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 "compat.h"
  42 #include "ctree.h"
  43 #include "disk-io.h"
  44 #include "transaction.h"
  45 #include "btrfs_inode.h"
  46 #include "ioctl.h"
  47 #include "print-tree.h"
  48 #include "volumes.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "compression.h"
  53 #include "locking.h"
  54 #include "free-space-cache.h"
  55 #include "inode-map.h"
  56
  57 struct btrfs_iget_args {
  58         u64 ino;
  59         struct btrfs_root *root;
  60 };
  61
  62 static const struct inode_operations btrfs_dir_inode_operations;
  63 static const struct inode_operations btrfs_symlink_inode_operations;
  64 static const struct inode_operations btrfs_dir_ro_inode_operations;
  65 static const struct inode_operations btrfs_special_inode_operations;
  66 static const struct inode_operations btrfs_file_inode_operations;
  67 static const struct address_space_operations btrfs_aops;
  68 static const struct address_space_operations btrfs_symlink_aops;
  69 static const struct file_operations btrfs_dir_file_operations;
  70 static struct extent_io_ops btrfs_extent_io_ops;
  71
  72 static struct kmem_cache *btrfs_inode_cachep;
  73 struct kmem_cache *btrfs_trans_handle_cachep;
  74 struct kmem_cache *btrfs_transaction_cachep;
  75 struct kmem_cache *btrfs_path_cachep;
  76 struct kmem_cache *btrfs_free_space_cachep;
  77
  78 #define S_SHIFT 12
  79 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  80         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  81         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  82         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  83         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  84         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  85         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  86         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  87 };
  88
  89 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  90 static int btrfs_truncate(struct inode *inode);
  91 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end);
  92 static noinline int cow_file_range(struct inode *inode,
  93                                    struct page *locked_page,
  94                                    u64 start, u64 end, int *page_started,
  95                                    unsigned long *nr_written, int unlock);
  96
  97 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
  98                                      struct inode *inode,  struct inode *dir,
  99                                      const struct qstr *qstr)
 100 {
 101         int err;
 102
 103         err = btrfs_init_acl(trans, inode, dir);
 104         if (!err)
 105                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 106         return err;
 107 }
 108
 109 /*
 110  * this does all the hard work for inserting an inline extent into
 111  * the btree.  The caller should have done a btrfs_drop_extents so that
 112  * no overlapping inline items exist in the btree
 113  */
 114 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 115                                 struct btrfs_root *root, struct inode *inode,
 116                                 u64 start, size_t size, size_t compressed_size,
 117                                 int compress_type,
 118                                 struct page **compressed_pages)
 119 {
 120         struct btrfs_key key;
 121         struct btrfs_path *path;
 122         struct extent_buffer *leaf;
 123         struct page *page = NULL;
 124         char *kaddr;
 125         unsigned long ptr;
 126         struct btrfs_file_extent_item *ei;
 127         int err = 0;
 128         int ret;
 129         size_t cur_size = size;
 130         size_t datasize;
 131         unsigned long offset;
 132
 133         if (compressed_size && compressed_pages)
 134                 cur_size = compressed_size;
 135
 136         path = btrfs_alloc_path();
 137         if (!path)
 138                 return -ENOMEM;
 139
 140         path->leave_spinning = 1;
 141
 142         key.objectid = btrfs_ino(inode);
 143         key.offset = start;
 144         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 145         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 146
 147         inode_add_bytes(inode, size);
 148         ret = btrfs_insert_empty_item(trans, root, path, &key,
 149                                       datasize);
 150         BUG_ON(ret);
 151         if (ret) {
 152                 err = ret;
 153                 goto fail;
 154         }
 155         leaf = path->nodes[0];
 156         ei = btrfs_item_ptr(leaf, path->slots[0],
 157                             struct btrfs_file_extent_item);
 158         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 159         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 160         btrfs_set_file_extent_encryption(leaf, ei, 0);
 161         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 162         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 163         ptr = btrfs_file_extent_inline_start(ei);
 164
 165         if (compress_type != BTRFS_COMPRESS_NONE) {
 166                 struct page *cpage;
 167                 int i = 0;
 168                 while (compressed_size > 0) {
 169                         cpage = compressed_pages[i];
 170                         cur_size = min_t(unsigned long, compressed_size,
 171                                        PAGE_CACHE_SIZE);
 172
 173                         kaddr = kmap_atomic(cpage, KM_USER0);
 174                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 175                         kunmap_atomic(kaddr, KM_USER0);
 176
 177                         i++;
 178                         ptr += cur_size;
 179                         compressed_size -= cur_size;
 180                 }
 181                 btrfs_set_file_extent_compression(leaf, ei,
 182                                                   compress_type);
 183         } else {
 184                 page = find_get_page(inode->i_mapping,
 185                                      start >> PAGE_CACHE_SHIFT);
 186                 btrfs_set_file_extent_compression(leaf, ei, 0);
 187                 kaddr = kmap_atomic(page, KM_USER0);
 188                 offset = start & (PAGE_CACHE_SIZE - 1);
 189                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 190                 kunmap_atomic(kaddr, KM_USER0);
 191                 page_cache_release(page);
 192         }
 193         btrfs_mark_buffer_dirty(leaf);
 194         btrfs_free_path(path);
 195
 196         /*
 197          * we're an inline extent, so nobody can
 198          * extend the file past i_size without locking
 199          * a page we already have locked.
 200          *
 201          * We must do any isize and inode updates
 202          * before we unlock the pages.  Otherwise we
 203          * could end up racing with unlink.
 204          */
 205         BTRFS_I(inode)->disk_i_size = inode->i_size;
 206         btrfs_update_inode(trans, root, inode);
 207
 208         return 0;
 209 fail:
 210         btrfs_free_path(path);
 211         return err;
 212 }
 213
 214
 215 /*
 216  * conditionally insert an inline extent into the file.  This
 217  * does the checks required to make sure the data is small enough
 218  * to fit as an inline extent.
 219  */
 220 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 221                                  struct btrfs_root *root,
 222                                  struct inode *inode, u64 start, u64 end,
 223                                  size_t compressed_size, int compress_type,
 224                                  struct page **compressed_pages)
 225 {
 226         u64 isize = i_size_read(inode);
 227         u64 actual_end = min(end + 1, isize);
 228         u64 inline_len = actual_end - start;
 229         u64 aligned_end = (end + root->sectorsize - 1) &
 230                         ~((u64)root->sectorsize - 1);
 231         u64 hint_byte;
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, inode, start, aligned_end,
 249                                  &hint_byte, 1);
 250         BUG_ON(ret);
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         BUG_ON(ret);
 258         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 259         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 260         return 0;
 261 }
 262
 263 struct async_extent {
 264         u64 start;
 265         u64 ram_size;
 266         u64 compressed_size;
 267         struct page **pages;
 268         unsigned long nr_pages;
 269         int compress_type;
 270         struct list_head list;
 271 };
 272
 273 struct async_cow {
 274         struct inode *inode;
 275         struct btrfs_root *root;
 276         struct page *locked_page;
 277         u64 start;
 278         u64 end;
 279         struct list_head extents;
 280         struct btrfs_work work;
 281 };
 282
 283 static noinline int add_async_extent(struct async_cow *cow,
 284                                      u64 start, u64 ram_size,
 285                                      u64 compressed_size,
 286                                      struct page **pages,
 287                                      unsigned long nr_pages,
 288                                      int compress_type)
 289 {
 290         struct async_extent *async_extent;
 291
 292         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 293         BUG_ON(!async_extent);
 294         async_extent->start = start;
 295         async_extent->ram_size = ram_size;
 296         async_extent->compressed_size = compressed_size;
 297         async_extent->pages = pages;
 298         async_extent->nr_pages = nr_pages;
 299         async_extent->compress_type = compress_type;
 300         list_add_tail(&async_extent->list, &cow->extents);
 301         return 0;
 302 }
 303
 304 /*
 305  * we create compressed extents in two phases.  The first
 306  * phase compresses a range of pages that have already been
 307  * locked (both pages and state bits are locked).
 308  *
 309  * This is done inside an ordered work queue, and the compression
 310  * is spread across many cpus.  The actual IO submission is step
 311  * two, and the ordered work queue takes care of making sure that
 312  * happens in the same order things were put onto the queue by
 313  * writepages and friends.
 314  *
 315  * If this code finds it can't get good compression, it puts an
 316  * entry onto the work queue to write the uncompressed bytes.  This
 317  * makes sure that both compressed inodes and uncompressed inodes
 318  * are written in the same order that pdflush sent them down.
 319  */
 320 static noinline int compress_file_range(struct inode *inode,
 321                                         struct page *locked_page,
 322                                         u64 start, u64 end,
 323                                         struct async_cow *async_cow,
 324                                         int *num_added)
 325 {
 326         struct btrfs_root *root = BTRFS_I(inode)->root;
 327         struct btrfs_trans_handle *trans;
 328         u64 num_bytes;
 329         u64 blocksize = root->sectorsize;
 330         u64 actual_end;
 331         u64 isize = i_size_read(inode);
 332         int ret = 0;
 333         struct page **pages = NULL;
 334         unsigned long nr_pages;
 335         unsigned long nr_pages_ret = 0;
 336         unsigned long total_compressed = 0;
 337         unsigned long total_in = 0;
 338         unsigned long max_compressed = 128 * 1024;
 339         unsigned long max_uncompressed = 128 * 1024;
 340         int i;
 341         int will_compress;
 342         int compress_type = root->fs_info->compress_type;
 343
 344         /* if this is a small write inside eof, kick off a defragbot */
 345         if (end <= BTRFS_I(inode)->disk_i_size && (end - start + 1) < 16 * 1024)
 346                 btrfs_add_inode_defrag(NULL, inode);
 347
 348         actual_end = min_t(u64, isize, end + 1);
 349 again:
 350         will_compress = 0;
 351         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 352         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 353
 354         /*
 355          * we don't want to send crud past the end of i_size through
 356          * compression, that's just a waste of CPU time.  So, if the
 357          * end of the file is before the start of our current
 358          * requested range of bytes, we bail out to the uncompressed
 359          * cleanup code that can deal with all of this.
 360          *
 361          * It isn't really the fastest way to fix things, but this is a
 362          * very uncommon corner.
 363          */
 364         if (actual_end <= start)
 365                 goto cleanup_and_bail_uncompressed;
 366
 367         total_compressed = actual_end - start;
 368
 369         /* we want to make sure that amount of ram required to uncompress
 370          * an extent is reasonable, so we limit the total size in ram
 371          * of a compressed extent to 128k.  This is a crucial number
 372          * because it also controls how easily we can spread reads across
 373          * cpus for decompression.
 374          *
 375          * We also want to make sure the amount of IO required to do
 376          * a random read is reasonably small, so we limit the size of
 377          * a compressed extent to 128k.
 378          */
 379         total_compressed = min(total_compressed, max_uncompressed);
 380         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 381         num_bytes = max(blocksize,  num_bytes);
 382         total_in = 0;
 383         ret = 0;
 384
 385         /*
 386          * we do compression for mount -o compress and when the
 387          * inode has not been flagged as nocompress.  This flag can
 388          * change at any time if we discover bad compression ratios.
 389          */
 390         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 391             (btrfs_test_opt(root, COMPRESS) ||
 392              (BTRFS_I(inode)->force_compress) ||
 393              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 394                 WARN_ON(pages);
 395                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 396                 if (!pages) {
 397                         /* just bail out to the uncompressed code */
 398                         goto cont;
 399                 }
 400
 401                 if (BTRFS_I(inode)->force_compress)
 402                         compress_type = BTRFS_I(inode)->force_compress;
 403
 404                 ret = btrfs_compress_pages(compress_type,
 405                                            inode->i_mapping, start,
 406                                            total_compressed, pages,
 407                                            nr_pages, &nr_pages_ret,
 408                                            &total_in,
 409                                            &total_compressed,
 410                                            max_compressed);
 411
 412                 if (!ret) {
 413                         unsigned long offset = total_compressed &
 414                                 (PAGE_CACHE_SIZE - 1);
 415                         struct page *page = pages[nr_pages_ret - 1];
 416                         char *kaddr;
 417
 418                         /* zero the tail end of the last page, we might be
 419                          * sending it down to disk
 420                          */
 421                         if (offset) {
 422                                 kaddr = kmap_atomic(page, KM_USER0);
 423                                 memset(kaddr + offset, 0,
 424                                        PAGE_CACHE_SIZE - offset);
 425                                 kunmap_atomic(kaddr, KM_USER0);
 426                         }
 427                         will_compress = 1;
 428                 }
 429         }
 430 cont:
 431         if (start == 0) {
 432                 trans = btrfs_join_transaction(root);
 433                 BUG_ON(IS_ERR(trans));
 434                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 435
 436                 /* lets try to make an inline extent */
 437                 if (ret || total_in < (actual_end - start)) {
 438                         /* we didn't compress the entire range, try
 439                          * to make an uncompressed inline extent.
 440                          */
 441                         ret = cow_file_range_inline(trans, root, inode,
 442                                                     start, end, 0, 0, NULL);
 443                 } else {
 444                         /* try making a compressed inline extent */
 445                         ret = cow_file_range_inline(trans, root, inode,
 446                                                     start, end,
 447                                                     total_compressed,
 448                                                     compress_type, pages);
 449                 }
 450                 if (ret == 0) {
 451                         /*
 452                          * inline extent creation worked, we don't need
 453                          * to create any more async work items.  Unlock
 454                          * and free up our temp pages.
 455                          */
 456                         extent_clear_unlock_delalloc(inode,
 457                              &BTRFS_I(inode)->io_tree,
 458                              start, end, NULL,
 459                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 460                              EXTENT_CLEAR_DELALLOC |
 461                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 462
 463                         btrfs_end_transaction(trans, root);
 464                         goto free_pages_out;
 465                 }
 466                 btrfs_end_transaction(trans, root);
 467         }
 468
 469         if (will_compress) {
 470                 /*
 471                  * we aren't doing an inline extent round the compressed size
 472                  * up to a block size boundary so the allocator does sane
 473                  * things
 474                  */
 475                 total_compressed = (total_compressed + blocksize - 1) &
 476                         ~(blocksize - 1);
 477
 478                 /*
 479                  * one last check to make sure the compression is really a
 480                  * win, compare the page count read with the blocks on disk
 481                  */
 482                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 483                         ~(PAGE_CACHE_SIZE - 1);
 484                 if (total_compressed >= total_in) {
 485                         will_compress = 0;
 486                 } else {
 487                         num_bytes = total_in;
 488                 }
 489         }
 490         if (!will_compress && pages) {
 491                 /*
 492                  * the compression code ran but failed to make things smaller,
 493                  * free any pages it allocated and our page pointer array
 494                  */
 495                 for (i = 0; i < nr_pages_ret; i++) {
 496                         WARN_ON(pages[i]->mapping);
 497                         page_cache_release(pages[i]);
 498                 }
 499                 kfree(pages);
 500                 pages = NULL;
 501                 total_compressed = 0;
 502                 nr_pages_ret = 0;
 503
 504                 /* flag the file so we don't compress in the future */
 505                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 506                     !(BTRFS_I(inode)->force_compress)) {
 507                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 508                 }
 509         }
 510         if (will_compress) {
 511                 *num_added += 1;
 512
 513                 /* the async work queues will take care of doing actual
 514                  * allocation on disk for these compressed pages,
 515                  * and will submit them to the elevator.
 516                  */
 517                 add_async_extent(async_cow, start, num_bytes,
 518                                  total_compressed, pages, nr_pages_ret,
 519                                  compress_type);
 520
 521                 if (start + num_bytes < end) {
 522                         start += num_bytes;
 523                         pages = NULL;
 524                         cond_resched();
 525                         goto again;
 526                 }
 527         } else {
 528 cleanup_and_bail_uncompressed:
 529                 /*
 530                  * No compression, but we still need to write the pages in
 531                  * the file we've been given so far.  redirty the locked
 532                  * page if it corresponds to our extent and set things up
 533                  * for the async work queue to run cow_file_range to do
 534                  * the normal delalloc dance
 535                  */
 536                 if (page_offset(locked_page) >= start &&
 537                     page_offset(locked_page) <= end) {
 538                         __set_page_dirty_nobuffers(locked_page);
 539                         /* unlocked later on in the async handlers */
 540                 }
 541                 add_async_extent(async_cow, start, end - start + 1,
 542                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 543                 *num_added += 1;
 544         }
 545
 546 out:
 547         return 0;
 548
 549 free_pages_out:
 550         for (i = 0; i < nr_pages_ret; i++) {
 551                 WARN_ON(pages[i]->mapping);
 552                 page_cache_release(pages[i]);
 553         }
 554         kfree(pages);
 555
 556         goto out;
 557 }
 558
 559 /*
 560  * phase two of compressed writeback.  This is the ordered portion
 561  * of the code, which only gets called in the order the work was
 562  * queued.  We walk all the async extents created by compress_file_range
 563  * and send them down to the disk.
 564  */
 565 static noinline int submit_compressed_extents(struct inode *inode,
 566                                               struct async_cow *async_cow)
 567 {
 568         struct async_extent *async_extent;
 569         u64 alloc_hint = 0;
 570         struct btrfs_trans_handle *trans;
 571         struct btrfs_key ins;
 572         struct extent_map *em;
 573         struct btrfs_root *root = BTRFS_I(inode)->root;
 574         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 575         struct extent_io_tree *io_tree;
 576         int ret = 0;
 577
 578         if (list_empty(&async_cow->extents))
 579                 return 0;
 580
 581
 582         while (!list_empty(&async_cow->extents)) {
 583                 async_extent = list_entry(async_cow->extents.next,
 584                                           struct async_extent, list);
 585                 list_del(&async_extent->list);
 586
 587                 io_tree = &BTRFS_I(inode)->io_tree;
 588
 589 retry:
 590                 /* did the compression code fall back to uncompressed IO? */
 591                 if (!async_extent->pages) {
 592                         int page_started = 0;
 593                         unsigned long nr_written = 0;
 594
 595                         lock_extent(io_tree, async_extent->start,
 596                                          async_extent->start +
 597                                          async_extent->ram_size - 1, GFP_NOFS);
 598
 599                         /* allocate blocks */
 600                         ret = cow_file_range(inode, async_cow->locked_page,
 601                                              async_extent->start,
 602                                              async_extent->start +
 603                                              async_extent->ram_size - 1,
 604                                              &page_started, &nr_written, 0);
 605
 606                         /*
 607                          * if page_started, cow_file_range inserted an
 608                          * inline extent and took care of all the unlocking
 609                          * and IO for us.  Otherwise, we need to submit
 610                          * all those pages down to the drive.
 611                          */
 612                         if (!page_started && !ret)
 613                                 extent_write_locked_range(io_tree,
 614                                                   inode, async_extent->start,
 615                                                   async_extent->start +
 616                                                   async_extent->ram_size - 1,
 617                                                   btrfs_get_extent,
 618                                                   WB_SYNC_ALL);
 619                         kfree(async_extent);
 620                         cond_resched();
 621                         continue;
 622                 }
 623
 624                 lock_extent(io_tree, async_extent->start,
 625                             async_extent->start + async_extent->ram_size - 1,
 626                             GFP_NOFS);
 627
 628                 trans = btrfs_join_transaction(root);
 629                 BUG_ON(IS_ERR(trans));
 630                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 631                 ret = btrfs_reserve_extent(trans, root,
 632                                            async_extent->compressed_size,
 633                                            async_extent->compressed_size,
 634                                            0, alloc_hint,
 635                                            (u64)-1, &ins, 1);
 636                 btrfs_end_transaction(trans, root);
 637
 638                 if (ret) {
 639                         int i;
 640                         for (i = 0; i < async_extent->nr_pages; i++) {
 641                                 WARN_ON(async_extent->pages[i]->mapping);
 642                                 page_cache_release(async_extent->pages[i]);
 643                         }
 644                         kfree(async_extent->pages);
 645                         async_extent->nr_pages = 0;
 646                         async_extent->pages = NULL;
 647                         unlock_extent(io_tree, async_extent->start,
 648                                       async_extent->start +
 649                                       async_extent->ram_size - 1, GFP_NOFS);
 650                         goto retry;
 651                 }
 652
 653                 /*
 654                  * here we're doing allocation and writeback of the
 655                  * compressed pages
 656                  */
 657                 btrfs_drop_extent_cache(inode, async_extent->start,
 658                                         async_extent->start +
 659                                         async_extent->ram_size - 1, 0);
 660
 661                 em = alloc_extent_map();
 662                 BUG_ON(!em);
 663                 em->start = async_extent->start;
 664                 em->len = async_extent->ram_size;
 665                 em->orig_start = em->start;
 666
 667                 em->block_start = ins.objectid;
 668                 em->block_len = ins.offset;
 669                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 670                 em->compress_type = async_extent->compress_type;
 671                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 672                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 673
 674                 while (1) {
 675                         write_lock(&em_tree->lock);
 676                         ret = add_extent_mapping(em_tree, em);
 677                         write_unlock(&em_tree->lock);
 678                         if (ret != -EEXIST) {
 679                                 free_extent_map(em);
 680                                 break;
 681                         }
 682                         btrfs_drop_extent_cache(inode, async_extent->start,
 683                                                 async_extent->start +
 684                                                 async_extent->ram_size - 1, 0);
 685                 }
 686
 687                 ret = btrfs_add_ordered_extent_compress(inode,
 688                                                 async_extent->start,
 689                                                 ins.objectid,
 690                                                 async_extent->ram_size,
 691                                                 ins.offset,
 692                                                 BTRFS_ORDERED_COMPRESSED,
 693                                                 async_extent->compress_type);
 694                 BUG_ON(ret);
 695
 696                 /*
 697                  * clear dirty, set writeback and unlock the pages.
 698                  */
 699                 extent_clear_unlock_delalloc(inode,
 700                                 &BTRFS_I(inode)->io_tree,
 701                                 async_extent->start,
 702                                 async_extent->start +
 703                                 async_extent->ram_size - 1,
 704                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 705                                 EXTENT_CLEAR_UNLOCK |
 706                                 EXTENT_CLEAR_DELALLOC |
 707                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 708
 709                 ret = btrfs_submit_compressed_write(inode,
 710                                     async_extent->start,
 711                                     async_extent->ram_size,
 712                                     ins.objectid,
 713                                     ins.offset, async_extent->pages,
 714                                     async_extent->nr_pages);
 715
 716                 BUG_ON(ret);
 717                 alloc_hint = ins.objectid + ins.offset;
 718                 kfree(async_extent);
 719                 cond_resched();
 720         }
 721
 722         return 0;
 723 }
 724
 725 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 726                                       u64 num_bytes)
 727 {
 728         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 729         struct extent_map *em;
 730         u64 alloc_hint = 0;
 731
 732         read_lock(&em_tree->lock);
 733         em = search_extent_mapping(em_tree, start, num_bytes);
 734         if (em) {
 735                 /*
 736                  * if block start isn't an actual block number then find the
 737                  * first block in this inode and use that as a hint.  If that
 738                  * block is also bogus then just don't worry about it.
 739                  */
 740                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 741                         free_extent_map(em);
 742                         em = search_extent_mapping(em_tree, 0, 0);
 743                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 744                                 alloc_hint = em->block_start;
 745                         if (em)
 746                                 free_extent_map(em);
 747                 } else {
 748                         alloc_hint = em->block_start;
 749                         free_extent_map(em);
 750                 }
 751         }
 752         read_unlock(&em_tree->lock);
 753
 754         return alloc_hint;
 755 }
 756
 757 /*
 758  * when extent_io.c finds a delayed allocation range in the file,
 759  * the call backs end up in this code.  The basic idea is to
 760  * allocate extents on disk for the range, and create ordered data structs
 761  * in ram to track those extents.
 762  *
 763  * locked_page is the page that writepage had locked already.  We use
 764  * it to make sure we don't do extra locks or unlocks.
 765  *
 766  * *page_started is set to one if we unlock locked_page and do everything
 767  * required to start IO on it.  It may be clean and already done with
 768  * IO when we return.
 769  */
 770 static noinline int cow_file_range(struct inode *inode,
 771                                    struct page *locked_page,
 772                                    u64 start, u64 end, int *page_started,
 773                                    unsigned long *nr_written,
 774                                    int unlock)
 775 {
 776         struct btrfs_root *root = BTRFS_I(inode)->root;
 777         struct btrfs_trans_handle *trans;
 778         u64 alloc_hint = 0;
 779         u64 num_bytes;
 780         unsigned long ram_size;
 781         u64 disk_num_bytes;
 782         u64 cur_alloc_size;
 783         u64 blocksize = root->sectorsize;
 784         struct btrfs_key ins;
 785         struct extent_map *em;
 786         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 787         int ret = 0;
 788
 789         BUG_ON(btrfs_is_free_space_inode(root, inode));
 790         trans = btrfs_join_transaction(root);
 791         BUG_ON(IS_ERR(trans));
 792         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 793
 794         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 795         num_bytes = max(blocksize,  num_bytes);
 796         disk_num_bytes = num_bytes;
 797         ret = 0;
 798
 799         /* if this is a small write inside eof, kick off defrag */
 800         if (end <= BTRFS_I(inode)->disk_i_size && num_bytes < 64 * 1024)
 801                 btrfs_add_inode_defrag(trans, inode);
 802
 803         if (start == 0) {
 804                 /* lets try to make an inline extent */
 805                 ret = cow_file_range_inline(trans, root, inode,
 806                                             start, end, 0, 0, NULL);
 807                 if (ret == 0) {
 808                         extent_clear_unlock_delalloc(inode,
 809                                      &BTRFS_I(inode)->io_tree,
 810                                      start, end, NULL,
 811                                      EXTENT_CLEAR_UNLOCK_PAGE |
 812                                      EXTENT_CLEAR_UNLOCK |
 813                                      EXTENT_CLEAR_DELALLOC |
 814                                      EXTENT_CLEAR_DIRTY |
 815                                      EXTENT_SET_WRITEBACK |
 816                                      EXTENT_END_WRITEBACK);
 817
 818                         *nr_written = *nr_written +
 819                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 820                         *page_started = 1;
 821                         ret = 0;
 822                         goto out;
 823                 }
 824         }
 825
 826         BUG_ON(disk_num_bytes >
 827                btrfs_super_total_bytes(root->fs_info->super_copy));
 828
 829         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 830         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 831
 832         while (disk_num_bytes > 0) {
 833                 unsigned long op;
 834
 835                 cur_alloc_size = disk_num_bytes;
 836                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 837                                            root->sectorsize, 0, alloc_hint,
 838                                            (u64)-1, &ins, 1);
 839                 BUG_ON(ret);
 840
 841                 em = alloc_extent_map();
 842                 BUG_ON(!em);
 843                 em->start = start;
 844                 em->orig_start = em->start;
 845                 ram_size = ins.offset;
 846                 em->len = ins.offset;
 847
 848                 em->block_start = ins.objectid;
 849                 em->block_len = ins.offset;
 850                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 851                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 852
 853                 while (1) {
 854                         write_lock(&em_tree->lock);
 855                         ret = add_extent_mapping(em_tree, em);
 856                         write_unlock(&em_tree->lock);
 857                         if (ret != -EEXIST) {
 858                                 free_extent_map(em);
 859                                 break;
 860                         }
 861                         btrfs_drop_extent_cache(inode, start,
 862                                                 start + ram_size - 1, 0);
 863                 }
 864
 865                 cur_alloc_size = ins.offset;
 866                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 867                                                ram_size, cur_alloc_size, 0);
 868                 BUG_ON(ret);
 869
 870                 if (root->root_key.objectid ==
 871                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 872                         ret = btrfs_reloc_clone_csums(inode, start,
 873                                                       cur_alloc_size);
 874                         BUG_ON(ret);
 875                 }
 876
 877                 if (disk_num_bytes < cur_alloc_size)
 878                         break;
 879
 880                 /* we're not doing compressed IO, don't unlock the first
 881                  * page (which the caller expects to stay locked), don't
 882                  * clear any dirty bits and don't set any writeback bits
 883                  *
 884                  * Do set the Private2 bit so we know this page was properly
 885                  * setup for writepage
 886                  */
 887                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 888                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 889                         EXTENT_SET_PRIVATE2;
 890
 891                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 892                                              start, start + ram_size - 1,
 893                                              locked_page, op);
 894                 disk_num_bytes -= cur_alloc_size;
 895                 num_bytes -= cur_alloc_size;
 896                 alloc_hint = ins.objectid + ins.offset;
 897                 start += cur_alloc_size;
 898         }
 899 out:
 900         ret = 0;
 901         btrfs_end_transaction(trans, root);
 902
 903         return ret;
 904 }
 905
 906 /*
 907  * work queue call back to started compression on a file and pages
 908  */
 909 static noinline void async_cow_start(struct btrfs_work *work)
 910 {
 911         struct async_cow *async_cow;
 912         int num_added = 0;
 913         async_cow = container_of(work, struct async_cow, work);
 914
 915         compress_file_range(async_cow->inode, async_cow->locked_page,
 916                             async_cow->start, async_cow->end, async_cow,
 917                             &num_added);
 918         if (num_added == 0)
 919                 async_cow->inode = NULL;
 920 }
 921
 922 /*
 923  * work queue call back to submit previously compressed pages
 924  */
 925 static noinline void async_cow_submit(struct btrfs_work *work)
 926 {
 927         struct async_cow *async_cow;
 928         struct btrfs_root *root;
 929         unsigned long nr_pages;
 930
 931         async_cow = container_of(work, struct async_cow, work);
 932
 933         root = async_cow->root;
 934         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
 935                 PAGE_CACHE_SHIFT;
 936
 937         atomic_sub(nr_pages, &root->fs_info->async_delalloc_pages);
 938
 939         if (atomic_read(&root->fs_info->async_delalloc_pages) <
 940             5 * 1042 * 1024 &&
 941             waitqueue_active(&root->fs_info->async_submit_wait))
 942                 wake_up(&root->fs_info->async_submit_wait);
 943
 944         if (async_cow->inode)
 945                 submit_compressed_extents(async_cow->inode, async_cow);
 946 }
 947
 948 static noinline void async_cow_free(struct btrfs_work *work)
 949 {
 950         struct async_cow *async_cow;
 951         async_cow = container_of(work, struct async_cow, work);
 952         kfree(async_cow);
 953 }
 954
 955 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
 956                                 u64 start, u64 end, int *page_started,
 957                                 unsigned long *nr_written)
 958 {
 959         struct async_cow *async_cow;
 960         struct btrfs_root *root = BTRFS_I(inode)->root;
 961         unsigned long nr_pages;
 962         u64 cur_end;
 963         int limit = 10 * 1024 * 1042;
 964
 965         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
 966                          1, 0, NULL, GFP_NOFS);
 967         while (start < end) {
 968                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
 969                 BUG_ON(!async_cow);
 970                 async_cow->inode = inode;
 971                 async_cow->root = root;
 972                 async_cow->locked_page = locked_page;
 973                 async_cow->start = start;
 974
 975                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
 976                         cur_end = end;
 977                 else
 978                         cur_end = min(end, start + 512 * 1024 - 1);
 979
 980                 async_cow->end = cur_end;
 981                 INIT_LIST_HEAD(&async_cow->extents);
 982
 983                 async_cow->work.func = async_cow_start;
 984                 async_cow->work.ordered_func = async_cow_submit;
 985                 async_cow->work.ordered_free = async_cow_free;
 986                 async_cow->work.flags = 0;
 987
 988                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
 989                         PAGE_CACHE_SHIFT;
 990                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
 991
 992                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
 993                                    &async_cow->work);
 994
 995                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
 996                         wait_event(root->fs_info->async_submit_wait,
 997                            (atomic_read(&root->fs_info->async_delalloc_pages) <
 998                             limit));
 999                 }
1000
1001                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1002                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1003                         wait_event(root->fs_info->async_submit_wait,
1004                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1005                            0));
1006                 }
1007
1008                 *nr_written += nr_pages;
1009                 start = cur_end + 1;
1010         }
1011         *page_started = 1;
1012         return 0;
1013 }
1014
1015 static noinline int csum_exist_in_range(struct btrfs_root *root,
1016                                         u64 bytenr, u64 num_bytes)
1017 {
1018         int ret;
1019         struct btrfs_ordered_sum *sums;
1020         LIST_HEAD(list);
1021
1022         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1023                                        bytenr + num_bytes - 1, &list, 0);
1024         if (ret == 0 && list_empty(&list))
1025                 return 0;
1026
1027         while (!list_empty(&list)) {
1028                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1029                 list_del(&sums->list);
1030                 kfree(sums);
1031         }
1032         return 1;
1033 }
1034
1035 /*
1036  * when nowcow writeback call back.  This checks for snapshots or COW copies
1037  * of the extents that exist in the file, and COWs the file as required.
1038  *
1039  * If no cow copies or snapshots exist, we write directly to the existing
1040  * blocks on disk
1041  */
1042 static noinline int run_delalloc_nocow(struct inode *inode,
1043                                        struct page *locked_page,
1044                               u64 start, u64 end, int *page_started, int force,
1045                               unsigned long *nr_written)
1046 {
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         struct btrfs_trans_handle *trans;
1049         struct extent_buffer *leaf;
1050         struct btrfs_path *path;
1051         struct btrfs_file_extent_item *fi;
1052         struct btrfs_key found_key;
1053         u64 cow_start;
1054         u64 cur_offset;
1055         u64 extent_end;
1056         u64 extent_offset;
1057         u64 disk_bytenr;
1058         u64 num_bytes;
1059         int extent_type;
1060         int ret;
1061         int type;
1062         int nocow;
1063         int check_prev = 1;
1064         bool nolock;
1065         u64 ino = btrfs_ino(inode);
1066
1067         path = btrfs_alloc_path();
1068         if (!path)
1069                 return -ENOMEM;
1070
1071         nolock = btrfs_is_free_space_inode(root, inode);
1072
1073         if (nolock)
1074                 trans = btrfs_join_transaction_nolock(root);
1075         else
1076                 trans = btrfs_join_transaction(root);
1077
1078         BUG_ON(IS_ERR(trans));
1079         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1080
1081         cow_start = (u64)-1;
1082         cur_offset = start;
1083         while (1) {
1084                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1085                                                cur_offset, 0);
1086                 BUG_ON(ret < 0);
1087                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1088                         leaf = path->nodes[0];
1089                         btrfs_item_key_to_cpu(leaf, &found_key,
1090                                               path->slots[0] - 1);
1091                         if (found_key.objectid == ino &&
1092                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1093                                 path->slots[0]--;
1094                 }
1095                 check_prev = 0;
1096 next_slot:
1097                 leaf = path->nodes[0];
1098                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1099                         ret = btrfs_next_leaf(root, path);
1100                         if (ret < 0)
1101                                 BUG_ON(1);
1102                         if (ret > 0)
1103                                 break;
1104                         leaf = path->nodes[0];
1105                 }
1106
1107                 nocow = 0;
1108                 disk_bytenr = 0;
1109                 num_bytes = 0;
1110                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1111
1112                 if (found_key.objectid > ino ||
1113                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1114                     found_key.offset > end)
1115                         break;
1116
1117                 if (found_key.offset > cur_offset) {
1118                         extent_end = found_key.offset;
1119                         extent_type = 0;
1120                         goto out_check;
1121                 }
1122
1123                 fi = btrfs_item_ptr(leaf, path->slots[0],
1124                                     struct btrfs_file_extent_item);
1125                 extent_type = btrfs_file_extent_type(leaf, fi);
1126
1127                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1128                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1129                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1130                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1131                         extent_end = found_key.offset +
1132                                 btrfs_file_extent_num_bytes(leaf, fi);
1133                         if (extent_end <= start) {
1134                                 path->slots[0]++;
1135                                 goto next_slot;
1136                         }
1137                         if (disk_bytenr == 0)
1138                                 goto out_check;
1139                         if (btrfs_file_extent_compression(leaf, fi) ||
1140                             btrfs_file_extent_encryption(leaf, fi) ||
1141                             btrfs_file_extent_other_encoding(leaf, fi))
1142                                 goto out_check;
1143                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1144                                 goto out_check;
1145                         if (btrfs_extent_readonly(root, disk_bytenr))
1146                                 goto out_check;
1147                         if (btrfs_cross_ref_exist(trans, root, ino,
1148                                                   found_key.offset -
1149                                                   extent_offset, disk_bytenr))
1150                                 goto out_check;
1151                         disk_bytenr += extent_offset;
1152                         disk_bytenr += cur_offset - found_key.offset;
1153                         num_bytes = min(end + 1, extent_end) - cur_offset;
1154                         /*
1155                          * force cow if csum exists in the range.
1156                          * this ensure that csum for a given extent are
1157                          * either valid or do not exist.
1158                          */
1159                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1160                                 goto out_check;
1161                         nocow = 1;
1162                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1163                         extent_end = found_key.offset +
1164                                 btrfs_file_extent_inline_len(leaf, fi);
1165                         extent_end = ALIGN(extent_end, root->sectorsize);
1166                 } else {
1167                         BUG_ON(1);
1168                 }
1169 out_check:
1170                 if (extent_end <= start) {
1171                         path->slots[0]++;
1172                         goto next_slot;
1173                 }
1174                 if (!nocow) {
1175                         if (cow_start == (u64)-1)
1176                                 cow_start = cur_offset;
1177                         cur_offset = extent_end;
1178                         if (cur_offset > end)
1179                                 break;
1180                         path->slots[0]++;
1181                         goto next_slot;
1182                 }
1183
1184                 btrfs_release_path(path);
1185                 if (cow_start != (u64)-1) {
1186                         ret = cow_file_range(inode, locked_page, cow_start,
1187                                         found_key.offset - 1, page_started,
1188                                         nr_written, 1);
1189                         BUG_ON(ret);
1190                         cow_start = (u64)-1;
1191                 }
1192
1193                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1194                         struct extent_map *em;
1195                         struct extent_map_tree *em_tree;
1196                         em_tree = &BTRFS_I(inode)->extent_tree;
1197                         em = alloc_extent_map();
1198                         BUG_ON(!em);
1199                         em->start = cur_offset;
1200                         em->orig_start = em->start;
1201                         em->len = num_bytes;
1202                         em->block_len = num_bytes;
1203                         em->block_start = disk_bytenr;
1204                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1205                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1206                         while (1) {
1207                                 write_lock(&em_tree->lock);
1208                                 ret = add_extent_mapping(em_tree, em);
1209                                 write_unlock(&em_tree->lock);
1210                                 if (ret != -EEXIST) {
1211                                         free_extent_map(em);
1212                                         break;
1213                                 }
1214                                 btrfs_drop_extent_cache(inode, em->start,
1215                                                 em->start + em->len - 1, 0);
1216                         }
1217                         type = BTRFS_ORDERED_PREALLOC;
1218                 } else {
1219                         type = BTRFS_ORDERED_NOCOW;
1220                 }
1221
1222                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1223                                                num_bytes, num_bytes, type);
1224                 BUG_ON(ret);
1225
1226                 if (root->root_key.objectid ==
1227                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1228                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1229                                                       num_bytes);
1230                         BUG_ON(ret);
1231                 }
1232
1233                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1234                                 cur_offset, cur_offset + num_bytes - 1,
1235                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1236                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1237                                 EXTENT_SET_PRIVATE2);
1238                 cur_offset = extent_end;
1239                 if (cur_offset > end)
1240                         break;
1241         }
1242         btrfs_release_path(path);
1243
1244         if (cur_offset <= end && cow_start == (u64)-1)
1245                 cow_start = cur_offset;
1246         if (cow_start != (u64)-1) {
1247                 ret = cow_file_range(inode, locked_page, cow_start, end,
1248                                      page_started, nr_written, 1);
1249                 BUG_ON(ret);
1250         }
1251
1252         if (nolock) {
1253                 ret = btrfs_end_transaction_nolock(trans, root);
1254                 BUG_ON(ret);
1255         } else {
1256                 ret = btrfs_end_transaction(trans, root);
1257                 BUG_ON(ret);
1258         }
1259         btrfs_free_path(path);
1260         return 0;
1261 }
1262
1263 /*
1264  * extent_io.c call back to do delayed allocation processing
1265  */
1266 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1267                               u64 start, u64 end, int *page_started,
1268                               unsigned long *nr_written)
1269 {
1270         int ret;
1271         struct btrfs_root *root = BTRFS_I(inode)->root;
1272
1273         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW)
1274                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1275                                          page_started, 1, nr_written);
1276         else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC)
1277                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1278                                          page_started, 0, nr_written);
1279         else if (!btrfs_test_opt(root, COMPRESS) &&
1280                  !(BTRFS_I(inode)->force_compress) &&
1281                  !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))
1282                 ret = cow_file_range(inode, locked_page, start, end,
1283                                       page_started, nr_written, 1);
1284         else
1285                 ret = cow_file_range_async(inode, locked_page, start, end,
1286                                            page_started, nr_written);
1287         return ret;
1288 }
1289
1290 static void btrfs_split_extent_hook(struct inode *inode,
1291                                     struct extent_state *orig, u64 split)
1292 {
1293         /* not delalloc, ignore it */
1294         if (!(orig->state & EXTENT_DELALLOC))
1295                 return;
1296
1297         spin_lock(&BTRFS_I(inode)->lock);
1298         BTRFS_I(inode)->outstanding_extents++;
1299         spin_unlock(&BTRFS_I(inode)->lock);
1300 }
1301
1302 /*
1303  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1304  * extents so we can keep track of new extents that are just merged onto old
1305  * extents, such as when we are doing sequential writes, so we can properly
1306  * account for the metadata space we'll need.
1307  */
1308 static void btrfs_merge_extent_hook(struct inode *inode,
1309                                     struct extent_state *new,
1310                                     struct extent_state *other)
1311 {
1312         /* not delalloc, ignore it */
1313         if (!(other->state & EXTENT_DELALLOC))
1314                 return;
1315
1316         spin_lock(&BTRFS_I(inode)->lock);
1317         BTRFS_I(inode)->outstanding_extents--;
1318         spin_unlock(&BTRFS_I(inode)->lock);
1319 }
1320
1321 /*
1322  * extent_io.c set_bit_hook, used to track delayed allocation
1323  * bytes in this file, and to maintain the list of inodes that
1324  * have pending delalloc work to be done.
1325  */
1326 static void btrfs_set_bit_hook(struct inode *inode,
1327                                struct extent_state *state, int *bits)
1328 {
1329
1330         /*
1331          * set_bit and clear bit hooks normally require _irqsave/restore
1332          * but in this case, we are only testing for the DELALLOC
1333          * bit, which is only set or cleared with irqs on
1334          */
1335         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1336                 struct btrfs_root *root = BTRFS_I(inode)->root;
1337                 u64 len = state->end + 1 - state->start;
1338                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1339
1340                 if (*bits & EXTENT_FIRST_DELALLOC) {
1341                         *bits &= ~EXTENT_FIRST_DELALLOC;
1342                 } else {
1343                         spin_lock(&BTRFS_I(inode)->lock);
1344                         BTRFS_I(inode)->outstanding_extents++;
1345                         spin_unlock(&BTRFS_I(inode)->lock);
1346                 }
1347
1348                 spin_lock(&root->fs_info->delalloc_lock);
1349                 BTRFS_I(inode)->delalloc_bytes += len;
1350                 root->fs_info->delalloc_bytes += len;
1351                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1352                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1353                                       &root->fs_info->delalloc_inodes);
1354                 }
1355                 spin_unlock(&root->fs_info->delalloc_lock);
1356         }
1357 }
1358
1359 /*
1360  * extent_io.c clear_bit_hook, see set_bit_hook for why
1361  */
1362 static void btrfs_clear_bit_hook(struct inode *inode,
1363                                  struct extent_state *state, int *bits)
1364 {
1365         /*
1366          * set_bit and clear bit hooks normally require _irqsave/restore
1367          * but in this case, we are only testing for the DELALLOC
1368          * bit, which is only set or cleared with irqs on
1369          */
1370         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1371                 struct btrfs_root *root = BTRFS_I(inode)->root;
1372                 u64 len = state->end + 1 - state->start;
1373                 bool do_list = !btrfs_is_free_space_inode(root, inode);
1374
1375                 if (*bits & EXTENT_FIRST_DELALLOC) {
1376                         *bits &= ~EXTENT_FIRST_DELALLOC;
1377                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1378                         spin_lock(&BTRFS_I(inode)->lock);
1379                         BTRFS_I(inode)->outstanding_extents--;
1380                         spin_unlock(&BTRFS_I(inode)->lock);
1381                 }
1382
1383                 if (*bits & EXTENT_DO_ACCOUNTING)
1384                         btrfs_delalloc_release_metadata(inode, len);
1385
1386                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1387                     && do_list)
1388                         btrfs_free_reserved_data_space(inode, len);
1389
1390                 spin_lock(&root->fs_info->delalloc_lock);
1391                 root->fs_info->delalloc_bytes -= len;
1392                 BTRFS_I(inode)->delalloc_bytes -= len;
1393
1394                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1395                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1396                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1397                 }
1398                 spin_unlock(&root->fs_info->delalloc_lock);
1399         }
1400 }
1401
1402 /*
1403  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1404  * we don't create bios that span stripes or chunks
1405  */
1406 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1407                          size_t size, struct bio *bio,
1408                          unsigned long bio_flags)
1409 {
1410         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1411         struct btrfs_mapping_tree *map_tree;
1412         u64 logical = (u64)bio->bi_sector << 9;
1413         u64 length = 0;
1414         u64 map_length;
1415         int ret;
1416
1417         if (bio_flags & EXTENT_BIO_COMPRESSED)
1418                 return 0;
1419
1420         length = bio->bi_size;
1421         map_tree = &root->fs_info->mapping_tree;
1422         map_length = length;
1423         ret = btrfs_map_block(map_tree, READ, logical,
1424                               &map_length, NULL, 0);
1425
1426         if (map_length < length + size)
1427                 return 1;
1428         return ret;
1429 }
1430
1431 /*
1432  * in order to insert checksums into the metadata in large chunks,
1433  * we wait until bio submission time.   All the pages in the bio are
1434  * checksummed and sums are attached onto the ordered extent record.
1435  *
1436  * At IO completion time the cums attached on the ordered extent record
1437  * are inserted into the btree
1438  */
1439 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1440                                     struct bio *bio, int mirror_num,
1441                                     unsigned long bio_flags,
1442                                     u64 bio_offset)
1443 {
1444         struct btrfs_root *root = BTRFS_I(inode)->root;
1445         int ret = 0;
1446
1447         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1448         BUG_ON(ret);
1449         return 0;
1450 }
1451
1452 /*
1453  * in order to insert checksums into the metadata in large chunks,
1454  * we wait until bio submission time.   All the pages in the bio are
1455  * checksummed and sums are attached onto the ordered extent record.
1456  *
1457  * At IO completion time the cums attached on the ordered extent record
1458  * are inserted into the btree
1459  */
1460 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1461                           int mirror_num, unsigned long bio_flags,
1462                           u64 bio_offset)
1463 {
1464         struct btrfs_root *root = BTRFS_I(inode)->root;
1465         return btrfs_map_bio(root, rw, bio, mirror_num, 1);
1466 }
1467
1468 /*
1469  * extent_io.c submission hook. This does the right thing for csum calculation
1470  * on write, or reading the csums from the tree before a read
1471  */
1472 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1473                           int mirror_num, unsigned long bio_flags,
1474                           u64 bio_offset)
1475 {
1476         struct btrfs_root *root = BTRFS_I(inode)->root;
1477         int ret = 0;
1478         int skip_sum;
1479
1480         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1481
1482         if (btrfs_is_free_space_inode(root, inode))
1483                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
1484         else
1485                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
1486         BUG_ON(ret);
1487
1488         if (!(rw & REQ_WRITE)) {
1489                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1490                         return btrfs_submit_compressed_read(inode, bio,
1491                                                     mirror_num, bio_flags);
1492                 } else if (!skip_sum) {
1493                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1494                         if (ret)
1495                                 return ret;
1496                 }
1497                 goto mapit;
1498         } else if (!skip_sum) {
1499                 /* csum items have already been cloned */
1500                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1501                         goto mapit;
1502                 /* we're doing a write, do the async checksumming */
1503                 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1504                                    inode, rw, bio, mirror_num,
1505                                    bio_flags, bio_offset,
1506                                    __btrfs_submit_bio_start,
1507                                    __btrfs_submit_bio_done);
1508         }
1509
1510 mapit:
1511         return btrfs_map_bio(root, rw, bio, mirror_num, 0);
1512 }
1513
1514 /*
1515  * given a list of ordered sums record them in the inode.  This happens
1516  * at IO completion time based on sums calculated at bio submission time.
1517  */
1518 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1519                              struct inode *inode, u64 file_offset,
1520                              struct list_head *list)
1521 {
1522         struct btrfs_ordered_sum *sum;
1523
1524         list_for_each_entry(sum, list, list) {
1525                 btrfs_csum_file_blocks(trans,
1526                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1527         }
1528         return 0;
1529 }
1530
1531 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1532                               struct extent_state **cached_state)
1533 {
1534         if ((end & (PAGE_CACHE_SIZE - 1)) == 0)
1535                 WARN_ON(1);
1536         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1537                                    cached_state, GFP_NOFS);
1538 }
1539
1540 /* see btrfs_writepage_start_hook for details on why this is required */
1541 struct btrfs_writepage_fixup {
1542         struct page *page;
1543         struct btrfs_work work;
1544 };
1545
1546 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1547 {
1548         struct btrfs_writepage_fixup *fixup;
1549         struct btrfs_ordered_extent *ordered;
1550         struct extent_state *cached_state = NULL;
1551         struct page *page;
1552         struct inode *inode;
1553         u64 page_start;
1554         u64 page_end;
1555
1556         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1557         page = fixup->page;
1558 again:
1559         lock_page(page);
1560         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1561                 ClearPageChecked(page);
1562                 goto out_page;
1563         }
1564
1565         inode = page->mapping->host;
1566         page_start = page_offset(page);
1567         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1568
1569         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1570                          &cached_state, GFP_NOFS);
1571
1572         /* already ordered? We're done */
1573         if (PagePrivate2(page))
1574                 goto out;
1575
1576         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1577         if (ordered) {
1578                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1579                                      page_end, &cached_state, GFP_NOFS);
1580                 unlock_page(page);
1581                 btrfs_start_ordered_extent(inode, ordered, 1);
1582                 goto again;
1583         }
1584
1585         BUG();
1586         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1587         ClearPageChecked(page);
1588 out:
1589         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1590                              &cached_state, GFP_NOFS);
1591 out_page:
1592         unlock_page(page);
1593         page_cache_release(page);
1594         kfree(fixup);
1595 }
1596
1597 /*
1598  * There are a few paths in the higher layers of the kernel that directly
1599  * set the page dirty bit without asking the filesystem if it is a
1600  * good idea.  This causes problems because we want to make sure COW
1601  * properly happens and the data=ordered rules are followed.
1602  *
1603  * In our case any range that doesn't have the ORDERED bit set
1604  * hasn't been properly setup for IO.  We kick off an async process
1605  * to fix it up.  The async helper will wait for ordered extents, set
1606  * the delalloc bit and make it safe to write the page.
1607  */
1608 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1609 {
1610         struct inode *inode = page->mapping->host;
1611         struct btrfs_writepage_fixup *fixup;
1612         struct btrfs_root *root = BTRFS_I(inode)->root;
1613
1614         /* this page is properly in the ordered list */
1615         if (TestClearPagePrivate2(page))
1616                 return 0;
1617
1618         if (PageChecked(page))
1619                 return -EAGAIN;
1620
1621         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1622         if (!fixup)
1623                 return -EAGAIN;
1624
1625         SetPageChecked(page);
1626         page_cache_get(page);
1627         fixup->work.func = btrfs_writepage_fixup_worker;
1628         fixup->page = page;
1629         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1630         return -EAGAIN;
1631 }
1632
1633 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1634                                        struct inode *inode, u64 file_pos,
1635                                        u64 disk_bytenr, u64 disk_num_bytes,
1636                                        u64 num_bytes, u64 ram_bytes,
1637                                        u8 compression, u8 encryption,
1638                                        u16 other_encoding, int extent_type)
1639 {
1640         struct btrfs_root *root = BTRFS_I(inode)->root;
1641         struct btrfs_file_extent_item *fi;
1642         struct btrfs_path *path;
1643         struct extent_buffer *leaf;
1644         struct btrfs_key ins;
1645         u64 hint;
1646         int ret;
1647
1648         path = btrfs_alloc_path();
1649         if (!path)
1650                 return -ENOMEM;
1651
1652         path->leave_spinning = 1;
1653
1654         /*
1655          * we may be replacing one extent in the tree with another.
1656          * The new extent is pinned in the extent map, and we don't want
1657          * to drop it from the cache until it is completely in the btree.
1658          *
1659          * So, tell btrfs_drop_extents to leave this extent in the cache.
1660          * the caller is expected to unpin it and allow it to be merged
1661          * with the others.
1662          */
1663         ret = btrfs_drop_extents(trans, inode, file_pos, file_pos + num_bytes,
1664                                  &hint, 0);
1665         BUG_ON(ret);
1666
1667         ins.objectid = btrfs_ino(inode);
1668         ins.offset = file_pos;
1669         ins.type = BTRFS_EXTENT_DATA_KEY;
1670         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1671         BUG_ON(ret);
1672         leaf = path->nodes[0];
1673         fi = btrfs_item_ptr(leaf, path->slots[0],
1674                             struct btrfs_file_extent_item);
1675         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1676         btrfs_set_file_extent_type(leaf, fi, extent_type);
1677         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1678         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1679         btrfs_set_file_extent_offset(leaf, fi, 0);
1680         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1681         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1682         btrfs_set_file_extent_compression(leaf, fi, compression);
1683         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1684         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1685
1686         btrfs_unlock_up_safe(path, 1);
1687         btrfs_set_lock_blocking(leaf);
1688
1689         btrfs_mark_buffer_dirty(leaf);
1690
1691         inode_add_bytes(inode, num_bytes);
1692
1693         ins.objectid = disk_bytenr;
1694         ins.offset = disk_num_bytes;
1695         ins.type = BTRFS_EXTENT_ITEM_KEY;
1696         ret = btrfs_alloc_reserved_file_extent(trans, root,
1697                                         root->root_key.objectid,
1698                                         btrfs_ino(inode), file_pos, &ins);
1699         BUG_ON(ret);
1700         btrfs_free_path(path);
1701
1702         return 0;
1703 }
1704
1705 /*
1706  * helper function for btrfs_finish_ordered_io, this
1707  * just reads in some of the csum leaves to prime them into ram
1708  * before we start the transaction.  It limits the amount of btree
1709  * reads required while inside the transaction.
1710  */
1711 /* as ordered data IO finishes, this gets called so we can finish
1712  * an ordered extent if the range of bytes in the file it covers are
1713  * fully written.
1714  */
1715 static int btrfs_finish_ordered_io(struct inode *inode, u64 start, u64 end)
1716 {
1717         struct btrfs_root *root = BTRFS_I(inode)->root;
1718         struct btrfs_trans_handle *trans = NULL;
1719         struct btrfs_ordered_extent *ordered_extent = NULL;
1720         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1721         struct extent_state *cached_state = NULL;
1722         int compress_type = 0;
1723         int ret;
1724         bool nolock;
1725
1726         ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
1727                                              end - start + 1);
1728         if (!ret)
1729                 return 0;
1730         BUG_ON(!ordered_extent);
1731
1732         nolock = btrfs_is_free_space_inode(root, inode);
1733
1734         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1735                 BUG_ON(!list_empty(&ordered_extent->list));
1736                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1737                 if (!ret) {
1738                         if (nolock)
1739                                 trans = btrfs_join_transaction_nolock(root);
1740                         else
1741                                 trans = btrfs_join_transaction(root);
1742                         BUG_ON(IS_ERR(trans));
1743                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1744                         ret = btrfs_update_inode(trans, root, inode);
1745                         BUG_ON(ret);
1746                 }
1747                 goto out;
1748         }
1749
1750         lock_extent_bits(io_tree, ordered_extent->file_offset,
1751                          ordered_extent->file_offset + ordered_extent->len - 1,
1752                          0, &cached_state, GFP_NOFS);
1753
1754         if (nolock)
1755                 trans = btrfs_join_transaction_nolock(root);
1756         else
1757                 trans = btrfs_join_transaction(root);
1758         BUG_ON(IS_ERR(trans));
1759         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1760
1761         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1762                 compress_type = ordered_extent->compress_type;
1763         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1764                 BUG_ON(compress_type);
1765                 ret = btrfs_mark_extent_written(trans, inode,
1766                                                 ordered_extent->file_offset,
1767                                                 ordered_extent->file_offset +
1768                                                 ordered_extent->len);
1769                 BUG_ON(ret);
1770         } else {
1771                 BUG_ON(root == root->fs_info->tree_root);
1772                 ret = insert_reserved_file_extent(trans, inode,
1773                                                 ordered_extent->file_offset,
1774                                                 ordered_extent->start,
1775                                                 ordered_extent->disk_len,
1776                                                 ordered_extent->len,
1777                                                 ordered_extent->len,
1778                                                 compress_type, 0, 0,
1779                                                 BTRFS_FILE_EXTENT_REG);
1780                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1781                                    ordered_extent->file_offset,
1782                                    ordered_extent->len);
1783                 BUG_ON(ret);
1784         }
1785         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1786                              ordered_extent->file_offset +
1787                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1788
1789         add_pending_csums(trans, inode, ordered_extent->file_offset,
1790                           &ordered_extent->list);
1791
1792         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1793         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1794                 ret = btrfs_update_inode(trans, root, inode);
1795                 BUG_ON(ret);
1796         }
1797         ret = 0;
1798 out:
1799         if (root != root->fs_info->tree_root)
1800                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1801         if (trans) {
1802                 if (nolock)
1803                         btrfs_end_transaction_nolock(trans, root);
1804                 else
1805                         btrfs_end_transaction(trans, root);
1806         }
1807
1808         /* once for us */
1809         btrfs_put_ordered_extent(ordered_extent);
1810         /* once for the tree */
1811         btrfs_put_ordered_extent(ordered_extent);
1812
1813         return 0;
1814 }
1815
1816 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
1817                                 struct extent_state *state, int uptodate)
1818 {
1819         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
1820
1821         ClearPagePrivate2(page);
1822         return btrfs_finish_ordered_io(page->mapping->host, start, end);
1823 }
1824
1825 /*
1826  * When IO fails, either with EIO or csum verification fails, we
1827  * try other mirrors that might have a good copy of the data.  This
1828  * io_failure_record is used to record state as we go through all the
1829  * mirrors.  If another mirror has good data, the page is set up to date
1830  * and things continue.  If a good mirror can't be found, the original
1831  * bio end_io callback is called to indicate things have failed.
1832  */
1833 struct io_failure_record {
1834         struct page *page;
1835         u64 start;
1836         u64 len;
1837         u64 logical;
1838         unsigned long bio_flags;
1839         int last_mirror;
1840 };
1841
1842 static int btrfs_io_failed_hook(struct bio *failed_bio,
1843                          struct page *page, u64 start, u64 end,
1844                          struct extent_state *state)
1845 {
1846         struct io_failure_record *failrec = NULL;
1847         u64 private;
1848         struct extent_map *em;
1849         struct inode *inode = page->mapping->host;
1850         struct extent_io_tree *failure_tree = &BTRFS_I(inode)->io_failure_tree;
1851         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
1852         struct bio *bio;
1853         int num_copies;
1854         int ret;
1855         int rw;
1856         u64 logical;
1857
1858         ret = get_state_private(failure_tree, start, &private);
1859         if (ret) {
1860                 failrec = kmalloc(sizeof(*failrec), GFP_NOFS);
1861                 if (!failrec)
1862                         return -ENOMEM;
1863                 failrec->start = start;
1864                 failrec->len = end - start + 1;
1865                 failrec->last_mirror = 0;
1866                 failrec->bio_flags = 0;
1867
1868                 read_lock(&em_tree->lock);
1869                 em = lookup_extent_mapping(em_tree, start, failrec->len);
1870                 if (em->start > start || em->start + em->len < start) {
1871                         free_extent_map(em);
1872                         em = NULL;
1873                 }
1874                 read_unlock(&em_tree->lock);
1875
1876                 if (IS_ERR_OR_NULL(em)) {
1877                         kfree(failrec);
1878                         return -EIO;
1879                 }
1880                 logical = start - em->start;
1881                 logical = em->block_start + logical;
1882                 if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
1883                         logical = em->block_start;
1884                         failrec->bio_flags = EXTENT_BIO_COMPRESSED;
1885                         extent_set_compress_type(&failrec->bio_flags,
1886                                                  em->compress_type);
1887                 }
1888                 failrec->logical = logical;
1889                 free_extent_map(em);
1890                 set_extent_bits(failure_tree, start, end, EXTENT_LOCKED |
1891                                 EXTENT_DIRTY, GFP_NOFS);
1892                 set_state_private(failure_tree, start,
1893                                  (u64)(unsigned long)failrec);
1894         } else {
1895                 failrec = (struct io_failure_record *)(unsigned long)private;
1896         }
1897         num_copies = btrfs_num_copies(
1898                               &BTRFS_I(inode)->root->fs_info->mapping_tree,
1899                               failrec->logical, failrec->len);
1900         failrec->last_mirror++;
1901         if (!state) {
1902                 spin_lock(&BTRFS_I(inode)->io_tree.lock);
1903                 state = find_first_extent_bit_state(&BTRFS_I(inode)->io_tree,
1904                                                     failrec->start,
1905                                                     EXTENT_LOCKED);
1906                 if (state && state->start != failrec->start)
1907                         state = NULL;
1908                 spin_unlock(&BTRFS_I(inode)->io_tree.lock);
1909         }
1910         if (!state || failrec->last_mirror > num_copies) {
1911                 set_state_private(failure_tree, failrec->start, 0);
1912                 clear_extent_bits(failure_tree, failrec->start,
1913                                   failrec->start + failrec->len - 1,
1914                                   EXTENT_LOCKED | EXTENT_DIRTY, GFP_NOFS);
1915                 kfree(failrec);
1916                 return -EIO;
1917         }
1918         bio = bio_alloc(GFP_NOFS, 1);
1919         bio->bi_private = state;
1920         bio->bi_end_io = failed_bio->bi_end_io;
1921         bio->bi_sector = failrec->logical >> 9;
1922         bio->bi_bdev = failed_bio->bi_bdev;
1923         bio->bi_size = 0;
1924
1925         bio_add_page(bio, page, failrec->len, start - page_offset(page));
1926         if (failed_bio->bi_rw & REQ_WRITE)
1927                 rw = WRITE;
1928         else
1929                 rw = READ;
1930
1931         ret = BTRFS_I(inode)->io_tree.ops->submit_bio_hook(inode, rw, bio,
1932                                                       failrec->last_mirror,
1933                                                       failrec->bio_flags, 0);
1934         return ret;
1935 }
1936
1937 /*
1938  * each time an IO finishes, we do a fast check in the IO failure tree
1939  * to see if we need to process or clean up an io_failure_record
1940  */
1941 static int btrfs_clean_io_failures(struct inode *inode, u64 start)
1942 {
1943         u64 private;
1944         u64 private_failure;
1945         struct io_failure_record *failure;
1946         int ret;
1947
1948         private = 0;
1949         if (count_range_bits(&BTRFS_I(inode)->io_failure_tree, &private,
1950                              (u64)-1, 1, EXTENT_DIRTY, 0)) {
1951                 ret = get_state_private(&BTRFS_I(inode)->io_failure_tree,
1952                                         start, &private_failure);
1953                 if (ret == 0) {
1954                         failure = (struct io_failure_record *)(unsigned long)
1955                                    private_failure;
1956                         set_state_private(&BTRFS_I(inode)->io_failure_tree,
1957                                           failure->start, 0);
1958                         clear_extent_bits(&BTRFS_I(inode)->io_failure_tree,
1959                                           failure->start,
1960                                           failure->start + failure->len - 1,
1961                                           EXTENT_DIRTY | EXTENT_LOCKED,
1962                                           GFP_NOFS);
1963                         kfree(failure);
1964                 }
1965         }
1966         return 0;
1967 }
1968
1969 /*
1970  * when reads are done, we need to check csums to verify the data is correct
1971  * if there's a match, we allow the bio to finish.  If not, we go through
1972  * the io_failure_record routines to find good copies
1973  */
1974 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
1975                                struct extent_state *state)
1976 {
1977         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
1978         struct inode *inode = page->mapping->host;
1979         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1980         char *kaddr;
1981         u64 private = ~(u32)0;
1982         int ret;
1983         struct btrfs_root *root = BTRFS_I(inode)->root;
1984         u32 csum = ~(u32)0;
1985
1986         if (PageChecked(page)) {
1987                 ClearPageChecked(page);
1988                 goto good;
1989         }
1990
1991         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
1992                 goto good;
1993
1994         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
1995             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
1996                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
1997                                   GFP_NOFS);
1998                 return 0;
1999         }
2000
2001         if (state && state->start == start) {
2002                 private = state->private;
2003                 ret = 0;
2004         } else {
2005                 ret = get_state_private(io_tree, start, &private);
2006         }
2007         kaddr = kmap_atomic(page, KM_USER0);
2008         if (ret)
2009                 goto zeroit;
2010
2011         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2012         btrfs_csum_final(csum, (char *)&csum);
2013         if (csum != private)
2014                 goto zeroit;
2015
2016         kunmap_atomic(kaddr, KM_USER0);
2017 good:
2018         /* if the io failure tree for this inode is non-empty,
2019          * check to see if we've recovered from a failed IO
2020          */
2021         btrfs_clean_io_failures(inode, start);
2022         return 0;
2023
2024 zeroit:
2025         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2026                        "private %llu\n",
2027                        (unsigned long long)btrfs_ino(page->mapping->host),
2028                        (unsigned long long)start, csum,
2029                        (unsigned long long)private);
2030         memset(kaddr + offset, 1, end - start + 1);
2031         flush_dcache_page(page);
2032         kunmap_atomic(kaddr, KM_USER0);
2033         if (private == 0)
2034                 return 0;
2035         return -EIO;
2036 }
2037
2038 struct delayed_iput {
2039         struct list_head list;
2040         struct inode *inode;
2041 };
2042
2043 void btrfs_add_delayed_iput(struct inode *inode)
2044 {
2045         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2046         struct delayed_iput *delayed;
2047
2048         if (atomic_add_unless(&inode->i_count, -1, 1))
2049                 return;
2050
2051         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2052         delayed->inode = inode;
2053
2054         spin_lock(&fs_info->delayed_iput_lock);
2055         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2056         spin_unlock(&fs_info->delayed_iput_lock);
2057 }
2058
2059 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2060 {
2061         LIST_HEAD(list);
2062         struct btrfs_fs_info *fs_info = root->fs_info;
2063         struct delayed_iput *delayed;
2064         int empty;
2065
2066         spin_lock(&fs_info->delayed_iput_lock);
2067         empty = list_empty(&fs_info->delayed_iputs);
2068         spin_unlock(&fs_info->delayed_iput_lock);
2069         if (empty)
2070                 return;
2071
2072         down_read(&root->fs_info->cleanup_work_sem);
2073         spin_lock(&fs_info->delayed_iput_lock);
2074         list_splice_init(&fs_info->delayed_iputs, &list);
2075         spin_unlock(&fs_info->delayed_iput_lock);
2076
2077         while (!list_empty(&list)) {
2078                 delayed = list_entry(list.next, struct delayed_iput, list);
2079                 list_del(&delayed->list);
2080                 iput(delayed->inode);
2081                 kfree(delayed);
2082         }
2083         up_read(&root->fs_info->cleanup_work_sem);
2084 }
2085
2086 enum btrfs_orphan_cleanup_state {
2087         ORPHAN_CLEANUP_STARTED  = 1,
2088         ORPHAN_CLEANUP_DONE     = 2,
2089 };
2090
2091 /*
2092  * This is called in transaction commmit time. If there are no orphan
2093  * files in the subvolume, it removes orphan item and frees block_rsv
2094  * structure.
2095  */
2096 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2097                               struct btrfs_root *root)
2098 {
2099         int ret;
2100
2101         if (!list_empty(&root->orphan_list) ||
2102             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2103                 return;
2104
2105         if (root->orphan_item_inserted &&
2106             btrfs_root_refs(&root->root_item) > 0) {
2107                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2108                                             root->root_key.objectid);
2109                 BUG_ON(ret);
2110                 root->orphan_item_inserted = 0;
2111         }
2112
2113         if (root->orphan_block_rsv) {
2114                 WARN_ON(root->orphan_block_rsv->size > 0);
2115                 btrfs_free_block_rsv(root, root->orphan_block_rsv);
2116                 root->orphan_block_rsv = NULL;
2117         }
2118 }
2119
2120 /*
2121  * This creates an orphan entry for the given inode in case something goes
2122  * wrong in the middle of an unlink/truncate.
2123  *
2124  * NOTE: caller of this function should reserve 5 units of metadata for
2125  *       this function.
2126  */
2127 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2128 {
2129         struct btrfs_root *root = BTRFS_I(inode)->root;
2130         struct btrfs_block_rsv *block_rsv = NULL;
2131         int reserve = 0;
2132         int insert = 0;
2133         int ret;
2134
2135         if (!root->orphan_block_rsv) {
2136                 block_rsv = btrfs_alloc_block_rsv(root);
2137                 if (!block_rsv)
2138                         return -ENOMEM;
2139         }
2140
2141         spin_lock(&root->orphan_lock);
2142         if (!root->orphan_block_rsv) {
2143                 root->orphan_block_rsv = block_rsv;
2144         } else if (block_rsv) {
2145                 btrfs_free_block_rsv(root, block_rsv);
2146                 block_rsv = NULL;
2147         }
2148
2149         if (list_empty(&BTRFS_I(inode)->i_orphan)) {
2150                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2151 #if 0
2152                 /*
2153                  * For proper ENOSPC handling, we should do orphan
2154                  * cleanup when mounting. But this introduces backward
2155                  * compatibility issue.
2156                  */
2157                 if (!xchg(&root->orphan_item_inserted, 1))
2158                         insert = 2;
2159                 else
2160                         insert = 1;
2161 #endif
2162                 insert = 1;
2163         }
2164
2165         if (!BTRFS_I(inode)->orphan_meta_reserved) {
2166                 BTRFS_I(inode)->orphan_meta_reserved = 1;
2167                 reserve = 1;
2168         }
2169         spin_unlock(&root->orphan_lock);
2170
2171         /* grab metadata reservation from transaction handle */
2172         if (reserve) {
2173                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2174                 BUG_ON(ret);
2175         }
2176
2177         /* insert an orphan item to track this unlinked/truncated file */
2178         if (insert >= 1) {
2179                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2180                 BUG_ON(ret);
2181         }
2182
2183         /* insert an orphan item to track subvolume contains orphan files */
2184         if (insert >= 2) {
2185                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2186                                                root->root_key.objectid);
2187                 BUG_ON(ret);
2188         }
2189         return 0;
2190 }
2191
2192 /*
2193  * We have done the truncate/delete so we can go ahead and remove the orphan
2194  * item for this particular inode.
2195  */
2196 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2197 {
2198         struct btrfs_root *root = BTRFS_I(inode)->root;
2199         int delete_item = 0;
2200         int release_rsv = 0;
2201         int ret = 0;
2202
2203         spin_lock(&root->orphan_lock);
2204         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
2205                 list_del_init(&BTRFS_I(inode)->i_orphan);
2206                 delete_item = 1;
2207         }
2208
2209         if (BTRFS_I(inode)->orphan_meta_reserved) {
2210                 BTRFS_I(inode)->orphan_meta_reserved = 0;
2211                 release_rsv = 1;
2212         }
2213         spin_unlock(&root->orphan_lock);
2214
2215         if (trans && delete_item) {
2216                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2217                 BUG_ON(ret);
2218         }
2219
2220         if (release_rsv)
2221                 btrfs_orphan_release_metadata(inode);
2222
2223         return 0;
2224 }
2225
2226 /*
2227  * this cleans up any orphans that may be left on the list from the last use
2228  * of this root.
2229  */
2230 int btrfs_orphan_cleanup(struct btrfs_root *root)
2231 {
2232         struct btrfs_path *path;
2233         struct extent_buffer *leaf;
2234         struct btrfs_key key, found_key;
2235         struct btrfs_trans_handle *trans;
2236         struct inode *inode;
2237         u64 last_objectid = 0;
2238         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2239
2240         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2241                 return 0;
2242
2243         path = btrfs_alloc_path();
2244         if (!path) {
2245                 ret = -ENOMEM;
2246                 goto out;
2247         }
2248         path->reada = -1;
2249
2250         key.objectid = BTRFS_ORPHAN_OBJECTID;
2251         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2252         key.offset = (u64)-1;
2253
2254         while (1) {
2255                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2256                 if (ret < 0)
2257                         goto out;
2258
2259                 /*
2260                  * if ret == 0 means we found what we were searching for, which
2261                  * is weird, but possible, so only screw with path if we didn't
2262                  * find the key and see if we have stuff that matches
2263                  */
2264                 if (ret > 0) {
2265                         ret = 0;
2266                         if (path->slots[0] == 0)
2267                                 break;
2268                         path->slots[0]--;
2269                 }
2270
2271                 /* pull out the item */
2272                 leaf = path->nodes[0];
2273                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2274
2275                 /* make sure the item matches what we want */
2276                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2277                         break;
2278                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2279                         break;
2280
2281                 /* release the path since we're done with it */
2282                 btrfs_release_path(path);
2283
2284                 /*
2285                  * this is where we are basically btrfs_lookup, without the
2286                  * crossing root thing.  we store the inode number in the
2287                  * offset of the orphan item.
2288                  */
2289
2290                 if (found_key.offset == last_objectid) {
2291                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2292                                "stopping orphan cleanup\n");
2293                         ret = -EINVAL;
2294                         goto out;
2295                 }
2296
2297                 last_objectid = found_key.offset;
2298
2299                 found_key.objectid = found_key.offset;
2300                 found_key.type = BTRFS_INODE_ITEM_KEY;
2301                 found_key.offset = 0;
2302                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2303                 ret = PTR_RET(inode);
2304                 if (ret && ret != -ESTALE)
2305                         goto out;
2306
2307                 /*
2308                  * Inode is already gone but the orphan item is still there,
2309                  * kill the orphan item.
2310                  */
2311                 if (ret == -ESTALE) {
2312                         trans = btrfs_start_transaction(root, 1);
2313                         if (IS_ERR(trans)) {
2314                                 ret = PTR_ERR(trans);
2315                                 goto out;
2316                         }
2317                         ret = btrfs_del_orphan_item(trans, root,
2318                                                     found_key.objectid);
2319                         BUG_ON(ret);
2320                         btrfs_end_transaction(trans, root);
2321                         continue;
2322                 }
2323
2324                 /*
2325                  * add this inode to the orphan list so btrfs_orphan_del does
2326                  * the proper thing when we hit it
2327                  */
2328                 spin_lock(&root->orphan_lock);
2329                 list_add(&BTRFS_I(inode)->i_orphan, &root->orphan_list);
2330                 spin_unlock(&root->orphan_lock);
2331
2332                 /* if we have links, this was a truncate, lets do that */
2333                 if (inode->i_nlink) {
2334                         if (!S_ISREG(inode->i_mode)) {
2335                                 WARN_ON(1);
2336                                 iput(inode);
2337                                 continue;
2338                         }
2339                         nr_truncate++;
2340                         ret = btrfs_truncate(inode);
2341                 } else {
2342                         nr_unlink++;
2343                 }
2344
2345                 /* this will do delete_inode and everything for us */
2346                 iput(inode);
2347                 if (ret)
2348                         goto out;
2349         }
2350         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2351
2352         if (root->orphan_block_rsv)
2353                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2354                                         (u64)-1);
2355
2356         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2357                 trans = btrfs_join_transaction(root);
2358                 if (!IS_ERR(trans))
2359                         btrfs_end_transaction(trans, root);
2360         }
2361
2362         if (nr_unlink)
2363                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2364         if (nr_truncate)
2365                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2366
2367 out:
2368         if (ret)
2369                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2370         btrfs_free_path(path);
2371         return ret;
2372 }
2373
2374 /*
2375  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2376  * don't find any xattrs, we know there can't be any acls.
2377  *
2378  * slot is the slot the inode is in, objectid is the objectid of the inode
2379  */
2380 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2381                                           int slot, u64 objectid)
2382 {
2383         u32 nritems = btrfs_header_nritems(leaf);
2384         struct btrfs_key found_key;
2385         int scanned = 0;
2386
2387         slot++;
2388         while (slot < nritems) {
2389                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2390
2391                 /* we found a different objectid, there must not be acls */
2392                 if (found_key.objectid != objectid)
2393                         return 0;
2394
2395                 /* we found an xattr, assume we've got an acl */
2396                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2397                         return 1;
2398
2399                 /*
2400                  * we found a key greater than an xattr key, there can't
2401                  * be any acls later on
2402                  */
2403                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2404                         return 0;
2405
2406                 slot++;
2407                 scanned++;
2408
2409                 /*
2410                  * it goes inode, inode backrefs, xattrs, extents,
2411                  * so if there are a ton of hard links to an inode there can
2412                  * be a lot of backrefs.  Don't waste time searching too hard,
2413                  * this is just an optimization
2414                  */
2415                 if (scanned >= 8)
2416                         break;
2417         }
2418         /* we hit the end of the leaf before we found an xattr or
2419          * something larger than an xattr.  We have to assume the inode
2420          * has acls
2421          */
2422         return 1;
2423 }
2424
2425 /*
2426  * read an inode from the btree into the in-memory inode
2427  */
2428 static void btrfs_read_locked_inode(struct inode *inode)
2429 {
2430         struct btrfs_path *path;
2431         struct extent_buffer *leaf;
2432         struct btrfs_inode_item *inode_item;
2433         struct btrfs_timespec *tspec;
2434         struct btrfs_root *root = BTRFS_I(inode)->root;
2435         struct btrfs_key location;
2436         int maybe_acls;
2437         u32 rdev;
2438         int ret;
2439         bool filled = false;
2440
2441         ret = btrfs_fill_inode(inode, &rdev);
2442         if (!ret)
2443                 filled = true;
2444
2445         path = btrfs_alloc_path();
2446         if (!path)
2447                 goto make_bad;
2448
2449         path->leave_spinning = 1;
2450         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2451
2452         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2453         if (ret)
2454                 goto make_bad;
2455
2456         leaf = path->nodes[0];
2457
2458         if (filled)
2459                 goto cache_acl;
2460
2461         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2462                                     struct btrfs_inode_item);
2463         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2464         inode->i_nlink = btrfs_inode_nlink(leaf, inode_item);
2465         inode->i_uid = btrfs_inode_uid(leaf, inode_item);
2466         inode->i_gid = btrfs_inode_gid(leaf, inode_item);
2467         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2468
2469         tspec = btrfs_inode_atime(inode_item);
2470         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2471         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2472
2473         tspec = btrfs_inode_mtime(inode_item);
2474         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2475         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2476
2477         tspec = btrfs_inode_ctime(inode_item);
2478         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2479         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2480
2481         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2482         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2483         BTRFS_I(inode)->sequence = btrfs_inode_sequence(leaf, inode_item);
2484         inode->i_generation = BTRFS_I(inode)->generation;
2485         inode->i_rdev = 0;
2486         rdev = btrfs_inode_rdev(leaf, inode_item);
2487
2488         BTRFS_I(inode)->index_cnt = (u64)-1;
2489         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2490 cache_acl:
2491         /*
2492          * try to precache a NULL acl entry for files that don't have
2493          * any xattrs or acls
2494          */
2495         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2496                                            btrfs_ino(inode));
2497         if (!maybe_acls)
2498                 cache_no_acl(inode);
2499
2500         btrfs_free_path(path);
2501
2502         switch (inode->i_mode & S_IFMT) {
2503         case S_IFREG:
2504                 inode->i_mapping->a_ops = &btrfs_aops;
2505                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2506                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2507                 inode->i_fop = &btrfs_file_operations;
2508                 inode->i_op = &btrfs_file_inode_operations;
2509                 break;
2510         case S_IFDIR:
2511                 inode->i_fop = &btrfs_dir_file_operations;
2512                 if (root == root->fs_info->tree_root)
2513                         inode->i_op = &btrfs_dir_ro_inode_operations;
2514                 else
2515                         inode->i_op = &btrfs_dir_inode_operations;
2516                 break;
2517         case S_IFLNK:
2518                 inode->i_op = &btrfs_symlink_inode_operations;
2519                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2520                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2521                 break;
2522         default:
2523                 inode->i_op = &btrfs_special_inode_operations;
2524                 init_special_inode(inode, inode->i_mode, rdev);
2525                 break;
2526         }
2527
2528         btrfs_update_iflags(inode);
2529         return;
2530
2531 make_bad:
2532         btrfs_free_path(path);
2533         make_bad_inode(inode);
2534 }
2535
2536 /*
2537  * given a leaf and an inode, copy the inode fields into the leaf
2538  */
2539 static void fill_inode_item(struct btrfs_trans_handle *trans,
2540                             struct extent_buffer *leaf,
2541                             struct btrfs_inode_item *item,
2542                             struct inode *inode)
2543 {
2544         btrfs_set_inode_uid(leaf, item, inode->i_uid);
2545         btrfs_set_inode_gid(leaf, item, inode->i_gid);
2546         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2547         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2548         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2549
2550         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2551                                inode->i_atime.tv_sec);
2552         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2553                                 inode->i_atime.tv_nsec);
2554
2555         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2556                                inode->i_mtime.tv_sec);
2557         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2558                                 inode->i_mtime.tv_nsec);
2559
2560         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2561                                inode->i_ctime.tv_sec);
2562         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2563                                 inode->i_ctime.tv_nsec);
2564
2565         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2566         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2567         btrfs_set_inode_sequence(leaf, item, BTRFS_I(inode)->sequence);
2568         btrfs_set_inode_transid(leaf, item, trans->transid);
2569         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2570         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2571         btrfs_set_inode_block_group(leaf, item, 0);
2572 }
2573
2574 /*
2575  * copy everything in the in-memory inode into the btree.
2576  */
2577 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2578                                 struct btrfs_root *root, struct inode *inode)
2579 {
2580         struct btrfs_inode_item *inode_item;
2581         struct btrfs_path *path;
2582         struct extent_buffer *leaf;
2583         int ret;
2584
2585         /*
2586          * If the inode is a free space inode, we can deadlock during commit
2587          * if we put it into the delayed code.
2588          *
2589          * The data relocation inode should also be directly updated
2590          * without delay
2591          */
2592         if (!btrfs_is_free_space_inode(root, inode)
2593             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2594                 ret = btrfs_delayed_update_inode(trans, root, inode);
2595                 if (!ret)
2596                         btrfs_set_inode_last_trans(trans, inode);
2597                 return ret;
2598         }
2599
2600         path = btrfs_alloc_path();
2601         if (!path)
2602                 return -ENOMEM;
2603
2604         path->leave_spinning = 1;
2605         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2606                                  1);
2607         if (ret) {
2608                 if (ret > 0)
2609                         ret = -ENOENT;
2610                 goto failed;
2611         }
2612
2613         btrfs_unlock_up_safe(path, 1);
2614         leaf = path->nodes[0];
2615         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2616                                     struct btrfs_inode_item);
2617
2618         fill_inode_item(trans, leaf, inode_item, inode);
2619         btrfs_mark_buffer_dirty(leaf);
2620         btrfs_set_inode_last_trans(trans, inode);
2621         ret = 0;
2622 failed:
2623         btrfs_free_path(path);
2624         return ret;
2625 }
2626
2627 /*
2628  * unlink helper that gets used here in inode.c and in the tree logging
2629  * recovery code.  It remove a link in a directory with a given name, and
2630  * also drops the back refs in the inode to the directory
2631  */
2632 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2633                                 struct btrfs_root *root,
2634                                 struct inode *dir, struct inode *inode,
2635                                 const char *name, int name_len)
2636 {
2637         struct btrfs_path *path;
2638         int ret = 0;
2639         struct extent_buffer *leaf;
2640         struct btrfs_dir_item *di;
2641         struct btrfs_key key;
2642         u64 index;
2643         u64 ino = btrfs_ino(inode);
2644         u64 dir_ino = btrfs_ino(dir);
2645
2646         path = btrfs_alloc_path();
2647         if (!path) {
2648                 ret = -ENOMEM;
2649                 goto out;
2650         }
2651
2652         path->leave_spinning = 1;
2653         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2654                                     name, name_len, -1);
2655         if (IS_ERR(di)) {
2656                 ret = PTR_ERR(di);
2657                 goto err;
2658         }
2659         if (!di) {
2660                 ret = -ENOENT;
2661                 goto err;
2662         }
2663         leaf = path->nodes[0];
2664         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2665         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2666         if (ret)
2667                 goto err;
2668         btrfs_release_path(path);
2669
2670         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2671                                   dir_ino, &index);
2672         if (ret) {
2673                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2674                        "inode %llu parent %llu\n", name_len, name,
2675                        (unsigned long long)ino, (unsigned long long)dir_ino);
2676                 goto err;
2677         }
2678
2679         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2680         if (ret)
2681                 goto err;
2682
2683         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2684                                          inode, dir_ino);
2685         BUG_ON(ret != 0 && ret != -ENOENT);
2686
2687         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2688                                            dir, index);
2689         if (ret == -ENOENT)
2690                 ret = 0;
2691 err:
2692         btrfs_free_path(path);
2693         if (ret)
2694                 goto out;
2695
2696         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2697         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2698         btrfs_update_inode(trans, root, dir);
2699 out:
2700         return ret;
2701 }
2702
2703 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2704                        struct btrfs_root *root,
2705                        struct inode *dir, struct inode *inode,
2706                        const char *name, int name_len)
2707 {
2708         int ret;
2709         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2710         if (!ret) {
2711                 btrfs_drop_nlink(inode);
2712                 ret = btrfs_update_inode(trans, root, inode);
2713         }
2714         return ret;
2715 }
2716
2717
2718 /* helper to check if there is any shared block in the path */
2719 static int check_path_shared(struct btrfs_root *root,
2720                              struct btrfs_path *path)
2721 {
2722         struct extent_buffer *eb;
2723         int level;
2724         u64 refs = 1;
2725
2726         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2727                 int ret;
2728
2729                 if (!path->nodes[level])
2730                         break;
2731                 eb = path->nodes[level];
2732                 if (!btrfs_block_can_be_shared(root, eb))
2733                         continue;
2734                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2735                                                &refs, NULL);
2736                 if (refs > 1)
2737                         return 1;
2738         }
2739         return 0;
2740 }
2741
2742 /*
2743  * helper to start transaction for unlink and rmdir.
2744  *
2745  * unlink and rmdir are special in btrfs, they do not always free space.
2746  * so in enospc case, we should make sure they will free space before
2747  * allowing them to use the global metadata reservation.
2748  */
2749 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2750                                                        struct dentry *dentry)
2751 {
2752         struct btrfs_trans_handle *trans;
2753         struct btrfs_root *root = BTRFS_I(dir)->root;
2754         struct btrfs_path *path;
2755         struct btrfs_inode_ref *ref;
2756         struct btrfs_dir_item *di;
2757         struct inode *inode = dentry->d_inode;
2758         u64 index;
2759         int check_link = 1;
2760         int err = -ENOSPC;
2761         int ret;
2762         u64 ino = btrfs_ino(inode);
2763         u64 dir_ino = btrfs_ino(dir);
2764
2765         /*
2766          * 1 for the possible orphan item
2767          * 1 for the dir item
2768          * 1 for the dir index
2769          * 1 for the inode ref
2770          * 1 for the inode ref in the tree log
2771          * 2 for the dir entries in the log
2772          * 1 for the inode
2773          */
2774         trans = btrfs_start_transaction(root, 8);
2775         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2776                 return trans;
2777
2778         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2779                 return ERR_PTR(-ENOSPC);
2780
2781         /* check if there is someone else holds reference */
2782         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2783                 return ERR_PTR(-ENOSPC);
2784
2785         if (atomic_read(&inode->i_count) > 2)
2786                 return ERR_PTR(-ENOSPC);
2787
2788         if (xchg(&root->fs_info->enospc_unlink, 1))
2789                 return ERR_PTR(-ENOSPC);
2790
2791         path = btrfs_alloc_path();
2792         if (!path) {
2793                 root->fs_info->enospc_unlink = 0;
2794                 return ERR_PTR(-ENOMEM);
2795         }
2796
2797         /* 1 for the orphan item */
2798         trans = btrfs_start_transaction(root, 1);
2799         if (IS_ERR(trans)) {
2800                 btrfs_free_path(path);
2801                 root->fs_info->enospc_unlink = 0;
2802                 return trans;
2803         }
2804
2805         path->skip_locking = 1;
2806         path->search_commit_root = 1;
2807
2808         ret = btrfs_lookup_inode(trans, root, path,
2809                                 &BTRFS_I(dir)->location, 0);
2810         if (ret < 0) {
2811                 err = ret;
2812                 goto out;
2813         }
2814         if (ret == 0) {
2815                 if (check_path_shared(root, path))
2816                         goto out;
2817         } else {
2818                 check_link = 0;
2819         }
2820         btrfs_release_path(path);
2821
2822         ret = btrfs_lookup_inode(trans, root, path,
2823                                 &BTRFS_I(inode)->location, 0);
2824         if (ret < 0) {
2825                 err = ret;
2826                 goto out;
2827         }
2828         if (ret == 0) {
2829                 if (check_path_shared(root, path))
2830                         goto out;
2831         } else {
2832                 check_link = 0;
2833         }
2834         btrfs_release_path(path);
2835
2836         if (ret == 0 && S_ISREG(inode->i_mode)) {
2837                 ret = btrfs_lookup_file_extent(trans, root, path,
2838                                                ino, (u64)-1, 0);
2839                 if (ret < 0) {
2840                         err = ret;
2841                         goto out;
2842                 }
2843                 BUG_ON(ret == 0);
2844                 if (check_path_shared(root, path))
2845                         goto out;
2846                 btrfs_release_path(path);
2847         }
2848
2849         if (!check_link) {
2850                 err = 0;
2851                 goto out;
2852         }
2853
2854         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2855                                 dentry->d_name.name, dentry->d_name.len, 0);
2856         if (IS_ERR(di)) {
2857                 err = PTR_ERR(di);
2858                 goto out;
2859         }
2860         if (di) {
2861                 if (check_path_shared(root, path))
2862                         goto out;
2863         } else {
2864                 err = 0;
2865                 goto out;
2866         }
2867         btrfs_release_path(path);
2868
2869         ref = btrfs_lookup_inode_ref(trans, root, path,
2870                                 dentry->d_name.name, dentry->d_name.len,
2871                                 ino, dir_ino, 0);
2872         if (IS_ERR(ref)) {
2873                 err = PTR_ERR(ref);
2874                 goto out;
2875         }
2876         BUG_ON(!ref);
2877         if (check_path_shared(root, path))
2878                 goto out;
2879         index = btrfs_inode_ref_index(path->nodes[0], ref);
2880         btrfs_release_path(path);
2881
2882         /*
2883          * This is a commit root search, if we can lookup inode item and other
2884          * relative items in the commit root, it means the transaction of
2885          * dir/file creation has been committed, and the dir index item that we
2886          * delay to insert has also been inserted into the commit root. So
2887          * we needn't worry about the delayed insertion of the dir index item
2888          * here.
2889          */
2890         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
2891                                 dentry->d_name.name, dentry->d_name.len, 0);
2892         if (IS_ERR(di)) {
2893                 err = PTR_ERR(di);
2894                 goto out;
2895         }
2896         BUG_ON(ret == -ENOENT);
2897         if (check_path_shared(root, path))
2898                 goto out;
2899
2900         err = 0;
2901 out:
2902         btrfs_free_path(path);
2903         /* Migrate the orphan reservation over */
2904         if (!err)
2905                 err = btrfs_block_rsv_migrate(trans->block_rsv,
2906                                 &root->fs_info->global_block_rsv,
2907                                 trans->bytes_reserved);
2908
2909         if (err) {
2910                 btrfs_end_transaction(trans, root);
2911                 root->fs_info->enospc_unlink = 0;
2912                 return ERR_PTR(err);
2913         }
2914
2915         trans->block_rsv = &root->fs_info->global_block_rsv;
2916         return trans;
2917 }
2918
2919 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
2920                                struct btrfs_root *root)
2921 {
2922         if (trans->block_rsv == &root->fs_info->global_block_rsv) {
2923                 btrfs_block_rsv_release(root, trans->block_rsv,
2924                                         trans->bytes_reserved);
2925                 trans->block_rsv = &root->fs_info->trans_block_rsv;
2926                 BUG_ON(!root->fs_info->enospc_unlink);
2927                 root->fs_info->enospc_unlink = 0;
2928         }
2929         btrfs_end_transaction_throttle(trans, root);
2930 }
2931
2932 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
2933 {
2934         struct btrfs_root *root = BTRFS_I(dir)->root;
2935         struct btrfs_trans_handle *trans;
2936         struct inode *inode = dentry->d_inode;
2937         int ret;
2938         unsigned long nr = 0;
2939
2940         trans = __unlink_start_trans(dir, dentry);
2941         if (IS_ERR(trans))
2942                 return PTR_ERR(trans);
2943
2944         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
2945
2946         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
2947                                  dentry->d_name.name, dentry->d_name.len);
2948         if (ret)
2949                 goto out;
2950
2951         if (inode->i_nlink == 0) {
2952                 ret = btrfs_orphan_add(trans, inode);
2953                 if (ret)
2954                         goto out;
2955         }
2956
2957 out:
2958         nr = trans->blocks_used;
2959         __unlink_end_trans(trans, root);
2960         btrfs_btree_balance_dirty(root, nr);
2961         return ret;
2962 }
2963
2964 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
2965                         struct btrfs_root *root,
2966                         struct inode *dir, u64 objectid,
2967                         const char *name, int name_len)
2968 {
2969         struct btrfs_path *path;
2970         struct extent_buffer *leaf;
2971         struct btrfs_dir_item *di;
2972         struct btrfs_key key;
2973         u64 index;
2974         int ret;
2975         u64 dir_ino = btrfs_ino(dir);
2976
2977         path = btrfs_alloc_path();
2978         if (!path)
2979                 return -ENOMEM;
2980
2981         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2982                                    name, name_len, -1);
2983         BUG_ON(IS_ERR_OR_NULL(di));
2984
2985         leaf = path->nodes[0];
2986         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2987         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
2988         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2989         BUG_ON(ret);
2990         btrfs_release_path(path);
2991
2992         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
2993                                  objectid, root->root_key.objectid,
2994                                  dir_ino, &index, name, name_len);
2995         if (ret < 0) {
2996                 BUG_ON(ret != -ENOENT);
2997                 di = btrfs_search_dir_index_item(root, path, dir_ino,
2998                                                  name, name_len);
2999                 BUG_ON(IS_ERR_OR_NULL(di));
3000
3001                 leaf = path->nodes[0];
3002                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3003                 btrfs_release_path(path);
3004                 index = key.offset;
3005         }
3006         btrfs_release_path(path);
3007
3008         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3009         BUG_ON(ret);
3010
3011         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3012         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3013         ret = btrfs_update_inode(trans, root, dir);
3014         BUG_ON(ret);
3015
3016         btrfs_free_path(path);
3017         return 0;
3018 }
3019
3020 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3021 {
3022         struct inode *inode = dentry->d_inode;
3023         int err = 0;
3024         struct btrfs_root *root = BTRFS_I(dir)->root;
3025         struct btrfs_trans_handle *trans;
3026         unsigned long nr = 0;
3027
3028         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE ||
3029             btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3030                 return -ENOTEMPTY;
3031
3032         trans = __unlink_start_trans(dir, dentry);
3033         if (IS_ERR(trans))
3034                 return PTR_ERR(trans);
3035
3036         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3037                 err = btrfs_unlink_subvol(trans, root, dir,
3038                                           BTRFS_I(inode)->location.objectid,
3039                                           dentry->d_name.name,
3040                                           dentry->d_name.len);
3041                 goto out;
3042         }
3043
3044         err = btrfs_orphan_add(trans, inode);
3045         if (err)
3046                 goto out;
3047
3048         /* now the directory is empty */
3049         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3050                                  dentry->d_name.name, dentry->d_name.len);
3051         if (!err)
3052                 btrfs_i_size_write(inode, 0);
3053 out:
3054         nr = trans->blocks_used;
3055         __unlink_end_trans(trans, root);
3056         btrfs_btree_balance_dirty(root, nr);
3057
3058         return err;
3059 }
3060
3061 /*
3062  * this can truncate away extent items, csum items and directory items.
3063  * It starts at a high offset and removes keys until it can't find
3064  * any higher than new_size
3065  *
3066  * csum items that cross the new i_size are truncated to the new size
3067  * as well.
3068  *
3069  * min_type is the minimum key type to truncate down to.  If set to 0, this
3070  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3071  */
3072 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3073                                struct btrfs_root *root,
3074                                struct inode *inode,
3075                                u64 new_size, u32 min_type)
3076 {
3077         struct btrfs_path *path;
3078         struct extent_buffer *leaf;
3079         struct btrfs_file_extent_item *fi;
3080         struct btrfs_key key;
3081         struct btrfs_key found_key;
3082         u64 extent_start = 0;
3083         u64 extent_num_bytes = 0;
3084         u64 extent_offset = 0;
3085         u64 item_end = 0;
3086         u64 mask = root->sectorsize - 1;
3087         u32 found_type = (u8)-1;
3088         int found_extent;
3089         int del_item;
3090         int pending_del_nr = 0;
3091         int pending_del_slot = 0;
3092         int extent_type = -1;
3093         int encoding;
3094         int ret;
3095         int err = 0;
3096         u64 ino = btrfs_ino(inode);
3097
3098         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3099
3100         path = btrfs_alloc_path();
3101         if (!path)
3102                 return -ENOMEM;
3103         path->reada = -1;
3104
3105         if (root->ref_cows || root == root->fs_info->tree_root)
3106                 btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
3107
3108         /*
3109          * This function is also used to drop the items in the log tree before
3110          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3111          * it is used to drop the loged items. So we shouldn't kill the delayed
3112          * items.
3113          */
3114         if (min_type == 0 && root == BTRFS_I(inode)->root)
3115                 btrfs_kill_delayed_inode_items(inode);
3116
3117         key.objectid = ino;
3118         key.offset = (u64)-1;
3119         key.type = (u8)-1;
3120
3121 search_again:
3122         path->leave_spinning = 1;
3123         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3124         if (ret < 0) {
3125                 err = ret;
3126                 goto out;
3127         }
3128
3129         if (ret > 0) {
3130                 /* there are no items in the tree for us to truncate, we're
3131                  * done
3132                  */
3133                 if (path->slots[0] == 0)
3134                         goto out;
3135                 path->slots[0]--;
3136         }
3137
3138         while (1) {
3139                 fi = NULL;
3140                 leaf = path->nodes[0];
3141                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3142                 found_type = btrfs_key_type(&found_key);
3143                 encoding = 0;
3144
3145                 if (found_key.objectid != ino)
3146                         break;
3147
3148                 if (found_type < min_type)
3149                         break;
3150
3151                 item_end = found_key.offset;
3152                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3153                         fi = btrfs_item_ptr(leaf, path->slots[0],
3154                                             struct btrfs_file_extent_item);
3155                         extent_type = btrfs_file_extent_type(leaf, fi);
3156                         encoding = btrfs_file_extent_compression(leaf, fi);
3157                         encoding |= btrfs_file_extent_encryption(leaf, fi);
3158                         encoding |= btrfs_file_extent_other_encoding(leaf, fi);
3159
3160                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3161                                 item_end +=
3162                                     btrfs_file_extent_num_bytes(leaf, fi);
3163                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3164                                 item_end += btrfs_file_extent_inline_len(leaf,
3165                                                                          fi);
3166                         }
3167                         item_end--;
3168                 }
3169                 if (found_type > min_type) {
3170                         del_item = 1;
3171                 } else {
3172                         if (item_end < new_size)
3173                                 break;
3174                         if (found_key.offset >= new_size)
3175                                 del_item = 1;
3176                         else
3177                                 del_item = 0;
3178                 }
3179                 found_extent = 0;
3180                 /* FIXME, shrink the extent if the ref count is only 1 */
3181                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3182                         goto delete;
3183
3184                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3185                         u64 num_dec;
3186                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3187                         if (!del_item && !encoding) {
3188                                 u64 orig_num_bytes =
3189                                         btrfs_file_extent_num_bytes(leaf, fi);
3190                                 extent_num_bytes = new_size -
3191                                         found_key.offset + root->sectorsize - 1;
3192                                 extent_num_bytes = extent_num_bytes &
3193                                         ~((u64)root->sectorsize - 1);
3194                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3195                                                          extent_num_bytes);
3196                                 num_dec = (orig_num_bytes -
3197                                            extent_num_bytes);
3198                                 if (root->ref_cows && extent_start != 0)
3199                                         inode_sub_bytes(inode, num_dec);
3200                                 btrfs_mark_buffer_dirty(leaf);
3201                         } else {
3202                                 extent_num_bytes =
3203                                         btrfs_file_extent_disk_num_bytes(leaf,
3204                                                                          fi);
3205                                 extent_offset = found_key.offset -
3206                                         btrfs_file_extent_offset(leaf, fi);
3207
3208                                 /* FIXME blocksize != 4096 */
3209                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3210                                 if (extent_start != 0) {
3211                                         found_extent = 1;
3212                                         if (root->ref_cows)
3213                                                 inode_sub_bytes(inode, num_dec);
3214                                 }
3215                         }
3216                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3217                         /*
3218                          * we can't truncate inline items that have had
3219                          * special encodings
3220                          */
3221                         if (!del_item &&
3222                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3223                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3224                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3225                                 u32 size = new_size - found_key.offset;
3226
3227                                 if (root->ref_cows) {
3228                                         inode_sub_bytes(inode, item_end + 1 -
3229                                                         new_size);
3230                                 }
3231                                 size =
3232                                     btrfs_file_extent_calc_inline_size(size);
3233                                 ret = btrfs_truncate_item(trans, root, path,
3234                                                           size, 1);
3235                         } else if (root->ref_cows) {
3236                                 inode_sub_bytes(inode, item_end + 1 -
3237                                                 found_key.offset);
3238                         }
3239                 }
3240 delete:
3241                 if (del_item) {
3242                         if (!pending_del_nr) {
3243                                 /* no pending yet, add ourselves */
3244                                 pending_del_slot = path->slots[0];
3245                                 pending_del_nr = 1;
3246                         } else if (pending_del_nr &&
3247                                    path->slots[0] + 1 == pending_del_slot) {
3248                                 /* hop on the pending chunk */
3249                                 pending_del_nr++;
3250                                 pending_del_slot = path->slots[0];
3251                         } else {
3252                                 BUG();
3253                         }
3254                 } else {
3255                         break;
3256                 }
3257                 if (found_extent && (root->ref_cows ||
3258                                      root == root->fs_info->tree_root)) {
3259                         btrfs_set_path_blocking(path);
3260                         ret = btrfs_free_extent(trans, root, extent_start,
3261                                                 extent_num_bytes, 0,
3262                                                 btrfs_header_owner(leaf),
3263                                                 ino, extent_offset);
3264                         BUG_ON(ret);
3265                 }
3266
3267                 if (found_type == BTRFS_INODE_ITEM_KEY)
3268                         break;
3269
3270                 if (path->slots[0] == 0 ||
3271                     path->slots[0] != pending_del_slot) {
3272                         if (root->ref_cows &&
3273                             BTRFS_I(inode)->location.objectid !=
3274                                                 BTRFS_FREE_INO_OBJECTID) {
3275                                 err = -EAGAIN;
3276                                 goto out;
3277                         }
3278                         if (pending_del_nr) {
3279                                 ret = btrfs_del_items(trans, root, path,
3280                                                 pending_del_slot,
3281                                                 pending_del_nr);
3282                                 BUG_ON(ret);
3283                                 pending_del_nr = 0;
3284                         }
3285                         btrfs_release_path(path);
3286                         goto search_again;
3287                 } else {
3288                         path->slots[0]--;
3289                 }
3290         }
3291 out:
3292         if (pending_del_nr) {
3293                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3294                                       pending_del_nr);
3295                 BUG_ON(ret);
3296         }
3297         btrfs_free_path(path);
3298         return err;
3299 }
3300
3301 /*
3302  * taken from block_truncate_page, but does cow as it zeros out
3303  * any bytes left in the last page in the file.
3304  */
3305 static int btrfs_truncate_page(struct address_space *mapping, loff_t from)
3306 {
3307         struct inode *inode = mapping->host;
3308         struct btrfs_root *root = BTRFS_I(inode)->root;
3309         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3310         struct btrfs_ordered_extent *ordered;
3311         struct extent_state *cached_state = NULL;
3312         char *kaddr;
3313         u32 blocksize = root->sectorsize;
3314         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3315         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3316         struct page *page;
3317         gfp_t mask = btrfs_alloc_write_mask(mapping);
3318         int ret = 0;
3319         u64 page_start;
3320         u64 page_end;
3321
3322         if ((offset & (blocksize - 1)) == 0)
3323                 goto out;
3324         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3325         if (ret)
3326                 goto out;
3327
3328         ret = -ENOMEM;
3329 again:
3330         page = find_or_create_page(mapping, index, mask);
3331         if (!page) {
3332                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3333                 goto out;
3334         }
3335
3336         page_start = page_offset(page);
3337         page_end = page_start + PAGE_CACHE_SIZE - 1;
3338
3339         if (!PageUptodate(page)) {
3340                 ret = btrfs_readpage(NULL, page);
3341                 lock_page(page);
3342                 if (page->mapping != mapping) {
3343                         unlock_page(page);
3344                         page_cache_release(page);
3345                         goto again;
3346                 }
3347                 if (!PageUptodate(page)) {
3348                         ret = -EIO;
3349                         goto out_unlock;
3350                 }
3351         }
3352         wait_on_page_writeback(page);
3353
3354         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
3355                          GFP_NOFS);
3356         set_page_extent_mapped(page);
3357
3358         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3359         if (ordered) {
3360                 unlock_extent_cached(io_tree, page_start, page_end,
3361                                      &cached_state, GFP_NOFS);
3362                 unlock_page(page);
3363                 page_cache_release(page);
3364                 btrfs_start_ordered_extent(inode, ordered, 1);
3365                 btrfs_put_ordered_extent(ordered);
3366                 goto again;
3367         }
3368
3369         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3370                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
3371                           0, 0, &cached_state, GFP_NOFS);
3372
3373         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3374                                         &cached_state);
3375         if (ret) {
3376                 unlock_extent_cached(io_tree, page_start, page_end,
3377                                      &cached_state, GFP_NOFS);
3378                 goto out_unlock;
3379         }
3380
3381         ret = 0;
3382         if (offset != PAGE_CACHE_SIZE) {
3383                 kaddr = kmap(page);
3384                 memset(kaddr + offset, 0, PAGE_CACHE_SIZE - offset);
3385                 flush_dcache_page(page);
3386                 kunmap(page);
3387         }
3388         ClearPageChecked(page);
3389         set_page_dirty(page);
3390         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3391                              GFP_NOFS);
3392
3393 out_unlock:
3394         if (ret)
3395                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3396         unlock_page(page);
3397         page_cache_release(page);
3398 out:
3399         return ret;
3400 }
3401
3402 /*
3403  * This function puts in dummy file extents for the area we're creating a hole
3404  * for.  So if we are truncating this file to a larger size we need to insert
3405  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3406  * the range between oldsize and size
3407  */
3408 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3409 {
3410         struct btrfs_trans_handle *trans;
3411         struct btrfs_root *root = BTRFS_I(inode)->root;
3412         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3413         struct extent_map *em = NULL;
3414         struct extent_state *cached_state = NULL;
3415         u64 mask = root->sectorsize - 1;
3416         u64 hole_start = (oldsize + mask) & ~mask;
3417         u64 block_end = (size + mask) & ~mask;
3418         u64 last_byte;
3419         u64 cur_offset;
3420         u64 hole_size;
3421         int err = 0;
3422
3423         if (size <= hole_start)
3424                 return 0;
3425
3426         while (1) {
3427                 struct btrfs_ordered_extent *ordered;
3428                 btrfs_wait_ordered_range(inode, hole_start,
3429                                          block_end - hole_start);
3430                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3431                                  &cached_state, GFP_NOFS);
3432                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3433                 if (!ordered)
3434                         break;
3435                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3436                                      &cached_state, GFP_NOFS);
3437                 btrfs_put_ordered_extent(ordered);
3438         }
3439
3440         cur_offset = hole_start;
3441         while (1) {
3442                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3443                                 block_end - cur_offset, 0);
3444                 BUG_ON(IS_ERR_OR_NULL(em));
3445                 last_byte = min(extent_map_end(em), block_end);
3446                 last_byte = (last_byte + mask) & ~mask;
3447                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3448                         u64 hint_byte = 0;
3449                         hole_size = last_byte - cur_offset;
3450
3451                         trans = btrfs_start_transaction(root, 2);
3452                         if (IS_ERR(trans)) {
3453                                 err = PTR_ERR(trans);
3454                                 break;
3455                         }
3456
3457                         err = btrfs_drop_extents(trans, inode, cur_offset,
3458                                                  cur_offset + hole_size,
3459                                                  &hint_byte, 1);
3460                         if (err) {
3461                                 btrfs_end_transaction(trans, root);
3462                                 break;
3463                         }
3464
3465                         err = btrfs_insert_file_extent(trans, root,
3466                                         btrfs_ino(inode), cur_offset, 0,
3467                                         0, hole_size, 0, hole_size,
3468                                         0, 0, 0);
3469                         if (err) {
3470                                 btrfs_end_transaction(trans, root);
3471                                 break;
3472                         }
3473
3474                         btrfs_drop_extent_cache(inode, hole_start,
3475                                         last_byte - 1, 0);
3476
3477                         btrfs_end_transaction(trans, root);
3478                 }
3479                 free_extent_map(em);
3480                 em = NULL;
3481                 cur_offset = last_byte;
3482                 if (cur_offset >= block_end)
3483                         break;
3484         }
3485
3486         free_extent_map(em);
3487         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3488                              GFP_NOFS);
3489         return err;
3490 }
3491
3492 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3493 {
3494         loff_t oldsize = i_size_read(inode);
3495         int ret;
3496
3497         if (newsize == oldsize)
3498                 return 0;
3499
3500         if (newsize > oldsize) {
3501                 i_size_write(inode, newsize);
3502                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3503                 truncate_pagecache(inode, oldsize, newsize);
3504                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3505                 if (ret) {
3506                         btrfs_setsize(inode, oldsize);
3507                         return ret;
3508                 }
3509
3510                 mark_inode_dirty(inode);
3511         } else {
3512
3513                 /*
3514                  * We're truncating a file that used to have good data down to
3515                  * zero. Make sure it gets into the ordered flush list so that
3516                  * any new writes get down to disk quickly.
3517                  */
3518                 if (newsize == 0)
3519                         BTRFS_I(inode)->ordered_data_close = 1;
3520
3521                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3522                 truncate_setsize(inode, newsize);
3523                 ret = btrfs_truncate(inode);
3524         }
3525
3526         return ret;
3527 }
3528
3529 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3530 {
3531         struct inode *inode = dentry->d_inode;
3532         struct btrfs_root *root = BTRFS_I(inode)->root;
3533         int err;
3534
3535         if (btrfs_root_readonly(root))
3536                 return -EROFS;
3537
3538         err = inode_change_ok(inode, attr);
3539         if (err)
3540                 return err;
3541
3542         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3543                 err = btrfs_setsize(inode, attr->ia_size);
3544                 if (err)
3545                         return err;
3546         }
3547
3548         if (attr->ia_valid) {
3549                 setattr_copy(inode, attr);
3550                 mark_inode_dirty(inode);
3551
3552                 if (attr->ia_valid & ATTR_MODE)
3553                         err = btrfs_acl_chmod(inode);
3554         }
3555
3556         return err;
3557 }
3558
3559 void btrfs_evict_inode(struct inode *inode)
3560 {
3561         struct btrfs_trans_handle *trans;
3562         struct btrfs_root *root = BTRFS_I(inode)->root;
3563         struct btrfs_block_rsv *rsv, *global_rsv;
3564         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3565         unsigned long nr;
3566         int ret;
3567
3568         trace_btrfs_inode_evict(inode);
3569
3570         truncate_inode_pages(&inode->i_data, 0);
3571         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3572                                btrfs_is_free_space_inode(root, inode)))
3573                 goto no_delete;
3574
3575         if (is_bad_inode(inode)) {
3576                 btrfs_orphan_del(NULL, inode);
3577                 goto no_delete;
3578         }
3579         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3580         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3581
3582         if (root->fs_info->log_root_recovering) {
3583                 BUG_ON(!list_empty(&BTRFS_I(inode)->i_orphan));
3584                 goto no_delete;
3585         }
3586
3587         if (inode->i_nlink > 0) {
3588                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3589                 goto no_delete;
3590         }
3591
3592         rsv = btrfs_alloc_block_rsv(root);
3593         if (!rsv) {
3594                 btrfs_orphan_del(NULL, inode);
3595                 goto no_delete;
3596         }
3597         rsv->size = min_size;
3598         global_rsv = &root->fs_info->global_block_rsv;
3599
3600         btrfs_i_size_write(inode, 0);
3601
3602         /*
3603          * This is a bit simpler than btrfs_truncate since
3604          *
3605          * 1) We've already reserved our space for our orphan item in the
3606          *    unlink.
3607          * 2) We're going to delete the inode item, so we don't need to update
3608          *    it at all.
3609          *
3610          * So we just need to reserve some slack space in case we add bytes when
3611          * doing the truncate.
3612          */
3613         while (1) {
3614                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
3615
3616                 /*
3617                  * Try and steal from the global reserve since we will
3618                  * likely not use this space anyway, we want to try as
3619                  * hard as possible to get this to work.
3620                  */
3621                 if (ret)
3622                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3623
3624                 if (ret) {
3625                         printk(KERN_WARNING "Could not get space for a "
3626                                "delete, will truncate on mount %d\n", ret);
3627                         btrfs_orphan_del(NULL, inode);
3628                         btrfs_free_block_rsv(root, rsv);
3629                         goto no_delete;
3630                 }
3631
3632                 trans = btrfs_start_transaction(root, 0);
3633                 if (IS_ERR(trans)) {
3634                         btrfs_orphan_del(NULL, inode);
3635                         btrfs_free_block_rsv(root, rsv);
3636                         goto no_delete;
3637                 }
3638
3639                 trans->block_rsv = rsv;
3640
3641                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3642                 if (ret != -EAGAIN)
3643                         break;
3644
3645                 nr = trans->blocks_used;
3646                 btrfs_end_transaction(trans, root);
3647                 trans = NULL;
3648                 btrfs_btree_balance_dirty(root, nr);
3649         }
3650
3651         btrfs_free_block_rsv(root, rsv);
3652
3653         if (ret == 0) {
3654                 trans->block_rsv = root->orphan_block_rsv;
3655                 ret = btrfs_orphan_del(trans, inode);
3656                 BUG_ON(ret);
3657         }
3658
3659         trans->block_rsv = &root->fs_info->trans_block_rsv;
3660         if (!(root == root->fs_info->tree_root ||
3661               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3662                 btrfs_return_ino(root, btrfs_ino(inode));
3663
3664         nr = trans->blocks_used;
3665         btrfs_end_transaction(trans, root);
3666         btrfs_btree_balance_dirty(root, nr);
3667 no_delete:
3668         end_writeback(inode);
3669         return;
3670 }
3671
3672 /*
3673  * this returns the key found in the dir entry in the location pointer.
3674  * If no dir entries were found, location->objectid is 0.
3675  */
3676 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3677                                struct btrfs_key *location)
3678 {
3679         const char *name = dentry->d_name.name;
3680         int namelen = dentry->d_name.len;
3681         struct btrfs_dir_item *di;
3682         struct btrfs_path *path;
3683         struct btrfs_root *root = BTRFS_I(dir)->root;
3684         int ret = 0;
3685
3686         path = btrfs_alloc_path();
3687         if (!path)
3688                 return -ENOMEM;
3689
3690         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3691                                     namelen, 0);
3692         if (IS_ERR(di))
3693                 ret = PTR_ERR(di);
3694
3695         if (IS_ERR_OR_NULL(di))
3696                 goto out_err;
3697
3698         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3699 out:
3700         btrfs_free_path(path);
3701         return ret;
3702 out_err:
3703         location->objectid = 0;
3704         goto out;
3705 }
3706
3707 /*
3708  * when we hit a tree root in a directory, the btrfs part of the inode
3709  * needs to be changed to reflect the root directory of the tree root.  This
3710  * is kind of like crossing a mount point.
3711  */
3712 static int fixup_tree_root_location(struct btrfs_root *root,
3713                                     struct inode *dir,
3714                                     struct dentry *dentry,
3715                                     struct btrfs_key *location,
3716                                     struct btrfs_root **sub_root)
3717 {
3718         struct btrfs_path *path;
3719         struct btrfs_root *new_root;
3720         struct btrfs_root_ref *ref;
3721         struct extent_buffer *leaf;
3722         int ret;
3723         int err = 0;
3724
3725         path = btrfs_alloc_path();
3726         if (!path) {
3727                 err = -ENOMEM;
3728                 goto out;
3729         }
3730
3731         err = -ENOENT;
3732         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3733                                   BTRFS_I(dir)->root->root_key.objectid,
3734                                   location->objectid);
3735         if (ret) {
3736                 if (ret < 0)
3737                         err = ret;
3738                 goto out;
3739         }
3740
3741         leaf = path->nodes[0];
3742         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3743         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3744             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3745                 goto out;
3746
3747         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3748                                    (unsigned long)(ref + 1),
3749                                    dentry->d_name.len);
3750         if (ret)
3751                 goto out;
3752
3753         btrfs_release_path(path);
3754
3755         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
3756         if (IS_ERR(new_root)) {
3757                 err = PTR_ERR(new_root);
3758                 goto out;
3759         }
3760
3761         if (btrfs_root_refs(&new_root->root_item) == 0) {
3762                 err = -ENOENT;
3763                 goto out;
3764         }
3765
3766         *sub_root = new_root;
3767         location->objectid = btrfs_root_dirid(&new_root->root_item);
3768         location->type = BTRFS_INODE_ITEM_KEY;
3769         location->offset = 0;
3770         err = 0;
3771 out:
3772         btrfs_free_path(path);
3773         return err;
3774 }
3775
3776 static void inode_tree_add(struct inode *inode)
3777 {
3778         struct btrfs_root *root = BTRFS_I(inode)->root;
3779         struct btrfs_inode *entry;
3780         struct rb_node **p;
3781         struct rb_node *parent;
3782         u64 ino = btrfs_ino(inode);
3783 again:
3784         p = &root->inode_tree.rb_node;
3785         parent = NULL;
3786
3787         if (inode_unhashed(inode))
3788                 return;
3789
3790         spin_lock(&root->inode_lock);
3791         while (*p) {
3792                 parent = *p;
3793                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
3794
3795                 if (ino < btrfs_ino(&entry->vfs_inode))
3796                         p = &parent->rb_left;
3797                 else if (ino > btrfs_ino(&entry->vfs_inode))
3798                         p = &parent->rb_right;
3799                 else {
3800                         WARN_ON(!(entry->vfs_inode.i_state &
3801                                   (I_WILL_FREE | I_FREEING)));
3802                         rb_erase(parent, &root->inode_tree);
3803                         RB_CLEAR_NODE(parent);
3804                         spin_unlock(&root->inode_lock);
3805                         goto again;
3806                 }
3807         }
3808         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
3809         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3810         spin_unlock(&root->inode_lock);
3811 }
3812
3813 static void inode_tree_del(struct inode *inode)
3814 {
3815         struct btrfs_root *root = BTRFS_I(inode)->root;
3816         int empty = 0;
3817
3818         spin_lock(&root->inode_lock);
3819         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
3820                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
3821                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
3822                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3823         }
3824         spin_unlock(&root->inode_lock);
3825
3826         /*
3827          * Free space cache has inodes in the tree root, but the tree root has a
3828          * root_refs of 0, so this could end up dropping the tree root as a
3829          * snapshot, so we need the extra !root->fs_info->tree_root check to
3830          * make sure we don't drop it.
3831          */
3832         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
3833             root != root->fs_info->tree_root) {
3834                 synchronize_srcu(&root->fs_info->subvol_srcu);
3835                 spin_lock(&root->inode_lock);
3836                 empty = RB_EMPTY_ROOT(&root->inode_tree);
3837                 spin_unlock(&root->inode_lock);
3838                 if (empty)
3839                         btrfs_add_dead_root(root);
3840         }
3841 }
3842
3843 int btrfs_invalidate_inodes(struct btrfs_root *root)
3844 {
3845         struct rb_node *node;
3846         struct rb_node *prev;
3847         struct btrfs_inode *entry;
3848         struct inode *inode;
3849         u64 objectid = 0;
3850
3851         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
3852
3853         spin_lock(&root->inode_lock);
3854 again:
3855         node = root->inode_tree.rb_node;
3856         prev = NULL;
3857         while (node) {
3858                 prev = node;
3859                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3860
3861                 if (objectid < btrfs_ino(&entry->vfs_inode))
3862                         node = node->rb_left;
3863                 else if (objectid > btrfs_ino(&entry->vfs_inode))
3864                         node = node->rb_right;
3865                 else
3866                         break;
3867         }
3868         if (!node) {
3869                 while (prev) {
3870                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
3871                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
3872                                 node = prev;
3873                                 break;
3874                         }
3875                         prev = rb_next(prev);
3876                 }
3877         }
3878         while (node) {
3879                 entry = rb_entry(node, struct btrfs_inode, rb_node);
3880                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
3881                 inode = igrab(&entry->vfs_inode);
3882                 if (inode) {
3883                         spin_unlock(&root->inode_lock);
3884                         if (atomic_read(&inode->i_count) > 1)
3885                                 d_prune_aliases(inode);
3886                         /*
3887                          * btrfs_drop_inode will have it removed from
3888                          * the inode cache when its usage count
3889                          * hits zero.
3890                          */
3891                         iput(inode);
3892                         cond_resched();
3893                         spin_lock(&root->inode_lock);
3894                         goto again;
3895                 }
3896
3897                 if (cond_resched_lock(&root->inode_lock))
3898                         goto again;
3899
3900                 node = rb_next(node);
3901         }
3902         spin_unlock(&root->inode_lock);
3903         return 0;
3904 }
3905
3906 static int btrfs_init_locked_inode(struct inode *inode, void *p)
3907 {
3908         struct btrfs_iget_args *args = p;
3909         inode->i_ino = args->ino;
3910         BTRFS_I(inode)->root = args->root;
3911         btrfs_set_inode_space_info(args->root, inode);
3912         return 0;
3913 }
3914
3915 static int btrfs_find_actor(struct inode *inode, void *opaque)
3916 {
3917         struct btrfs_iget_args *args = opaque;
3918         return args->ino == btrfs_ino(inode) &&
3919                 args->root == BTRFS_I(inode)->root;
3920 }
3921
3922 static struct inode *btrfs_iget_locked(struct super_block *s,
3923                                        u64 objectid,
3924                                        struct btrfs_root *root)
3925 {
3926         struct inode *inode;
3927         struct btrfs_iget_args args;
3928         args.ino = objectid;
3929         args.root = root;
3930
3931         inode = iget5_locked(s, objectid, btrfs_find_actor,
3932                              btrfs_init_locked_inode,
3933                              (void *)&args);
3934         return inode;
3935 }
3936
3937 /* Get an inode object given its location and corresponding root.
3938  * Returns in *is_new if the inode was read from disk
3939  */
3940 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
3941                          struct btrfs_root *root, int *new)
3942 {
3943         struct inode *inode;
3944
3945         inode = btrfs_iget_locked(s, location->objectid, root);
3946         if (!inode)
3947                 return ERR_PTR(-ENOMEM);
3948
3949         if (inode->i_state & I_NEW) {
3950                 BTRFS_I(inode)->root = root;
3951                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
3952                 btrfs_read_locked_inode(inode);
3953                 if (!is_bad_inode(inode)) {
3954                         inode_tree_add(inode);
3955                         unlock_new_inode(inode);
3956                         if (new)
3957                                 *new = 1;
3958                 } else {
3959                         unlock_new_inode(inode);
3960                         iput(inode);
3961                         inode = ERR_PTR(-ESTALE);
3962                 }
3963         }
3964
3965         return inode;
3966 }
3967
3968 static struct inode *new_simple_dir(struct super_block *s,
3969                                     struct btrfs_key *key,
3970                                     struct btrfs_root *root)
3971 {
3972         struct inode *inode = new_inode(s);
3973
3974         if (!inode)
3975                 return ERR_PTR(-ENOMEM);
3976
3977         BTRFS_I(inode)->root = root;
3978         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
3979         BTRFS_I(inode)->dummy_inode = 1;
3980
3981         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
3982         inode->i_op = &simple_dir_inode_operations;
3983         inode->i_fop = &simple_dir_operations;
3984         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
3985         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
3986
3987         return inode;
3988 }
3989
3990 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
3991 {
3992         struct inode *inode;
3993         struct btrfs_root *root = BTRFS_I(dir)->root;
3994         struct btrfs_root *sub_root = root;
3995         struct btrfs_key location;
3996         int index;
3997         int ret = 0;
3998
3999         if (dentry->d_name.len > BTRFS_NAME_LEN)
4000                 return ERR_PTR(-ENAMETOOLONG);
4001
4002         if (unlikely(d_need_lookup(dentry))) {
4003                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4004                 kfree(dentry->d_fsdata);
4005                 dentry->d_fsdata = NULL;
4006                 /* This thing is hashed, drop it for now */
4007                 d_drop(dentry);
4008         } else {
4009                 ret = btrfs_inode_by_name(dir, dentry, &location);
4010         }
4011
4012         if (ret < 0)
4013                 return ERR_PTR(ret);
4014
4015         if (location.objectid == 0)
4016                 return NULL;
4017
4018         if (location.type == BTRFS_INODE_ITEM_KEY) {
4019                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4020                 return inode;
4021         }
4022
4023         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4024
4025         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4026         ret = fixup_tree_root_location(root, dir, dentry,
4027                                        &location, &sub_root);
4028         if (ret < 0) {
4029                 if (ret != -ENOENT)
4030                         inode = ERR_PTR(ret);
4031                 else
4032                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4033         } else {
4034                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4035         }
4036         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4037
4038         if (!IS_ERR(inode) && root != sub_root) {
4039                 down_read(&root->fs_info->cleanup_work_sem);
4040                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4041                         ret = btrfs_orphan_cleanup(sub_root);
4042                 up_read(&root->fs_info->cleanup_work_sem);
4043                 if (ret)
4044                         inode = ERR_PTR(ret);
4045         }
4046
4047         return inode;
4048 }
4049
4050 static int btrfs_dentry_delete(const struct dentry *dentry)
4051 {
4052         struct btrfs_root *root;
4053
4054         if (!dentry->d_inode && !IS_ROOT(dentry))
4055                 dentry = dentry->d_parent;
4056
4057         if (dentry->d_inode) {
4058                 root = BTRFS_I(dentry->d_inode)->root;
4059                 if (btrfs_root_refs(&root->root_item) == 0)
4060                         return 1;
4061         }
4062         return 0;
4063 }
4064
4065 static void btrfs_dentry_release(struct dentry *dentry)
4066 {
4067         if (dentry->d_fsdata)
4068                 kfree(dentry->d_fsdata);
4069 }
4070
4071 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4072                                    struct nameidata *nd)
4073 {
4074         struct dentry *ret;
4075
4076         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4077         if (unlikely(d_need_lookup(dentry))) {
4078                 spin_lock(&dentry->d_lock);
4079                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4080                 spin_unlock(&dentry->d_lock);
4081         }
4082         return ret;
4083 }
4084
4085 unsigned char btrfs_filetype_table[] = {
4086         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4087 };
4088
4089 static int btrfs_real_readdir(struct file *filp, void *dirent,
4090                               filldir_t filldir)
4091 {
4092         struct inode *inode = filp->f_dentry->d_inode;
4093         struct btrfs_root *root = BTRFS_I(inode)->root;
4094         struct btrfs_item *item;
4095         struct btrfs_dir_item *di;
4096         struct btrfs_key key;
4097         struct btrfs_key found_key;
4098         struct btrfs_path *path;
4099         struct list_head ins_list;
4100         struct list_head del_list;
4101         struct qstr q;
4102         int ret;
4103         struct extent_buffer *leaf;
4104         int slot;
4105         unsigned char d_type;
4106         int over = 0;
4107         u32 di_cur;
4108         u32 di_total;
4109         u32 di_len;
4110         int key_type = BTRFS_DIR_INDEX_KEY;
4111         char tmp_name[32];
4112         char *name_ptr;
4113         int name_len;
4114         int is_curr = 0;        /* filp->f_pos points to the current index? */
4115
4116         /* FIXME, use a real flag for deciding about the key type */
4117         if (root->fs_info->tree_root == root)
4118                 key_type = BTRFS_DIR_ITEM_KEY;
4119
4120         /* special case for "." */
4121         if (filp->f_pos == 0) {
4122                 over = filldir(dirent, ".", 1,
4123                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4124                 if (over)
4125                         return 0;
4126                 filp->f_pos = 1;
4127         }
4128         /* special case for .., just use the back ref */
4129         if (filp->f_pos == 1) {
4130                 u64 pino = parent_ino(filp->f_path.dentry);
4131                 over = filldir(dirent, "..", 2,
4132                                filp->f_pos, pino, DT_DIR);
4133                 if (over)
4134                         return 0;
4135                 filp->f_pos = 2;
4136         }
4137         path = btrfs_alloc_path();
4138         if (!path)
4139                 return -ENOMEM;
4140
4141         path->reada = 1;
4142
4143         if (key_type == BTRFS_DIR_INDEX_KEY) {
4144                 INIT_LIST_HEAD(&ins_list);
4145                 INIT_LIST_HEAD(&del_list);
4146                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4147         }
4148
4149         btrfs_set_key_type(&key, key_type);
4150         key.offset = filp->f_pos;
4151         key.objectid = btrfs_ino(inode);
4152
4153         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4154         if (ret < 0)
4155                 goto err;
4156
4157         while (1) {
4158                 leaf = path->nodes[0];
4159                 slot = path->slots[0];
4160                 if (slot >= btrfs_header_nritems(leaf)) {
4161                         ret = btrfs_next_leaf(root, path);
4162                         if (ret < 0)
4163                                 goto err;
4164                         else if (ret > 0)
4165                                 break;
4166                         continue;
4167                 }
4168
4169                 item = btrfs_item_nr(leaf, slot);
4170                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4171
4172                 if (found_key.objectid != key.objectid)
4173                         break;
4174                 if (btrfs_key_type(&found_key) != key_type)
4175                         break;
4176                 if (found_key.offset < filp->f_pos)
4177                         goto next;
4178                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4179                     btrfs_should_delete_dir_index(&del_list,
4180                                                   found_key.offset))
4181                         goto next;
4182
4183                 filp->f_pos = found_key.offset;
4184                 is_curr = 1;
4185
4186                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4187                 di_cur = 0;
4188                 di_total = btrfs_item_size(leaf, item);
4189
4190                 while (di_cur < di_total) {
4191                         struct btrfs_key location;
4192                         struct dentry *tmp;
4193
4194                         if (verify_dir_item(root, leaf, di))
4195                                 break;
4196
4197                         name_len = btrfs_dir_name_len(leaf, di);
4198                         if (name_len <= sizeof(tmp_name)) {
4199                                 name_ptr = tmp_name;
4200                         } else {
4201                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4202                                 if (!name_ptr) {
4203                                         ret = -ENOMEM;
4204                                         goto err;
4205                                 }
4206                         }
4207                         read_extent_buffer(leaf, name_ptr,
4208                                            (unsigned long)(di + 1), name_len);
4209
4210                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4211                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4212
4213                         q.name = name_ptr;
4214                         q.len = name_len;
4215                         q.hash = full_name_hash(q.name, q.len);
4216                         tmp = d_lookup(filp->f_dentry, &q);
4217                         if (!tmp) {
4218                                 struct btrfs_key *newkey;
4219
4220                                 newkey = kzalloc(sizeof(struct btrfs_key),
4221                                                  GFP_NOFS);
4222                                 if (!newkey)
4223                                         goto no_dentry;
4224                                 tmp = d_alloc(filp->f_dentry, &q);
4225                                 if (!tmp) {
4226                                         kfree(newkey);
4227                                         dput(tmp);
4228                                         goto no_dentry;
4229                                 }
4230                                 memcpy(newkey, &location,
4231                                        sizeof(struct btrfs_key));
4232                                 tmp->d_fsdata = newkey;
4233                                 tmp->d_flags |= DCACHE_NEED_LOOKUP;
4234                                 d_rehash(tmp);
4235                                 dput(tmp);
4236                         } else {
4237                                 dput(tmp);
4238                         }
4239 no_dentry:
4240                         /* is this a reference to our own snapshot? If so
4241                          * skip it
4242                          */
4243                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4244                             location.objectid == root->root_key.objectid) {
4245                                 over = 0;
4246                                 goto skip;
4247                         }
4248                         over = filldir(dirent, name_ptr, name_len,
4249                                        found_key.offset, location.objectid,
4250                                        d_type);
4251
4252 skip:
4253                         if (name_ptr != tmp_name)
4254                                 kfree(name_ptr);
4255
4256                         if (over)
4257                                 goto nopos;
4258                         di_len = btrfs_dir_name_len(leaf, di) +
4259                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4260                         di_cur += di_len;
4261                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4262                 }
4263 next:
4264                 path->slots[0]++;
4265         }
4266
4267         if (key_type == BTRFS_DIR_INDEX_KEY) {
4268                 if (is_curr)
4269                         filp->f_pos++;
4270                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4271                                                       &ins_list);
4272                 if (ret)
4273                         goto nopos;
4274         }
4275
4276         /* Reached end of directory/root. Bump pos past the last item. */
4277         if (key_type == BTRFS_DIR_INDEX_KEY)
4278                 /*
4279                  * 32-bit glibc will use getdents64, but then strtol -
4280                  * so the last number we can serve is this.
4281                  */
4282                 filp->f_pos = 0x7fffffff;
4283         else
4284                 filp->f_pos++;
4285 nopos:
4286         ret = 0;
4287 err:
4288         if (key_type == BTRFS_DIR_INDEX_KEY)
4289                 btrfs_put_delayed_items(&ins_list, &del_list);
4290         btrfs_free_path(path);
4291         return ret;
4292 }
4293
4294 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4295 {
4296         struct btrfs_root *root = BTRFS_I(inode)->root;
4297         struct btrfs_trans_handle *trans;
4298         int ret = 0;
4299         bool nolock = false;
4300
4301         if (BTRFS_I(inode)->dummy_inode)
4302                 return 0;
4303
4304         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(root, inode))
4305                 nolock = true;
4306
4307         if (wbc->sync_mode == WB_SYNC_ALL) {
4308                 if (nolock)
4309                         trans = btrfs_join_transaction_nolock(root);
4310                 else
4311                         trans = btrfs_join_transaction(root);
4312                 if (IS_ERR(trans))
4313                         return PTR_ERR(trans);
4314                 if (nolock)
4315                         ret = btrfs_end_transaction_nolock(trans, root);
4316                 else
4317                         ret = btrfs_commit_transaction(trans, root);
4318         }
4319         return ret;
4320 }
4321
4322 /*
4323  * This is somewhat expensive, updating the tree every time the
4324  * inode changes.  But, it is most likely to find the inode in cache.
4325  * FIXME, needs more benchmarking...there are no reasons other than performance
4326  * to keep or drop this code.
4327  */
4328 void btrfs_dirty_inode(struct inode *inode, int flags)
4329 {
4330         struct btrfs_root *root = BTRFS_I(inode)->root;
4331         struct btrfs_trans_handle *trans;
4332         int ret;
4333
4334         if (BTRFS_I(inode)->dummy_inode)
4335                 return;
4336
4337         trans = btrfs_join_transaction(root);
4338         BUG_ON(IS_ERR(trans));
4339
4340         ret = btrfs_update_inode(trans, root, inode);
4341         if (ret && ret == -ENOSPC) {
4342                 /* whoops, lets try again with the full transaction */
4343                 btrfs_end_transaction(trans, root);
4344                 trans = btrfs_start_transaction(root, 1);
4345                 if (IS_ERR(trans)) {
4346                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4347                                        "dirty  inode %llu error %ld\n",
4348                                        (unsigned long long)btrfs_ino(inode),
4349                                        PTR_ERR(trans));
4350                         return;
4351                 }
4352
4353                 ret = btrfs_update_inode(trans, root, inode);
4354                 if (ret) {
4355                         printk_ratelimited(KERN_ERR "btrfs: fail to "
4356                                        "dirty  inode %llu error %d\n",
4357                                        (unsigned long long)btrfs_ino(inode),
4358                                        ret);
4359                 }
4360         }
4361         btrfs_end_transaction(trans, root);
4362         if (BTRFS_I(inode)->delayed_node)
4363                 btrfs_balance_delayed_items(root);
4364 }
4365
4366 /*
4367  * find the highest existing sequence number in a directory
4368  * and then set the in-memory index_cnt variable to reflect
4369  * free sequence numbers
4370  */
4371 static int btrfs_set_inode_index_count(struct inode *inode)
4372 {
4373         struct btrfs_root *root = BTRFS_I(inode)->root;
4374         struct btrfs_key key, found_key;
4375         struct btrfs_path *path;
4376         struct extent_buffer *leaf;
4377         int ret;
4378
4379         key.objectid = btrfs_ino(inode);
4380         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4381         key.offset = (u64)-1;
4382
4383         path = btrfs_alloc_path();
4384         if (!path)
4385                 return -ENOMEM;
4386
4387         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4388         if (ret < 0)
4389                 goto out;
4390         /* FIXME: we should be able to handle this */
4391         if (ret == 0)
4392                 goto out;
4393         ret = 0;
4394
4395         /*
4396          * MAGIC NUMBER EXPLANATION:
4397          * since we search a directory based on f_pos we have to start at 2
4398          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4399          * else has to start at 2
4400          */
4401         if (path->slots[0] == 0) {
4402                 BTRFS_I(inode)->index_cnt = 2;
4403                 goto out;
4404         }
4405
4406         path->slots[0]--;
4407
4408         leaf = path->nodes[0];
4409         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4410
4411         if (found_key.objectid != btrfs_ino(inode) ||
4412             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4413                 BTRFS_I(inode)->index_cnt = 2;
4414                 goto out;
4415         }
4416
4417         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4418 out:
4419         btrfs_free_path(path);
4420         return ret;
4421 }
4422
4423 /*
4424  * helper to find a free sequence number in a given directory.  This current
4425  * code is very simple, later versions will do smarter things in the btree
4426  */
4427 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4428 {
4429         int ret = 0;
4430
4431         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4432                 ret = btrfs_inode_delayed_dir_index_count(dir);
4433                 if (ret) {
4434                         ret = btrfs_set_inode_index_count(dir);
4435                         if (ret)
4436                                 return ret;
4437                 }
4438         }
4439
4440         *index = BTRFS_I(dir)->index_cnt;
4441         BTRFS_I(dir)->index_cnt++;
4442
4443         return ret;
4444 }
4445
4446 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4447                                      struct btrfs_root *root,
4448                                      struct inode *dir,
4449                                      const char *name, int name_len,
4450                                      u64 ref_objectid, u64 objectid, int mode,
4451                                      u64 *index)
4452 {
4453         struct inode *inode;
4454         struct btrfs_inode_item *inode_item;
4455         struct btrfs_key *location;
4456         struct btrfs_path *path;
4457         struct btrfs_inode_ref *ref;
4458         struct btrfs_key key[2];
4459         u32 sizes[2];
4460         unsigned long ptr;
4461         int ret;
4462         int owner;
4463
4464         path = btrfs_alloc_path();
4465         if (!path)
4466                 return ERR_PTR(-ENOMEM);
4467
4468         inode = new_inode(root->fs_info->sb);
4469         if (!inode) {
4470                 btrfs_free_path(path);
4471                 return ERR_PTR(-ENOMEM);
4472         }
4473
4474         /*
4475          * we have to initialize this early, so we can reclaim the inode
4476          * number if we fail afterwards in this function.
4477          */
4478         inode->i_ino = objectid;
4479
4480         if (dir) {
4481                 trace_btrfs_inode_request(dir);
4482
4483                 ret = btrfs_set_inode_index(dir, index);
4484                 if (ret) {
4485                         btrfs_free_path(path);
4486                         iput(inode);
4487                         return ERR_PTR(ret);
4488                 }
4489         }
4490         /*
4491          * index_cnt is ignored for everything but a dir,
4492          * btrfs_get_inode_index_count has an explanation for the magic
4493          * number
4494          */
4495         BTRFS_I(inode)->index_cnt = 2;
4496         BTRFS_I(inode)->root = root;
4497         BTRFS_I(inode)->generation = trans->transid;
4498         inode->i_generation = BTRFS_I(inode)->generation;
4499         btrfs_set_inode_space_info(root, inode);
4500
4501         if (S_ISDIR(mode))
4502                 owner = 0;
4503         else
4504                 owner = 1;
4505
4506         key[0].objectid = objectid;
4507         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4508         key[0].offset = 0;
4509
4510         key[1].objectid = objectid;
4511         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4512         key[1].offset = ref_objectid;
4513
4514         sizes[0] = sizeof(struct btrfs_inode_item);
4515         sizes[1] = name_len + sizeof(*ref);
4516
4517         path->leave_spinning = 1;
4518         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4519         if (ret != 0)
4520                 goto fail;
4521
4522         inode_init_owner(inode, dir, mode);
4523         inode_set_bytes(inode, 0);
4524         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4525         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4526                                   struct btrfs_inode_item);
4527         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4528
4529         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4530                              struct btrfs_inode_ref);
4531         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4532         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4533         ptr = (unsigned long)(ref + 1);
4534         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4535
4536         btrfs_mark_buffer_dirty(path->nodes[0]);
4537         btrfs_free_path(path);
4538
4539         location = &BTRFS_I(inode)->location;
4540         location->objectid = objectid;
4541         location->offset = 0;
4542         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4543
4544         btrfs_inherit_iflags(inode, dir);
4545
4546         if (S_ISREG(mode)) {
4547                 if (btrfs_test_opt(root, NODATASUM))
4548                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4549                 if (btrfs_test_opt(root, NODATACOW) ||
4550                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4551                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4552         }
4553
4554         insert_inode_hash(inode);
4555         inode_tree_add(inode);
4556
4557         trace_btrfs_inode_new(inode);
4558         btrfs_set_inode_last_trans(trans, inode);
4559
4560         return inode;
4561 fail:
4562         if (dir)
4563                 BTRFS_I(dir)->index_cnt--;
4564         btrfs_free_path(path);
4565         iput(inode);
4566         return ERR_PTR(ret);
4567 }
4568
4569 static inline u8 btrfs_inode_type(struct inode *inode)
4570 {
4571         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4572 }
4573
4574 /*
4575  * utility function to add 'inode' into 'parent_inode' with
4576  * a give name and a given sequence number.
4577  * if 'add_backref' is true, also insert a backref from the
4578  * inode to the parent directory.
4579  */
4580 int btrfs_add_link(struct btrfs_trans_handle *trans,
4581                    struct inode *parent_inode, struct inode *inode,
4582                    const char *name, int name_len, int add_backref, u64 index)
4583 {
4584         int ret = 0;
4585         struct btrfs_key key;
4586         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4587         u64 ino = btrfs_ino(inode);
4588         u64 parent_ino = btrfs_ino(parent_inode);
4589
4590         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4591                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4592         } else {
4593                 key.objectid = ino;
4594                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4595                 key.offset = 0;
4596         }
4597
4598         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4599                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4600                                          key.objectid, root->root_key.objectid,
4601                                          parent_ino, index, name, name_len);
4602         } else if (add_backref) {
4603                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4604                                              parent_ino, index);
4605         }
4606
4607         if (ret == 0) {
4608                 ret = btrfs_insert_dir_item(trans, root, name, name_len,
4609                                             parent_inode, &key,
4610                                             btrfs_inode_type(inode), index);
4611                 BUG_ON(ret);
4612
4613                 btrfs_i_size_write(parent_inode, parent_inode->i_size +
4614                                    name_len * 2);
4615                 parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4616                 ret = btrfs_update_inode(trans, root, parent_inode);
4617         }
4618         return ret;
4619 }
4620
4621 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4622                             struct inode *dir, struct dentry *dentry,
4623                             struct inode *inode, int backref, u64 index)
4624 {
4625         int err = btrfs_add_link(trans, dir, inode,
4626                                  dentry->d_name.name, dentry->d_name.len,
4627                                  backref, index);
4628         if (!err) {
4629                 d_instantiate(dentry, inode);
4630                 return 0;
4631         }
4632         if (err > 0)
4633                 err = -EEXIST;
4634         return err;
4635 }
4636
4637 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4638                         int mode, dev_t rdev)
4639 {
4640         struct btrfs_trans_handle *trans;
4641         struct btrfs_root *root = BTRFS_I(dir)->root;
4642         struct inode *inode = NULL;
4643         int err;
4644         int drop_inode = 0;
4645         u64 objectid;
4646         unsigned long nr = 0;
4647         u64 index = 0;
4648
4649         if (!new_valid_dev(rdev))
4650                 return -EINVAL;
4651
4652         /*
4653          * 2 for inode item and ref
4654          * 2 for dir items
4655          * 1 for xattr if selinux is on
4656          */
4657         trans = btrfs_start_transaction(root, 5);
4658         if (IS_ERR(trans))
4659                 return PTR_ERR(trans);
4660
4661         err = btrfs_find_free_ino(root, &objectid);
4662         if (err)
4663                 goto out_unlock;
4664
4665         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4666                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4667                                 mode, &index);
4668         if (IS_ERR(inode)) {
4669                 err = PTR_ERR(inode);
4670                 goto out_unlock;
4671         }
4672
4673         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4674         if (err) {
4675                 drop_inode = 1;
4676                 goto out_unlock;
4677         }
4678
4679         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4680         if (err)
4681                 drop_inode = 1;
4682         else {
4683                 inode->i_op = &btrfs_special_inode_operations;
4684                 init_special_inode(inode, inode->i_mode, rdev);
4685                 btrfs_update_inode(trans, root, inode);
4686         }
4687 out_unlock:
4688         nr = trans->blocks_used;
4689         btrfs_end_transaction_throttle(trans, root);
4690         btrfs_btree_balance_dirty(root, nr);
4691         if (drop_inode) {
4692                 inode_dec_link_count(inode);
4693                 iput(inode);
4694         }
4695         return err;
4696 }
4697
4698 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4699                         int mode, struct nameidata *nd)
4700 {
4701         struct btrfs_trans_handle *trans;
4702         struct btrfs_root *root = BTRFS_I(dir)->root;
4703         struct inode *inode = NULL;
4704         int drop_inode = 0;
4705         int err;
4706         unsigned long nr = 0;
4707         u64 objectid;
4708         u64 index = 0;
4709
4710         /*
4711          * 2 for inode item and ref
4712          * 2 for dir items
4713          * 1 for xattr if selinux is on
4714          */
4715         trans = btrfs_start_transaction(root, 5);
4716         if (IS_ERR(trans))
4717                 return PTR_ERR(trans);
4718
4719         err = btrfs_find_free_ino(root, &objectid);
4720         if (err)
4721                 goto out_unlock;
4722
4723         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4724                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4725                                 mode, &index);
4726         if (IS_ERR(inode)) {
4727                 err = PTR_ERR(inode);
4728                 goto out_unlock;
4729         }
4730
4731         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4732         if (err) {
4733                 drop_inode = 1;
4734                 goto out_unlock;
4735         }
4736
4737         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4738         if (err)
4739                 drop_inode = 1;
4740         else {
4741                 inode->i_mapping->a_ops = &btrfs_aops;
4742                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
4743                 inode->i_fop = &btrfs_file_operations;
4744                 inode->i_op = &btrfs_file_inode_operations;
4745                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
4746         }
4747 out_unlock:
4748         nr = trans->blocks_used;
4749         btrfs_end_transaction_throttle(trans, root);
4750         if (drop_inode) {
4751                 inode_dec_link_count(inode);
4752                 iput(inode);
4753         }
4754         btrfs_btree_balance_dirty(root, nr);
4755         return err;
4756 }
4757
4758 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
4759                       struct dentry *dentry)
4760 {
4761         struct btrfs_trans_handle *trans;
4762         struct btrfs_root *root = BTRFS_I(dir)->root;
4763         struct inode *inode = old_dentry->d_inode;
4764         u64 index;
4765         unsigned long nr = 0;
4766         int err;
4767         int drop_inode = 0;
4768
4769         /* do not allow sys_link's with other subvols of the same device */
4770         if (root->objectid != BTRFS_I(inode)->root->objectid)
4771                 return -EXDEV;
4772
4773         if (inode->i_nlink == ~0U)
4774                 return -EMLINK;
4775
4776         err = btrfs_set_inode_index(dir, &index);
4777         if (err)
4778                 goto fail;
4779
4780         /*
4781          * 2 items for inode and inode ref
4782          * 2 items for dir items
4783          * 1 item for parent inode
4784          */
4785         trans = btrfs_start_transaction(root, 5);
4786         if (IS_ERR(trans)) {
4787                 err = PTR_ERR(trans);
4788                 goto fail;
4789         }
4790
4791         btrfs_inc_nlink(inode);
4792         inode->i_ctime = CURRENT_TIME;
4793         ihold(inode);
4794
4795         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
4796
4797         if (err) {
4798                 drop_inode = 1;
4799         } else {
4800                 struct dentry *parent = dentry->d_parent;
4801                 err = btrfs_update_inode(trans, root, inode);
4802                 BUG_ON(err);
4803                 btrfs_log_new_name(trans, inode, NULL, parent);
4804         }
4805
4806         nr = trans->blocks_used;
4807         btrfs_end_transaction_throttle(trans, root);
4808 fail:
4809         if (drop_inode) {
4810                 inode_dec_link_count(inode);
4811                 iput(inode);
4812         }
4813         btrfs_btree_balance_dirty(root, nr);
4814         return err;
4815 }
4816
4817 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
4818 {
4819         struct inode *inode = NULL;
4820         struct btrfs_trans_handle *trans;
4821         struct btrfs_root *root = BTRFS_I(dir)->root;
4822         int err = 0;
4823         int drop_on_err = 0;
4824         u64 objectid = 0;
4825         u64 index = 0;
4826         unsigned long nr = 1;
4827
4828         /*
4829          * 2 items for inode and ref
4830          * 2 items for dir items
4831          * 1 for xattr if selinux is on
4832          */
4833         trans = btrfs_start_transaction(root, 5);
4834         if (IS_ERR(trans))
4835                 return PTR_ERR(trans);
4836
4837         err = btrfs_find_free_ino(root, &objectid);
4838         if (err)
4839                 goto out_fail;
4840
4841         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4842                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4843                                 S_IFDIR | mode, &index);
4844         if (IS_ERR(inode)) {
4845                 err = PTR_ERR(inode);
4846                 goto out_fail;
4847         }
4848
4849         drop_on_err = 1;
4850
4851         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4852         if (err)
4853                 goto out_fail;
4854
4855         inode->i_op = &btrfs_dir_inode_operations;
4856         inode->i_fop = &btrfs_dir_file_operations;
4857
4858         btrfs_i_size_write(inode, 0);
4859         err = btrfs_update_inode(trans, root, inode);
4860         if (err)
4861                 goto out_fail;
4862
4863         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
4864                              dentry->d_name.len, 0, index);
4865         if (err)
4866                 goto out_fail;
4867
4868         d_instantiate(dentry, inode);
4869         drop_on_err = 0;
4870
4871 out_fail:
4872         nr = trans->blocks_used;
4873         btrfs_end_transaction_throttle(trans, root);
4874         if (drop_on_err)
4875                 iput(inode);
4876         btrfs_btree_balance_dirty(root, nr);
4877         return err;
4878 }
4879
4880 /* helper for btfs_get_extent.  Given an existing extent in the tree,
4881  * and an extent that you want to insert, deal with overlap and insert
4882  * the new extent into the tree.
4883  */
4884 static int merge_extent_mapping(struct extent_map_tree *em_tree,
4885                                 struct extent_map *existing,
4886                                 struct extent_map *em,
4887                                 u64 map_start, u64 map_len)
4888 {
4889         u64 start_diff;
4890
4891         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
4892         start_diff = map_start - em->start;
4893         em->start = map_start;
4894         em->len = map_len;
4895         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
4896             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
4897                 em->block_start += start_diff;
4898                 em->block_len -= start_diff;
4899         }
4900         return add_extent_mapping(em_tree, em);
4901 }
4902
4903 static noinline int uncompress_inline(struct btrfs_path *path,
4904                                       struct inode *inode, struct page *page,
4905                                       size_t pg_offset, u64 extent_offset,
4906                                       struct btrfs_file_extent_item *item)
4907 {
4908         int ret;
4909         struct extent_buffer *leaf = path->nodes[0];
4910         char *tmp;
4911         size_t max_size;
4912         unsigned long inline_size;
4913         unsigned long ptr;
4914         int compress_type;
4915
4916         WARN_ON(pg_offset != 0);
4917         compress_type = btrfs_file_extent_compression(leaf, item);
4918         max_size = btrfs_file_extent_ram_bytes(leaf, item);
4919         inline_size = btrfs_file_extent_inline_item_len(leaf,
4920                                         btrfs_item_nr(leaf, path->slots[0]));
4921         tmp = kmalloc(inline_size, GFP_NOFS);
4922         if (!tmp)
4923                 return -ENOMEM;
4924         ptr = btrfs_file_extent_inline_start(item);
4925
4926         read_extent_buffer(leaf, tmp, ptr, inline_size);
4927
4928         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
4929         ret = btrfs_decompress(compress_type, tmp, page,
4930                                extent_offset, inline_size, max_size);
4931         if (ret) {
4932                 char *kaddr = kmap_atomic(page, KM_USER0);
4933                 unsigned long copy_size = min_t(u64,
4934                                   PAGE_CACHE_SIZE - pg_offset,
4935                                   max_size - extent_offset);
4936                 memset(kaddr + pg_offset, 0, copy_size);
4937                 kunmap_atomic(kaddr, KM_USER0);
4938         }
4939         kfree(tmp);
4940         return 0;
4941 }
4942
4943 /*
4944  * a bit scary, this does extent mapping from logical file offset to the disk.
4945  * the ugly parts come from merging extents from the disk with the in-ram
4946  * representation.  This gets more complex because of the data=ordered code,
4947  * where the in-ram extents might be locked pending data=ordered completion.
4948  *
4949  * This also copies inline extents directly into the page.
4950  */
4951
4952 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
4953                                     size_t pg_offset, u64 start, u64 len,
4954                                     int create)
4955 {
4956         int ret;
4957         int err = 0;
4958         u64 bytenr;
4959         u64 extent_start = 0;
4960         u64 extent_end = 0;
4961         u64 objectid = btrfs_ino(inode);
4962         u32 found_type;
4963         struct btrfs_path *path = NULL;
4964         struct btrfs_root *root = BTRFS_I(inode)->root;
4965         struct btrfs_file_extent_item *item;
4966         struct extent_buffer *leaf;
4967         struct btrfs_key found_key;
4968         struct extent_map *em = NULL;
4969         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
4970         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
4971         struct btrfs_trans_handle *trans = NULL;
4972         int compress_type;
4973
4974 again:
4975         read_lock(&em_tree->lock);
4976         em = lookup_extent_mapping(em_tree, start, len);
4977         if (em)
4978                 em->bdev = root->fs_info->fs_devices->latest_bdev;
4979         read_unlock(&em_tree->lock);
4980
4981         if (em) {
4982                 if (em->start > start || em->start + em->len <= start)
4983                         free_extent_map(em);
4984                 else if (em->block_start == EXTENT_MAP_INLINE && page)
4985                         free_extent_map(em);
4986                 else
4987                         goto out;
4988         }
4989         em = alloc_extent_map();
4990         if (!em) {
4991                 err = -ENOMEM;
4992                 goto out;
4993         }
4994         em->bdev = root->fs_info->fs_devices->latest_bdev;
4995         em->start = EXTENT_MAP_HOLE;
4996         em->orig_start = EXTENT_MAP_HOLE;
4997         em->len = (u64)-1;
4998         em->block_len = (u64)-1;
4999
5000         if (!path) {
5001                 path = btrfs_alloc_path();
5002                 if (!path) {
5003                         err = -ENOMEM;
5004                         goto out;
5005                 }
5006                 /*
5007                  * Chances are we'll be called again, so go ahead and do
5008                  * readahead
5009                  */
5010                 path->reada = 1;
5011         }
5012
5013         ret = btrfs_lookup_file_extent(trans, root, path,
5014                                        objectid, start, trans != NULL);
5015         if (ret < 0) {
5016                 err = ret;
5017                 goto out;
5018         }
5019
5020         if (ret != 0) {
5021                 if (path->slots[0] == 0)
5022                         goto not_found;
5023                 path->slots[0]--;
5024         }
5025
5026         leaf = path->nodes[0];
5027         item = btrfs_item_ptr(leaf, path->slots[0],
5028                               struct btrfs_file_extent_item);
5029         /* are we inside the extent that was found? */
5030         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5031         found_type = btrfs_key_type(&found_key);
5032         if (found_key.objectid != objectid ||
5033             found_type != BTRFS_EXTENT_DATA_KEY) {
5034                 goto not_found;
5035         }
5036
5037         found_type = btrfs_file_extent_type(leaf, item);
5038         extent_start = found_key.offset;
5039         compress_type = btrfs_file_extent_compression(leaf, item);
5040         if (found_type == BTRFS_FILE_EXTENT_REG ||
5041             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5042                 extent_end = extent_start +
5043                        btrfs_file_extent_num_bytes(leaf, item);
5044         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5045                 size_t size;
5046                 size = btrfs_file_extent_inline_len(leaf, item);
5047                 extent_end = (extent_start + size + root->sectorsize - 1) &
5048                         ~((u64)root->sectorsize - 1);
5049         }
5050
5051         if (start >= extent_end) {
5052                 path->slots[0]++;
5053                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5054                         ret = btrfs_next_leaf(root, path);
5055                         if (ret < 0) {
5056                                 err = ret;
5057                                 goto out;
5058                         }
5059                         if (ret > 0)
5060                                 goto not_found;
5061                         leaf = path->nodes[0];
5062                 }
5063                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5064                 if (found_key.objectid != objectid ||
5065                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5066                         goto not_found;
5067                 if (start + len <= found_key.offset)
5068                         goto not_found;
5069                 em->start = start;
5070                 em->len = found_key.offset - start;
5071                 goto not_found_em;
5072         }
5073
5074         if (found_type == BTRFS_FILE_EXTENT_REG ||
5075             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5076                 em->start = extent_start;
5077                 em->len = extent_end - extent_start;
5078                 em->orig_start = extent_start -
5079                                  btrfs_file_extent_offset(leaf, item);
5080                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5081                 if (bytenr == 0) {
5082                         em->block_start = EXTENT_MAP_HOLE;
5083                         goto insert;
5084                 }
5085                 if (compress_type != BTRFS_COMPRESS_NONE) {
5086                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5087                         em->compress_type = compress_type;
5088                         em->block_start = bytenr;
5089                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5090                                                                          item);
5091                 } else {
5092                         bytenr += btrfs_file_extent_offset(leaf, item);
5093                         em->block_start = bytenr;
5094                         em->block_len = em->len;
5095                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5096                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5097                 }
5098                 goto insert;
5099         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5100                 unsigned long ptr;
5101                 char *map;
5102                 size_t size;
5103                 size_t extent_offset;
5104                 size_t copy_size;
5105
5106                 em->block_start = EXTENT_MAP_INLINE;
5107                 if (!page || create) {
5108                         em->start = extent_start;
5109                         em->len = extent_end - extent_start;
5110                         goto out;
5111                 }
5112
5113                 size = btrfs_file_extent_inline_len(leaf, item);
5114                 extent_offset = page_offset(page) + pg_offset - extent_start;
5115                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5116                                 size - extent_offset);
5117                 em->start = extent_start + extent_offset;
5118                 em->len = (copy_size + root->sectorsize - 1) &
5119                         ~((u64)root->sectorsize - 1);
5120                 em->orig_start = EXTENT_MAP_INLINE;
5121                 if (compress_type) {
5122                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5123                         em->compress_type = compress_type;
5124                 }
5125                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5126                 if (create == 0 && !PageUptodate(page)) {
5127                         if (btrfs_file_extent_compression(leaf, item) !=
5128                             BTRFS_COMPRESS_NONE) {
5129                                 ret = uncompress_inline(path, inode, page,
5130                                                         pg_offset,
5131                                                         extent_offset, item);
5132                                 BUG_ON(ret);
5133                         } else {
5134                                 map = kmap(page);
5135                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5136                                                    copy_size);
5137                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5138                                         memset(map + pg_offset + copy_size, 0,
5139                                                PAGE_CACHE_SIZE - pg_offset -
5140                                                copy_size);
5141                                 }
5142                                 kunmap(page);
5143                         }
5144                         flush_dcache_page(page);
5145                 } else if (create && PageUptodate(page)) {
5146                         WARN_ON(1);
5147                         if (!trans) {
5148                                 kunmap(page);
5149                                 free_extent_map(em);
5150                                 em = NULL;
5151
5152                                 btrfs_release_path(path);
5153                                 trans = btrfs_join_transaction(root);
5154
5155                                 if (IS_ERR(trans))
5156                                         return ERR_CAST(trans);
5157                                 goto again;
5158                         }
5159                         map = kmap(page);
5160                         write_extent_buffer(leaf, map + pg_offset, ptr,
5161                                             copy_size);
5162                         kunmap(page);
5163                         btrfs_mark_buffer_dirty(leaf);
5164                 }
5165                 set_extent_uptodate(io_tree, em->start,
5166                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5167                 goto insert;
5168         } else {
5169                 printk(KERN_ERR "btrfs unknown found_type %d\n", found_type);
5170                 WARN_ON(1);
5171         }
5172 not_found:
5173         em->start = start;
5174         em->len = len;
5175 not_found_em:
5176         em->block_start = EXTENT_MAP_HOLE;
5177         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5178 insert:
5179         btrfs_release_path(path);
5180         if (em->start > start || extent_map_end(em) <= start) {
5181                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5182                        "[%llu %llu]\n", (unsigned long long)em->start,
5183                        (unsigned long long)em->len,
5184                        (unsigned long long)start,
5185                        (unsigned long long)len);
5186                 err = -EIO;
5187                 goto out;
5188         }
5189
5190         err = 0;
5191         write_lock(&em_tree->lock);
5192         ret = add_extent_mapping(em_tree, em);
5193         /* it is possible that someone inserted the extent into the tree
5194          * while we had the lock dropped.  It is also possible that
5195          * an overlapping map exists in the tree
5196          */
5197         if (ret == -EEXIST) {
5198                 struct extent_map *existing;
5199
5200                 ret = 0;
5201
5202                 existing = lookup_extent_mapping(em_tree, start, len);
5203                 if (existing && (existing->start > start ||
5204                     existing->start + existing->len <= start)) {
5205                         free_extent_map(existing);
5206                         existing = NULL;
5207                 }
5208                 if (!existing) {
5209                         existing = lookup_extent_mapping(em_tree, em->start,
5210                                                          em->len);
5211                         if (existing) {
5212                                 err = merge_extent_mapping(em_tree, existing,
5213                                                            em, start,
5214                                                            root->sectorsize);
5215                                 free_extent_map(existing);
5216                                 if (err) {
5217                                         free_extent_map(em);
5218                                         em = NULL;
5219                                 }
5220                         } else {
5221                                 err = -EIO;
5222                                 free_extent_map(em);
5223                                 em = NULL;
5224                         }
5225                 } else {
5226                         free_extent_map(em);
5227                         em = existing;
5228                         err = 0;
5229                 }
5230         }
5231         write_unlock(&em_tree->lock);
5232 out:
5233
5234         trace_btrfs_get_extent(root, em);
5235
5236         if (path)
5237                 btrfs_free_path(path);
5238         if (trans) {
5239                 ret = btrfs_end_transaction(trans, root);
5240                 if (!err)
5241                         err = ret;
5242         }
5243         if (err) {
5244                 free_extent_map(em);
5245                 return ERR_PTR(err);
5246         }
5247         return em;
5248 }
5249
5250 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5251                                            size_t pg_offset, u64 start, u64 len,
5252                                            int create)
5253 {
5254         struct extent_map *em;
5255         struct extent_map *hole_em = NULL;
5256         u64 range_start = start;
5257         u64 end;
5258         u64 found;
5259         u64 found_end;
5260         int err = 0;
5261
5262         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5263         if (IS_ERR(em))
5264                 return em;
5265         if (em) {
5266                 /*
5267                  * if our em maps to a hole, there might
5268                  * actually be delalloc bytes behind it
5269                  */
5270                 if (em->block_start != EXTENT_MAP_HOLE)
5271                         return em;
5272                 else
5273                         hole_em = em;
5274         }
5275
5276         /* check to see if we've wrapped (len == -1 or similar) */
5277         end = start + len;
5278         if (end < start)
5279                 end = (u64)-1;
5280         else
5281                 end -= 1;
5282
5283         em = NULL;
5284
5285         /* ok, we didn't find anything, lets look for delalloc */
5286         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5287                                  end, len, EXTENT_DELALLOC, 1);
5288         found_end = range_start + found;
5289         if (found_end < range_start)
5290                 found_end = (u64)-1;
5291
5292         /*
5293          * we didn't find anything useful, return
5294          * the original results from get_extent()
5295          */
5296         if (range_start > end || found_end <= start) {
5297                 em = hole_em;
5298                 hole_em = NULL;
5299                 goto out;
5300         }
5301
5302         /* adjust the range_start to make sure it doesn't
5303          * go backwards from the start they passed in
5304          */
5305         range_start = max(start,range_start);
5306         found = found_end - range_start;
5307
5308         if (found > 0) {
5309                 u64 hole_start = start;
5310                 u64 hole_len = len;
5311
5312                 em = alloc_extent_map();
5313                 if (!em) {
5314                         err = -ENOMEM;
5315                         goto out;
5316                 }
5317                 /*
5318                  * when btrfs_get_extent can't find anything it
5319                  * returns one huge hole
5320                  *
5321                  * make sure what it found really fits our range, and
5322                  * adjust to make sure it is based on the start from
5323                  * the caller
5324                  */
5325                 if (hole_em) {
5326                         u64 calc_end = extent_map_end(hole_em);
5327
5328                         if (calc_end <= start || (hole_em->start > end)) {
5329                                 free_extent_map(hole_em);
5330                                 hole_em = NULL;
5331                         } else {
5332                                 hole_start = max(hole_em->start, start);
5333                                 hole_len = calc_end - hole_start;
5334                         }
5335                 }
5336                 em->bdev = NULL;
5337                 if (hole_em && range_start > hole_start) {
5338                         /* our hole starts before our delalloc, so we
5339                          * have to return just the parts of the hole
5340                          * that go until  the delalloc starts
5341                          */
5342                         em->len = min(hole_len,
5343                                       range_start - hole_start);
5344                         em->start = hole_start;
5345                         em->orig_start = hole_start;
5346                         /*
5347                          * don't adjust block start at all,
5348                          * it is fixed at EXTENT_MAP_HOLE
5349                          */
5350                         em->block_start = hole_em->block_start;
5351                         em->block_len = hole_len;
5352                 } else {
5353                         em->start = range_start;
5354                         em->len = found;
5355                         em->orig_start = range_start;
5356                         em->block_start = EXTENT_MAP_DELALLOC;
5357                         em->block_len = found;
5358                 }
5359         } else if (hole_em) {
5360                 return hole_em;
5361         }
5362 out:
5363
5364         free_extent_map(hole_em);
5365         if (err) {
5366                 free_extent_map(em);
5367                 return ERR_PTR(err);
5368         }
5369         return em;
5370 }
5371
5372 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5373                                                   struct extent_map *em,
5374                                                   u64 start, u64 len)
5375 {
5376         struct btrfs_root *root = BTRFS_I(inode)->root;
5377         struct btrfs_trans_handle *trans;
5378         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5379         struct btrfs_key ins;
5380         u64 alloc_hint;
5381         int ret;
5382         bool insert = false;
5383
5384         /*
5385          * Ok if the extent map we looked up is a hole and is for the exact
5386          * range we want, there is no reason to allocate a new one, however if
5387          * it is not right then we need to free this one and drop the cache for
5388          * our range.
5389          */
5390         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5391             em->len != len) {
5392                 free_extent_map(em);
5393                 em = NULL;
5394                 insert = true;
5395                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5396         }
5397
5398         trans = btrfs_join_transaction(root);
5399         if (IS_ERR(trans))
5400                 return ERR_CAST(trans);
5401
5402         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5403                 btrfs_add_inode_defrag(trans, inode);
5404
5405         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5406
5407         alloc_hint = get_extent_allocation_hint(inode, start, len);
5408         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5409                                    alloc_hint, (u64)-1, &ins, 1);
5410         if (ret) {
5411                 em = ERR_PTR(ret);
5412                 goto out;
5413         }
5414
5415         if (!em) {
5416                 em = alloc_extent_map();
5417                 if (!em) {
5418                         em = ERR_PTR(-ENOMEM);
5419                         goto out;
5420                 }
5421         }
5422
5423         em->start = start;
5424         em->orig_start = em->start;
5425         em->len = ins.offset;
5426
5427         em->block_start = ins.objectid;
5428         em->block_len = ins.offset;
5429         em->bdev = root->fs_info->fs_devices->latest_bdev;
5430
5431         /*
5432          * We need to do this because if we're using the original em we searched
5433          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5434          */
5435         em->flags = 0;
5436         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5437
5438         while (insert) {
5439                 write_lock(&em_tree->lock);
5440                 ret = add_extent_mapping(em_tree, em);
5441                 write_unlock(&em_tree->lock);
5442                 if (ret != -EEXIST)
5443                         break;
5444                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5445         }
5446
5447         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5448                                            ins.offset, ins.offset, 0);
5449         if (ret) {
5450                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5451                 em = ERR_PTR(ret);
5452         }
5453 out:
5454         btrfs_end_transaction(trans, root);
5455         return em;
5456 }
5457
5458 /*
5459  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5460  * block must be cow'd
5461  */
5462 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5463                                       struct inode *inode, u64 offset, u64 len)
5464 {
5465         struct btrfs_path *path;
5466         int ret;
5467         struct extent_buffer *leaf;
5468         struct btrfs_root *root = BTRFS_I(inode)->root;
5469         struct btrfs_file_extent_item *fi;
5470         struct btrfs_key key;
5471         u64 disk_bytenr;
5472         u64 backref_offset;
5473         u64 extent_end;
5474         u64 num_bytes;
5475         int slot;
5476         int found_type;
5477
5478         path = btrfs_alloc_path();
5479         if (!path)
5480                 return -ENOMEM;
5481
5482         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5483                                        offset, 0);
5484         if (ret < 0)
5485                 goto out;
5486
5487         slot = path->slots[0];
5488         if (ret == 1) {
5489                 if (slot == 0) {
5490                         /* can't find the item, must cow */
5491                         ret = 0;
5492                         goto out;
5493                 }
5494                 slot--;
5495         }
5496         ret = 0;
5497         leaf = path->nodes[0];
5498         btrfs_item_key_to_cpu(leaf, &key, slot);
5499         if (key.objectid != btrfs_ino(inode) ||
5500             key.type != BTRFS_EXTENT_DATA_KEY) {
5501                 /* not our file or wrong item type, must cow */
5502                 goto out;
5503         }
5504
5505         if (key.offset > offset) {
5506                 /* Wrong offset, must cow */
5507                 goto out;
5508         }
5509
5510         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5511         found_type = btrfs_file_extent_type(leaf, fi);
5512         if (found_type != BTRFS_FILE_EXTENT_REG &&
5513             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5514                 /* not a regular extent, must cow */
5515                 goto out;
5516         }
5517         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5518         backref_offset = btrfs_file_extent_offset(leaf, fi);
5519
5520         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5521         if (extent_end < offset + len) {
5522                 /* extent doesn't include our full range, must cow */
5523                 goto out;
5524         }
5525
5526         if (btrfs_extent_readonly(root, disk_bytenr))
5527                 goto out;
5528
5529         /*
5530          * look for other files referencing this extent, if we
5531          * find any we must cow
5532          */
5533         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5534                                   key.offset - backref_offset, disk_bytenr))
5535                 goto out;
5536
5537         /*
5538          * adjust disk_bytenr and num_bytes to cover just the bytes
5539          * in this extent we are about to write.  If there
5540          * are any csums in that range we have to cow in order
5541          * to keep the csums correct
5542          */
5543         disk_bytenr += backref_offset;
5544         disk_bytenr += offset - key.offset;
5545         num_bytes = min(offset + len, extent_end) - offset;
5546         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5547                                 goto out;
5548         /*
5549          * all of the above have passed, it is safe to overwrite this extent
5550          * without cow
5551          */
5552         ret = 1;
5553 out:
5554         btrfs_free_path(path);
5555         return ret;
5556 }
5557
5558 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5559                                    struct buffer_head *bh_result, int create)
5560 {
5561         struct extent_map *em;
5562         struct btrfs_root *root = BTRFS_I(inode)->root;
5563         u64 start = iblock << inode->i_blkbits;
5564         u64 len = bh_result->b_size;
5565         struct btrfs_trans_handle *trans;
5566
5567         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5568         if (IS_ERR(em))
5569                 return PTR_ERR(em);
5570
5571         /*
5572          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5573          * io.  INLINE is special, and we could probably kludge it in here, but
5574          * it's still buffered so for safety lets just fall back to the generic
5575          * buffered path.
5576          *
5577          * For COMPRESSED we _have_ to read the entire extent in so we can
5578          * decompress it, so there will be buffering required no matter what we
5579          * do, so go ahead and fallback to buffered.
5580          *
5581          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5582          * to buffered IO.  Don't blame me, this is the price we pay for using
5583          * the generic code.
5584          */
5585         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5586             em->block_start == EXTENT_MAP_INLINE) {
5587                 free_extent_map(em);
5588                 return -ENOTBLK;
5589         }
5590
5591         /* Just a good old fashioned hole, return */
5592         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5593                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5594                 free_extent_map(em);
5595                 /* DIO will do one hole at a time, so just unlock a sector */
5596                 unlock_extent(&BTRFS_I(inode)->io_tree, start,
5597                               start + root->sectorsize - 1, GFP_NOFS);
5598                 return 0;
5599         }
5600
5601         /*
5602          * We don't allocate a new extent in the following cases
5603          *
5604          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
5605          * existing extent.
5606          * 2) The extent is marked as PREALLOC.  We're good to go here and can
5607          * just use the extent.
5608          *
5609          */
5610         if (!create) {
5611                 len = em->len - (start - em->start);
5612                 goto map;
5613         }
5614
5615         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
5616             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
5617              em->block_start != EXTENT_MAP_HOLE)) {
5618                 int type;
5619                 int ret;
5620                 u64 block_start;
5621
5622                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5623                         type = BTRFS_ORDERED_PREALLOC;
5624                 else
5625                         type = BTRFS_ORDERED_NOCOW;
5626                 len = min(len, em->len - (start - em->start));
5627                 block_start = em->block_start + (start - em->start);
5628
5629                 /*
5630                  * we're not going to log anything, but we do need
5631                  * to make sure the current transaction stays open
5632                  * while we look for nocow cross refs
5633                  */
5634                 trans = btrfs_join_transaction(root);
5635                 if (IS_ERR(trans))
5636                         goto must_cow;
5637
5638                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
5639                         ret = btrfs_add_ordered_extent_dio(inode, start,
5640                                            block_start, len, len, type);
5641                         btrfs_end_transaction(trans, root);
5642                         if (ret) {
5643                                 free_extent_map(em);
5644                                 return ret;
5645                         }
5646                         goto unlock;
5647                 }
5648                 btrfs_end_transaction(trans, root);
5649         }
5650 must_cow:
5651         /*
5652          * this will cow the extent, reset the len in case we changed
5653          * it above
5654          */
5655         len = bh_result->b_size;
5656         em = btrfs_new_extent_direct(inode, em, start, len);
5657         if (IS_ERR(em))
5658                 return PTR_ERR(em);
5659         len = min(len, em->len - (start - em->start));
5660 unlock:
5661         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, start + len - 1,
5662                           EXTENT_LOCKED | EXTENT_DELALLOC | EXTENT_DIRTY, 1,
5663                           0, NULL, GFP_NOFS);
5664 map:
5665         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
5666                 inode->i_blkbits;
5667         bh_result->b_size = len;
5668         bh_result->b_bdev = em->bdev;
5669         set_buffer_mapped(bh_result);
5670         if (create && !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5671                 set_buffer_new(bh_result);
5672
5673         free_extent_map(em);
5674
5675         return 0;
5676 }
5677
5678 struct btrfs_dio_private {
5679         struct inode *inode;
5680         u64 logical_offset;
5681         u64 disk_bytenr;
5682         u64 bytes;
5683         u32 *csums;
5684         void *private;
5685
5686         /* number of bios pending for this dio */
5687         atomic_t pending_bios;
5688
5689         /* IO errors */
5690         int errors;
5691
5692         struct bio *orig_bio;
5693 };
5694
5695 static void btrfs_endio_direct_read(struct bio *bio, int err)
5696 {
5697         struct btrfs_dio_private *dip = bio->bi_private;
5698         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
5699         struct bio_vec *bvec = bio->bi_io_vec;
5700         struct inode *inode = dip->inode;
5701         struct btrfs_root *root = BTRFS_I(inode)->root;
5702         u64 start;
5703         u32 *private = dip->csums;
5704
5705         start = dip->logical_offset;
5706         do {
5707                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
5708                         struct page *page = bvec->bv_page;
5709                         char *kaddr;
5710                         u32 csum = ~(u32)0;
5711                         unsigned long flags;
5712
5713                         local_irq_save(flags);
5714                         kaddr = kmap_atomic(page, KM_IRQ0);
5715                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
5716                                                csum, bvec->bv_len);
5717                         btrfs_csum_final(csum, (char *)&csum);
5718                         kunmap_atomic(kaddr, KM_IRQ0);
5719                         local_irq_restore(flags);
5720
5721                         flush_dcache_page(bvec->bv_page);
5722                         if (csum != *private) {
5723                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
5724                                       " %llu csum %u private %u\n",
5725                                       (unsigned long long)btrfs_ino(inode),
5726                                       (unsigned long long)start,
5727                                       csum, *private);
5728                                 err = -EIO;
5729                         }
5730                 }
5731
5732                 start += bvec->bv_len;
5733                 private++;
5734                 bvec++;
5735         } while (bvec <= bvec_end);
5736
5737         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
5738                       dip->logical_offset + dip->bytes - 1, GFP_NOFS);
5739         bio->bi_private = dip->private;
5740
5741         kfree(dip->csums);
5742         kfree(dip);
5743
5744         /* If we had a csum failure make sure to clear the uptodate flag */
5745         if (err)
5746                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5747         dio_end_io(bio, err);
5748 }
5749
5750 static void btrfs_endio_direct_write(struct bio *bio, int err)
5751 {
5752         struct btrfs_dio_private *dip = bio->bi_private;
5753         struct inode *inode = dip->inode;
5754         struct btrfs_root *root = BTRFS_I(inode)->root;
5755         struct btrfs_trans_handle *trans;
5756         struct btrfs_ordered_extent *ordered = NULL;
5757         struct extent_state *cached_state = NULL;
5758         u64 ordered_offset = dip->logical_offset;
5759         u64 ordered_bytes = dip->bytes;
5760         int ret;
5761
5762         if (err)
5763                 goto out_done;
5764 again:
5765         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
5766                                                    &ordered_offset,
5767                                                    ordered_bytes);
5768         if (!ret)
5769                 goto out_test;
5770
5771         BUG_ON(!ordered);
5772
5773         trans = btrfs_join_transaction(root);
5774         if (IS_ERR(trans)) {
5775                 err = -ENOMEM;
5776                 goto out;
5777         }
5778         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5779
5780         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags)) {
5781                 ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5782                 if (!ret)
5783                         err = btrfs_update_inode(trans, root, inode);
5784                 goto out;
5785         }
5786
5787         lock_extent_bits(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5788                          ordered->file_offset + ordered->len - 1, 0,
5789                          &cached_state, GFP_NOFS);
5790
5791         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) {
5792                 ret = btrfs_mark_extent_written(trans, inode,
5793                                                 ordered->file_offset,
5794                                                 ordered->file_offset +
5795                                                 ordered->len);
5796                 if (ret) {
5797                         err = ret;
5798                         goto out_unlock;
5799                 }
5800         } else {
5801                 ret = insert_reserved_file_extent(trans, inode,
5802                                                   ordered->file_offset,
5803                                                   ordered->start,
5804                                                   ordered->disk_len,
5805                                                   ordered->len,
5806                                                   ordered->len,
5807                                                   0, 0, 0,
5808                                                   BTRFS_FILE_EXTENT_REG);
5809                 unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
5810                                    ordered->file_offset, ordered->len);
5811                 if (ret) {
5812                         err = ret;
5813                         WARN_ON(1);
5814                         goto out_unlock;
5815                 }
5816         }
5817
5818         add_pending_csums(trans, inode, ordered->file_offset, &ordered->list);
5819         ret = btrfs_ordered_update_i_size(inode, 0, ordered);
5820         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags))
5821                 btrfs_update_inode(trans, root, inode);
5822         ret = 0;
5823 out_unlock:
5824         unlock_extent_cached(&BTRFS_I(inode)->io_tree, ordered->file_offset,
5825                              ordered->file_offset + ordered->len - 1,
5826                              &cached_state, GFP_NOFS);
5827 out:
5828         btrfs_delalloc_release_metadata(inode, ordered->len);
5829         btrfs_end_transaction(trans, root);
5830         ordered_offset = ordered->file_offset + ordered->len;
5831         btrfs_put_ordered_extent(ordered);
5832         btrfs_put_ordered_extent(ordered);
5833
5834 out_test:
5835         /*
5836          * our bio might span multiple ordered extents.  If we haven't
5837          * completed the accounting for the whole dio, go back and try again
5838          */
5839         if (ordered_offset < dip->logical_offset + dip->bytes) {
5840                 ordered_bytes = dip->logical_offset + dip->bytes -
5841                         ordered_offset;
5842                 goto again;
5843         }
5844 out_done:
5845         bio->bi_private = dip->private;
5846
5847         kfree(dip->csums);
5848         kfree(dip);
5849
5850         /* If we had an error make sure to clear the uptodate flag */
5851         if (err)
5852                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
5853         dio_end_io(bio, err);
5854 }
5855
5856 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
5857                                     struct bio *bio, int mirror_num,
5858                                     unsigned long bio_flags, u64 offset)
5859 {
5860         int ret;
5861         struct btrfs_root *root = BTRFS_I(inode)->root;
5862         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
5863         BUG_ON(ret);
5864         return 0;
5865 }
5866
5867 static void btrfs_end_dio_bio(struct bio *bio, int err)
5868 {
5869         struct btrfs_dio_private *dip = bio->bi_private;
5870
5871         if (err) {
5872                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
5873                       "sector %#Lx len %u err no %d\n",
5874                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
5875                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
5876                 dip->errors = 1;
5877
5878                 /*
5879                  * before atomic variable goto zero, we must make sure
5880                  * dip->errors is perceived to be set.
5881                  */
5882                 smp_mb__before_atomic_dec();
5883         }
5884
5885         /* if there are more bios still pending for this dio, just exit */
5886         if (!atomic_dec_and_test(&dip->pending_bios))
5887                 goto out;
5888
5889         if (dip->errors)
5890                 bio_io_error(dip->orig_bio);
5891         else {
5892                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
5893                 bio_endio(dip->orig_bio, 0);
5894         }
5895 out:
5896         bio_put(bio);
5897 }
5898
5899 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
5900                                        u64 first_sector, gfp_t gfp_flags)
5901 {
5902         int nr_vecs = bio_get_nr_vecs(bdev);
5903         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
5904 }
5905
5906 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
5907                                          int rw, u64 file_offset, int skip_sum,
5908                                          u32 *csums, int async_submit)
5909 {
5910         int write = rw & REQ_WRITE;
5911         struct btrfs_root *root = BTRFS_I(inode)->root;
5912         int ret;
5913
5914         bio_get(bio);
5915         ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
5916         if (ret)
5917                 goto err;
5918
5919         if (skip_sum)
5920                 goto map;
5921
5922         if (write && async_submit) {
5923                 ret = btrfs_wq_submit_bio(root->fs_info,
5924                                    inode, rw, bio, 0, 0,
5925                                    file_offset,
5926                                    __btrfs_submit_bio_start_direct_io,
5927                                    __btrfs_submit_bio_done);
5928                 goto err;
5929         } else if (write) {
5930                 /*
5931                  * If we aren't doing async submit, calculate the csum of the
5932                  * bio now.
5933                  */
5934                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
5935                 if (ret)
5936                         goto err;
5937         } else if (!skip_sum) {
5938                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio,
5939                                           file_offset, csums);
5940                 if (ret)
5941                         goto err;
5942         }
5943
5944 map:
5945         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
5946 err:
5947         bio_put(bio);
5948         return ret;
5949 }
5950
5951 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
5952                                     int skip_sum)
5953 {
5954         struct inode *inode = dip->inode;
5955         struct btrfs_root *root = BTRFS_I(inode)->root;
5956         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5957         struct bio *bio;
5958         struct bio *orig_bio = dip->orig_bio;
5959         struct bio_vec *bvec = orig_bio->bi_io_vec;
5960         u64 start_sector = orig_bio->bi_sector;
5961         u64 file_offset = dip->logical_offset;
5962         u64 submit_len = 0;
5963         u64 map_length;
5964         int nr_pages = 0;
5965         u32 *csums = dip->csums;
5966         int ret = 0;
5967         int async_submit = 0;
5968         int write = rw & REQ_WRITE;
5969
5970         map_length = orig_bio->bi_size;
5971         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
5972                               &map_length, NULL, 0);
5973         if (ret) {
5974                 bio_put(orig_bio);
5975                 return -EIO;
5976         }
5977
5978         if (map_length >= orig_bio->bi_size) {
5979                 bio = orig_bio;
5980                 goto submit;
5981         }
5982
5983         async_submit = 1;
5984         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
5985         if (!bio)
5986                 return -ENOMEM;
5987         bio->bi_private = dip;
5988         bio->bi_end_io = btrfs_end_dio_bio;
5989         atomic_inc(&dip->pending_bios);
5990
5991         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
5992                 if (unlikely(map_length < submit_len + bvec->bv_len ||
5993                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5994                                  bvec->bv_offset) < bvec->bv_len)) {
5995                         /*
5996                          * inc the count before we submit the bio so
5997                          * we know the end IO handler won't happen before
5998                          * we inc the count. Otherwise, the dip might get freed
5999                          * before we're done setting it up
6000                          */
6001                         atomic_inc(&dip->pending_bios);
6002                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6003                                                      file_offset, skip_sum,
6004                                                      csums, async_submit);
6005                         if (ret) {
6006                                 bio_put(bio);
6007                                 atomic_dec(&dip->pending_bios);
6008                                 goto out_err;
6009                         }
6010
6011                         /* Write's use the ordered csums */
6012                         if (!write && !skip_sum)
6013                                 csums = csums + nr_pages;
6014                         start_sector += submit_len >> 9;
6015                         file_offset += submit_len;
6016
6017                         submit_len = 0;
6018                         nr_pages = 0;
6019
6020                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6021                                                   start_sector, GFP_NOFS);
6022                         if (!bio)
6023                                 goto out_err;
6024                         bio->bi_private = dip;
6025                         bio->bi_end_io = btrfs_end_dio_bio;
6026
6027                         map_length = orig_bio->bi_size;
6028                         ret = btrfs_map_block(map_tree, READ, start_sector << 9,
6029                                               &map_length, NULL, 0);
6030                         if (ret) {
6031                                 bio_put(bio);
6032                                 goto out_err;
6033                         }
6034                 } else {
6035                         submit_len += bvec->bv_len;
6036                         nr_pages ++;
6037                         bvec++;
6038                 }
6039         }
6040
6041 submit:
6042         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6043                                      csums, async_submit);
6044         if (!ret)
6045                 return 0;
6046
6047         bio_put(bio);
6048 out_err:
6049         dip->errors = 1;
6050         /*
6051          * before atomic variable goto zero, we must
6052          * make sure dip->errors is perceived to be set.
6053          */
6054         smp_mb__before_atomic_dec();
6055         if (atomic_dec_and_test(&dip->pending_bios))
6056                 bio_io_error(dip->orig_bio);
6057
6058         /* bio_end_io() will handle error, so we needn't return it */
6059         return 0;
6060 }
6061
6062 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6063                                 loff_t file_offset)
6064 {
6065         struct btrfs_root *root = BTRFS_I(inode)->root;
6066         struct btrfs_dio_private *dip;
6067         struct bio_vec *bvec = bio->bi_io_vec;
6068         int skip_sum;
6069         int write = rw & REQ_WRITE;
6070         int ret = 0;
6071
6072         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6073
6074         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6075         if (!dip) {
6076                 ret = -ENOMEM;
6077                 goto free_ordered;
6078         }
6079         dip->csums = NULL;
6080
6081         /* Write's use the ordered csum stuff, so we don't need dip->csums */
6082         if (!write && !skip_sum) {
6083                 dip->csums = kmalloc(sizeof(u32) * bio->bi_vcnt, GFP_NOFS);
6084                 if (!dip->csums) {
6085                         kfree(dip);
6086                         ret = -ENOMEM;
6087                         goto free_ordered;
6088                 }
6089         }
6090
6091         dip->private = bio->bi_private;
6092         dip->inode = inode;
6093         dip->logical_offset = file_offset;
6094
6095         dip->bytes = 0;
6096         do {
6097                 dip->bytes += bvec->bv_len;
6098                 bvec++;
6099         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6100
6101         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6102         bio->bi_private = dip;
6103         dip->errors = 0;
6104         dip->orig_bio = bio;
6105         atomic_set(&dip->pending_bios, 0);
6106
6107         if (write)
6108                 bio->bi_end_io = btrfs_endio_direct_write;
6109         else
6110                 bio->bi_end_io = btrfs_endio_direct_read;
6111
6112         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6113         if (!ret)
6114                 return;
6115 free_ordered:
6116         /*
6117          * If this is a write, we need to clean up the reserved space and kill
6118          * the ordered extent.
6119          */
6120         if (write) {
6121                 struct btrfs_ordered_extent *ordered;
6122                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6123                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6124                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6125                         btrfs_free_reserved_extent(root, ordered->start,
6126                                                    ordered->disk_len);
6127                 btrfs_put_ordered_extent(ordered);
6128                 btrfs_put_ordered_extent(ordered);
6129         }
6130         bio_endio(bio, ret);
6131 }
6132
6133 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6134                         const struct iovec *iov, loff_t offset,
6135                         unsigned long nr_segs)
6136 {
6137         int seg;
6138         int i;
6139         size_t size;
6140         unsigned long addr;
6141         unsigned blocksize_mask = root->sectorsize - 1;
6142         ssize_t retval = -EINVAL;
6143         loff_t end = offset;
6144
6145         if (offset & blocksize_mask)
6146                 goto out;
6147
6148         /* Check the memory alignment.  Blocks cannot straddle pages */
6149         for (seg = 0; seg < nr_segs; seg++) {
6150                 addr = (unsigned long)iov[seg].iov_base;
6151                 size = iov[seg].iov_len;
6152                 end += size;
6153                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6154                         goto out;
6155
6156                 /* If this is a write we don't need to check anymore */
6157                 if (rw & WRITE)
6158                         continue;
6159
6160                 /*
6161                  * Check to make sure we don't have duplicate iov_base's in this
6162                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6163                  * when reading back.
6164                  */
6165                 for (i = seg + 1; i < nr_segs; i++) {
6166                         if (iov[seg].iov_base == iov[i].iov_base)
6167                                 goto out;
6168                 }
6169         }
6170         retval = 0;
6171 out:
6172         return retval;
6173 }
6174 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6175                         const struct iovec *iov, loff_t offset,
6176                         unsigned long nr_segs)
6177 {
6178         struct file *file = iocb->ki_filp;
6179         struct inode *inode = file->f_mapping->host;
6180         struct btrfs_ordered_extent *ordered;
6181         struct extent_state *cached_state = NULL;
6182         u64 lockstart, lockend;
6183         ssize_t ret;
6184         int writing = rw & WRITE;
6185         int write_bits = 0;
6186         size_t count = iov_length(iov, nr_segs);
6187
6188         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6189                             offset, nr_segs)) {
6190                 return 0;
6191         }
6192
6193         lockstart = offset;
6194         lockend = offset + count - 1;
6195
6196         if (writing) {
6197                 ret = btrfs_delalloc_reserve_space(inode, count);
6198                 if (ret)
6199                         goto out;
6200         }
6201
6202         while (1) {
6203                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6204                                  0, &cached_state, GFP_NOFS);
6205                 /*
6206                  * We're concerned with the entire range that we're going to be
6207                  * doing DIO to, so we need to make sure theres no ordered
6208                  * extents in this range.
6209                  */
6210                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
6211                                                      lockend - lockstart + 1);
6212                 if (!ordered)
6213                         break;
6214                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6215                                      &cached_state, GFP_NOFS);
6216                 btrfs_start_ordered_extent(inode, ordered, 1);
6217                 btrfs_put_ordered_extent(ordered);
6218                 cond_resched();
6219         }
6220
6221         /*
6222          * we don't use btrfs_set_extent_delalloc because we don't want
6223          * the dirty or uptodate bits
6224          */
6225         if (writing) {
6226                 write_bits = EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING;
6227                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6228                                      EXTENT_DELALLOC, 0, NULL, &cached_state,
6229                                      GFP_NOFS);
6230                 if (ret) {
6231                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6232                                          lockend, EXTENT_LOCKED | write_bits,
6233                                          1, 0, &cached_state, GFP_NOFS);
6234                         goto out;
6235                 }
6236         }
6237
6238         free_extent_state(cached_state);
6239         cached_state = NULL;
6240
6241         ret = __blockdev_direct_IO(rw, iocb, inode,
6242                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6243                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6244                    btrfs_submit_direct, 0);
6245
6246         if (ret < 0 && ret != -EIOCBQUEUED) {
6247                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset,
6248                               offset + iov_length(iov, nr_segs) - 1,
6249                               EXTENT_LOCKED | write_bits, 1, 0,
6250                               &cached_state, GFP_NOFS);
6251         } else if (ret >= 0 && ret < iov_length(iov, nr_segs)) {
6252                 /*
6253                  * We're falling back to buffered, unlock the section we didn't
6254                  * do IO on.
6255                  */
6256                 clear_extent_bit(&BTRFS_I(inode)->io_tree, offset + ret,
6257                               offset + iov_length(iov, nr_segs) - 1,
6258                               EXTENT_LOCKED | write_bits, 1, 0,
6259                               &cached_state, GFP_NOFS);
6260         }
6261 out:
6262         free_extent_state(cached_state);
6263         return ret;
6264 }
6265
6266 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6267                 __u64 start, __u64 len)
6268 {
6269         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6270 }
6271
6272 int btrfs_readpage(struct file *file, struct page *page)
6273 {
6274         struct extent_io_tree *tree;
6275         tree = &BTRFS_I(page->mapping->host)->io_tree;
6276         return extent_read_full_page(tree, page, btrfs_get_extent);
6277 }
6278
6279 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6280 {
6281         struct extent_io_tree *tree;
6282
6283
6284         if (current->flags & PF_MEMALLOC) {
6285                 redirty_page_for_writepage(wbc, page);
6286                 unlock_page(page);
6287                 return 0;
6288         }
6289         tree = &BTRFS_I(page->mapping->host)->io_tree;
6290         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6291 }
6292
6293 int btrfs_writepages(struct address_space *mapping,
6294                      struct writeback_control *wbc)
6295 {
6296         struct extent_io_tree *tree;
6297
6298         tree = &BTRFS_I(mapping->host)->io_tree;
6299         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6300 }
6301
6302 static int
6303 btrfs_readpages(struct file *file, struct address_space *mapping,
6304                 struct list_head *pages, unsigned nr_pages)
6305 {
6306         struct extent_io_tree *tree;
6307         tree = &BTRFS_I(mapping->host)->io_tree;
6308         return extent_readpages(tree, mapping, pages, nr_pages,
6309                                 btrfs_get_extent);
6310 }
6311 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6312 {
6313         struct extent_io_tree *tree;
6314         struct extent_map_tree *map;
6315         int ret;
6316
6317         tree = &BTRFS_I(page->mapping->host)->io_tree;
6318         map = &BTRFS_I(page->mapping->host)->extent_tree;
6319         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6320         if (ret == 1) {
6321                 ClearPagePrivate(page);
6322                 set_page_private(page, 0);
6323                 page_cache_release(page);
6324         }
6325         return ret;
6326 }
6327
6328 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6329 {
6330         if (PageWriteback(page) || PageDirty(page))
6331                 return 0;
6332         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6333 }
6334
6335 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6336 {
6337         struct extent_io_tree *tree;
6338         struct btrfs_ordered_extent *ordered;
6339         struct extent_state *cached_state = NULL;
6340         u64 page_start = page_offset(page);
6341         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6342
6343
6344         /*
6345          * we have the page locked, so new writeback can't start,
6346          * and the dirty bit won't be cleared while we are here.
6347          *
6348          * Wait for IO on this page so that we can safely clear
6349          * the PagePrivate2 bit and do ordered accounting
6350          */
6351         wait_on_page_writeback(page);
6352
6353         tree = &BTRFS_I(page->mapping->host)->io_tree;
6354         if (offset) {
6355                 btrfs_releasepage(page, GFP_NOFS);
6356                 return;
6357         }
6358         lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6359                          GFP_NOFS);
6360         ordered = btrfs_lookup_ordered_extent(page->mapping->host,
6361                                            page_offset(page));
6362         if (ordered) {
6363                 /*
6364                  * IO on this page will never be started, so we need
6365                  * to account for any ordered extents now
6366                  */
6367                 clear_extent_bit(tree, page_start, page_end,
6368                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6369                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING, 1, 0,
6370                                  &cached_state, GFP_NOFS);
6371                 /*
6372                  * whoever cleared the private bit is responsible
6373                  * for the finish_ordered_io
6374                  */
6375                 if (TestClearPagePrivate2(page)) {
6376                         btrfs_finish_ordered_io(page->mapping->host,
6377                                                 page_start, page_end);
6378                 }
6379                 btrfs_put_ordered_extent(ordered);
6380                 cached_state = NULL;
6381                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state,
6382                                  GFP_NOFS);
6383         }
6384         clear_extent_bit(tree, page_start, page_end,
6385                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6386                  EXTENT_DO_ACCOUNTING, 1, 1, &cached_state, GFP_NOFS);
6387         __btrfs_releasepage(page, GFP_NOFS);
6388
6389         ClearPageChecked(page);
6390         if (PagePrivate(page)) {
6391                 ClearPagePrivate(page);
6392                 set_page_private(page, 0);
6393                 page_cache_release(page);
6394         }
6395 }
6396
6397 /*
6398  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6399  * called from a page fault handler when a page is first dirtied. Hence we must
6400  * be careful to check for EOF conditions here. We set the page up correctly
6401  * for a written page which means we get ENOSPC checking when writing into
6402  * holes and correct delalloc and unwritten extent mapping on filesystems that
6403  * support these features.
6404  *
6405  * We are not allowed to take the i_mutex here so we have to play games to
6406  * protect against truncate races as the page could now be beyond EOF.  Because
6407  * vmtruncate() writes the inode size before removing pages, once we have the
6408  * page lock we can determine safely if the page is beyond EOF. If it is not
6409  * beyond EOF, then the page is guaranteed safe against truncation until we
6410  * unlock the page.
6411  */
6412 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6413 {
6414         struct page *page = vmf->page;
6415         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6416         struct btrfs_root *root = BTRFS_I(inode)->root;
6417         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6418         struct btrfs_ordered_extent *ordered;
6419         struct extent_state *cached_state = NULL;
6420         char *kaddr;
6421         unsigned long zero_start;
6422         loff_t size;
6423         int ret;
6424         u64 page_start;
6425         u64 page_end;
6426
6427         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6428         if (ret) {
6429                 if (ret == -ENOMEM)
6430                         ret = VM_FAULT_OOM;
6431                 else /* -ENOSPC, -EIO, etc */
6432                         ret = VM_FAULT_SIGBUS;
6433                 goto out;
6434         }
6435
6436         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6437 again:
6438         lock_page(page);
6439         size = i_size_read(inode);
6440         page_start = page_offset(page);
6441         page_end = page_start + PAGE_CACHE_SIZE - 1;
6442
6443         if ((page->mapping != inode->i_mapping) ||
6444             (page_start >= size)) {
6445                 /* page got truncated out from underneath us */
6446                 goto out_unlock;
6447         }
6448         wait_on_page_writeback(page);
6449
6450         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state,
6451                          GFP_NOFS);
6452         set_page_extent_mapped(page);
6453
6454         /*
6455          * we can't set the delalloc bits if there are pending ordered
6456          * extents.  Drop our locks and wait for them to finish
6457          */
6458         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6459         if (ordered) {
6460                 unlock_extent_cached(io_tree, page_start, page_end,
6461                                      &cached_state, GFP_NOFS);
6462                 unlock_page(page);
6463                 btrfs_start_ordered_extent(inode, ordered, 1);
6464                 btrfs_put_ordered_extent(ordered);
6465                 goto again;
6466         }
6467
6468         /*
6469          * XXX - page_mkwrite gets called every time the page is dirtied, even
6470          * if it was already dirty, so for space accounting reasons we need to
6471          * clear any delalloc bits for the range we are fixing to save.  There
6472          * is probably a better way to do this, but for now keep consistent with
6473          * prepare_pages in the normal write path.
6474          */
6475         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6476                           EXTENT_DIRTY | EXTENT_DELALLOC | EXTENT_DO_ACCOUNTING,
6477                           0, 0, &cached_state, GFP_NOFS);
6478
6479         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6480                                         &cached_state);
6481         if (ret) {
6482                 unlock_extent_cached(io_tree, page_start, page_end,
6483                                      &cached_state, GFP_NOFS);
6484                 ret = VM_FAULT_SIGBUS;
6485                 goto out_unlock;
6486         }
6487         ret = 0;
6488
6489         /* page is wholly or partially inside EOF */
6490         if (page_start + PAGE_CACHE_SIZE > size)
6491                 zero_start = size & ~PAGE_CACHE_MASK;
6492         else
6493                 zero_start = PAGE_CACHE_SIZE;
6494
6495         if (zero_start != PAGE_CACHE_SIZE) {
6496                 kaddr = kmap(page);
6497                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6498                 flush_dcache_page(page);
6499                 kunmap(page);
6500         }
6501         ClearPageChecked(page);
6502         set_page_dirty(page);
6503         SetPageUptodate(page);
6504
6505         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6506         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6507
6508         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6509
6510 out_unlock:
6511         if (!ret)
6512                 return VM_FAULT_LOCKED;
6513         unlock_page(page);
6514         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6515 out:
6516         return ret;
6517 }
6518
6519 static int btrfs_truncate(struct inode *inode)
6520 {
6521         struct btrfs_root *root = BTRFS_I(inode)->root;
6522         struct btrfs_block_rsv *rsv;
6523         int ret;
6524         int err = 0;
6525         struct btrfs_trans_handle *trans;
6526         unsigned long nr;
6527         u64 mask = root->sectorsize - 1;
6528         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6529
6530         ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
6531         if (ret)
6532                 return ret;
6533
6534         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6535         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6536
6537         /*
6538          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6539          * 3 things going on here
6540          *
6541          * 1) We need to reserve space for our orphan item and the space to
6542          * delete our orphan item.  Lord knows we don't want to have a dangling
6543          * orphan item because we didn't reserve space to remove it.
6544          *
6545          * 2) We need to reserve space to update our inode.
6546          *
6547          * 3) We need to have something to cache all the space that is going to
6548          * be free'd up by the truncate operation, but also have some slack
6549          * space reserved in case it uses space during the truncate (thank you
6550          * very much snapshotting).
6551          *
6552          * And we need these to all be seperate.  The fact is we can use alot of
6553          * space doing the truncate, and we have no earthly idea how much space
6554          * we will use, so we need the truncate reservation to be seperate so it
6555          * doesn't end up using space reserved for updating the inode or
6556          * removing the orphan item.  We also need to be able to stop the
6557          * transaction and start a new one, which means we need to be able to
6558          * update the inode several times, and we have no idea of knowing how
6559          * many times that will be, so we can't just reserve 1 item for the
6560          * entirety of the opration, so that has to be done seperately as well.
6561          * Then there is the orphan item, which does indeed need to be held on
6562          * to for the whole operation, and we need nobody to touch this reserved
6563          * space except the orphan code.
6564          *
6565          * So that leaves us with
6566          *
6567          * 1) root->orphan_block_rsv - for the orphan deletion.
6568          * 2) rsv - for the truncate reservation, which we will steal from the
6569          * transaction reservation.
6570          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6571          * updating the inode.
6572          */
6573         rsv = btrfs_alloc_block_rsv(root);
6574         if (!rsv)
6575                 return -ENOMEM;
6576         rsv->size = min_size;
6577
6578         /*
6579          * 1 for the truncate slack space
6580          * 1 for the orphan item we're going to add
6581          * 1 for the orphan item deletion
6582          * 1 for updating the inode.
6583          */
6584         trans = btrfs_start_transaction(root, 4);
6585         if (IS_ERR(trans)) {
6586                 err = PTR_ERR(trans);
6587                 goto out;
6588         }
6589
6590         /* Migrate the slack space for the truncate to our reserve */
6591         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6592                                       min_size);
6593         BUG_ON(ret);
6594
6595         ret = btrfs_orphan_add(trans, inode);
6596         if (ret) {
6597                 btrfs_end_transaction(trans, root);
6598                 goto out;
6599         }
6600
6601         /*
6602          * setattr is responsible for setting the ordered_data_close flag,
6603          * but that is only tested during the last file release.  That
6604          * could happen well after the next commit, leaving a great big
6605          * window where new writes may get lost if someone chooses to write
6606          * to this file after truncating to zero
6607          *
6608          * The inode doesn't have any dirty data here, and so if we commit
6609          * this is a noop.  If someone immediately starts writing to the inode
6610          * it is very likely we'll catch some of their writes in this
6611          * transaction, and the commit will find this file on the ordered
6612          * data list with good things to send down.
6613          *
6614          * This is a best effort solution, there is still a window where
6615          * using truncate to replace the contents of the file will
6616          * end up with a zero length file after a crash.
6617          */
6618         if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
6619                 btrfs_add_ordered_operation(trans, root, inode);
6620
6621         while (1) {
6622                 ret = btrfs_block_rsv_refill(root, rsv, min_size);
6623                 if (ret) {
6624                         /*
6625                          * This can only happen with the original transaction we
6626                          * started above, every other time we shouldn't have a
6627                          * transaction started yet.
6628                          */
6629                         if (ret == -EAGAIN)
6630                                 goto end_trans;
6631                         err = ret;
6632                         break;
6633                 }
6634
6635                 if (!trans) {
6636                         /* Just need the 1 for updating the inode */
6637                         trans = btrfs_start_transaction(root, 1);
6638                         if (IS_ERR(trans)) {
6639                                 err = PTR_ERR(trans);
6640                                 goto out;
6641                         }
6642                 }
6643
6644                 trans->block_rsv = rsv;
6645
6646                 ret = btrfs_truncate_inode_items(trans, root, inode,
6647                                                  inode->i_size,
6648                                                  BTRFS_EXTENT_DATA_KEY);
6649                 if (ret != -EAGAIN) {
6650                         err = ret;
6651                         break;
6652                 }
6653
6654                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6655                 ret = btrfs_update_inode(trans, root, inode);
6656                 if (ret) {
6657                         err = ret;
6658                         break;
6659                 }
6660 end_trans:
6661                 nr = trans->blocks_used;
6662                 btrfs_end_transaction(trans, root);
6663                 trans = NULL;
6664                 btrfs_btree_balance_dirty(root, nr);
6665         }
6666
6667         if (ret == 0 && inode->i_nlink > 0) {
6668                 trans->block_rsv = root->orphan_block_rsv;
6669                 ret = btrfs_orphan_del(trans, inode);
6670                 if (ret)
6671                         err = ret;
6672         } else if (ret && inode->i_nlink > 0) {
6673                 /*
6674                  * Failed to do the truncate, remove us from the in memory
6675                  * orphan list.
6676                  */
6677                 ret = btrfs_orphan_del(NULL, inode);
6678         }
6679
6680         trans->block_rsv = &root->fs_info->trans_block_rsv;
6681         ret = btrfs_update_inode(trans, root, inode);
6682         if (ret && !err)
6683                 err = ret;
6684
6685         nr = trans->blocks_used;
6686         ret = btrfs_end_transaction_throttle(trans, root);
6687         btrfs_btree_balance_dirty(root, nr);
6688
6689 out:
6690         btrfs_free_block_rsv(root, rsv);
6691
6692         if (ret && !err)
6693                 err = ret;
6694
6695         return err;
6696 }
6697
6698 /*
6699  * create a new subvolume directory/inode (helper for the ioctl).
6700  */
6701 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
6702                              struct btrfs_root *new_root, u64 new_dirid)
6703 {
6704         struct inode *inode;
6705         int err;
6706         u64 index = 0;
6707
6708         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2, new_dirid,
6709                                 new_dirid, S_IFDIR | 0700, &index);
6710         if (IS_ERR(inode))
6711                 return PTR_ERR(inode);
6712         inode->i_op = &btrfs_dir_inode_operations;
6713         inode->i_fop = &btrfs_dir_file_operations;
6714
6715         inode->i_nlink = 1;
6716         btrfs_i_size_write(inode, 0);
6717
6718         err = btrfs_update_inode(trans, new_root, inode);
6719         BUG_ON(err);
6720
6721         iput(inode);
6722         return 0;
6723 }
6724
6725 struct inode *btrfs_alloc_inode(struct super_block *sb)
6726 {
6727         struct btrfs_inode *ei;
6728         struct inode *inode;
6729
6730         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
6731         if (!ei)
6732                 return NULL;
6733
6734         ei->root = NULL;
6735         ei->space_info = NULL;
6736         ei->generation = 0;
6737         ei->sequence = 0;
6738         ei->last_trans = 0;
6739         ei->last_sub_trans = 0;
6740         ei->logged_trans = 0;
6741         ei->delalloc_bytes = 0;
6742         ei->disk_i_size = 0;
6743         ei->flags = 0;
6744         ei->csum_bytes = 0;
6745         ei->index_cnt = (u64)-1;
6746         ei->last_unlink_trans = 0;
6747
6748         spin_lock_init(&ei->lock);
6749         ei->outstanding_extents = 0;
6750         ei->reserved_extents = 0;
6751
6752         ei->ordered_data_close = 0;
6753         ei->orphan_meta_reserved = 0;
6754         ei->dummy_inode = 0;
6755         ei->in_defrag = 0;
6756         ei->force_compress = BTRFS_COMPRESS_NONE;
6757
6758         ei->delayed_node = NULL;
6759
6760         inode = &ei->vfs_inode;
6761         extent_map_tree_init(&ei->extent_tree);
6762         extent_io_tree_init(&ei->io_tree, &inode->i_data);
6763         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
6764         mutex_init(&ei->log_mutex);
6765         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
6766         INIT_LIST_HEAD(&ei->i_orphan);
6767         INIT_LIST_HEAD(&ei->delalloc_inodes);
6768         INIT_LIST_HEAD(&ei->ordered_operations);
6769         RB_CLEAR_NODE(&ei->rb_node);
6770
6771         return inode;
6772 }
6773
6774 static void btrfs_i_callback(struct rcu_head *head)
6775 {
6776         struct inode *inode = container_of(head, struct inode, i_rcu);
6777         INIT_LIST_HEAD(&inode->i_dentry);
6778         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
6779 }
6780
6781 void btrfs_destroy_inode(struct inode *inode)
6782 {
6783         struct btrfs_ordered_extent *ordered;
6784         struct btrfs_root *root = BTRFS_I(inode)->root;
6785
6786         WARN_ON(!list_empty(&inode->i_dentry));
6787         WARN_ON(inode->i_data.nrpages);
6788         WARN_ON(BTRFS_I(inode)->outstanding_extents);
6789         WARN_ON(BTRFS_I(inode)->reserved_extents);
6790         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
6791         WARN_ON(BTRFS_I(inode)->csum_bytes);
6792
6793         /*
6794          * This can happen where we create an inode, but somebody else also
6795          * created the same inode and we need to destroy the one we already
6796          * created.
6797          */
6798         if (!root)
6799                 goto free;
6800
6801         /*
6802          * Make sure we're properly removed from the ordered operation
6803          * lists.
6804          */
6805         smp_mb();
6806         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
6807                 spin_lock(&root->fs_info->ordered_extent_lock);
6808                 list_del_init(&BTRFS_I(inode)->ordered_operations);
6809                 spin_unlock(&root->fs_info->ordered_extent_lock);
6810         }
6811
6812         spin_lock(&root->orphan_lock);
6813         if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
6814                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
6815                        (unsigned long long)btrfs_ino(inode));
6816                 list_del_init(&BTRFS_I(inode)->i_orphan);
6817         }
6818         spin_unlock(&root->orphan_lock);
6819
6820         while (1) {
6821                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
6822                 if (!ordered)
6823                         break;
6824                 else {
6825                         printk(KERN_ERR "btrfs found ordered "
6826                                "extent %llu %llu on inode cleanup\n",
6827                                (unsigned long long)ordered->file_offset,
6828                                (unsigned long long)ordered->len);
6829                         btrfs_remove_ordered_extent(inode, ordered);
6830                         btrfs_put_ordered_extent(ordered);
6831                         btrfs_put_ordered_extent(ordered);
6832                 }
6833         }
6834         inode_tree_del(inode);
6835         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
6836 free:
6837         btrfs_remove_delayed_node(inode);
6838         call_rcu(&inode->i_rcu, btrfs_i_callback);
6839 }
6840
6841 int btrfs_drop_inode(struct inode *inode)
6842 {
6843         struct btrfs_root *root = BTRFS_I(inode)->root;
6844
6845         if (btrfs_root_refs(&root->root_item) == 0 &&
6846             !btrfs_is_free_space_inode(root, inode))
6847                 return 1;
6848         else
6849                 return generic_drop_inode(inode);
6850 }
6851
6852 static void init_once(void *foo)
6853 {
6854         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
6855
6856         inode_init_once(&ei->vfs_inode);
6857 }
6858
6859 void btrfs_destroy_cachep(void)
6860 {
6861         if (btrfs_inode_cachep)
6862                 kmem_cache_destroy(btrfs_inode_cachep);
6863         if (btrfs_trans_handle_cachep)
6864                 kmem_cache_destroy(btrfs_trans_handle_cachep);
6865         if (btrfs_transaction_cachep)
6866                 kmem_cache_destroy(btrfs_transaction_cachep);
6867         if (btrfs_path_cachep)
6868                 kmem_cache_destroy(btrfs_path_cachep);
6869         if (btrfs_free_space_cachep)
6870                 kmem_cache_destroy(btrfs_free_space_cachep);
6871 }
6872
6873 int btrfs_init_cachep(void)
6874 {
6875         btrfs_inode_cachep = kmem_cache_create("btrfs_inode_cache",
6876                         sizeof(struct btrfs_inode), 0,
6877                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
6878         if (!btrfs_inode_cachep)
6879                 goto fail;
6880
6881         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle_cache",
6882                         sizeof(struct btrfs_trans_handle), 0,
6883                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6884         if (!btrfs_trans_handle_cachep)
6885                 goto fail;
6886
6887         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction_cache",
6888                         sizeof(struct btrfs_transaction), 0,
6889                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6890         if (!btrfs_transaction_cachep)
6891                 goto fail;
6892
6893         btrfs_path_cachep = kmem_cache_create("btrfs_path_cache",
6894                         sizeof(struct btrfs_path), 0,
6895                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6896         if (!btrfs_path_cachep)
6897                 goto fail;
6898
6899         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space_cache",
6900                         sizeof(struct btrfs_free_space), 0,
6901                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
6902         if (!btrfs_free_space_cachep)
6903                 goto fail;
6904
6905         return 0;
6906 fail:
6907         btrfs_destroy_cachep();
6908         return -ENOMEM;
6909 }
6910
6911 static int btrfs_getattr(struct vfsmount *mnt,
6912                          struct dentry *dentry, struct kstat *stat)
6913 {
6914         struct inode *inode = dentry->d_inode;
6915         generic_fillattr(inode, stat);
6916         stat->dev = BTRFS_I(inode)->root->anon_dev;
6917         stat->blksize = PAGE_CACHE_SIZE;
6918         stat->blocks = (inode_get_bytes(inode) +
6919                         BTRFS_I(inode)->delalloc_bytes) >> 9;
6920         return 0;
6921 }
6922
6923 /*
6924  * If a file is moved, it will inherit the cow and compression flags of the new
6925  * directory.
6926  */
6927 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
6928 {
6929         struct btrfs_inode *b_dir = BTRFS_I(dir);
6930         struct btrfs_inode *b_inode = BTRFS_I(inode);
6931
6932         if (b_dir->flags & BTRFS_INODE_NODATACOW)
6933                 b_inode->flags |= BTRFS_INODE_NODATACOW;
6934         else
6935                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
6936
6937         if (b_dir->flags & BTRFS_INODE_COMPRESS)
6938                 b_inode->flags |= BTRFS_INODE_COMPRESS;
6939         else
6940                 b_inode->flags &= ~BTRFS_INODE_COMPRESS;
6941 }
6942
6943 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
6944                            struct inode *new_dir, struct dentry *new_dentry)
6945 {
6946         struct btrfs_trans_handle *trans;
6947         struct btrfs_root *root = BTRFS_I(old_dir)->root;
6948         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
6949         struct inode *new_inode = new_dentry->d_inode;
6950         struct inode *old_inode = old_dentry->d_inode;
6951         struct timespec ctime = CURRENT_TIME;
6952         u64 index = 0;
6953         u64 root_objectid;
6954         int ret;
6955         u64 old_ino = btrfs_ino(old_inode);
6956
6957         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
6958                 return -EPERM;
6959
6960         /* we only allow rename subvolume link between subvolumes */
6961         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
6962                 return -EXDEV;
6963
6964         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
6965             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
6966                 return -ENOTEMPTY;
6967
6968         if (S_ISDIR(old_inode->i_mode) && new_inode &&
6969             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
6970                 return -ENOTEMPTY;
6971         /*
6972          * we're using rename to replace one file with another.
6973          * and the replacement file is large.  Start IO on it now so
6974          * we don't add too much work to the end of the transaction
6975          */
6976         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
6977             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
6978                 filemap_flush(old_inode->i_mapping);
6979
6980         /* close the racy window with snapshot create/destroy ioctl */
6981         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
6982                 down_read(&root->fs_info->subvol_sem);
6983         /*
6984          * We want to reserve the absolute worst case amount of items.  So if
6985          * both inodes are subvols and we need to unlink them then that would
6986          * require 4 item modifications, but if they are both normal inodes it
6987          * would require 5 item modifications, so we'll assume their normal
6988          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
6989          * should cover the worst case number of items we'll modify.
6990          */
6991         trans = btrfs_start_transaction(root, 20);
6992         if (IS_ERR(trans)) {
6993                 ret = PTR_ERR(trans);
6994                 goto out_notrans;
6995         }
6996
6997         if (dest != root)
6998                 btrfs_record_root_in_trans(trans, dest);
6999
7000         ret = btrfs_set_inode_index(new_dir, &index);
7001         if (ret)
7002                 goto out_fail;
7003
7004         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7005                 /* force full log commit if subvolume involved. */
7006                 root->fs_info->last_trans_log_full_commit = trans->transid;
7007         } else {
7008                 ret = btrfs_insert_inode_ref(trans, dest,
7009                                              new_dentry->d_name.name,
7010                                              new_dentry->d_name.len,
7011                                              old_ino,
7012                                              btrfs_ino(new_dir), index);
7013                 if (ret)
7014                         goto out_fail;
7015                 /*
7016                  * this is an ugly little race, but the rename is required
7017                  * to make sure that if we crash, the inode is either at the
7018                  * old name or the new one.  pinning the log transaction lets
7019                  * us make sure we don't allow a log commit to come in after
7020                  * we unlink the name but before we add the new name back in.
7021                  */
7022                 btrfs_pin_log_trans(root);
7023         }
7024         /*
7025          * make sure the inode gets flushed if it is replacing
7026          * something.
7027          */
7028         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7029                 btrfs_add_ordered_operation(trans, root, old_inode);
7030
7031         old_dir->i_ctime = old_dir->i_mtime = ctime;
7032         new_dir->i_ctime = new_dir->i_mtime = ctime;
7033         old_inode->i_ctime = ctime;
7034
7035         if (old_dentry->d_parent != new_dentry->d_parent)
7036                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7037
7038         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7039                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7040                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7041                                         old_dentry->d_name.name,
7042                                         old_dentry->d_name.len);
7043         } else {
7044                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7045                                         old_dentry->d_inode,
7046                                         old_dentry->d_name.name,
7047                                         old_dentry->d_name.len);
7048                 if (!ret)
7049                         ret = btrfs_update_inode(trans, root, old_inode);
7050         }
7051         BUG_ON(ret);
7052
7053         if (new_inode) {
7054                 new_inode->i_ctime = CURRENT_TIME;
7055                 if (unlikely(btrfs_ino(new_inode) ==
7056                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7057                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7058                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7059                                                 root_objectid,
7060                                                 new_dentry->d_name.name,
7061                                                 new_dentry->d_name.len);
7062                         BUG_ON(new_inode->i_nlink == 0);
7063                 } else {
7064                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7065                                                  new_dentry->d_inode,
7066                                                  new_dentry->d_name.name,
7067                                                  new_dentry->d_name.len);
7068                 }
7069                 BUG_ON(ret);
7070                 if (new_inode->i_nlink == 0) {
7071                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7072                         BUG_ON(ret);
7073                 }
7074         }
7075
7076         fixup_inode_flags(new_dir, old_inode);
7077
7078         ret = btrfs_add_link(trans, new_dir, old_inode,
7079                              new_dentry->d_name.name,
7080                              new_dentry->d_name.len, 0, index);
7081         BUG_ON(ret);
7082
7083         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7084                 struct dentry *parent = new_dentry->d_parent;
7085                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7086                 btrfs_end_log_trans(root);
7087         }
7088 out_fail:
7089         btrfs_end_transaction_throttle(trans, root);
7090 out_notrans:
7091         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7092                 up_read(&root->fs_info->subvol_sem);
7093
7094         return ret;
7095 }
7096
7097 /*
7098  * some fairly slow code that needs optimization. This walks the list
7099  * of all the inodes with pending delalloc and forces them to disk.
7100  */
7101 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7102 {
7103         struct list_head *head = &root->fs_info->delalloc_inodes;
7104         struct btrfs_inode *binode;
7105         struct inode *inode;
7106
7107         if (root->fs_info->sb->s_flags & MS_RDONLY)
7108                 return -EROFS;
7109
7110         spin_lock(&root->fs_info->delalloc_lock);
7111         while (!list_empty(head)) {
7112                 binode = list_entry(head->next, struct btrfs_inode,
7113                                     delalloc_inodes);
7114                 inode = igrab(&binode->vfs_inode);
7115                 if (!inode)
7116                         list_del_init(&binode->delalloc_inodes);
7117                 spin_unlock(&root->fs_info->delalloc_lock);
7118                 if (inode) {
7119                         filemap_flush(inode->i_mapping);
7120                         if (delay_iput)
7121                                 btrfs_add_delayed_iput(inode);
7122                         else
7123                                 iput(inode);
7124                 }
7125                 cond_resched();
7126                 spin_lock(&root->fs_info->delalloc_lock);
7127         }
7128         spin_unlock(&root->fs_info->delalloc_lock);
7129
7130         /* the filemap_flush will queue IO into the worker threads, but
7131          * we have to make sure the IO is actually started and that
7132          * ordered extents get created before we return
7133          */
7134         atomic_inc(&root->fs_info->async_submit_draining);
7135         while (atomic_read(&root->fs_info->nr_async_submits) ||
7136               atomic_read(&root->fs_info->async_delalloc_pages)) {
7137                 wait_event(root->fs_info->async_submit_wait,
7138                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7139                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7140         }
7141         atomic_dec(&root->fs_info->async_submit_draining);
7142         return 0;
7143 }
7144
7145 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7146                          const char *symname)
7147 {
7148         struct btrfs_trans_handle *trans;
7149         struct btrfs_root *root = BTRFS_I(dir)->root;
7150         struct btrfs_path *path;
7151         struct btrfs_key key;
7152         struct inode *inode = NULL;
7153         int err;
7154         int drop_inode = 0;
7155         u64 objectid;
7156         u64 index = 0 ;
7157         int name_len;
7158         int datasize;
7159         unsigned long ptr;
7160         struct btrfs_file_extent_item *ei;
7161         struct extent_buffer *leaf;
7162         unsigned long nr = 0;
7163
7164         name_len = strlen(symname) + 1;
7165         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7166                 return -ENAMETOOLONG;
7167
7168         /*
7169          * 2 items for inode item and ref
7170          * 2 items for dir items
7171          * 1 item for xattr if selinux is on
7172          */
7173         trans = btrfs_start_transaction(root, 5);
7174         if (IS_ERR(trans))
7175                 return PTR_ERR(trans);
7176
7177         err = btrfs_find_free_ino(root, &objectid);
7178         if (err)
7179                 goto out_unlock;
7180
7181         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7182                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7183                                 S_IFLNK|S_IRWXUGO, &index);
7184         if (IS_ERR(inode)) {
7185                 err = PTR_ERR(inode);
7186                 goto out_unlock;
7187         }
7188
7189         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7190         if (err) {
7191                 drop_inode = 1;
7192                 goto out_unlock;
7193         }
7194
7195         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7196         if (err)
7197                 drop_inode = 1;
7198         else {
7199                 inode->i_mapping->a_ops = &btrfs_aops;
7200                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7201                 inode->i_fop = &btrfs_file_operations;
7202                 inode->i_op = &btrfs_file_inode_operations;
7203                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7204         }
7205         if (drop_inode)
7206                 goto out_unlock;
7207
7208         path = btrfs_alloc_path();
7209         if (!path) {
7210                 err = -ENOMEM;
7211                 drop_inode = 1;
7212                 goto out_unlock;
7213         }
7214         key.objectid = btrfs_ino(inode);
7215         key.offset = 0;
7216         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7217         datasize = btrfs_file_extent_calc_inline_size(name_len);
7218         err = btrfs_insert_empty_item(trans, root, path, &key,
7219                                       datasize);
7220         if (err) {
7221                 drop_inode = 1;
7222                 btrfs_free_path(path);
7223                 goto out_unlock;
7224         }
7225         leaf = path->nodes[0];
7226         ei = btrfs_item_ptr(leaf, path->slots[0],
7227                             struct btrfs_file_extent_item);
7228         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7229         btrfs_set_file_extent_type(leaf, ei,
7230                                    BTRFS_FILE_EXTENT_INLINE);
7231         btrfs_set_file_extent_encryption(leaf, ei, 0);
7232         btrfs_set_file_extent_compression(leaf, ei, 0);
7233         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7234         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7235
7236         ptr = btrfs_file_extent_inline_start(ei);
7237         write_extent_buffer(leaf, symname, ptr, name_len);
7238         btrfs_mark_buffer_dirty(leaf);
7239         btrfs_free_path(path);
7240
7241         inode->i_op = &btrfs_symlink_inode_operations;
7242         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7243         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7244         inode_set_bytes(inode, name_len);
7245         btrfs_i_size_write(inode, name_len - 1);
7246         err = btrfs_update_inode(trans, root, inode);
7247         if (err)
7248                 drop_inode = 1;
7249
7250 out_unlock:
7251         nr = trans->blocks_used;
7252         btrfs_end_transaction_throttle(trans, root);
7253         if (drop_inode) {
7254                 inode_dec_link_count(inode);
7255                 iput(inode);
7256         }
7257         btrfs_btree_balance_dirty(root, nr);
7258         return err;
7259 }
7260
7261 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7262                                        u64 start, u64 num_bytes, u64 min_size,
7263                                        loff_t actual_len, u64 *alloc_hint,
7264                                        struct btrfs_trans_handle *trans)
7265 {
7266         struct btrfs_root *root = BTRFS_I(inode)->root;
7267         struct btrfs_key ins;
7268         u64 cur_offset = start;
7269         u64 i_size;
7270         int ret = 0;
7271         bool own_trans = true;
7272
7273         if (trans)
7274                 own_trans = false;
7275         while (num_bytes > 0) {
7276                 if (own_trans) {
7277                         trans = btrfs_start_transaction(root, 3);
7278                         if (IS_ERR(trans)) {
7279                                 ret = PTR_ERR(trans);
7280                                 break;
7281                         }
7282                 }
7283
7284                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7285                                            0, *alloc_hint, (u64)-1, &ins, 1);
7286                 if (ret) {
7287                         if (own_trans)
7288                                 btrfs_end_transaction(trans, root);
7289                         break;
7290                 }
7291
7292                 ret = insert_reserved_file_extent(trans, inode,
7293                                                   cur_offset, ins.objectid,
7294                                                   ins.offset, ins.offset,
7295                                                   ins.offset, 0, 0, 0,
7296                                                   BTRFS_FILE_EXTENT_PREALLOC);
7297                 BUG_ON(ret);
7298                 btrfs_drop_extent_cache(inode, cur_offset,
7299                                         cur_offset + ins.offset -1, 0);
7300
7301                 num_bytes -= ins.offset;
7302                 cur_offset += ins.offset;
7303                 *alloc_hint = ins.objectid + ins.offset;
7304
7305                 inode->i_ctime = CURRENT_TIME;
7306                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7307                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7308                     (actual_len > inode->i_size) &&
7309                     (cur_offset > inode->i_size)) {
7310                         if (cur_offset > actual_len)
7311                                 i_size = actual_len;
7312                         else
7313                                 i_size = cur_offset;
7314                         i_size_write(inode, i_size);
7315                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7316                 }
7317
7318                 ret = btrfs_update_inode(trans, root, inode);
7319                 BUG_ON(ret);
7320
7321                 if (own_trans)
7322                         btrfs_end_transaction(trans, root);
7323         }
7324         return ret;
7325 }
7326
7327 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7328                               u64 start, u64 num_bytes, u64 min_size,
7329                               loff_t actual_len, u64 *alloc_hint)
7330 {
7331         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7332                                            min_size, actual_len, alloc_hint,
7333                                            NULL);
7334 }
7335
7336 int btrfs_prealloc_file_range_trans(struct inode *inode,
7337                                     struct btrfs_trans_handle *trans, int mode,
7338                                     u64 start, u64 num_bytes, u64 min_size,
7339                                     loff_t actual_len, u64 *alloc_hint)
7340 {
7341         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7342                                            min_size, actual_len, alloc_hint, trans);
7343 }
7344
7345 static int btrfs_set_page_dirty(struct page *page)
7346 {
7347         return __set_page_dirty_nobuffers(page);
7348 }
7349
7350 static int btrfs_permission(struct inode *inode, int mask)
7351 {
7352         struct btrfs_root *root = BTRFS_I(inode)->root;
7353         umode_t mode = inode->i_mode;
7354
7355         if (mask & MAY_WRITE &&
7356             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7357                 if (btrfs_root_readonly(root))
7358                         return -EROFS;
7359                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7360                         return -EACCES;
7361         }
7362         return generic_permission(inode, mask);
7363 }
7364
7365 static const struct inode_operations btrfs_dir_inode_operations = {
7366         .getattr        = btrfs_getattr,
7367         .lookup         = btrfs_lookup,
7368         .create         = btrfs_create,
7369         .unlink         = btrfs_unlink,
7370         .link           = btrfs_link,
7371         .mkdir          = btrfs_mkdir,
7372         .rmdir          = btrfs_rmdir,
7373         .rename         = btrfs_rename,
7374         .symlink        = btrfs_symlink,
7375         .setattr        = btrfs_setattr,
7376         .mknod          = btrfs_mknod,
7377         .setxattr       = btrfs_setxattr,
7378         .getxattr       = btrfs_getxattr,
7379         .listxattr      = btrfs_listxattr,
7380         .removexattr    = btrfs_removexattr,
7381         .permission     = btrfs_permission,
7382         .get_acl        = btrfs_get_acl,
7383 };
7384 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7385         .lookup         = btrfs_lookup,
7386         .permission     = btrfs_permission,
7387         .get_acl        = btrfs_get_acl,
7388 };
7389
7390 static const struct file_operations btrfs_dir_file_operations = {
7391         .llseek         = generic_file_llseek,
7392         .read           = generic_read_dir,
7393         .readdir        = btrfs_real_readdir,
7394         .unlocked_ioctl = btrfs_ioctl,
7395 #ifdef CONFIG_COMPAT
7396         .compat_ioctl   = btrfs_ioctl,
7397 #endif
7398         .release        = btrfs_release_file,
7399         .fsync          = btrfs_sync_file,
7400 };
7401
7402 static struct extent_io_ops btrfs_extent_io_ops = {
7403         .fill_delalloc = run_delalloc_range,
7404         .submit_bio_hook = btrfs_submit_bio_hook,
7405         .merge_bio_hook = btrfs_merge_bio_hook,
7406         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7407         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7408         .writepage_start_hook = btrfs_writepage_start_hook,
7409         .readpage_io_failed_hook = btrfs_io_failed_hook,
7410         .set_bit_hook = btrfs_set_bit_hook,
7411         .clear_bit_hook = btrfs_clear_bit_hook,
7412         .merge_extent_hook = btrfs_merge_extent_hook,
7413         .split_extent_hook = btrfs_split_extent_hook,
7414 };
7415
7416 /*
7417  * btrfs doesn't support the bmap operation because swapfiles
7418  * use bmap to make a mapping of extents in the file.  They assume
7419  * these extents won't change over the life of the file and they
7420  * use the bmap result to do IO directly to the drive.
7421  *
7422  * the btrfs bmap call would return logical addresses that aren't
7423  * suitable for IO and they also will change frequently as COW
7424  * operations happen.  So, swapfile + btrfs == corruption.
7425  *
7426  * For now we're avoiding this by dropping bmap.
7427  */
7428 static const struct address_space_operations btrfs_aops = {
7429         .readpage       = btrfs_readpage,
7430         .writepage      = btrfs_writepage,
7431         .writepages     = btrfs_writepages,
7432         .readpages      = btrfs_readpages,
7433         .direct_IO      = btrfs_direct_IO,
7434         .invalidatepage = btrfs_invalidatepage,
7435         .releasepage    = btrfs_releasepage,
7436         .set_page_dirty = btrfs_set_page_dirty,
7437         .error_remove_page = generic_error_remove_page,
7438 };
7439
7440 static const struct address_space_operations btrfs_symlink_aops = {
7441         .readpage       = btrfs_readpage,
7442         .writepage      = btrfs_writepage,
7443         .invalidatepage = btrfs_invalidatepage,
7444         .releasepage    = btrfs_releasepage,
7445 };
7446
7447 static const struct inode_operations btrfs_file_inode_operations = {
7448         .getattr        = btrfs_getattr,
7449         .setattr        = btrfs_setattr,
7450         .setxattr       = btrfs_setxattr,
7451         .getxattr       = btrfs_getxattr,
7452         .listxattr      = btrfs_listxattr,
7453         .removexattr    = btrfs_removexattr,
7454         .permission     = btrfs_permission,
7455         .fiemap         = btrfs_fiemap,
7456         .get_acl        = btrfs_get_acl,
7457 };
7458 static const struct inode_operations btrfs_special_inode_operations = {
7459         .getattr        = btrfs_getattr,
7460         .setattr        = btrfs_setattr,
7461         .permission     = btrfs_permission,
7462         .setxattr       = btrfs_setxattr,
7463         .getxattr       = btrfs_getxattr,
7464         .listxattr      = btrfs_listxattr,
7465         .removexattr    = btrfs_removexattr,
7466         .get_acl        = btrfs_get_acl,
7467 };
7468 static const struct inode_operations btrfs_symlink_inode_operations = {
7469         .readlink       = generic_readlink,
7470         .follow_link    = page_follow_link_light,
7471         .put_link       = page_put_link,
7472         .getattr        = btrfs_getattr,
7473         .permission     = btrfs_permission,
7474         .setxattr       = btrfs_setxattr,
7475         .getxattr       = btrfs_getxattr,
7476         .listxattr      = btrfs_listxattr,
7477         .removexattr    = btrfs_removexattr,
7478         .get_acl        = btrfs_get_acl,
7479 };
7480
7481 const struct dentry_operations btrfs_dentry_operations = {
7482         .d_delete       = btrfs_dentry_delete,
7483         .d_release      = btrfs_dentry_release,
7484 };