err = btrfs_add_link(trans, dir, inode,
4762                                  dentry->d_name.name, dentry->d_name.len,
4763                                  backref, index);
4764         if (err > 0)
4765                 err = -EEXIST;
4766         return err;
4767 }
4768
4769 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4770                         umode_t mode, dev_t rdev)
4771 {
4772         struct btrfs_trans_handle *trans;
4773         struct btrfs_root *root = BTRFS_I(dir)->root;
4774         struct inode *inode = NULL;
4775         int err;
4776         int drop_inode = 0;
4777         u64 objectid;
4778         unsigned long nr = 0;
4779         u64 index = 0;
4780
4781         if (!new_valid_dev(rdev))
4782                 return -EINVAL;
4783
4784         /*
4785          * 2 for inode item and ref
4786          * 2 for dir items
4787          * 1 for xattr if selinux is on
4788          */
4789         trans = btrfs_start_transaction(root, 5);
4790         if (IS_ERR(trans))
4791                 return PTR_ERR(trans);
4792
4793         err = btrfs_find_free_ino(root, &objectid);
4794         if (err)
4795                 goto out_unlock;
4796
4797         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4798                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4799                                 mode, &index);
4800         if (IS_ERR(inode)) {
4801                 err = PTR_ERR(inode);
4802                 goto out_unlock;
4803         }
4804
4805         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4806         if (err) {
4807                 drop_inode = 1;
4808                 goto out_unlock;
4809         }
4810
4811         /*
4812         * If the active LSM wants to access the inode during
4813         * d_instantiate it needs these. Smack checks to see
4814         * if the filesystem supports xattrs by looking at the
4815         * ops vector.
4816         */
4817
4818         inode->i_op = &btrfs_special_inode_operations;
4819         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4820         if (err)
4821                 drop_inode = 1;
4822         else {
4823                 init_special_inode(inode, inode->i_mode, rdev);
4824                 btrfs_update_inode(trans, root, inode);
4825                 d_instantiate(dentry, inode);
4826         }
4827 out_unlock:
4828         nr = trans->blocks_used;
4829         btrfs_end_transaction(trans, root);
4830         btrfs_btree_balance_dirty(root, nr);
4831         if (drop_inode) {
4832                 inode_dec_link_count(inode);
4833                 iput(inode);
4834         }
4835         return err;
4836 }
4837
4838 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4839                         umode_t mode, struct nameidata *nd)
4840 {
4841         struct btrfs_trans_handle *trans;
4842         struct btrfs_root *root = BTRFS_I(dir)->root;
4843         struct inode *inode = NULL;
4844         int drop_inode = 0;
4845         int err;
4846         unsigned long nr = 0;
4847         u64 objectid;
4848         u64 index = 0;
4849
4850         /*
4851          * 2 for inode item and ref
4852          * 2 for dir items
4853          * 1 for xattr if selinux is on
4854          */
4855         trans = btrfs_start_transaction(root, 5);
4856         if (IS_ERR(trans))
4857                 return PTR_ERR(trans);
4858
4859         err = btrfs_find_free_ino(root, &objectid);
4860         if (err)
4861                 goto out_unlock;
4862
4863         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4864                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4865                                 mode, &index);
4866         if (IS_ERR(inode)) {
4867                 err = PTR_ERR(inode);
4868                 goto out_unlock;
4869         }
4870
4871         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4872         if (err) {
4873                 drop_inode = 1;
4874                 goto out_unlock;
4875         }
4876
4877         /*
4878         * If the active LSM wants to access the inode during
4879         * d_instantiate it needs these. Smack checks to see
4880         * if the filesystem supports xattrs by looking at the
4881         * ops vector.
4882         */
4883         inode->i_fop = &btrfs_file_operations;
4884         inode->i_op = &btrfs_file_inode_operations;
4885
4886         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4887         if (err)
4888                 drop_inode = 1;
4889         else {
4890                 inode->i_mapping->a_ops = &btrfs_aops;
4891                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4892                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4893                 d_instantiate(dentry, inode);
4894         }
4895 out_unlock:
4896         nr = trans->blocks_used;
4897         btrfs_end_transaction(trans, root);
4898         if (drop_inode) {
4899                 inode_dec_link_count(inode);
4900                 iput(inode);
4901         }
4902         btrfs_btree_balance_dirty(root, nr);
4903         return err;
4904 }
4905
4906 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4907                       struct dentry *dentry)
4908 {
4909         struct btrfs_trans_handle *trans;
4910         struct btrfs_root *root = BTRFS_I(dir)->root;
4911         struct inode *inode = old_dentry->d_inode;
4912         u64 index;
4913         unsigned long nr = 0;
4914         int err;
4915         int drop_inode = 0;
4916
4917         /* do not allow sys_link's with other subvols of the same device */
4918         if (root->objectid != BTRFS_I(inode)->root->objectid)
4919                 return -EXDEV;
4920
4921         if (inode->i_nlink == ~0U)
4922                 return -EMLINK;
4923
4924         err = btrfs_set_inode_index(dir, &index);
4925         if (err)
4926                 goto fail;
4927
4928         /*
4929          * 2 items for inode and inode ref
4930          * 2 items for dir items
4931          * 1 item for parent inode
4932          */
4933         trans = btrfs_start_transaction(root, 5);
4934         if (IS_ERR(trans)) {
4935                 err = PTR_ERR(trans);
4936                 goto fail;
4937         }
4938
4939         btrfs_inc_nlink(inode);
4940         inode->i_ctime = CURRENT_TIME;
4941         ihold(inode);
4942
4943         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4944
4945         if (err) {
4946                 drop_inode = 1;
4947         } else {
4948                 struct dentry *parent = dentry->d_parent;
4949                 err = btrfs_update_inode(trans, root, inode);
4950                 if (err)
4951                         goto fail;
4952                 d_instantiate(dentry, inode);
4953                 btrfs_log_new_name(trans, inode, NULL, parent);
4954         }
4955
4956         nr = trans->blocks_used;
4957         btrfs_end_transaction(trans, root);
4958 fail:
4959         if (drop_inode) {
4960                 inode_dec_link_count(inode);
4961                 iput(inode);
4962         }
4963         btrfs_btree_balance_dirty(root, nr);
4964         return err;
4965 }
4966
4967 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
4968 {
4969         struct inode *inode = NULL;
4970         struct btrfs_trans_handle *trans;
4971         struct btrfs_root *root = BTRFS_I(dir)->root;
4972         int err = 0;
4973         int drop_on_err = 0;
4974         u64 objectid = 0;
4975         u64 index = 0;
4976         unsigned long nr = 1;
4977
4978         /*
4979          * 2 items for inode and ref
4980          * 2 items for dir items
4981          * 1 for xattr if selinux is on
4982          */
4983         trans = btrfs_start_transaction(root, 5);
4984         if (IS_ERR(trans))
4985                 return PTR_ERR(trans);
4986
4987         err = btrfs_find_free_ino(root, &objectid);
4988         if (err)
4989                 goto out_fail;
4990
4991         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4992                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4993                                 S_IFDIR | mode, &index);
4994         if (IS_ERR(inode)) {
4995                 err = PTR_ERR(inode);
4996                 goto out_fail;
4997         }
4998
4999         drop_on_err = 1;
5000
5001         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5002         if (err)
5003                 goto out_fail;
5004
5005         inode->i_op = &btrfs_dir_inode_operations;
5006         inode->i_fop = &btrfs_dir_file_operations;
5007
5008         btrfs_i_size_write(inode, 0);
5009         err = btrfs_update_inode(trans, root, inode);
5010         if (err)
5011                 goto out_fail;
5012
5013         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5014                              dentry->d_name.len, 0, index);
5015         if (err)
5016                 goto out_fail;
5017
5018         d_instantiate(dentry, inode);
5019         drop_on_err = 0;
5020
5021 out_fail:
5022         nr = trans->blocks_used;
5023         btrfs_end_transaction(trans, root);
5024         if (drop_on_err)
5025                 iput(inode);
5026         btrfs_btree_balance_dirty(root, nr);
5027         return err;
5028 }
5029
5030 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5031  * and an extent that you want to insert, deal with overlap and insert
5032  * the new extent into the tree.
5033  */
5034 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5035                                 struct extent_map *existing,
5036                                 struct extent_map *em,
5037                                 u64 map_start, u64 map_len)
5038 {
5039         u64 start_diff;
5040
5041         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5042         start_diff = map_start - em->start;
5043         em->start = map_start;
5044         em->len = map_len;
5045         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5046             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5047                 em->block_start += start_diff;
5048                 em->block_len -= start_diff;
5049         }
5050         return add_extent_mapping(em_tree, em);
5051 }
5052
5053 static noinline int uncompress_inline(struct btrfs_path *path,
5054                                       struct inode *inode, struct page *page,
5055                                       size_t pg_offset, u64 extent_offset,
5056                                       struct btrfs_file_extent_item *item)
5057 {
5058         int ret;
5059         struct extent_buffer *leaf = path->nodes[0];
5060         char *tmp;
5061         size_t max_size;
5062         unsigned long inline_size;
5063         unsigned long ptr;
5064         int compress_type;
5065
5066         WARN_ON(pg_offset != 0);
5067         compress_type = btrfs_file_extent_compression(leaf, item);
5068         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5069         inline_size = btrfs_file_extent_inline_item_len(leaf,
5070                                         btrfs_item_nr(leaf, path->slots[0]));
5071         tmp = kmalloc(inline_size, GFP_NOFS);
5072         if (!tmp)
5073                 return -ENOMEM;
5074         ptr = btrfs_file_extent_inline_start(item);
5075
5076         read_extent_buffer(leaf, tmp, ptr, inline_size);
5077
5078         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5079         ret = btrfs_decompress(compress_type, tmp, page,
5080                                extent_offset, inline_size, max_size);
5081         if (ret) {
5082                 char *kaddr = kmap_atomic(page);
5083                 unsigned long copy_size = min_t(u64,
5084                                   PAGE_CACHE_SIZE - pg_offset,
5085                                   max_size - extent_offset);
5086                 memset(kaddr + pg_offset, 0, copy_size);
5087                 kunmap_atomic(kaddr);
5088         }
5089         kfree(tmp);
5090         return 0;
5091 }
5092
5093 /*
5094  * a bit scary, this does extent mapping from logical file offset to the disk.
5095  * the ugly parts come from merging extents from the disk with the in-ram
5096  * representation.  This gets more complex because of the data=ordered code,
5097  * where the in-ram extents might be locked pending data=ordered completion.
5098  *
5099  * This also copies inline extents directly into the page.
5100  */
5101
5102 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5103                                     size_t pg_offset, u64 start, u64 len,
5104                                     int create)
5105 {
5106         int ret;
5107         int err = 0;
5108         u64 bytenr;
5109         u64 extent_start = 0;
5110         u64 extent_end = 0;
5111         u64 objectid = btrfs_ino(inode);
5112         u32 found_type;
5113         struct btrfs_path *path = NULL;
5114         struct btrfs_root *root = BTRFS_I(inode)->root;
5115         struct btrfs_file_extent_item *item;
5116         struct extent_buffer *leaf;
5117         struct btrfs_key found_key;
5118         struct extent_map *em = NULL;
5119         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5120         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5121         struct btrfs_trans_handle *trans = NULL;
5122         int compress_type;
5123
5124 again:
5125         read_lock(&em_tree->lock);
5126         em = lookup_extent_mapping(em_tree, start, len);
5127         if (em)
5128                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5129         read_unlock(&em_tree->lock);
5130
5131         if (em) {
5132                 if (em->start > start || em->start + em->len <= start)
5133                         free_extent_map(em);
5134                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5135                         free_extent_map(em);
5136                 else
5137                         goto out;
5138         }
5139         em = alloc_extent_map();
5140         if (!em) {
5141                 err = -ENOMEM;
5142                 goto out;
5143         }
5144         em->bdev = root->fs_info->fs_devices->latest_bdev;
5145         em->start = EXTENT_MAP_HOLE;
5146         em->orig_start = EXTENT_MAP_HOLE;
5147         em->len = (u64)-1;
5148         em->block_len = (u64)-1;
5149
5150         if (!path) {
5151                 path = btrfs_alloc_path();
5152                 if (!path) {
5153                         err = -ENOMEM;
5154                         goto out;
5155                 }
5156                 /*
5157                  * Chances are we'll be called again, so go ahead and do
5158                  * readahead
5159                  */
5160                 path->reada = 1;
5161         }
5162
5163         ret = btrfs_lookup_file_extent(trans, root, path,
5164                                        objectid, start, trans != NULL);
5165         if (ret < 0) {
5166                 err = ret;
5167                 goto out;
5168         }
5169
5170         if (ret != 0) {
5171                 if (path->slots[0] == 0)
5172                         goto not_found;
5173                 path->slots[0]--;
5174         }
5175
5176         leaf = path->nodes[0];
5177         item = btrfs_item_ptr(leaf, path->slots[0],
5178                               struct btrfs_file_extent_item);
5179         /* are we inside the extent that was found? */
5180         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5181         found_type = btrfs_key_type(&found_key);
5182         if (found_key.objectid != objectid ||
5183             found_type != BTRFS_EXTENT_DATA_KEY) {
5184                 goto not_found;
5185         }
5186
5187         found_type = btrfs_file_extent_type(leaf, item);
5188         extent_start = found_key.offset;
5189         compress_type = btrfs_file_extent_compression(leaf, item);
5190         if (found_type == BTRFS_FILE_EXTENT_REG ||
5191             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5192                 extent_end = extent_start +
5193                        btrfs_file_extent_num_bytes(leaf, item);
5194         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5195                 size_t size;
5196                 size = btrfs_file_extent_inline_len(leaf, item);
5197                 extent_end = (extent_start + size + root->sectorsize - 1) &
5198                         ~((u64)root->sectorsize - 1);
5199         }
5200
5201         if (start >= extent_end) {
5202                 path->slots[0]++;
5203                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5204                         ret = btrfs_next_leaf(root, path);
5205                         if (ret < 0) {
5206                                 err = ret;
5207                                 goto out;
5208                         }
5209                         if (ret > 0)
5210                                 goto not_found;
5211                         leaf = path->nodes[0];
5212                 }
5213                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5214                 if (found_key.objectid != objectid ||
5215                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5216                         goto not_found;
5217                 if (start + len <= found_key.offset)
5218                         goto not_found;
5219                 em->start = start;
5220                 em->len = found_key.offset - start;
5221                 goto not_found_em;
5222         }
5223
5224         if (found_type == BTRFS_FILE_EXTENT_REG ||
5225             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5226                 em->start = extent_start;
5227                 em->len = extent_end - extent_start;
5228                 em->orig_start = extent_start -
5229                                  btrfs_file_extent_offset(leaf, item);
5230                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5231                 if (bytenr == 0) {
5232                         em->block_start = EXTENT_MAP_HOLE;
5233                         goto insert;
5234                 }
5235                 if (compress_type != BTRFS_COMPRESS_NONE) {
5236                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5237                         em->compress_type = compress_type;
5238                         em->block_start = bytenr;
5239                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5240                                                                          item);
5241                 } else {
5242                         bytenr += btrfs_file_extent_offset(leaf, item);
5243                         em->block_start = bytenr;
5244                         em->block_len = em->len;
5245                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5246                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5247                 }
5248                 goto insert;
5249         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5250                 unsigned long ptr;
5251                 char *map;
5252                 size_t size;
5253                 size_t extent_offset;
5254                 size_t copy_size;
5255
5256                 em->block_start = EXTENT_MAP_INLINE;
5257                 if (!page || create) {
5258                         em->start = extent_start;
5259                         em->len = extent_end - extent_start;
5260                         goto out;
5261                 }
5262
5263                 size = btrfs_file_extent_inline_len(leaf, item);
5264                 extent_offset = page_offset(page) + pg_offset - extent_start;
5265                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5266                                 size - extent_offset);
5267                 em->start = extent_start + extent_offset;
5268                 em->len = (copy_size + root->sectorsize - 1) &
5269                         ~((u64)root->sectorsize - 1);
5270                 em->orig_start = EXTENT_MAP_INLINE;
5271                 if (compress_type) {
5272                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5273                         em->compress_type = compress_type;
5274                 }
5275                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5276                 if (create == 0 && !PageUptodate(page)) {
5277                         if (btrfs_file_extent_compression(leaf, item) !=
5278                             BTRFS_COMPRESS_NONE) {
5279                                 ret = uncompress_inline(path, inode, page,
5280                                                         pg_offset,
5281                                                         extent_offset, item);
5282                                 BUG_ON(ret); /* -ENOMEM */
5283                         } else {
5284                                 map = kmap(page);
5285                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5286                                                    copy_size);
5287                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5288                                         memset(map + pg_offset + copy_size, 0,
5289                                                PAGE_CACHE_SIZE - pg_offset -
5290                                                copy_size);
5291                                 }
5292                                 kunmap(page);
5293                         }
5294                         flush_dcache_page(page);
5295                 } else if (create && PageUptodate(page)) {
5296                         BUG();
5297                         if (!trans) {
5298                                 kunmap(page);
5299                                 free_extent_map(em);
5300                                 em = NULL;
5301
5302                                 btrfs_release_path(path);
5303                                 trans = btrfs_join_transaction(root);
5304
5305                                 if (IS_ERR(trans))
5306                                         return ERR_CAST(trans);
5307                                 goto again;
5308                         }
5309                         map = kmap(page);
5310                         write_extent_buffer(leaf, map + pg_offset, ptr,
5311                                             copy_size);
5312                         kunmap(page);
5313                         btrfs_mark_buffer_dirty(leaf);
5314                 }
5315                 set_extent_uptodate(io_tree, em->start,
5316                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5317                 goto insert;
5318         } else {
5319                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5320                 WARN_ON(1);
5321         }
5322 not_found:
5323         em->start = start;
5324         em->len = len;
5325 not_found_em:
5326         em->block_start = EXTENT_MAP_HOLE;
5327         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5328 insert:
5329         btrfs_release_path(path);
5330         if (em->start > start || extent_map_end(em) <= start) {
5331                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5332                        "[%llu %llu]\n", (unsigned long long)em->start,
5333                        (unsigned long long)em->len,
5334                        (unsigned long long)start,
5335                        (unsigned long long)len);
5336                 err = -EIO;
5337                 goto out;
5338         }
5339
5340         err = 0;
5341         write_lock(&em_tree->lock);
5342         ret = add_extent_mapping(em_tree, em);
5343         /* it is possible that someone inserted the extent into the tree
5344          * while we had the lock dropped.  It is also possible that
5345          * an overlapping map exists in the tree
5346          */
5347         if (ret == -EEXIST) {
5348                 struct extent_map *existing;
5349
5350                 ret = 0;
5351
5352                 existing = lookup_extent_mapping(em_tree, start, len);
5353                 if (existing && (existing->start > start ||
5354                     existing->start + existing->len <= start)) {
5355                         free_extent_map(existing);
5356                         existing = NULL;
5357                 }
5358                 if (!existing) {
5359                         existing = lookup_extent_mapping(em_tree, em->start,
5360                                                          em->len);
5361                         if (existing) {
5362                                 err = merge_extent_mapping(em_tree, existing,
5363                                                            em, start,
5364                                                            root->sectorsize);
5365                                 free_extent_map(existing);
5366                                 if (err) {
5367                                         free_extent_map(em);
5368                                         em = NULL;
5369                                 }
5370                         } else {
5371                                 err = -EIO;
5372                                 free_extent_map(em);
5373                                 em = NULL;
5374                         }
5375                 } else {
5376                         free_extent_map(em);
5377                         em = existing;
5378                         err = 0;
5379                 }
5380         }
5381         write_unlock(&em_tree->lock);
5382 out:
5383
5384         trace_btrfs_get_extent(root, em);
5385
5386         if (path)
5387                 btrfs_free_path(path);
5388         if (trans) {
5389                 ret = btrfs_end_transaction(trans, root);
5390                 if (!err)
5391                         err = ret;
5392         }
5393         if (err) {
5394                 free_extent_map(em);
5395                 return ERR_PTR(err);
5396         }
5397         BUG_ON(!em); /* Error is always set */
5398         return em;
5399 }
5400
5401 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5402                                            size_t pg_offset, u64 start, u64 len,
5403                                            int create)
5404 {
5405         struct extent_map *em;
5406         struct extent_map *hole_em = NULL;
5407         u64 range_start = start;
5408         u64 end;
5409         u64 found;
5410         u64 found_end;
5411         int err = 0;
5412
5413         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5414         if (IS_ERR(em))
5415                 return em;
5416         if (em) {
5417                 /*
5418                  * if our em maps to a hole, there might
5419                  * actually be delalloc bytes behind it
5420                  */
5421                 if (em->block_start != EXTENT_MAP_HOLE)
5422                         return em;
5423                 else
5424                         hole_em = em;
5425         }
5426
5427         /* check to see if we've wrapped (len == -1 or similar) */
5428         end = start + len;
5429         if (end < start)
5430                 end = (u64)-1;
5431         else
5432                 end -= 1;
5433
5434         em = NULL;
5435
5436         /* ok, we didn't find anything, lets look for delalloc */
5437         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5438                                  end, len, EXTENT_DELALLOC, 1);
5439         found_end = range_start + found;
5440         if (found_end < range_start)
5441                 found_end = (u64)-1;
5442
5443         /*
5444          * we didn't find anything useful, return
5445          * the original results from get_extent()
5446          */
5447         if (range_start > end || found_end <= start) {
5448                 em = hole_em;
5449                 hole_em = NULL;
5450                 goto out;
5451         }
5452
5453         /* adjust the range_start to make sure it doesn't
5454          * go backwards from the start they passed in
5455          */
5456         range_start = max(start,range_start);
5457         found = found_end - range_start;
5458
5459         if (found > 0) {
5460                 u64 hole_start = start;
5461                 u64 hole_len = len;
5462
5463                 em = alloc_extent_map();
5464                 if (!em) {
5465                         err = -ENOMEM;
5466                         goto out;
5467                 }
5468                 /*
5469                  * when btrfs_get_extent can't find anything it
5470                  * returns one huge hole
5471                  *
5472                  * make sure what it found really fits our range, and
5473                  * adjust to make sure it is based on the start from
5474                  * the caller
5475                  */
5476                 if (hole_em) {
5477                         u64 calc_end = extent_map_end(hole_em);
5478
5479                         if (calc_end <= start || (hole_em->start > end)) {
5480                                 free_extent_map(hole_em);
5481                                 hole_em = NULL;
5482                         } else {
5483                                 hole_start = max(hole_em->start, start);
5484                                 hole_len = calc_end - hole_start;
5485                         }
5486                 }
5487                 em->bdev = NULL;
5488                 if (hole_em && range_start > hole_start) {
5489                         /* our hole starts before our delalloc, so we
5490                          * have to return just the parts of the hole
5491                          * that go until  the delalloc starts
5492                          */
5493                         em->len = min(hole_len,
5494                                       range_start - hole_start);
5495                         em->start = hole_start;
5496                         em->orig_start = hole_start;
5497                         /*
5498                          * don't adjust block start at all,
5499                          * it is fixed at EXTENT_MAP_HOLE
5500                          */
5501                         em->block_start = hole_em->block_start;
5502                         em->block_len = hole_len;
5503                 } else {
5504                         em->start = range_start;
5505                         em->len = found;
5506                         em->orig_start = range_start;
5507                         em->block_start = EXTENT_MAP_DELALLOC;
5508                         em->block_len = found;
5509                 }
5510         } else if (hole_em) {
5511                 return hole_em;
5512         }
5513 out:
5514
5515         free_extent_map(hole_em);
5516         if (err) {
5517                 free_extent_map(em);
5518                 return ERR_PTR(err);
5519         }
5520         return em;
5521 }
5522
5523 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5524                                                   struct extent_map *em,
5525                                                   u64 start, u64 len)
5526 {
5527         struct btrfs_root *root = BTRFS_I(inode)->root;
5528         struct btrfs_trans_handle *trans;
5529         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5530         struct btrfs_key ins;
5531         u64 alloc_hint;
5532         int ret;
5533         bool insert = false;
5534
5535         /*
5536          * Ok if the extent map we looked up is a hole and is for the exact
5537          * range we want, there is no reason to allocate a new one, however if
5538          * it is not right then we need to free this one and drop the cache for
5539          * our range.
5540          */
5541         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5542             em->len != len) {
5543                 free_extent_map(em);
5544                 em = NULL;
5545                 insert = true;
5546                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5547         }
5548
5549         trans = btrfs_join_transaction(root);
5550         if (IS_ERR(trans))
5551                 return ERR_CAST(trans);
5552
5553         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5554                 btrfs_add_inode_defrag(trans, inode);
5555
5556         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5557
5558         alloc_hint = get_extent_allocation_hint(inode, start, len);
5559         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5560                                    alloc_hint, &ins, 1);
5561         if (ret) {
5562                 em = ERR_PTR(ret);
5563                 goto out;
5564         }
5565
5566         if (!em) {
5567                 em = alloc_extent_map();
5568                 if (!em) {
5569                         em = ERR_PTR(-ENOMEM);
5570                         goto out;
5571                 }
5572         }
5573
5574         em->start = start;
5575         em->orig_start = em->start;
5576         em->len = ins.offset;
5577
5578         em->block_start = ins.objectid;
5579         em->block_len = ins.offset;
5580         em->bdev = root->fs_info->fs_devices->latest_bdev;
5581
5582         /*
5583          * We need to do this because if we're using the original em we searched
5584          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5585          */
5586         em->flags = 0;
5587         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5588
5589         while (insert) {
5590                 write_lock(&em_tree->lock);
5591                 ret = add_extent_mapping(em_tree, em);
5592                 write_unlock(&em_tree->lock);
5593                 if (ret != -EEXIST)
5594                         break;
5595                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5596         }
5597
5598         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5599                                            ins.offset, ins.offset, 0);
5600         if (ret) {
5601                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5602                 em = ERR_PTR(ret);
5603         }
5604 out:
5605         btrfs_end_transaction(trans, root);
5606         return em;
5607 }
5608
5609 /*
5610  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5611  * block must be cow'd
5612  */
5613 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5614                                       struct inode *inode, u64 offset, u64 len)
5615 {
5616         struct btrfs_path *path;
5617         int ret;
5618         struct extent_buffer *leaf;
5619         struct btrfs_root *root = BTRFS_I(inode)->root;
5620         struct btrfs_file_extent_item *fi;
5621         struct btrfs_key key;
5622         u64 disk_bytenr;
5623         u64 backref_offset;
5624         u64 extent_end;
5625         u64 num_bytes;
5626         int slot;
5627         int found_type;
5628
5629         path = btrfs_alloc_path();
5630         if (!path)
5631                 return -ENOMEM;
5632
5633         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5634                                        offset, 0);
5635         if (ret < 0)
5636                 goto out;
5637
5638         slot = path->slots[0];
5639         if (ret == 1) {
5640                 if (slot == 0) {
5641                         /* can't find the item, must cow */
5642                         ret = 0;
5643                         goto out;
5644                 }
5645                 slot--;
5646         }
5647         ret = 0;
5648         leaf = path->nodes[0];
5649         btrfs_item_key_to_cpu(leaf, &key, slot);
5650         if (key.objectid != btrfs_ino(inode) ||
5651             key.type != BTRFS_EXTENT_DATA_KEY) {
5652                 /* not our file or wrong item type, must cow */
5653                 goto out;
5654         }
5655
5656         if (key.offset > offset) {
5657                 /* Wrong offset, must cow */
5658                 goto out;
5659         }
5660
5661         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5662         found_type = btrfs_file_extent_type(leaf, fi);
5663         if (found_type != BTRFS_FILE_EXTENT_REG &&
5664             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5665                 /* not a regular extent, must cow */
5666                 goto out;
5667         }
5668         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5669         backref_offset = btrfs_file_extent_offset(leaf, fi);
5670
5671         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5672         if (extent_end < offset + len) {
5673                 /* extent doesn't include our full range, must cow */
5674                 goto out;
5675         }
5676
5677         if (btrfs_extent_readonly(root, disk_bytenr))
5678                 goto out;
5679
5680         /*
5681          * look for other files referencing this extent, if we
5682          * find any we must cow
5683          */
5684         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5685                                   key.offset - backref_offset, disk_bytenr))
5686                 goto out;
5687
5688         /*
5689          * adjust disk_bytenr and num_bytes to cover just the bytes
5690          * in this extent we are about to write.  If there
5691          * are any csums in that range we have to cow in order
5692          * to keep the csums correct
5693          */
5694         disk_bytenr += backref_offset;
5695         disk_bytenr += offset - key.offset;
5696         num_bytes = min(offset + len, extent_end) - offset;
5697         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5698                                 goto out;
5699         /*
5700          * all of the above have passed, it is safe to overwrite this extent
5701          * without cow
5702          */
5703         ret = 1;
5704 out:
5705         btrfs_free_path(path);
5706         return ret;
5707 }
5708
5709 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5710                                    struct buffer_head *bh_result, int create)
5711 {
5712         struct extent_map *em;
5713         struct btrfs_root *root = BTRFS_I(inode)->root;
5714         u64 start = iblock << inode->i_blkbits;
5715         u64 len = bh_result->b_size;
5716         struct btrfs_trans_handle *trans;
5717
5718         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5719         if (IS_ERR(em))
5720                 return PTR_ERR(em);
5721
5722         /*
5723          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5724          * io.  INLINE is special, and we could probably kludge it in here, but
5725          * it's still buffered so for safety lets just fall back to the generic
5726          * buffered path.
5727          *
5728          * For COMPRESSED we _have_ to read the entire extent in so we can
5729          * decompress it, so there will be buffering required no matter what we
5730          * do, so go ahead and fallback to buffered.
5731          *
5732          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5733          * to buffered IO.  Don't blame me, this is the price we pay for using
5734          * the generic code.
5735          */
5736         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5737             em->block_start == EXTENT_MAP_INLINE) {
5738                 free_extent_map(em);
5739                 return -ENOTBLK;
5740         }
5741
5742         /* Just a good old fashioned hole, return */
5743         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5744                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5745                 free_extent_map(em);
5746                 /* DIO will do one hole at a time, so just unlock a sector */
5747                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5748                               start + root->sectorsize - 1);
5749                 return 0;
5750         }
5751
5752         /*
5753          * We don't allocate a new extent in the following cases
5754          *
5755          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5756          * existing extent.
5757          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5758          * just use the extent.
5759          *
5760          */
5761         if (!create) {
5762                 len = em->len - (start - em->start);
5763                 goto map;
5764         }
5765
5766         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5767             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5768              em->block_start != EXTENT_MAP_HOLE)) {
5769                 int type;
5770                 int ret;
5771                 u64 block_start;
5772
5773                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5774                         type = BTRFS_ORDERED_PREALLOC;
5775                 else
5776                         type = BTRFS_ORDERED_NOCOW;
5777                 len = min(len, em->len - (start - em->start));
5778                 block_start = em->block_start + (start - em->start);
5779
5780                 /*
5781                  * we're not going to log anything, but we do need
5782                  * to make sure the current transaction stays open
5783                  * while we look for nocow cross refs
5784                  */
5785                 trans = btrfs_join_transaction(root);
5786                 if (IS_ERR(trans))
5787                         goto must_cow;
5788
5789                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5790                         ret = btrfs_add_ordered_extent_dio(inode, start,
5791                                            block_start, len, len, type);
5792                         btrfs_end_transaction(trans, root);
5793                         if (ret) {
5794                                 free_extent_map(em);
5795                                 return ret;
5796                         }
5797                         goto unlock;
5798                 }
5799                 btrfs_end_transaction(trans, root);
5800         }
5801 must_cow:
5802         /*
5803          * this will cow the extent, reset the len in case we changed
5804          * it above
5805          */
5806         len = bh_result->b_size;
5807         em = btrfs_new_extent_direct(inode, em, start, len);
5808         if (IS_ERR(em))
5809                 return PTR_ERR(em);
5810         len = min(len, em->len - (start - em->start));
5811 unlock:
5812         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5813                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5814                           0, NULL, GFP_NOFS);
5815 map:
5816         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5817                 inode->i_blkbits;
5818         bh_result->b_size = len;
5819         bh_result->b_bdev = em->bdev;
5820         set_buffer_mapped(bh_result);
5821         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5822                 set_buffer_new(bh_result);
5823
5824         free_extent_map(em);
5825
5826         return 0;
5827 }
5828
5829 struct btrfs_dio_private {
5830         struct inode *inode;
5831         u64 logical_offset;
5832         u64 disk_bytenr;
5833         u64 bytes;
5834         u32 *csums;
5835         void *private;
5836
5837         /* number of bios pending for this dio */
5838         atomic_t pending_bios;
5839
5840         /* IO errors */
5841         int errors;
5842
5843         struct bio *orig_bio;
5844 };
5845
5846 static void btrfs_endio_direct_read(struct bio *bio, int err)
5847 {
5848         struct btrfs_dio_private *dip = bio->bi_private;
5849         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5850         struct bio_vec *bvec = bio->bi_io_vec;
5851         struct inode *inode = dip->inode;
5852         struct btrfs_root *root = BTRFS_I(inode)->root;
5853         u64 start;
5854         u32 *private = dip->csums;
5855
5856         start = dip->logical_offset;
5857         do {
5858                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5859                         struct page *page = bvec->bv_page;
5860                         char *kaddr;
5861                         u32 csum = ~(u32)0;
5862                         unsigned long flags;
5863
5864                         local_irq_save(flags);
5865                         kaddr = kmap_atomic(page);
5866                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5867                                                csum, bvec->bv_len);
5868                         btrfs_csum_final(csum, (char *)&csum);
5869                         kunmap_atomic(kaddr);
5870                         local_irq_restore(flags);
5871
5872                         flush_dcache_page(bvec->bv_page);
5873                         if (csum != *private) {
5874                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5875                                       " %llu csum %u private %u\n",
5876                                       (unsigned long long)btrfs_ino(inode),
5877                                       (unsigned long long)start,
5878                                       csum, *private);
5879                                 err = -EIO;
5880                         }
5881                 }
5882
5883                 start += bvec->bv_len;
5884                 private++;
5885                 bvec++;
5886         } while (bvec <= bvec_end);
5887
5888         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5889                       dip->logical_offset + dip->bytes - 1);
5890         bio->bi_private = dip->private;
5891
5892         kfree(dip->csums);
5893         kfree(dip);
5894
5895         /* If we had a csum failure make sure to clear the uptodate flag */
5896         if (err)
5897                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5898         dio_end_io(bio, err);
5899 }
5900
5901 static void btrfs_endio_direct_write(struct bio *bio, int err)
5902 {
5903         struct btrfs_dio_private *dip = bio->bi_private;
5904         struct inode *inode = dip->inode;
5905         struct btrfs_root *root = BTRFS_I(inode)->root;
5906         struct btrfs_trans_handle *trans;
5907         struct btrfs_ordered_extent *ordered = NULL;
5908         struct extent_state *cached_state = NULL;
5909         u64 ordered_offset = dip->logical_offset;
5910         u64 ordered_bytes = dip->bytes;
5911         int ret;
5912
5913         if (err)
5914                 goto out_done;
5915 again:
5916         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5917                                                    &ordered_offset,
5918                                                    ordered_bytes);
5919         if (!ret)
5920                 goto out_test;
5921
5922         BUG_ON(!ordered);
5923
5924         trans = btrfs_join_transaction(root);
5925         if (IS_ERR(trans)) {
5926                 err = -ENOMEM;
5927                 goto out;
5928         }
5929         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5930
5931         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5932                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5933                 if (!ret)
5934                         err = btrfs_update_inode_fallback(trans, root, inode);
5935                 goto out;
5936         }
5937
5938         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5939                          ordered->file_offset + ordered->len - 1, 0,
5940                          &cached_state);
5941
5942         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5943                 ret = btrfs_mark_extent_written(trans, inode,
5944                                                 ordered->file_offset,
5945                                                 ordered->file_offset +
5946                                                 ordered->len);
5947                 if (ret) {
5948                         err = ret;
5949                         goto out_unlock;
5950                 }
5951         } else {
5952                 ret = insert_reserved_file_extent(trans, inode,
5953                                                   ordered->file_offset,
5954                                                   ordered->start,
5955                                                   ordered->disk_len,
5956                                                   ordered->len,
5957                                                   ordered->len,
5958                                                   0, 0, 0,
5959                                                   BTRFS_FILE_EXTENT_REG);
5960                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5961                                    ordered->file_offset, ordered->len);
5962                 if (ret) {
5963                         err = ret;
5964                         WARN_ON(1);
5965                         goto out_unlock;
5966                 }
5967         }
5968
5969         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5970         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5971         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5972                 btrfs_update_inode_fallback(trans, root, inode);
5973         ret = 0;
5974 out_unlock:
5975         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5976                              ordered->file_offset + ordered->len - 1,
5977                              &cached_state, GFP_NOFS);
5978 out:
5979         btrfs_delalloc_release_metadata(inode, ordered->len);
5980         btrfs_end_transaction(trans, root);
5981         ordered_offset = ordered->file_offset + ordered->len;
5982         btrfs_put_ordered_extent(ordered);
5983         btrfs_put_ordered_extent(ordered);
5984
5985 out_test:
5986         /*
5987          * our bio might span multiple ordered extents.  If we haven't
5988          * completed the accounting for the whole dio, go back and try again
5989          */
5990         if (ordered_offset < dip->logical_offset + dip->bytes) {
5991                 ordered_bytes = dip->logical_offset + dip->bytes -
5992                         ordered_offset;
5993                 goto again;
5994         }
5995 out_done:
5996         bio->bi_private = dip->private;
5997
5998         kfree(dip->csums);
5999         kfree(dip);
6000
6001         /* If we had an error make sure to clear the uptodate flag */
6002         if (err)
6003                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6004         dio_end_io(bio, err);
6005 }
6006
6007 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6008                                     struct bio *bio, int mirror_num,
6009                                     unsigned long bio_flags, u64 offset)
6010 {
6011         int ret;
6012         struct btrfs_root *root = BTRFS_I(inode)->root;
6013         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6014         BUG_ON(ret); /* -ENOMEM */
6015         return 0;
6016 }
6017
6018 static void btrfs_end_dio_bio(struct bio *bio, int err)
6019 {
6020         struct btrfs_dio_private *dip = bio->bi_private;
6021
6022         if (err) {
6023                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6024                       "sector %#Lx len %u err no %d\n",
6025                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6026                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6027                 dip->errors = 1;
6028
6029                 /*
6030                  * before atomic variable goto zero, we must make sure
6031                  * dip->errors is perceived to be set.
6032                  */
6033                 smp_mb__before_atomic_dec();
6034         }
6035
6036         /* if there are more bios still pending for this dio, just exit */
6037         if (!atomic_dec_and_test(&dip->pending_bios))
6038                 goto out;
6039
6040         if (dip->errors)
6041                 bio_io_error(dip->orig_bio);
6042         else {
6043                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6044                 bio_endio(dip->orig_bio, 0);
6045         }
6046 out:
6047         bio_put(bio);
6048 }
6049
6050 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6051                                        u64 first_sector, gfp_t gfp_flags)
6052 {
6053         int nr_vecs = bio_get_nr_vecs(bdev);
6054         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6055 }
6056
6057 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6058                                          int rw, u64 file_offset, int skip_sum,
6059                                          u32 *csums, int async_submit)
6060 {
6061         int write = rw & REQ_WRITE;
6062         struct btrfs_root *root = BTRFS_I(inode)->root;
6063         int ret;
6064
6065         bio_get(bio);
6066         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6067         if (ret)
6068                 goto err;
6069
6070         if (skip_sum)
6071                 goto map;
6072
6073         if (write && async_submit) {
6074                 ret = btrfs_wq_submit_bio(root->fs_info,
6075                                    inode, rw, bio, 0, 0,
6076                                    file_offset,
6077                                    __btrfs_submit_bio_start_direct_io,
6078                                    __btrfs_submit_bio_done);
6079                 goto err;
6080         } else if (write) {
6081                 /*
6082                  * If we aren't doing async submit, calculate the csum of the
6083                  * bio now.
6084                  */
6085                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6086                 if (ret)
6087                         goto err;
6088         } else if (!skip_sum) {
6089                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
6090                                           file_offset, csums);
6091                 if (ret)
6092                         goto err;
6093         }
6094
6095 map:
6096         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6097 err:
6098         bio_put(bio);
6099         return ret;
6100 }
6101
6102 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6103                                     int skip_sum)
6104 {
6105         struct inode *inode = dip->inode;
6106         struct btrfs_root *root = BTRFS_I(inode)->root;
6107         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
6108         struct bio *bio;
6109         struct bio *orig_bio = dip->orig_bio;
6110         struct bio_vec *bvec = orig_bio->bi_io_vec;
6111         u64 start_sector = orig_bio->bi_sector;
6112         u64 file_offset = dip->logical_offset;
6113         u64 submit_len = 0;
6114         u64 map_length;
6115         int nr_pages = 0;
6116         u32 *csums = dip->csums;
6117         int ret = 0;
6118         int async_submit = 0;
6119         int write = rw & REQ_WRITE;
6120
6121         map_length = orig_bio->bi_size;
6122         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6123                               &map_length, NULL, 0);
6124         if (ret) {
6125                 bio_put(orig_bio);
6126                 return -EIO;
6127         }
6128
6129         if (map_length >= orig_bio->bi_size) {
6130                 bio = orig_bio;
6131                 goto submit;
6132         }
6133
6134         async_submit = 1;
6135         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6136         if (!bio)
6137                 return -ENOMEM;
6138         bio->bi_private = dip;
6139         bio->bi_end_io = btrfs_end_dio_bio;
6140         atomic_inc(&dip->pending_bios);
6141
6142         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6143                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6144                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6145                                  bvec->bv_offset) < bvec->bv_len)) {
6146                         /*
6147                          * inc the count before we submit the bio so
6148                          * we know the end IO handler won't happen before
6149                          * we inc the count. Otherwise, the dip might get freed
6150                          * before we're done setting it up
6151                          */
6152                         atomic_inc(&dip->pending_bios);
6153                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6154                                                      file_offset, skip_sum,
6155                                                      csums, async_submit);
6156                         if (ret) {
6157                                 bio_put(bio);
6158                                 atomic_dec(&dip->pending_bios);
6159                                 goto out_err;
6160                         }
6161
6162                         /* Write's use the ordered csums */
6163                         if (!write && !skip_sum)
6164                                 csums = csums + nr_pages;
6165                         start_sector += submit_len >> 9;
6166                         file_offset += submit_len;
6167
6168                         submit_len = 0;
6169                         nr_pages = 0;
6170
6171                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6172                                                   start_sector, GFP_NOFS);
6173                         if (!bio)
6174                                 goto out_err;
6175                         bio->bi_private = dip;
6176                         bio->bi_end_io = btrfs_end_dio_bio;
6177
6178                         map_length = orig_bio->bi_size;
6179                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6180                                               &map_length, NULL, 0);
6181                         if (ret) {
6182                                 bio_put(bio);
6183                                 goto out_err;
6184                         }
6185                 } else {
6186                         submit_len += bvec->bv_len;
6187                         nr_pages ++;
6188                         bvec++;
6189                 }
6190         }
6191
6192 submit:
6193         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6194                                      csums, async_submit);
6195         if (!ret)
6196                 return 0;
6197
6198         bio_put(bio);
6199 out_err:
6200         dip->errors = 1;
6201         /*
6202          * before atomic variable goto zero, we must
6203          * make sure dip->errors is perceived to be set.
6204          */
6205         smp_mb__before_atomic_dec();
6206         if (atomic_dec_and_test(&dip->pending_bios))
6207                 bio_io_error(dip->orig_bio);
6208
6209         /* bio_end_io() will handle error, so we needn't return it */
6210         return 0;
6211 }
6212
6213 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6214                                 loff_t file_offset)
6215 {
6216         struct btrfs_root *root = BTRFS_I(inode)->root;
6217         struct btrfs_dio_private *dip;
6218         struct bio_vec *bvec = bio->bi_io_vec;
6219         int skip_sum;
6220         int write = rw & REQ_WRITE;
6221         int ret = 0;
6222
6223         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6224
6225         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6226         if (!dip) {
6227                 ret = -ENOMEM;
6228                 goto free_ordered;
6229         }
6230         dip->csums = NULL;
6231
6232         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6233         if (!write && !skip_sum) {
6234                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6235                 if (!dip->csums) {
6236                         kfree(dip);
6237                         ret = -ENOMEM;
6238                         goto free_ordered;
6239                 }
6240         }
6241
6242         dip->private = bio->bi_private;
6243         dip->inode = inode;
6244         dip->logical_offset = file_offset;
6245
6246         dip->bytes = 0;
6247         do {
6248                 dip->bytes += bvec->bv_len;
6249                 bvec++;
6250         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6251
6252         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6253         bio->bi_private = dip;
6254         dip->errors = 0;
6255         dip->orig_bio = bio;
6256         atomic_set(&dip->pending_bios, 0);
6257
6258         if (write)
6259                 bio->bi_end_io = btrfs_endio_direct_write;
6260         else
6261                 bio->bi_end_io = btrfs_endio_direct_read;
6262
6263         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6264         if (!ret)
6265                 return;
6266 free_ordered:
6267         /*
6268          * If this is a write, we need to clean up the reserved space and kill
6269          * the ordered extent.
6270          */
6271         if (write) {
6272                 struct btrfs_ordered_extent *ordered;
6273                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6274                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6275                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6276                         btrfs_free_reserved_extent(root, ordered->start,
6277                                                    ordered->disk_len);
6278                 btrfs_put_ordered_extent(ordered);
6279                 btrfs_put_ordered_extent(ordered);
6280         }
6281         bio_endio(bio, ret);
6282 }
6283
6284 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6285                         const struct iovec *iov, loff_t offset,
6286                         unsigned long nr_segs)
6287 {
6288         int seg;
6289         int i;
6290         size_t size;
6291         unsigned long addr;
6292         unsigned blocksize_mask = root->sectorsize - 1;
6293         ssize_t retval = -EINVAL;
6294         loff_t end = offset;
6295
6296         if (offset & blocksize_mask)
6297                 goto out;
6298
6299         /* Check the memory alignment.  Blocks cannot straddle pages */
6300         for (seg = 0; seg < nr_segs; seg++) {
6301                 addr = (unsigned long)iov[seg].iov_base;
6302                 size = iov[seg].iov_len;
6303                 end += size;
6304                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6305                         goto out;
6306
6307                 /* If this is a write we don't need to check anymore */
6308                 if (rw & WRITE)
6309                         continue;
6310
6311                 /*
6312                  * Check to make sure we don't have duplicate iov_base's in this
6313                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6314                  * when reading back.
6315                  */
6316                 for (i = seg + 1; i < nr_segs; i++) {
6317                         if (iov[seg].iov_base == iov[i].iov_base)
6318                                 goto out;
6319                 }
6320         }
6321         retval = 0;
6322 out:
6323         return retval;
6324 }
6325 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6326                         const struct iovec *iov, loff_t offset,
6327                         unsigned long nr_segs)
6328 {
6329         struct file *file = iocb->ki_filp;
6330         struct inode *inode = file->f_mapping->host;
6331         struct btrfs_ordered_extent *ordered;
6332         struct extent_state *cached_state = NULL;
6333         u64 lockstart, lockend;
6334         ssize_t ret;
6335         int writing = rw & WRITE;
6336         int write_bits = 0;
6337         size_t count = iov_length(iov, nr_segs);
6338
6339         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6340                             offset, nr_segs)) {
6341                 return 0;
6342         }
6343
6344         lockstart = offset;
6345         lockend = offset + count - 1;
6346
6347         if (writing) {
6348                 ret = btrfs_delalloc_reserve_space(inode, count);
6349                 if (ret)
6350                         goto out;
6351         }
6352
6353         while (1) {
6354                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6355                                  0, &cached_state);
6356                 /*
6357                  * We're concerned with the entire range that we're going to be
6358                  * doing DIO to, so we need to make sure theres no ordered
6359                  * extents in this range.
6360                  */
6361                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6362                                                      lockend - lockstart + 1);
6363                 if (!ordered)
6364                         break;
6365                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6366                                      &cached_state, GFP_NOFS);
6367                 btrfs_start_ordered_extent(inode, ordered, 1);
6368                 btrfs_put_ordered_extent(ordered);
6369                 cond_resched();
6370         }
6371
6372         /*
6373          * we don't use btrfs_set_extent_delalloc because we don't want
6374          * the dirty or uptodate bits
6375          */
6376         if (writing) {
6377                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6378                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6379                                      EXTENT_DELALLOC, NULL, &cached_state,
6380                                      GFP_NOFS);
6381                 if (ret) {
6382                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6383                                          lockend, EXTENT_LOCKED | write_bits,
6384                                          1, 0, &cached_state, GFP_NOFS);
6385                         goto out;
6386                 }
6387         }
6388
6389         free_extent_state(cached_state);
6390         cached_state = NULL;
6391
6392         ret = __blockdev_direct_IO(rw, iocb, inode,
6393                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6394                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6395                    btrfs_submit_direct, 0);
6396
6397         if (ret < 0 && ret != -EIOCBQUEUED) {
6398                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6399                               offset + iov_length(iov, nr_segs) - 1,
6400                               EXTENT_LOCKED | write_bits, 1, 0,
6401                               &cached_state, GFP_NOFS);
6402         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6403                 /*
6404                  * We're falling back to buffered, unlock the section we didn't
6405                  * do IO on.
6406                  */
6407                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6408                               offset + iov_length(iov, nr_segs) - 1,
6409                               EXTENT_LOCKED | write_bits, 1, 0,
6410                               &cached_state, GFP_NOFS);
6411         }
6412 out:
6413         free_extent_state(cached_state);
6414         return ret;
6415 }
6416
6417 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6418                 __u64 start, __u64 len)
6419 {
6420         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6421 }
6422
6423 int btrfs_readpage(struct file *file, struct page *page)
6424 {
6425         struct extent_io_tree *tree;
6426         tree = &BTRFS_I(page->mapping->host)->io_tree;
6427         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6428 }
6429
6430 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6431 {
6432         struct extent_io_tree *tree;
6433
6434
6435         if (current->flags & PF_MEMALLOC) {
6436                 redirty_page_for_writepage(wbc, page);
6437                 unlock_page(page);
6438                 return 0;
6439         }
6440         tree = &BTRFS_I(page->mapping->host)->io_tree;
6441         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6442 }
6443
6444 int btrfs_writepages(struct address_space *mapping,
6445                      struct writeback_control *wbc)
6446 {
6447         struct extent_io_tree *tree;
6448
6449         tree = &BTRFS_I(mapping->host)->io_tree;
6450         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6451 }
6452
6453 static int
6454 btrfs_readpages(struct file *file, struct address_space *mapping,
6455                 struct list_head *pages, unsigned nr_pages)
6456 {
6457         struct extent_io_tree *tree;
6458         tree = &BTRFS_I(mapping->host)->io_tree;
6459         return extent_readpages(tree, mapping, pages, nr_pages,
6460                                 btrfs_get_extent);
6461 }
6462 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6463 {
6464         struct extent_io_tree *tree;
6465         struct extent_map_tree *map;
6466         int ret;
6467
6468         tree = &BTRFS_I(page->mapping->host)->io_tree;
6469         map = &BTRFS_I(page->mapping->host)->extent_tree;
6470         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6471         if (ret == 1) {
6472                 ClearPagePrivate(page);
6473                 set_page_private(page, 0);
6474                 page_cache_release(page);
6475         }
6476         return ret;
6477 }
6478
6479 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6480 {
6481         if (PageWriteback(page) || PageDirty(page))
6482                 return 0;
6483         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6484 }
6485
6486 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6487 {
6488         struct extent_io_tree *tree;
6489         struct btrfs_ordered_extent *ordered;
6490         struct extent_state *cached_state = NULL;
6491         u64 page_start = page_offset(page);
6492         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6493
6494
6495         /*
6496          * we have the page locked, so new writeback can't start,
6497          * and the dirty bit won't be cleared while we are here.
6498          *
6499          * Wait for IO on this page so that we can safely clear
6500          * the PagePrivate2 bit and do ordered accounting
6501          */
6502         wait_on_page_writeback(page);
6503
6504         tree = &BTRFS_I(page->mapping->host)->io_tree;
6505         if (offset) {
6506                 btrfs_releasepage(page, GFP_NOFS);
6507                 return;
6508         }
6509         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6510         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6511                                            page_offset(page));
6512         if (ordered) {
6513                 /*
6514                  * IO on this page will never be started, so we need
6515                  * to account for any ordered extents now
6516                  */
6517                 clear_extent_bit(tree, page_start, page_end,
6518                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6519                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6520                                  &cached_state, GFP_NOFS);
6521                 /*
6522                  * whoever cleared the private bit is responsible
6523                  * for the finish_ordered_io
6524                  */
6525                 if (TestClearPagePrivate2(page)) {
6526                         btrfs_finish_ordered_io(page->mapping->host,
6527                                                 page_start, page_end);
6528                 }
6529                 btrfs_put_ordered_extent(ordered);
6530                 cached_state = NULL;
6531                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6532         }
6533         clear_extent_bit(tree, page_start, page_end,
6534                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6535                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6536         __btrfs_releasepage(page, GFP_NOFS);
6537
6538         ClearPageChecked(page);
6539         if (PagePrivate(page)) {
6540                 ClearPagePrivate(page);
6541                 set_page_private(page, 0);
6542                 page_cache_release(page);
6543         }
6544 }
6545
6546 /*
6547  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6548  * called from a page fault handler when a page is first dirtied. Hence we must
6549  * be careful to check for EOF conditions here. We set the page up correctly
6550  * for a written page which means we get ENOSPC checking when writing into
6551  * holes and correct delalloc and unwritten extent mapping on filesystems that
6552  * support these features.
6553  *
6554  * We are not allowed to take the i_mutex here so we have to play games to
6555  * protect against truncate races as the page could now be beyond EOF.  Because
6556  * vmtruncate() writes the inode size before removing pages, once we have the
6557  * page lock we can determine safely if the page is beyond EOF. If it is not
6558  * beyond EOF, then the page is guaranteed safe against truncation until we
6559  * unlock the page.
6560  */
6561 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6562 {
6563         struct page *page = vmf->page;
6564         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6565         struct btrfs_root *root = BTRFS_I(inode)->root;
6566         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6567         struct btrfs_ordered_extent *ordered;
6568         struct extent_state *cached_state = NULL;
6569         char *kaddr;
6570         unsigned long zero_start;
6571         loff_t size;
6572         int ret;
6573         int reserved = 0;
6574         u64 page_start;
6575         u64 page_end;
6576
6577         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6578         if (!ret) {
6579                 ret = btrfs_update_time(vma->vm_file);
6580                 reserved = 1;
6581         }
6582         if (ret) {
6583                 if (ret == -ENOMEM)
6584                         ret = VM_FAULT_OOM;
6585                 else /* -ENOSPC, -EIO, etc */
6586                         ret = VM_FAULT_SIGBUS;
6587                 if (reserved)
6588                         goto out;
6589                 goto out_noreserve;
6590         }
6591
6592         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6593 again:
6594         lock_page(page);
6595         size = i_size_read(inode);
6596         page_start = page_offset(page);
6597         page_end = page_start + PAGE_CACHE_SIZE - 1;
6598
6599         if ((page->mapping != inode->i_mapping) ||
6600             (page_start >= size)) {
6601                 /* page got truncated out from underneath us */
6602                 goto out_unlock;
6603         }
6604         wait_on_page_writeback(page);
6605
6606         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6607         set_page_extent_mapped(page);
6608
6609         /*
6610          * we can't set the delalloc bits if there are pending ordered
6611          * extents.  Drop our locks and wait for them to finish
6612          */
6613         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6614         if (ordered) {
6615                 unlock_extent_cached(io_tree, page_start, page_end,
6616                                      &cached_state, GFP_NOFS);
6617                 unlock_page(page);
6618                 btrfs_start_ordered_extent(inode, ordered, 1);
6619                 btrfs_put_ordered_extent(ordered);
6620                 goto again;
6621         }
6622
6623         /*
6624          * XXX - page_mkwrite gets called every time the page is dirtied, even
6625          * if it was already dirty, so for space accounting reasons we need to
6626          * clear any delalloc bits for the range we are fixing to save.  There
6627          * is probably a better way to do this, but for now keep consistent with
6628          * prepare_pages in the normal write path.
6629          */
6630         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6631                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6632                           0, 0, &cached_state, GFP_NOFS);
6633
6634         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6635                                         &cached_state);
6636         if (ret) {
6637                 unlock_extent_cached(io_tree, page_start, page_end,
6638                                      &cached_state, GFP_NOFS);
6639                 ret = VM_FAULT_SIGBUS;
6640                 goto out_unlock;
6641         }
6642         ret = 0;
6643
6644         /* page is wholly or partially inside EOF */
6645         if (page_start + PAGE_CACHE_SIZE > size)
6646                 zero_start = size & ~PAGE_CACHE_MASK;
6647         else
6648                 zero_start = PAGE_CACHE_SIZE;
6649
6650         if (zero_start != PAGE_CACHE_SIZE) {
6651                 kaddr = kmap(page);
6652                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6653                 flush_dcache_page(page);
6654                 kunmap(page);
6655         }
6656         ClearPageChecked(page);
6657         set_page_dirty(page);
6658         SetPageUptodate(page);
6659
6660         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6661         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6662
6663         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6664
6665 out_unlock:
6666         if (!ret)
6667                 return VM_FAULT_LOCKED;
6668         unlock_page(page);
6669 out:
6670         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6671 out_noreserve:
6672         return ret;
6673 }
6674
6675 static int btrfs_truncate(struct inode *inode)
6676 {
6677         struct btrfs_root *root = BTRFS_I(inode)->root;
6678         struct btrfs_block_rsv *rsv;
6679         int ret;
6680         int err = 0;
6681         struct btrfs_trans_handle *trans;
6682         unsigned long nr;
6683         u64 mask = root->sectorsize - 1;
6684         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6685
6686         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6687         if (ret)
6688                 return ret;
6689
6690         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6691         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6692
6693         /*
6694          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6695          * 3 things going on here
6696          *
6697          * 1) We need to reserve space for our orphan item and the space to
6698          * delete our orphan item.  Lord knows we don't want to have a dangling
6699          * orphan item because we didn't reserve space to remove it.
6700          *
6701          * 2) We need to reserve space to update our inode.
6702          *
6703          * 3) We need to have something to cache all the space that is going to
6704          * be free'd up by the truncate operation, but also have some slack
6705          * space reserved in case it uses space during the truncate (thank you
6706          * very much snapshotting).
6707          *
6708          * And we need these to all be seperate.  The fact is we can use alot of
6709          * space doing the truncate, and we have no earthly idea how much space
6710          * we will use, so we need the truncate reservation to be seperate so it
6711          * doesn't end up using space reserved for updating the inode or
6712          * removing the orphan item.  We also need to be able to stop the
6713          * transaction and start a new one, which means we need to be able to
6714          * update the inode several times, and we have no idea of knowing how
6715          * many times that will be, so we can't just reserve 1 item for the
6716          * entirety of the opration, so that has to be done seperately as well.
6717          * Then there is the orphan item, which does indeed need to be held on
6718          * to for the whole operation, and we need nobody to touch this reserved
6719          * space except the orphan code.
6720          *
6721          * So that leaves us with
6722          *
6723          * 1) root->orphan_block_rsv - for the orphan deletion.
6724          * 2) rsv - for the truncate reservation, which we will steal from the
6725          * transaction reservation.
6726          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6727          * updating the inode.
6728          */
6729         rsv = btrfs_alloc_block_rsv(root);
6730         if (!rsv)
6731                 return -ENOMEM;
6732         rsv->size = min_size;
6733
6734         /*
6735          * 1 for the truncate slack space
6736          * 1 for the orphan item we're going to add
6737          * 1 for the orphan item deletion
6738          * 1 for updating the inode.
6739          */
6740         trans = btrfs_start_transaction(root, 4);
6741         if (IS_ERR(trans)) {
6742                 err = PTR_ERR(trans);
6743                 goto out;
6744         }
6745
6746         /* Migrate the slack space for the truncate to our reserve */
6747         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6748                                       min_size);
6749         BUG_ON(ret);
6750
6751         ret = btrfs_orphan_add(trans, inode);
6752         if (ret) {
6753                 btrfs_end_transaction(trans, root);
6754                 goto out;
6755         }
6756
6757         /*
6758          * setattr is responsible for setting the ordered_data_close flag,
6759          * but that is only tested during the last file release.  That
6760          * could happen well after the next commit, leaving a great big
6761          * window where new writes may get lost if someone chooses to write
6762          * to this file after truncating to zero
6763          *
6764          * The inode doesn't have any dirty data here, and so if we commit
6765          * this is a noop.  If someone immediately starts writing to the inode
6766          * it is very likely we'll catch some of their writes in this
6767          * transaction, and the commit will find this file on the ordered
6768          * data list with good things to send down.
6769          *
6770          * This is a best effort solution, there is still a window where
6771          * using truncate to replace the contents of the file will
6772          * end up with a zero length file after a crash.
6773          */
6774         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6775                 btrfs_add_ordered_operation(trans, root, inode);
6776
6777         while (1) {
6778                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6779                 if (ret) {
6780                         /*
6781                          * This can only happen with the original transaction we
6782                          * started above, every other time we shouldn't have a
6783                          * transaction started yet.
6784                          */
6785                         if (ret == -EAGAIN)
6786                                 goto end_trans;
6787                         err = ret;
6788                         break;
6789                 }
6790
6791                 if (!trans) {
6792                         /* Just need the 1 for updating the inode */
6793                         trans = btrfs_start_transaction(root, 1);
6794                         if (IS_ERR(trans)) {
6795                                 ret = err = PTR_ERR(trans);
6796                                 trans = NULL;
6797                                 break;
6798                         }
6799                 }
6800
6801                 trans->block_rsv = rsv;
6802
6803                 ret = btrfs_truncate_inode_items(trans, root, inode,
6804                                                  inode->i_size,
6805                                                  BTRFS_EXTENT_DATA_KEY);
6806                 if (ret != -EAGAIN) {
6807                         err = ret;
6808                         break;
6809                 }
6810
6811                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6812                 ret = btrfs_update_inode(trans, root, inode);
6813                 if (ret) {
6814                         err = ret;
6815                         break;
6816                 }
6817 end_trans:
6818                 nr = trans->blocks_used;
6819                 btrfs_end_transaction(trans, root);
6820                 trans = NULL;
6821                 btrfs_btree_balance_dirty(root, nr);
6822         }
6823
6824         if (ret == 0 && inode->i_nlink > 0) {
6825                 trans->block_rsv = root->orphan_block_rsv;
6826                 ret = btrfs_orphan_del(trans, inode);
6827                 if (ret)
6828                         err = ret;
6829         } else if (ret && inode->i_nlink > 0) {
6830                 /*
6831                  * Failed to do the truncate, remove us from the in memory
6832                  * orphan list.
6833                  */
6834                 ret = btrfs_orphan_del(NULL, inode);
6835         }
6836
6837         if (trans) {
6838                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6839                 ret = btrfs_update_inode(trans, root, inode);
6840                 if (ret && !err)
6841                         err = ret;
6842
6843                 nr = trans->blocks_used;
6844                 ret = btrfs_end_transaction(trans, root);
6845                 btrfs_btree_balance_dirty(root, nr);
6846         }
6847
6848 out:
6849         btrfs_free_block_rsv(root, rsv);
6850
6851         if (ret && !err)
6852                 err = ret;
6853
6854         return err;
6855 }
6856
6857 /*
6858  * create a new subvolume directory/inode (helper for the ioctl).
6859  */
6860 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6861                              struct btrfs_root *new_root, u64 new_dirid)
6862 {
6863         struct inode *inode;
6864         int err;
6865         u64 index = 0;
6866
6867         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
6868                                 new_dirid, new_dirid,
6869                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
6870                                 &index);
6871         if (IS_ERR(inode))
6872                 return PTR_ERR(inode);
6873         inode->i_op = &btrfs_dir_inode_operations;
6874         inode->i_fop = &btrfs_dir_file_operations;
6875
6876         set_nlink(inode, 1);
6877         btrfs_i_size_write(inode, 0);
6878
6879         err = btrfs_update_inode(trans, new_root, inode);
6880
6881         iput(inode);
6882         return err;
6883 }
6884
6885 struct inode *btrfs_alloc_inode(struct super_block *sb)
6886 {
6887         struct btrfs_inode *ei;
6888         struct inode *inode;
6889
6890         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6891         if (!ei)
6892                 return NULL;
6893
6894         ei->root = NULL;
6895         ei->space_info = NULL;
6896         ei->generation = 0;
6897         ei->sequence = 0;
6898         ei->last_trans = 0;
6899         ei->last_sub_trans = 0;
6900         ei->logged_trans = 0;
6901         ei->delalloc_bytes = 0;
6902         ei->disk_i_size = 0;
6903         ei->flags = 0;
6904         ei->csum_bytes = 0;
6905         ei->index_cnt = (u64)-1;
6906         ei->last_unlink_trans = 0;
6907
6908         spin_lock_init(&ei->lock);
6909         ei->outstanding_extents = 0;
6910         ei->reserved_extents = 0;
6911
6912         ei->ordered_data_close = 0;
6913         ei->orphan_meta_reserved = 0;
6914         ei->dummy_inode = 0;
6915         ei->in_defrag = 0;
6916         ei->delalloc_meta_reserved = 0;
6917         ei->force_compress = BTRFS_COMPRESS_NONE;
6918
6919         ei->delayed_node = NULL;
6920
6921         inode = &ei->vfs_inode;
6922         extent_map_tree_init(&ei->extent_tree);
6923         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6924         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6925         ei->io_tree.track_uptodate = 1;
6926         ei->io_failure_tree.track_uptodate = 1;
6927         mutex_init(&ei->log_mutex);
6928         mutex_init(&ei->delalloc_mutex);
6929         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6930         INIT_LIST_HEAD(&ei->i_orphan);
6931         INIT_LIST_HEAD(&ei->delalloc_inodes);
6932         INIT_LIST_HEAD(&ei->ordered_operations);
6933         RB_CLEAR_NODE(&ei->rb_node);
6934
6935         return inode;
6936 }
6937
6938 static void btrfs_i_callback(struct rcu_head *head)
6939 {
6940         struct inode *inode = container_of(head, struct inode, i_rcu);
6941         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6942 }
6943
6944 void btrfs_destroy_inode(struct inode *inode)
6945 {
6946         struct btrfs_ordered_extent *ordered;
6947         struct btrfs_root *root = BTRFS_I(inode)->root;
6948
6949         WARN_ON(!list_empty(&inode->i_dentry));
6950         WARN_ON(inode->i_data.nrpages);
6951         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6952         WARN_ON(BTRFS_I(inode)->reserved_extents);
6953         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6954         WARN_ON(BTRFS_I(inode)->csum_bytes);
6955
6956         /*
6957          * This can happen where we create an inode, but somebody else also
6958          * created the same inode and we need to destroy the one we already
6959          * created.
6960          */
6961         if (!root)
6962                 goto free;
6963
6964         /*
6965          * Make sure we're properly removed from the ordered operation
6966          * lists.
6967          */
6968         smp_mb();
6969         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6970                 spin_lock(&root->fs_info->ordered_extent_lock);
6971                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6972                 spin_unlock(&root->fs_info->ordered_extent_lock);
6973         }
6974
6975         spin_lock(&root->orphan_lock);
6976         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6977                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6978                        (unsigned long long)btrfs_ino(inode));
6979                 list_del_init(&BTRFS_I(inode)->i_orphan);
6980         }
6981         spin_unlock(&root->orphan_lock);
6982
6983         while (1) {
6984                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6985                 if (!ordered)
6986                         break;
6987                 else {
6988                         printk(KERN_ERR "btrfs found ordered "
6989                                "extent %llu %llu on inode cleanup\n",
6990                                (unsigned long long)ordered->file_offset,
6991                                (unsigned long long)ordered->len);
6992                         btrfs_remove_ordered_extent(inode, ordered);
6993                         btrfs_put_ordered_extent(ordered);
6994                         btrfs_put_ordered_extent(ordered);
6995                 }
6996         }
6997         inode_tree_del(inode);
6998         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6999 free:
7000         btrfs_remove_delayed_node(inode);
7001         call_rcu(&inode->i_rcu, btrfs_i_callback);
7002 }
7003
7004 int btrfs_drop_inode(struct inode *inode)
7005 {
7006         struct btrfs_root *root = BTRFS_I(inode)->root;
7007
7008         if (btrfs_root_refs(&root->root_item) == 0 &&
7009             !btrfs_is_free_space_inode(root, inode))
7010                 return 1;
7011         else
7012                 return generic_drop_inode(inode);
7013 }
7014
7015 static void init_once(void *foo)
7016 {
7017         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7018
7019         inode_init_once(&ei->vfs_inode);
7020 }
7021
7022 void btrfs_destroy_cachep(void)
7023 {
7024         if (btrfs_inode_cachep)
7025                 kmem_cache_destroy(btrfs_inode_cachep);
7026         if (btrfs_trans_handle_cachep)
7027                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7028         if (btrfs_transaction_cachep)
7029                 kmem_cache_destroy(btrfs_transaction_cachep);
7030         if (btrfs_path_cachep)
7031                 kmem_cache_destroy(btrfs_path_cachep);
7032         if (btrfs_free_space_cachep)
7033                 kmem_cache_destroy(btrfs_free_space_cachep);
7034 }
7035
7036 int btrfs_init_cachep(void)
7037 {
7038         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
7039                         sizeof(struct btrfs_inode), 0,
7040                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7041         if (!btrfs_inode_cachep)
7042                 goto fail;
7043
7044         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
7045                         sizeof(struct btrfs_trans_handle), 0,
7046                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7047         if (!btrfs_trans_handle_cachep)
7048                 goto fail;
7049
7050         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
7051                         sizeof(struct btrfs_transaction), 0,
7052                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7053         if (!btrfs_transaction_cachep)
7054                 goto fail;
7055
7056         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
7057                         sizeof(struct btrfs_path), 0,
7058                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7059         if (!btrfs_path_cachep)
7060                 goto fail;
7061
7062         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
7063                         sizeof(struct btrfs_free_space), 0,
7064                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7065         if (!btrfs_free_space_cachep)
7066                 goto fail;
7067
7068         return 0;
7069 fail:
7070         btrfs_destroy_cachep();
7071         return -ENOMEM;
7072 }
7073
7074 static int btrfs_getattr(struct vfsmount *mnt,
7075                          struct dentry *dentry, struct kstat *stat)
7076 {
7077         struct inode *inode = dentry->d_inode;
7078         u32 blocksize = inode->i_sb->s_blocksize;
7079
7080         generic_fillattr(inode, stat);
7081         stat->dev = BTRFS_I(inode)->root->anon_dev;
7082         stat->blksize = PAGE_CACHE_SIZE;
7083         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7084                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7085         return 0;
7086 }
7087
7088 /*
7089  * If a file is moved, it will inherit the cow and compression flags of the new
7090  * directory.
7091  */
7092 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7093 {
7094         struct btrfs_inode *b_dir = BTRFS_I(dir);
7095         struct btrfs_inode *b_inode = BTRFS_I(inode);
7096
7097         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7098                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7099         else
7100                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7101
7102         if (b_dir->flags & BTRFS_INODE_COMPRESS)
7103                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7104         else
7105                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
7106 }
7107
7108 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7109                            struct inode *new_dir, struct dentry *new_dentry)
7110 {
7111         struct btrfs_trans_handle *trans;
7112         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7113         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7114         struct inode *new_inode = new_dentry->d_inode;
7115         struct inode *old_inode = old_dentry->d_inode;
7116         struct timespec ctime = CURRENT_TIME;
7117         u64 index = 0;
7118         u64 root_objectid;
7119         int ret;
7120         u64 old_ino = btrfs_ino(old_inode);
7121
7122         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7123                 return -EPERM;
7124
7125         /* we only allow rename subvolume link between subvolumes */
7126         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7127                 return -EXDEV;
7128
7129         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7130             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7131                 return -ENOTEMPTY;
7132
7133         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7134             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7135                 return -ENOTEMPTY;
7136         /*
7137          * we're using rename to replace one file with another.
7138          * and the replacement file is large.  Start IO on it now so
7139          * we don't add too much work to the end of the transaction
7140          */
7141         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7142             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7143                 filemap_flush(old_inode->i_mapping);
7144
7145         /* close the racy window with snapshot create/destroy ioctl */
7146         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7147                 down_read(&root->fs_info->subvol_sem);
7148         /*
7149          * We want to reserve the absolute worst case amount of items.  So if
7150          * both inodes are subvols and we need to unlink them then that would
7151          * require 4 item modifications, but if they are both normal inodes it
7152          * would require 5 item modifications, so we'll assume their normal
7153          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7154          * should cover the worst case number of items we'll modify.
7155          */
7156         trans = btrfs_start_transaction(root, 20);
7157         if (IS_ERR(trans)) {
7158                 ret = PTR_ERR(trans);
7159                 goto out_notrans;
7160         }
7161
7162         if (dest != root)
7163                 btrfs_record_root_in_trans(trans, dest);
7164
7165         ret = btrfs_set_inode_index(new_dir, &index);
7166         if (ret)
7167                 goto out_fail;
7168
7169         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7170                 /* force full log commit if subvolume involved. */
7171                 root->fs_info->last_trans_log_full_commit = trans->transid;
7172         } else {
7173                 ret = btrfs_insert_inode_ref(trans, dest,
7174                                              new_dentry->d_name.name,
7175                                              new_dentry->d_name.len,
7176                                              old_ino,
7177                                              btrfs_ino(new_dir), index);
7178                 if (ret)
7179                         goto out_fail;
7180                 /*
7181                  * this is an ugly little race, but the rename is required
7182                  * to make sure that if we crash, the inode is either at the
7183                  * old name or the new one.  pinning the log transaction lets
7184                  * us make sure we don't allow a log commit to come in after
7185                  * we unlink the name but before we add the new name back in.
7186                  */
7187                 btrfs_pin_log_trans(root);
7188         }
7189         /*
7190          * make sure the inode gets flushed if it is replacing
7191          * something.
7192          */
7193         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7194                 btrfs_add_ordered_operation(trans, root, old_inode);
7195
7196         old_dir->i_ctime = old_dir->i_mtime = ctime;
7197         new_dir->i_ctime = new_dir->i_mtime = ctime;
7198         old_inode->i_ctime = ctime;
7199
7200         if (old_dentry->d_parent != new_dentry->d_parent)
7201                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7202
7203         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7204                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7205                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7206                                         old_dentry->d_name.name,
7207                                         old_dentry->d_name.len);
7208         } else {
7209                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7210                                         old_dentry->d_inode,
7211                                         old_dentry->d_name.name,
7212                                         old_dentry->d_name.len);
7213                 if (!ret)
7214                         ret = btrfs_update_inode(trans, root, old_inode);
7215         }
7216         if (ret) {
7217                 btrfs_abort_transaction(trans, root, ret);
7218                 goto out_fail;
7219         }
7220
7221         if (new_inode) {
7222                 new_inode->i_ctime = CURRENT_TIME;
7223                 if (unlikely(btrfs_ino(new_inode) ==
7224                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7225                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7226                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7227                                                 root_objectid,
7228                                                 new_dentry->d_name.name,
7229                                                 new_dentry->d_name.len);
7230                         BUG_ON(new_inode->i_nlink == 0);
7231                 } else {
7232                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7233                                                  new_dentry->d_inode,
7234                                                  new_dentry->d_name.name,
7235                                                  new_dentry->d_name.len);
7236                 }
7237                 if (!ret && new_inode->i_nlink == 0) {
7238                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7239                         BUG_ON(ret);
7240                 }
7241                 if (ret) {
7242                         btrfs_abort_transaction(trans, root, ret);
7243                         goto out_fail;
7244                 }
7245         }
7246
7247         fixup_inode_flags(new_dir, old_inode);
7248
7249         ret = btrfs_add_link(trans, new_dir, old_inode,
7250                              new_dentry->d_name.name,
7251                              new_dentry->d_name.len, 0, index);
7252         if (ret) {
7253                 btrfs_abort_transaction(trans, root, ret);
7254                 goto out_fail;
7255         }
7256
7257         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7258                 struct dentry *parent = new_dentry->d_parent;
7259                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7260                 btrfs_end_log_trans(root);
7261         }
7262 out_fail:
7263         btrfs_end_transaction(trans, root);
7264 out_notrans:
7265         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7266                 up_read(&root->fs_info->subvol_sem);
7267
7268         return ret;
7269 }
7270
7271 /*
7272  * some fairly slow code that needs optimization. This walks the list
7273  * of all the inodes with pending delalloc and forces them to disk.
7274  */
7275 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7276 {
7277         struct list_head *head = &root->fs_info->delalloc_inodes;
7278         struct btrfs_inode *binode;
7279         struct inode *inode;
7280
7281         if (root->fs_info->sb->s_flags & MS_RDONLY)
7282                 return -EROFS;
7283
7284         spin_lock(&root->fs_info->delalloc_lock);
7285         while (!list_empty(head)) {
7286                 binode = list_entry(head->next, struct btrfs_inode,
7287                                     delalloc_inodes);
7288                 inode = igrab(&binode->vfs_inode);
7289                 if (!inode)
7290                         list_del_init(&binode->delalloc_inodes);
7291                 spin_unlock(&root->fs_info->delalloc_lock);
7292                 if (inode) {
7293                         filemap_flush(inode->i_mapping);
7294                         if (delay_iput)
7295                                 btrfs_add_delayed_iput(inode);
7296                         else
7297                                 iput(inode);
7298                 }
7299                 cond_resched();
7300                 spin_lock(&root->fs_info->delalloc_lock);
7301         }
7302         spin_unlock(&root->fs_info->delalloc_lock);
7303
7304         /* the filemap_flush will queue IO into the worker threads, but
7305          * we have to make sure the IO is actually started and that
7306          * ordered extents get created before we return
7307          */
7308         atomic_inc(&root->fs_info->async_submit_draining);
7309         while (atomic_read(&root->fs_info->nr_async_submits) ||
7310               atomic_read(&root->fs_info->async_delalloc_pages)) {
7311                 wait_event(root->fs_info->async_submit_wait,
7312                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7313                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7314         }
7315         atomic_dec(&root->fs_info->async_submit_draining);
7316         return 0;
7317 }
7318
7319 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7320                          const char *symname)
7321 {
7322         struct btrfs_trans_handle *trans;
7323         struct btrfs_root *root = BTRFS_I(dir)->root;
7324         struct btrfs_path *path;
7325         struct btrfs_key key;
7326         struct inode *inode = NULL;
7327         int err;
7328         int drop_inode = 0;
7329         u64 objectid;
7330         u64 index = 0 ;
7331         int name_len;
7332         int datasize;
7333         unsigned long ptr;
7334         struct btrfs_file_extent_item *ei;
7335         struct extent_buffer *leaf;
7336         unsigned long nr = 0;
7337
7338         name_len = strlen(symname) + 1;
7339         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7340                 return -ENAMETOOLONG;
7341
7342         /*
7343          * 2 items for inode item and ref
7344          * 2 items for dir items
7345          * 1 item for xattr if selinux is on
7346          */
7347         trans = btrfs_start_transaction(root, 5);
7348         if (IS_ERR(trans))
7349                 return PTR_ERR(trans);
7350
7351         err = btrfs_find_free_ino(root, &objectid);
7352         if (err)
7353                 goto out_unlock;
7354
7355         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7356                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7357                                 S_IFLNK|S_IRWXUGO, &index);
7358         if (IS_ERR(inode)) {
7359                 err = PTR_ERR(inode);
7360                 goto out_unlock;
7361         }
7362
7363         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7364         if (err) {
7365                 drop_inode = 1;
7366                 goto out_unlock;
7367         }
7368
7369         /*
7370         * If the active LSM wants to access the inode during
7371         * d_instantiate it needs these. Smack checks to see
7372         * if the filesystem supports xattrs by looking at the
7373         * ops vector.
7374         */
7375         inode->i_fop = &btrfs_file_operations;
7376         inode->i_op = &btrfs_file_inode_operations;
7377
7378         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7379         if (err)
7380                 drop_inode = 1;
7381         else {
7382                 inode->i_mapping->a_ops = &btrfs_aops;
7383                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7384                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7385         }
7386         if (drop_inode)
7387                 goto out_unlock;
7388
7389         path = btrfs_alloc_path();
7390         if (!path) {
7391                 err = -ENOMEM;
7392                 drop_inode = 1;
7393                 goto out_unlock;
7394         }
7395         key.objectid = btrfs_ino(inode);
7396         key.offset = 0;
7397         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7398         datasize = btrfs_file_extent_calc_inline_size(name_len);
7399         err = btrfs_insert_empty_item(trans, root, path, &key,
7400                                       datasize);
7401         if (err) {
7402                 drop_inode = 1;
7403                 btrfs_free_path(path);
7404                 goto out_unlock;
7405         }
7406         leaf = path->nodes[0];
7407         ei = btrfs_item_ptr(leaf, path->slots[0],
7408                             struct btrfs_file_extent_item);
7409         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7410         btrfs_set_file_extent_type(leaf, ei,
7411                                    BTRFS_FILE_EXTENT_INLINE);
7412         btrfs_set_file_extent_encryption(leaf, ei, 0);
7413         btrfs_set_file_extent_compression(leaf, ei, 0);
7414         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7415         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7416
7417         ptr = btrfs_file_extent_inline_start(ei);
7418         write_extent_buffer(leaf, symname, ptr, name_len);
7419         btrfs_mark_buffer_dirty(leaf);
7420         btrfs_free_path(path);
7421
7422         inode->i_op = &btrfs_symlink_inode_operations;
7423         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7424         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7425         inode_set_bytes(inode, name_len);
7426         btrfs_i_size_write(inode, name_len - 1);
7427         err = btrfs_update_inode(trans, root, inode);
7428         if (err)
7429                 drop_inode = 1;
7430
7431 out_unlock:
7432         if (!err)
7433                 d_instantiate(dentry, inode);
7434         nr = trans->blocks_used;
7435         btrfs_end_transaction(trans, root);
7436         if (drop_inode) {
7437                 inode_dec_link_count(inode);
7438                 iput(inode);
7439         }
7440         btrfs_btree_balance_dirty(root, nr);
7441         return err;
7442 }
7443
7444 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7445                                        u64 start, u64 num_bytes, u64 min_size,
7446                                        loff_t actual_len, u64 *alloc_hint,
7447                                        struct btrfs_trans_handle *trans)
7448 {
7449         struct btrfs_root *root = BTRFS_I(inode)->root;
7450         struct btrfs_key ins;
7451         u64 cur_offset = start;
7452         u64 i_size;
7453         int ret = 0;
7454         bool own_trans = true;
7455
7456         if (trans)
7457                 own_trans = false;
7458         while (num_bytes > 0) {
7459                 if (own_trans) {
7460                         trans = btrfs_start_transaction(root, 3);
7461                         if (IS_ERR(trans)) {
7462                                 ret = PTR_ERR(trans);
7463                                 break;
7464                         }
7465                 }
7466
7467                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7468                                            0, *alloc_hint, &ins, 1);
7469                 if (ret) {
7470                         if (own_trans)
7471                                 btrfs_end_transaction(trans, root);
7472                         break;
7473                 }
7474
7475                 ret = insert_reserved_file_extent(trans, inode,
7476                                                   cur_offset, ins.objectid,
7477                                                   ins.offset, ins.offset,
7478                                                   ins.offset, 0, 0, 0,
7479                                                   BTRFS_FILE_EXTENT_PREALLOC);
7480                 if (ret) {
7481                         btrfs_abort_transaction(trans, root, ret);
7482                         if (own_trans)
7483                                 btrfs_end_transaction(trans, root);
7484                         break;
7485                 }
7486                 btrfs_drop_extent_cache(inode, cur_offset,
7487                                         cur_offset + ins.offset -1, 0);
7488
7489                 num_bytes -= ins.offset;
7490                 cur_offset += ins.offset;
7491                 *alloc_hint = ins.objectid + ins.offset;
7492
7493                 inode->i_ctime = CURRENT_TIME;
7494                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7495                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7496                     (actual_len > inode->i_size) &&
7497                     (cur_offset > inode->i_size)) {
7498                         if (cur_offset > actual_len)
7499                                 i_size = actual_len;
7500                         else
7501                                 i_size = cur_offset;
7502                         i_size_write(inode, i_size);
7503                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7504                 }
7505
7506                 ret = btrfs_update_inode(trans, root, inode);
7507
7508                 if (ret) {
7509                         btrfs_abort_transaction(trans, root, ret);
7510                         if (own_trans)
7511                                 btrfs_end_transaction(trans, root);
7512                         break;
7513                 }
7514
7515                 if (own_trans)
7516                         btrfs_end_transaction(trans, root);
7517         }
7518         return ret;
7519 }
7520
7521 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7522                               u64 start, u64 num_bytes, u64 min_size,
7523                               loff_t actual_len, u64 *alloc_hint)
7524 {
7525         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7526                                            min_size, actual_len, alloc_hint,
7527                                            NULL);
7528 }
7529
7530 int btrfs_prealloc_file_range_trans(struct inode *inode,
7531                                     struct btrfs_trans_handle *trans, int mode,
7532                                     u64 start, u64 num_bytes, u64 min_size,
7533                                     loff_t actual_len, u64 *alloc_hint)
7534 {
7535         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7536                                            min_size, actual_len, alloc_hint, trans);
7537 }
7538
7539 static int btrfs_set_page_dirty(struct page *page)
7540 {
7541         return __set_page_dirty_nobuffers(page);
7542 }
7543
7544 static int btrfs_permission(struct inode *inode, int mask)
7545 {
7546         struct btrfs_root *root = BTRFS_I(inode)->root;
7547         umode_t mode = inode->i_mode;
7548
7549         if (mask & MAY_WRITE &&
7550             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7551                 if (btrfs_root_readonly(root))
7552                         return -EROFS;
7553                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7554                         return -EACCES;
7555         }
7556         return generic_permission(inode, mask);
7557 }
7558
7559 static const struct inode_operations btrfs_dir_inode_operations = {
7560         .getattr        = btrfs_getattr,
7561         .lookup         = btrfs_lookup,
7562         .create         = btrfs_create,
7563         .unlink         = btrfs_unlink,
7564         .link           = btrfs_link,
7565         .mkdir          = btrfs_mkdir,
7566         .rmdir          = btrfs_rmdir,
7567         .rename         = btrfs_rename,
7568         .symlink        = btrfs_symlink,
7569         .setattr        = btrfs_setattr,
7570         .mknod          = btrfs_mknod,
7571         .setxattr       = btrfs_setxattr,
7572         .getxattr       = btrfs_getxattr,
7573         .listxattr      = btrfs_listxattr,
7574         .removexattr    = btrfs_removexattr,
7575         .permission     = btrfs_permission,
7576         .get_acl        = btrfs_get_acl,
7577 };
7578 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7579         .lookup         = btrfs_lookup,
7580         .permission     = btrfs_permission,
7581         .get_acl        = btrfs_get_acl,
7582 };
7583
7584 static const struct file_operations btrfs_dir_file_operations = {
7585         .llseek         = generic_file_llseek,
7586         .read           = generic_read_dir,
7587         .readdir        = btrfs_real_readdir,
7588         .unlocked_ioctl = btrfs_ioctl,
7589 #ifdef CONFIG_COMPAT
7590         .compat_ioctl   = btrfs_ioctl,
7591 #endif
7592         .release        = btrfs_release_file,
7593         .fsync          = btrfs_sync_file,
7594 };
7595
7596 static struct extent_io_ops btrfs_extent_io_ops = {
7597         .fill_delalloc = run_delalloc_range,
7598         .submit_bio_hook = btrfs_submit_bio_hook,
7599         .merge_bio_hook = btrfs_merge_bio_hook,
7600         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7601         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7602         .writepage_start_hook = btrfs_writepage_start_hook,
7603         .set_bit_hook = btrfs_set_bit_hook,
7604         .clear_bit_hook = btrfs_clear_bit_hook,
7605         .merge_extent_hook = btrfs_merge_extent_hook,
7606         .split_extent_hook = btrfs_split_extent_hook,
7607 };
7608
7609 /*
7610  * btrfs doesn't support the bmap operation because swapfiles
7611  * use bmap to make a mapping of extents in the file.  They assume
7612  * these extents won't change over the life of the file and they
7613  * use the bmap result to do IO directly to the drive.
7614  *
7615  * the btrfs bmap call would return logical addresses that aren't
7616  * suitable for IO and they also will change frequently as COW
7617  * operations happen.  So, swapfile + btrfs == corruption.
7618  *
7619  * For now we're avoiding this by dropping bmap.
7620  */
7621 static const struct address_space_operations btrfs_aops = {
7622         .readpage       = btrfs_readpage,
7623         .writepage      = btrfs_writepage,
7624         .writepages     = btrfs_writepages,
7625         .readpages      = btrfs_readpages,
7626         .direct_IO      = btrfs_direct_IO,
7627         .invalidatepage = btrfs_invalidatepage,
7628         .releasepage    = btrfs_releasepage,
7629         .set_page_dirty = btrfs_set_page_dirty,
7630         .error_remove_page = generic_error_remove_page,
7631 };
7632
7633 static const struct address_space_operations btrfs_symlink_aops = {
7634         .readpage       = btrfs_readpage,
7635         .writepage      = btrfs_writepage,
7636         .invalidatepage = btrfs_invalidatepage,
7637         .releasepage    = btrfs_releasepage,
7638 };
7639
7640 static const struct inode_operations btrfs_file_inode_operations = {
7641         .getattr        = btrfs_getattr,
7642         .setattr        = btrfs_setattr,
7643         .setxattr       = btrfs_setxattr,
7644         .getxattr       = btrfs_getxattr,
7645         .listxattr      = btrfs_listxattr,
7646         .removexattr    = btrfs_removexattr,
7647         .permission     = btrfs_permission,
7648         .fiemap         = btrfs_fiemap,
7649         .get_acl        = btrfs_get_acl,
7650 };
7651 static const struct inode_operations btrfs_special_inode_operations = {
7652         .getattr        = btrfs_getattr,
7653         .setattr        = btrfs_setattr,
7654         .permission     = btrfs_permission,
7655         .setxattr       = btrfs_setxattr,
7656         .getxattr       = btrfs_getxattr,
7657         .listxattr      = btrfs_listxattr,
7658         .removexattr    = btrfs_removexattr,
7659         .get_acl        = btrfs_get_acl,
7660 };
7661 static const struct inode_operations btrfs_symlink_inode_operations = {
7662         .readlink       = generic_readlink,
7663         .follow_link    = page_follow_link_light,
7664         .put_link       = page_put_link,
7665         .getattr        = btrfs_getattr,
7666         .setattr        = btrfs_setattr,
7667         .permission     = btrfs_permission,
7668         .setxattr       = btrfs_setxattr,
7669         .getxattr       = btrfs_getxattr,
7670         .listxattr      = btrfs_listxattr,
7671         .removexattr    = btrfs_removexattr,
7672         .get_acl        = btrfs_get_acl,
7673 };
7674
7675 const struct dentry_operations btrfs_dentry_operations = {
7676         .d_delete       = btrfs_dentry_delete,
7677         .d_release      = btrfs_dentry_release,
7678 };