fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include "compat.h"
  44 #include "ctree.h"
  45 #include "disk-io.h"
  46 #include "transaction.h"
  47 #include "btrfs_inode.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
  99                                            u64 len, u64 orig_start,
 100                                            u64 block_start, u64 block_len,
 101                                            u64 orig_block_len, int type);
 102
 103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 104                                      struct inode *inode,  struct inode *dir,
 105                                      const struct qstr *qstr)
 106 {
 107         int err;
 108
 109         err = btrfs_init_acl(trans, inode, dir);
 110         if (!err)
 111                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 112         return err;
 113 }
 114
 115 /*
 116  * this does all the hard work for inserting an inline extent into
 117  * the btree.  The caller should have done a btrfs_drop_extents so that
 118  * no overlapping inline items exist in the btree
 119  */
 120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 121                                 struct btrfs_root *root, struct inode *inode,
 122                                 u64 start, size_t size, size_t compressed_size,
 123                                 int compress_type,
 124                                 struct page **compressed_pages)
 125 {
 126         struct btrfs_key key;
 127         struct btrfs_path *path;
 128         struct extent_buffer *leaf;
 129         struct page *page = NULL;
 130         char *kaddr;
 131         unsigned long ptr;
 132         struct btrfs_file_extent_item *ei;
 133         int err = 0;
 134         int ret;
 135         size_t cur_size = size;
 136         size_t datasize;
 137         unsigned long offset;
 138
 139         if (compressed_size && compressed_pages)
 140                 cur_size = compressed_size;
 141
 142         path = btrfs_alloc_path();
 143         if (!path)
 144                 return -ENOMEM;
 145
 146         path->leave_spinning = 1;
 147
 148         key.objectid = btrfs_ino(inode);
 149         key.offset = start;
 150         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 151         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 152
 153         inode_add_bytes(inode, size);
 154         ret = btrfs_insert_empty_item(trans, root, path, &key,
 155                                       datasize);
 156         if (ret) {
 157                 err = ret;
 158                 goto fail;
 159         }
 160         leaf = path->nodes[0];
 161         ei = btrfs_item_ptr(leaf, path->slots[0],
 162                             struct btrfs_file_extent_item);
 163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 164         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 165         btrfs_set_file_extent_encryption(leaf, ei, 0);
 166         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 167         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 168         ptr = btrfs_file_extent_inline_start(ei);
 169
 170         if (compress_type != BTRFS_COMPRESS_NONE) {
 171                 struct page *cpage;
 172                 int i = 0;
 173                 while (compressed_size > 0) {
 174                         cpage = compressed_pages[i];
 175                         cur_size = min_t(unsigned long, compressed_size,
 176                                        PAGE_CACHE_SIZE);
 177
 178                         kaddr = kmap_atomic(cpage);
 179                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 180                         kunmap_atomic(kaddr);
 181
 182                         i++;
 183                         ptr += cur_size;
 184                         compressed_size -= cur_size;
 185                 }
 186                 btrfs_set_file_extent_compression(leaf, ei,
 187                                                   compress_type);
 188         } else {
 189                 page = find_get_page(inode->i_mapping,
 190                                      start >> PAGE_CACHE_SHIFT);
 191                 btrfs_set_file_extent_compression(leaf, ei, 0);
 192                 kaddr = kmap_atomic(page);
 193                 offset = start & (PAGE_CACHE_SIZE - 1);
 194                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 195                 kunmap_atomic(kaddr);
 196                 page_cache_release(page);
 197         }
 198         btrfs_mark_buffer_dirty(leaf);
 199         btrfs_free_path(path);
 200
 201         /*
 202          * we're an inline extent, so nobody can
 203          * extend the file past i_size without locking
 204          * a page we already have locked.
 205          *
 206          * We must do any isize and inode updates
 207          * before we unlock the pages.  Otherwise we
 208          * could end up racing with unlink.
 209          */
 210         BTRFS_I(inode)->disk_i_size = inode->i_size;
 211         ret = btrfs_update_inode(trans, root, inode);
 212
 213         return ret;
 214 fail:
 215         btrfs_free_path(path);
 216         return err;
 217 }
 218
 219
 220 /*
 221  * conditionally insert an inline extent into the file.  This
 222  * does the checks required to make sure the data is small enough
 223  * to fit as an inline extent.
 224  */
 225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 226                                  struct btrfs_root *root,
 227                                  struct inode *inode, u64 start, u64 end,
 228                                  size_t compressed_size, int compress_type,
 229                                  struct page **compressed_pages)
 230 {
 231         u64 isize = i_size_read(inode);
 232         u64 actual_end = min(end + 1, isize);
 233         u64 inline_len = actual_end - start;
 234         u64 aligned_end = (end + root->sectorsize - 1) &
 235                         ~((u64)root->sectorsize - 1);
 236         u64 data_len = inline_len;
 237         int ret;
 238
 239         if (compressed_size)
 240                 data_len = compressed_size;
 241
 242         if (start > 0 ||
 243             actual_end >= PAGE_CACHE_SIZE ||
 244             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 245             (!compressed_size &&
 246             (actual_end & (root->sectorsize - 1)) == 0) ||
 247             end + 1 < isize ||
 248             data_len > root->fs_info->max_inline) {
 249                 return 1;
 250         }
 251
 252         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 253         if (ret)
 254                 return ret;
 255
 256         if (isize > actual_end)
 257                 inline_len = min_t(u64, isize, actual_end);
 258         ret = insert_inline_extent(trans, root, inode, start,
 259                                    inline_len, compressed_size,
 260                                    compress_type, compressed_pages);
 261         if (ret && ret != -ENOSPC) {
 262                 btrfs_abort_transaction(trans, root, ret);
 263                 return ret;
 264         } else if (ret == -ENOSPC) {
 265                 return 1;
 266         }
 267
 268         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 269         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 270         return 0;
 271 }
 272
 273 struct async_extent {
 274         u64 start;
 275         u64 ram_size;
 276         u64 compressed_size;
 277         struct page **pages;
 278         unsigned long nr_pages;
 279         int compress_type;
 280         struct list_head list;
 281 };
 282
 283 struct async_cow {
 284         struct inode *inode;
 285         struct btrfs_root *root;
 286         struct page *locked_page;
 287         u64 start;
 288         u64 end;
 289         struct list_head extents;
 290         struct btrfs_work work;
 291 };
 292
 293 static noinline int add_async_extent(struct async_cow *cow,
 294                                      u64 start, u64 ram_size,
 295                                      u64 compressed_size,
 296                                      struct page **pages,
 297                                      unsigned long nr_pages,
 298                                      int compress_type)
 299 {
 300         struct async_extent *async_extent;
 301
 302         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 303         BUG_ON(!async_extent); /* -ENOMEM */
 304         async_extent->start = start;
 305         async_extent->ram_size = ram_size;
 306         async_extent->compressed_size = compressed_size;
 307         async_extent->pages = pages;
 308         async_extent->nr_pages = nr_pages;
 309         async_extent->compress_type = compress_type;
 310         list_add_tail(&async_extent->list, &cow->extents);
 311         return 0;
 312 }
 313
 314 /*
 315  * we create compressed extents in two phases.  The first
 316  * phase compresses a range of pages that have already been
 317  * locked (both pages and state bits are locked).
 318  *
 319  * This is done inside an ordered work queue, and the compression
 320  * is spread across many cpus.  The actual IO submission is step
 321  * two, and the ordered work queue takes care of making sure that
 322  * happens in the same order things were put onto the queue by
 323  * writepages and friends.
 324  *
 325  * If this code finds it can't get good compression, it puts an
 326  * entry onto the work queue to write the uncompressed bytes.  This
 327  * makes sure that both compressed inodes and uncompressed inodes
 328  * are written in the same order that the flusher thread sent them
 329  * down.
 330  */
 331 static noinline int compress_file_range(struct inode *inode,
 332                                         struct page *locked_page,
 333                                         u64 start, u64 end,
 334                                         struct async_cow *async_cow,
 335                                         int *num_added)
 336 {
 337         struct btrfs_root *root = BTRFS_I(inode)->root;
 338         struct btrfs_trans_handle *trans;
 339         u64 num_bytes;
 340         u64 blocksize = root->sectorsize;
 341         u64 actual_end;
 342         u64 isize = i_size_read(inode);
 343         int ret = 0;
 344         struct page **pages = NULL;
 345         unsigned long nr_pages;
 346         unsigned long nr_pages_ret = 0;
 347         unsigned long total_compressed = 0;
 348         unsigned long total_in = 0;
 349         unsigned long max_compressed = 128 * 1024;
 350         unsigned long max_uncompressed = 128 * 1024;
 351         int i;
 352         int will_compress;
 353         int compress_type = root->fs_info->compress_type;
 354
 355         /* if this is a small write inside eof, kick off a defrag */
 356         if ((end - start + 1) < 16 * 1024 &&
 357             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 358                 btrfs_add_inode_defrag(NULL, inode);
 359
 360         actual_end = min_t(u64, isize, end + 1);
 361 again:
 362         will_compress = 0;
 363         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 364         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 365
 366         /*
 367          * we don't want to send crud past the end of i_size through
 368          * compression, that's just a waste of CPU time.  So, if the
 369          * end of the file is before the start of our current
 370          * requested range of bytes, we bail out to the uncompressed
 371          * cleanup code that can deal with all of this.
 372          *
 373          * It isn't really the fastest way to fix things, but this is a
 374          * very uncommon corner.
 375          */
 376         if (actual_end <= start)
 377                 goto cleanup_and_bail_uncompressed;
 378
 379         total_compressed = actual_end - start;
 380
 381         /* we want to make sure that amount of ram required to uncompress
 382          * an extent is reasonable, so we limit the total size in ram
 383          * of a compressed extent to 128k.  This is a crucial number
 384          * because it also controls how easily we can spread reads across
 385          * cpus for decompression.
 386          *
 387          * We also want to make sure the amount of IO required to do
 388          * a random read is reasonably small, so we limit the size of
 389          * a compressed extent to 128k.
 390          */
 391         total_compressed = min(total_compressed, max_uncompressed);
 392         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 393         num_bytes = max(blocksize,  num_bytes);
 394         total_in = 0;
 395         ret = 0;
 396
 397         /*
 398          * we do compression for mount -o compress and when the
 399          * inode has not been flagged as nocompress.  This flag can
 400          * change at any time if we discover bad compression ratios.
 401          */
 402         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 403             (btrfs_test_opt(root, COMPRESS) ||
 404              (BTRFS_I(inode)->force_compress) ||
 405              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 406                 WARN_ON(pages);
 407                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 408                 if (!pages) {
 409                         /* just bail out to the uncompressed code */
 410                         goto cont;
 411                 }
 412
 413                 if (BTRFS_I(inode)->force_compress)
 414                         compress_type = BTRFS_I(inode)->force_compress;
 415
 416                 ret = btrfs_compress_pages(compress_type,
 417                                            inode->i_mapping, start,
 418                                            total_compressed, pages,
 419                                            nr_pages, &nr_pages_ret,
 420                                            &total_in,
 421                                            &total_compressed,
 422                                            max_compressed);
 423
 424                 if (!ret) {
 425                         unsigned long offset = total_compressed &
 426                                 (PAGE_CACHE_SIZE - 1);
 427                         struct page *page = pages[nr_pages_ret - 1];
 428                         char *kaddr;
 429
 430                         /* zero the tail end of the last page, we might be
 431                          * sending it down to disk
 432                          */
 433                         if (offset) {
 434                                 kaddr = kmap_atomic(page);
 435                                 memset(kaddr + offset, 0,
 436                                        PAGE_CACHE_SIZE - offset);
 437                                 kunmap_atomic(kaddr);
 438                         }
 439                         will_compress = 1;
 440                 }
 441         }
 442 cont:
 443         if (start == 0) {
 444                 trans = btrfs_join_transaction(root);
 445                 if (IS_ERR(trans)) {
 446                         ret = PTR_ERR(trans);
 447                         trans = NULL;
 448                         goto cleanup_and_out;
 449                 }
 450                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 451
 452                 /* lets try to make an inline extent */
 453                 if (ret || total_in < (actual_end - start)) {
 454                         /* we didn't compress the entire range, try
 455                          * to make an uncompressed inline extent.
 456                          */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end, 0, 0, NULL);
 459                 } else {
 460                         /* try making a compressed inline extent */
 461                         ret = cow_file_range_inline(trans, root, inode,
 462                                                     start, end,
 463                                                     total_compressed,
 464                                                     compress_type, pages);
 465                 }
 466                 if (ret <= 0) {
 467                         /*
 468                          * inline extent creation worked or returned error,
 469                          * we don't need to create any more async work items.
 470                          * Unlock and free up our temp pages.
 471                          */
 472                         extent_clear_unlock_delalloc(inode,
 473                              &BTRFS_I(inode)->io_tree,
 474                              start, end, NULL,
 475                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 476                              EXTENT_CLEAR_DELALLOC |
 477                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 478
 479                         btrfs_end_transaction(trans, root);
 480                         goto free_pages_out;
 481                 }
 482                 btrfs_end_transaction(trans, root);
 483         }
 484
 485         if (will_compress) {
 486                 /*
 487                  * we aren't doing an inline extent round the compressed size
 488                  * up to a block size boundary so the allocator does sane
 489                  * things
 490                  */
 491                 total_compressed = (total_compressed + blocksize - 1) &
 492                         ~(blocksize - 1);
 493
 494                 /*
 495                  * one last check to make sure the compression is really a
 496                  * win, compare the page count read with the blocks on disk
 497                  */
 498                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 499                         ~(PAGE_CACHE_SIZE - 1);
 500                 if (total_compressed >= total_in) {
 501                         will_compress = 0;
 502                 } else {
 503                         num_bytes = total_in;
 504                 }
 505         }
 506         if (!will_compress && pages) {
 507                 /*
 508                  * the compression code ran but failed to make things smaller,
 509                  * free any pages it allocated and our page pointer array
 510                  */
 511                 for (i = 0; i < nr_pages_ret; i++) {
 512                         WARN_ON(pages[i]->mapping);
 513                         page_cache_release(pages[i]);
 514                 }
 515                 kfree(pages);
 516                 pages = NULL;
 517                 total_compressed = 0;
 518                 nr_pages_ret = 0;
 519
 520                 /* flag the file so we don't compress in the future */
 521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 522                     !(BTRFS_I(inode)->force_compress)) {
 523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 524                 }
 525         }
 526         if (will_compress) {
 527                 *num_added += 1;
 528
 529                 /* the async work queues will take care of doing actual
 530                  * allocation on disk for these compressed pages,
 531                  * and will submit them to the elevator.
 532                  */
 533                 add_async_extent(async_cow, start, num_bytes,
 534                                  total_compressed, pages, nr_pages_ret,
 535                                  compress_type);
 536
 537                 if (start + num_bytes < end) {
 538                         start += num_bytes;
 539                         pages = NULL;
 540                         cond_resched();
 541                         goto again;
 542                 }
 543         } else {
 544 cleanup_and_bail_uncompressed:
 545                 /*
 546                  * No compression, but we still need to write the pages in
 547                  * the file we've been given so far.  redirty the locked
 548                  * page if it corresponds to our extent and set things up
 549                  * for the async work queue to run cow_file_range to do
 550                  * the normal delalloc dance
 551                  */
 552                 if (page_offset(locked_page) >= start &&
 553                     page_offset(locked_page) <= end) {
 554                         __set_page_dirty_nobuffers(locked_page);
 555                         /* unlocked later on in the async handlers */
 556                 }
 557                 add_async_extent(async_cow, start, end - start + 1,
 558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 559                 *num_added += 1;
 560         }
 561
 562 out:
 563         return ret;
 564
 565 free_pages_out:
 566         for (i = 0; i < nr_pages_ret; i++) {
 567                 WARN_ON(pages[i]->mapping);
 568                 page_cache_release(pages[i]);
 569         }
 570         kfree(pages);
 571
 572         goto out;
 573
 574 cleanup_and_out:
 575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 576                                      start, end, NULL,
 577                                      EXTENT_CLEAR_UNLOCK_PAGE |
 578                                      EXTENT_CLEAR_DIRTY |
 579                                      EXTENT_CLEAR_DELALLOC |
 580                                      EXTENT_SET_WRITEBACK |
 581                                      EXTENT_END_WRITEBACK);
 582         if (!trans || IS_ERR(trans))
 583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 584         else
 585                 btrfs_abort_transaction(trans, root, ret);
 586         goto free_pages_out;
 587 }
 588
 589 /*
 590  * phase two of compressed writeback.  This is the ordered portion
 591  * of the code, which only gets called in the order the work was
 592  * queued.  We walk all the async extents created by compress_file_range
 593  * and send them down to the disk.
 594  */
 595 static noinline int submit_compressed_extents(struct inode *inode,
 596                                               struct async_cow *async_cow)
 597 {
 598         struct async_extent *async_extent;
 599         u64 alloc_hint = 0;
 600         struct btrfs_trans_handle *trans;
 601         struct btrfs_key ins;
 602         struct extent_map *em;
 603         struct btrfs_root *root = BTRFS_I(inode)->root;
 604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 605         struct extent_io_tree *io_tree;
 606         int ret = 0;
 607
 608         if (list_empty(&async_cow->extents))
 609                 return 0;
 610
 611 again:
 612         while (!list_empty(&async_cow->extents)) {
 613                 async_extent = list_entry(async_cow->extents.next,
 614                                           struct async_extent, list);
 615                 list_del(&async_extent->list);
 616
 617                 io_tree = &BTRFS_I(inode)->io_tree;
 618
 619 retry:
 620                 /* did the compression code fall back to uncompressed IO? */
 621                 if (!async_extent->pages) {
 622                         int page_started = 0;
 623                         unsigned long nr_written = 0;
 624
 625                         lock_extent(io_tree, async_extent->start,
 626                                          async_extent->start +
 627                                          async_extent->ram_size - 1);
 628
 629                         /* allocate blocks */
 630                         ret = cow_file_range(inode, async_cow->locked_page,
 631                                              async_extent->start,
 632                                              async_extent->start +
 633                                              async_extent->ram_size - 1,
 634                                              &page_started, &nr_written, 0);
 635
 636                         /* JDM XXX */
 637
 638                         /*
 639                          * if page_started, cow_file_range inserted an
 640                          * inline extent and took care of all the unlocking
 641                          * and IO for us.  Otherwise, we need to submit
 642                          * all those pages down to the drive.
 643                          */
 644                         if (!page_started && !ret)
 645                                 extent_write_locked_range(io_tree,
 646                                                   inode, async_extent->start,
 647                                                   async_extent->start +
 648                                                   async_extent->ram_size - 1,
 649                                                   btrfs_get_extent,
 650                                                   WB_SYNC_ALL);
 651                         else if (ret)
 652                                 unlock_page(async_cow->locked_page);
 653                         kfree(async_extent);
 654                         cond_resched();
 655                         continue;
 656                 }
 657
 658                 lock_extent(io_tree, async_extent->start,
 659                             async_extent->start + async_extent->ram_size - 1);
 660
 661                 trans = btrfs_join_transaction(root);
 662                 if (IS_ERR(trans)) {
 663                         ret = PTR_ERR(trans);
 664                 } else {
 665                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 666                         ret = btrfs_reserve_extent(trans, root,
 667                                            async_extent->compressed_size,
 668                                            async_extent->compressed_size,
 669                                            0, alloc_hint, &ins, 1);
 670                         if (ret && ret != -ENOSPC)
 671                                 btrfs_abort_transaction(trans, root, ret);
 672                         btrfs_end_transaction(trans, root);
 673                 }
 674
 675                 if (ret) {
 676                         int i;
 677
 678                         for (i = 0; i < async_extent->nr_pages; i++) {
 679                                 WARN_ON(async_extent->pages[i]->mapping);
 680                                 page_cache_release(async_extent->pages[i]);
 681                         }
 682                         kfree(async_extent->pages);
 683                         async_extent->nr_pages = 0;
 684                         async_extent->pages = NULL;
 685
 686                         if (ret == -ENOSPC)
 687                                 goto retry;
 688                         goto out_free;
 689                 }
 690
 691                 /*
 692                  * here we're doing allocation and writeback of the
 693                  * compressed pages
 694                  */
 695                 btrfs_drop_extent_cache(inode, async_extent->start,
 696                                         async_extent->start +
 697                                         async_extent->ram_size - 1, 0);
 698
 699                 em = alloc_extent_map();
 700                 if (!em)
 701                         goto out_free_reserve;
 702                 em->start = async_extent->start;
 703                 em->len = async_extent->ram_size;
 704                 em->orig_start = em->start;
 705                 em->mod_start = em->start;
 706                 em->mod_len = em->len;
 707
 708                 em->block_start = ins.objectid;
 709                 em->block_len = ins.offset;
 710                 em->orig_block_len = ins.offset;
 711                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 712                 em->compress_type = async_extent->compress_type;
 713                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 714                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 715                 em->generation = -1;
 716
 717                 while (1) {
 718                         write_lock(&em_tree->lock);
 719                         ret = add_extent_mapping(em_tree, em);
 720                         if (!ret)
 721                                 list_move(&em->list,
 722                                           &em_tree->modified_extents);
 723                         write_unlock(&em_tree->lock);
 724                         if (ret != -EEXIST) {
 725                                 free_extent_map(em);
 726                                 break;
 727                         }
 728                         btrfs_drop_extent_cache(inode, async_extent->start,
 729                                                 async_extent->start +
 730                                                 async_extent->ram_size - 1, 0);
 731                 }
 732
 733                 if (ret)
 734                         goto out_free_reserve;
 735
 736                 ret = btrfs_add_ordered_extent_compress(inode,
 737                                                 async_extent->start,
 738                                                 ins.objectid,
 739                                                 async_extent->ram_size,
 740                                                 ins.offset,
 741                                                 BTRFS_ORDERED_COMPRESSED,
 742                                                 async_extent->compress_type);
 743                 if (ret)
 744                         goto out_free_reserve;
 745
 746                 /*
 747                  * clear dirty, set writeback and unlock the pages.
 748                  */
 749                 extent_clear_unlock_delalloc(inode,
 750                                 &BTRFS_I(inode)->io_tree,
 751                                 async_extent->start,
 752                                 async_extent->start +
 753                                 async_extent->ram_size - 1,
 754                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 755                                 EXTENT_CLEAR_UNLOCK |
 756                                 EXTENT_CLEAR_DELALLOC |
 757                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 758
 759                 ret = btrfs_submit_compressed_write(inode,
 760                                     async_extent->start,
 761                                     async_extent->ram_size,
 762                                     ins.objectid,
 763                                     ins.offset, async_extent->pages,
 764                                     async_extent->nr_pages);
 765                 alloc_hint = ins.objectid + ins.offset;
 766                 kfree(async_extent);
 767                 if (ret)
 768                         goto out;
 769                 cond_resched();
 770         }
 771         ret = 0;
 772 out:
 773         return ret;
 774 out_free_reserve:
 775         btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
 776 out_free:
 777         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 778                                      async_extent->start,
 779                                      async_extent->start +
 780                                      async_extent->ram_size - 1,
 781                                      NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 782                                      EXTENT_CLEAR_UNLOCK |
 783                                      EXTENT_CLEAR_DELALLOC |
 784                                      EXTENT_CLEAR_DIRTY |
 785                                      EXTENT_SET_WRITEBACK |
 786                                      EXTENT_END_WRITEBACK);
 787         kfree(async_extent);
 788         goto again;
 789 }
 790
 791 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 792                                       u64 num_bytes)
 793 {
 794         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 795         struct extent_map *em;
 796         u64 alloc_hint = 0;
 797
 798         read_lock(&em_tree->lock);
 799         em = search_extent_mapping(em_tree, start, num_bytes);
 800         if (em) {
 801                 /*
 802                  * if block start isn't an actual block number then find the
 803                  * first block in this inode and use that as a hint.  If that
 804                  * block is also bogus then just don't worry about it.
 805                  */
 806                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 807                         free_extent_map(em);
 808                         em = search_extent_mapping(em_tree, 0, 0);
 809                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 810                                 alloc_hint = em->block_start;
 811                         if (em)
 812                                 free_extent_map(em);
 813                 } else {
 814                         alloc_hint = em->block_start;
 815                         free_extent_map(em);
 816                 }
 817         }
 818         read_unlock(&em_tree->lock);
 819
 820         return alloc_hint;
 821 }
 822
 823 /*
 824  * when extent_io.c finds a delayed allocation range in the file,
 825  * the call backs end up in this code.  The basic idea is to
 826  * allocate extents on disk for the range, and create ordered data structs
 827  * in ram to track those extents.
 828  *
 829  * locked_page is the page that writepage had locked already.  We use
 830  * it to make sure we don't do extra locks or unlocks.
 831  *
 832  * *page_started is set to one if we unlock locked_page and do everything
 833  * required to start IO on it.  It may be clean and already done with
 834  * IO when we return.
 835  */
 836 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 837                                      struct inode *inode,
 838                                      struct btrfs_root *root,
 839                                      struct page *locked_page,
 840                                      u64 start, u64 end, int *page_started,
 841                                      unsigned long *nr_written,
 842                                      int unlock)
 843 {
 844         u64 alloc_hint = 0;
 845         u64 num_bytes;
 846         unsigned long ram_size;
 847         u64 disk_num_bytes;
 848         u64 cur_alloc_size;
 849         u64 blocksize = root->sectorsize;
 850         struct btrfs_key ins;
 851         struct extent_map *em;
 852         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 853         int ret = 0;
 854
 855         BUG_ON(btrfs_is_free_space_inode(inode));
 856
 857         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 858         num_bytes = max(blocksize,  num_bytes);
 859         disk_num_bytes = num_bytes;
 860
 861         /* if this is a small write inside eof, kick off defrag */
 862         if (num_bytes < 64 * 1024 &&
 863             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 864                 btrfs_add_inode_defrag(trans, inode);
 865
 866         if (start == 0) {
 867                 /* lets try to make an inline extent */
 868                 ret = cow_file_range_inline(trans, root, inode,
 869                                             start, end, 0, 0, NULL);
 870                 if (ret == 0) {
 871                         extent_clear_unlock_delalloc(inode,
 872                                      &BTRFS_I(inode)->io_tree,
 873                                      start, end, NULL,
 874                                      EXTENT_CLEAR_UNLOCK_PAGE |
 875                                      EXTENT_CLEAR_UNLOCK |
 876                                      EXTENT_CLEAR_DELALLOC |
 877                                      EXTENT_CLEAR_DIRTY |
 878                                      EXTENT_SET_WRITEBACK |
 879                                      EXTENT_END_WRITEBACK);
 880
 881                         *nr_written = *nr_written +
 882                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 883                         *page_started = 1;
 884                         goto out;
 885                 } else if (ret < 0) {
 886                         btrfs_abort_transaction(trans, root, ret);
 887                         goto out_unlock;
 888                 }
 889         }
 890
 891         BUG_ON(disk_num_bytes >
 892                btrfs_super_total_bytes(root->fs_info->super_copy));
 893
 894         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 895         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 896
 897         while (disk_num_bytes > 0) {
 898                 unsigned long op;
 899
 900                 cur_alloc_size = disk_num_bytes;
 901                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 902                                            root->sectorsize, 0, alloc_hint,
 903                                            &ins, 1);
 904                 if (ret < 0) {
 905                         btrfs_abort_transaction(trans, root, ret);
 906                         goto out_unlock;
 907                 }
 908
 909                 em = alloc_extent_map();
 910                 BUG_ON(!em); /* -ENOMEM */
 911                 em->start = start;
 912                 em->orig_start = em->start;
 913                 ram_size = ins.offset;
 914                 em->len = ins.offset;
 915                 em->mod_start = em->start;
 916                 em->mod_len = em->len;
 917
 918                 em->block_start = ins.objectid;
 919                 em->block_len = ins.offset;
 920                 em->orig_block_len = ins.offset;
 921                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 922                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 923                 em->generation = -1;
 924
 925                 while (1) {
 926                         write_lock(&em_tree->lock);
 927                         ret = add_extent_mapping(em_tree, em);
 928                         if (!ret)
 929                                 list_move(&em->list,
 930                                           &em_tree->modified_extents);
 931                         write_unlock(&em_tree->lock);
 932                         if (ret != -EEXIST) {
 933                                 free_extent_map(em);
 934                                 break;
 935                         }
 936                         btrfs_drop_extent_cache(inode, start,
 937                                                 start + ram_size - 1, 0);
 938                 }
 939
 940                 cur_alloc_size = ins.offset;
 941                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 942                                                ram_size, cur_alloc_size, 0);
 943                 BUG_ON(ret); /* -ENOMEM */
 944
 945                 if (root->root_key.objectid ==
 946                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 947                         ret = btrfs_reloc_clone_csums(inode, start,
 948                                                       cur_alloc_size);
 949                         if (ret) {
 950                                 btrfs_abort_transaction(trans, root, ret);
 951                                 goto out_unlock;
 952                         }
 953                 }
 954
 955                 if (disk_num_bytes < cur_alloc_size)
 956                         break;
 957
 958                 /* we're not doing compressed IO, don't unlock the first
 959                  * page (which the caller expects to stay locked), don't
 960                  * clear any dirty bits and don't set any writeback bits
 961                  *
 962                  * Do set the Private2 bit so we know this page was properly
 963                  * setup for writepage
 964                  */
 965                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 966                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 967                         EXTENT_SET_PRIVATE2;
 968
 969                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 970                                              start, start + ram_size - 1,
 971                                              locked_page, op);
 972                 disk_num_bytes -= cur_alloc_size;
 973                 num_bytes -= cur_alloc_size;
 974                 alloc_hint = ins.objectid + ins.offset;
 975                 start += cur_alloc_size;
 976         }
 977 out:
 978         return ret;
 979
 980 out_unlock:
 981         extent_clear_unlock_delalloc(inode,
 982                      &BTRFS_I(inode)->io_tree,
 983                      start, end, locked_page,
 984                      EXTENT_CLEAR_UNLOCK_PAGE |
 985                      EXTENT_CLEAR_UNLOCK |
 986                      EXTENT_CLEAR_DELALLOC |
 987                      EXTENT_CLEAR_DIRTY |
 988                      EXTENT_SET_WRITEBACK |
 989                      EXTENT_END_WRITEBACK);
 990
 991         goto out;
 992 }
 993
 994 static noinline int cow_file_range(struct inode *inode,
 995                                    struct page *locked_page,
 996                                    u64 start, u64 end, int *page_started,
 997                                    unsigned long *nr_written,
 998                                    int unlock)
 999 {
1000         struct btrfs_trans_handle *trans;
1001         struct btrfs_root *root = BTRFS_I(inode)->root;
1002         int ret;
1003
1004         trans = btrfs_join_transaction(root);
1005         if (IS_ERR(trans)) {
1006                 extent_clear_unlock_delalloc(inode,
1007                              &BTRFS_I(inode)->io_tree,
1008                              start, end, locked_page,
1009                              EXTENT_CLEAR_UNLOCK_PAGE |
1010                              EXTENT_CLEAR_UNLOCK |
1011                              EXTENT_CLEAR_DELALLOC |
1012                              EXTENT_CLEAR_DIRTY |
1013                              EXTENT_SET_WRITEBACK |
1014                              EXTENT_END_WRITEBACK);
1015                 return PTR_ERR(trans);
1016         }
1017         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1018
1019         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1020                                page_started, nr_written, unlock);
1021
1022         btrfs_end_transaction(trans, root);
1023
1024         return ret;
1025 }
1026
1027 /*
1028  * work queue call back to started compression on a file and pages
1029  */
1030 static noinline void async_cow_start(struct btrfs_work *work)
1031 {
1032         struct async_cow *async_cow;
1033         int num_added = 0;
1034         async_cow = container_of(work, struct async_cow, work);
1035
1036         compress_file_range(async_cow->inode, async_cow->locked_page,
1037                             async_cow->start, async_cow->end, async_cow,
1038                             &num_added);
1039         if (num_added == 0) {
1040                 btrfs_add_delayed_iput(async_cow->inode);
1041                 async_cow->inode = NULL;
1042         }
1043 }
1044
1045 /*
1046  * work queue call back to submit previously compressed pages
1047  */
1048 static noinline void async_cow_submit(struct btrfs_work *work)
1049 {
1050         struct async_cow *async_cow;
1051         struct btrfs_root *root;
1052         unsigned long nr_pages;
1053
1054         async_cow = container_of(work, struct async_cow, work);
1055
1056         root = async_cow->root;
1057         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1058                 PAGE_CACHE_SHIFT;
1059
1060         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1061             5 * 1024 * 1024 &&
1062             waitqueue_active(&root->fs_info->async_submit_wait))
1063                 wake_up(&root->fs_info->async_submit_wait);
1064
1065         if (async_cow->inode)
1066                 submit_compressed_extents(async_cow->inode, async_cow);
1067 }
1068
1069 static noinline void async_cow_free(struct btrfs_work *work)
1070 {
1071         struct async_cow *async_cow;
1072         async_cow = container_of(work, struct async_cow, work);
1073         if (async_cow->inode)
1074                 btrfs_add_delayed_iput(async_cow->inode);
1075         kfree(async_cow);
1076 }
1077
1078 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1079                                 u64 start, u64 end, int *page_started,
1080                                 unsigned long *nr_written)
1081 {
1082         struct async_cow *async_cow;
1083         struct btrfs_root *root = BTRFS_I(inode)->root;
1084         unsigned long nr_pages;
1085         u64 cur_end;
1086         int limit = 10 * 1024 * 1024;
1087
1088         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1089                          1, 0, NULL, GFP_NOFS);
1090         while (start < end) {
1091                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1092                 BUG_ON(!async_cow); /* -ENOMEM */
1093                 async_cow->inode = igrab(inode);
1094                 async_cow->root = root;
1095                 async_cow->locked_page = locked_page;
1096                 async_cow->start = start;
1097
1098                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1099                         cur_end = end;
1100                 else
1101                         cur_end = min(end, start + 512 * 1024 - 1);
1102
1103                 async_cow->end = cur_end;
1104                 INIT_LIST_HEAD(&async_cow->extents);
1105
1106                 async_cow->work.func = async_cow_start;
1107                 async_cow->work.ordered_func = async_cow_submit;
1108                 async_cow->work.ordered_free = async_cow_free;
1109                 async_cow->work.flags = 0;
1110
1111                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1112                         PAGE_CACHE_SHIFT;
1113                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1114
1115                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1116                                    &async_cow->work);
1117
1118                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1119                         wait_event(root->fs_info->async_submit_wait,
1120                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1121                             limit));
1122                 }
1123
1124                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1125                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1126                         wait_event(root->fs_info->async_submit_wait,
1127                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1128                            0));
1129                 }
1130
1131                 *nr_written += nr_pages;
1132                 start = cur_end + 1;
1133         }
1134         *page_started = 1;
1135         return 0;
1136 }
1137
1138 static noinline int csum_exist_in_range(struct btrfs_root *root,
1139                                         u64 bytenr, u64 num_bytes)
1140 {
1141         int ret;
1142         struct btrfs_ordered_sum *sums;
1143         LIST_HEAD(list);
1144
1145         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1146                                        bytenr + num_bytes - 1, &list, 0);
1147         if (ret == 0 && list_empty(&list))
1148                 return 0;
1149
1150         while (!list_empty(&list)) {
1151                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1152                 list_del(&sums->list);
1153                 kfree(sums);
1154         }
1155         return 1;
1156 }
1157
1158 /*
1159  * when nowcow writeback call back.  This checks for snapshots or COW copies
1160  * of the extents that exist in the file, and COWs the file as required.
1161  *
1162  * If no cow copies or snapshots exist, we write directly to the existing
1163  * blocks on disk
1164  */
1165 static noinline int run_delalloc_nocow(struct inode *inode,
1166                                        struct page *locked_page,
1167                               u64 start, u64 end, int *page_started, int force,
1168                               unsigned long *nr_written)
1169 {
1170         struct btrfs_root *root = BTRFS_I(inode)->root;
1171         struct btrfs_trans_handle *trans;
1172         struct extent_buffer *leaf;
1173         struct btrfs_path *path;
1174         struct btrfs_file_extent_item *fi;
1175         struct btrfs_key found_key;
1176         u64 cow_start;
1177         u64 cur_offset;
1178         u64 extent_end;
1179         u64 extent_offset;
1180         u64 disk_bytenr;
1181         u64 num_bytes;
1182         u64 disk_num_bytes;
1183         int extent_type;
1184         int ret, err;
1185         int type;
1186         int nocow;
1187         int check_prev = 1;
1188         bool nolock;
1189         u64 ino = btrfs_ino(inode);
1190
1191         path = btrfs_alloc_path();
1192         if (!path) {
1193                 extent_clear_unlock_delalloc(inode,
1194                              &BTRFS_I(inode)->io_tree,
1195                              start, end, locked_page,
1196                              EXTENT_CLEAR_UNLOCK_PAGE |
1197                              EXTENT_CLEAR_UNLOCK |
1198                              EXTENT_CLEAR_DELALLOC |
1199                              EXTENT_CLEAR_DIRTY |
1200                              EXTENT_SET_WRITEBACK |
1201                              EXTENT_END_WRITEBACK);
1202                 return -ENOMEM;
1203         }
1204
1205         nolock = btrfs_is_free_space_inode(inode);
1206
1207         if (nolock)
1208                 trans = btrfs_join_transaction_nolock(root);
1209         else
1210                 trans = btrfs_join_transaction(root);
1211
1212         if (IS_ERR(trans)) {
1213                 extent_clear_unlock_delalloc(inode,
1214                              &BTRFS_I(inode)->io_tree,
1215                              start, end, locked_page,
1216                              EXTENT_CLEAR_UNLOCK_PAGE |
1217                              EXTENT_CLEAR_UNLOCK |
1218                              EXTENT_CLEAR_DELALLOC |
1219                              EXTENT_CLEAR_DIRTY |
1220                              EXTENT_SET_WRITEBACK |
1221                              EXTENT_END_WRITEBACK);
1222                 btrfs_free_path(path);
1223                 return PTR_ERR(trans);
1224         }
1225
1226         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1227
1228         cow_start = (u64)-1;
1229         cur_offset = start;
1230         while (1) {
1231                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1232                                                cur_offset, 0);
1233                 if (ret < 0) {
1234                         btrfs_abort_transaction(trans, root, ret);
1235                         goto error;
1236                 }
1237                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1238                         leaf = path->nodes[0];
1239                         btrfs_item_key_to_cpu(leaf, &found_key,
1240                                               path->slots[0] - 1);
1241                         if (found_key.objectid == ino &&
1242                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1243                                 path->slots[0]--;
1244                 }
1245                 check_prev = 0;
1246 next_slot:
1247                 leaf = path->nodes[0];
1248                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1249                         ret = btrfs_next_leaf(root, path);
1250                         if (ret < 0) {
1251                                 btrfs_abort_transaction(trans, root, ret);
1252                                 goto error;
1253                         }
1254                         if (ret > 0)
1255                                 break;
1256                         leaf = path->nodes[0];
1257                 }
1258
1259                 nocow = 0;
1260                 disk_bytenr = 0;
1261                 num_bytes = 0;
1262                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1263
1264                 if (found_key.objectid > ino ||
1265                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1266                     found_key.offset > end)
1267                         break;
1268
1269                 if (found_key.offset > cur_offset) {
1270                         extent_end = found_key.offset;
1271                         extent_type = 0;
1272                         goto out_check;
1273                 }
1274
1275                 fi = btrfs_item_ptr(leaf, path->slots[0],
1276                                     struct btrfs_file_extent_item);
1277                 extent_type = btrfs_file_extent_type(leaf, fi);
1278
1279                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1280                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1281                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1282                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1283                         extent_end = found_key.offset +
1284                                 btrfs_file_extent_num_bytes(leaf, fi);
1285                         disk_num_bytes =
1286                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1287                         if (extent_end <= start) {
1288                                 path->slots[0]++;
1289                                 goto next_slot;
1290                         }
1291                         if (disk_bytenr == 0)
1292                                 goto out_check;
1293                         if (btrfs_file_extent_compression(leaf, fi) ||
1294                             btrfs_file_extent_encryption(leaf, fi) ||
1295                             btrfs_file_extent_other_encoding(leaf, fi))
1296                                 goto out_check;
1297                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1298                                 goto out_check;
1299                         if (btrfs_extent_readonly(root, disk_bytenr))
1300                                 goto out_check;
1301                         if (btrfs_cross_ref_exist(trans, root, ino,
1302                                                   found_key.offset -
1303                                                   extent_offset, disk_bytenr))
1304                                 goto out_check;
1305                         disk_bytenr += extent_offset;
1306                         disk_bytenr += cur_offset - found_key.offset;
1307                         num_bytes = min(end + 1, extent_end) - cur_offset;
1308                         /*
1309                          * force cow if csum exists in the range.
1310                          * this ensure that csum for a given extent are
1311                          * either valid or do not exist.
1312                          */
1313                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1314                                 goto out_check;
1315                         nocow = 1;
1316                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1317                         extent_end = found_key.offset +
1318                                 btrfs_file_extent_inline_len(leaf, fi);
1319                         extent_end = ALIGN(extent_end, root->sectorsize);
1320                 } else {
1321                         BUG_ON(1);
1322                 }
1323 out_check:
1324                 if (extent_end <= start) {
1325                         path->slots[0]++;
1326                         goto next_slot;
1327                 }
1328                 if (!nocow) {
1329                         if (cow_start == (u64)-1)
1330                                 cow_start = cur_offset;
1331                         cur_offset = extent_end;
1332                         if (cur_offset > end)
1333                                 break;
1334                         path->slots[0]++;
1335                         goto next_slot;
1336                 }
1337
1338                 btrfs_release_path(path);
1339                 if (cow_start != (u64)-1) {
1340                         ret = __cow_file_range(trans, inode, root, locked_page,
1341                                                cow_start, found_key.offset - 1,
1342                                                page_started, nr_written, 1);
1343                         if (ret) {
1344                                 btrfs_abort_transaction(trans, root, ret);
1345                                 goto error;
1346                         }
1347                         cow_start = (u64)-1;
1348                 }
1349
1350                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1351                         struct extent_map *em;
1352                         struct extent_map_tree *em_tree;
1353                         em_tree = &BTRFS_I(inode)->extent_tree;
1354                         em = alloc_extent_map();
1355                         BUG_ON(!em); /* -ENOMEM */
1356                         em->start = cur_offset;
1357                         em->orig_start = found_key.offset - extent_offset;
1358                         em->len = num_bytes;
1359                         em->block_len = num_bytes;
1360                         em->block_start = disk_bytenr;
1361                         em->orig_block_len = disk_num_bytes;
1362                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1363                         em->mod_start = em->start;
1364                         em->mod_len = em->len;
1365                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1366                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1367                         em->generation = -1;
1368                         while (1) {
1369                                 write_lock(&em_tree->lock);
1370                                 ret = add_extent_mapping(em_tree, em);
1371                                 if (!ret)
1372                                         list_move(&em->list,
1373                                                   &em_tree->modified_extents);
1374                                 write_unlock(&em_tree->lock);
1375                                 if (ret != -EEXIST) {
1376                                         free_extent_map(em);
1377                                         break;
1378                                 }
1379                                 btrfs_drop_extent_cache(inode, em->start,
1380                                                 em->start + em->len - 1, 0);
1381                         }
1382                         type = BTRFS_ORDERED_PREALLOC;
1383                 } else {
1384                         type = BTRFS_ORDERED_NOCOW;
1385                 }
1386
1387                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1388                                                num_bytes, num_bytes, type);
1389                 BUG_ON(ret); /* -ENOMEM */
1390
1391                 if (root->root_key.objectid ==
1392                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1393                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1394                                                       num_bytes);
1395                         if (ret) {
1396                                 btrfs_abort_transaction(trans, root, ret);
1397                                 goto error;
1398                         }
1399                 }
1400
1401                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1402                                 cur_offset, cur_offset + num_bytes - 1,
1403                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1404                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1405                                 EXTENT_SET_PRIVATE2);
1406                 cur_offset = extent_end;
1407                 if (cur_offset > end)
1408                         break;
1409         }
1410         btrfs_release_path(path);
1411
1412         if (cur_offset <= end && cow_start == (u64)-1) {
1413                 cow_start = cur_offset;
1414                 cur_offset = end;
1415         }
1416
1417         if (cow_start != (u64)-1) {
1418                 ret = __cow_file_range(trans, inode, root, locked_page,
1419                                        cow_start, end,
1420                                        page_started, nr_written, 1);
1421                 if (ret) {
1422                         btrfs_abort_transaction(trans, root, ret);
1423                         goto error;
1424                 }
1425         }
1426
1427 error:
1428         err = btrfs_end_transaction(trans, root);
1429         if (!ret)
1430                 ret = err;
1431
1432         if (ret && cur_offset < end)
1433                 extent_clear_unlock_delalloc(inode,
1434                              &BTRFS_I(inode)->io_tree,
1435                              cur_offset, end, locked_page,
1436                              EXTENT_CLEAR_UNLOCK_PAGE |
1437                              EXTENT_CLEAR_UNLOCK |
1438                              EXTENT_CLEAR_DELALLOC |
1439                              EXTENT_CLEAR_DIRTY |
1440                              EXTENT_SET_WRITEBACK |
1441                              EXTENT_END_WRITEBACK);
1442
1443         btrfs_free_path(path);
1444         return ret;
1445 }
1446
1447 /*
1448  * extent_io.c call back to do delayed allocation processing
1449  */
1450 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1451                               u64 start, u64 end, int *page_started,
1452                               unsigned long *nr_written)
1453 {
1454         int ret;
1455         struct btrfs_root *root = BTRFS_I(inode)->root;
1456
1457         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1458                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1459                                          page_started, 1, nr_written);
1460         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1461                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1462                                          page_started, 0, nr_written);
1463         } else if (!btrfs_test_opt(root, COMPRESS) &&
1464                    !(BTRFS_I(inode)->force_compress) &&
1465                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1466                 ret = cow_file_range(inode, locked_page, start, end,
1467                                       page_started, nr_written, 1);
1468         } else {
1469                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1470                         &BTRFS_I(inode)->runtime_flags);
1471                 ret = cow_file_range_async(inode, locked_page, start, end,
1472                                            page_started, nr_written);
1473         }
1474         return ret;
1475 }
1476
1477 static void btrfs_split_extent_hook(struct inode *inode,
1478                                     struct extent_state *orig, u64 split)
1479 {
1480         /* not delalloc, ignore it */
1481         if (!(orig->state & EXTENT_DELALLOC))
1482                 return;
1483
1484         spin_lock(&BTRFS_I(inode)->lock);
1485         BTRFS_I(inode)->outstanding_extents++;
1486         spin_unlock(&BTRFS_I(inode)->lock);
1487 }
1488
1489 /*
1490  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1491  * extents so we can keep track of new extents that are just merged onto old
1492  * extents, such as when we are doing sequential writes, so we can properly
1493  * account for the metadata space we'll need.
1494  */
1495 static void btrfs_merge_extent_hook(struct inode *inode,
1496                                     struct extent_state *new,
1497                                     struct extent_state *other)
1498 {
1499         /* not delalloc, ignore it */
1500         if (!(other->state & EXTENT_DELALLOC))
1501                 return;
1502
1503         spin_lock(&BTRFS_I(inode)->lock);
1504         BTRFS_I(inode)->outstanding_extents--;
1505         spin_unlock(&BTRFS_I(inode)->lock);
1506 }
1507
1508 /*
1509  * extent_io.c set_bit_hook, used to track delayed allocation
1510  * bytes in this file, and to maintain the list of inodes that
1511  * have pending delalloc work to be done.
1512  */
1513 static void btrfs_set_bit_hook(struct inode *inode,
1514                                struct extent_state *state, int *bits)
1515 {
1516
1517         /*
1518          * set_bit and clear bit hooks normally require _irqsave/restore
1519          * but in this case, we are only testing for the DELALLOC
1520          * bit, which is only set or cleared with irqs on
1521          */
1522         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1523                 struct btrfs_root *root = BTRFS_I(inode)->root;
1524                 u64 len = state->end + 1 - state->start;
1525                 bool do_list = !btrfs_is_free_space_inode(inode);
1526
1527                 if (*bits & EXTENT_FIRST_DELALLOC) {
1528                         *bits &= ~EXTENT_FIRST_DELALLOC;
1529                 } else {
1530                         spin_lock(&BTRFS_I(inode)->lock);
1531                         BTRFS_I(inode)->outstanding_extents++;
1532                         spin_unlock(&BTRFS_I(inode)->lock);
1533                 }
1534
1535                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1536                                      root->fs_info->delalloc_batch);
1537                 spin_lock(&BTRFS_I(inode)->lock);
1538                 BTRFS_I(inode)->delalloc_bytes += len;
1539                 if (do_list && !test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1540                                          &BTRFS_I(inode)->runtime_flags)) {
1541                         spin_lock(&root->fs_info->delalloc_lock);
1542                         if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1543                                 list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1544                                               &root->fs_info->delalloc_inodes);
1545                                 set_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1546                                         &BTRFS_I(inode)->runtime_flags);
1547                         }
1548                         spin_unlock(&root->fs_info->delalloc_lock);
1549                 }
1550                 spin_unlock(&BTRFS_I(inode)->lock);
1551         }
1552 }
1553
1554 /*
1555  * extent_io.c clear_bit_hook, see set_bit_hook for why
1556  */
1557 static void btrfs_clear_bit_hook(struct inode *inode,
1558                                  struct extent_state *state, int *bits)
1559 {
1560         /*
1561          * set_bit and clear bit hooks normally require _irqsave/restore
1562          * but in this case, we are only testing for the DELALLOC
1563          * bit, which is only set or cleared with irqs on
1564          */
1565         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1566                 struct btrfs_root *root = BTRFS_I(inode)->root;
1567                 u64 len = state->end + 1 - state->start;
1568                 bool do_list = !btrfs_is_free_space_inode(inode);
1569
1570                 if (*bits & EXTENT_FIRST_DELALLOC) {
1571                         *bits &= ~EXTENT_FIRST_DELALLOC;
1572                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1573                         spin_lock(&BTRFS_I(inode)->lock);
1574                         BTRFS_I(inode)->outstanding_extents--;
1575                         spin_unlock(&BTRFS_I(inode)->lock);
1576                 }
1577
1578                 if (*bits & EXTENT_DO_ACCOUNTING)
1579                         btrfs_delalloc_release_metadata(inode, len);
1580
1581                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1582                     && do_list)
1583                         btrfs_free_reserved_data_space(inode, len);
1584
1585                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1586                                      root->fs_info->delalloc_batch);
1587                 spin_lock(&BTRFS_I(inode)->lock);
1588                 BTRFS_I(inode)->delalloc_bytes -= len;
1589                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1590                     test_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1591                              &BTRFS_I(inode)->runtime_flags)) {
1592                         spin_lock(&root->fs_info->delalloc_lock);
1593                         if (!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1594                                 list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1595                                 clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
1596                                           &BTRFS_I(inode)->runtime_flags);
1597                         }
1598                         spin_unlock(&root->fs_info->delalloc_lock);
1599                 }
1600                 spin_unlock(&BTRFS_I(inode)->lock);
1601         }
1602 }
1603
1604 /*
1605  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1606  * we don't create bios that span stripes or chunks
1607  */
1608 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1609                          size_t size, struct bio *bio,
1610                          unsigned long bio_flags)
1611 {
1612         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1613         u64 logical = (u64)bio->bi_sector << 9;
1614         u64 length = 0;
1615         u64 map_length;
1616         int ret;
1617
1618         if (bio_flags & EXTENT_BIO_COMPRESSED)
1619                 return 0;
1620
1621         length = bio->bi_size;
1622         map_length = length;
1623         ret = btrfs_map_block(root->fs_info, READ, logical,
1624                               &map_length, NULL, 0);
1625         /* Will always return 0 with map_multi == NULL */
1626         BUG_ON(ret < 0);
1627         if (map_length < length + size)
1628                 return 1;
1629         return 0;
1630 }
1631
1632 /*
1633  * in order to insert checksums into the metadata in large chunks,
1634  * we wait until bio submission time.   All the pages in the bio are
1635  * checksummed and sums are attached onto the ordered extent record.
1636  *
1637  * At IO completion time the cums attached on the ordered extent record
1638  * are inserted into the btree
1639  */
1640 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1641                                     struct bio *bio, int mirror_num,
1642                                     unsigned long bio_flags,
1643                                     u64 bio_offset)
1644 {
1645         struct btrfs_root *root = BTRFS_I(inode)->root;
1646         int ret = 0;
1647
1648         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1649         BUG_ON(ret); /* -ENOMEM */
1650         return 0;
1651 }
1652
1653 /*
1654  * in order to insert checksums into the metadata in large chunks,
1655  * we wait until bio submission time.   All the pages in the bio are
1656  * checksummed and sums are attached onto the ordered extent record.
1657  *
1658  * At IO completion time the cums attached on the ordered extent record
1659  * are inserted into the btree
1660  */
1661 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1662                           int mirror_num, unsigned long bio_flags,
1663                           u64 bio_offset)
1664 {
1665         struct btrfs_root *root = BTRFS_I(inode)->root;
1666         int ret;
1667
1668         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1669         if (ret)
1670                 bio_endio(bio, ret);
1671         return ret;
1672 }
1673
1674 /*
1675  * extent_io.c submission hook. This does the right thing for csum calculation
1676  * on write, or reading the csums from the tree before a read
1677  */
1678 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1679                           int mirror_num, unsigned long bio_flags,
1680                           u64 bio_offset)
1681 {
1682         struct btrfs_root *root = BTRFS_I(inode)->root;
1683         int ret = 0;
1684         int skip_sum;
1685         int metadata = 0;
1686         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1687
1688         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1689
1690         if (btrfs_is_free_space_inode(inode))
1691                 metadata = 2;
1692
1693         if (!(rw & REQ_WRITE)) {
1694                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1695                 if (ret)
1696                         goto out;
1697
1698                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1699                         ret = btrfs_submit_compressed_read(inode, bio,
1700                                                            mirror_num,
1701                                                            bio_flags);
1702                         goto out;
1703                 } else if (!skip_sum) {
1704                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1705                         if (ret)
1706                                 goto out;
1707                 }
1708                 goto mapit;
1709         } else if (async && !skip_sum) {
1710                 /* csum items have already been cloned */
1711                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1712                         goto mapit;
1713                 /* we're doing a write, do the async checksumming */
1714                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1715                                    inode, rw, bio, mirror_num,
1716                                    bio_flags, bio_offset,
1717                                    __btrfs_submit_bio_start,
1718                                    __btrfs_submit_bio_done);
1719                 goto out;
1720         } else if (!skip_sum) {
1721                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1722                 if (ret)
1723                         goto out;
1724         }
1725
1726 mapit:
1727         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1728
1729 out:
1730         if (ret < 0)
1731                 bio_endio(bio, ret);
1732         return ret;
1733 }
1734
1735 /*
1736  * given a list of ordered sums record them in the inode.  This happens
1737  * at IO completion time based on sums calculated at bio submission time.
1738  */
1739 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1740                              struct inode *inode, u64 file_offset,
1741                              struct list_head *list)
1742 {
1743         struct btrfs_ordered_sum *sum;
1744
1745         list_for_each_entry(sum, list, list) {
1746                 btrfs_csum_file_blocks(trans,
1747                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1748         }
1749         return 0;
1750 }
1751
1752 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1753                               struct extent_state **cached_state)
1754 {
1755         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1756         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1757                                    cached_state, GFP_NOFS);
1758 }
1759
1760 /* see btrfs_writepage_start_hook for details on why this is required */
1761 struct btrfs_writepage_fixup {
1762         struct page *page;
1763         struct btrfs_work work;
1764 };
1765
1766 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1767 {
1768         struct btrfs_writepage_fixup *fixup;
1769         struct btrfs_ordered_extent *ordered;
1770         struct extent_state *cached_state = NULL;
1771         struct page *page;
1772         struct inode *inode;
1773         u64 page_start;
1774         u64 page_end;
1775         int ret;
1776
1777         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1778         page = fixup->page;
1779 again:
1780         lock_page(page);
1781         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1782                 ClearPageChecked(page);
1783                 goto out_page;
1784         }
1785
1786         inode = page->mapping->host;
1787         page_start = page_offset(page);
1788         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1789
1790         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1791                          &cached_state);
1792
1793         /* already ordered? We're done */
1794         if (PagePrivate2(page))
1795                 goto out;
1796
1797         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1798         if (ordered) {
1799                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1800                                      page_end, &cached_state, GFP_NOFS);
1801                 unlock_page(page);
1802                 btrfs_start_ordered_extent(inode, ordered, 1);
1803                 btrfs_put_ordered_extent(ordered);
1804                 goto again;
1805         }
1806
1807         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1808         if (ret) {
1809                 mapping_set_error(page->mapping, ret);
1810                 end_extent_writepage(page, ret, page_start, page_end);
1811                 ClearPageChecked(page);
1812                 goto out;
1813          }
1814
1815         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1816         ClearPageChecked(page);
1817         set_page_dirty(page);
1818 out:
1819         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1820                              &cached_state, GFP_NOFS);
1821 out_page:
1822         unlock_page(page);
1823         page_cache_release(page);
1824         kfree(fixup);
1825 }
1826
1827 /*
1828  * There are a few paths in the higher layers of the kernel that directly
1829  * set the page dirty bit without asking the filesystem if it is a
1830  * good idea.  This causes problems because we want to make sure COW
1831  * properly happens and the data=ordered rules are followed.
1832  *
1833  * In our case any range that doesn't have the ORDERED bit set
1834  * hasn't been properly setup for IO.  We kick off an async process
1835  * to fix it up.  The async helper will wait for ordered extents, set
1836  * the delalloc bit and make it safe to write the page.
1837  */
1838 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1839 {
1840         struct inode *inode = page->mapping->host;
1841         struct btrfs_writepage_fixup *fixup;
1842         struct btrfs_root *root = BTRFS_I(inode)->root;
1843
1844         /* this page is properly in the ordered list */
1845         if (TestClearPagePrivate2(page))
1846                 return 0;
1847
1848         if (PageChecked(page))
1849                 return -EAGAIN;
1850
1851         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1852         if (!fixup)
1853                 return -EAGAIN;
1854
1855         SetPageChecked(page);
1856         page_cache_get(page);
1857         fixup->work.func = btrfs_writepage_fixup_worker;
1858         fixup->page = page;
1859         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1860         return -EBUSY;
1861 }
1862
1863 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1864                                        struct inode *inode, u64 file_pos,
1865                                        u64 disk_bytenr, u64 disk_num_bytes,
1866                                        u64 num_bytes, u64 ram_bytes,
1867                                        u8 compression, u8 encryption,
1868                                        u16 other_encoding, int extent_type)
1869 {
1870         struct btrfs_root *root = BTRFS_I(inode)->root;
1871         struct btrfs_file_extent_item *fi;
1872         struct btrfs_path *path;
1873         struct extent_buffer *leaf;
1874         struct btrfs_key ins;
1875         int ret;
1876
1877         path = btrfs_alloc_path();
1878         if (!path)
1879                 return -ENOMEM;
1880
1881         path->leave_spinning = 1;
1882
1883         /*
1884          * we may be replacing one extent in the tree with another.
1885          * The new extent is pinned in the extent map, and we don't want
1886          * to drop it from the cache until it is completely in the btree.
1887          *
1888          * So, tell btrfs_drop_extents to leave this extent in the cache.
1889          * the caller is expected to unpin it and allow it to be merged
1890          * with the others.
1891          */
1892         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1893                                  file_pos + num_bytes, 0);
1894         if (ret)
1895                 goto out;
1896
1897         ins.objectid = btrfs_ino(inode);
1898         ins.offset = file_pos;
1899         ins.type = BTRFS_EXTENT_DATA_KEY;
1900         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1901         if (ret)
1902                 goto out;
1903         leaf = path->nodes[0];
1904         fi = btrfs_item_ptr(leaf, path->slots[0],
1905                             struct btrfs_file_extent_item);
1906         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1907         btrfs_set_file_extent_type(leaf, fi, extent_type);
1908         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1909         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1910         btrfs_set_file_extent_offset(leaf, fi, 0);
1911         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1912         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1913         btrfs_set_file_extent_compression(leaf, fi, compression);
1914         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1915         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1916
1917         btrfs_mark_buffer_dirty(leaf);
1918         btrfs_release_path(path);
1919
1920         inode_add_bytes(inode, num_bytes);
1921
1922         ins.objectid = disk_bytenr;
1923         ins.offset = disk_num_bytes;
1924         ins.type = BTRFS_EXTENT_ITEM_KEY;
1925         ret = btrfs_alloc_reserved_file_extent(trans, root,
1926                                         root->root_key.objectid,
1927                                         btrfs_ino(inode), file_pos, &ins);
1928 out:
1929         btrfs_free_path(path);
1930
1931         return ret;
1932 }
1933
1934 /*
1935  * helper function for btrfs_finish_ordered_io, this
1936  * just reads in some of the csum leaves to prime them into ram
1937  * before we start the transaction.  It limits the amount of btree
1938  * reads required while inside the transaction.
1939  */
1940 /* as ordered data IO finishes, this gets called so we can finish
1941  * an ordered extent if the range of bytes in the file it covers are
1942  * fully written.
1943  */
1944 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1945 {
1946         struct inode *inode = ordered_extent->inode;
1947         struct btrfs_root *root = BTRFS_I(inode)->root;
1948         struct btrfs_trans_handle *trans = NULL;
1949         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1950         struct extent_state *cached_state = NULL;
1951         int compress_type = 0;
1952         int ret;
1953         bool nolock;
1954
1955         nolock = btrfs_is_free_space_inode(inode);
1956
1957         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1958                 ret = -EIO;
1959                 goto out;
1960         }
1961
1962         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1963                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1964                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1965                 if (nolock)
1966                         trans = btrfs_join_transaction_nolock(root);
1967                 else
1968                         trans = btrfs_join_transaction(root);
1969                 if (IS_ERR(trans)) {
1970                         ret = PTR_ERR(trans);
1971                         trans = NULL;
1972                         goto out;
1973                 }
1974                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1975                 ret = btrfs_update_inode_fallback(trans, root, inode);
1976                 if (ret) /* -ENOMEM or corruption */
1977                         btrfs_abort_transaction(trans, root, ret);
1978                 goto out;
1979         }
1980
1981         lock_extent_bits(io_tree, ordered_extent->file_offset,
1982                          ordered_extent->file_offset + ordered_extent->len - 1,
1983                          0, &cached_state);
1984
1985         if (nolock)
1986                 trans = btrfs_join_transaction_nolock(root);
1987         else
1988                 trans = btrfs_join_transaction(root);
1989         if (IS_ERR(trans)) {
1990                 ret = PTR_ERR(trans);
1991                 trans = NULL;
1992                 goto out_unlock;
1993         }
1994         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1995
1996         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1997                 compress_type = ordered_extent->compress_type;
1998         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1999                 BUG_ON(compress_type);
2000                 ret = btrfs_mark_extent_written(trans, inode,
2001                                                 ordered_extent->file_offset,
2002                                                 ordered_extent->file_offset +
2003                                                 ordered_extent->len);
2004         } else {
2005                 BUG_ON(root == root->fs_info->tree_root);
2006                 ret = insert_reserved_file_extent(trans, inode,
2007                                                 ordered_extent->file_offset,
2008                                                 ordered_extent->start,
2009                                                 ordered_extent->disk_len,
2010                                                 ordered_extent->len,
2011                                                 ordered_extent->len,
2012                                                 compress_type, 0, 0,
2013                                                 BTRFS_FILE_EXTENT_REG);
2014         }
2015         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
2016                            ordered_extent->file_offset, ordered_extent->len,
2017                            trans->transid);
2018         if (ret < 0) {
2019                 btrfs_abort_transaction(trans, root, ret);
2020                 goto out_unlock;
2021         }
2022
2023         add_pending_csums(trans, inode, ordered_extent->file_offset,
2024                           &ordered_extent->list);
2025
2026         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
2027         ret = btrfs_update_inode_fallback(trans, root, inode);
2028         if (ret) { /* -ENOMEM or corruption */
2029                 btrfs_abort_transaction(trans, root, ret);
2030                 goto out_unlock;
2031         }
2032         ret = 0;
2033 out_unlock:
2034         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2035                              ordered_extent->file_offset +
2036                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2037 out:
2038         if (root != root->fs_info->tree_root)
2039                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2040         if (trans)
2041                 btrfs_end_transaction(trans, root);
2042
2043         if (ret) {
2044                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2045                                       ordered_extent->file_offset +
2046                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2047
2048                 /*
2049                  * If the ordered extent had an IOERR or something else went
2050                  * wrong we need to return the space for this ordered extent
2051                  * back to the allocator.
2052                  */
2053                 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags) &&
2054                     !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags))
2055                         btrfs_free_reserved_extent(root, ordered_extent->start,
2056                                                    ordered_extent->disk_len);
2057         }
2058
2059
2060         /*
2061          * This needs to be done to make sure anybody waiting knows we are done
2062          * updating everything for this ordered extent.
2063          */
2064         btrfs_remove_ordered_extent(inode, ordered_extent);
2065
2066         /* once for us */
2067         btrfs_put_ordered_extent(ordered_extent);
2068         /* once for the tree */
2069         btrfs_put_ordered_extent(ordered_extent);
2070
2071         return ret;
2072 }
2073
2074 static void finish_ordered_fn(struct btrfs_work *work)
2075 {
2076         struct btrfs_ordered_extent *ordered_extent;
2077         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2078         btrfs_finish_ordered_io(ordered_extent);
2079 }
2080
2081 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2082                                 struct extent_state *state, int uptodate)
2083 {
2084         struct inode *inode = page->mapping->host;
2085         struct btrfs_root *root = BTRFS_I(inode)->root;
2086         struct btrfs_ordered_extent *ordered_extent = NULL;
2087         struct btrfs_workers *workers;
2088
2089         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2090
2091         ClearPagePrivate2(page);
2092         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2093                                             end - start + 1, uptodate))
2094                 return 0;
2095
2096         ordered_extent->work.func = finish_ordered_fn;
2097         ordered_extent->work.flags = 0;
2098
2099         if (btrfs_is_free_space_inode(inode))
2100                 workers = &root->fs_info->endio_freespace_worker;
2101         else
2102                 workers = &root->fs_info->endio_write_workers;
2103         btrfs_queue_worker(workers, &ordered_extent->work);
2104
2105         return 0;
2106 }
2107
2108 /*
2109  * when reads are done, we need to check csums to verify the data is correct
2110  * if there's a match, we allow the bio to finish.  If not, the code in
2111  * extent_io.c will try to find good copies for us.
2112  */
2113 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2114                                struct extent_state *state, int mirror)
2115 {
2116         size_t offset = start - page_offset(page);
2117         struct inode *inode = page->mapping->host;
2118         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2119         char *kaddr;
2120         u64 private = ~(u32)0;
2121         int ret;
2122         struct btrfs_root *root = BTRFS_I(inode)->root;
2123         u32 csum = ~(u32)0;
2124
2125         if (PageChecked(page)) {
2126                 ClearPageChecked(page);
2127                 goto good;
2128         }
2129
2130         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2131                 goto good;
2132
2133         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2134             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2135                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2136                                   GFP_NOFS);
2137                 return 0;
2138         }
2139
2140         if (state && state->start == start) {
2141                 private = state->private;
2142                 ret = 0;
2143         } else {
2144                 ret = get_state_private(io_tree, start, &private);
2145         }
2146         kaddr = kmap_atomic(page);
2147         if (ret)
2148                 goto zeroit;
2149
2150         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2151         btrfs_csum_final(csum, (char *)&csum);
2152         if (csum != private)
2153                 goto zeroit;
2154
2155         kunmap_atomic(kaddr);
2156 good:
2157         return 0;
2158
2159 zeroit:
2160         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2161                        "private %llu\n",
2162                        (unsigned long long)btrfs_ino(page->mapping->host),
2163                        (unsigned long long)start, csum,
2164                        (unsigned long long)private);
2165         memset(kaddr + offset, 1, end - start + 1);
2166         flush_dcache_page(page);
2167         kunmap_atomic(kaddr);
2168         if (private == 0)
2169                 return 0;
2170         return -EIO;
2171 }
2172
2173 struct delayed_iput {
2174         struct list_head list;
2175         struct inode *inode;
2176 };
2177
2178 /* JDM: If this is fs-wide, why can't we add a pointer to
2179  * btrfs_inode instead and avoid the allocation? */
2180 void btrfs_add_delayed_iput(struct inode *inode)
2181 {
2182         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2183         struct delayed_iput *delayed;
2184
2185         if (atomic_add_unless(&inode->i_count, -1, 1))
2186                 return;
2187
2188         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2189         delayed->inode = inode;
2190
2191         spin_lock(&fs_info->delayed_iput_lock);
2192         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2193         spin_unlock(&fs_info->delayed_iput_lock);
2194 }
2195
2196 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2197 {
2198         LIST_HEAD(list);
2199         struct btrfs_fs_info *fs_info = root->fs_info;
2200         struct delayed_iput *delayed;
2201         int empty;
2202
2203         spin_lock(&fs_info->delayed_iput_lock);
2204         empty = list_empty(&fs_info->delayed_iputs);
2205         spin_unlock(&fs_info->delayed_iput_lock);
2206         if (empty)
2207                 return;
2208
2209         spin_lock(&fs_info->delayed_iput_lock);
2210         list_splice_init(&fs_info->delayed_iputs, &list);
2211         spin_unlock(&fs_info->delayed_iput_lock);
2212
2213         while (!list_empty(&list)) {
2214                 delayed = list_entry(list.next, struct delayed_iput, list);
2215                 list_del(&delayed->list);
2216                 iput(delayed->inode);
2217                 kfree(delayed);
2218         }
2219 }
2220
2221 /*
2222  * This is called in transaction commit time. If there are no orphan
2223  * files in the subvolume, it removes orphan item and frees block_rsv
2224  * structure.
2225  */
2226 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2227                               struct btrfs_root *root)
2228 {
2229         struct btrfs_block_rsv *block_rsv;
2230         int ret;
2231
2232         if (atomic_read(&root->orphan_inodes) ||
2233             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2234                 return;
2235
2236         spin_lock(&root->orphan_lock);
2237         if (atomic_read(&root->orphan_inodes)) {
2238                 spin_unlock(&root->orphan_lock);
2239                 return;
2240         }
2241
2242         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2243                 spin_unlock(&root->orphan_lock);
2244                 return;
2245         }
2246
2247         block_rsv = root->orphan_block_rsv;
2248         root->orphan_block_rsv = NULL;
2249         spin_unlock(&root->orphan_lock);
2250
2251         if (root->orphan_item_inserted &&
2252             btrfs_root_refs(&root->root_item) > 0) {
2253                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2254                                             root->root_key.objectid);
2255                 BUG_ON(ret);
2256                 root->orphan_item_inserted = 0;
2257         }
2258
2259         if (block_rsv) {
2260                 WARN_ON(block_rsv->size > 0);
2261                 btrfs_free_block_rsv(root, block_rsv);
2262         }
2263 }
2264
2265 /*
2266  * This creates an orphan entry for the given inode in case something goes
2267  * wrong in the middle of an unlink/truncate.
2268  *
2269  * NOTE: caller of this function should reserve 5 units of metadata for
2270  *       this function.
2271  */
2272 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2273 {
2274         struct btrfs_root *root = BTRFS_I(inode)->root;
2275         struct btrfs_block_rsv *block_rsv = NULL;
2276         int reserve = 0;
2277         int insert = 0;
2278         int ret;
2279
2280         if (!root->orphan_block_rsv) {
2281                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2282                 if (!block_rsv)
2283                         return -ENOMEM;
2284         }
2285
2286         spin_lock(&root->orphan_lock);
2287         if (!root->orphan_block_rsv) {
2288                 root->orphan_block_rsv = block_rsv;
2289         } else if (block_rsv) {
2290                 btrfs_free_block_rsv(root, block_rsv);
2291                 block_rsv = NULL;
2292         }
2293
2294         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2295                               &BTRFS_I(inode)->runtime_flags)) {
2296 #if 0
2297                 /*
2298                  * For proper ENOSPC handling, we should do orphan
2299                  * cleanup when mounting. But this introduces backward
2300                  * compatibility issue.
2301                  */
2302                 if (!xchg(&root->orphan_item_inserted, 1))
2303                         insert = 2;
2304                 else
2305                         insert = 1;
2306 #endif
2307                 insert = 1;
2308                 atomic_inc(&root->orphan_inodes);
2309         }
2310
2311         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2312                               &BTRFS_I(inode)->runtime_flags))
2313                 reserve = 1;
2314         spin_unlock(&root->orphan_lock);
2315
2316         /* grab metadata reservation from transaction handle */
2317         if (reserve) {
2318                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2319                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2320         }
2321
2322         /* insert an orphan item to track this unlinked/truncated file */
2323         if (insert >= 1) {
2324                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2325                 if (ret && ret != -EEXIST) {
2326                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2327                                   &BTRFS_I(inode)->runtime_flags);
2328                         btrfs_abort_transaction(trans, root, ret);
2329                         return ret;
2330                 }
2331                 ret = 0;
2332         }
2333
2334         /* insert an orphan item to track subvolume contains orphan files */
2335         if (insert >= 2) {
2336                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2337                                                root->root_key.objectid);
2338                 if (ret && ret != -EEXIST) {
2339                         btrfs_abort_transaction(trans, root, ret);
2340                         return ret;
2341                 }
2342         }
2343         return 0;
2344 }
2345
2346 /*
2347  * We have done the truncate/delete so we can go ahead and remove the orphan
2348  * item for this particular inode.
2349  */
2350 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2351 {
2352         struct btrfs_root *root = BTRFS_I(inode)->root;
2353         int delete_item = 0;
2354         int release_rsv = 0;
2355         int ret = 0;
2356
2357         spin_lock(&root->orphan_lock);
2358         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2359                                &BTRFS_I(inode)->runtime_flags))
2360                 delete_item = 1;
2361
2362         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2363                                &BTRFS_I(inode)->runtime_flags))
2364                 release_rsv = 1;
2365         spin_unlock(&root->orphan_lock);
2366
2367         if (trans && delete_item) {
2368                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2369                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2370         }
2371
2372         if (release_rsv) {
2373                 btrfs_orphan_release_metadata(inode);
2374                 atomic_dec(&root->orphan_inodes);
2375         }
2376
2377         return 0;
2378 }
2379
2380 /*
2381  * this cleans up any orphans that may be left on the list from the last use
2382  * of this root.
2383  */
2384 int btrfs_orphan_cleanup(struct btrfs_root *root)
2385 {
2386         struct btrfs_path *path;
2387         struct extent_buffer *leaf;
2388         struct btrfs_key key, found_key;
2389         struct btrfs_trans_handle *trans;
2390         struct inode *inode;
2391         u64 last_objectid = 0;
2392         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2393
2394         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2395                 return 0;
2396
2397         path = btrfs_alloc_path();
2398         if (!path) {
2399                 ret = -ENOMEM;
2400                 goto out;
2401         }
2402         path->reada = -1;
2403
2404         key.objectid = BTRFS_ORPHAN_OBJECTID;
2405         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2406         key.offset = (u64)-1;
2407
2408         while (1) {
2409                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2410                 if (ret < 0)
2411                         goto out;
2412
2413                 /*
2414                  * if ret == 0 means we found what we were searching for, which
2415                  * is weird, but possible, so only screw with path if we didn't
2416                  * find the key and see if we have stuff that matches
2417                  */
2418                 if (ret > 0) {
2419                         ret = 0;
2420                         if (path->slots[0] == 0)
2421                                 break;
2422                         path->slots[0]--;
2423                 }
2424
2425                 /* pull out the item */
2426                 leaf = path->nodes[0];
2427                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2428
2429                 /* make sure the item matches what we want */
2430                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2431                         break;
2432                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2433                         break;
2434
2435                 /* release the path since we're done with it */
2436                 btrfs_release_path(path);
2437
2438                 /*
2439                  * this is where we are basically btrfs_lookup, without the
2440                  * crossing root thing.  we store the inode number in the
2441                  * offset of the orphan item.
2442                  */
2443
2444                 if (found_key.offset == last_objectid) {
2445                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2446                                "stopping orphan cleanup\n");
2447                         ret = -EINVAL;
2448                         goto out;
2449                 }
2450
2451                 last_objectid = found_key.offset;
2452
2453                 found_key.objectid = found_key.offset;
2454                 found_key.type = BTRFS_INODE_ITEM_KEY;
2455                 found_key.offset = 0;
2456                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2457                 ret = PTR_RET(inode);
2458                 if (ret && ret != -ESTALE)
2459                         goto out;
2460
2461                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2462                         struct btrfs_root *dead_root;
2463                         struct btrfs_fs_info *fs_info = root->fs_info;
2464                         int is_dead_root = 0;
2465
2466                         /*
2467                          * this is an orphan in the tree root. Currently these
2468                          * could come from 2 sources:
2469                          *  a) a snapshot deletion in progress
2470                          *  b) a free space cache inode
2471                          * We need to distinguish those two, as the snapshot
2472                          * orphan must not get deleted.
2473                          * find_dead_roots already ran before us, so if this
2474                          * is a snapshot deletion, we should find the root
2475                          * in the dead_roots list
2476                          */
2477                         spin_lock(&fs_info->trans_lock);
2478                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2479                                             root_list) {
2480                                 if (dead_root->root_key.objectid ==
2481                                     found_key.objectid) {
2482                                         is_dead_root = 1;
2483                                         break;
2484                                 }
2485                         }
2486                         spin_unlock(&fs_info->trans_lock);
2487                         if (is_dead_root) {
2488                                 /* prevent this orphan from being found again */
2489                                 key.offset = found_key.objectid - 1;
2490                                 continue;
2491                         }
2492                 }
2493                 /*
2494                  * Inode is already gone but the orphan item is still there,
2495                  * kill the orphan item.
2496                  */
2497                 if (ret == -ESTALE) {
2498                         trans = btrfs_start_transaction(root, 1);
2499                         if (IS_ERR(trans)) {
2500                                 ret = PTR_ERR(trans);
2501                                 goto out;
2502                         }
2503                         printk(KERN_ERR "auto deleting %Lu\n",
2504                                found_key.objectid);
2505                         ret = btrfs_del_orphan_item(trans, root,
2506                                                     found_key.objectid);
2507                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2508                         btrfs_end_transaction(trans, root);
2509                         continue;
2510                 }
2511
2512                 /*
2513                  * add this inode to the orphan list so btrfs_orphan_del does
2514                  * the proper thing when we hit it
2515                  */
2516                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2517                         &BTRFS_I(inode)->runtime_flags);
2518                 atomic_inc(&root->orphan_inodes);
2519
2520                 /* if we have links, this was a truncate, lets do that */
2521                 if (inode->i_nlink) {
2522                         if (!S_ISREG(inode->i_mode)) {
2523                                 WARN_ON(1);
2524                                 iput(inode);
2525                                 continue;
2526                         }
2527                         nr_truncate++;
2528
2529                         /* 1 for the orphan item deletion. */
2530                         trans = btrfs_start_transaction(root, 1);
2531                         if (IS_ERR(trans)) {
2532                                 ret = PTR_ERR(trans);
2533                                 goto out;
2534                         }
2535                         ret = btrfs_orphan_add(trans, inode);
2536                         btrfs_end_transaction(trans, root);
2537                         if (ret)
2538                                 goto out;
2539
2540                         ret = btrfs_truncate(inode);
2541                         if (ret)
2542                                 btrfs_orphan_del(NULL, inode);
2543                 } else {
2544                         nr_unlink++;
2545                 }
2546
2547                 /* this will do delete_inode and everything for us */
2548                 iput(inode);
2549                 if (ret)
2550                         goto out;
2551         }
2552         /* release the path since we're done with it */
2553         btrfs_release_path(path);
2554
2555         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2556
2557         if (root->orphan_block_rsv)
2558                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2559                                         (u64)-1);
2560
2561         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2562                 trans = btrfs_join_transaction(root);
2563                 if (!IS_ERR(trans))
2564                         btrfs_end_transaction(trans, root);
2565         }
2566
2567         if (nr_unlink)
2568                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2569         if (nr_truncate)
2570                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2571
2572 out:
2573         if (ret)
2574                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2575         btrfs_free_path(path);
2576         return ret;
2577 }
2578
2579 /*
2580  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2581  * don't find any xattrs, we know there can't be any acls.
2582  *
2583  * slot is the slot the inode is in, objectid is the objectid of the inode
2584  */
2585 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2586                                           int slot, u64 objectid)
2587 {
2588         u32 nritems = btrfs_header_nritems(leaf);
2589         struct btrfs_key found_key;
2590         int scanned = 0;
2591
2592         slot++;
2593         while (slot < nritems) {
2594                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2595
2596                 /* we found a different objectid, there must not be acls */
2597                 if (found_key.objectid != objectid)
2598                         return 0;
2599
2600                 /* we found an xattr, assume we've got an acl */
2601                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2602                         return 1;
2603
2604                 /*
2605                  * we found a key greater than an xattr key, there can't
2606                  * be any acls later on
2607                  */
2608                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2609                         return 0;
2610
2611                 slot++;
2612                 scanned++;
2613
2614                 /*
2615                  * it goes inode, inode backrefs, xattrs, extents,
2616                  * so if there are a ton of hard links to an inode there can
2617                  * be a lot of backrefs.  Don't waste time searching too hard,
2618                  * this is just an optimization
2619                  */
2620                 if (scanned >= 8)
2621                         break;
2622         }
2623         /* we hit the end of the leaf before we found an xattr or
2624          * something larger than an xattr.  We have to assume the inode
2625          * has acls
2626          */
2627         return 1;
2628 }
2629
2630 /*
2631  * read an inode from the btree into the in-memory inode
2632  */
2633 static void btrfs_read_locked_inode(struct inode *inode)
2634 {
2635         struct btrfs_path *path;
2636         struct extent_buffer *leaf;
2637         struct btrfs_inode_item *inode_item;
2638         struct btrfs_timespec *tspec;
2639         struct btrfs_root *root = BTRFS_I(inode)->root;
2640         struct btrfs_key location;
2641         int maybe_acls;
2642         u32 rdev;
2643         int ret;
2644         bool filled = false;
2645
2646         ret = btrfs_fill_inode(inode, &rdev);
2647         if (!ret)
2648                 filled = true;
2649
2650         path = btrfs_alloc_path();
2651         if (!path)
2652                 goto make_bad;
2653
2654         path->leave_spinning = 1;
2655         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2656
2657         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2658         if (ret)
2659                 goto make_bad;
2660
2661         leaf = path->nodes[0];
2662
2663         if (filled)
2664                 goto cache_acl;
2665
2666         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2667                                     struct btrfs_inode_item);
2668         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2669         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2670         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2671         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2672         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2673
2674         tspec = btrfs_inode_atime(inode_item);
2675         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2676         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2677
2678         tspec = btrfs_inode_mtime(inode_item);
2679         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2680         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2681
2682         tspec = btrfs_inode_ctime(inode_item);
2683         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2684         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2685
2686         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2687         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2688         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2689
2690         /*
2691          * If we were modified in the current generation and evicted from memory
2692          * and then re-read we need to do a full sync since we don't have any
2693          * idea about which extents were modified before we were evicted from
2694          * cache.
2695          */
2696         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2697                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2698                         &BTRFS_I(inode)->runtime_flags);
2699
2700         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2701         inode->i_generation = BTRFS_I(inode)->generation;
2702         inode->i_rdev = 0;
2703         rdev = btrfs_inode_rdev(leaf, inode_item);
2704
2705         BTRFS_I(inode)->index_cnt = (u64)-1;
2706         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2707 cache_acl:
2708         /*
2709          * try to precache a NULL acl entry for files that don't have
2710          * any xattrs or acls
2711          */
2712         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2713                                            btrfs_ino(inode));
2714         if (!maybe_acls)
2715                 cache_no_acl(inode);
2716
2717         btrfs_free_path(path);
2718
2719         switch (inode->i_mode & S_IFMT) {
2720         case S_IFREG:
2721                 inode->i_mapping->a_ops = &btrfs_aops;
2722                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2723                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2724                 inode->i_fop = &btrfs_file_operations;
2725                 inode->i_op = &btrfs_file_inode_operations;
2726                 break;
2727         case S_IFDIR:
2728                 inode->i_fop = &btrfs_dir_file_operations;
2729                 if (root == root->fs_info->tree_root)
2730                         inode->i_op = &btrfs_dir_ro_inode_operations;
2731                 else
2732                         inode->i_op = &btrfs_dir_inode_operations;
2733                 break;
2734         case S_IFLNK:
2735                 inode->i_op = &btrfs_symlink_inode_operations;
2736                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2737                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2738                 break;
2739         default:
2740                 inode->i_op = &btrfs_special_inode_operations;
2741                 init_special_inode(inode, inode->i_mode, rdev);
2742                 break;
2743         }
2744
2745         btrfs_update_iflags(inode);
2746         return;
2747
2748 make_bad:
2749         btrfs_free_path(path);
2750         make_bad_inode(inode);
2751 }
2752
2753 /*
2754  * given a leaf and an inode, copy the inode fields into the leaf
2755  */
2756 static void fill_inode_item(struct btrfs_trans_handle *trans,
2757                             struct extent_buffer *leaf,
2758                             struct btrfs_inode_item *item,
2759                             struct inode *inode)
2760 {
2761         struct btrfs_map_token token;
2762
2763         btrfs_init_map_token(&token);
2764
2765         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2766         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2767         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
2768                                    &token);
2769         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2770         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2771
2772         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2773                                      inode->i_atime.tv_sec, &token);
2774         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2775                                       inode->i_atime.tv_nsec, &token);
2776
2777         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2778                                      inode->i_mtime.tv_sec, &token);
2779         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2780                                       inode->i_mtime.tv_nsec, &token);
2781
2782         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2783                                      inode->i_ctime.tv_sec, &token);
2784         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2785                                       inode->i_ctime.tv_nsec, &token);
2786
2787         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2788                                      &token);
2789         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
2790                                          &token);
2791         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2792         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2793         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
2794         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
2795         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
2796 }
2797
2798 /*
2799  * copy everything in the in-memory inode into the btree.
2800  */
2801 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2802                                 struct btrfs_root *root, struct inode *inode)
2803 {
2804         struct btrfs_inode_item *inode_item;
2805         struct btrfs_path *path;
2806         struct extent_buffer *leaf;
2807         int ret;
2808
2809         path = btrfs_alloc_path();
2810         if (!path)
2811                 return -ENOMEM;
2812
2813         path->leave_spinning = 1;
2814         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2815                                  1);
2816         if (ret) {
2817                 if (ret > 0)
2818                         ret = -ENOENT;
2819                 goto failed;
2820         }
2821
2822         btrfs_unlock_up_safe(path, 1);
2823         leaf = path->nodes[0];
2824         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2825                                     struct btrfs_inode_item);
2826
2827         fill_inode_item(trans, leaf, inode_item, inode);
2828         btrfs_mark_buffer_dirty(leaf);
2829         btrfs_set_inode_last_trans(trans, inode);
2830         ret = 0;
2831 failed:
2832         btrfs_free_path(path);
2833         return ret;
2834 }
2835
2836 /*
2837  * copy everything in the in-memory inode into the btree.
2838  */
2839 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2840                                 struct btrfs_root *root, struct inode *inode)
2841 {
2842         int ret;
2843
2844         /*
2845          * If the inode is a free space inode, we can deadlock during commit
2846          * if we put it into the delayed code.
2847          *
2848          * The data relocation inode should also be directly updated
2849          * without delay
2850          */
2851         if (!btrfs_is_free_space_inode(inode)
2852             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2853                 btrfs_update_root_times(trans, root);
2854
2855                 ret = btrfs_delayed_update_inode(trans, root, inode);
2856                 if (!ret)
2857                         btrfs_set_inode_last_trans(trans, inode);
2858                 return ret;
2859         }
2860
2861         return btrfs_update_inode_item(trans, root, inode);
2862 }
2863
2864 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2865                                          struct btrfs_root *root,
2866                                          struct inode *inode)
2867 {
2868         int ret;
2869
2870         ret = btrfs_update_inode(trans, root, inode);
2871         if (ret == -ENOSPC)
2872                 return btrfs_update_inode_item(trans, root, inode);
2873         return ret;
2874 }
2875
2876 /*
2877  * unlink helper that gets used here in inode.c and in the tree logging
2878  * recovery code.  It remove a link in a directory with a given name, and
2879  * also drops the back refs in the inode to the directory
2880  */
2881 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2882                                 struct btrfs_root *root,
2883                                 struct inode *dir, struct inode *inode,
2884                                 const char *name, int name_len)
2885 {
2886         struct btrfs_path *path;
2887         int ret = 0;
2888         struct extent_buffer *leaf;
2889         struct btrfs_dir_item *di;
2890         struct btrfs_key key;
2891         u64 index;
2892         u64 ino = btrfs_ino(inode);
2893         u64 dir_ino = btrfs_ino(dir);
2894
2895         path = btrfs_alloc_path();
2896         if (!path) {
2897                 ret = -ENOMEM;
2898                 goto out;
2899         }
2900
2901         path->leave_spinning = 1;
2902         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2903                                     name, name_len, -1);
2904         if (IS_ERR(di)) {
2905                 ret = PTR_ERR(di);
2906                 goto err;
2907         }
2908         if (!di) {
2909                 ret = -ENOENT;
2910                 goto err;
2911         }
2912         leaf = path->nodes[0];
2913         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2914         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2915         if (ret)
2916                 goto err;
2917         btrfs_release_path(path);
2918
2919         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2920                                   dir_ino, &index);
2921         if (ret) {
2922                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2923                        "inode %llu parent %llu\n", name_len, name,
2924                        (unsigned long long)ino, (unsigned long long)dir_ino);
2925                 btrfs_abort_transaction(trans, root, ret);
2926                 goto err;
2927         }
2928
2929         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2930         if (ret) {
2931                 btrfs_abort_transaction(trans, root, ret);
2932                 goto err;
2933         }
2934
2935         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2936                                          inode, dir_ino);
2937         if (ret != 0 && ret != -ENOENT) {
2938                 btrfs_abort_transaction(trans, root, ret);
2939                 goto err;
2940         }
2941
2942         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2943                                            dir, index);
2944         if (ret == -ENOENT)
2945                 ret = 0;
2946 err:
2947         btrfs_free_path(path);
2948         if (ret)
2949                 goto out;
2950
2951         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2952         inode_inc_iversion(inode);
2953         inode_inc_iversion(dir);
2954         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2955         ret = btrfs_update_inode(trans, root, dir);
2956 out:
2957         return ret;
2958 }
2959
2960 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2961                        struct btrfs_root *root,
2962                        struct inode *dir, struct inode *inode,
2963                        const char *name, int name_len)
2964 {
2965         int ret;
2966         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2967         if (!ret) {
2968                 btrfs_drop_nlink(inode);
2969                 ret = btrfs_update_inode(trans, root, inode);
2970         }
2971         return ret;
2972 }
2973
2974
2975 /* helper to check if there is any shared block in the path */
2976 static int check_path_shared(struct btrfs_root *root,
2977                              struct btrfs_path *path)
2978 {
2979         struct extent_buffer *eb;
2980         int level;
2981         u64 refs = 1;
2982
2983         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2984                 int ret;
2985
2986                 if (!path->nodes[level])
2987                         break;
2988                 eb = path->nodes[level];
2989                 if (!btrfs_block_can_be_shared(root, eb))
2990                         continue;
2991                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2992                                                &refs, NULL);
2993                 if (refs > 1)
2994                         return 1;
2995         }
2996         return 0;
2997 }
2998
2999 /*
3000  * helper to start transaction for unlink and rmdir.
3001  *
3002  * unlink and rmdir are special in btrfs, they do not always free space.
3003  * so in enospc case, we should make sure they will free space before
3004  * allowing them to use the global metadata reservation.
3005  */
3006 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
3007                                                        struct dentry *dentry)
3008 {
3009         struct btrfs_trans_handle *trans;
3010         struct btrfs_root *root = BTRFS_I(dir)->root;
3011         struct btrfs_path *path;
3012         struct btrfs_dir_item *di;
3013         struct inode *inode = dentry->d_inode;
3014         u64 index;
3015         int check_link = 1;
3016         int err = -ENOSPC;
3017         int ret;
3018         u64 ino = btrfs_ino(inode);
3019         u64 dir_ino = btrfs_ino(dir);
3020
3021         /*
3022          * 1 for the possible orphan item
3023          * 1 for the dir item
3024          * 1 for the dir index
3025          * 1 for the inode ref
3026          * 1 for the inode ref in the tree log
3027          * 2 for the dir entries in the log
3028          * 1 for the inode
3029          */
3030         trans = btrfs_start_transaction(root, 8);
3031         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
3032                 return trans;
3033
3034         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
3035                 return ERR_PTR(-ENOSPC);
3036
3037         /* check if there is someone else holds reference */
3038         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
3039                 return ERR_PTR(-ENOSPC);
3040
3041         if (atomic_read(&inode->i_count) > 2)
3042                 return ERR_PTR(-ENOSPC);
3043
3044         if (xchg(&root->fs_info->enospc_unlink, 1))
3045                 return ERR_PTR(-ENOSPC);
3046
3047         path = btrfs_alloc_path();
3048         if (!path) {
3049                 root->fs_info->enospc_unlink = 0;
3050                 return ERR_PTR(-ENOMEM);
3051         }
3052
3053         /* 1 for the orphan item */
3054         trans = btrfs_start_transaction(root, 1);
3055         if (IS_ERR(trans)) {
3056                 btrfs_free_path(path);
3057                 root->fs_info->enospc_unlink = 0;
3058                 return trans;
3059         }
3060
3061         path->skip_locking = 1;
3062         path->search_commit_root = 1;
3063
3064         ret = btrfs_lookup_inode(trans, root, path,
3065                                 &BTRFS_I(dir)->location, 0);
3066         if (ret < 0) {
3067                 err = ret;
3068                 goto out;
3069         }
3070         if (ret == 0) {
3071                 if (check_path_shared(root, path))
3072                         goto out;
3073         } else {
3074                 check_link = 0;
3075         }
3076         btrfs_release_path(path);
3077
3078         ret = btrfs_lookup_inode(trans, root, path,
3079                                 &BTRFS_I(inode)->location, 0);
3080         if (ret < 0) {
3081                 err = ret;
3082                 goto out;
3083         }
3084         if (ret == 0) {
3085                 if (check_path_shared(root, path))
3086                         goto out;
3087         } else {
3088                 check_link = 0;
3089         }
3090         btrfs_release_path(path);
3091
3092         if (ret == 0 && S_ISREG(inode->i_mode)) {
3093                 ret = btrfs_lookup_file_extent(trans, root, path,
3094                                                ino, (u64)-1, 0);
3095                 if (ret < 0) {
3096                         err = ret;
3097                         goto out;
3098                 }
3099                 BUG_ON(ret == 0); /* Corruption */
3100                 if (check_path_shared(root, path))
3101                         goto out;
3102                 btrfs_release_path(path);
3103         }
3104
3105         if (!check_link) {
3106                 err = 0;
3107                 goto out;
3108         }
3109
3110         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3111                                 dentry->d_name.name, dentry->d_name.len, 0);
3112         if (IS_ERR(di)) {
3113                 err = PTR_ERR(di);
3114                 goto out;
3115         }
3116         if (di) {
3117                 if (check_path_shared(root, path))
3118                         goto out;
3119         } else {
3120                 err = 0;
3121                 goto out;
3122         }
3123         btrfs_release_path(path);
3124
3125         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3126                                         dentry->d_name.len, ino, dir_ino, 0,
3127                                         &index);
3128         if (ret) {
3129                 err = ret;
3130                 goto out;
3131         }
3132
3133         if (check_path_shared(root, path))
3134                 goto out;
3135
3136         btrfs_release_path(path);
3137
3138         /*
3139          * This is a commit root search, if we can lookup inode item and other
3140          * relative items in the commit root, it means the transaction of
3141          * dir/file creation has been committed, and the dir index item that we
3142          * delay to insert has also been inserted into the commit root. So
3143          * we needn't worry about the delayed insertion of the dir index item
3144          * here.
3145          */
3146         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3147                                 dentry->d_name.name, dentry->d_name.len, 0);
3148         if (IS_ERR(di)) {
3149                 err = PTR_ERR(di);
3150                 goto out;
3151         }
3152         BUG_ON(ret == -ENOENT);
3153         if (check_path_shared(root, path))
3154                 goto out;
3155
3156         err = 0;
3157 out:
3158         btrfs_free_path(path);
3159         /* Migrate the orphan reservation over */
3160         if (!err)
3161                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3162                                 &root->fs_info->global_block_rsv,
3163                                 trans->bytes_reserved);
3164
3165         if (err) {
3166                 btrfs_end_transaction(trans, root);
3167                 root->fs_info->enospc_unlink = 0;
3168                 return ERR_PTR(err);
3169         }
3170
3171         trans->block_rsv = &root->fs_info->global_block_rsv;
3172         return trans;
3173 }
3174
3175 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3176                                struct btrfs_root *root)
3177 {
3178         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3179                 btrfs_block_rsv_release(root, trans->block_rsv,
3180                                         trans->bytes_reserved);
3181                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3182                 BUG_ON(!root->fs_info->enospc_unlink);
3183                 root->fs_info->enospc_unlink = 0;
3184         }
3185         btrfs_end_transaction(trans, root);
3186 }
3187
3188 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3189 {
3190         struct btrfs_root *root = BTRFS_I(dir)->root;
3191         struct btrfs_trans_handle *trans;
3192         struct inode *inode = dentry->d_inode;
3193         int ret;
3194
3195         trans = __unlink_start_trans(dir, dentry);
3196         if (IS_ERR(trans))
3197                 return PTR_ERR(trans);
3198
3199         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3200
3201         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3202                                  dentry->d_name.name, dentry->d_name.len);
3203         if (ret)
3204                 goto out;
3205
3206         if (inode->i_nlink == 0) {
3207                 ret = btrfs_orphan_add(trans, inode);
3208                 if (ret)
3209                         goto out;
3210         }
3211
3212 out:
3213         __unlink_end_trans(trans, root);
3214         btrfs_btree_balance_dirty(root);
3215         return ret;
3216 }
3217
3218 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3219                         struct btrfs_root *root,
3220                         struct inode *dir, u64 objectid,
3221                         const char *name, int name_len)
3222 {
3223         struct btrfs_path *path;
3224         struct extent_buffer *leaf;
3225         struct btrfs_dir_item *di;
3226         struct btrfs_key key;
3227         u64 index;
3228         int ret;
3229         u64 dir_ino = btrfs_ino(dir);
3230
3231         path = btrfs_alloc_path();
3232         if (!path)
3233                 return -ENOMEM;
3234
3235         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3236                                    name, name_len, -1);
3237         if (IS_ERR_OR_NULL(di)) {
3238                 if (!di)
3239                         ret = -ENOENT;
3240                 else
3241                         ret = PTR_ERR(di);
3242                 goto out;
3243         }
3244
3245         leaf = path->nodes[0];
3246         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3247         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3248         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3249         if (ret) {
3250                 btrfs_abort_transaction(trans, root, ret);
3251                 goto out;
3252         }
3253         btrfs_release_path(path);
3254
3255         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3256                                  objectid, root->root_key.objectid,
3257                                  dir_ino, &index, name, name_len);
3258         if (ret < 0) {
3259                 if (ret != -ENOENT) {
3260                         btrfs_abort_transaction(trans, root, ret);
3261                         goto out;
3262                 }
3263                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3264                                                  name, name_len);
3265                 if (IS_ERR_OR_NULL(di)) {
3266                         if (!di)
3267                                 ret = -ENOENT;
3268                         else
3269                                 ret = PTR_ERR(di);
3270                         btrfs_abort_transaction(trans, root, ret);
3271                         goto out;
3272                 }
3273
3274                 leaf = path->nodes[0];
3275                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3276                 btrfs_release_path(path);
3277                 index = key.offset;
3278         }
3279         btrfs_release_path(path);
3280
3281         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3282         if (ret) {
3283                 btrfs_abort_transaction(trans, root, ret);
3284                 goto out;
3285         }
3286
3287         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3288         inode_inc_iversion(dir);
3289         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3290         ret = btrfs_update_inode_fallback(trans, root, dir);
3291         if (ret)
3292                 btrfs_abort_transaction(trans, root, ret);
3293 out:
3294         btrfs_free_path(path);
3295         return ret;
3296 }
3297
3298 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3299 {
3300         struct inode *inode = dentry->d_inode;
3301         int err = 0;
3302         struct btrfs_root *root = BTRFS_I(dir)->root;
3303         struct btrfs_trans_handle *trans;
3304
3305         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3306                 return -ENOTEMPTY;
3307         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3308                 return -EPERM;
3309
3310         trans = __unlink_start_trans(dir, dentry);
3311         if (IS_ERR(trans))
3312                 return PTR_ERR(trans);
3313
3314         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3315                 err = btrfs_unlink_subvol(trans, root, dir,
3316                                           BTRFS_I(inode)->location.objectid,
3317                                           dentry->d_name.name,
3318                                           dentry->d_name.len);
3319                 goto out;
3320         }
3321
3322         err = btrfs_orphan_add(trans, inode);
3323         if (err)
3324                 goto out;
3325
3326         /* now the directory is empty */
3327         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3328                                  dentry->d_name.name, dentry->d_name.len);
3329         if (!err)
3330                 btrfs_i_size_write(inode, 0);
3331 out:
3332         __unlink_end_trans(trans, root);
3333         btrfs_btree_balance_dirty(root);
3334
3335         return err;
3336 }
3337
3338 /*
3339  * this can truncate away extent items, csum items and directory items.
3340  * It starts at a high offset and removes keys until it can't find
3341  * any higher than new_size
3342  *
3343  * csum items that cross the new i_size are truncated to the new size
3344  * as well.
3345  *
3346  * min_type is the minimum key type to truncate down to.  If set to 0, this
3347  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3348  */
3349 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3350                                struct btrfs_root *root,
3351                                struct inode *inode,
3352                                u64 new_size, u32 min_type)
3353 {
3354         struct btrfs_path *path;
3355         struct extent_buffer *leaf;
3356         struct btrfs_file_extent_item *fi;
3357         struct btrfs_key key;
3358         struct btrfs_key found_key;
3359         u64 extent_start = 0;
3360         u64 extent_num_bytes = 0;
3361         u64 extent_offset = 0;
3362         u64 item_end = 0;
3363         u64 mask = root->sectorsize - 1;
3364         u32 found_type = (u8)-1;
3365         int found_extent;
3366         int del_item;
3367         int pending_del_nr = 0;
3368         int pending_del_slot = 0;
3369         int extent_type = -1;
3370         int ret;
3371         int err = 0;
3372         u64 ino = btrfs_ino(inode);
3373
3374         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3375
3376         path = btrfs_alloc_path();
3377         if (!path)
3378                 return -ENOMEM;
3379         path->reada = -1;
3380
3381         /*
3382          * We want to drop from the next block forward in case this new size is
3383          * not block aligned since we will be keeping the last block of the
3384          * extent just the way it is.
3385          */
3386         if (root->ref_cows || root == root->fs_info->tree_root)
3387                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3388
3389         /*
3390          * This function is also used to drop the items in the log tree before
3391          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3392          * it is used to drop the loged items. So we shouldn't kill the delayed
3393          * items.
3394          */
3395         if (min_type == 0 && root == BTRFS_I(inode)->root)
3396                 btrfs_kill_delayed_inode_items(inode);
3397
3398         key.objectid = ino;
3399         key.offset = (u64)-1;
3400         key.type = (u8)-1;
3401
3402 search_again:
3403         path->leave_spinning = 1;
3404         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3405         if (ret < 0) {
3406                 err = ret;
3407                 goto out;
3408         }
3409
3410         if (ret > 0) {
3411                 /* there are no items in the tree for us to truncate, we're
3412                  * done
3413                  */
3414                 if (path->slots[0] == 0)
3415                         goto out;
3416                 path->slots[0]--;
3417         }
3418
3419         while (1) {
3420                 fi = NULL;
3421                 leaf = path->nodes[0];
3422                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3423                 found_type = btrfs_key_type(&found_key);
3424
3425                 if (found_key.objectid != ino)
3426                         break;
3427
3428                 if (found_type < min_type)
3429                         break;
3430
3431                 item_end = found_key.offset;
3432                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3433                         fi = btrfs_item_ptr(leaf, path->slots[0],
3434                                             struct btrfs_file_extent_item);
3435                         extent_type = btrfs_file_extent_type(leaf, fi);
3436                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3437                                 item_end +=
3438                                     btrfs_file_extent_num_bytes(leaf, fi);
3439                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3440                                 item_end += btrfs_file_extent_inline_len(leaf,
3441                                                                          fi);
3442                         }
3443                         item_end--;
3444                 }
3445                 if (found_type > min_type) {
3446                         del_item = 1;
3447                 } else {
3448                         if (item_end < new_size)
3449                                 break;
3450                         if (found_key.offset >= new_size)
3451                                 del_item = 1;
3452                         else
3453                                 del_item = 0;
3454                 }
3455                 found_extent = 0;
3456                 /* FIXME, shrink the extent if the ref count is only 1 */
3457                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3458                         goto delete;
3459
3460                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3461                         u64 num_dec;
3462                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3463                         if (!del_item) {
3464                                 u64 orig_num_bytes =
3465                                         btrfs_file_extent_num_bytes(leaf, fi);
3466                                 extent_num_bytes = new_size -
3467                                         found_key.offset + root->sectorsize - 1;
3468                                 extent_num_bytes = extent_num_bytes &
3469                                         ~((u64)root->sectorsize - 1);
3470                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3471                                                          extent_num_bytes);
3472                                 num_dec = (orig_num_bytes -
3473                                            extent_num_bytes);
3474                                 if (root->ref_cows && extent_start != 0)
3475                                         inode_sub_bytes(inode, num_dec);
3476                                 btrfs_mark_buffer_dirty(leaf);
3477                         } else {
3478                                 extent_num_bytes =
3479                                         btrfs_file_extent_disk_num_bytes(leaf,
3480                                                                          fi);
3481                                 extent_offset = found_key.offset -
3482                                         btrfs_file_extent_offset(leaf, fi);
3483
3484                                 /* FIXME blocksize != 4096 */
3485                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3486                                 if (extent_start != 0) {
3487                                         found_extent = 1;
3488                                         if (root->ref_cows)
3489                                                 inode_sub_bytes(inode, num_dec);
3490                                 }
3491                         }
3492                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3493                         /*
3494                          * we can't truncate inline items that have had
3495                          * special encodings
3496                          */
3497                         if (!del_item &&
3498                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3499                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3500                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3501                                 u32 size = new_size - found_key.offset;
3502
3503                                 if (root->ref_cows) {
3504                                         inode_sub_bytes(inode, item_end + 1 -
3505                                                         new_size);
3506                                 }
3507                                 size =
3508                                     btrfs_file_extent_calc_inline_size(size);
3509                                 btrfs_truncate_item(trans, root, path,
3510                                                     size, 1);
3511                         } else if (root->ref_cows) {
3512                                 inode_sub_bytes(inode, item_end + 1 -
3513                                                 found_key.offset);
3514                         }
3515                 }
3516 delete:
3517                 if (del_item) {
3518                         if (!pending_del_nr) {
3519                                 /* no pending yet, add ourselves */
3520                                 pending_del_slot = path->slots[0];
3521                                 pending_del_nr = 1;
3522                         } else if (pending_del_nr &&
3523                                    path->slots[0] + 1 == pending_del_slot) {
3524                                 /* hop on the pending chunk */
3525                                 pending_del_nr++;
3526                                 pending_del_slot = path->slots[0];
3527                         } else {
3528                                 BUG();
3529                         }
3530                 } else {
3531                         break;
3532                 }
3533                 if (found_extent && (root->ref_cows ||
3534                                      root == root->fs_info->tree_root)) {
3535                         btrfs_set_path_blocking(path);
3536                         ret = btrfs_free_extent(trans, root, extent_start,
3537                                                 extent_num_bytes, 0,
3538                                                 btrfs_header_owner(leaf),
3539                                                 ino, extent_offset, 0);
3540                         BUG_ON(ret);
3541                 }
3542
3543                 if (found_type == BTRFS_INODE_ITEM_KEY)
3544                         break;
3545
3546                 if (path->slots[0] == 0 ||
3547                     path->slots[0] != pending_del_slot) {
3548                         if (pending_del_nr) {
3549                                 ret = btrfs_del_items(trans, root, path,
3550                                                 pending_del_slot,
3551                                                 pending_del_nr);
3552                                 if (ret) {
3553                                         btrfs_abort_transaction(trans,
3554                                                                 root, ret);
3555                                         goto error;
3556                                 }
3557                                 pending_del_nr = 0;
3558                         }
3559                         btrfs_release_path(path);
3560                         goto search_again;
3561                 } else {
3562                         path->slots[0]--;
3563                 }
3564         }
3565 out:
3566         if (pending_del_nr) {
3567                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3568                                       pending_del_nr);
3569                 if (ret)
3570                         btrfs_abort_transaction(trans, root, ret);
3571         }
3572 error:
3573         btrfs_free_path(path);
3574         return err;
3575 }
3576
3577 /*
3578  * btrfs_truncate_page - read, zero a chunk and write a page
3579  * @inode - inode that we're zeroing
3580  * @from - the offset to start zeroing
3581  * @len - the length to zero, 0 to zero the entire range respective to the
3582  *      offset
3583  * @front - zero up to the offset instead of from the offset on
3584  *
3585  * This will find the page for the "from" offset and cow the page and zero the
3586  * part we want to zero.  This is used with truncate and hole punching.
3587  */
3588 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3589                         int front)
3590 {
3591         struct address_space *mapping = inode->i_mapping;
3592         struct btrfs_root *root = BTRFS_I(inode)->root;
3593         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3594         struct btrfs_ordered_extent *ordered;
3595         struct extent_state *cached_state = NULL;
3596         char *kaddr;
3597         u32 blocksize = root->sectorsize;
3598         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3599         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3600         struct page *page;
3601         gfp_t mask = btrfs_alloc_write_mask(mapping);
3602         int ret = 0;
3603         u64 page_start;
3604         u64 page_end;
3605
3606         if ((offset & (blocksize - 1)) == 0 &&
3607             (!len || ((len & (blocksize - 1)) == 0)))
3608                 goto out;
3609         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3610         if (ret)
3611                 goto out;
3612
3613 again:
3614         page = find_or_create_page(mapping, index, mask);
3615         if (!page) {
3616                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3617                 ret = -ENOMEM;
3618                 goto out;
3619         }
3620
3621         page_start = page_offset(page);
3622         page_end = page_start + PAGE_CACHE_SIZE - 1;
3623
3624         if (!PageUptodate(page)) {
3625                 ret = btrfs_readpage(NULL, page);
3626                 lock_page(page);
3627                 if (page->mapping != mapping) {
3628                         unlock_page(page);
3629                         page_cache_release(page);
3630                         goto again;
3631                 }
3632                 if (!PageUptodate(page)) {
3633                         ret = -EIO;
3634                         goto out_unlock;
3635                 }
3636         }
3637         wait_on_page_writeback(page);
3638
3639         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3640         set_page_extent_mapped(page);
3641
3642         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3643         if (ordered) {
3644                 unlock_extent_cached(io_tree, page_start, page_end,
3645                                      &cached_state, GFP_NOFS);
3646                 unlock_page(page);
3647                 page_cache_release(page);
3648                 btrfs_start_ordered_extent(inode, ordered, 1);
3649                 btrfs_put_ordered_extent(ordered);
3650                 goto again;
3651         }
3652
3653         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3654                           EXTENT_DIRTY | EXTENT_DELALLOC |
3655                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3656                           0, 0, &cached_state, GFP_NOFS);
3657
3658         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3659                                         &cached_state);
3660         if (ret) {
3661                 unlock_extent_cached(io_tree, page_start, page_end,
3662                                      &cached_state, GFP_NOFS);
3663                 goto out_unlock;
3664         }
3665
3666         if (offset != PAGE_CACHE_SIZE) {
3667                 if (!len)
3668                         len = PAGE_CACHE_SIZE - offset;
3669                 kaddr = kmap(page);
3670                 if (front)
3671                         memset(kaddr, 0, offset);
3672                 else
3673                         memset(kaddr + offset, 0, len);
3674                 flush_dcache_page(page);
3675                 kunmap(page);
3676         }
3677         ClearPageChecked(page);
3678         set_page_dirty(page);
3679         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3680                              GFP_NOFS);
3681
3682 out_unlock:
3683         if (ret)
3684                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3685         unlock_page(page);
3686         page_cache_release(page);
3687 out:
3688         return ret;
3689 }
3690
3691 /*
3692  * This function puts in dummy file extents for the area we're creating a hole
3693  * for.  So if we are truncating this file to a larger size we need to insert
3694  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3695  * the range between oldsize and size
3696  */
3697 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3698 {
3699         struct btrfs_trans_handle *trans;
3700         struct btrfs_root *root = BTRFS_I(inode)->root;
3701         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3702         struct extent_map *em = NULL;
3703         struct extent_state *cached_state = NULL;
3704         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3705         u64 mask = root->sectorsize - 1;
3706         u64 hole_start = (oldsize + mask) & ~mask;
3707         u64 block_end = (size + mask) & ~mask;
3708         u64 last_byte;
3709         u64 cur_offset;
3710         u64 hole_size;
3711         int err = 0;
3712
3713         if (size <= hole_start)
3714                 return 0;
3715
3716         while (1) {
3717                 struct btrfs_ordered_extent *ordered;
3718                 btrfs_wait_ordered_range(inode, hole_start,
3719                                          block_end - hole_start);
3720                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3721                                  &cached_state);
3722                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3723                 if (!ordered)
3724                         break;
3725                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3726                                      &cached_state, GFP_NOFS);
3727                 btrfs_put_ordered_extent(ordered);
3728         }
3729
3730         cur_offset = hole_start;
3731         while (1) {
3732                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3733                                 block_end - cur_offset, 0);
3734                 if (IS_ERR(em)) {
3735                         err = PTR_ERR(em);
3736                         em = NULL;
3737                         break;
3738                 }
3739                 last_byte = min(extent_map_end(em), block_end);
3740                 last_byte = (last_byte + mask) & ~mask;
3741                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3742                         struct extent_map *hole_em;
3743                         hole_size = last_byte - cur_offset;
3744
3745                         trans = btrfs_start_transaction(root, 3);
3746                         if (IS_ERR(trans)) {
3747                                 err = PTR_ERR(trans);
3748                                 break;
3749                         }
3750
3751                         err = btrfs_drop_extents(trans, root, inode,
3752                                                  cur_offset,
3753                                                  cur_offset + hole_size, 1);
3754                         if (err) {
3755                                 btrfs_abort_transaction(trans, root, err);
3756                                 btrfs_end_transaction(trans, root);
3757                                 break;
3758                         }
3759
3760                         err = btrfs_insert_file_extent(trans, root,
3761                                         btrfs_ino(inode), cur_offset, 0,
3762                                         0, hole_size, 0, hole_size,
3763                                         0, 0, 0);
3764                         if (err) {
3765                                 btrfs_abort_transaction(trans, root, err);
3766                                 btrfs_end_transaction(trans, root);
3767                                 break;
3768                         }
3769
3770                         btrfs_drop_extent_cache(inode, cur_offset,
3771                                                 cur_offset + hole_size - 1, 0);
3772                         hole_em = alloc_extent_map();
3773                         if (!hole_em) {
3774                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3775                                         &BTRFS_I(inode)->runtime_flags);
3776                                 goto next;
3777                         }
3778                         hole_em->start = cur_offset;
3779                         hole_em->len = hole_size;
3780                         hole_em->orig_start = cur_offset;
3781
3782                         hole_em->block_start = EXTENT_MAP_HOLE;
3783                         hole_em->block_len = 0;
3784                         hole_em->orig_block_len = 0;
3785                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3786                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3787                         hole_em->generation = trans->transid;
3788
3789                         while (1) {
3790                                 write_lock(&em_tree->lock);
3791                                 err = add_extent_mapping(em_tree, hole_em);
3792                                 if (!err)
3793                                         list_move(&hole_em->list,
3794                                                   &em_tree->modified_extents);
3795                                 write_unlock(&em_tree->lock);
3796                                 if (err != -EEXIST)
3797                                         break;
3798                                 btrfs_drop_extent_cache(inode, cur_offset,
3799                                                         cur_offset +
3800                                                         hole_size - 1, 0);
3801                         }
3802                         free_extent_map(hole_em);
3803 next:
3804                         btrfs_update_inode(trans, root, inode);
3805                         btrfs_end_transaction(trans, root);
3806                 }
3807                 free_extent_map(em);
3808                 em = NULL;
3809                 cur_offset = last_byte;
3810                 if (cur_offset >= block_end)
3811                         break;
3812         }
3813
3814         free_extent_map(em);
3815         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3816                              GFP_NOFS);
3817         return err;
3818 }
3819
3820 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
3821 {
3822         struct btrfs_root *root = BTRFS_I(inode)->root;
3823         struct btrfs_trans_handle *trans;
3824         loff_t oldsize = i_size_read(inode);
3825         loff_t newsize = attr->ia_size;
3826         int mask = attr->ia_valid;
3827         int ret;
3828
3829         if (newsize == oldsize)
3830                 return 0;
3831
3832         /*
3833          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3834          * special case where we need to update the times despite not having
3835          * these flags set.  For all other operations the VFS set these flags
3836          * explicitly if it wants a timestamp update.
3837          */
3838         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3839                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3840
3841         if (newsize > oldsize) {
3842                 truncate_pagecache(inode, oldsize, newsize);
3843                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3844                 if (ret)
3845                         return ret;
3846
3847                 trans = btrfs_start_transaction(root, 1);
3848                 if (IS_ERR(trans))
3849                         return PTR_ERR(trans);
3850
3851                 i_size_write(inode, newsize);
3852                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3853                 ret = btrfs_update_inode(trans, root, inode);
3854                 btrfs_end_transaction(trans, root);
3855         } else {
3856
3857                 /*
3858                  * We're truncating a file that used to have good data down to
3859                  * zero. Make sure it gets into the ordered flush list so that
3860                  * any new writes get down to disk quickly.
3861                  */
3862                 if (newsize == 0)
3863                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3864                                 &BTRFS_I(inode)->runtime_flags);
3865
3866                 /*
3867                  * 1 for the orphan item we're going to add
3868                  * 1 for the orphan item deletion.
3869                  */
3870                 trans = btrfs_start_transaction(root, 2);
3871                 if (IS_ERR(trans))
3872                         return PTR_ERR(trans);
3873
3874                 /*
3875                  * We need to do this in case we fail at _any_ point during the
3876                  * actual truncate.  Once we do the truncate_setsize we could
3877                  * invalidate pages which forces any outstanding ordered io to
3878                  * be instantly completed which will give us extents that need
3879                  * to be truncated.  If we fail to get an orphan inode down we
3880                  * could have left over extents that were never meant to live,
3881                  * so we need to garuntee from this point on that everything
3882                  * will be consistent.
3883                  */
3884                 ret = btrfs_orphan_add(trans, inode);
3885                 btrfs_end_transaction(trans, root);
3886                 if (ret)
3887                         return ret;
3888
3889                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3890                 truncate_setsize(inode, newsize);
3891                 ret = btrfs_truncate(inode);
3892                 if (ret && inode->i_nlink)
3893                         btrfs_orphan_del(NULL, inode);
3894         }
3895
3896         return ret;
3897 }
3898
3899 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3900 {
3901         struct inode *inode = dentry->d_inode;
3902         struct btrfs_root *root = BTRFS_I(inode)->root;
3903         int err;
3904
3905         if (btrfs_root_readonly(root))
3906                 return -EROFS;
3907
3908         err = inode_change_ok(inode, attr);
3909         if (err)
3910                 return err;
3911
3912         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3913                 err = btrfs_setsize(inode, attr);
3914                 if (err)
3915                         return err;
3916         }
3917
3918         if (attr->ia_valid) {
3919                 setattr_copy(inode, attr);
3920                 inode_inc_iversion(inode);
3921                 err = btrfs_dirty_inode(inode);
3922
3923                 if (!err && attr->ia_valid & ATTR_MODE)
3924                         err = btrfs_acl_chmod(inode);
3925         }
3926
3927         return err;
3928 }
3929
3930 void btrfs_evict_inode(struct inode *inode)
3931 {
3932         struct btrfs_trans_handle *trans;
3933         struct btrfs_root *root = BTRFS_I(inode)->root;
3934         struct btrfs_block_rsv *rsv, *global_rsv;
3935         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3936         int ret;
3937
3938         trace_btrfs_inode_evict(inode);
3939
3940         truncate_inode_pages(&inode->i_data, 0);
3941         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3942                                btrfs_is_free_space_inode(inode)))
3943                 goto no_delete;
3944
3945         if (is_bad_inode(inode)) {
3946                 btrfs_orphan_del(NULL, inode);
3947                 goto no_delete;
3948         }
3949         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3950         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3951
3952         if (root->fs_info->log_root_recovering) {
3953                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3954                                  &BTRFS_I(inode)->runtime_flags));
3955                 goto no_delete;
3956         }
3957
3958         if (inode->i_nlink > 0) {
3959                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3960                 goto no_delete;
3961         }
3962
3963         ret = btrfs_commit_inode_delayed_inode(inode);
3964         if (ret) {
3965                 btrfs_orphan_del(NULL, inode);
3966                 goto no_delete;
3967         }
3968
3969         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3970         if (!rsv) {
3971                 btrfs_orphan_del(NULL, inode);
3972                 goto no_delete;
3973         }
3974         rsv->size = min_size;
3975         rsv->failfast = 1;
3976         global_rsv = &root->fs_info->global_block_rsv;
3977
3978         btrfs_i_size_write(inode, 0);
3979
3980         /*
3981          * This is a bit simpler than btrfs_truncate since we've already
3982          * reserved our space for our orphan item in the unlink, so we just
3983          * need to reserve some slack space in case we add bytes and update
3984          * inode item when doing the truncate.
3985          */
3986         while (1) {
3987                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3988                                              BTRFS_RESERVE_FLUSH_LIMIT);
3989
3990                 /*
3991                  * Try and steal from the global reserve since we will
3992                  * likely not use this space anyway, we want to try as
3993                  * hard as possible to get this to work.
3994                  */
3995                 if (ret)
3996                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3997
3998                 if (ret) {
3999                         printk(KERN_WARNING "Could not get space for a "
4000                                "delete, will truncate on mount %d\n", ret);
4001                         btrfs_orphan_del(NULL, inode);
4002                         btrfs_free_block_rsv(root, rsv);
4003                         goto no_delete;
4004                 }
4005
4006                 trans = btrfs_join_transaction(root);
4007                 if (IS_ERR(trans)) {
4008                         btrfs_orphan_del(NULL, inode);
4009                         btrfs_free_block_rsv(root, rsv);
4010                         goto no_delete;
4011                 }
4012
4013                 trans->block_rsv = rsv;
4014
4015                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
4016                 if (ret != -ENOSPC)
4017                         break;
4018
4019                 trans->block_rsv = &root->fs_info->trans_block_rsv;
4020                 btrfs_end_transaction(trans, root);
4021                 trans = NULL;
4022                 btrfs_btree_balance_dirty(root);
4023         }
4024
4025         btrfs_free_block_rsv(root, rsv);
4026
4027         if (ret == 0) {
4028                 trans->block_rsv = root->orphan_block_rsv;
4029                 ret = btrfs_orphan_del(trans, inode);
4030                 BUG_ON(ret);
4031         }
4032
4033         trans->block_rsv = &root->fs_info->trans_block_rsv;
4034         if (!(root == root->fs_info->tree_root ||
4035               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
4036                 btrfs_return_ino(root, btrfs_ino(inode));
4037
4038         btrfs_end_transaction(trans, root);
4039         btrfs_btree_balance_dirty(root);
4040 no_delete:
4041         clear_inode(inode);
4042         return;
4043 }
4044
4045 /*
4046  * this returns the key found in the dir entry in the location pointer.
4047  * If no dir entries were found, location->objectid is 0.
4048  */
4049 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4050                                struct btrfs_key *location)
4051 {
4052         const char *name = dentry->d_name.name;
4053         int namelen = dentry->d_name.len;
4054         struct btrfs_dir_item *di;
4055         struct btrfs_path *path;
4056         struct btrfs_root *root = BTRFS_I(dir)->root;
4057         int ret = 0;
4058
4059         path = btrfs_alloc_path();
4060         if (!path)
4061                 return -ENOMEM;
4062
4063         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4064                                     namelen, 0);
4065         if (IS_ERR(di))
4066                 ret = PTR_ERR(di);
4067
4068         if (IS_ERR_OR_NULL(di))
4069                 goto out_err;
4070
4071         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4072 out:
4073         btrfs_free_path(path);
4074         return ret;
4075 out_err:
4076         location->objectid = 0;
4077         goto out;
4078 }
4079
4080 /*
4081  * when we hit a tree root in a directory, the btrfs part of the inode
4082  * needs to be changed to reflect the root directory of the tree root.  This
4083  * is kind of like crossing a mount point.
4084  */
4085 static int fixup_tree_root_location(struct btrfs_root *root,
4086                                     struct inode *dir,
4087                                     struct dentry *dentry,
4088                                     struct btrfs_key *location,
4089                                     struct btrfs_root **sub_root)
4090 {
4091         struct btrfs_path *path;
4092         struct btrfs_root *new_root;
4093         struct btrfs_root_ref *ref;
4094         struct extent_buffer *leaf;
4095         int ret;
4096         int err = 0;
4097
4098         path = btrfs_alloc_path();
4099         if (!path) {
4100                 err = -ENOMEM;
4101                 goto out;
4102         }
4103
4104         err = -ENOENT;
4105         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4106                                   BTRFS_I(dir)->root->root_key.objectid,
4107                                   location->objectid);
4108         if (ret) {
4109                 if (ret < 0)
4110                         err = ret;
4111                 goto out;
4112         }
4113
4114         leaf = path->nodes[0];
4115         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4116         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4117             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4118                 goto out;
4119
4120         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4121                                    (unsigned long)(ref + 1),
4122                                    dentry->d_name.len);
4123         if (ret)
4124                 goto out;
4125
4126         btrfs_release_path(path);
4127
4128         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4129         if (IS_ERR(new_root)) {
4130                 err = PTR_ERR(new_root);
4131                 goto out;
4132         }
4133
4134         if (btrfs_root_refs(&new_root->root_item) == 0) {
4135                 err = -ENOENT;
4136                 goto out;
4137         }
4138
4139         *sub_root = new_root;
4140         location->objectid = btrfs_root_dirid(&new_root->root_item);
4141         location->type = BTRFS_INODE_ITEM_KEY;
4142         location->offset = 0;
4143         err = 0;
4144 out:
4145         btrfs_free_path(path);
4146         return err;
4147 }
4148
4149 static void inode_tree_add(struct inode *inode)
4150 {
4151         struct btrfs_root *root = BTRFS_I(inode)->root;
4152         struct btrfs_inode *entry;
4153         struct rb_node **p;
4154         struct rb_node *parent;
4155         u64 ino = btrfs_ino(inode);
4156 again:
4157         p = &root->inode_tree.rb_node;
4158         parent = NULL;
4159
4160         if (inode_unhashed(inode))
4161                 return;
4162
4163         spin_lock(&root->inode_lock);
4164         while (*p) {
4165                 parent = *p;
4166                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4167
4168                 if (ino < btrfs_ino(&entry->vfs_inode))
4169                         p = &parent->rb_left;
4170                 else if (ino > btrfs_ino(&entry->vfs_inode))
4171                         p = &parent->rb_right;
4172                 else {
4173                         WARN_ON(!(entry->vfs_inode.i_state &
4174                                   (I_WILL_FREE | I_FREEING)));
4175                         rb_erase(parent, &root->inode_tree);
4176                         RB_CLEAR_NODE(parent);
4177                         spin_unlock(&root->inode_lock);
4178                         goto again;
4179                 }
4180         }
4181         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4182         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4183         spin_unlock(&root->inode_lock);
4184 }
4185
4186 static void inode_tree_del(struct inode *inode)
4187 {
4188         struct btrfs_root *root = BTRFS_I(inode)->root;
4189         int empty = 0;
4190
4191         spin_lock(&root->inode_lock);
4192         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4193                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4194                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4195                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4196         }
4197         spin_unlock(&root->inode_lock);
4198
4199         /*
4200          * Free space cache has inodes in the tree root, but the tree root has a
4201          * root_refs of 0, so this could end up dropping the tree root as a
4202          * snapshot, so we need the extra !root->fs_info->tree_root check to
4203          * make sure we don't drop it.
4204          */
4205         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4206             root != root->fs_info->tree_root) {
4207                 synchronize_srcu(&root->fs_info->subvol_srcu);
4208                 spin_lock(&root->inode_lock);
4209                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4210                 spin_unlock(&root->inode_lock);
4211                 if (empty)
4212                         btrfs_add_dead_root(root);
4213         }
4214 }
4215
4216 void btrfs_invalidate_inodes(struct btrfs_root *root)
4217 {
4218         struct rb_node *node;
4219         struct rb_node *prev;
4220         struct btrfs_inode *entry;
4221         struct inode *inode;
4222         u64 objectid = 0;
4223
4224         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4225
4226         spin_lock(&root->inode_lock);
4227 again:
4228         node = root->inode_tree.rb_node;
4229         prev = NULL;
4230         while (node) {
4231                 prev = node;
4232                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4233
4234                 if (objectid < btrfs_ino(&entry->vfs_inode))
4235                         node = node->rb_left;
4236                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4237                         node = node->rb_right;
4238                 else
4239                         break;
4240         }
4241         if (!node) {
4242                 while (prev) {
4243                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4244                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4245                                 node = prev;
4246                                 break;
4247                         }
4248                         prev = rb_next(prev);
4249                 }
4250         }
4251         while (node) {
4252                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4253                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4254                 inode = igrab(&entry->vfs_inode);
4255                 if (inode) {
4256                         spin_unlock(&root->inode_lock);
4257                         if (atomic_read(&inode->i_count) > 1)
4258                                 d_prune_aliases(inode);
4259                         /*
4260                          * btrfs_drop_inode will have it removed from
4261                          * the inode cache when its usage count
4262                          * hits zero.
4263                          */
4264                         iput(inode);
4265                         cond_resched();
4266                         spin_lock(&root->inode_lock);
4267                         goto again;
4268                 }
4269
4270                 if (cond_resched_lock(&root->inode_lock))
4271                         goto again;
4272
4273                 node = rb_next(node);
4274         }
4275         spin_unlock(&root->inode_lock);
4276 }
4277
4278 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4279 {
4280         struct btrfs_iget_args *args = p;
4281         inode->i_ino = args->ino;
4282         BTRFS_I(inode)->root = args->root;
4283         return 0;
4284 }
4285
4286 static int btrfs_find_actor(struct inode *inode, void *opaque)
4287 {
4288         struct btrfs_iget_args *args = opaque;
4289         return args->ino == btrfs_ino(inode) &&
4290                 args->root == BTRFS_I(inode)->root;
4291 }
4292
4293 static struct inode *btrfs_iget_locked(struct super_block *s,
4294                                        u64 objectid,
4295                                        struct btrfs_root *root)
4296 {
4297         struct inode *inode;
4298         struct btrfs_iget_args args;
4299         args.ino = objectid;
4300         args.root = root;
4301
4302         inode = iget5_locked(s, objectid, btrfs_find_actor,
4303                              btrfs_init_locked_inode,
4304                              (void *)&args);
4305         return inode;
4306 }
4307
4308 /* Get an inode object given its location and corresponding root.
4309  * Returns in *is_new if the inode was read from disk
4310  */
4311 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4312                          struct btrfs_root *root, int *new)
4313 {
4314         struct inode *inode;
4315
4316         inode = btrfs_iget_locked(s, location->objectid, root);
4317         if (!inode)
4318                 return ERR_PTR(-ENOMEM);
4319
4320         if (inode->i_state & I_NEW) {
4321                 BTRFS_I(inode)->root = root;
4322                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4323                 btrfs_read_locked_inode(inode);
4324                 if (!is_bad_inode(inode)) {
4325                         inode_tree_add(inode);
4326                         unlock_new_inode(inode);
4327                         if (new)
4328                                 *new = 1;
4329                 } else {
4330                         unlock_new_inode(inode);
4331                         iput(inode);
4332                         inode = ERR_PTR(-ESTALE);
4333                 }
4334         }
4335
4336         return inode;
4337 }
4338
4339 static struct inode *new_simple_dir(struct super_block *s,
4340                                     struct btrfs_key *key,
4341                                     struct btrfs_root *root)
4342 {
4343         struct inode *inode = new_inode(s);
4344
4345         if (!inode)
4346                 return ERR_PTR(-ENOMEM);
4347
4348         BTRFS_I(inode)->root = root;
4349         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4350         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4351
4352         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4353         inode->i_op = &btrfs_dir_ro_inode_operations;
4354         inode->i_fop = &simple_dir_operations;
4355         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4356         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4357
4358         return inode;
4359 }
4360
4361 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4362 {
4363         struct inode *inode;
4364         struct btrfs_root *root = BTRFS_I(dir)->root;
4365         struct btrfs_root *sub_root = root;
4366         struct btrfs_key location;
4367         int index;
4368         int ret = 0;
4369
4370         if (dentry->d_name.len > BTRFS_NAME_LEN)
4371                 return ERR_PTR(-ENAMETOOLONG);
4372
4373         if (unlikely(d_need_lookup(dentry))) {
4374                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4375                 kfree(dentry->d_fsdata);
4376                 dentry->d_fsdata = NULL;
4377                 /* This thing is hashed, drop it for now */
4378                 d_drop(dentry);
4379         } else {
4380                 ret = btrfs_inode_by_name(dir, dentry, &location);
4381         }
4382
4383         if (ret < 0)
4384                 return ERR_PTR(ret);
4385
4386         if (location.objectid == 0)
4387                 return NULL;
4388
4389         if (location.type == BTRFS_INODE_ITEM_KEY) {
4390                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4391                 return inode;
4392         }
4393
4394         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4395
4396         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4397         ret = fixup_tree_root_location(root, dir, dentry,
4398                                        &location, &sub_root);
4399         if (ret < 0) {
4400                 if (ret != -ENOENT)
4401                         inode = ERR_PTR(ret);
4402                 else
4403                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4404         } else {
4405                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4406         }
4407         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4408
4409         if (!IS_ERR(inode) && root != sub_root) {
4410                 down_read(&root->fs_info->cleanup_work_sem);
4411                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4412                         ret = btrfs_orphan_cleanup(sub_root);
4413                 up_read(&root->fs_info->cleanup_work_sem);
4414                 if (ret)
4415                         inode = ERR_PTR(ret);
4416         }
4417
4418         return inode;
4419 }
4420
4421 static int btrfs_dentry_delete(const struct dentry *dentry)
4422 {
4423         struct btrfs_root *root;
4424         struct inode *inode = dentry->d_inode;
4425
4426         if (!inode && !IS_ROOT(dentry))
4427                 inode = dentry->d_parent->d_inode;
4428
4429         if (inode) {
4430                 root = BTRFS_I(inode)->root;
4431                 if (btrfs_root_refs(&root->root_item) == 0)
4432                         return 1;
4433
4434                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4435                         return 1;
4436         }
4437         return 0;
4438 }
4439
4440 static void btrfs_dentry_release(struct dentry *dentry)
4441 {
4442         if (dentry->d_fsdata)
4443                 kfree(dentry->d_fsdata);
4444 }
4445
4446 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4447                                    unsigned int flags)
4448 {
4449         struct dentry *ret;
4450
4451         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4452         if (unlikely(d_need_lookup(dentry))) {
4453                 spin_lock(&dentry->d_lock);
4454                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4455                 spin_unlock(&dentry->d_lock);
4456         }
4457         return ret;
4458 }
4459
4460 unsigned char btrfs_filetype_table[] = {
4461         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4462 };
4463
4464 static int btrfs_real_readdir(struct file *filp, void *dirent,
4465                               filldir_t filldir)
4466 {
4467         struct inode *inode = filp->f_dentry->d_inode;
4468         struct btrfs_root *root = BTRFS_I(inode)->root;
4469         struct btrfs_item *item;
4470         struct btrfs_dir_item *di;
4471         struct btrfs_key key;
4472         struct btrfs_key found_key;
4473         struct btrfs_path *path;
4474         struct list_head ins_list;
4475         struct list_head del_list;
4476         int ret;
4477         struct extent_buffer *leaf;
4478         int slot;
4479         unsigned char d_type;
4480         int over = 0;
4481         u32 di_cur;
4482         u32 di_total;
4483         u32 di_len;
4484         int key_type = BTRFS_DIR_INDEX_KEY;
4485         char tmp_name[32];
4486         char *name_ptr;
4487         int name_len;
4488         int is_curr = 0;        /* filp->f_pos points to the current index? */
4489
4490         /* FIXME, use a real flag for deciding about the key type */
4491         if (root->fs_info->tree_root == root)
4492                 key_type = BTRFS_DIR_ITEM_KEY;
4493
4494         /* special case for "." */
4495         if (filp->f_pos == 0) {
4496                 over = filldir(dirent, ".", 1,
4497                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4498                 if (over)
4499                         return 0;
4500                 filp->f_pos = 1;
4501         }
4502         /* special case for .., just use the back ref */
4503         if (filp->f_pos == 1) {
4504                 u64 pino = parent_ino(filp->f_path.dentry);
4505                 over = filldir(dirent, "..", 2,
4506                                filp->f_pos, pino, DT_DIR);
4507                 if (over)
4508                         return 0;
4509                 filp->f_pos = 2;
4510         }
4511         path = btrfs_alloc_path();
4512         if (!path)
4513                 return -ENOMEM;
4514
4515         path->reada = 1;
4516
4517         if (key_type == BTRFS_DIR_INDEX_KEY) {
4518                 INIT_LIST_HEAD(&ins_list);
4519                 INIT_LIST_HEAD(&del_list);
4520                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4521         }
4522
4523         btrfs_set_key_type(&key, key_type);
4524         key.offset = filp->f_pos;
4525         key.objectid = btrfs_ino(inode);
4526
4527         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4528         if (ret < 0)
4529                 goto err;
4530
4531         while (1) {
4532                 leaf = path->nodes[0];
4533                 slot = path->slots[0];
4534                 if (slot >= btrfs_header_nritems(leaf)) {
4535                         ret = btrfs_next_leaf(root, path);
4536                         if (ret < 0)
4537                                 goto err;
4538                         else if (ret > 0)
4539                                 break;
4540                         continue;
4541                 }
4542
4543                 item = btrfs_item_nr(leaf, slot);
4544                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4545
4546                 if (found_key.objectid != key.objectid)
4547                         break;
4548                 if (btrfs_key_type(&found_key) != key_type)
4549                         break;
4550                 if (found_key.offset < filp->f_pos)
4551                         goto next;
4552                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4553                     btrfs_should_delete_dir_index(&del_list,
4554                                                   found_key.offset))
4555                         goto next;
4556
4557                 filp->f_pos = found_key.offset;
4558                 is_curr = 1;
4559
4560                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4561                 di_cur = 0;
4562                 di_total = btrfs_item_size(leaf, item);
4563
4564                 while (di_cur < di_total) {
4565                         struct btrfs_key location;
4566
4567                         if (verify_dir_item(root, leaf, di))
4568                                 break;
4569
4570                         name_len = btrfs_dir_name_len(leaf, di);
4571                         if (name_len <= sizeof(tmp_name)) {
4572                                 name_ptr = tmp_name;
4573                         } else {
4574                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4575                                 if (!name_ptr) {
4576                                         ret = -ENOMEM;
4577                                         goto err;
4578                                 }
4579                         }
4580                         read_extent_buffer(leaf, name_ptr,
4581                                            (unsigned long)(di + 1), name_len);
4582
4583                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4584                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4585
4586
4587                         /* is this a reference to our own snapshot? If so
4588                          * skip it.
4589                          *
4590                          * In contrast to old kernels, we insert the snapshot's
4591                          * dir item and dir index after it has been created, so
4592                          * we won't find a reference to our own snapshot. We
4593                          * still keep the following code for backward
4594                          * compatibility.
4595                          */
4596                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4597                             location.objectid == root->root_key.objectid) {
4598                                 over = 0;
4599                                 goto skip;
4600                         }
4601                         over = filldir(dirent, name_ptr, name_len,
4602                                        found_key.offset, location.objectid,
4603                                        d_type);
4604
4605 skip:
4606                         if (name_ptr != tmp_name)
4607                                 kfree(name_ptr);
4608
4609                         if (over)
4610                                 goto nopos;
4611                         di_len = btrfs_dir_name_len(leaf, di) +
4612                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4613                         di_cur += di_len;
4614                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4615                 }
4616 next:
4617                 path->slots[0]++;
4618         }
4619
4620         if (key_type == BTRFS_DIR_INDEX_KEY) {
4621                 if (is_curr)
4622                         filp->f_pos++;
4623                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4624                                                       &ins_list);
4625                 if (ret)
4626                         goto nopos;
4627         }
4628
4629         /* Reached end of directory/root. Bump pos past the last item. */
4630         if (key_type == BTRFS_DIR_INDEX_KEY)
4631                 /*
4632                  * 32-bit glibc will use getdents64, but then strtol -
4633                  * so the last number we can serve is this.
4634                  */
4635                 filp->f_pos = 0x7fffffff;
4636         else
4637                 filp->f_pos++;
4638 nopos:
4639         ret = 0;
4640 err:
4641         if (key_type == BTRFS_DIR_INDEX_KEY)
4642                 btrfs_put_delayed_items(&ins_list, &del_list);
4643         btrfs_free_path(path);
4644         return ret;
4645 }
4646
4647 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4648 {
4649         struct btrfs_root *root = BTRFS_I(inode)->root;
4650         struct btrfs_trans_handle *trans;
4651         int ret = 0;
4652         bool nolock = false;
4653
4654         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4655                 return 0;
4656
4657         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4658                 nolock = true;
4659
4660         if (wbc->sync_mode == WB_SYNC_ALL) {
4661                 if (nolock)
4662                         trans = btrfs_join_transaction_nolock(root);
4663                 else
4664                         trans = btrfs_join_transaction(root);
4665                 if (IS_ERR(trans))
4666                         return PTR_ERR(trans);
4667                 ret = btrfs_commit_transaction(trans, root);
4668         }
4669         return ret;
4670 }
4671
4672 /*
4673  * This is somewhat expensive, updating the tree every time the
4674  * inode changes.  But, it is most likely to find the inode in cache.
4675  * FIXME, needs more benchmarking...there are no reasons other than performance
4676  * to keep or drop this code.
4677  */
4678 int btrfs_dirty_inode(struct inode *inode)
4679 {
4680         struct btrfs_root *root = BTRFS_I(inode)->root;
4681         struct btrfs_trans_handle *trans;
4682         int ret;
4683
4684         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4685                 return 0;
4686
4687         trans = btrfs_join_transaction(root);
4688         if (IS_ERR(trans))
4689                 return PTR_ERR(trans);
4690
4691         ret = btrfs_update_inode(trans, root, inode);
4692         if (ret && ret == -ENOSPC) {
4693                 /* whoops, lets try again with the full transaction */
4694                 btrfs_end_transaction(trans, root);
4695                 trans = btrfs_start_transaction(root, 1);
4696                 if (IS_ERR(trans))
4697                         return PTR_ERR(trans);
4698
4699                 ret = btrfs_update_inode(trans, root, inode);
4700         }
4701         btrfs_end_transaction(trans, root);
4702         if (BTRFS_I(inode)->delayed_node)
4703                 btrfs_balance_delayed_items(root);
4704
4705         return ret;
4706 }
4707
4708 /*
4709  * This is a copy of file_update_time.  We need this so we can return error on
4710  * ENOSPC for updating the inode in the case of file write and mmap writes.
4711  */
4712 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4713                              int flags)
4714 {
4715         struct btrfs_root *root = BTRFS_I(inode)->root;
4716
4717         if (btrfs_root_readonly(root))
4718                 return -EROFS;
4719
4720         if (flags & S_VERSION)
4721                 inode_inc_iversion(inode);
4722         if (flags & S_CTIME)
4723                 inode->i_ctime = *now;
4724         if (flags & S_MTIME)
4725                 inode->i_mtime = *now;
4726         if (flags & S_ATIME)
4727                 inode->i_atime = *now;
4728         return btrfs_dirty_inode(inode);
4729 }
4730
4731 /*
4732  * find the highest existing sequence number in a directory
4733  * and then set the in-memory index_cnt variable to reflect
4734  * free sequence numbers
4735  */
4736 static int btrfs_set_inode_index_count(struct inode *inode)
4737 {
4738         struct btrfs_root *root = BTRFS_I(inode)->root;
4739         struct btrfs_key key, found_key;
4740         struct btrfs_path *path;
4741         struct extent_buffer *leaf;
4742         int ret;
4743
4744         key.objectid = btrfs_ino(inode);
4745         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4746         key.offset = (u64)-1;
4747
4748         path = btrfs_alloc_path();
4749         if (!path)
4750                 return -ENOMEM;
4751
4752         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4753         if (ret < 0)
4754                 goto out;
4755         /* FIXME: we should be able to handle this */
4756         if (ret == 0)
4757                 goto out;
4758         ret = 0;
4759
4760         /*
4761          * MAGIC NUMBER EXPLANATION:
4762          * since we search a directory based on f_pos we have to start at 2
4763          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4764          * else has to start at 2
4765          */
4766         if (path->slots[0] == 0) {
4767                 BTRFS_I(inode)->index_cnt = 2;
4768                 goto out;
4769         }
4770
4771         path->slots[0]--;
4772
4773         leaf = path->nodes[0];
4774         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4775
4776         if (found_key.objectid != btrfs_ino(inode) ||
4777             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4778                 BTRFS_I(inode)->index_cnt = 2;
4779                 goto out;
4780         }
4781
4782         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4783 out:
4784         btrfs_free_path(path);
4785         return ret;
4786 }
4787
4788 /*
4789  * helper to find a free sequence number in a given directory.  This current
4790  * code is very simple, later versions will do smarter things in the btree
4791  */
4792 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4793 {
4794         int ret = 0;
4795
4796         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4797                 ret = btrfs_inode_delayed_dir_index_count(dir);
4798                 if (ret) {
4799                         ret = btrfs_set_inode_index_count(dir);
4800                         if (ret)
4801                                 return ret;
4802                 }
4803         }
4804
4805         *index = BTRFS_I(dir)->index_cnt;
4806         BTRFS_I(dir)->index_cnt++;
4807
4808         return ret;
4809 }
4810
4811 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4812                                      struct btrfs_root *root,
4813                                      struct inode *dir,
4814                                      const char *name, int name_len,
4815                                      u64 ref_objectid, u64 objectid,
4816                                      umode_t mode, u64 *index)
4817 {
4818         struct inode *inode;
4819         struct btrfs_inode_item *inode_item;
4820         struct btrfs_key *location;
4821         struct btrfs_path *path;
4822         struct btrfs_inode_ref *ref;
4823         struct btrfs_key key[2];
4824         u32 sizes[2];
4825         unsigned long ptr;
4826         int ret;
4827         int owner;
4828
4829         path = btrfs_alloc_path();
4830         if (!path)
4831                 return ERR_PTR(-ENOMEM);
4832
4833         inode = new_inode(root->fs_info->sb);
4834         if (!inode) {
4835                 btrfs_free_path(path);
4836                 return ERR_PTR(-ENOMEM);
4837         }
4838
4839         /*
4840          * we have to initialize this early, so we can reclaim the inode
4841          * number if we fail afterwards in this function.
4842          */
4843         inode->i_ino = objectid;
4844
4845         if (dir) {
4846                 trace_btrfs_inode_request(dir);
4847
4848                 ret = btrfs_set_inode_index(dir, index);
4849                 if (ret) {
4850                         btrfs_free_path(path);
4851                         iput(inode);
4852                         return ERR_PTR(ret);
4853                 }
4854         }
4855         /*
4856          * index_cnt is ignored for everything but a dir,
4857          * btrfs_get_inode_index_count has an explanation for the magic
4858          * number
4859          */
4860         BTRFS_I(inode)->index_cnt = 2;
4861         BTRFS_I(inode)->root = root;
4862         BTRFS_I(inode)->generation = trans->transid;
4863         inode->i_generation = BTRFS_I(inode)->generation;
4864
4865         /*
4866          * We could have gotten an inode number from somebody who was fsynced
4867          * and then removed in this same transaction, so let's just set full
4868          * sync since it will be a full sync anyway and this will blow away the
4869          * old info in the log.
4870          */
4871         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4872
4873         if (S_ISDIR(mode))
4874                 owner = 0;
4875         else
4876                 owner = 1;
4877
4878         key[0].objectid = objectid;
4879         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4880         key[0].offset = 0;
4881
4882         /*
4883          * Start new inodes with an inode_ref. This is slightly more
4884          * efficient for small numbers of hard links since they will
4885          * be packed into one item. Extended refs will kick in if we
4886          * add more hard links than can fit in the ref item.
4887          */
4888         key[1].objectid = objectid;
4889         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4890         key[1].offset = ref_objectid;
4891
4892         sizes[0] = sizeof(struct btrfs_inode_item);
4893         sizes[1] = name_len + sizeof(*ref);
4894
4895         path->leave_spinning = 1;
4896         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4897         if (ret != 0)
4898                 goto fail;
4899
4900         inode_init_owner(inode, dir, mode);
4901         inode_set_bytes(inode, 0);
4902         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4903         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4904                                   struct btrfs_inode_item);
4905         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4906                              sizeof(*inode_item));
4907         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4908
4909         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4910                              struct btrfs_inode_ref);
4911         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4912         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4913         ptr = (unsigned long)(ref + 1);
4914         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4915
4916         btrfs_mark_buffer_dirty(path->nodes[0]);
4917         btrfs_free_path(path);
4918
4919         location = &BTRFS_I(inode)->location;
4920         location->objectid = objectid;
4921         location->offset = 0;
4922         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4923
4924         btrfs_inherit_iflags(inode, dir);
4925
4926         if (S_ISREG(mode)) {
4927                 if (btrfs_test_opt(root, NODATASUM))
4928                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4929                 if (btrfs_test_opt(root, NODATACOW))
4930                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4931         }
4932
4933         insert_inode_hash(inode);
4934         inode_tree_add(inode);
4935
4936         trace_btrfs_inode_new(inode);
4937         btrfs_set_inode_last_trans(trans, inode);
4938
4939         btrfs_update_root_times(trans, root);
4940
4941         return inode;
4942 fail:
4943         if (dir)
4944                 BTRFS_I(dir)->index_cnt--;
4945         btrfs_free_path(path);
4946         iput(inode);
4947         return ERR_PTR(ret);
4948 }
4949
4950 static inline u8 btrfs_inode_type(struct inode *inode)
4951 {
4952         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4953 }
4954
4955 /*
4956  * utility function to add 'inode' into 'parent_inode' with
4957  * a give name and a given sequence number.
4958  * if 'add_backref' is true, also insert a backref from the
4959  * inode to the parent directory.
4960  */
4961 int btrfs_add_link(struct btrfs_trans_handle *trans,
4962                    struct inode *parent_inode, struct inode *inode,
4963                    const char *name, int name_len, int add_backref, u64 index)
4964 {
4965         int ret = 0;
4966         struct btrfs_key key;
4967         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4968         u64 ino = btrfs_ino(inode);
4969         u64 parent_ino = btrfs_ino(parent_inode);
4970
4971         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4972                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4973         } else {
4974                 key.objectid = ino;
4975                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4976                 key.offset = 0;
4977         }
4978
4979         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4980                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4981                                          key.objectid, root->root_key.objectid,
4982                                          parent_ino, index, name, name_len);
4983         } else if (add_backref) {
4984                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4985                                              parent_ino, index);
4986         }
4987
4988         /* Nothing to clean up yet */
4989         if (ret)
4990                 return ret;
4991
4992         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4993                                     parent_inode, &key,
4994                                     btrfs_inode_type(inode), index);
4995         if (ret == -EEXIST || ret == -EOVERFLOW)
4996                 goto fail_dir_item;
4997         else if (ret) {
4998                 btrfs_abort_transaction(trans, root, ret);
4999                 return ret;
5000         }
5001
5002         btrfs_i_size_write(parent_inode, parent_inode->i_size +
5003                            name_len * 2);
5004         inode_inc_iversion(parent_inode);
5005         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
5006         ret = btrfs_update_inode(trans, root, parent_inode);
5007         if (ret)
5008                 btrfs_abort_transaction(trans, root, ret);
5009         return ret;
5010
5011 fail_dir_item:
5012         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
5013                 u64 local_index;
5014                 int err;
5015                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
5016                                  key.objectid, root->root_key.objectid,
5017                                  parent_ino, &local_index, name, name_len);
5018
5019         } else if (add_backref) {
5020                 u64 local_index;
5021                 int err;
5022
5023                 err = btrfs_del_inode_ref(trans, root, name, name_len,
5024                                           ino, parent_ino, &local_index);
5025         }
5026         return ret;
5027 }
5028
5029 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
5030                             struct inode *dir, struct dentry *dentry,
5031                             struct inode *inode, int backref, u64 index)
5032 {
5033         int err = btrfs_add_link(trans, dir, inode,
5034                                  dentry->d_name.name, dentry->d_name.len,
5035                                  backref, index);
5036         if (err > 0)
5037                 err = -EEXIST;
5038         return err;
5039 }
5040
5041 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5042                         umode_t mode, dev_t rdev)
5043 {
5044         struct btrfs_trans_handle *trans;
5045         struct btrfs_root *root = BTRFS_I(dir)->root;
5046         struct inode *inode = NULL;
5047         int err;
5048         int drop_inode = 0;
5049         u64 objectid;
5050         u64 index = 0;
5051
5052         if (!new_valid_dev(rdev))
5053                 return -EINVAL;
5054
5055         /*
5056          * 2 for inode item and ref
5057          * 2 for dir items
5058          * 1 for xattr if selinux is on
5059          */
5060         trans = btrfs_start_transaction(root, 5);
5061         if (IS_ERR(trans))
5062                 return PTR_ERR(trans);
5063
5064         err = btrfs_find_free_ino(root, &objectid);
5065         if (err)
5066                 goto out_unlock;
5067
5068         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5069                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5070                                 mode, &index);
5071         if (IS_ERR(inode)) {
5072                 err = PTR_ERR(inode);
5073                 goto out_unlock;
5074         }
5075
5076         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5077         if (err) {
5078                 drop_inode = 1;
5079                 goto out_unlock;
5080         }
5081
5082         /*
5083         * If the active LSM wants to access the inode during
5084         * d_instantiate it needs these. Smack checks to see
5085         * if the filesystem supports xattrs by looking at the
5086         * ops vector.
5087         */
5088
5089         inode->i_op = &btrfs_special_inode_operations;
5090         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5091         if (err)
5092                 drop_inode = 1;
5093         else {
5094                 init_special_inode(inode, inode->i_mode, rdev);
5095                 btrfs_update_inode(trans, root, inode);
5096                 d_instantiate(dentry, inode);
5097         }
5098 out_unlock:
5099         btrfs_end_transaction(trans, root);
5100         btrfs_btree_balance_dirty(root);
5101         if (drop_inode) {
5102                 inode_dec_link_count(inode);
5103                 iput(inode);
5104         }
5105         return err;
5106 }
5107
5108 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5109                         umode_t mode, bool excl)
5110 {
5111         struct btrfs_trans_handle *trans;
5112         struct btrfs_root *root = BTRFS_I(dir)->root;
5113         struct inode *inode = NULL;
5114         int drop_inode_on_err = 0;
5115         int err;
5116         u64 objectid;
5117         u64 index = 0;
5118
5119         /*
5120          * 2 for inode item and ref
5121          * 2 for dir items
5122          * 1 for xattr if selinux is on
5123          */
5124         trans = btrfs_start_transaction(root, 5);
5125         if (IS_ERR(trans))
5126                 return PTR_ERR(trans);
5127
5128         err = btrfs_find_free_ino(root, &objectid);
5129         if (err)
5130                 goto out_unlock;
5131
5132         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5133                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5134                                 mode, &index);
5135         if (IS_ERR(inode)) {
5136                 err = PTR_ERR(inode);
5137                 goto out_unlock;
5138         }
5139         drop_inode_on_err = 1;
5140
5141         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5142         if (err)
5143                 goto out_unlock;
5144
5145         err = btrfs_update_inode(trans, root, inode);
5146         if (err)
5147                 goto out_unlock;
5148
5149         /*
5150         * If the active LSM wants to access the inode during
5151         * d_instantiate it needs these. Smack checks to see
5152         * if the filesystem supports xattrs by looking at the
5153         * ops vector.
5154         */
5155         inode->i_fop = &btrfs_file_operations;
5156         inode->i_op = &btrfs_file_inode_operations;
5157
5158         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5159         if (err)
5160                 goto out_unlock;
5161
5162         inode->i_mapping->a_ops = &btrfs_aops;
5163         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5164         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5165         d_instantiate(dentry, inode);
5166
5167 out_unlock:
5168         btrfs_end_transaction(trans, root);
5169         if (err && drop_inode_on_err) {
5170                 inode_dec_link_count(inode);
5171                 iput(inode);
5172         }
5173         btrfs_btree_balance_dirty(root);
5174         return err;
5175 }
5176
5177 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5178                       struct dentry *dentry)
5179 {
5180         struct btrfs_trans_handle *trans;
5181         struct btrfs_root *root = BTRFS_I(dir)->root;
5182         struct inode *inode = old_dentry->d_inode;
5183         u64 index;
5184         int err;
5185         int drop_inode = 0;
5186
5187         /* do not allow sys_link's with other subvols of the same device */
5188         if (root->objectid != BTRFS_I(inode)->root->objectid)
5189                 return -EXDEV;
5190
5191         if (inode->i_nlink >= BTRFS_LINK_MAX)
5192                 return -EMLINK;
5193
5194         err = btrfs_set_inode_index(dir, &index);
5195         if (err)
5196                 goto fail;
5197
5198         /*
5199          * 2 items for inode and inode ref
5200          * 2 items for dir items
5201          * 1 item for parent inode
5202          */
5203         trans = btrfs_start_transaction(root, 5);
5204         if (IS_ERR(trans)) {
5205                 err = PTR_ERR(trans);
5206                 goto fail;
5207         }
5208
5209         btrfs_inc_nlink(inode);
5210         inode_inc_iversion(inode);
5211         inode->i_ctime = CURRENT_TIME;
5212         ihold(inode);
5213         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5214
5215         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5216
5217         if (err) {
5218                 drop_inode = 1;
5219         } else {
5220                 struct dentry *parent = dentry->d_parent;
5221                 err = btrfs_update_inode(trans, root, inode);
5222                 if (err)
5223                         goto fail;
5224                 d_instantiate(dentry, inode);
5225                 btrfs_log_new_name(trans, inode, NULL, parent);
5226         }
5227
5228         btrfs_end_transaction(trans, root);
5229 fail:
5230         if (drop_inode) {
5231                 inode_dec_link_count(inode);
5232                 iput(inode);
5233         }
5234         btrfs_btree_balance_dirty(root);
5235         return err;
5236 }
5237
5238 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5239 {
5240         struct inode *inode = NULL;
5241         struct btrfs_trans_handle *trans;
5242         struct btrfs_root *root = BTRFS_I(dir)->root;
5243         int err = 0;
5244         int drop_on_err = 0;
5245         u64 objectid = 0;
5246         u64 index = 0;
5247
5248         /*
5249          * 2 items for inode and ref
5250          * 2 items for dir items
5251          * 1 for xattr if selinux is on
5252          */
5253         trans = btrfs_start_transaction(root, 5);
5254         if (IS_ERR(trans))
5255                 return PTR_ERR(trans);
5256
5257         err = btrfs_find_free_ino(root, &objectid);
5258         if (err)
5259                 goto out_fail;
5260
5261         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5262                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5263                                 S_IFDIR | mode, &index);
5264         if (IS_ERR(inode)) {
5265                 err = PTR_ERR(inode);
5266                 goto out_fail;
5267         }
5268
5269         drop_on_err = 1;
5270
5271         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5272         if (err)
5273                 goto out_fail;
5274
5275         inode->i_op = &btrfs_dir_inode_operations;
5276         inode->i_fop = &btrfs_dir_file_operations;
5277
5278         btrfs_i_size_write(inode, 0);
5279         err = btrfs_update_inode(trans, root, inode);
5280         if (err)
5281                 goto out_fail;
5282
5283         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5284                              dentry->d_name.len, 0, index);
5285         if (err)
5286                 goto out_fail;
5287
5288         d_instantiate(dentry, inode);
5289         drop_on_err = 0;
5290
5291 out_fail:
5292         btrfs_end_transaction(trans, root);
5293         if (drop_on_err)
5294                 iput(inode);
5295         btrfs_btree_balance_dirty(root);
5296         return err;
5297 }
5298
5299 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5300  * and an extent that you want to insert, deal with overlap and insert
5301  * the new extent into the tree.
5302  */
5303 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5304                                 struct extent_map *existing,
5305                                 struct extent_map *em,
5306                                 u64 map_start, u64 map_len)
5307 {
5308         u64 start_diff;
5309
5310         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5311         start_diff = map_start - em->start;
5312         em->start = map_start;
5313         em->len = map_len;
5314         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5315             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5316                 em->block_start += start_diff;
5317                 em->block_len -= start_diff;
5318         }
5319         return add_extent_mapping(em_tree, em);
5320 }
5321
5322 static noinline int uncompress_inline(struct btrfs_path *path,
5323                                       struct inode *inode, struct page *page,
5324                                       size_t pg_offset, u64 extent_offset,
5325                                       struct btrfs_file_extent_item *item)
5326 {
5327         int ret;
5328         struct extent_buffer *leaf = path->nodes[0];
5329         char *tmp;
5330         size_t max_size;
5331         unsigned long inline_size;
5332         unsigned long ptr;
5333         int compress_type;
5334
5335         WARN_ON(pg_offset != 0);
5336         compress_type = btrfs_file_extent_compression(leaf, item);
5337         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5338         inline_size = btrfs_file_extent_inline_item_len(leaf,
5339                                         btrfs_item_nr(leaf, path->slots[0]));
5340         tmp = kmalloc(inline_size, GFP_NOFS);
5341         if (!tmp)
5342                 return -ENOMEM;
5343         ptr = btrfs_file_extent_inline_start(item);
5344
5345         read_extent_buffer(leaf, tmp, ptr, inline_size);
5346
5347         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5348         ret = btrfs_decompress(compress_type, tmp, page,
5349                                extent_offset, inline_size, max_size);
5350         if (ret) {
5351                 char *kaddr = kmap_atomic(page);
5352                 unsigned long copy_size = min_t(u64,
5353                                   PAGE_CACHE_SIZE - pg_offset,
5354                                   max_size - extent_offset);
5355                 memset(kaddr + pg_offset, 0, copy_size);
5356                 kunmap_atomic(kaddr);
5357         }
5358         kfree(tmp);
5359         return 0;
5360 }
5361
5362 /*
5363  * a bit scary, this does extent mapping from logical file offset to the disk.
5364  * the ugly parts come from merging extents from the disk with the in-ram
5365  * representation.  This gets more complex because of the data=ordered code,
5366  * where the in-ram extents might be locked pending data=ordered completion.
5367  *
5368  * This also copies inline extents directly into the page.
5369  */
5370
5371 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5372                                     size_t pg_offset, u64 start, u64 len,
5373                                     int create)
5374 {
5375         int ret;
5376         int err = 0;
5377         u64 bytenr;
5378         u64 extent_start = 0;
5379         u64 extent_end = 0;
5380         u64 objectid = btrfs_ino(inode);
5381         u32 found_type;
5382         struct btrfs_path *path = NULL;
5383         struct btrfs_root *root = BTRFS_I(inode)->root;
5384         struct btrfs_file_extent_item *item;
5385         struct extent_buffer *leaf;
5386         struct btrfs_key found_key;
5387         struct extent_map *em = NULL;
5388         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5389         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5390         struct btrfs_trans_handle *trans = NULL;
5391         int compress_type;
5392
5393 again:
5394         read_lock(&em_tree->lock);
5395         em = lookup_extent_mapping(em_tree, start, len);
5396         if (em)
5397                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5398         read_unlock(&em_tree->lock);
5399
5400         if (em) {
5401                 if (em->start > start || em->start + em->len <= start)
5402                         free_extent_map(em);
5403                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5404                         free_extent_map(em);
5405                 else
5406                         goto out;
5407         }
5408         em = alloc_extent_map();
5409         if (!em) {
5410                 err = -ENOMEM;
5411                 goto out;
5412         }
5413         em->bdev = root->fs_info->fs_devices->latest_bdev;
5414         em->start = EXTENT_MAP_HOLE;
5415         em->orig_start = EXTENT_MAP_HOLE;
5416         em->len = (u64)-1;
5417         em->block_len = (u64)-1;
5418
5419         if (!path) {
5420                 path = btrfs_alloc_path();
5421                 if (!path) {
5422                         err = -ENOMEM;
5423                         goto out;
5424                 }
5425                 /*
5426                  * Chances are we'll be called again, so go ahead and do
5427                  * readahead
5428                  */
5429                 path->reada = 1;
5430         }
5431
5432         ret = btrfs_lookup_file_extent(trans, root, path,
5433                                        objectid, start, trans != NULL);
5434         if (ret < 0) {
5435                 err = ret;
5436                 goto out;
5437         }
5438
5439         if (ret != 0) {
5440                 if (path->slots[0] == 0)
5441                         goto not_found;
5442                 path->slots[0]--;
5443         }
5444
5445         leaf = path->nodes[0];
5446         item = btrfs_item_ptr(leaf, path->slots[0],
5447                               struct btrfs_file_extent_item);
5448         /* are we inside the extent that was found? */
5449         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5450         found_type = btrfs_key_type(&found_key);
5451         if (found_key.objectid != objectid ||
5452             found_type != BTRFS_EXTENT_DATA_KEY) {
5453                 goto not_found;
5454         }
5455
5456         found_type = btrfs_file_extent_type(leaf, item);
5457         extent_start = found_key.offset;
5458         compress_type = btrfs_file_extent_compression(leaf, item);
5459         if (found_type == BTRFS_FILE_EXTENT_REG ||
5460             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5461                 extent_end = extent_start +
5462                        btrfs_file_extent_num_bytes(leaf, item);
5463         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5464                 size_t size;
5465                 size = btrfs_file_extent_inline_len(leaf, item);
5466                 extent_end = (extent_start + size + root->sectorsize - 1) &
5467                         ~((u64)root->sectorsize - 1);
5468         }
5469
5470         if (start >= extent_end) {
5471                 path->slots[0]++;
5472                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5473                         ret = btrfs_next_leaf(root, path);
5474                         if (ret < 0) {
5475                                 err = ret;
5476                                 goto out;
5477                         }
5478                         if (ret > 0)
5479                                 goto not_found;
5480                         leaf = path->nodes[0];
5481                 }
5482                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5483                 if (found_key.objectid != objectid ||
5484                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5485                         goto not_found;
5486                 if (start + len <= found_key.offset)
5487                         goto not_found;
5488                 em->start = start;
5489                 em->orig_start = start;
5490                 em->len = found_key.offset - start;
5491                 goto not_found_em;
5492         }
5493
5494         if (found_type == BTRFS_FILE_EXTENT_REG ||
5495             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5496                 em->start = extent_start;
5497                 em->len = extent_end - extent_start;
5498                 em->orig_start = extent_start -
5499                                  btrfs_file_extent_offset(leaf, item);
5500                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5501                                                                       item);
5502                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5503                 if (bytenr == 0) {
5504                         em->block_start = EXTENT_MAP_HOLE;
5505                         goto insert;
5506                 }
5507                 if (compress_type != BTRFS_COMPRESS_NONE) {
5508                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5509                         em->compress_type = compress_type;
5510                         em->block_start = bytenr;
5511                         em->block_len = em->orig_block_len;
5512                 } else {
5513                         bytenr += btrfs_file_extent_offset(leaf, item);
5514                         em->block_start = bytenr;
5515                         em->block_len = em->len;
5516                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5517                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5518                 }
5519                 goto insert;
5520         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5521                 unsigned long ptr;
5522                 char *map;
5523                 size_t size;
5524                 size_t extent_offset;
5525                 size_t copy_size;
5526
5527                 em->block_start = EXTENT_MAP_INLINE;
5528                 if (!page || create) {
5529                         em->start = extent_start;
5530                         em->len = extent_end - extent_start;
5531                         goto out;
5532                 }
5533
5534                 size = btrfs_file_extent_inline_len(leaf, item);
5535                 extent_offset = page_offset(page) + pg_offset - extent_start;
5536                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5537                                 size - extent_offset);
5538                 em->start = extent_start + extent_offset;
5539                 em->len = (copy_size + root->sectorsize - 1) &
5540                         ~((u64)root->sectorsize - 1);
5541                 em->orig_block_len = em->len;
5542                 em->orig_start = em->start;
5543                 if (compress_type) {
5544                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5545                         em->compress_type = compress_type;
5546                 }
5547                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5548                 if (create == 0 && !PageUptodate(page)) {
5549                         if (btrfs_file_extent_compression(leaf, item) !=
5550                             BTRFS_COMPRESS_NONE) {
5551                                 ret = uncompress_inline(path, inode, page,
5552                                                         pg_offset,
5553                                                         extent_offset, item);
5554                                 BUG_ON(ret); /* -ENOMEM */
5555                         } else {
5556                                 map = kmap(page);
5557                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5558                                                    copy_size);
5559                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5560                                         memset(map + pg_offset + copy_size, 0,
5561                                                PAGE_CACHE_SIZE - pg_offset -
5562                                                copy_size);
5563                                 }
5564                                 kunmap(page);
5565                         }
5566                         flush_dcache_page(page);
5567                 } else if (create && PageUptodate(page)) {
5568                         BUG();
5569                         if (!trans) {
5570                                 kunmap(page);
5571                                 free_extent_map(em);
5572                                 em = NULL;
5573
5574                                 btrfs_release_path(path);
5575                                 trans = btrfs_join_transaction(root);
5576
5577                                 if (IS_ERR(trans))
5578                                         return ERR_CAST(trans);
5579                                 goto again;
5580                         }
5581                         map = kmap(page);
5582                         write_extent_buffer(leaf, map + pg_offset, ptr,
5583                                             copy_size);
5584                         kunmap(page);
5585                         btrfs_mark_buffer_dirty(leaf);
5586                 }
5587                 set_extent_uptodate(io_tree, em->start,
5588                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5589                 goto insert;
5590         } else {
5591                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5592         }
5593 not_found:
5594         em->start = start;
5595         em->orig_start = start;
5596         em->len = len;
5597 not_found_em:
5598         em->block_start = EXTENT_MAP_HOLE;
5599         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5600 insert:
5601         btrfs_release_path(path);
5602         if (em->start > start || extent_map_end(em) <= start) {
5603                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5604                        "[%llu %llu]\n", (unsigned long long)em->start,
5605                        (unsigned long long)em->len,
5606                        (unsigned long long)start,
5607                        (unsigned long long)len);
5608                 err = -EIO;
5609                 goto out;
5610         }
5611
5612         err = 0;
5613         write_lock(&em_tree->lock);
5614         ret = add_extent_mapping(em_tree, em);
5615         /* it is possible that someone inserted the extent into the tree
5616          * while we had the lock dropped.  It is also possible that
5617          * an overlapping map exists in the tree
5618          */
5619         if (ret == -EEXIST) {
5620                 struct extent_map *existing;
5621
5622                 ret = 0;
5623
5624                 existing = lookup_extent_mapping(em_tree, start, len);
5625                 if (existing && (existing->start > start ||
5626                     existing->start + existing->len <= start)) {
5627                         free_extent_map(existing);
5628                         existing = NULL;
5629                 }
5630                 if (!existing) {
5631                         existing = lookup_extent_mapping(em_tree, em->start,
5632                                                          em->len);
5633                         if (existing) {
5634                                 err = merge_extent_mapping(em_tree, existing,
5635                                                            em, start,
5636                                                            root->sectorsize);
5637                                 free_extent_map(existing);
5638                                 if (err) {
5639                                         free_extent_map(em);
5640                                         em = NULL;
5641                                 }
5642                         } else {
5643                                 err = -EIO;
5644                                 free_extent_map(em);
5645                                 em = NULL;
5646                         }
5647                 } else {
5648                         free_extent_map(em);
5649                         em = existing;
5650                         err = 0;
5651                 }
5652         }
5653         write_unlock(&em_tree->lock);
5654 out:
5655
5656         if (em)
5657                 trace_btrfs_get_extent(root, em);
5658
5659         if (path)
5660                 btrfs_free_path(path);
5661         if (trans) {
5662                 ret = btrfs_end_transaction(trans, root);
5663                 if (!err)
5664                         err = ret;
5665         }
5666         if (err) {
5667                 free_extent_map(em);
5668                 return ERR_PTR(err);
5669         }
5670         BUG_ON(!em); /* Error is always set */
5671         return em;
5672 }
5673
5674 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5675                                            size_t pg_offset, u64 start, u64 len,
5676                                            int create)
5677 {
5678         struct extent_map *em;
5679         struct extent_map *hole_em = NULL;
5680         u64 range_start = start;
5681         u64 end;
5682         u64 found;
5683         u64 found_end;
5684         int err = 0;
5685
5686         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5687         if (IS_ERR(em))
5688                 return em;
5689         if (em) {
5690                 /*
5691                  * if our em maps to
5692                  * -  a hole or
5693                  * -  a pre-alloc extent,
5694                  * there might actually be delalloc bytes behind it.
5695                  */
5696                 if (em->block_start != EXTENT_MAP_HOLE &&
5697                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5698                         return em;
5699                 else
5700                         hole_em = em;
5701         }
5702
5703         /* check to see if we've wrapped (len == -1 or similar) */
5704         end = start + len;
5705         if (end < start)
5706                 end = (u64)-1;
5707         else
5708                 end -= 1;
5709
5710         em = NULL;
5711
5712         /* ok, we didn't find anything, lets look for delalloc */
5713         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5714                                  end, len, EXTENT_DELALLOC, 1);
5715         found_end = range_start + found;
5716         if (found_end < range_start)
5717                 found_end = (u64)-1;
5718
5719         /*
5720          * we didn't find anything useful, return
5721          * the original results from get_extent()
5722          */
5723         if (range_start > end || found_end <= start) {
5724                 em = hole_em;
5725                 hole_em = NULL;
5726                 goto out;
5727         }
5728
5729         /* adjust the range_start to make sure it doesn't
5730          * go backwards from the start they passed in
5731          */
5732         range_start = max(start,range_start);
5733         found = found_end - range_start;
5734
5735         if (found > 0) {
5736                 u64 hole_start = start;
5737                 u64 hole_len = len;
5738
5739                 em = alloc_extent_map();
5740                 if (!em) {
5741                         err = -ENOMEM;
5742                         goto out;
5743                 }
5744                 /*
5745                  * when btrfs_get_extent can't find anything it
5746                  * returns one huge hole
5747                  *
5748                  * make sure what it found really fits our range, and
5749                  * adjust to make sure it is based on the start from
5750                  * the caller
5751                  */
5752                 if (hole_em) {
5753                         u64 calc_end = extent_map_end(hole_em);
5754
5755                         if (calc_end <= start || (hole_em->start > end)) {
5756                                 free_extent_map(hole_em);
5757                                 hole_em = NULL;
5758                         } else {
5759                                 hole_start = max(hole_em->start, start);
5760                                 hole_len = calc_end - hole_start;
5761                         }
5762                 }
5763                 em->bdev = NULL;
5764                 if (hole_em && range_start > hole_start) {
5765                         /* our hole starts before our delalloc, so we
5766                          * have to return just the parts of the hole
5767                          * that go until  the delalloc starts
5768                          */
5769                         em->len = min(hole_len,
5770                                       range_start - hole_start);
5771                         em->start = hole_start;
5772                         em->orig_start = hole_start;
5773                         /*
5774                          * don't adjust block start at all,
5775                          * it is fixed at EXTENT_MAP_HOLE
5776                          */
5777                         em->block_start = hole_em->block_start;
5778                         em->block_len = hole_len;
5779                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5780                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5781                 } else {
5782                         em->start = range_start;
5783                         em->len = found;
5784                         em->orig_start = range_start;
5785                         em->block_start = EXTENT_MAP_DELALLOC;
5786                         em->block_len = found;
5787                 }
5788         } else if (hole_em) {
5789                 return hole_em;
5790         }
5791 out:
5792
5793         free_extent_map(hole_em);
5794         if (err) {
5795                 free_extent_map(em);
5796                 return ERR_PTR(err);
5797         }
5798         return em;
5799 }
5800
5801 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5802                                                   u64 start, u64 len)
5803 {
5804         struct btrfs_root *root = BTRFS_I(inode)->root;
5805         struct btrfs_trans_handle *trans;
5806         struct extent_map *em;
5807         struct btrfs_key ins;
5808         u64 alloc_hint;
5809         int ret;
5810
5811         trans = btrfs_join_transaction(root);
5812         if (IS_ERR(trans))
5813                 return ERR_CAST(trans);
5814
5815         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5816
5817         alloc_hint = get_extent_allocation_hint(inode, start, len);
5818         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5819                                    alloc_hint, &ins, 1);
5820         if (ret) {
5821                 em = ERR_PTR(ret);
5822                 goto out;
5823         }
5824
5825         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5826                               ins.offset, ins.offset, 0);
5827         if (IS_ERR(em))
5828                 goto out;
5829
5830         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5831                                            ins.offset, ins.offset, 0);
5832         if (ret) {
5833                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5834                 em = ERR_PTR(ret);
5835         }
5836 out:
5837         btrfs_end_transaction(trans, root);
5838         return em;
5839 }
5840
5841 /*
5842  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5843  * block must be cow'd
5844  */
5845 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5846                                       struct inode *inode, u64 offset, u64 len)
5847 {
5848         struct btrfs_path *path;
5849         int ret;
5850         struct extent_buffer *leaf;
5851         struct btrfs_root *root = BTRFS_I(inode)->root;
5852         struct btrfs_file_extent_item *fi;
5853         struct btrfs_key key;
5854         u64 disk_bytenr;
5855         u64 backref_offset;
5856         u64 extent_end;
5857         u64 num_bytes;
5858         int slot;
5859         int found_type;
5860
5861         path = btrfs_alloc_path();
5862         if (!path)
5863                 return -ENOMEM;
5864
5865         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5866                                        offset, 0);
5867         if (ret < 0)
5868                 goto out;
5869
5870         slot = path->slots[0];
5871         if (ret == 1) {
5872                 if (slot == 0) {
5873                         /* can't find the item, must cow */
5874                         ret = 0;
5875                         goto out;
5876                 }
5877                 slot--;
5878         }
5879         ret = 0;
5880         leaf = path->nodes[0];
5881         btrfs_item_key_to_cpu(leaf, &key, slot);
5882         if (key.objectid != btrfs_ino(inode) ||
5883             key.type != BTRFS_EXTENT_DATA_KEY) {
5884                 /* not our file or wrong item type, must cow */
5885                 goto out;
5886         }
5887
5888         if (key.offset > offset) {
5889                 /* Wrong offset, must cow */
5890                 goto out;
5891         }
5892
5893         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5894         found_type = btrfs_file_extent_type(leaf, fi);
5895         if (found_type != BTRFS_FILE_EXTENT_REG &&
5896             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5897                 /* not a regular extent, must cow */
5898                 goto out;
5899         }
5900         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5901         backref_offset = btrfs_file_extent_offset(leaf, fi);
5902
5903         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5904         if (extent_end < offset + len) {
5905                 /* extent doesn't include our full range, must cow */
5906                 goto out;
5907         }
5908
5909         if (btrfs_extent_readonly(root, disk_bytenr))
5910                 goto out;
5911
5912         /*
5913          * look for other files referencing this extent, if we
5914          * find any we must cow
5915          */
5916         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5917                                   key.offset - backref_offset, disk_bytenr))
5918                 goto out;
5919
5920         /*
5921          * adjust disk_bytenr and num_bytes to cover just the bytes
5922          * in this extent we are about to write.  If there
5923          * are any csums in that range we have to cow in order
5924          * to keep the csums correct
5925          */
5926         disk_bytenr += backref_offset;
5927         disk_bytenr += offset - key.offset;
5928         num_bytes = min(offset + len, extent_end) - offset;
5929         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5930                                 goto out;
5931         /*
5932          * all of the above have passed, it is safe to overwrite this extent
5933          * without cow
5934          */
5935         ret = 1;
5936 out:
5937         btrfs_free_path(path);
5938         return ret;
5939 }
5940
5941 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5942                               struct extent_state **cached_state, int writing)
5943 {
5944         struct btrfs_ordered_extent *ordered;
5945         int ret = 0;
5946
5947         while (1) {
5948                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5949                                  0, cached_state);
5950                 /*
5951                  * We're concerned with the entire range that we're going to be
5952                  * doing DIO to, so we need to make sure theres no ordered
5953                  * extents in this range.
5954                  */
5955                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5956                                                      lockend - lockstart + 1);
5957
5958                 /*
5959                  * We need to make sure there are no buffered pages in this
5960                  * range either, we could have raced between the invalidate in
5961                  * generic_file_direct_write and locking the extent.  The
5962                  * invalidate needs to happen so that reads after a write do not
5963                  * get stale data.
5964                  */
5965                 if (!ordered && (!writing ||
5966                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5967                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5968                                     *cached_state)))
5969                         break;
5970
5971                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5972                                      cached_state, GFP_NOFS);
5973
5974                 if (ordered) {
5975                         btrfs_start_ordered_extent(inode, ordered, 1);
5976                         btrfs_put_ordered_extent(ordered);
5977                 } else {
5978                         /* Screw you mmap */
5979                         ret = filemap_write_and_wait_range(inode->i_mapping,
5980                                                            lockstart,
5981                                                            lockend);
5982                         if (ret)
5983                                 break;
5984
5985                         /*
5986                          * If we found a page that couldn't be invalidated just
5987                          * fall back to buffered.
5988                          */
5989                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5990                                         lockstart >> PAGE_CACHE_SHIFT,
5991                                         lockend >> PAGE_CACHE_SHIFT);
5992                         if (ret)
5993                                 break;
5994                 }
5995
5996                 cond_resched();
5997         }
5998
5999         return ret;
6000 }
6001
6002 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
6003                                            u64 len, u64 orig_start,
6004                                            u64 block_start, u64 block_len,
6005                                            u64 orig_block_len, int type)
6006 {
6007         struct extent_map_tree *em_tree;
6008         struct extent_map *em;
6009         struct btrfs_root *root = BTRFS_I(inode)->root;
6010         int ret;
6011
6012         em_tree = &BTRFS_I(inode)->extent_tree;
6013         em = alloc_extent_map();
6014         if (!em)
6015                 return ERR_PTR(-ENOMEM);
6016
6017         em->start = start;
6018         em->orig_start = orig_start;
6019         em->mod_start = start;
6020         em->mod_len = len;
6021         em->len = len;
6022         em->block_len = block_len;
6023         em->block_start = block_start;
6024         em->bdev = root->fs_info->fs_devices->latest_bdev;
6025         em->orig_block_len = orig_block_len;
6026         em->generation = -1;
6027         set_bit(EXTENT_FLAG_PINNED, &em->flags);
6028         if (type == BTRFS_ORDERED_PREALLOC)
6029                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
6030
6031         do {
6032                 btrfs_drop_extent_cache(inode, em->start,
6033                                 em->start + em->len - 1, 0);
6034                 write_lock(&em_tree->lock);
6035                 ret = add_extent_mapping(em_tree, em);
6036                 if (!ret)
6037                         list_move(&em->list,
6038                                   &em_tree->modified_extents);
6039                 write_unlock(&em_tree->lock);
6040         } while (ret == -EEXIST);
6041
6042         if (ret) {
6043                 free_extent_map(em);
6044                 return ERR_PTR(ret);
6045         }
6046
6047         return em;
6048 }
6049
6050
6051 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6052                                    struct buffer_head *bh_result, int create)
6053 {
6054         struct extent_map *em;
6055         struct btrfs_root *root = BTRFS_I(inode)->root;
6056         struct extent_state *cached_state = NULL;
6057         u64 start = iblock << inode->i_blkbits;
6058         u64 lockstart, lockend;
6059         u64 len = bh_result->b_size;
6060         struct btrfs_trans_handle *trans;
6061         int unlock_bits = EXTENT_LOCKED;
6062         int ret;
6063
6064         if (create) {
6065                 ret = btrfs_delalloc_reserve_space(inode, len);
6066                 if (ret)
6067                         return ret;
6068                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6069         } else {
6070                 len = min_t(u64, len, root->sectorsize);
6071         }
6072
6073         lockstart = start;
6074         lockend = start + len - 1;
6075
6076         /*
6077          * If this errors out it's because we couldn't invalidate pagecache for
6078          * this range and we need to fallback to buffered.
6079          */
6080         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6081                 return -ENOTBLK;
6082
6083         if (create) {
6084                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6085                                      lockend, EXTENT_DELALLOC, NULL,
6086                                      &cached_state, GFP_NOFS);
6087                 if (ret)
6088                         goto unlock_err;
6089         }
6090
6091         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6092         if (IS_ERR(em)) {
6093                 ret = PTR_ERR(em);
6094                 goto unlock_err;
6095         }
6096
6097         /*
6098          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6099          * io.  INLINE is special, and we could probably kludge it in here, but
6100          * it's still buffered so for safety lets just fall back to the generic
6101          * buffered path.
6102          *
6103          * For COMPRESSED we _have_ to read the entire extent in so we can
6104          * decompress it, so there will be buffering required no matter what we
6105          * do, so go ahead and fallback to buffered.
6106          *
6107          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6108          * to buffered IO.  Don't blame me, this is the price we pay for using
6109          * the generic code.
6110          */
6111         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6112             em->block_start == EXTENT_MAP_INLINE) {
6113                 free_extent_map(em);
6114                 ret = -ENOTBLK;
6115                 goto unlock_err;
6116         }
6117
6118         /* Just a good old fashioned hole, return */
6119         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6120                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6121                 free_extent_map(em);
6122                 ret = 0;
6123                 goto unlock_err;
6124         }
6125
6126         /*
6127          * We don't allocate a new extent in the following cases
6128          *
6129          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6130          * existing extent.
6131          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6132          * just use the extent.
6133          *
6134          */
6135         if (!create) {
6136                 len = min(len, em->len - (start - em->start));
6137                 lockstart = start + len;
6138                 goto unlock;
6139         }
6140
6141         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6142             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6143              em->block_start != EXTENT_MAP_HOLE)) {
6144                 int type;
6145                 int ret;
6146                 u64 block_start;
6147
6148                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6149                         type = BTRFS_ORDERED_PREALLOC;
6150                 else
6151                         type = BTRFS_ORDERED_NOCOW;
6152                 len = min(len, em->len - (start - em->start));
6153                 block_start = em->block_start + (start - em->start);
6154
6155                 /*
6156                  * we're not going to log anything, but we do need
6157                  * to make sure the current transaction stays open
6158                  * while we look for nocow cross refs
6159                  */
6160                 trans = btrfs_join_transaction(root);
6161                 if (IS_ERR(trans))
6162                         goto must_cow;
6163
6164                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6165                         u64 orig_start = em->orig_start;
6166                         u64 orig_block_len = em->orig_block_len;
6167
6168                         if (type == BTRFS_ORDERED_PREALLOC) {
6169                                 free_extent_map(em);
6170                                 em = create_pinned_em(inode, start, len,
6171                                                        orig_start,
6172                                                        block_start, len,
6173                                                        orig_block_len, type);
6174                                 if (IS_ERR(em)) {
6175                                         btrfs_end_transaction(trans, root);
6176                                         goto unlock_err;
6177                                 }
6178                         }
6179
6180                         ret = btrfs_add_ordered_extent_dio(inode, start,
6181                                            block_start, len, len, type);
6182                         btrfs_end_transaction(trans, root);
6183                         if (ret) {
6184                                 free_extent_map(em);
6185                                 goto unlock_err;
6186                         }
6187                         goto unlock;
6188                 }
6189                 btrfs_end_transaction(trans, root);
6190         }
6191 must_cow:
6192         /*
6193          * this will cow the extent, reset the len in case we changed
6194          * it above
6195          */
6196         len = bh_result->b_size;
6197         free_extent_map(em);
6198         em = btrfs_new_extent_direct(inode, start, len);
6199         if (IS_ERR(em)) {
6200                 ret = PTR_ERR(em);
6201                 goto unlock_err;
6202         }
6203         len = min(len, em->len - (start - em->start));
6204 unlock:
6205         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6206                 inode->i_blkbits;
6207         bh_result->b_size = len;
6208         bh_result->b_bdev = em->bdev;
6209         set_buffer_mapped(bh_result);
6210         if (create) {
6211                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6212                         set_buffer_new(bh_result);
6213
6214                 /*
6215                  * Need to update the i_size under the extent lock so buffered
6216                  * readers will get the updated i_size when we unlock.
6217                  */
6218                 if (start + len > i_size_read(inode))
6219                         i_size_write(inode, start + len);
6220         }
6221
6222         /*
6223          * In the case of write we need to clear and unlock the entire range,
6224          * in the case of read we need to unlock only the end area that we
6225          * aren't using if there is any left over space.
6226          */
6227         if (lockstart < lockend) {
6228                 if (create && len < lockend - lockstart) {
6229                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6230                                          lockstart + len - 1,
6231                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6232                                          &cached_state, GFP_NOFS);
6233                         /*
6234                          * Beside unlock, we also need to cleanup reserved space
6235                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6236                          */
6237                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6238                                          lockstart + len, lockend,
6239                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6240                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6241                 } else {
6242                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6243                                          lockend, unlock_bits, 1, 0,
6244                                          &cached_state, GFP_NOFS);
6245                 }
6246         } else {
6247                 free_extent_state(cached_state);
6248         }
6249
6250         free_extent_map(em);
6251
6252         return 0;
6253
6254 unlock_err:
6255         if (create)
6256                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6257
6258         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6259                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6260         return ret;
6261 }
6262
6263 struct btrfs_dio_private {
6264         struct inode *inode;
6265         u64 logical_offset;
6266         u64 disk_bytenr;
6267         u64 bytes;
6268         void *private;
6269
6270         /* number of bios pending for this dio */
6271         atomic_t pending_bios;
6272
6273         /* IO errors */
6274         int errors;
6275
6276         struct bio *orig_bio;
6277 };
6278
6279 static void btrfs_endio_direct_read(struct bio *bio, int err)
6280 {
6281         struct btrfs_dio_private *dip = bio->bi_private;
6282         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6283         struct bio_vec *bvec = bio->bi_io_vec;
6284         struct inode *inode = dip->inode;
6285         struct btrfs_root *root = BTRFS_I(inode)->root;
6286         u64 start;
6287
6288         start = dip->logical_offset;
6289         do {
6290                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6291                         struct page *page = bvec->bv_page;
6292                         char *kaddr;
6293                         u32 csum = ~(u32)0;
6294                         u64 private = ~(u32)0;
6295                         unsigned long flags;
6296
6297                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6298                                               start, &private))
6299                                 goto failed;
6300                         local_irq_save(flags);
6301                         kaddr = kmap_atomic(page);
6302                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6303                                                csum, bvec->bv_len);
6304                         btrfs_csum_final(csum, (char *)&csum);
6305                         kunmap_atomic(kaddr);
6306                         local_irq_restore(flags);
6307
6308                         flush_dcache_page(bvec->bv_page);
6309                         if (csum != private) {
6310 failed:
6311                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6312                                       " %llu csum %u private %u\n",
6313                                       (unsigned long long)btrfs_ino(inode),
6314                                       (unsigned long long)start,
6315                                       csum, (unsigned)private);
6316                                 err = -EIO;
6317                         }
6318                 }
6319
6320                 start += bvec->bv_len;
6321                 bvec++;
6322         } while (bvec <= bvec_end);
6323
6324         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6325                       dip->logical_offset + dip->bytes - 1);
6326         bio->bi_private = dip->private;
6327
6328         kfree(dip);
6329
6330         /* If we had a csum failure make sure to clear the uptodate flag */
6331         if (err)
6332                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6333         dio_end_io(bio, err);
6334 }
6335
6336 static void btrfs_endio_direct_write(struct bio *bio, int err)
6337 {
6338         struct btrfs_dio_private *dip = bio->bi_private;
6339         struct inode *inode = dip->inode;
6340         struct btrfs_root *root = BTRFS_I(inode)->root;
6341         struct btrfs_ordered_extent *ordered = NULL;
6342         u64 ordered_offset = dip->logical_offset;
6343         u64 ordered_bytes = dip->bytes;
6344         int ret;
6345
6346         if (err)
6347                 goto out_done;
6348 again:
6349         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6350                                                    &ordered_offset,
6351                                                    ordered_bytes, !err);
6352         if (!ret)
6353                 goto out_test;
6354
6355         ordered->work.func = finish_ordered_fn;
6356         ordered->work.flags = 0;
6357         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6358                            &ordered->work);
6359 out_test:
6360         /*
6361          * our bio might span multiple ordered extents.  If we haven't
6362          * completed the accounting for the whole dio, go back and try again
6363          */
6364         if (ordered_offset < dip->logical_offset + dip->bytes) {
6365                 ordered_bytes = dip->logical_offset + dip->bytes -
6366                         ordered_offset;
6367                 ordered = NULL;
6368                 goto again;
6369         }
6370 out_done:
6371         bio->bi_private = dip->private;
6372
6373         kfree(dip);
6374
6375         /* If we had an error make sure to clear the uptodate flag */
6376         if (err)
6377                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6378         dio_end_io(bio, err);
6379 }
6380
6381 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6382                                     struct bio *bio, int mirror_num,
6383                                     unsigned long bio_flags, u64 offset)
6384 {
6385         int ret;
6386         struct btrfs_root *root = BTRFS_I(inode)->root;
6387         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6388         BUG_ON(ret); /* -ENOMEM */
6389         return 0;
6390 }
6391
6392 static void btrfs_end_dio_bio(struct bio *bio, int err)
6393 {
6394         struct btrfs_dio_private *dip = bio->bi_private;
6395
6396         if (err) {
6397                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6398                       "sector %#Lx len %u err no %d\n",
6399                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6400                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6401                 dip->errors = 1;
6402
6403                 /*
6404                  * before atomic variable goto zero, we must make sure
6405                  * dip->errors is perceived to be set.
6406                  */
6407                 smp_mb__before_atomic_dec();
6408         }
6409
6410         /* if there are more bios still pending for this dio, just exit */
6411         if (!atomic_dec_and_test(&dip->pending_bios))
6412                 goto out;
6413
6414         if (dip->errors)
6415                 bio_io_error(dip->orig_bio);
6416         else {
6417                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6418                 bio_endio(dip->orig_bio, 0);
6419         }
6420 out:
6421         bio_put(bio);
6422 }
6423
6424 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6425                                        u64 first_sector, gfp_t gfp_flags)
6426 {
6427         int nr_vecs = bio_get_nr_vecs(bdev);
6428         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6429 }
6430
6431 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6432                                          int rw, u64 file_offset, int skip_sum,
6433                                          int async_submit)
6434 {
6435         int write = rw & REQ_WRITE;
6436         struct btrfs_root *root = BTRFS_I(inode)->root;
6437         int ret;
6438
6439         if (async_submit)
6440                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6441
6442         bio_get(bio);
6443
6444         if (!write) {
6445                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6446                 if (ret)
6447                         goto err;
6448         }
6449
6450         if (skip_sum)
6451                 goto map;
6452
6453         if (write && async_submit) {
6454                 ret = btrfs_wq_submit_bio(root->fs_info,
6455                                    inode, rw, bio, 0, 0,
6456                                    file_offset,
6457                                    __btrfs_submit_bio_start_direct_io,
6458                                    __btrfs_submit_bio_done);
6459                 goto err;
6460         } else if (write) {
6461                 /*
6462                  * If we aren't doing async submit, calculate the csum of the
6463                  * bio now.
6464                  */
6465                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6466                 if (ret)
6467                         goto err;
6468         } else if (!skip_sum) {
6469                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6470                 if (ret)
6471                         goto err;
6472         }
6473
6474 map:
6475         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6476 err:
6477         bio_put(bio);
6478         return ret;
6479 }
6480
6481 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6482                                     int skip_sum)
6483 {
6484         struct inode *inode = dip->inode;
6485         struct btrfs_root *root = BTRFS_I(inode)->root;
6486         struct bio *bio;
6487         struct bio *orig_bio = dip->orig_bio;
6488         struct bio_vec *bvec = orig_bio->bi_io_vec;
6489         u64 start_sector = orig_bio->bi_sector;
6490         u64 file_offset = dip->logical_offset;
6491         u64 submit_len = 0;
6492         u64 map_length;
6493         int nr_pages = 0;
6494         int ret = 0;
6495         int async_submit = 0;
6496
6497         map_length = orig_bio->bi_size;
6498         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6499                               &map_length, NULL, 0);
6500         if (ret) {
6501                 bio_put(orig_bio);
6502                 return -EIO;
6503         }
6504
6505         if (map_length >= orig_bio->bi_size) {
6506                 bio = orig_bio;
6507                 goto submit;
6508         }
6509
6510         async_submit = 1;
6511         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6512         if (!bio)
6513                 return -ENOMEM;
6514         bio->bi_private = dip;
6515         bio->bi_end_io = btrfs_end_dio_bio;
6516         atomic_inc(&dip->pending_bios);
6517
6518         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6519                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6520                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6521                                  bvec->bv_offset) < bvec->bv_len)) {
6522                         /*
6523                          * inc the count before we submit the bio so
6524                          * we know the end IO handler won't happen before
6525                          * we inc the count. Otherwise, the dip might get freed
6526                          * before we're done setting it up
6527                          */
6528                         atomic_inc(&dip->pending_bios);
6529                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6530                                                      file_offset, skip_sum,
6531                                                      async_submit);
6532                         if (ret) {
6533                                 bio_put(bio);
6534                                 atomic_dec(&dip->pending_bios);
6535                                 goto out_err;
6536                         }
6537
6538                         start_sector += submit_len >> 9;
6539                         file_offset += submit_len;
6540
6541                         submit_len = 0;
6542                         nr_pages = 0;
6543
6544                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6545                                                   start_sector, GFP_NOFS);
6546                         if (!bio)
6547                                 goto out_err;
6548                         bio->bi_private = dip;
6549                         bio->bi_end_io = btrfs_end_dio_bio;
6550
6551                         map_length = orig_bio->bi_size;
6552                         ret = btrfs_map_block(root->fs_info, READ,
6553                                               start_sector << 9,
6554                                               &map_length, NULL, 0);
6555                         if (ret) {
6556                                 bio_put(bio);
6557                                 goto out_err;
6558                         }
6559                 } else {
6560                         submit_len += bvec->bv_len;
6561                         nr_pages ++;
6562                         bvec++;
6563                 }
6564         }
6565
6566 submit:
6567         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6568                                      async_submit);
6569         if (!ret)
6570                 return 0;
6571
6572         bio_put(bio);
6573 out_err:
6574         dip->errors = 1;
6575         /*
6576          * before atomic variable goto zero, we must
6577          * make sure dip->errors is perceived to be set.
6578          */
6579         smp_mb__before_atomic_dec();
6580         if (atomic_dec_and_test(&dip->pending_bios))
6581                 bio_io_error(dip->orig_bio);
6582
6583         /* bio_end_io() will handle error, so we needn't return it */
6584         return 0;
6585 }
6586
6587 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6588                                 loff_t file_offset)
6589 {
6590         struct btrfs_root *root = BTRFS_I(inode)->root;
6591         struct btrfs_dio_private *dip;
6592         struct bio_vec *bvec = bio->bi_io_vec;
6593         int skip_sum;
6594         int write = rw & REQ_WRITE;
6595         int ret = 0;
6596
6597         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6598
6599         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6600         if (!dip) {
6601                 ret = -ENOMEM;
6602                 goto free_ordered;
6603         }
6604
6605         dip->private = bio->bi_private;
6606         dip->inode = inode;
6607         dip->logical_offset = file_offset;
6608
6609         dip->bytes = 0;
6610         do {
6611                 dip->bytes += bvec->bv_len;
6612                 bvec++;
6613         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6614
6615         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6616         bio->bi_private = dip;
6617         dip->errors = 0;
6618         dip->orig_bio = bio;
6619         atomic_set(&dip->pending_bios, 0);
6620
6621         if (write)
6622                 bio->bi_end_io = btrfs_endio_direct_write;
6623         else
6624                 bio->bi_end_io = btrfs_endio_direct_read;
6625
6626         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6627         if (!ret)
6628                 return;
6629 free_ordered:
6630         /*
6631          * If this is a write, we need to clean up the reserved space and kill
6632          * the ordered extent.
6633          */
6634         if (write) {
6635                 struct btrfs_ordered_extent *ordered;
6636                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6637                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6638                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6639                         btrfs_free_reserved_extent(root, ordered->start,
6640                                                    ordered->disk_len);
6641                 btrfs_put_ordered_extent(ordered);
6642                 btrfs_put_ordered_extent(ordered);
6643         }
6644         bio_endio(bio, ret);
6645 }
6646
6647 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6648                         const struct iovec *iov, loff_t offset,
6649                         unsigned long nr_segs)
6650 {
6651         int seg;
6652         int i;
6653         size_t size;
6654         unsigned long addr;
6655         unsigned blocksize_mask = root->sectorsize - 1;
6656         ssize_t retval = -EINVAL;
6657         loff_t end = offset;
6658
6659         if (offset & blocksize_mask)
6660                 goto out;
6661
6662         /* Check the memory alignment.  Blocks cannot straddle pages */
6663         for (seg = 0; seg < nr_segs; seg++) {
6664                 addr = (unsigned long)iov[seg].iov_base;
6665                 size = iov[seg].iov_len;
6666                 end += size;
6667                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6668                         goto out;
6669
6670                 /* If this is a write we don't need to check anymore */
6671                 if (rw & WRITE)
6672                         continue;
6673
6674                 /*
6675                  * Check to make sure we don't have duplicate iov_base's in this
6676                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6677                  * when reading back.
6678                  */
6679                 for (i = seg + 1; i < nr_segs; i++) {
6680                         if (iov[seg].iov_base == iov[i].iov_base)
6681                                 goto out;
6682                 }
6683         }
6684         retval = 0;
6685 out:
6686         return retval;
6687 }
6688
6689 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6690                         const struct iovec *iov, loff_t offset,
6691                         unsigned long nr_segs)
6692 {
6693         struct file *file = iocb->ki_filp;
6694         struct inode *inode = file->f_mapping->host;
6695
6696         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6697                             offset, nr_segs))
6698                 return 0;
6699
6700         return __blockdev_direct_IO(rw, iocb, inode,
6701                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6702                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6703                    btrfs_submit_direct, 0);
6704 }
6705
6706 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6707
6708 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6709                 __u64 start, __u64 len)
6710 {
6711         int     ret;
6712
6713         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6714         if (ret)
6715                 return ret;
6716
6717         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6718 }
6719
6720 int btrfs_readpage(struct file *file, struct page *page)
6721 {
6722         struct extent_io_tree *tree;
6723         tree = &BTRFS_I(page->mapping->host)->io_tree;
6724         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6725 }
6726
6727 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6728 {
6729         struct extent_io_tree *tree;
6730
6731
6732         if (current->flags & PF_MEMALLOC) {
6733                 redirty_page_for_writepage(wbc, page);
6734                 unlock_page(page);
6735                 return 0;
6736         }
6737         tree = &BTRFS_I(page->mapping->host)->io_tree;
6738         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6739 }
6740
6741 int btrfs_writepages(struct address_space *mapping,
6742                      struct writeback_control *wbc)
6743 {
6744         struct extent_io_tree *tree;
6745
6746         tree = &BTRFS_I(mapping->host)->io_tree;
6747         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6748 }
6749
6750 static int
6751 btrfs_readpages(struct file *file, struct address_space *mapping,
6752                 struct list_head *pages, unsigned nr_pages)
6753 {
6754         struct extent_io_tree *tree;
6755         tree = &BTRFS_I(mapping->host)->io_tree;
6756         return extent_readpages(tree, mapping, pages, nr_pages,
6757                                 btrfs_get_extent);
6758 }
6759 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6760 {
6761         struct extent_io_tree *tree;
6762         struct extent_map_tree *map;
6763         int ret;
6764
6765         tree = &BTRFS_I(page->mapping->host)->io_tree;
6766         map = &BTRFS_I(page->mapping->host)->extent_tree;
6767         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6768         if (ret == 1) {
6769                 ClearPagePrivate(page);
6770                 set_page_private(page, 0);
6771                 page_cache_release(page);
6772         }
6773         return ret;
6774 }
6775
6776 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6777 {
6778         if (PageWriteback(page) || PageDirty(page))
6779                 return 0;
6780         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6781 }
6782
6783 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6784 {
6785         struct inode *inode = page->mapping->host;
6786         struct extent_io_tree *tree;
6787         struct btrfs_ordered_extent *ordered;
6788         struct extent_state *cached_state = NULL;
6789         u64 page_start = page_offset(page);
6790         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6791
6792         /*
6793          * we have the page locked, so new writeback can't start,
6794          * and the dirty bit won't be cleared while we are here.
6795          *
6796          * Wait for IO on this page so that we can safely clear
6797          * the PagePrivate2 bit and do ordered accounting
6798          */
6799         wait_on_page_writeback(page);
6800
6801         tree = &BTRFS_I(inode)->io_tree;
6802         if (offset) {
6803                 btrfs_releasepage(page, GFP_NOFS);
6804                 return;
6805         }
6806         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6807         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
6808         if (ordered) {
6809                 /*
6810                  * IO on this page will never be started, so we need
6811                  * to account for any ordered extents now
6812                  */
6813                 clear_extent_bit(tree, page_start, page_end,
6814                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6815                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6816                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6817                 /*
6818                  * whoever cleared the private bit is responsible
6819                  * for the finish_ordered_io
6820                  */
6821                 if (TestClearPagePrivate2(page) &&
6822                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6823                                                    PAGE_CACHE_SIZE, 1)) {
6824                         btrfs_finish_ordered_io(ordered);
6825                 }
6826                 btrfs_put_ordered_extent(ordered);
6827                 cached_state = NULL;
6828                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6829         }
6830         clear_extent_bit(tree, page_start, page_end,
6831                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6832                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6833                  &cached_state, GFP_NOFS);
6834         __btrfs_releasepage(page, GFP_NOFS);
6835
6836         ClearPageChecked(page);
6837         if (PagePrivate(page)) {
6838                 ClearPagePrivate(page);
6839                 set_page_private(page, 0);
6840                 page_cache_release(page);
6841         }
6842 }
6843
6844 /*
6845  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6846  * called from a page fault handler when a page is first dirtied. Hence we must
6847  * be careful to check for EOF conditions here. We set the page up correctly
6848  * for a written page which means we get ENOSPC checking when writing into
6849  * holes and correct delalloc and unwritten extent mapping on filesystems that
6850  * support these features.
6851  *
6852  * We are not allowed to take the i_mutex here so we have to play games to
6853  * protect against truncate races as the page could now be beyond EOF.  Because
6854  * vmtruncate() writes the inode size before removing pages, once we have the
6855  * page lock we can determine safely if the page is beyond EOF. If it is not
6856  * beyond EOF, then the page is guaranteed safe against truncation until we
6857  * unlock the page.
6858  */
6859 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6860 {
6861         struct page *page = vmf->page;
6862         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6863         struct btrfs_root *root = BTRFS_I(inode)->root;
6864         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6865         struct btrfs_ordered_extent *ordered;
6866         struct extent_state *cached_state = NULL;
6867         char *kaddr;
6868         unsigned long zero_start;
6869         loff_t size;
6870         int ret;
6871         int reserved = 0;
6872         u64 page_start;
6873         u64 page_end;
6874
6875         sb_start_pagefault(inode->i_sb);
6876         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6877         if (!ret) {
6878                 ret = file_update_time(vma->vm_file);
6879                 reserved = 1;
6880         }
6881         if (ret) {
6882                 if (ret == -ENOMEM)
6883                         ret = VM_FAULT_OOM;
6884                 else /* -ENOSPC, -EIO, etc */
6885                         ret = VM_FAULT_SIGBUS;
6886                 if (reserved)
6887                         goto out;
6888                 goto out_noreserve;
6889         }
6890
6891         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6892 again:
6893         lock_page(page);
6894         size = i_size_read(inode);
6895         page_start = page_offset(page);
6896         page_end = page_start + PAGE_CACHE_SIZE - 1;
6897
6898         if ((page->mapping != inode->i_mapping) ||
6899             (page_start >= size)) {
6900                 /* page got truncated out from underneath us */
6901                 goto out_unlock;
6902         }
6903         wait_on_page_writeback(page);
6904
6905         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6906         set_page_extent_mapped(page);
6907
6908         /*
6909          * we can't set the delalloc bits if there are pending ordered
6910          * extents.  Drop our locks and wait for them to finish
6911          */
6912         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6913         if (ordered) {
6914                 unlock_extent_cached(io_tree, page_start, page_end,
6915                                      &cached_state, GFP_NOFS);
6916                 unlock_page(page);
6917                 btrfs_start_ordered_extent(inode, ordered, 1);
6918                 btrfs_put_ordered_extent(ordered);
6919                 goto again;
6920         }
6921
6922         /*
6923          * XXX - page_mkwrite gets called every time the page is dirtied, even
6924          * if it was already dirty, so for space accounting reasons we need to
6925          * clear any delalloc bits for the range we are fixing to save.  There
6926          * is probably a better way to do this, but for now keep consistent with
6927          * prepare_pages in the normal write path.
6928          */
6929         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6930                           EXTENT_DIRTY | EXTENT_DELALLOC |
6931                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6932                           0, 0, &cached_state, GFP_NOFS);
6933
6934         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6935                                         &cached_state);
6936         if (ret) {
6937                 unlock_extent_cached(io_tree, page_start, page_end,
6938                                      &cached_state, GFP_NOFS);
6939                 ret = VM_FAULT_SIGBUS;
6940                 goto out_unlock;
6941         }
6942         ret = 0;
6943
6944         /* page is wholly or partially inside EOF */
6945         if (page_start + PAGE_CACHE_SIZE > size)
6946                 zero_start = size & ~PAGE_CACHE_MASK;
6947         else
6948                 zero_start = PAGE_CACHE_SIZE;
6949
6950         if (zero_start != PAGE_CACHE_SIZE) {
6951                 kaddr = kmap(page);
6952                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6953                 flush_dcache_page(page);
6954                 kunmap(page);
6955         }
6956         ClearPageChecked(page);
6957         set_page_dirty(page);
6958         SetPageUptodate(page);
6959
6960         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6961         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6962         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6963
6964         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6965
6966 out_unlock:
6967         if (!ret) {
6968                 sb_end_pagefault(inode->i_sb);
6969                 return VM_FAULT_LOCKED;
6970         }
6971         unlock_page(page);
6972 out:
6973         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6974 out_noreserve:
6975         sb_end_pagefault(inode->i_sb);
6976         return ret;
6977 }
6978
6979 static int btrfs_truncate(struct inode *inode)
6980 {
6981         struct btrfs_root *root = BTRFS_I(inode)->root;
6982         struct btrfs_block_rsv *rsv;
6983         int ret;
6984         int err = 0;
6985         struct btrfs_trans_handle *trans;
6986         u64 mask = root->sectorsize - 1;
6987         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6988
6989         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6990         if (ret)
6991                 return ret;
6992
6993         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6994         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6995
6996         /*
6997          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6998          * 3 things going on here
6999          *
7000          * 1) We need to reserve space for our orphan item and the space to
7001          * delete our orphan item.  Lord knows we don't want to have a dangling
7002          * orphan item because we didn't reserve space to remove it.
7003          *
7004          * 2) We need to reserve space to update our inode.
7005          *
7006          * 3) We need to have something to cache all the space that is going to
7007          * be free'd up by the truncate operation, but also have some slack
7008          * space reserved in case it uses space during the truncate (thank you
7009          * very much snapshotting).
7010          *
7011          * And we need these to all be seperate.  The fact is we can use alot of
7012          * space doing the truncate, and we have no earthly idea how much space
7013          * we will use, so we need the truncate reservation to be seperate so it
7014          * doesn't end up using space reserved for updating the inode or
7015          * removing the orphan item.  We also need to be able to stop the
7016          * transaction and start a new one, which means we need to be able to
7017          * update the inode several times, and we have no idea of knowing how
7018          * many times that will be, so we can't just reserve 1 item for the
7019          * entirety of the opration, so that has to be done seperately as well.
7020          * Then there is the orphan item, which does indeed need to be held on
7021          * to for the whole operation, and we need nobody to touch this reserved
7022          * space except the orphan code.
7023          *
7024          * So that leaves us with
7025          *
7026          * 1) root->orphan_block_rsv - for the orphan deletion.
7027          * 2) rsv - for the truncate reservation, which we will steal from the
7028          * transaction reservation.
7029          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
7030          * updating the inode.
7031          */
7032         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
7033         if (!rsv)
7034                 return -ENOMEM;
7035         rsv->size = min_size;
7036         rsv->failfast = 1;
7037
7038         /*
7039          * 1 for the truncate slack space
7040          * 1 for updating the inode.
7041          */
7042         trans = btrfs_start_transaction(root, 2);
7043         if (IS_ERR(trans)) {
7044                 err = PTR_ERR(trans);
7045                 goto out;
7046         }
7047
7048         /* Migrate the slack space for the truncate to our reserve */
7049         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7050                                       min_size);
7051         BUG_ON(ret);
7052
7053         /*
7054          * setattr is responsible for setting the ordered_data_close flag,
7055          * but that is only tested during the last file release.  That
7056          * could happen well after the next commit, leaving a great big
7057          * window where new writes may get lost if someone chooses to write
7058          * to this file after truncating to zero
7059          *
7060          * The inode doesn't have any dirty data here, and so if we commit
7061          * this is a noop.  If someone immediately starts writing to the inode
7062          * it is very likely we'll catch some of their writes in this
7063          * transaction, and the commit will find this file on the ordered
7064          * data list with good things to send down.
7065          *
7066          * This is a best effort solution, there is still a window where
7067          * using truncate to replace the contents of the file will
7068          * end up with a zero length file after a crash.
7069          */
7070         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7071                                            &BTRFS_I(inode)->runtime_flags))
7072                 btrfs_add_ordered_operation(trans, root, inode);
7073
7074         /*
7075          * So if we truncate and then write and fsync we normally would just
7076          * write the extents that changed, which is a problem if we need to
7077          * first truncate that entire inode.  So set this flag so we write out
7078          * all of the extents in the inode to the sync log so we're completely
7079          * safe.
7080          */
7081         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7082         trans->block_rsv = rsv;
7083
7084         while (1) {
7085                 ret = btrfs_truncate_inode_items(trans, root, inode,
7086                                                  inode->i_size,
7087                                                  BTRFS_EXTENT_DATA_KEY);
7088                 if (ret != -ENOSPC) {
7089                         err = ret;
7090                         break;
7091                 }
7092
7093                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7094                 ret = btrfs_update_inode(trans, root, inode);
7095                 if (ret) {
7096                         err = ret;
7097                         break;
7098                 }
7099
7100                 btrfs_end_transaction(trans, root);
7101                 btrfs_btree_balance_dirty(root);
7102
7103                 trans = btrfs_start_transaction(root, 2);
7104                 if (IS_ERR(trans)) {
7105                         ret = err = PTR_ERR(trans);
7106                         trans = NULL;
7107                         break;
7108                 }
7109
7110                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7111                                               rsv, min_size);
7112                 BUG_ON(ret);    /* shouldn't happen */
7113                 trans->block_rsv = rsv;
7114         }
7115
7116         if (ret == 0 && inode->i_nlink > 0) {
7117                 trans->block_rsv = root->orphan_block_rsv;
7118                 ret = btrfs_orphan_del(trans, inode);
7119                 if (ret)
7120                         err = ret;
7121         }
7122
7123         if (trans) {
7124                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7125                 ret = btrfs_update_inode(trans, root, inode);
7126                 if (ret && !err)
7127                         err = ret;
7128
7129                 ret = btrfs_end_transaction(trans, root);
7130                 btrfs_btree_balance_dirty(root);
7131         }
7132
7133 out:
7134         btrfs_free_block_rsv(root, rsv);
7135
7136         if (ret && !err)
7137                 err = ret;
7138
7139         return err;
7140 }
7141
7142 /*
7143  * create a new subvolume directory/inode (helper for the ioctl).
7144  */
7145 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7146                              struct btrfs_root *new_root, u64 new_dirid)
7147 {
7148         struct inode *inode;
7149         int err;
7150         u64 index = 0;
7151
7152         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7153                                 new_dirid, new_dirid,
7154                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7155                                 &index);
7156         if (IS_ERR(inode))
7157                 return PTR_ERR(inode);
7158         inode->i_op = &btrfs_dir_inode_operations;
7159         inode->i_fop = &btrfs_dir_file_operations;
7160
7161         set_nlink(inode, 1);
7162         btrfs_i_size_write(inode, 0);
7163
7164         err = btrfs_update_inode(trans, new_root, inode);
7165
7166         iput(inode);
7167         return err;
7168 }
7169
7170 struct inode *btrfs_alloc_inode(struct super_block *sb)
7171 {
7172         struct btrfs_inode *ei;
7173         struct inode *inode;
7174
7175         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7176         if (!ei)
7177                 return NULL;
7178
7179         ei->root = NULL;
7180         ei->generation = 0;
7181         ei->last_trans = 0;
7182         ei->last_sub_trans = 0;
7183         ei->logged_trans = 0;
7184         ei->delalloc_bytes = 0;
7185         ei->disk_i_size = 0;
7186         ei->flags = 0;
7187         ei->csum_bytes = 0;
7188         ei->index_cnt = (u64)-1;
7189         ei->last_unlink_trans = 0;
7190         ei->last_log_commit = 0;
7191
7192         spin_lock_init(&ei->lock);
7193         ei->outstanding_extents = 0;
7194         ei->reserved_extents = 0;
7195
7196         ei->runtime_flags = 0;
7197         ei->force_compress = BTRFS_COMPRESS_NONE;
7198
7199         ei->delayed_node = NULL;
7200
7201         inode = &ei->vfs_inode;
7202         extent_map_tree_init(&ei->extent_tree);
7203         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7204         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7205         ei->io_tree.track_uptodate = 1;
7206         ei->io_failure_tree.track_uptodate = 1;
7207         atomic_set(&ei->sync_writers, 0);
7208         mutex_init(&ei->log_mutex);
7209         mutex_init(&ei->delalloc_mutex);
7210         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7211         INIT_LIST_HEAD(&ei->delalloc_inodes);
7212         INIT_LIST_HEAD(&ei->ordered_operations);
7213         RB_CLEAR_NODE(&ei->rb_node);
7214
7215         return inode;
7216 }
7217
7218 static void btrfs_i_callback(struct rcu_head *head)
7219 {
7220         struct inode *inode = container_of(head, struct inode, i_rcu);
7221         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7222 }
7223
7224 void btrfs_destroy_inode(struct inode *inode)
7225 {
7226         struct btrfs_ordered_extent *ordered;
7227         struct btrfs_root *root = BTRFS_I(inode)->root;
7228
7229         WARN_ON(!hlist_empty(&inode->i_dentry));
7230         WARN_ON(inode->i_data.nrpages);
7231         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7232         WARN_ON(BTRFS_I(inode)->reserved_extents);
7233         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7234         WARN_ON(BTRFS_I(inode)->csum_bytes);
7235
7236         /*
7237          * This can happen where we create an inode, but somebody else also
7238          * created the same inode and we need to destroy the one we already
7239          * created.
7240          */
7241         if (!root)
7242                 goto free;
7243
7244         /*
7245          * Make sure we're properly removed from the ordered operation
7246          * lists.
7247          */
7248         smp_mb();
7249         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7250                 spin_lock(&root->fs_info->ordered_extent_lock);
7251                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7252                 spin_unlock(&root->fs_info->ordered_extent_lock);
7253         }
7254
7255         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7256                      &BTRFS_I(inode)->runtime_flags)) {
7257                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7258                        (unsigned long long)btrfs_ino(inode));
7259                 atomic_dec(&root->orphan_inodes);
7260         }
7261
7262         while (1) {
7263                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7264                 if (!ordered)
7265                         break;
7266                 else {
7267                         printk(KERN_ERR "btrfs found ordered "
7268                                "extent %llu %llu on inode cleanup\n",
7269                                (unsigned long long)ordered->file_offset,
7270                                (unsigned long long)ordered->len);
7271                         btrfs_remove_ordered_extent(inode, ordered);
7272                         btrfs_put_ordered_extent(ordered);
7273                         btrfs_put_ordered_extent(ordered);
7274                 }
7275         }
7276         inode_tree_del(inode);
7277         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7278 free:
7279         btrfs_remove_delayed_node(inode);
7280         call_rcu(&inode->i_rcu, btrfs_i_callback);
7281 }
7282
7283 int btrfs_drop_inode(struct inode *inode)
7284 {
7285         struct btrfs_root *root = BTRFS_I(inode)->root;
7286
7287         if (btrfs_root_refs(&root->root_item) == 0 &&
7288             !btrfs_is_free_space_inode(inode))
7289                 return 1;
7290         else
7291                 return generic_drop_inode(inode);
7292 }
7293
7294 static void init_once(void *foo)
7295 {
7296         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7297
7298         inode_init_once(&ei->vfs_inode);
7299 }
7300
7301 void btrfs_destroy_cachep(void)
7302 {
7303         /*
7304          * Make sure all delayed rcu free inodes are flushed before we
7305          * destroy cache.
7306          */
7307         rcu_barrier();
7308         if (btrfs_inode_cachep)
7309                 kmem_cache_destroy(btrfs_inode_cachep);
7310         if (btrfs_trans_handle_cachep)
7311                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7312         if (btrfs_transaction_cachep)
7313                 kmem_cache_destroy(btrfs_transaction_cachep);
7314         if (btrfs_path_cachep)
7315                 kmem_cache_destroy(btrfs_path_cachep);
7316         if (btrfs_free_space_cachep)
7317                 kmem_cache_destroy(btrfs_free_space_cachep);
7318         if (btrfs_delalloc_work_cachep)
7319                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7320 }
7321
7322 int btrfs_init_cachep(void)
7323 {
7324         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7325                         sizeof(struct btrfs_inode), 0,
7326                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7327         if (!btrfs_inode_cachep)
7328                 goto fail;
7329
7330         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7331                         sizeof(struct btrfs_trans_handle), 0,
7332                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7333         if (!btrfs_trans_handle_cachep)
7334                 goto fail;
7335
7336         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7337                         sizeof(struct btrfs_transaction), 0,
7338                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7339         if (!btrfs_transaction_cachep)
7340                 goto fail;
7341
7342         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7343                         sizeof(struct btrfs_path), 0,
7344                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7345         if (!btrfs_path_cachep)
7346                 goto fail;
7347
7348         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7349                         sizeof(struct btrfs_free_space), 0,
7350                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7351         if (!btrfs_free_space_cachep)
7352                 goto fail;
7353
7354         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7355                         sizeof(struct btrfs_delalloc_work), 0,
7356                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7357                         NULL);
7358         if (!btrfs_delalloc_work_cachep)
7359                 goto fail;
7360
7361         return 0;
7362 fail:
7363         btrfs_destroy_cachep();
7364         return -ENOMEM;
7365 }
7366
7367 static int btrfs_getattr(struct vfsmount *mnt,
7368                          struct dentry *dentry, struct kstat *stat)
7369 {
7370         u64 delalloc_bytes;
7371         struct inode *inode = dentry->d_inode;
7372         u32 blocksize = inode->i_sb->s_blocksize;
7373
7374         generic_fillattr(inode, stat);
7375         stat->dev = BTRFS_I(inode)->root->anon_dev;
7376         stat->blksize = PAGE_CACHE_SIZE;
7377
7378         spin_lock(&BTRFS_I(inode)->lock);
7379         delalloc_bytes = BTRFS_I(inode)->delalloc_bytes;
7380         spin_unlock(&BTRFS_I(inode)->lock);
7381         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7382                         ALIGN(delalloc_bytes, blocksize)) >> 9;
7383         return 0;
7384 }
7385
7386 /*
7387  * If a file is moved, it will inherit the cow and compression flags of the new
7388  * directory.
7389  */
7390 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7391 {
7392         struct btrfs_inode *b_dir = BTRFS_I(dir);
7393         struct btrfs_inode *b_inode = BTRFS_I(inode);
7394
7395         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7396                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7397         else
7398                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7399
7400         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7401                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7402                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7403         } else {
7404                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7405                                     BTRFS_INODE_NOCOMPRESS);
7406         }
7407 }
7408
7409 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7410                            struct inode *new_dir, struct dentry *new_dentry)
7411 {
7412         struct btrfs_trans_handle *trans;
7413         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7414         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7415         struct inode *new_inode = new_dentry->d_inode;
7416         struct inode *old_inode = old_dentry->d_inode;
7417         struct timespec ctime = CURRENT_TIME;
7418         u64 index = 0;
7419         u64 root_objectid;
7420         int ret;
7421         u64 old_ino = btrfs_ino(old_inode);
7422
7423         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7424                 return -EPERM;
7425
7426         /* we only allow rename subvolume link between subvolumes */
7427         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7428                 return -EXDEV;
7429
7430         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7431             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7432                 return -ENOTEMPTY;
7433
7434         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7435             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7436                 return -ENOTEMPTY;
7437
7438
7439         /* check for collisions, even if the  name isn't there */
7440         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7441                              new_dentry->d_name.name,
7442                              new_dentry->d_name.len);
7443
7444         if (ret) {
7445                 if (ret == -EEXIST) {
7446                         /* we shouldn't get
7447                          * eexist without a new_inode */
7448                         if (!new_inode) {
7449                                 WARN_ON(1);
7450                                 return ret;
7451                         }
7452                 } else {
7453                         /* maybe -EOVERFLOW */
7454                         return ret;
7455                 }
7456         }
7457         ret = 0;
7458
7459         /*
7460          * we're using rename to replace one file with another.
7461          * and the replacement file is large.  Start IO on it now so
7462          * we don't add too much work to the end of the transaction
7463          */
7464         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7465             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7466                 filemap_flush(old_inode->i_mapping);
7467
7468         /* close the racy window with snapshot create/destroy ioctl */
7469         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7470                 down_read(&root->fs_info->subvol_sem);
7471         /*
7472          * We want to reserve the absolute worst case amount of items.  So if
7473          * both inodes are subvols and we need to unlink them then that would
7474          * require 4 item modifications, but if they are both normal inodes it
7475          * would require 5 item modifications, so we'll assume their normal
7476          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7477          * should cover the worst case number of items we'll modify.
7478          */
7479         trans = btrfs_start_transaction(root, 20);
7480         if (IS_ERR(trans)) {
7481                 ret = PTR_ERR(trans);
7482                 goto out_notrans;
7483         }
7484
7485         if (dest != root)
7486                 btrfs_record_root_in_trans(trans, dest);
7487
7488         ret = btrfs_set_inode_index(new_dir, &index);
7489         if (ret)
7490                 goto out_fail;
7491
7492         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7493                 /* force full log commit if subvolume involved. */
7494                 root->fs_info->last_trans_log_full_commit = trans->transid;
7495         } else {
7496                 ret = btrfs_insert_inode_ref(trans, dest,
7497                                              new_dentry->d_name.name,
7498                                              new_dentry->d_name.len,
7499                                              old_ino,
7500                                              btrfs_ino(new_dir), index);
7501                 if (ret)
7502                         goto out_fail;
7503                 /*
7504                  * this is an ugly little race, but the rename is required
7505                  * to make sure that if we crash, the inode is either at the
7506                  * old name or the new one.  pinning the log transaction lets
7507                  * us make sure we don't allow a log commit to come in after
7508                  * we unlink the name but before we add the new name back in.
7509                  */
7510                 btrfs_pin_log_trans(root);
7511         }
7512         /*
7513          * make sure the inode gets flushed if it is replacing
7514          * something.
7515          */
7516         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7517                 btrfs_add_ordered_operation(trans, root, old_inode);
7518
7519         inode_inc_iversion(old_dir);
7520         inode_inc_iversion(new_dir);
7521         inode_inc_iversion(old_inode);
7522         old_dir->i_ctime = old_dir->i_mtime = ctime;
7523         new_dir->i_ctime = new_dir->i_mtime = ctime;
7524         old_inode->i_ctime = ctime;
7525
7526         if (old_dentry->d_parent != new_dentry->d_parent)
7527                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7528
7529         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7530                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7531                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7532                                         old_dentry->d_name.name,
7533                                         old_dentry->d_name.len);
7534         } else {
7535                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7536                                         old_dentry->d_inode,
7537                                         old_dentry->d_name.name,
7538                                         old_dentry->d_name.len);
7539                 if (!ret)
7540                         ret = btrfs_update_inode(trans, root, old_inode);
7541         }
7542         if (ret) {
7543                 btrfs_abort_transaction(trans, root, ret);
7544                 goto out_fail;
7545         }
7546
7547         if (new_inode) {
7548                 inode_inc_iversion(new_inode);
7549                 new_inode->i_ctime = CURRENT_TIME;
7550                 if (unlikely(btrfs_ino(new_inode) ==
7551                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7552                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7553                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7554                                                 root_objectid,
7555                                                 new_dentry->d_name.name,
7556                                                 new_dentry->d_name.len);
7557                         BUG_ON(new_inode->i_nlink == 0);
7558                 } else {
7559                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7560                                                  new_dentry->d_inode,
7561                                                  new_dentry->d_name.name,
7562                                                  new_dentry->d_name.len);
7563                 }
7564                 if (!ret && new_inode->i_nlink == 0) {
7565                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7566                         BUG_ON(ret);
7567                 }
7568                 if (ret) {
7569                         btrfs_abort_transaction(trans, root, ret);
7570                         goto out_fail;
7571                 }
7572         }
7573
7574         fixup_inode_flags(new_dir, old_inode);
7575
7576         ret = btrfs_add_link(trans, new_dir, old_inode,
7577                              new_dentry->d_name.name,
7578                              new_dentry->d_name.len, 0, index);
7579         if (ret) {
7580                 btrfs_abort_transaction(trans, root, ret);
7581                 goto out_fail;
7582         }
7583
7584         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7585                 struct dentry *parent = new_dentry->d_parent;
7586                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7587                 btrfs_end_log_trans(root);
7588         }
7589 out_fail:
7590         btrfs_end_transaction(trans, root);
7591 out_notrans:
7592         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7593                 up_read(&root->fs_info->subvol_sem);
7594
7595         return ret;
7596 }
7597
7598 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7599 {
7600         struct btrfs_delalloc_work *delalloc_work;
7601
7602         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7603                                      work);
7604         if (delalloc_work->wait)
7605                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7606         else
7607                 filemap_flush(delalloc_work->inode->i_mapping);
7608
7609         if (delalloc_work->delay_iput)
7610                 btrfs_add_delayed_iput(delalloc_work->inode);
7611         else
7612                 iput(delalloc_work->inode);
7613         complete(&delalloc_work->completion);
7614 }
7615
7616 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7617                                                     int wait, int delay_iput)
7618 {
7619         struct btrfs_delalloc_work *work;
7620
7621         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7622         if (!work)
7623                 return NULL;
7624
7625         init_completion(&work->completion);
7626         INIT_LIST_HEAD(&work->list);
7627         work->inode = inode;
7628         work->wait = wait;
7629         work->delay_iput = delay_iput;
7630         work->work.func = btrfs_run_delalloc_work;
7631
7632         return work;
7633 }
7634
7635 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7636 {
7637         wait_for_completion(&work->completion);
7638         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7639 }
7640
7641 /*
7642  * some fairly slow code that needs optimization. This walks the list
7643  * of all the inodes with pending delalloc and forces them to disk.
7644  */
7645 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7646 {
7647         struct btrfs_inode *binode;
7648         struct inode *inode;
7649         struct btrfs_delalloc_work *work, *next;
7650         struct list_head works;
7651         struct list_head splice;
7652         int ret = 0;
7653
7654         if (root->fs_info->sb->s_flags & MS_RDONLY)
7655                 return -EROFS;
7656
7657         INIT_LIST_HEAD(&works);
7658         INIT_LIST_HEAD(&splice);
7659
7660         spin_lock(&root->fs_info->delalloc_lock);
7661         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7662         while (!list_empty(&splice)) {
7663                 binode = list_entry(splice.next, struct btrfs_inode,
7664                                     delalloc_inodes);
7665
7666                 list_del_init(&binode->delalloc_inodes);
7667
7668                 inode = igrab(&binode->vfs_inode);
7669                 if (!inode) {
7670                         clear_bit(BTRFS_INODE_IN_DELALLOC_LIST,
7671                                   &binode->runtime_flags);
7672                         continue;
7673                 }
7674
7675                 list_add_tail(&binode->delalloc_inodes,
7676                               &root->fs_info->delalloc_inodes);
7677                 spin_unlock(&root->fs_info->delalloc_lock);
7678
7679                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7680                 if (unlikely(!work)) {
7681                         ret = -ENOMEM;
7682                         goto out;
7683                 }
7684                 list_add_tail(&work->list, &works);
7685                 btrfs_queue_worker(&root->fs_info->flush_workers,
7686                                    &work->work);
7687
7688                 cond_resched();
7689                 spin_lock(&root->fs_info->delalloc_lock);
7690         }
7691         spin_unlock(&root->fs_info->delalloc_lock);
7692
7693         list_for_each_entry_safe(work, next, &works, list) {
7694                 list_del_init(&work->list);
7695                 btrfs_wait_and_free_delalloc_work(work);
7696         }
7697
7698         /* the filemap_flush will queue IO into the worker threads, but
7699          * we have to make sure the IO is actually started and that
7700          * ordered extents get created before we return
7701          */
7702         atomic_inc(&root->fs_info->async_submit_draining);
7703         while (atomic_read(&root->fs_info->nr_async_submits) ||
7704               atomic_read(&root->fs_info->async_delalloc_pages)) {
7705                 wait_event(root->fs_info->async_submit_wait,
7706                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7707                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7708         }
7709         atomic_dec(&root->fs_info->async_submit_draining);
7710         return 0;
7711 out:
7712         list_for_each_entry_safe(work, next, &works, list) {
7713                 list_del_init(&work->list);
7714                 btrfs_wait_and_free_delalloc_work(work);
7715         }
7716
7717         if (!list_empty_careful(&splice)) {
7718                 spin_lock(&root->fs_info->delalloc_lock);
7719                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7720                 spin_unlock(&root->fs_info->delalloc_lock);
7721         }
7722         return ret;
7723 }
7724
7725 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7726                          const char *symname)
7727 {
7728         struct btrfs_trans_handle *trans;
7729         struct btrfs_root *root = BTRFS_I(dir)->root;
7730         struct btrfs_path *path;
7731         struct btrfs_key key;
7732         struct inode *inode = NULL;
7733         int err;
7734         int drop_inode = 0;
7735         u64 objectid;
7736         u64 index = 0 ;
7737         int name_len;
7738         int datasize;
7739         unsigned long ptr;
7740         struct btrfs_file_extent_item *ei;
7741         struct extent_buffer *leaf;
7742
7743         name_len = strlen(symname) + 1;
7744         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7745                 return -ENAMETOOLONG;
7746
7747         /*
7748          * 2 items for inode item and ref
7749          * 2 items for dir items
7750          * 1 item for xattr if selinux is on
7751          */
7752         trans = btrfs_start_transaction(root, 5);
7753         if (IS_ERR(trans))
7754                 return PTR_ERR(trans);
7755
7756         err = btrfs_find_free_ino(root, &objectid);
7757         if (err)
7758                 goto out_unlock;
7759
7760         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7761                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7762                                 S_IFLNK|S_IRWXUGO, &index);
7763         if (IS_ERR(inode)) {
7764                 err = PTR_ERR(inode);
7765                 goto out_unlock;
7766         }
7767
7768         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7769         if (err) {
7770                 drop_inode = 1;
7771                 goto out_unlock;
7772         }
7773
7774         /*
7775         * If the active LSM wants to access the inode during
7776         * d_instantiate it needs these. Smack checks to see
7777         * if the filesystem supports xattrs by looking at the
7778         * ops vector.
7779         */
7780         inode->i_fop = &btrfs_file_operations;
7781         inode->i_op = &btrfs_file_inode_operations;
7782
7783         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7784         if (err)
7785                 drop_inode = 1;
7786         else {
7787                 inode->i_mapping->a_ops = &btrfs_aops;
7788                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7789                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7790         }
7791         if (drop_inode)
7792                 goto out_unlock;
7793
7794         path = btrfs_alloc_path();
7795         if (!path) {
7796                 err = -ENOMEM;
7797                 drop_inode = 1;
7798                 goto out_unlock;
7799         }
7800         key.objectid = btrfs_ino(inode);
7801         key.offset = 0;
7802         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7803         datasize = btrfs_file_extent_calc_inline_size(name_len);
7804         err = btrfs_insert_empty_item(trans, root, path, &key,
7805                                       datasize);
7806         if (err) {
7807                 drop_inode = 1;
7808                 btrfs_free_path(path);
7809                 goto out_unlock;
7810         }
7811         leaf = path->nodes[0];
7812         ei = btrfs_item_ptr(leaf, path->slots[0],
7813                             struct btrfs_file_extent_item);
7814         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7815         btrfs_set_file_extent_type(leaf, ei,
7816                                    BTRFS_FILE_EXTENT_INLINE);
7817         btrfs_set_file_extent_encryption(leaf, ei, 0);
7818         btrfs_set_file_extent_compression(leaf, ei, 0);
7819         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7820         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7821
7822         ptr = btrfs_file_extent_inline_start(ei);
7823         write_extent_buffer(leaf, symname, ptr, name_len);
7824         btrfs_mark_buffer_dirty(leaf);
7825         btrfs_free_path(path);
7826
7827         inode->i_op = &btrfs_symlink_inode_operations;
7828         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7829         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7830         inode_set_bytes(inode, name_len);
7831         btrfs_i_size_write(inode, name_len - 1);
7832         err = btrfs_update_inode(trans, root, inode);
7833         if (err)
7834                 drop_inode = 1;
7835
7836 out_unlock:
7837         if (!err)
7838                 d_instantiate(dentry, inode);
7839         btrfs_end_transaction(trans, root);
7840         if (drop_inode) {
7841                 inode_dec_link_count(inode);
7842                 iput(inode);
7843         }
7844         btrfs_btree_balance_dirty(root);
7845         return err;
7846 }
7847
7848 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7849                                        u64 start, u64 num_bytes, u64 min_size,
7850                                        loff_t actual_len, u64 *alloc_hint,
7851                                        struct btrfs_trans_handle *trans)
7852 {
7853         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7854         struct extent_map *em;
7855         struct btrfs_root *root = BTRFS_I(inode)->root;
7856         struct btrfs_key ins;
7857         u64 cur_offset = start;
7858         u64 i_size;
7859         int ret = 0;
7860         bool own_trans = true;
7861
7862         if (trans)
7863                 own_trans = false;
7864         while (num_bytes > 0) {
7865                 if (own_trans) {
7866                         trans = btrfs_start_transaction(root, 3);
7867                         if (IS_ERR(trans)) {
7868                                 ret = PTR_ERR(trans);
7869                                 break;
7870                         }
7871                 }
7872
7873                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7874                                            0, *alloc_hint, &ins, 1);
7875                 if (ret) {
7876                         if (own_trans)
7877                                 btrfs_end_transaction(trans, root);
7878                         break;
7879                 }
7880
7881                 ret = insert_reserved_file_extent(trans, inode,
7882                                                   cur_offset, ins.objectid,
7883                                                   ins.offset, ins.offset,
7884                                                   ins.offset, 0, 0, 0,
7885                                                   BTRFS_FILE_EXTENT_PREALLOC);
7886                 if (ret) {
7887                         btrfs_abort_transaction(trans, root, ret);
7888                         if (own_trans)
7889                                 btrfs_end_transaction(trans, root);
7890                         break;
7891                 }
7892                 btrfs_drop_extent_cache(inode, cur_offset,
7893                                         cur_offset + ins.offset -1, 0);
7894
7895                 em = alloc_extent_map();
7896                 if (!em) {
7897                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7898                                 &BTRFS_I(inode)->runtime_flags);
7899                         goto next;
7900                 }
7901
7902                 em->start = cur_offset;
7903                 em->orig_start = cur_offset;
7904                 em->len = ins.offset;
7905                 em->block_start = ins.objectid;
7906                 em->block_len = ins.offset;
7907                 em->orig_block_len = ins.offset;
7908                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7909                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7910                 em->generation = trans->transid;
7911
7912                 while (1) {
7913                         write_lock(&em_tree->lock);
7914                         ret = add_extent_mapping(em_tree, em);
7915                         if (!ret)
7916                                 list_move(&em->list,
7917                                           &em_tree->modified_extents);
7918                         write_unlock(&em_tree->lock);
7919                         if (ret != -EEXIST)
7920                                 break;
7921                         btrfs_drop_extent_cache(inode, cur_offset,
7922                                                 cur_offset + ins.offset - 1,
7923                                                 0);
7924                 }
7925                 free_extent_map(em);
7926 next:
7927                 num_bytes -= ins.offset;
7928                 cur_offset += ins.offset;
7929                 *alloc_hint = ins.objectid + ins.offset;
7930
7931                 inode_inc_iversion(inode);
7932                 inode->i_ctime = CURRENT_TIME;
7933                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7934                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7935                     (actual_len > inode->i_size) &&
7936                     (cur_offset > inode->i_size)) {
7937                         if (cur_offset > actual_len)
7938                                 i_size = actual_len;
7939                         else
7940                                 i_size = cur_offset;
7941                         i_size_write(inode, i_size);
7942                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7943                 }
7944
7945                 ret = btrfs_update_inode(trans, root, inode);
7946
7947                 if (ret) {
7948                         btrfs_abort_transaction(trans, root, ret);
7949                         if (own_trans)
7950                                 btrfs_end_transaction(trans, root);
7951                         break;
7952                 }
7953
7954                 if (own_trans)
7955                         btrfs_end_transaction(trans, root);
7956         }
7957         return ret;
7958 }
7959
7960 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7961                               u64 start, u64 num_bytes, u64 min_size,
7962                               loff_t actual_len, u64 *alloc_hint)
7963 {
7964         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7965                                            min_size, actual_len, alloc_hint,
7966                                            NULL);
7967 }
7968
7969 int btrfs_prealloc_file_range_trans(struct inode *inode,
7970                                     struct btrfs_trans_handle *trans, int mode,
7971                                     u64 start, u64 num_bytes, u64 min_size,
7972                                     loff_t actual_len, u64 *alloc_hint)
7973 {
7974         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7975                                            min_size, actual_len, alloc_hint, trans);
7976 }
7977
7978 static int btrfs_set_page_dirty(struct page *page)
7979 {
7980         return __set_page_dirty_nobuffers(page);
7981 }
7982
7983 static int btrfs_permission(struct inode *inode, int mask)
7984 {
7985         struct btrfs_root *root = BTRFS_I(inode)->root;
7986         umode_t mode = inode->i_mode;
7987
7988         if (mask & MAY_WRITE &&
7989             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7990                 if (btrfs_root_readonly(root))
7991                         return -EROFS;
7992                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7993                         return -EACCES;
7994         }
7995         return generic_permission(inode, mask);
7996 }
7997
7998 static const struct inode_operations btrfs_dir_inode_operations = {
7999         .getattr        = btrfs_getattr,
8000         .lookup         = btrfs_lookup,
8001         .create         = btrfs_create,
8002         .unlink         = btrfs_unlink,
8003         .link           = btrfs_link,
8004         .mkdir          = btrfs_mkdir,
8005         .rmdir          = btrfs_rmdir,
8006         .rename         = btrfs_rename,
8007         .symlink        = btrfs_symlink,
8008         .setattr        = btrfs_setattr,
8009         .mknod          = btrfs_mknod,
8010         .setxattr       = btrfs_setxattr,
8011         .getxattr       = btrfs_getxattr,
8012         .listxattr      = btrfs_listxattr,
8013         .removexattr    = btrfs_removexattr,
8014         .permission     = btrfs_permission,
8015         .get_acl        = btrfs_get_acl,
8016 };
8017 static const struct inode_operations btrfs_dir_ro_inode_operations = {
8018         .lookup         = btrfs_lookup,
8019         .permission     = btrfs_permission,
8020         .get_acl        = btrfs_get_acl,
8021 };
8022
8023 static const struct file_operations btrfs_dir_file_operations = {
8024         .llseek         = generic_file_llseek,
8025         .read           = generic_read_dir,
8026         .readdir        = btrfs_real_readdir,
8027         .unlocked_ioctl = btrfs_ioctl,
8028 #ifdef CONFIG_COMPAT
8029         .compat_ioctl   = btrfs_ioctl,
8030 #endif
8031         .release        = btrfs_release_file,
8032         .fsync          = btrfs_sync_file,
8033 };
8034
8035 static struct extent_io_ops btrfs_extent_io_ops = {
8036         .fill_delalloc = run_delalloc_range,
8037         .submit_bio_hook = btrfs_submit_bio_hook,
8038         .merge_bio_hook = btrfs_merge_bio_hook,
8039         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
8040         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
8041         .writepage_start_hook = btrfs_writepage_start_hook,
8042         .set_bit_hook = btrfs_set_bit_hook,
8043         .clear_bit_hook = btrfs_clear_bit_hook,
8044         .merge_extent_hook = btrfs_merge_extent_hook,
8045         .split_extent_hook = btrfs_split_extent_hook,
8046 };
8047
8048 /*
8049  * btrfs doesn't support the bmap operation because swapfiles
8050  * use bmap to make a mapping of extents in the file.  They assume
8051  * these extents won't change over the life of the file and they
8052  * use the bmap result to do IO directly to the drive.
8053  *
8054  * the btrfs bmap call would return logical addresses that aren't
8055  * suitable for IO and they also will change frequently as COW
8056  * operations happen.  So, swapfile + btrfs == corruption.
8057  *
8058  * For now we're avoiding this by dropping bmap.
8059  */
8060 static const struct address_space_operations btrfs_aops = {
8061         .readpage       = btrfs_readpage,
8062         .writepage      = btrfs_writepage,
8063         .writepages     = btrfs_writepages,
8064         .readpages      = btrfs_readpages,
8065         .direct_IO      = btrfs_direct_IO,
8066         .invalidatepage = btrfs_invalidatepage,
8067         .releasepage    = btrfs_releasepage,
8068         .set_page_dirty = btrfs_set_page_dirty,
8069         .error_remove_page = generic_error_remove_page,
8070 };
8071
8072 static const struct address_space_operations btrfs_symlink_aops = {
8073         .readpage       = btrfs_readpage,
8074         .writepage      = btrfs_writepage,
8075         .invalidatepage = btrfs_invalidatepage,
8076         .releasepage    = btrfs_releasepage,
8077 };
8078
8079 static const struct inode_operations btrfs_file_inode_operations = {
8080         .getattr        = btrfs_getattr,
8081         .setattr        = btrfs_setattr,
8082         .setxattr       = btrfs_setxattr,
8083         .getxattr       = btrfs_getxattr,
8084         .listxattr      = btrfs_listxattr,
8085         .removexattr    = btrfs_removexattr,
8086         .permission     = btrfs_permission,
8087         .fiemap         = btrfs_fiemap,
8088         .get_acl        = btrfs_get_acl,
8089         .update_time    = btrfs_update_time,
8090 };
8091 static const struct inode_operations btrfs_special_inode_operations = {
8092         .getattr        = btrfs_getattr,
8093         .setattr        = btrfs_setattr,
8094         .permission     = btrfs_permission,
8095         .setxattr       = btrfs_setxattr,
8096         .getxattr       = btrfs_getxattr,
8097         .listxattr      = btrfs_listxattr,
8098         .removexattr    = btrfs_removexattr,
8099         .get_acl        = btrfs_get_acl,
8100         .update_time    = btrfs_update_time,
8101 };
8102 static const struct inode_operations btrfs_symlink_inode_operations = {
8103         .readlink       = generic_readlink,
8104         .follow_link    = page_follow_link_light,
8105         .put_link       = page_put_link,
8106         .getattr        = btrfs_getattr,
8107         .setattr        = btrfs_setattr,
8108         .permission     = btrfs_permission,
8109         .setxattr       = btrfs_setxattr,
8110         .getxattr       = btrfs_getxattr,
8111         .listxattr      = btrfs_listxattr,
8112         .removexattr    = btrfs_removexattr,
8113         .get_acl        = btrfs_get_acl,
8114         .update_time    = btrfs_update_time,
8115 };
8116
8117 const struct dentry_operations btrfs_dentry_operations = {
8118         .d_delete       = btrfs_dentry_delete,
8119         .d_release      = btrfs_dentry_release,
8120 };