fs/btrfs/inode.c

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