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, loff_t newsize);
  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                         ret = btrfs_truncate(inode);
2482                 } else {
2483                         nr_unlink++;
2484                 }
2485
2486                 /* this will do delete_inode and everything for us */
2487                 iput(inode);
2488                 if (ret)
2489                         goto out;
2490         }
2491         /* release the path since we're done with it */
2492         btrfs_release_path(path);
2493
2494         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2495
2496         if (root->orphan_block_rsv)
2497                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2498                                         (u64)-1);
2499
2500         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2501                 trans = btrfs_join_transaction(root);
2502                 if (!IS_ERR(trans))
2503                         btrfs_end_transaction(trans, root);
2504         }
2505
2506         if (nr_unlink)
2507                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2508         if (nr_truncate)
2509                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2510
2511 out:
2512         if (ret)
2513                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2514         btrfs_free_path(path);
2515         return ret;
2516 }
2517
2518 /*
2519  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2520  * don't find any xattrs, we know there can't be any acls.
2521  *
2522  * slot is the slot the inode is in, objectid is the objectid of the inode
2523  */
2524 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2525                                           int slot, u64 objectid)
2526 {
2527         u32 nritems = btrfs_header_nritems(leaf);
2528         struct btrfs_key found_key;
2529         int scanned = 0;
2530
2531         slot++;
2532         while (slot < nritems) {
2533                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2534
2535                 /* we found a different objectid, there must not be acls */
2536                 if (found_key.objectid != objectid)
2537                         return 0;
2538
2539                 /* we found an xattr, assume we've got an acl */
2540                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2541                         return 1;
2542
2543                 /*
2544                  * we found a key greater than an xattr key, there can't
2545                  * be any acls later on
2546                  */
2547                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2548                         return 0;
2549
2550                 slot++;
2551                 scanned++;
2552
2553                 /*
2554                  * it goes inode, inode backrefs, xattrs, extents,
2555                  * so if there are a ton of hard links to an inode there can
2556                  * be a lot of backrefs.  Don't waste time searching too hard,
2557                  * this is just an optimization
2558                  */
2559                 if (scanned >= 8)
2560                         break;
2561         }
2562         /* we hit the end of the leaf before we found an xattr or
2563          * something larger than an xattr.  We have to assume the inode
2564          * has acls
2565          */
2566         return 1;
2567 }
2568
2569 /*
2570  * read an inode from the btree into the in-memory inode
2571  */
2572 static void btrfs_read_locked_inode(struct inode *inode)
2573 {
2574         struct btrfs_path *path;
2575         struct extent_buffer *leaf;
2576         struct btrfs_inode_item *inode_item;
2577         struct btrfs_timespec *tspec;
2578         struct btrfs_root *root = BTRFS_I(inode)->root;
2579         struct btrfs_key location;
2580         int maybe_acls;
2581         u32 rdev;
2582         int ret;
2583         bool filled = false;
2584
2585         ret = btrfs_fill_inode(inode, &rdev);
2586         if (!ret)
2587                 filled = true;
2588
2589         path = btrfs_alloc_path();
2590         if (!path)
2591                 goto make_bad;
2592
2593         path->leave_spinning = 1;
2594         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2595
2596         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2597         if (ret)
2598                 goto make_bad;
2599
2600         leaf = path->nodes[0];
2601
2602         if (filled)
2603                 goto cache_acl;
2604
2605         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2606                                     struct btrfs_inode_item);
2607         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2608         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2609         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2610         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2611         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2612
2613         tspec = btrfs_inode_atime(inode_item);
2614         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2615         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2616
2617         tspec = btrfs_inode_mtime(inode_item);
2618         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2619         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2620
2621         tspec = btrfs_inode_ctime(inode_item);
2622         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2623         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2624
2625         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2626         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2627         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2628
2629         /*
2630          * If we were modified in the current generation and evicted from memory
2631          * and then re-read we need to do a full sync since we don't have any
2632          * idea about which extents were modified before we were evicted from
2633          * cache.
2634          */
2635         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2636                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2637                         &BTRFS_I(inode)->runtime_flags);
2638
2639         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2640         inode->i_generation = BTRFS_I(inode)->generation;
2641         inode->i_rdev = 0;
2642         rdev = btrfs_inode_rdev(leaf, inode_item);
2643
2644         BTRFS_I(inode)->index_cnt = (u64)-1;
2645         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2646 cache_acl:
2647         /*
2648          * try to precache a NULL acl entry for files that don't have
2649          * any xattrs or acls
2650          */
2651         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2652                                            btrfs_ino(inode));
2653         if (!maybe_acls)
2654                 cache_no_acl(inode);
2655
2656         btrfs_free_path(path);
2657
2658         switch (inode->i_mode & S_IFMT) {
2659         case S_IFREG:
2660                 inode->i_mapping->a_ops = &btrfs_aops;
2661                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2662                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2663                 inode->i_fop = &btrfs_file_operations;
2664                 inode->i_op = &btrfs_file_inode_operations;
2665                 break;
2666         case S_IFDIR:
2667                 inode->i_fop = &btrfs_dir_file_operations;
2668                 if (root == root->fs_info->tree_root)
2669                         inode->i_op = &btrfs_dir_ro_inode_operations;
2670                 else
2671                         inode->i_op = &btrfs_dir_inode_operations;
2672                 break;
2673         case S_IFLNK:
2674                 inode->i_op = &btrfs_symlink_inode_operations;
2675                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2676                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2677                 break;
2678         default:
2679                 inode->i_op = &btrfs_special_inode_operations;
2680                 init_special_inode(inode, inode->i_mode, rdev);
2681                 break;
2682         }
2683
2684         btrfs_update_iflags(inode);
2685         return;
2686
2687 make_bad:
2688         btrfs_free_path(path);
2689         make_bad_inode(inode);
2690 }
2691
2692 /*
2693  * given a leaf and an inode, copy the inode fields into the leaf
2694  */
2695 static void fill_inode_item(struct btrfs_trans_handle *trans,
2696                             struct extent_buffer *leaf,
2697                             struct btrfs_inode_item *item,
2698                             struct inode *inode)
2699 {
2700         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2701         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2702         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2703         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2704         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2705
2706         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2707                                inode->i_atime.tv_sec);
2708         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2709                                 inode->i_atime.tv_nsec);
2710
2711         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2712                                inode->i_mtime.tv_sec);
2713         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2714                                 inode->i_mtime.tv_nsec);
2715
2716         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2717                                inode->i_ctime.tv_sec);
2718         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2719                                 inode->i_ctime.tv_nsec);
2720
2721         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2722         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2723         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2724         btrfs_set_inode_transid(leaf, item, trans->transid);
2725         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2726         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2727         btrfs_set_inode_block_group(leaf, item, 0);
2728 }
2729
2730 /*
2731  * copy everything in the in-memory inode into the btree.
2732  */
2733 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2734                                 struct btrfs_root *root, struct inode *inode)
2735 {
2736         struct btrfs_inode_item *inode_item;
2737         struct btrfs_path *path;
2738         struct extent_buffer *leaf;
2739         int ret;
2740
2741         path = btrfs_alloc_path();
2742         if (!path)
2743                 return -ENOMEM;
2744
2745         path->leave_spinning = 1;
2746         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2747                                  1);
2748         if (ret) {
2749                 if (ret > 0)
2750                         ret = -ENOENT;
2751                 goto failed;
2752         }
2753
2754         btrfs_unlock_up_safe(path, 1);
2755         leaf = path->nodes[0];
2756         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2757                                     struct btrfs_inode_item);
2758
2759         fill_inode_item(trans, leaf, inode_item, inode);
2760         btrfs_mark_buffer_dirty(leaf);
2761         btrfs_set_inode_last_trans(trans, inode);
2762         ret = 0;
2763 failed:
2764         btrfs_free_path(path);
2765         return ret;
2766 }
2767
2768 /*
2769  * copy everything in the in-memory inode into the btree.
2770  */
2771 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2772                                 struct btrfs_root *root, struct inode *inode)
2773 {
2774         int ret;
2775
2776         /*
2777          * If the inode is a free space inode, we can deadlock during commit
2778          * if we put it into the delayed code.
2779          *
2780          * The data relocation inode should also be directly updated
2781          * without delay
2782          */
2783         if (!btrfs_is_free_space_inode(inode)
2784             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2785                 btrfs_update_root_times(trans, root);
2786
2787                 ret = btrfs_delayed_update_inode(trans, root, inode);
2788                 if (!ret)
2789                         btrfs_set_inode_last_trans(trans, inode);
2790                 return ret;
2791         }
2792
2793         return btrfs_update_inode_item(trans, root, inode);
2794 }
2795
2796 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2797                                          struct btrfs_root *root,
2798                                          struct inode *inode)
2799 {
2800         int ret;
2801
2802         ret = btrfs_update_inode(trans, root, inode);
2803         if (ret == -ENOSPC)
2804                 return btrfs_update_inode_item(trans, root, inode);
2805         return ret;
2806 }
2807
2808 /*
2809  * unlink helper that gets used here in inode.c and in the tree logging
2810  * recovery code.  It remove a link in a directory with a given name, and
2811  * also drops the back refs in the inode to the directory
2812  */
2813 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2814                                 struct btrfs_root *root,
2815                                 struct inode *dir, struct inode *inode,
2816                                 const char *name, int name_len)
2817 {
2818         struct btrfs_path *path;
2819         int ret = 0;
2820         struct extent_buffer *leaf;
2821         struct btrfs_dir_item *di;
2822         struct btrfs_key key;
2823         u64 index;
2824         u64 ino = btrfs_ino(inode);
2825         u64 dir_ino = btrfs_ino(dir);
2826
2827         path = btrfs_alloc_path();
2828         if (!path) {
2829                 ret = -ENOMEM;
2830                 goto out;
2831         }
2832
2833         path->leave_spinning = 1;
2834         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2835                                     name, name_len, -1);
2836         if (IS_ERR(di)) {
2837                 ret = PTR_ERR(di);
2838                 goto err;
2839         }
2840         if (!di) {
2841                 ret = -ENOENT;
2842                 goto err;
2843         }
2844         leaf = path->nodes[0];
2845         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2846         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2847         if (ret)
2848                 goto err;
2849         btrfs_release_path(path);
2850
2851         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2852                                   dir_ino, &index);
2853         if (ret) {
2854                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2855                        "inode %llu parent %llu\n", name_len, name,
2856                        (unsigned long long)ino, (unsigned long long)dir_ino);
2857                 btrfs_abort_transaction(trans, root, ret);
2858                 goto err;
2859         }
2860
2861         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2862         if (ret) {
2863                 btrfs_abort_transaction(trans, root, ret);
2864                 goto err;
2865         }
2866
2867         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2868                                          inode, dir_ino);
2869         if (ret != 0 && ret != -ENOENT) {
2870                 btrfs_abort_transaction(trans, root, ret);
2871                 goto err;
2872         }
2873
2874         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2875                                            dir, index);
2876         if (ret == -ENOENT)
2877                 ret = 0;
2878 err:
2879         btrfs_free_path(path);
2880         if (ret)
2881                 goto out;
2882
2883         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2884         inode_inc_iversion(inode);
2885         inode_inc_iversion(dir);
2886         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2887         ret = btrfs_update_inode(trans, root, dir);
2888 out:
2889         return ret;
2890 }
2891
2892 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2893                        struct btrfs_root *root,
2894                        struct inode *dir, struct inode *inode,
2895                        const char *name, int name_len)
2896 {
2897         int ret;
2898         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2899         if (!ret) {
2900                 btrfs_drop_nlink(inode);
2901                 ret = btrfs_update_inode(trans, root, inode);
2902         }
2903         return ret;
2904 }
2905
2906
2907 /* helper to check if there is any shared block in the path */
2908 static int check_path_shared(struct btrfs_root *root,
2909                              struct btrfs_path *path)
2910 {
2911         struct extent_buffer *eb;
2912         int level;
2913         u64 refs = 1;
2914
2915         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2916                 int ret;
2917
2918                 if (!path->nodes[level])
2919                         break;
2920                 eb = path->nodes[level];
2921                 if (!btrfs_block_can_be_shared(root, eb))
2922                         continue;
2923                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2924                                                &refs, NULL);
2925                 if (refs > 1)
2926                         return 1;
2927         }
2928         return 0;
2929 }
2930
2931 /*
2932  * helper to start transaction for unlink and rmdir.
2933  *
2934  * unlink and rmdir are special in btrfs, they do not always free space.
2935  * so in enospc case, we should make sure they will free space before
2936  * allowing them to use the global metadata reservation.
2937  */
2938 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2939                                                        struct dentry *dentry)
2940 {
2941         struct btrfs_trans_handle *trans;
2942         struct btrfs_root *root = BTRFS_I(dir)->root;
2943         struct btrfs_path *path;
2944         struct btrfs_dir_item *di;
2945         struct inode *inode = dentry->d_inode;
2946         u64 index;
2947         int check_link = 1;
2948         int err = -ENOSPC;
2949         int ret;
2950         u64 ino = btrfs_ino(inode);
2951         u64 dir_ino = btrfs_ino(dir);
2952
2953         /*
2954          * 1 for the possible orphan item
2955          * 1 for the dir item
2956          * 1 for the dir index
2957          * 1 for the inode ref
2958          * 1 for the inode ref in the tree log
2959          * 2 for the dir entries in the log
2960          * 1 for the inode
2961          */
2962         trans = btrfs_start_transaction(root, 8);
2963         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2964                 return trans;
2965
2966         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2967                 return ERR_PTR(-ENOSPC);
2968
2969         /* check if there is someone else holds reference */
2970         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2971                 return ERR_PTR(-ENOSPC);
2972
2973         if (atomic_read(&inode->i_count) > 2)
2974                 return ERR_PTR(-ENOSPC);
2975
2976         if (xchg(&root->fs_info->enospc_unlink, 1))
2977                 return ERR_PTR(-ENOSPC);
2978
2979         path = btrfs_alloc_path();
2980         if (!path) {
2981                 root->fs_info->enospc_unlink = 0;
2982                 return ERR_PTR(-ENOMEM);
2983         }
2984
2985         /* 1 for the orphan item */
2986         trans = btrfs_start_transaction(root, 1);
2987         if (IS_ERR(trans)) {
2988                 btrfs_free_path(path);
2989                 root->fs_info->enospc_unlink = 0;
2990                 return trans;
2991         }
2992
2993         path->skip_locking = 1;
2994         path->search_commit_root = 1;
2995
2996         ret = btrfs_lookup_inode(trans, root, path,
2997                                 &BTRFS_I(dir)->location, 0);
2998         if (ret < 0) {
2999                 err = ret;
3000                 goto out;
3001         }
3002         if (ret == 0) {
3003                 if (check_path_shared(root, path))
3004                         goto out;
3005         } else {
3006                 check_link = 0;
3007         }
3008         btrfs_release_path(path);
3009
3010         ret = btrfs_lookup_inode(trans, root, path,
3011                                 &BTRFS_I(inode)->location, 0);
3012         if (ret < 0) {
3013                 err = ret;
3014                 goto out;
3015         }
3016         if (ret == 0) {
3017                 if (check_path_shared(root, path))
3018                         goto out;
3019         } else {
3020                 check_link = 0;
3021         }
3022         btrfs_release_path(path);
3023
3024         if (ret == 0 && S_ISREG(inode->i_mode)) {
3025                 ret = btrfs_lookup_file_extent(trans, root, path,
3026                                                ino, (u64)-1, 0);
3027                 if (ret < 0) {
3028                         err = ret;
3029                         goto out;
3030                 }
3031                 BUG_ON(ret == 0); /* Corruption */
3032                 if (check_path_shared(root, path))
3033                         goto out;
3034                 btrfs_release_path(path);
3035         }
3036
3037         if (!check_link) {
3038                 err = 0;
3039                 goto out;
3040         }
3041
3042         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3043                                 dentry->d_name.name, dentry->d_name.len, 0);
3044         if (IS_ERR(di)) {
3045                 err = PTR_ERR(di);
3046                 goto out;
3047         }
3048         if (di) {
3049                 if (check_path_shared(root, path))
3050                         goto out;
3051         } else {
3052                 err = 0;
3053                 goto out;
3054         }
3055         btrfs_release_path(path);
3056
3057         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3058                                         dentry->d_name.len, ino, dir_ino, 0,
3059                                         &index);
3060         if (ret) {
3061                 err = ret;
3062                 goto out;
3063         }
3064
3065         if (check_path_shared(root, path))
3066                 goto out;
3067
3068         btrfs_release_path(path);
3069
3070         /*
3071          * This is a commit root search, if we can lookup inode item and other
3072          * relative items in the commit root, it means the transaction of
3073          * dir/file creation has been committed, and the dir index item that we
3074          * delay to insert has also been inserted into the commit root. So
3075          * we needn't worry about the delayed insertion of the dir index item
3076          * here.
3077          */
3078         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3079                                 dentry->d_name.name, dentry->d_name.len, 0);
3080         if (IS_ERR(di)) {
3081                 err = PTR_ERR(di);
3082                 goto out;
3083         }
3084         BUG_ON(ret == -ENOENT);
3085         if (check_path_shared(root, path))
3086                 goto out;
3087
3088         err = 0;
3089 out:
3090         btrfs_free_path(path);
3091         /* Migrate the orphan reservation over */
3092         if (!err)
3093                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3094                                 &root->fs_info->global_block_rsv,
3095                                 trans->bytes_reserved);
3096
3097         if (err) {
3098                 btrfs_end_transaction(trans, root);
3099                 root->fs_info->enospc_unlink = 0;
3100                 return ERR_PTR(err);
3101         }
3102
3103         trans->block_rsv = &root->fs_info->global_block_rsv;
3104         return trans;
3105 }
3106
3107 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3108                                struct btrfs_root *root)
3109 {
3110         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3111                 btrfs_block_rsv_release(root, trans->block_rsv,
3112                                         trans->bytes_reserved);
3113                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3114                 BUG_ON(!root->fs_info->enospc_unlink);
3115                 root->fs_info->enospc_unlink = 0;
3116         }
3117         btrfs_end_transaction(trans, root);
3118 }
3119
3120 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3121 {
3122         struct btrfs_root *root = BTRFS_I(dir)->root;
3123         struct btrfs_trans_handle *trans;
3124         struct inode *inode = dentry->d_inode;
3125         int ret;
3126
3127         trans = __unlink_start_trans(dir, dentry);
3128         if (IS_ERR(trans))
3129                 return PTR_ERR(trans);
3130
3131         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3132
3133         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3134                                  dentry->d_name.name, dentry->d_name.len);
3135         if (ret)
3136                 goto out;
3137
3138         if (inode->i_nlink == 0) {
3139                 ret = btrfs_orphan_add(trans, inode);
3140                 if (ret)
3141                         goto out;
3142         }
3143
3144 out:
3145         __unlink_end_trans(trans, root);
3146         btrfs_btree_balance_dirty(root);
3147         return ret;
3148 }
3149
3150 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3151                         struct btrfs_root *root,
3152                         struct inode *dir, u64 objectid,
3153                         const char *name, int name_len)
3154 {
3155         struct btrfs_path *path;
3156         struct extent_buffer *leaf;
3157         struct btrfs_dir_item *di;
3158         struct btrfs_key key;
3159         u64 index;
3160         int ret;
3161         u64 dir_ino = btrfs_ino(dir);
3162
3163         path = btrfs_alloc_path();
3164         if (!path)
3165                 return -ENOMEM;
3166
3167         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3168                                    name, name_len, -1);
3169         if (IS_ERR_OR_NULL(di)) {
3170                 if (!di)
3171                         ret = -ENOENT;
3172                 else
3173                         ret = PTR_ERR(di);
3174                 goto out;
3175         }
3176
3177         leaf = path->nodes[0];
3178         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3179         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3180         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3181         if (ret) {
3182                 btrfs_abort_transaction(trans, root, ret);
3183                 goto out;
3184         }
3185         btrfs_release_path(path);
3186
3187         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3188                                  objectid, root->root_key.objectid,
3189                                  dir_ino, &index, name, name_len);
3190         if (ret < 0) {
3191                 if (ret != -ENOENT) {
3192                         btrfs_abort_transaction(trans, root, ret);
3193                         goto out;
3194                 }
3195                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3196                                                  name, name_len);
3197                 if (IS_ERR_OR_NULL(di)) {
3198                         if (!di)
3199                                 ret = -ENOENT;
3200                         else
3201                                 ret = PTR_ERR(di);
3202                         btrfs_abort_transaction(trans, root, ret);
3203                         goto out;
3204                 }
3205
3206                 leaf = path->nodes[0];
3207                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3208                 btrfs_release_path(path);
3209                 index = key.offset;
3210         }
3211         btrfs_release_path(path);
3212
3213         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3214         if (ret) {
3215                 btrfs_abort_transaction(trans, root, ret);
3216                 goto out;
3217         }
3218
3219         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3220         inode_inc_iversion(dir);
3221         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3222         ret = btrfs_update_inode_fallback(trans, root, dir);
3223         if (ret)
3224                 btrfs_abort_transaction(trans, root, ret);
3225 out:
3226         btrfs_free_path(path);
3227         return ret;
3228 }
3229
3230 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3231 {
3232         struct inode *inode = dentry->d_inode;
3233         int err = 0;
3234         struct btrfs_root *root = BTRFS_I(dir)->root;
3235         struct btrfs_trans_handle *trans;
3236
3237         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3238                 return -ENOTEMPTY;
3239         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3240                 return -EPERM;
3241
3242         trans = __unlink_start_trans(dir, dentry);
3243         if (IS_ERR(trans))
3244                 return PTR_ERR(trans);
3245
3246         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3247                 err = btrfs_unlink_subvol(trans, root, dir,
3248                                           BTRFS_I(inode)->location.objectid,
3249                                           dentry->d_name.name,
3250                                           dentry->d_name.len);
3251                 goto out;
3252         }
3253
3254         err = btrfs_orphan_add(trans, inode);
3255         if (err)
3256                 goto out;
3257
3258         /* now the directory is empty */
3259         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3260                                  dentry->d_name.name, dentry->d_name.len);
3261         if (!err)
3262                 btrfs_i_size_write(inode, 0);
3263 out:
3264         __unlink_end_trans(trans, root);
3265         btrfs_btree_balance_dirty(root);
3266
3267         return err;
3268 }
3269
3270 /*
3271  * this can truncate away extent items, csum items and directory items.
3272  * It starts at a high offset and removes keys until it can't find
3273  * any higher than new_size
3274  *
3275  * csum items that cross the new i_size are truncated to the new size
3276  * as well.
3277  *
3278  * min_type is the minimum key type to truncate down to.  If set to 0, this
3279  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3280  */
3281 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3282                                struct btrfs_root *root,
3283                                struct inode *inode,
3284                                u64 new_size, u32 min_type)
3285 {
3286         struct btrfs_path *path;
3287         struct extent_buffer *leaf;
3288         struct btrfs_file_extent_item *fi;
3289         struct btrfs_key key;
3290         struct btrfs_key found_key;
3291         u64 extent_start = 0;
3292         u64 extent_num_bytes = 0;
3293         u64 extent_offset = 0;
3294         u64 item_end = 0;
3295         u64 mask = root->sectorsize - 1;
3296         u32 found_type = (u8)-1;
3297         int found_extent;
3298         int del_item;
3299         int pending_del_nr = 0;
3300         int pending_del_slot = 0;
3301         int extent_type = -1;
3302         int ret;
3303         int err = 0;
3304         u64 ino = btrfs_ino(inode);
3305
3306         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3307
3308         path = btrfs_alloc_path();
3309         if (!path)
3310                 return -ENOMEM;
3311         path->reada = -1;
3312
3313         /*
3314          * We want to drop from the next block forward in case this new size is
3315          * not block aligned since we will be keeping the last block of the
3316          * extent just the way it is.
3317          */
3318         if (root->ref_cows || root == root->fs_info->tree_root)
3319                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3320
3321         /*
3322          * This function is also used to drop the items in the log tree before
3323          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3324          * it is used to drop the loged items. So we shouldn't kill the delayed
3325          * items.
3326          */
3327         if (min_type == 0 && root == BTRFS_I(inode)->root)
3328                 btrfs_kill_delayed_inode_items(inode);
3329
3330         key.objectid = ino;
3331         key.offset = (u64)-1;
3332         key.type = (u8)-1;
3333
3334 search_again:
3335         path->leave_spinning = 1;
3336         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3337         if (ret < 0) {
3338                 err = ret;
3339                 goto out;
3340         }
3341
3342         if (ret > 0) {
3343                 /* there are no items in the tree for us to truncate, we're
3344                  * done
3345                  */
3346                 if (path->slots[0] == 0)
3347                         goto out;
3348                 path->slots[0]--;
3349         }
3350
3351         while (1) {
3352                 fi = NULL;
3353                 leaf = path->nodes[0];
3354                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3355                 found_type = btrfs_key_type(&found_key);
3356
3357                 if (found_key.objectid != ino)
3358                         break;
3359
3360                 if (found_type < min_type)
3361                         break;
3362
3363                 item_end = found_key.offset;
3364                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3365                         fi = btrfs_item_ptr(leaf, path->slots[0],
3366                                             struct btrfs_file_extent_item);
3367                         extent_type = btrfs_file_extent_type(leaf, fi);
3368                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3369                                 item_end +=
3370                                     btrfs_file_extent_num_bytes(leaf, fi);
3371                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3372                                 item_end += btrfs_file_extent_inline_len(leaf,
3373                                                                          fi);
3374                         }
3375                         item_end--;
3376                 }
3377                 if (found_type > min_type) {
3378                         del_item = 1;
3379                 } else {
3380                         if (item_end < new_size)
3381                                 break;
3382                         if (found_key.offset >= new_size)
3383                                 del_item = 1;
3384                         else
3385                                 del_item = 0;
3386                 }
3387                 found_extent = 0;
3388                 /* FIXME, shrink the extent if the ref count is only 1 */
3389                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3390                         goto delete;
3391
3392                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3393                         u64 num_dec;
3394                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3395                         if (!del_item) {
3396                                 u64 orig_num_bytes =
3397                                         btrfs_file_extent_num_bytes(leaf, fi);
3398                                 extent_num_bytes = new_size -
3399                                         found_key.offset + root->sectorsize - 1;
3400                                 extent_num_bytes = extent_num_bytes &
3401                                         ~((u64)root->sectorsize - 1);
3402                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3403                                                          extent_num_bytes);
3404                                 num_dec = (orig_num_bytes -
3405                                            extent_num_bytes);
3406                                 if (root->ref_cows && extent_start != 0)
3407                                         inode_sub_bytes(inode, num_dec);
3408                                 btrfs_mark_buffer_dirty(leaf);
3409                         } else {
3410                                 extent_num_bytes =
3411                                         btrfs_file_extent_disk_num_bytes(leaf,
3412                                                                          fi);
3413                                 extent_offset = found_key.offset -
3414                                         btrfs_file_extent_offset(leaf, fi);
3415
3416                                 /* FIXME blocksize != 4096 */
3417                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3418                                 if (extent_start != 0) {
3419                                         found_extent = 1;
3420                                         if (root->ref_cows)
3421                                                 inode_sub_bytes(inode, num_dec);
3422                                 }
3423                         }
3424                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3425                         /*
3426                          * we can't truncate inline items that have had
3427                          * special encodings
3428                          */
3429                         if (!del_item &&
3430                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3431                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3432                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3433                                 u32 size = new_size - found_key.offset;
3434
3435                                 if (root->ref_cows) {
3436                                         inode_sub_bytes(inode, item_end + 1 -
3437                                                         new_size);
3438                                 }
3439                                 size =
3440                                     btrfs_file_extent_calc_inline_size(size);
3441                                 btrfs_truncate_item(trans, root, path,
3442                                                     size, 1);
3443                         } else if (root->ref_cows) {
3444                                 inode_sub_bytes(inode, item_end + 1 -
3445                                                 found_key.offset);
3446                         }
3447                 }
3448 delete:
3449                 if (del_item) {
3450                         if (!pending_del_nr) {
3451                                 /* no pending yet, add ourselves */
3452                                 pending_del_slot = path->slots[0];
3453                                 pending_del_nr = 1;
3454                         } else if (pending_del_nr &&
3455                                    path->slots[0] + 1 == pending_del_slot) {
3456                                 /* hop on the pending chunk */
3457                                 pending_del_nr++;
3458                                 pending_del_slot = path->slots[0];
3459                         } else {
3460                                 BUG();
3461                         }
3462                 } else {
3463                         break;
3464                 }
3465                 if (found_extent && (root->ref_cows ||
3466                                      root == root->fs_info->tree_root)) {
3467                         btrfs_set_path_blocking(path);
3468                         ret = btrfs_free_extent(trans, root, extent_start,
3469                                                 extent_num_bytes, 0,
3470                                                 btrfs_header_owner(leaf),
3471                                                 ino, extent_offset, 0);
3472                         BUG_ON(ret);
3473                 }
3474
3475                 if (found_type == BTRFS_INODE_ITEM_KEY)
3476                         break;
3477
3478                 if (path->slots[0] == 0 ||
3479                     path->slots[0] != pending_del_slot) {
3480                         if (pending_del_nr) {
3481                                 ret = btrfs_del_items(trans, root, path,
3482                                                 pending_del_slot,
3483                                                 pending_del_nr);
3484                                 if (ret) {
3485                                         btrfs_abort_transaction(trans,
3486                                                                 root, ret);
3487                                         goto error;
3488                                 }
3489                                 pending_del_nr = 0;
3490                         }
3491                         btrfs_release_path(path);
3492                         goto search_again;
3493                 } else {
3494                         path->slots[0]--;
3495                 }
3496         }
3497 out:
3498         if (pending_del_nr) {
3499                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3500                                       pending_del_nr);
3501                 if (ret)
3502                         btrfs_abort_transaction(trans, root, ret);
3503         }
3504 error:
3505         btrfs_free_path(path);
3506         return err;
3507 }
3508
3509 /*
3510  * btrfs_truncate_page - read, zero a chunk and write a page
3511  * @inode - inode that we're zeroing
3512  * @from - the offset to start zeroing
3513  * @len - the length to zero, 0 to zero the entire range respective to the
3514  *      offset
3515  * @front - zero up to the offset instead of from the offset on
3516  *
3517  * This will find the page for the "from" offset and cow the page and zero the
3518  * part we want to zero.  This is used with truncate and hole punching.
3519  */
3520 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3521                         int front)
3522 {
3523         struct address_space *mapping = inode->i_mapping;
3524         struct btrfs_root *root = BTRFS_I(inode)->root;
3525         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3526         struct btrfs_ordered_extent *ordered;
3527         struct extent_state *cached_state = NULL;
3528         char *kaddr;
3529         u32 blocksize = root->sectorsize;
3530         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3531         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3532         struct page *page;
3533         gfp_t mask = btrfs_alloc_write_mask(mapping);
3534         int ret = 0;
3535         u64 page_start;
3536         u64 page_end;
3537
3538         if ((offset & (blocksize - 1)) == 0 &&
3539             (!len || ((len & (blocksize - 1)) == 0)))
3540                 goto out;
3541         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3542         if (ret)
3543                 goto out;
3544
3545 again:
3546         page = find_or_create_page(mapping, index, mask);
3547         if (!page) {
3548                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3549                 ret = -ENOMEM;
3550                 goto out;
3551         }
3552
3553         page_start = page_offset(page);
3554         page_end = page_start + PAGE_CACHE_SIZE - 1;
3555
3556         if (!PageUptodate(page)) {
3557                 ret = btrfs_readpage(NULL, page);
3558                 lock_page(page);
3559                 if (page->mapping != mapping) {
3560                         unlock_page(page);
3561                         page_cache_release(page);
3562                         goto again;
3563                 }
3564                 if (!PageUptodate(page)) {
3565                         ret = -EIO;
3566                         goto out_unlock;
3567                 }
3568         }
3569         wait_on_page_writeback(page);
3570
3571         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3572         set_page_extent_mapped(page);
3573
3574         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3575         if (ordered) {
3576                 unlock_extent_cached(io_tree, page_start, page_end,
3577                                      &cached_state, GFP_NOFS);
3578                 unlock_page(page);
3579                 page_cache_release(page);
3580                 btrfs_start_ordered_extent(inode, ordered, 1);
3581                 btrfs_put_ordered_extent(ordered);
3582                 goto again;
3583         }
3584
3585         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3586                           EXTENT_DIRTY | EXTENT_DELALLOC |
3587                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3588                           0, 0, &cached_state, GFP_NOFS);
3589
3590         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3591                                         &cached_state);
3592         if (ret) {
3593                 unlock_extent_cached(io_tree, page_start, page_end,
3594                                      &cached_state, GFP_NOFS);
3595                 goto out_unlock;
3596         }
3597
3598         if (offset != PAGE_CACHE_SIZE) {
3599                 if (!len)
3600                         len = PAGE_CACHE_SIZE - offset;
3601                 kaddr = kmap(page);
3602                 if (front)
3603                         memset(kaddr, 0, offset);
3604                 else
3605                         memset(kaddr + offset, 0, len);
3606                 flush_dcache_page(page);
3607                 kunmap(page);
3608         }
3609         ClearPageChecked(page);
3610         set_page_dirty(page);
3611         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3612                              GFP_NOFS);
3613
3614 out_unlock:
3615         if (ret)
3616                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3617         unlock_page(page);
3618         page_cache_release(page);
3619 out:
3620         return ret;
3621 }
3622
3623 /*
3624  * This function puts in dummy file extents for the area we're creating a hole
3625  * for.  So if we are truncating this file to a larger size we need to insert
3626  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3627  * the range between oldsize and size
3628  */
3629 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3630 {
3631         struct btrfs_trans_handle *trans;
3632         struct btrfs_root *root = BTRFS_I(inode)->root;
3633         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3634         struct extent_map *em = NULL;
3635         struct extent_state *cached_state = NULL;
3636         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3637         u64 mask = root->sectorsize - 1;
3638         u64 hole_start = (oldsize + mask) & ~mask;
3639         u64 block_end = (size + mask) & ~mask;
3640         u64 last_byte;
3641         u64 cur_offset;
3642         u64 hole_size;
3643         int err = 0;
3644
3645         if (size <= hole_start)
3646                 return 0;
3647
3648         while (1) {
3649                 struct btrfs_ordered_extent *ordered;
3650                 btrfs_wait_ordered_range(inode, hole_start,
3651                                          block_end - hole_start);
3652                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3653                                  &cached_state);
3654                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3655                 if (!ordered)
3656                         break;
3657                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3658                                      &cached_state, GFP_NOFS);
3659                 btrfs_put_ordered_extent(ordered);
3660         }
3661
3662         cur_offset = hole_start;
3663         while (1) {
3664                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3665                                 block_end - cur_offset, 0);
3666                 if (IS_ERR(em)) {
3667                         err = PTR_ERR(em);
3668                         break;
3669                 }
3670                 last_byte = min(extent_map_end(em), block_end);
3671                 last_byte = (last_byte + mask) & ~mask;
3672                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3673                         struct extent_map *hole_em;
3674                         hole_size = last_byte - cur_offset;
3675
3676                         trans = btrfs_start_transaction(root, 3);
3677                         if (IS_ERR(trans)) {
3678                                 err = PTR_ERR(trans);
3679                                 break;
3680                         }
3681
3682                         err = btrfs_drop_extents(trans, root, inode,
3683                                                  cur_offset,
3684                                                  cur_offset + hole_size, 1);
3685                         if (err) {
3686                                 btrfs_abort_transaction(trans, root, err);
3687                                 btrfs_end_transaction(trans, root);
3688                                 break;
3689                         }
3690
3691                         err = btrfs_insert_file_extent(trans, root,
3692                                         btrfs_ino(inode), cur_offset, 0,
3693                                         0, hole_size, 0, hole_size,
3694                                         0, 0, 0);
3695                         if (err) {
3696                                 btrfs_abort_transaction(trans, root, err);
3697                                 btrfs_end_transaction(trans, root);
3698                                 break;
3699                         }
3700
3701                         btrfs_drop_extent_cache(inode, cur_offset,
3702                                                 cur_offset + hole_size - 1, 0);
3703                         hole_em = alloc_extent_map();
3704                         if (!hole_em) {
3705                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3706                                         &BTRFS_I(inode)->runtime_flags);
3707                                 goto next;
3708                         }
3709                         hole_em->start = cur_offset;
3710                         hole_em->len = hole_size;
3711                         hole_em->orig_start = cur_offset;
3712
3713                         hole_em->block_start = EXTENT_MAP_HOLE;
3714                         hole_em->block_len = 0;
3715                         hole_em->orig_block_len = 0;
3716                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3717                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3718                         hole_em->generation = trans->transid;
3719
3720                         while (1) {
3721                                 write_lock(&em_tree->lock);
3722                                 err = add_extent_mapping(em_tree, hole_em);
3723                                 if (!err)
3724                                         list_move(&hole_em->list,
3725                                                   &em_tree->modified_extents);
3726                                 write_unlock(&em_tree->lock);
3727                                 if (err != -EEXIST)
3728                                         break;
3729                                 btrfs_drop_extent_cache(inode, cur_offset,
3730                                                         cur_offset +
3731                                                         hole_size - 1, 0);
3732                         }
3733                         free_extent_map(hole_em);
3734 next:
3735                         btrfs_update_inode(trans, root, inode);
3736                         btrfs_end_transaction(trans, root);
3737                 }
3738                 free_extent_map(em);
3739                 em = NULL;
3740                 cur_offset = last_byte;
3741                 if (cur_offset >= block_end)
3742                         break;
3743         }
3744
3745         free_extent_map(em);
3746         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3747                              GFP_NOFS);
3748         return err;
3749 }
3750
3751 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3752 {
3753         struct btrfs_root *root = BTRFS_I(inode)->root;
3754         struct btrfs_trans_handle *trans;
3755         loff_t oldsize = i_size_read(inode);
3756         int ret;
3757
3758         if (newsize == oldsize)
3759                 return 0;
3760
3761         if (newsize > oldsize) {
3762                 truncate_pagecache(inode, oldsize, newsize);
3763                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3764                 if (ret)
3765                         return ret;
3766
3767                 trans = btrfs_start_transaction(root, 1);
3768                 if (IS_ERR(trans))
3769                         return PTR_ERR(trans);
3770
3771                 i_size_write(inode, newsize);
3772                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3773                 ret = btrfs_update_inode(trans, root, inode);
3774                 btrfs_end_transaction(trans, root);
3775         } else {
3776
3777                 /*
3778                  * We're truncating a file that used to have good data down to
3779                  * zero. Make sure it gets into the ordered flush list so that
3780                  * any new writes get down to disk quickly.
3781                  */
3782                 if (newsize == 0)
3783                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3784                                 &BTRFS_I(inode)->runtime_flags);
3785
3786                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3787                 truncate_setsize(inode, newsize);
3788                 ret = btrfs_truncate(inode);
3789         }
3790
3791         return ret;
3792 }
3793
3794 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3795 {
3796         struct inode *inode = dentry->d_inode;
3797         struct btrfs_root *root = BTRFS_I(inode)->root;
3798         int err;
3799
3800         if (btrfs_root_readonly(root))
3801                 return -EROFS;
3802
3803         err = inode_change_ok(inode, attr);
3804         if (err)
3805                 return err;
3806
3807         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3808                 err = btrfs_setsize(inode, attr->ia_size);
3809                 if (err)
3810                         return err;
3811         }
3812
3813         if (attr->ia_valid) {
3814                 setattr_copy(inode, attr);
3815                 inode_inc_iversion(inode);
3816                 err = btrfs_dirty_inode(inode);
3817
3818                 if (!err && attr->ia_valid & ATTR_MODE)
3819                         err = btrfs_acl_chmod(inode);
3820         }
3821
3822         return err;
3823 }
3824
3825 void btrfs_evict_inode(struct inode *inode)
3826 {
3827         struct btrfs_trans_handle *trans;
3828         struct btrfs_root *root = BTRFS_I(inode)->root;
3829         struct btrfs_block_rsv *rsv, *global_rsv;
3830         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3831         int ret;
3832
3833         trace_btrfs_inode_evict(inode);
3834
3835         truncate_inode_pages(&inode->i_data, 0);
3836         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3837                                btrfs_is_free_space_inode(inode)))
3838                 goto no_delete;
3839
3840         if (is_bad_inode(inode)) {
3841                 btrfs_orphan_del(NULL, inode);
3842                 goto no_delete;
3843         }
3844         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3845         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3846
3847         if (root->fs_info->log_root_recovering) {
3848                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3849                                  &BTRFS_I(inode)->runtime_flags));
3850                 goto no_delete;
3851         }
3852
3853         if (inode->i_nlink > 0) {
3854                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3855                 goto no_delete;
3856         }
3857
3858         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3859         if (!rsv) {
3860                 btrfs_orphan_del(NULL, inode);
3861                 goto no_delete;
3862         }
3863         rsv->size = min_size;
3864         rsv->failfast = 1;
3865         global_rsv = &root->fs_info->global_block_rsv;
3866
3867         btrfs_i_size_write(inode, 0);
3868
3869         /*
3870          * This is a bit simpler than btrfs_truncate since we've already
3871          * reserved our space for our orphan item in the unlink, so we just
3872          * need to reserve some slack space in case we add bytes and update
3873          * inode item when doing the truncate.
3874          */
3875         while (1) {
3876                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3877                                              BTRFS_RESERVE_FLUSH_LIMIT);
3878
3879                 /*
3880                  * Try and steal from the global reserve since we will
3881                  * likely not use this space anyway, we want to try as
3882                  * hard as possible to get this to work.
3883                  */
3884                 if (ret)
3885                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3886
3887                 if (ret) {
3888                         printk(KERN_WARNING "Could not get space for a "
3889                                "delete, will truncate on mount %d\n", ret);
3890                         btrfs_orphan_del(NULL, inode);
3891                         btrfs_free_block_rsv(root, rsv);
3892                         goto no_delete;
3893                 }
3894
3895                 trans = btrfs_start_transaction_lflush(root, 1);
3896                 if (IS_ERR(trans)) {
3897                         btrfs_orphan_del(NULL, inode);
3898                         btrfs_free_block_rsv(root, rsv);
3899                         goto no_delete;
3900                 }
3901
3902                 trans->block_rsv = rsv;
3903
3904                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3905                 if (ret != -ENOSPC)
3906                         break;
3907
3908                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3909                 ret = btrfs_update_inode(trans, root, inode);
3910                 BUG_ON(ret);
3911
3912                 btrfs_end_transaction(trans, root);
3913                 trans = NULL;
3914                 btrfs_btree_balance_dirty(root);
3915         }
3916
3917         btrfs_free_block_rsv(root, rsv);
3918
3919         if (ret == 0) {
3920                 trans->block_rsv = root->orphan_block_rsv;
3921                 ret = btrfs_orphan_del(trans, inode);
3922                 BUG_ON(ret);
3923         }
3924
3925         trans->block_rsv = &root->fs_info->trans_block_rsv;
3926         if (!(root == root->fs_info->tree_root ||
3927               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3928                 btrfs_return_ino(root, btrfs_ino(inode));
3929
3930         btrfs_end_transaction(trans, root);
3931         btrfs_btree_balance_dirty(root);
3932 no_delete:
3933         clear_inode(inode);
3934         return;
3935 }
3936
3937 /*
3938  * this returns the key found in the dir entry in the location pointer.
3939  * If no dir entries were found, location->objectid is 0.
3940  */
3941 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3942                                struct btrfs_key *location)
3943 {
3944         const char *name = dentry->d_name.name;
3945         int namelen = dentry->d_name.len;
3946         struct btrfs_dir_item *di;
3947         struct btrfs_path *path;
3948         struct btrfs_root *root = BTRFS_I(dir)->root;
3949         int ret = 0;
3950
3951         path = btrfs_alloc_path();
3952         if (!path)
3953                 return -ENOMEM;
3954
3955         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3956                                     namelen, 0);
3957         if (IS_ERR(di))
3958                 ret = PTR_ERR(di);
3959
3960         if (IS_ERR_OR_NULL(di))
3961                 goto out_err;
3962
3963         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3964 out:
3965         btrfs_free_path(path);
3966         return ret;
3967 out_err:
3968         location->objectid = 0;
3969         goto out;
3970 }
3971
3972 /*
3973  * when we hit a tree root in a directory, the btrfs part of the inode
3974  * needs to be changed to reflect the root directory of the tree root.  This
3975  * is kind of like crossing a mount point.
3976  */
3977 static int fixup_tree_root_location(struct btrfs_root *root,
3978                                     struct inode *dir,
3979                                     struct dentry *dentry,
3980                                     struct btrfs_key *location,
3981                                     struct btrfs_root **sub_root)
3982 {
3983         struct btrfs_path *path;
3984         struct btrfs_root *new_root;
3985         struct btrfs_root_ref *ref;
3986         struct extent_buffer *leaf;
3987         int ret;
3988         int err = 0;
3989
3990         path = btrfs_alloc_path();
3991         if (!path) {
3992                 err = -ENOMEM;
3993                 goto out;
3994         }
3995
3996         err = -ENOENT;
3997         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3998                                   BTRFS_I(dir)->root->root_key.objectid,
3999                                   location->objectid);
4000         if (ret) {
4001                 if (ret < 0)
4002                         err = ret;
4003                 goto out;
4004         }
4005
4006         leaf = path->nodes[0];
4007         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4008         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4009             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4010                 goto out;
4011
4012         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4013                                    (unsigned long)(ref + 1),
4014                                    dentry->d_name.len);
4015         if (ret)
4016                 goto out;
4017
4018         btrfs_release_path(path);
4019
4020         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4021         if (IS_ERR(new_root)) {
4022                 err = PTR_ERR(new_root);
4023                 goto out;
4024         }
4025
4026         if (btrfs_root_refs(&new_root->root_item) == 0) {
4027                 err = -ENOENT;
4028                 goto out;
4029         }
4030
4031         *sub_root = new_root;
4032         location->objectid = btrfs_root_dirid(&new_root->root_item);
4033         location->type = BTRFS_INODE_ITEM_KEY;
4034         location->offset = 0;
4035         err = 0;
4036 out:
4037         btrfs_free_path(path);
4038         return err;
4039 }
4040
4041 static void inode_tree_add(struct inode *inode)
4042 {
4043         struct btrfs_root *root = BTRFS_I(inode)->root;
4044         struct btrfs_inode *entry;
4045         struct rb_node **p;
4046         struct rb_node *parent;
4047         u64 ino = btrfs_ino(inode);
4048 again:
4049         p = &root->inode_tree.rb_node;
4050         parent = NULL;
4051
4052         if (inode_unhashed(inode))
4053                 return;
4054
4055         spin_lock(&root->inode_lock);
4056         while (*p) {
4057                 parent = *p;
4058                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4059
4060                 if (ino < btrfs_ino(&entry->vfs_inode))
4061                         p = &parent->rb_left;
4062                 else if (ino > btrfs_ino(&entry->vfs_inode))
4063                         p = &parent->rb_right;
4064                 else {
4065                         WARN_ON(!(entry->vfs_inode.i_state &
4066                                   (I_WILL_FREE | I_FREEING)));
4067                         rb_erase(parent, &root->inode_tree);
4068                         RB_CLEAR_NODE(parent);
4069                         spin_unlock(&root->inode_lock);
4070                         goto again;
4071                 }
4072         }
4073         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4074         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4075         spin_unlock(&root->inode_lock);
4076 }
4077
4078 static void inode_tree_del(struct inode *inode)
4079 {
4080         struct btrfs_root *root = BTRFS_I(inode)->root;
4081         int empty = 0;
4082
4083         spin_lock(&root->inode_lock);
4084         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4085                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4086                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4087                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4088         }
4089         spin_unlock(&root->inode_lock);
4090
4091         /*
4092          * Free space cache has inodes in the tree root, but the tree root has a
4093          * root_refs of 0, so this could end up dropping the tree root as a
4094          * snapshot, so we need the extra !root->fs_info->tree_root check to
4095          * make sure we don't drop it.
4096          */
4097         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4098             root != root->fs_info->tree_root) {
4099                 synchronize_srcu(&root->fs_info->subvol_srcu);
4100                 spin_lock(&root->inode_lock);
4101                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4102                 spin_unlock(&root->inode_lock);
4103                 if (empty)
4104                         btrfs_add_dead_root(root);
4105         }
4106 }
4107
4108 void btrfs_invalidate_inodes(struct btrfs_root *root)
4109 {
4110         struct rb_node *node;
4111         struct rb_node *prev;
4112         struct btrfs_inode *entry;
4113         struct inode *inode;
4114         u64 objectid = 0;
4115
4116         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4117
4118         spin_lock(&root->inode_lock);
4119 again:
4120         node = root->inode_tree.rb_node;
4121         prev = NULL;
4122         while (node) {
4123                 prev = node;
4124                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4125
4126                 if (objectid < btrfs_ino(&entry->vfs_inode))
4127                         node = node->rb_left;
4128                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4129                         node = node->rb_right;
4130                 else
4131                         break;
4132         }
4133         if (!node) {
4134                 while (prev) {
4135                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4136                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4137                                 node = prev;
4138                                 break;
4139                         }
4140                         prev = rb_next(prev);
4141                 }
4142         }
4143         while (node) {
4144                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4145                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4146                 inode = igrab(&entry->vfs_inode);
4147                 if (inode) {
4148                         spin_unlock(&root->inode_lock);
4149                         if (atomic_read(&inode->i_count) > 1)
4150                                 d_prune_aliases(inode);
4151                         /*
4152                          * btrfs_drop_inode will have it removed from
4153                          * the inode cache when its usage count
4154                          * hits zero.
4155                          */
4156                         iput(inode);
4157                         cond_resched();
4158                         spin_lock(&root->inode_lock);
4159                         goto again;
4160                 }
4161
4162                 if (cond_resched_lock(&root->inode_lock))
4163                         goto again;
4164
4165                 node = rb_next(node);
4166         }
4167         spin_unlock(&root->inode_lock);
4168 }
4169
4170 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4171 {
4172         struct btrfs_iget_args *args = p;
4173         inode->i_ino = args->ino;
4174         BTRFS_I(inode)->root = args->root;
4175         return 0;
4176 }
4177
4178 static int btrfs_find_actor(struct inode *inode, void *opaque)
4179 {
4180         struct btrfs_iget_args *args = opaque;
4181         return args->ino == btrfs_ino(inode) &&
4182                 args->root == BTRFS_I(inode)->root;
4183 }
4184
4185 static struct inode *btrfs_iget_locked(struct super_block *s,
4186                                        u64 objectid,
4187                                        struct btrfs_root *root)
4188 {
4189         struct inode *inode;
4190         struct btrfs_iget_args args;
4191         args.ino = objectid;
4192         args.root = root;
4193
4194         inode = iget5_locked(s, objectid, btrfs_find_actor,
4195                              btrfs_init_locked_inode,
4196                              (void *)&args);
4197         return inode;
4198 }
4199
4200 /* Get an inode object given its location and corresponding root.
4201  * Returns in *is_new if the inode was read from disk
4202  */
4203 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4204                          struct btrfs_root *root, int *new)
4205 {
4206         struct inode *inode;
4207
4208         inode = btrfs_iget_locked(s, location->objectid, root);
4209         if (!inode)
4210                 return ERR_PTR(-ENOMEM);
4211
4212         if (inode->i_state & I_NEW) {
4213                 BTRFS_I(inode)->root = root;
4214                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4215                 btrfs_read_locked_inode(inode);
4216                 if (!is_bad_inode(inode)) {
4217                         inode_tree_add(inode);
4218                         unlock_new_inode(inode);
4219                         if (new)
4220                                 *new = 1;
4221                 } else {
4222                         unlock_new_inode(inode);
4223                         iput(inode);
4224                         inode = ERR_PTR(-ESTALE);
4225                 }
4226         }
4227
4228         return inode;
4229 }
4230
4231 static struct inode *new_simple_dir(struct super_block *s,
4232                                     struct btrfs_key *key,
4233                                     struct btrfs_root *root)
4234 {
4235         struct inode *inode = new_inode(s);
4236
4237         if (!inode)
4238                 return ERR_PTR(-ENOMEM);
4239
4240         BTRFS_I(inode)->root = root;
4241         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4242         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4243
4244         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4245         inode->i_op = &btrfs_dir_ro_inode_operations;
4246         inode->i_fop = &simple_dir_operations;
4247         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4248         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4249
4250         return inode;
4251 }
4252
4253 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4254 {
4255         struct inode *inode;
4256         struct btrfs_root *root = BTRFS_I(dir)->root;
4257         struct btrfs_root *sub_root = root;
4258         struct btrfs_key location;
4259         int index;
4260         int ret = 0;
4261
4262         if (dentry->d_name.len > BTRFS_NAME_LEN)
4263                 return ERR_PTR(-ENAMETOOLONG);
4264
4265         ret = btrfs_inode_by_name(dir, dentry, &location);
4266         if (ret < 0)
4267                 return ERR_PTR(ret);
4268
4269         if (location.objectid == 0)
4270                 return NULL;
4271
4272         if (location.type == BTRFS_INODE_ITEM_KEY) {
4273                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4274                 return inode;
4275         }
4276
4277         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4278
4279         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4280         ret = fixup_tree_root_location(root, dir, dentry,
4281                                        &location, &sub_root);
4282         if (ret < 0) {
4283                 if (ret != -ENOENT)
4284                         inode = ERR_PTR(ret);
4285                 else
4286                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4287         } else {
4288                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4289         }
4290         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4291
4292         if (!IS_ERR(inode) && root != sub_root) {
4293                 down_read(&root->fs_info->cleanup_work_sem);
4294                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4295                         ret = btrfs_orphan_cleanup(sub_root);
4296                 up_read(&root->fs_info->cleanup_work_sem);
4297                 if (ret)
4298                         inode = ERR_PTR(ret);
4299         }
4300
4301         return inode;
4302 }
4303
4304 static int btrfs_dentry_delete(const struct dentry *dentry)
4305 {
4306         struct btrfs_root *root;
4307         struct inode *inode = dentry->d_inode;
4308
4309         if (!inode && !IS_ROOT(dentry))
4310                 inode = dentry->d_parent->d_inode;
4311
4312         if (inode) {
4313                 root = BTRFS_I(inode)->root;
4314                 if (btrfs_root_refs(&root->root_item) == 0)
4315                         return 1;
4316
4317                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4318                         return 1;
4319         }
4320         return 0;
4321 }
4322
4323 static void btrfs_dentry_release(struct dentry *dentry)
4324 {
4325         if (dentry->d_fsdata)
4326                 kfree(dentry->d_fsdata);
4327 }
4328
4329 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4330                                    unsigned int flags)
4331 {
4332         struct dentry *ret;
4333
4334         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4335         return ret;
4336 }
4337
4338 unsigned char btrfs_filetype_table[] = {
4339         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4340 };
4341
4342 static int btrfs_real_readdir(struct file *filp, void *dirent,
4343                               filldir_t filldir)
4344 {
4345         struct inode *inode = file_inode(filp);
4346         struct btrfs_root *root = BTRFS_I(inode)->root;
4347         struct btrfs_item *item;
4348         struct btrfs_dir_item *di;
4349         struct btrfs_key key;
4350         struct btrfs_key found_key;
4351         struct btrfs_path *path;
4352         struct list_head ins_list;
4353         struct list_head del_list;
4354         int ret;
4355         struct extent_buffer *leaf;
4356         int slot;
4357         unsigned char d_type;
4358         int over = 0;
4359         u32 di_cur;
4360         u32 di_total;
4361         u32 di_len;
4362         int key_type = BTRFS_DIR_INDEX_KEY;
4363         char tmp_name[32];
4364         char *name_ptr;
4365         int name_len;
4366         int is_curr = 0;        /* filp->f_pos points to the current index? */
4367
4368         /* FIXME, use a real flag for deciding about the key type */
4369         if (root->fs_info->tree_root == root)
4370                 key_type = BTRFS_DIR_ITEM_KEY;
4371
4372         /* special case for "." */
4373         if (filp->f_pos == 0) {
4374                 over = filldir(dirent, ".", 1,
4375                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4376                 if (over)
4377                         return 0;
4378                 filp->f_pos = 1;
4379         }
4380         /* special case for .., just use the back ref */
4381         if (filp->f_pos == 1) {
4382                 u64 pino = parent_ino(filp->f_path.dentry);
4383                 over = filldir(dirent, "..", 2,
4384                                filp->f_pos, pino, DT_DIR);
4385                 if (over)
4386                         return 0;
4387                 filp->f_pos = 2;
4388         }
4389         path = btrfs_alloc_path();
4390         if (!path)
4391                 return -ENOMEM;
4392
4393         path->reada = 1;
4394
4395         if (key_type == BTRFS_DIR_INDEX_KEY) {
4396                 INIT_LIST_HEAD(&ins_list);
4397                 INIT_LIST_HEAD(&del_list);
4398                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4399         }
4400
4401         btrfs_set_key_type(&key, key_type);
4402         key.offset = filp->f_pos;
4403         key.objectid = btrfs_ino(inode);
4404
4405         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4406         if (ret < 0)
4407                 goto err;
4408
4409         while (1) {
4410                 leaf = path->nodes[0];
4411                 slot = path->slots[0];
4412                 if (slot >= btrfs_header_nritems(leaf)) {
4413                         ret = btrfs_next_leaf(root, path);
4414                         if (ret < 0)
4415                                 goto err;
4416                         else if (ret > 0)
4417                                 break;
4418                         continue;
4419                 }
4420
4421                 item = btrfs_item_nr(leaf, slot);
4422                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4423
4424                 if (found_key.objectid != key.objectid)
4425                         break;
4426                 if (btrfs_key_type(&found_key) != key_type)
4427                         break;
4428                 if (found_key.offset < filp->f_pos)
4429                         goto next;
4430                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4431                     btrfs_should_delete_dir_index(&del_list,
4432                                                   found_key.offset))
4433                         goto next;
4434
4435                 filp->f_pos = found_key.offset;
4436                 is_curr = 1;
4437
4438                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4439                 di_cur = 0;
4440                 di_total = btrfs_item_size(leaf, item);
4441
4442                 while (di_cur < di_total) {
4443                         struct btrfs_key location;
4444
4445                         if (verify_dir_item(root, leaf, di))
4446                                 break;
4447
4448                         name_len = btrfs_dir_name_len(leaf, di);
4449                         if (name_len <= sizeof(tmp_name)) {
4450                                 name_ptr = tmp_name;
4451                         } else {
4452                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4453                                 if (!name_ptr) {
4454                                         ret = -ENOMEM;
4455                                         goto err;
4456                                 }
4457                         }
4458                         read_extent_buffer(leaf, name_ptr,
4459                                            (unsigned long)(di + 1), name_len);
4460
4461                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4462                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4463
4464
4465                         /* is this a reference to our own snapshot? If so
4466                          * skip it.
4467                          *
4468                          * In contrast to old kernels, we insert the snapshot's
4469                          * dir item and dir index after it has been created, so
4470                          * we won't find a reference to our own snapshot. We
4471                          * still keep the following code for backward
4472                          * compatibility.
4473                          */
4474                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4475                             location.objectid == root->root_key.objectid) {
4476                                 over = 0;
4477                                 goto skip;
4478                         }
4479                         over = filldir(dirent, name_ptr, name_len,
4480                                        found_key.offset, location.objectid,
4481                                        d_type);
4482
4483 skip:
4484                         if (name_ptr != tmp_name)
4485                                 kfree(name_ptr);
4486
4487                         if (over)
4488                                 goto nopos;
4489                         di_len = btrfs_dir_name_len(leaf, di) +
4490                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4491                         di_cur += di_len;
4492                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4493                 }
4494 next:
4495                 path->slots[0]++;
4496         }
4497
4498         if (key_type == BTRFS_DIR_INDEX_KEY) {
4499                 if (is_curr)
4500                         filp->f_pos++;
4501                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4502                                                       &ins_list);
4503                 if (ret)
4504                         goto nopos;
4505         }
4506
4507         /* Reached end of directory/root. Bump pos past the last item. */
4508         if (key_type == BTRFS_DIR_INDEX_KEY)
4509                 /*
4510                  * 32-bit glibc will use getdents64, but then strtol -
4511                  * so the last number we can serve is this.
4512                  */
4513                 filp->f_pos = 0x7fffffff;
4514         else
4515                 filp->f_pos++;
4516 nopos:
4517         ret = 0;
4518 err:
4519         if (key_type == BTRFS_DIR_INDEX_KEY)
4520                 btrfs_put_delayed_items(&ins_list, &del_list);
4521         btrfs_free_path(path);
4522         return ret;
4523 }
4524
4525 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4526 {
4527         struct btrfs_root *root = BTRFS_I(inode)->root;
4528         struct btrfs_trans_handle *trans;
4529         int ret = 0;
4530         bool nolock = false;
4531
4532         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4533                 return 0;
4534
4535         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4536                 nolock = true;
4537
4538         if (wbc->sync_mode == WB_SYNC_ALL) {
4539                 if (nolock)
4540                         trans = btrfs_join_transaction_nolock(root);
4541                 else
4542                         trans = btrfs_join_transaction(root);
4543                 if (IS_ERR(trans))
4544                         return PTR_ERR(trans);
4545                 ret = btrfs_commit_transaction(trans, root);
4546         }
4547         return ret;
4548 }
4549
4550 /*
4551  * This is somewhat expensive, updating the tree every time the
4552  * inode changes.  But, it is most likely to find the inode in cache.
4553  * FIXME, needs more benchmarking...there are no reasons other than performance
4554  * to keep or drop this code.
4555  */
4556 int btrfs_dirty_inode(struct inode *inode)
4557 {
4558         struct btrfs_root *root = BTRFS_I(inode)->root;
4559         struct btrfs_trans_handle *trans;
4560         int ret;
4561
4562         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4563                 return 0;
4564
4565         trans = btrfs_join_transaction(root);
4566         if (IS_ERR(trans))
4567                 return PTR_ERR(trans);
4568
4569         ret = btrfs_update_inode(trans, root, inode);
4570         if (ret && ret == -ENOSPC) {
4571                 /* whoops, lets try again with the full transaction */
4572                 btrfs_end_transaction(trans, root);
4573                 trans = btrfs_start_transaction(root, 1);
4574                 if (IS_ERR(trans))
4575                         return PTR_ERR(trans);
4576
4577                 ret = btrfs_update_inode(trans, root, inode);
4578         }
4579         btrfs_end_transaction(trans, root);
4580         if (BTRFS_I(inode)->delayed_node)
4581                 btrfs_balance_delayed_items(root);
4582
4583         return ret;
4584 }
4585
4586 /*
4587  * This is a copy of file_update_time.  We need this so we can return error on
4588  * ENOSPC for updating the inode in the case of file write and mmap writes.
4589  */
4590 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4591                              int flags)
4592 {
4593         struct btrfs_root *root = BTRFS_I(inode)->root;
4594
4595         if (btrfs_root_readonly(root))
4596                 return -EROFS;
4597
4598         if (flags & S_VERSION)
4599                 inode_inc_iversion(inode);
4600         if (flags & S_CTIME)
4601                 inode->i_ctime = *now;
4602         if (flags & S_MTIME)
4603                 inode->i_mtime = *now;
4604         if (flags & S_ATIME)
4605                 inode->i_atime = *now;
4606         return btrfs_dirty_inode(inode);
4607 }
4608
4609 /*
4610  * find the highest existing sequence number in a directory
4611  * and then set the in-memory index_cnt variable to reflect
4612  * free sequence numbers
4613  */
4614 static int btrfs_set_inode_index_count(struct inode *inode)
4615 {
4616         struct btrfs_root *root = BTRFS_I(inode)->root;
4617         struct btrfs_key key, found_key;
4618         struct btrfs_path *path;
4619         struct extent_buffer *leaf;
4620         int ret;
4621
4622         key.objectid = btrfs_ino(inode);
4623         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4624         key.offset = (u64)-1;
4625
4626         path = btrfs_alloc_path();
4627         if (!path)
4628                 return -ENOMEM;
4629
4630         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4631         if (ret < 0)
4632                 goto out;
4633         /* FIXME: we should be able to handle this */
4634         if (ret == 0)
4635                 goto out;
4636         ret = 0;
4637
4638         /*
4639          * MAGIC NUMBER EXPLANATION:
4640          * since we search a directory based on f_pos we have to start at 2
4641          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4642          * else has to start at 2
4643          */
4644         if (path->slots[0] == 0) {
4645                 BTRFS_I(inode)->index_cnt = 2;
4646                 goto out;
4647         }
4648
4649         path->slots[0]--;
4650
4651         leaf = path->nodes[0];
4652         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4653
4654         if (found_key.objectid != btrfs_ino(inode) ||
4655             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4656                 BTRFS_I(inode)->index_cnt = 2;
4657                 goto out;
4658         }
4659
4660         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4661 out:
4662         btrfs_free_path(path);
4663         return ret;
4664 }
4665
4666 /*
4667  * helper to find a free sequence number in a given directory.  This current
4668  * code is very simple, later versions will do smarter things in the btree
4669  */
4670 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4671 {
4672         int ret = 0;
4673
4674         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4675                 ret = btrfs_inode_delayed_dir_index_count(dir);
4676                 if (ret) {
4677                         ret = btrfs_set_inode_index_count(dir);
4678                         if (ret)
4679                                 return ret;
4680                 }
4681         }
4682
4683         *index = BTRFS_I(dir)->index_cnt;
4684         BTRFS_I(dir)->index_cnt++;
4685
4686         return ret;
4687 }
4688
4689 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4690                                      struct btrfs_root *root,
4691                                      struct inode *dir,
4692                                      const char *name, int name_len,
4693                                      u64 ref_objectid, u64 objectid,
4694                                      umode_t mode, u64 *index)
4695 {
4696         struct inode *inode;
4697         struct btrfs_inode_item *inode_item;
4698         struct btrfs_key *location;
4699         struct btrfs_path *path;
4700         struct btrfs_inode_ref *ref;
4701         struct btrfs_key key[2];
4702         u32 sizes[2];
4703         unsigned long ptr;
4704         int ret;
4705         int owner;
4706
4707         path = btrfs_alloc_path();
4708         if (!path)
4709                 return ERR_PTR(-ENOMEM);
4710
4711         inode = new_inode(root->fs_info->sb);
4712         if (!inode) {
4713                 btrfs_free_path(path);
4714                 return ERR_PTR(-ENOMEM);
4715         }
4716
4717         /*
4718          * we have to initialize this early, so we can reclaim the inode
4719          * number if we fail afterwards in this function.
4720          */
4721         inode->i_ino = objectid;
4722
4723         if (dir) {
4724                 trace_btrfs_inode_request(dir);
4725
4726                 ret = btrfs_set_inode_index(dir, index);
4727                 if (ret) {
4728                         btrfs_free_path(path);
4729                         iput(inode);
4730                         return ERR_PTR(ret);
4731                 }
4732         }
4733         /*
4734          * index_cnt is ignored for everything but a dir,
4735          * btrfs_get_inode_index_count has an explanation for the magic
4736          * number
4737          */
4738         BTRFS_I(inode)->index_cnt = 2;
4739         BTRFS_I(inode)->root = root;
4740         BTRFS_I(inode)->generation = trans->transid;
4741         inode->i_generation = BTRFS_I(inode)->generation;
4742
4743         /*
4744          * We could have gotten an inode number from somebody who was fsynced
4745          * and then removed in this same transaction, so let's just set full
4746          * sync since it will be a full sync anyway and this will blow away the
4747          * old info in the log.
4748          */
4749         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4750
4751         if (S_ISDIR(mode))
4752                 owner = 0;
4753         else
4754                 owner = 1;
4755
4756         key[0].objectid = objectid;
4757         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4758         key[0].offset = 0;
4759
4760         /*
4761          * Start new inodes with an inode_ref. This is slightly more
4762          * efficient for small numbers of hard links since they will
4763          * be packed into one item. Extended refs will kick in if we
4764          * add more hard links than can fit in the ref item.
4765          */
4766         key[1].objectid = objectid;
4767         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4768         key[1].offset = ref_objectid;
4769
4770         sizes[0] = sizeof(struct btrfs_inode_item);
4771         sizes[1] = name_len + sizeof(*ref);
4772
4773         path->leave_spinning = 1;
4774         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4775         if (ret != 0)
4776                 goto fail;
4777
4778         inode_init_owner(inode, dir, mode);
4779         inode_set_bytes(inode, 0);
4780         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4781         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4782                                   struct btrfs_inode_item);
4783         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4784                              sizeof(*inode_item));
4785         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4786
4787         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4788                              struct btrfs_inode_ref);
4789         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4790         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4791         ptr = (unsigned long)(ref + 1);
4792         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4793
4794         btrfs_mark_buffer_dirty(path->nodes[0]);
4795         btrfs_free_path(path);
4796
4797         location = &BTRFS_I(inode)->location;
4798         location->objectid = objectid;
4799         location->offset = 0;
4800         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4801
4802         btrfs_inherit_iflags(inode, dir);
4803
4804         if (S_ISREG(mode)) {
4805                 if (btrfs_test_opt(root, NODATASUM))
4806                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4807                 if (btrfs_test_opt(root, NODATACOW))
4808                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4809         }
4810
4811         insert_inode_hash(inode);
4812         inode_tree_add(inode);
4813
4814         trace_btrfs_inode_new(inode);
4815         btrfs_set_inode_last_trans(trans, inode);
4816
4817         btrfs_update_root_times(trans, root);
4818
4819         return inode;
4820 fail:
4821         if (dir)
4822                 BTRFS_I(dir)->index_cnt--;
4823         btrfs_free_path(path);
4824         iput(inode);
4825         return ERR_PTR(ret);
4826 }
4827
4828 static inline u8 btrfs_inode_type(struct inode *inode)
4829 {
4830         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4831 }
4832
4833 /*
4834  * utility function to add 'inode' into 'parent_inode' with
4835  * a give name and a given sequence number.
4836  * if 'add_backref' is true, also insert a backref from the
4837  * inode to the parent directory.
4838  */
4839 int btrfs_add_link(struct btrfs_trans_handle *trans,
4840                    struct inode *parent_inode, struct inode *inode,
4841                    const char *name, int name_len, int add_backref, u64 index)
4842 {
4843         int ret = 0;
4844         struct btrfs_key key;
4845         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4846         u64 ino = btrfs_ino(inode);
4847         u64 parent_ino = btrfs_ino(parent_inode);
4848
4849         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4850                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4851         } else {
4852                 key.objectid = ino;
4853                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4854                 key.offset = 0;
4855         }
4856
4857         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4858                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4859                                          key.objectid, root->root_key.objectid,
4860                                          parent_ino, index, name, name_len);
4861         } else if (add_backref) {
4862                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4863                                              parent_ino, index);
4864         }
4865
4866         /* Nothing to clean up yet */
4867         if (ret)
4868                 return ret;
4869
4870         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4871                                     parent_inode, &key,
4872                                     btrfs_inode_type(inode), index);
4873         if (ret == -EEXIST || ret == -EOVERFLOW)
4874                 goto fail_dir_item;
4875         else if (ret) {
4876                 btrfs_abort_transaction(trans, root, ret);
4877                 return ret;
4878         }
4879
4880         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4881                            name_len * 2);
4882         inode_inc_iversion(parent_inode);
4883         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4884         ret = btrfs_update_inode(trans, root, parent_inode);
4885         if (ret)
4886                 btrfs_abort_transaction(trans, root, ret);
4887         return ret;
4888
4889 fail_dir_item:
4890         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4891                 u64 local_index;
4892                 int err;
4893                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4894                                  key.objectid, root->root_key.objectid,
4895                                  parent_ino, &local_index, name, name_len);
4896
4897         } else if (add_backref) {
4898                 u64 local_index;
4899                 int err;
4900
4901                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4902                                           ino, parent_ino, &local_index);
4903         }
4904         return ret;
4905 }
4906
4907 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4908                             struct inode *dir, struct dentry *dentry,
4909                             struct inode *inode, int backref, u64 index)
4910 {
4911         int err = btrfs_add_link(trans, dir, inode,
4912                                  dentry->d_name.name, dentry->d_name.len,
4913                                  backref, index);
4914         if (err > 0)
4915                 err = -EEXIST;
4916         return err;
4917 }
4918
4919 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4920                         umode_t mode, dev_t rdev)
4921 {
4922         struct btrfs_trans_handle *trans;
4923         struct btrfs_root *root = BTRFS_I(dir)->root;
4924         struct inode *inode = NULL;
4925         int err;
4926         int drop_inode = 0;
4927         u64 objectid;
4928         u64 index = 0;
4929
4930         if (!new_valid_dev(rdev))
4931                 return -EINVAL;
4932
4933         /*
4934          * 2 for inode item and ref
4935          * 2 for dir items
4936          * 1 for xattr if selinux is on
4937          */
4938         trans = btrfs_start_transaction(root, 5);
4939         if (IS_ERR(trans))
4940                 return PTR_ERR(trans);
4941
4942         err = btrfs_find_free_ino(root, &objectid);
4943         if (err)
4944                 goto out_unlock;
4945
4946         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4947                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4948                                 mode, &index);
4949         if (IS_ERR(inode)) {
4950                 err = PTR_ERR(inode);
4951                 goto out_unlock;
4952         }
4953
4954         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4955         if (err) {
4956                 drop_inode = 1;
4957                 goto out_unlock;
4958         }
4959
4960         err = btrfs_update_inode(trans, root, inode);
4961         if (err) {
4962                 drop_inode = 1;
4963                 goto out_unlock;
4964         }
4965
4966         /*
4967         * If the active LSM wants to access the inode during
4968         * d_instantiate it needs these. Smack checks to see
4969         * if the filesystem supports xattrs by looking at the
4970         * ops vector.
4971         */
4972
4973         inode->i_op = &btrfs_special_inode_operations;
4974         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4975         if (err)
4976                 drop_inode = 1;
4977         else {
4978                 init_special_inode(inode, inode->i_mode, rdev);
4979                 btrfs_update_inode(trans, root, inode);
4980                 d_instantiate(dentry, inode);
4981         }
4982 out_unlock:
4983         btrfs_end_transaction(trans, root);
4984         btrfs_btree_balance_dirty(root);
4985         if (drop_inode) {
4986                 inode_dec_link_count(inode);
4987                 iput(inode);
4988         }
4989         return err;
4990 }
4991
4992 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4993                         umode_t mode, bool excl)
4994 {
4995         struct btrfs_trans_handle *trans;
4996         struct btrfs_root *root = BTRFS_I(dir)->root;
4997         struct inode *inode = NULL;
4998         int drop_inode_on_err = 0;
4999         int err;
5000         u64 objectid;
5001         u64 index = 0;
5002
5003         /*
5004          * 2 for inode item and ref
5005          * 2 for dir items
5006          * 1 for xattr if selinux is on
5007          */
5008         trans = btrfs_start_transaction(root, 5);
5009         if (IS_ERR(trans))
5010                 return PTR_ERR(trans);
5011
5012         err = btrfs_find_free_ino(root, &objectid);
5013         if (err)
5014                 goto out_unlock;
5015
5016         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5017                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5018                                 mode, &index);
5019         if (IS_ERR(inode)) {
5020                 err = PTR_ERR(inode);
5021                 goto out_unlock;
5022         }
5023         drop_inode_on_err = 1;
5024
5025         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5026         if (err)
5027                 goto out_unlock;
5028
5029         err = btrfs_update_inode(trans, root, inode);
5030         if (err)
5031                 goto out_unlock;
5032
5033         /*
5034         * If the active LSM wants to access the inode during
5035         * d_instantiate it needs these. Smack checks to see
5036         * if the filesystem supports xattrs by looking at the
5037         * ops vector.
5038         */
5039         inode->i_fop = &btrfs_file_operations;
5040         inode->i_op = &btrfs_file_inode_operations;
5041
5042         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5043         if (err)
5044                 goto out_unlock;
5045
5046         inode->i_mapping->a_ops = &btrfs_aops;
5047         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5048         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5049         d_instantiate(dentry, inode);
5050
5051 out_unlock:
5052         btrfs_end_transaction(trans, root);
5053         if (err && drop_inode_on_err) {
5054                 inode_dec_link_count(inode);
5055                 iput(inode);
5056         }
5057         btrfs_btree_balance_dirty(root);
5058         return err;
5059 }
5060
5061 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5062                       struct dentry *dentry)
5063 {
5064         struct btrfs_trans_handle *trans;
5065         struct btrfs_root *root = BTRFS_I(dir)->root;
5066         struct inode *inode = old_dentry->d_inode;
5067         u64 index;
5068         int err;
5069         int drop_inode = 0;
5070
5071         /* do not allow sys_link's with other subvols of the same device */
5072         if (root->objectid != BTRFS_I(inode)->root->objectid)
5073                 return -EXDEV;
5074
5075         if (inode->i_nlink >= BTRFS_LINK_MAX)
5076                 return -EMLINK;
5077
5078         err = btrfs_set_inode_index(dir, &index);
5079         if (err)
5080                 goto fail;
5081
5082         /*
5083          * 2 items for inode and inode ref
5084          * 2 items for dir items
5085          * 1 item for parent inode
5086          */
5087         trans = btrfs_start_transaction(root, 5);
5088         if (IS_ERR(trans)) {
5089                 err = PTR_ERR(trans);
5090                 goto fail;
5091         }
5092
5093         btrfs_inc_nlink(inode);
5094         inode_inc_iversion(inode);
5095         inode->i_ctime = CURRENT_TIME;
5096         ihold(inode);
5097         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5098
5099         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5100
5101         if (err) {
5102                 drop_inode = 1;
5103         } else {
5104                 struct dentry *parent = dentry->d_parent;
5105                 err = btrfs_update_inode(trans, root, inode);
5106                 if (err)
5107                         goto fail;
5108                 d_instantiate(dentry, inode);
5109                 btrfs_log_new_name(trans, inode, NULL, parent);
5110         }
5111
5112         btrfs_end_transaction(trans, root);
5113 fail:
5114         if (drop_inode) {
5115                 inode_dec_link_count(inode);
5116                 iput(inode);
5117         }
5118         btrfs_btree_balance_dirty(root);
5119         return err;
5120 }
5121
5122 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5123 {
5124         struct inode *inode = NULL;
5125         struct btrfs_trans_handle *trans;
5126         struct btrfs_root *root = BTRFS_I(dir)->root;
5127         int err = 0;
5128         int drop_on_err = 0;
5129         u64 objectid = 0;
5130         u64 index = 0;
5131
5132         /*
5133          * 2 items for inode and ref
5134          * 2 items for dir items
5135          * 1 for xattr if selinux is on
5136          */
5137         trans = btrfs_start_transaction(root, 5);
5138         if (IS_ERR(trans))
5139                 return PTR_ERR(trans);
5140
5141         err = btrfs_find_free_ino(root, &objectid);
5142         if (err)
5143                 goto out_fail;
5144
5145         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5146                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5147                                 S_IFDIR | mode, &index);
5148         if (IS_ERR(inode)) {
5149                 err = PTR_ERR(inode);
5150                 goto out_fail;
5151         }
5152
5153         drop_on_err = 1;
5154
5155         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5156         if (err)
5157                 goto out_fail;
5158
5159         inode->i_op = &btrfs_dir_inode_operations;
5160         inode->i_fop = &btrfs_dir_file_operations;
5161
5162         btrfs_i_size_write(inode, 0);
5163         err = btrfs_update_inode(trans, root, inode);
5164         if (err)
5165                 goto out_fail;
5166
5167         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5168                              dentry->d_name.len, 0, index);
5169         if (err)
5170                 goto out_fail;
5171
5172         d_instantiate(dentry, inode);
5173         drop_on_err = 0;
5174
5175 out_fail:
5176         btrfs_end_transaction(trans, root);
5177         if (drop_on_err)
5178                 iput(inode);
5179         btrfs_btree_balance_dirty(root);
5180         return err;
5181 }
5182
5183 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5184  * and an extent that you want to insert, deal with overlap and insert
5185  * the new extent into the tree.
5186  */
5187 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5188                                 struct extent_map *existing,
5189                                 struct extent_map *em,
5190                                 u64 map_start, u64 map_len)
5191 {
5192         u64 start_diff;
5193
5194         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5195         start_diff = map_start - em->start;
5196         em->start = map_start;
5197         em->len = map_len;
5198         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5199             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5200                 em->block_start += start_diff;
5201                 em->block_len -= start_diff;
5202         }
5203         return add_extent_mapping(em_tree, em);
5204 }
5205
5206 static noinline int uncompress_inline(struct btrfs_path *path,
5207                                       struct inode *inode, struct page *page,
5208                                       size_t pg_offset, u64 extent_offset,
5209                                       struct btrfs_file_extent_item *item)
5210 {
5211         int ret;
5212         struct extent_buffer *leaf = path->nodes[0];
5213         char *tmp;
5214         size_t max_size;
5215         unsigned long inline_size;
5216         unsigned long ptr;
5217         int compress_type;
5218
5219         WARN_ON(pg_offset != 0);
5220         compress_type = btrfs_file_extent_compression(leaf, item);
5221         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5222         inline_size = btrfs_file_extent_inline_item_len(leaf,
5223                                         btrfs_item_nr(leaf, path->slots[0]));
5224         tmp = kmalloc(inline_size, GFP_NOFS);
5225         if (!tmp)
5226                 return -ENOMEM;
5227         ptr = btrfs_file_extent_inline_start(item);
5228
5229         read_extent_buffer(leaf, tmp, ptr, inline_size);
5230
5231         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5232         ret = btrfs_decompress(compress_type, tmp, page,
5233                                extent_offset, inline_size, max_size);
5234         if (ret) {
5235                 char *kaddr = kmap_atomic(page);
5236                 unsigned long copy_size = min_t(u64,
5237                                   PAGE_CACHE_SIZE - pg_offset,
5238                                   max_size - extent_offset);
5239                 memset(kaddr + pg_offset, 0, copy_size);
5240                 kunmap_atomic(kaddr);
5241         }
5242         kfree(tmp);
5243         return 0;
5244 }
5245
5246 /*
5247  * a bit scary, this does extent mapping from logical file offset to the disk.
5248  * the ugly parts come from merging extents from the disk with the in-ram
5249  * representation.  This gets more complex because of the data=ordered code,
5250  * where the in-ram extents might be locked pending data=ordered completion.
5251  *
5252  * This also copies inline extents directly into the page.
5253  */
5254
5255 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5256                                     size_t pg_offset, u64 start, u64 len,
5257                                     int create)
5258 {
5259         int ret;
5260         int err = 0;
5261         u64 bytenr;
5262         u64 extent_start = 0;
5263         u64 extent_end = 0;
5264         u64 objectid = btrfs_ino(inode);
5265         u32 found_type;
5266         struct btrfs_path *path = NULL;
5267         struct btrfs_root *root = BTRFS_I(inode)->root;
5268         struct btrfs_file_extent_item *item;
5269         struct extent_buffer *leaf;
5270         struct btrfs_key found_key;
5271         struct extent_map *em = NULL;
5272         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5273         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5274         struct btrfs_trans_handle *trans = NULL;
5275         int compress_type;
5276
5277 again:
5278         read_lock(&em_tree->lock);
5279         em = lookup_extent_mapping(em_tree, start, len);
5280         if (em)
5281                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5282         read_unlock(&em_tree->lock);
5283
5284         if (em) {
5285                 if (em->start > start || em->start + em->len <= start)
5286                         free_extent_map(em);
5287                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5288                         free_extent_map(em);
5289                 else
5290                         goto out;
5291         }
5292         em = alloc_extent_map();
5293         if (!em) {
5294                 err = -ENOMEM;
5295                 goto out;
5296         }
5297         em->bdev = root->fs_info->fs_devices->latest_bdev;
5298         em->start = EXTENT_MAP_HOLE;
5299         em->orig_start = EXTENT_MAP_HOLE;
5300         em->len = (u64)-1;
5301         em->block_len = (u64)-1;
5302
5303         if (!path) {
5304                 path = btrfs_alloc_path();
5305                 if (!path) {
5306                         err = -ENOMEM;
5307                         goto out;
5308                 }
5309                 /*
5310                  * Chances are we'll be called again, so go ahead and do
5311                  * readahead
5312                  */
5313                 path->reada = 1;
5314         }
5315
5316         ret = btrfs_lookup_file_extent(trans, root, path,
5317                                        objectid, start, trans != NULL);
5318         if (ret < 0) {
5319                 err = ret;
5320                 goto out;
5321         }
5322
5323         if (ret != 0) {
5324                 if (path->slots[0] == 0)
5325                         goto not_found;
5326                 path->slots[0]--;
5327         }
5328
5329         leaf = path->nodes[0];
5330         item = btrfs_item_ptr(leaf, path->slots[0],
5331                               struct btrfs_file_extent_item);
5332         /* are we inside the extent that was found? */
5333         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5334         found_type = btrfs_key_type(&found_key);
5335         if (found_key.objectid != objectid ||
5336             found_type != BTRFS_EXTENT_DATA_KEY) {
5337                 goto not_found;
5338         }
5339
5340         found_type = btrfs_file_extent_type(leaf, item);
5341         extent_start = found_key.offset;
5342         compress_type = btrfs_file_extent_compression(leaf, item);
5343         if (found_type == BTRFS_FILE_EXTENT_REG ||
5344             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5345                 extent_end = extent_start +
5346                        btrfs_file_extent_num_bytes(leaf, item);
5347         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5348                 size_t size;
5349                 size = btrfs_file_extent_inline_len(leaf, item);
5350                 extent_end = (extent_start + size + root->sectorsize - 1) &
5351                         ~((u64)root->sectorsize - 1);
5352         }
5353
5354         if (start >= extent_end) {
5355                 path->slots[0]++;
5356                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5357                         ret = btrfs_next_leaf(root, path);
5358                         if (ret < 0) {
5359                                 err = ret;
5360                                 goto out;
5361                         }
5362                         if (ret > 0)
5363                                 goto not_found;
5364                         leaf = path->nodes[0];
5365                 }
5366                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5367                 if (found_key.objectid != objectid ||
5368                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5369                         goto not_found;
5370                 if (start + len <= found_key.offset)
5371                         goto not_found;
5372                 em->start = start;
5373                 em->orig_start = start;
5374                 em->len = found_key.offset - start;
5375                 goto not_found_em;
5376         }
5377
5378         if (found_type == BTRFS_FILE_EXTENT_REG ||
5379             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5380                 em->start = extent_start;
5381                 em->len = extent_end - extent_start;
5382                 em->orig_start = extent_start -
5383                                  btrfs_file_extent_offset(leaf, item);
5384                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5385                                                                       item);
5386                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5387                 if (bytenr == 0) {
5388                         em->block_start = EXTENT_MAP_HOLE;
5389                         goto insert;
5390                 }
5391                 if (compress_type != BTRFS_COMPRESS_NONE) {
5392                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5393                         em->compress_type = compress_type;
5394                         em->block_start = bytenr;
5395                         em->block_len = em->orig_block_len;
5396                 } else {
5397                         bytenr += btrfs_file_extent_offset(leaf, item);
5398                         em->block_start = bytenr;
5399                         em->block_len = em->len;
5400                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5401                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5402                 }
5403                 goto insert;
5404         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5405                 unsigned long ptr;
5406                 char *map;
5407                 size_t size;
5408                 size_t extent_offset;
5409                 size_t copy_size;
5410
5411                 em->block_start = EXTENT_MAP_INLINE;
5412                 if (!page || create) {
5413                         em->start = extent_start;
5414                         em->len = extent_end - extent_start;
5415                         goto out;
5416                 }
5417
5418                 size = btrfs_file_extent_inline_len(leaf, item);
5419                 extent_offset = page_offset(page) + pg_offset - extent_start;
5420                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5421                                 size - extent_offset);
5422                 em->start = extent_start + extent_offset;
5423                 em->len = (copy_size + root->sectorsize - 1) &
5424                         ~((u64)root->sectorsize - 1);
5425                 em->orig_block_len = em->len;
5426                 em->orig_start = em->start;
5427                 if (compress_type) {
5428                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5429                         em->compress_type = compress_type;
5430                 }
5431                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5432                 if (create == 0 && !PageUptodate(page)) {
5433                         if (btrfs_file_extent_compression(leaf, item) !=
5434                             BTRFS_COMPRESS_NONE) {
5435                                 ret = uncompress_inline(path, inode, page,
5436                                                         pg_offset,
5437                                                         extent_offset, item);
5438                                 BUG_ON(ret); /* -ENOMEM */
5439                         } else {
5440                                 map = kmap(page);
5441                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5442                                                    copy_size);
5443                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5444                                         memset(map + pg_offset + copy_size, 0,
5445                                                PAGE_CACHE_SIZE - pg_offset -
5446                                                copy_size);
5447                                 }
5448                                 kunmap(page);
5449                         }
5450                         flush_dcache_page(page);
5451                 } else if (create && PageUptodate(page)) {
5452                         BUG();
5453                         if (!trans) {
5454                                 kunmap(page);
5455                                 free_extent_map(em);
5456                                 em = NULL;
5457
5458                                 btrfs_release_path(path);
5459                                 trans = btrfs_join_transaction(root);
5460
5461                                 if (IS_ERR(trans))
5462                                         return ERR_CAST(trans);
5463                                 goto again;
5464                         }
5465                         map = kmap(page);
5466                         write_extent_buffer(leaf, map + pg_offset, ptr,
5467                                             copy_size);
5468                         kunmap(page);
5469                         btrfs_mark_buffer_dirty(leaf);
5470                 }
5471                 set_extent_uptodate(io_tree, em->start,
5472                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5473                 goto insert;
5474         } else {
5475                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5476         }
5477 not_found:
5478         em->start = start;
5479         em->orig_start = start;
5480         em->len = len;
5481 not_found_em:
5482         em->block_start = EXTENT_MAP_HOLE;
5483         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5484 insert:
5485         btrfs_release_path(path);
5486         if (em->start > start || extent_map_end(em) <= start) {
5487                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5488                        "[%llu %llu]\n", (unsigned long long)em->start,
5489                        (unsigned long long)em->len,
5490                        (unsigned long long)start,
5491                        (unsigned long long)len);
5492                 err = -EIO;
5493                 goto out;
5494         }
5495
5496         err = 0;
5497         write_lock(&em_tree->lock);
5498         ret = add_extent_mapping(em_tree, em);
5499         /* it is possible that someone inserted the extent into the tree
5500          * while we had the lock dropped.  It is also possible that
5501          * an overlapping map exists in the tree
5502          */
5503         if (ret == -EEXIST) {
5504                 struct extent_map *existing;
5505
5506                 ret = 0;
5507
5508                 existing = lookup_extent_mapping(em_tree, start, len);
5509                 if (existing && (existing->start > start ||
5510                     existing->start + existing->len <= start)) {
5511                         free_extent_map(existing);
5512                         existing = NULL;
5513                 }
5514                 if (!existing) {
5515                         existing = lookup_extent_mapping(em_tree, em->start,
5516                                                          em->len);
5517                         if (existing) {
5518                                 err = merge_extent_mapping(em_tree, existing,
5519                                                            em, start,
5520                                                            root->sectorsize);
5521                                 free_extent_map(existing);
5522                                 if (err) {
5523                                         free_extent_map(em);
5524                                         em = NULL;
5525                                 }
5526                         } else {
5527                                 err = -EIO;
5528                                 free_extent_map(em);
5529                                 em = NULL;
5530                         }
5531                 } else {
5532                         free_extent_map(em);
5533                         em = existing;
5534                         err = 0;
5535                 }
5536         }
5537         write_unlock(&em_tree->lock);
5538 out:
5539
5540         if (em)
5541                 trace_btrfs_get_extent(root, em);
5542
5543         if (path)
5544                 btrfs_free_path(path);
5545         if (trans) {
5546                 ret = btrfs_end_transaction(trans, root);
5547                 if (!err)
5548                         err = ret;
5549         }
5550         if (err) {
5551                 free_extent_map(em);
5552                 return ERR_PTR(err);
5553         }
5554         BUG_ON(!em); /* Error is always set */
5555         return em;
5556 }
5557
5558 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5559                                            size_t pg_offset, u64 start, u64 len,
5560                                            int create)
5561 {
5562         struct extent_map *em;
5563         struct extent_map *hole_em = NULL;
5564         u64 range_start = start;
5565         u64 end;
5566         u64 found;
5567         u64 found_end;
5568         int err = 0;
5569
5570         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5571         if (IS_ERR(em))
5572                 return em;
5573         if (em) {
5574                 /*
5575                  * if our em maps to a hole, there might
5576                  * actually be delalloc bytes behind it
5577                  */
5578                 if (em->block_start != EXTENT_MAP_HOLE)
5579                         return em;
5580                 else
5581                         hole_em = em;
5582         }
5583
5584         /* check to see if we've wrapped (len == -1 or similar) */
5585         end = start + len;
5586         if (end < start)
5587                 end = (u64)-1;
5588         else
5589                 end -= 1;
5590
5591         em = NULL;
5592
5593         /* ok, we didn't find anything, lets look for delalloc */
5594         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5595                                  end, len, EXTENT_DELALLOC, 1);
5596         found_end = range_start + found;
5597         if (found_end < range_start)
5598                 found_end = (u64)-1;
5599
5600         /*
5601          * we didn't find anything useful, return
5602          * the original results from get_extent()
5603          */
5604         if (range_start > end || found_end <= start) {
5605                 em = hole_em;
5606                 hole_em = NULL;
5607                 goto out;
5608         }
5609
5610         /* adjust the range_start to make sure it doesn't
5611          * go backwards from the start they passed in
5612          */
5613         range_start = max(start,range_start);
5614         found = found_end - range_start;
5615
5616         if (found > 0) {
5617                 u64 hole_start = start;
5618                 u64 hole_len = len;
5619
5620                 em = alloc_extent_map();
5621                 if (!em) {
5622                         err = -ENOMEM;
5623                         goto out;
5624                 }
5625                 /*
5626                  * when btrfs_get_extent can't find anything it
5627                  * returns one huge hole
5628                  *
5629                  * make sure what it found really fits our range, and
5630                  * adjust to make sure it is based on the start from
5631                  * the caller
5632                  */
5633                 if (hole_em) {
5634                         u64 calc_end = extent_map_end(hole_em);
5635
5636                         if (calc_end <= start || (hole_em->start > end)) {
5637                                 free_extent_map(hole_em);
5638                                 hole_em = NULL;
5639                         } else {
5640                                 hole_start = max(hole_em->start, start);
5641                                 hole_len = calc_end - hole_start;
5642                         }
5643                 }
5644                 em->bdev = NULL;
5645                 if (hole_em && range_start > hole_start) {
5646                         /* our hole starts before our delalloc, so we
5647                          * have to return just the parts of the hole
5648                          * that go until  the delalloc starts
5649                          */
5650                         em->len = min(hole_len,
5651                                       range_start - hole_start);
5652                         em->start = hole_start;
5653                         em->orig_start = hole_start;
5654                         /*
5655                          * don't adjust block start at all,
5656                          * it is fixed at EXTENT_MAP_HOLE
5657                          */
5658                         em->block_start = hole_em->block_start;
5659                         em->block_len = hole_len;
5660                 } else {
5661                         em->start = range_start;
5662                         em->len = found;
5663                         em->orig_start = range_start;
5664                         em->block_start = EXTENT_MAP_DELALLOC;
5665                         em->block_len = found;
5666                 }
5667         } else if (hole_em) {
5668                 return hole_em;
5669         }
5670 out:
5671
5672         free_extent_map(hole_em);
5673         if (err) {
5674                 free_extent_map(em);
5675                 return ERR_PTR(err);
5676         }
5677         return em;
5678 }
5679
5680 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5681                                                   u64 start, u64 len)
5682 {
5683         struct btrfs_root *root = BTRFS_I(inode)->root;
5684         struct btrfs_trans_handle *trans;
5685         struct extent_map *em;
5686         struct btrfs_key ins;
5687         u64 alloc_hint;
5688         int ret;
5689
5690         trans = btrfs_join_transaction(root);
5691         if (IS_ERR(trans))
5692                 return ERR_CAST(trans);
5693
5694         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5695
5696         alloc_hint = get_extent_allocation_hint(inode, start, len);
5697         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5698                                    alloc_hint, &ins, 1);
5699         if (ret) {
5700                 em = ERR_PTR(ret);
5701                 goto out;
5702         }
5703
5704         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5705                               ins.offset, ins.offset, 0);
5706         if (IS_ERR(em))
5707                 goto out;
5708
5709         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5710                                            ins.offset, ins.offset, 0);
5711         if (ret) {
5712                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5713                 em = ERR_PTR(ret);
5714         }
5715 out:
5716         btrfs_end_transaction(trans, root);
5717         return em;
5718 }
5719
5720 /*
5721  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5722  * block must be cow'd
5723  */
5724 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5725                                       struct inode *inode, u64 offset, u64 len)
5726 {
5727         struct btrfs_path *path;
5728         int ret;
5729         struct extent_buffer *leaf;
5730         struct btrfs_root *root = BTRFS_I(inode)->root;
5731         struct btrfs_file_extent_item *fi;
5732         struct btrfs_key key;
5733         u64 disk_bytenr;
5734         u64 backref_offset;
5735         u64 extent_end;
5736         u64 num_bytes;
5737         int slot;
5738         int found_type;
5739
5740         path = btrfs_alloc_path();
5741         if (!path)
5742                 return -ENOMEM;
5743
5744         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5745                                        offset, 0);
5746         if (ret < 0)
5747                 goto out;
5748
5749         slot = path->slots[0];
5750         if (ret == 1) {
5751                 if (slot == 0) {
5752                         /* can't find the item, must cow */
5753                         ret = 0;
5754                         goto out;
5755                 }
5756                 slot--;
5757         }
5758         ret = 0;
5759         leaf = path->nodes[0];
5760         btrfs_item_key_to_cpu(leaf, &key, slot);
5761         if (key.objectid != btrfs_ino(inode) ||
5762             key.type != BTRFS_EXTENT_DATA_KEY) {
5763                 /* not our file or wrong item type, must cow */
5764                 goto out;
5765         }
5766
5767         if (key.offset > offset) {
5768                 /* Wrong offset, must cow */
5769                 goto out;
5770         }
5771
5772         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5773         found_type = btrfs_file_extent_type(leaf, fi);
5774         if (found_type != BTRFS_FILE_EXTENT_REG &&
5775             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5776                 /* not a regular extent, must cow */
5777                 goto out;
5778         }
5779         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5780         backref_offset = btrfs_file_extent_offset(leaf, fi);
5781
5782         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5783         if (extent_end < offset + len) {
5784                 /* extent doesn't include our full range, must cow */
5785                 goto out;
5786         }
5787
5788         if (btrfs_extent_readonly(root, disk_bytenr))
5789                 goto out;
5790
5791         /*
5792          * look for other files referencing this extent, if we
5793          * find any we must cow
5794          */
5795         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5796                                   key.offset - backref_offset, disk_bytenr))
5797                 goto out;
5798
5799         /*
5800          * adjust disk_bytenr and num_bytes to cover just the bytes
5801          * in this extent we are about to write.  If there
5802          * are any csums in that range we have to cow in order
5803          * to keep the csums correct
5804          */
5805         disk_bytenr += backref_offset;
5806         disk_bytenr += offset - key.offset;
5807         num_bytes = min(offset + len, extent_end) - offset;
5808         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5809                                 goto out;
5810         /*
5811          * all of the above have passed, it is safe to overwrite this extent
5812          * without cow
5813          */
5814         ret = 1;
5815 out:
5816         btrfs_free_path(path);
5817         return ret;
5818 }
5819
5820 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5821                               struct extent_state **cached_state, int writing)
5822 {
5823         struct btrfs_ordered_extent *ordered;
5824         int ret = 0;
5825
5826         while (1) {
5827                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5828                                  0, cached_state);
5829                 /*
5830                  * We're concerned with the entire range that we're going to be
5831                  * doing DIO to, so we need to make sure theres no ordered
5832                  * extents in this range.
5833                  */
5834                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5835                                                      lockend - lockstart + 1);
5836
5837                 /*
5838                  * We need to make sure there are no buffered pages in this
5839                  * range either, we could have raced between the invalidate in
5840                  * generic_file_direct_write and locking the extent.  The
5841                  * invalidate needs to happen so that reads after a write do not
5842                  * get stale data.
5843                  */
5844                 if (!ordered && (!writing ||
5845                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5846                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5847                                     *cached_state)))
5848                         break;
5849
5850                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5851                                      cached_state, GFP_NOFS);
5852
5853                 if (ordered) {
5854                         btrfs_start_ordered_extent(inode, ordered, 1);
5855                         btrfs_put_ordered_extent(ordered);
5856                 } else {
5857                         /* Screw you mmap */
5858                         ret = filemap_write_and_wait_range(inode->i_mapping,
5859                                                            lockstart,
5860                                                            lockend);
5861                         if (ret)
5862                                 break;
5863
5864                         /*
5865                          * If we found a page that couldn't be invalidated just
5866                          * fall back to buffered.
5867                          */
5868                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5869                                         lockstart >> PAGE_CACHE_SHIFT,
5870                                         lockend >> PAGE_CACHE_SHIFT);
5871                         if (ret)
5872                                 break;
5873                 }
5874
5875                 cond_resched();
5876         }
5877
5878         return ret;
5879 }
5880
5881 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5882                                            u64 len, u64 orig_start,
5883                                            u64 block_start, u64 block_len,
5884                                            u64 orig_block_len, int type)
5885 {
5886         struct extent_map_tree *em_tree;
5887         struct extent_map *em;
5888         struct btrfs_root *root = BTRFS_I(inode)->root;
5889         int ret;
5890
5891         em_tree = &BTRFS_I(inode)->extent_tree;
5892         em = alloc_extent_map();
5893         if (!em)
5894                 return ERR_PTR(-ENOMEM);
5895
5896         em->start = start;
5897         em->orig_start = orig_start;
5898         em->len = len;
5899         em->block_len = block_len;
5900         em->block_start = block_start;
5901         em->bdev = root->fs_info->fs_devices->latest_bdev;
5902         em->orig_block_len = orig_block_len;
5903         em->generation = -1;
5904         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5905         if (type == BTRFS_ORDERED_PREALLOC)
5906                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5907
5908         do {
5909                 btrfs_drop_extent_cache(inode, em->start,
5910                                 em->start + em->len - 1, 0);
5911                 write_lock(&em_tree->lock);
5912                 ret = add_extent_mapping(em_tree, em);
5913                 if (!ret)
5914                         list_move(&em->list,
5915                                   &em_tree->modified_extents);
5916                 write_unlock(&em_tree->lock);
5917         } while (ret == -EEXIST);
5918
5919         if (ret) {
5920                 free_extent_map(em);
5921                 return ERR_PTR(ret);
5922         }
5923
5924         return em;
5925 }
5926
5927
5928 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5929                                    struct buffer_head *bh_result, int create)
5930 {
5931         struct extent_map *em;
5932         struct btrfs_root *root = BTRFS_I(inode)->root;
5933         struct extent_state *cached_state = NULL;
5934         u64 start = iblock << inode->i_blkbits;
5935         u64 lockstart, lockend;
5936         u64 len = bh_result->b_size;
5937         struct btrfs_trans_handle *trans;
5938         int unlock_bits = EXTENT_LOCKED;
5939         int ret;
5940
5941         if (create) {
5942                 ret = btrfs_delalloc_reserve_space(inode, len);
5943                 if (ret)
5944                         return ret;
5945                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5946         } else {
5947                 len = min_t(u64, len, root->sectorsize);
5948         }
5949
5950         lockstart = start;
5951         lockend = start + len - 1;
5952
5953         /*
5954          * If this errors out it's because we couldn't invalidate pagecache for
5955          * this range and we need to fallback to buffered.
5956          */
5957         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5958                 return -ENOTBLK;
5959
5960         if (create) {
5961                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5962                                      lockend, EXTENT_DELALLOC, NULL,
5963                                      &cached_state, GFP_NOFS);
5964                 if (ret)
5965                         goto unlock_err;
5966         }
5967
5968         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5969         if (IS_ERR(em)) {
5970                 ret = PTR_ERR(em);
5971                 goto unlock_err;
5972         }
5973
5974         /*
5975          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5976          * io.  INLINE is special, and we could probably kludge it in here, but
5977          * it's still buffered so for safety lets just fall back to the generic
5978          * buffered path.
5979          *
5980          * For COMPRESSED we _have_ to read the entire extent in so we can
5981          * decompress it, so there will be buffering required no matter what we
5982          * do, so go ahead and fallback to buffered.
5983          *
5984          * We return -ENOTBLK because thats what makes DIO go ahead and go back
5985          * to buffered IO.  Don't blame me, this is the price we pay for using
5986          * the generic code.
5987          */
5988         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
5989             em->block_start == EXTENT_MAP_INLINE) {
5990                 free_extent_map(em);
5991                 ret = -ENOTBLK;
5992                 goto unlock_err;
5993         }
5994
5995         /* Just a good old fashioned hole, return */
5996         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
5997                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
5998                 free_extent_map(em);
5999                 ret = 0;
6000                 goto unlock_err;
6001         }
6002
6003         /*
6004          * We don't allocate a new extent in the following cases
6005          *
6006          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6007          * existing extent.
6008          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6009          * just use the extent.
6010          *
6011          */
6012         if (!create) {
6013                 len = min(len, em->len - (start - em->start));
6014                 lockstart = start + len;
6015                 goto unlock;
6016         }
6017
6018         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6019             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6020              em->block_start != EXTENT_MAP_HOLE)) {
6021                 int type;
6022                 int ret;
6023                 u64 block_start;
6024
6025                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6026                         type = BTRFS_ORDERED_PREALLOC;
6027                 else
6028                         type = BTRFS_ORDERED_NOCOW;
6029                 len = min(len, em->len - (start - em->start));
6030                 block_start = em->block_start + (start - em->start);
6031
6032                 /*
6033                  * we're not going to log anything, but we do need
6034                  * to make sure the current transaction stays open
6035                  * while we look for nocow cross refs
6036                  */
6037                 trans = btrfs_join_transaction(root);
6038                 if (IS_ERR(trans))
6039                         goto must_cow;
6040
6041                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6042                         u64 orig_start = em->orig_start;
6043                         u64 orig_block_len = em->orig_block_len;
6044
6045                         if (type == BTRFS_ORDERED_PREALLOC) {
6046                                 free_extent_map(em);
6047                                 em = create_pinned_em(inode, start, len,
6048                                                        orig_start,
6049                                                        block_start, len,
6050                                                        orig_block_len, type);
6051                                 if (IS_ERR(em)) {
6052                                         btrfs_end_transaction(trans, root);
6053                                         goto unlock_err;
6054                                 }
6055                         }
6056
6057                         ret = btrfs_add_ordered_extent_dio(inode, start,
6058                                            block_start, len, len, type);
6059                         btrfs_end_transaction(trans, root);
6060                         if (ret) {
6061                                 free_extent_map(em);
6062                                 goto unlock_err;
6063                         }
6064                         goto unlock;
6065                 }
6066                 btrfs_end_transaction(trans, root);
6067         }
6068 must_cow:
6069         /*
6070          * this will cow the extent, reset the len in case we changed
6071          * it above
6072          */
6073         len = bh_result->b_size;
6074         free_extent_map(em);
6075         em = btrfs_new_extent_direct(inode, start, len);
6076         if (IS_ERR(em)) {
6077                 ret = PTR_ERR(em);
6078                 goto unlock_err;
6079         }
6080         len = min(len, em->len - (start - em->start));
6081 unlock:
6082         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6083                 inode->i_blkbits;
6084         bh_result->b_size = len;
6085         bh_result->b_bdev = em->bdev;
6086         set_buffer_mapped(bh_result);
6087         if (create) {
6088                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6089                         set_buffer_new(bh_result);
6090
6091                 /*
6092                  * Need to update the i_size under the extent lock so buffered
6093                  * readers will get the updated i_size when we unlock.
6094                  */
6095                 if (start + len > i_size_read(inode))
6096                         i_size_write(inode, start + len);
6097         }
6098
6099         /*
6100          * In the case of write we need to clear and unlock the entire range,
6101          * in the case of read we need to unlock only the end area that we
6102          * aren't using if there is any left over space.
6103          */
6104         if (lockstart < lockend) {
6105                 if (create && len < lockend - lockstart) {
6106                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6107                                          lockstart + len - 1,
6108                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6109                                          &cached_state, GFP_NOFS);
6110                         /*
6111                          * Beside unlock, we also need to cleanup reserved space
6112                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6113                          */
6114                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6115                                          lockstart + len, lockend,
6116                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6117                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6118                 } else {
6119                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6120                                          lockend, unlock_bits, 1, 0,
6121                                          &cached_state, GFP_NOFS);
6122                 }
6123         } else {
6124                 free_extent_state(cached_state);
6125         }
6126
6127         free_extent_map(em);
6128
6129         return 0;
6130
6131 unlock_err:
6132         if (create)
6133                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6134
6135         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6136                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6137         return ret;
6138 }
6139
6140 struct btrfs_dio_private {
6141         struct inode *inode;
6142         u64 logical_offset;
6143         u64 disk_bytenr;
6144         u64 bytes;
6145         void *private;
6146
6147         /* number of bios pending for this dio */
6148         atomic_t pending_bios;
6149
6150         /* IO errors */
6151         int errors;
6152
6153         struct bio *orig_bio;
6154 };
6155
6156 static void btrfs_endio_direct_read(struct bio *bio, int err)
6157 {
6158         struct btrfs_dio_private *dip = bio->bi_private;
6159         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6160         struct bio_vec *bvec = bio->bi_io_vec;
6161         struct inode *inode = dip->inode;
6162         struct btrfs_root *root = BTRFS_I(inode)->root;
6163         u64 start;
6164
6165         start = dip->logical_offset;
6166         do {
6167                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6168                         struct page *page = bvec->bv_page;
6169                         char *kaddr;
6170                         u32 csum = ~(u32)0;
6171                         u64 private = ~(u32)0;
6172                         unsigned long flags;
6173
6174                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6175                                               start, &private))
6176                                 goto failed;
6177                         local_irq_save(flags);
6178                         kaddr = kmap_atomic(page);
6179                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6180                                                csum, bvec->bv_len);
6181                         btrfs_csum_final(csum, (char *)&csum);
6182                         kunmap_atomic(kaddr);
6183                         local_irq_restore(flags);
6184
6185                         flush_dcache_page(bvec->bv_page);
6186                         if (csum != private) {
6187 failed:
6188                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6189                                       " %llu csum %u private %u\n",
6190                                       (unsigned long long)btrfs_ino(inode),
6191                                       (unsigned long long)start,
6192                                       csum, (unsigned)private);
6193                                 err = -EIO;
6194                         }
6195                 }
6196
6197                 start += bvec->bv_len;
6198                 bvec++;
6199         } while (bvec <= bvec_end);
6200
6201         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6202                       dip->logical_offset + dip->bytes - 1);
6203         bio->bi_private = dip->private;
6204
6205         kfree(dip);
6206
6207         /* If we had a csum failure make sure to clear the uptodate flag */
6208         if (err)
6209                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6210         dio_end_io(bio, err);
6211 }
6212
6213 static void btrfs_endio_direct_write(struct bio *bio, int err)
6214 {
6215         struct btrfs_dio_private *dip = bio->bi_private;
6216         struct inode *inode = dip->inode;
6217         struct btrfs_root *root = BTRFS_I(inode)->root;
6218         struct btrfs_ordered_extent *ordered = NULL;
6219         u64 ordered_offset = dip->logical_offset;
6220         u64 ordered_bytes = dip->bytes;
6221         int ret;
6222
6223         if (err)
6224                 goto out_done;
6225 again:
6226         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6227                                                    &ordered_offset,
6228                                                    ordered_bytes, !err);
6229         if (!ret)
6230                 goto out_test;
6231
6232         ordered->work.func = finish_ordered_fn;
6233         ordered->work.flags = 0;
6234         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6235                            &ordered->work);
6236 out_test:
6237         /*
6238          * our bio might span multiple ordered extents.  If we haven't
6239          * completed the accounting for the whole dio, go back and try again
6240          */
6241         if (ordered_offset < dip->logical_offset + dip->bytes) {
6242                 ordered_bytes = dip->logical_offset + dip->bytes -
6243                         ordered_offset;
6244                 ordered = NULL;
6245                 goto again;
6246         }
6247 out_done:
6248         bio->bi_private = dip->private;
6249
6250         kfree(dip);
6251
6252         /* If we had an error make sure to clear the uptodate flag */
6253         if (err)
6254                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6255         dio_end_io(bio, err);
6256 }
6257
6258 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6259                                     struct bio *bio, int mirror_num,
6260                                     unsigned long bio_flags, u64 offset)
6261 {
6262         int ret;
6263         struct btrfs_root *root = BTRFS_I(inode)->root;
6264         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6265         BUG_ON(ret); /* -ENOMEM */
6266         return 0;
6267 }
6268
6269 static void btrfs_end_dio_bio(struct bio *bio, int err)
6270 {
6271         struct btrfs_dio_private *dip = bio->bi_private;
6272
6273         if (err) {
6274                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6275                       "sector %#Lx len %u err no %d\n",
6276                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6277                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6278                 dip->errors = 1;
6279
6280                 /*
6281                  * before atomic variable goto zero, we must make sure
6282                  * dip->errors is perceived to be set.
6283                  */
6284                 smp_mb__before_atomic_dec();
6285         }
6286
6287         /* if there are more bios still pending for this dio, just exit */
6288         if (!atomic_dec_and_test(&dip->pending_bios))
6289                 goto out;
6290
6291         if (dip->errors)
6292                 bio_io_error(dip->orig_bio);
6293         else {
6294                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6295                 bio_endio(dip->orig_bio, 0);
6296         }
6297 out:
6298         bio_put(bio);
6299 }
6300
6301 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6302                                        u64 first_sector, gfp_t gfp_flags)
6303 {
6304         int nr_vecs = bio_get_nr_vecs(bdev);
6305         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6306 }
6307
6308 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6309                                          int rw, u64 file_offset, int skip_sum,
6310                                          int async_submit)
6311 {
6312         int write = rw & REQ_WRITE;
6313         struct btrfs_root *root = BTRFS_I(inode)->root;
6314         int ret;
6315
6316         if (async_submit)
6317                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6318
6319         bio_get(bio);
6320
6321         if (!write) {
6322                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6323                 if (ret)
6324                         goto err;
6325         }
6326
6327         if (skip_sum)
6328                 goto map;
6329
6330         if (write && async_submit) {
6331                 ret = btrfs_wq_submit_bio(root->fs_info,
6332                                    inode, rw, bio, 0, 0,
6333                                    file_offset,
6334                                    __btrfs_submit_bio_start_direct_io,
6335                                    __btrfs_submit_bio_done);
6336                 goto err;
6337         } else if (write) {
6338                 /*
6339                  * If we aren't doing async submit, calculate the csum of the
6340                  * bio now.
6341                  */
6342                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6343                 if (ret)
6344                         goto err;
6345         } else if (!skip_sum) {
6346                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6347                 if (ret)
6348                         goto err;
6349         }
6350
6351 map:
6352         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6353 err:
6354         bio_put(bio);
6355         return ret;
6356 }
6357
6358 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6359                                     int skip_sum)
6360 {
6361         struct inode *inode = dip->inode;
6362         struct btrfs_root *root = BTRFS_I(inode)->root;
6363         struct bio *bio;
6364         struct bio *orig_bio = dip->orig_bio;
6365         struct bio_vec *bvec = orig_bio->bi_io_vec;
6366         u64 start_sector = orig_bio->bi_sector;
6367         u64 file_offset = dip->logical_offset;
6368         u64 submit_len = 0;
6369         u64 map_length;
6370         int nr_pages = 0;
6371         int ret = 0;
6372         int async_submit = 0;
6373
6374         map_length = orig_bio->bi_size;
6375         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6376                               &map_length, NULL, 0);
6377         if (ret) {
6378                 bio_put(orig_bio);
6379                 return -EIO;
6380         }
6381
6382         if (map_length >= orig_bio->bi_size) {
6383                 bio = orig_bio;
6384                 goto submit;
6385         }
6386
6387         async_submit = 1;
6388         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6389         if (!bio)
6390                 return -ENOMEM;
6391         bio->bi_private = dip;
6392         bio->bi_end_io = btrfs_end_dio_bio;
6393         atomic_inc(&dip->pending_bios);
6394
6395         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6396                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6397                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6398                                  bvec->bv_offset) < bvec->bv_len)) {
6399                         /*
6400                          * inc the count before we submit the bio so
6401                          * we know the end IO handler won't happen before
6402                          * we inc the count. Otherwise, the dip might get freed
6403                          * before we're done setting it up
6404                          */
6405                         atomic_inc(&dip->pending_bios);
6406                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6407                                                      file_offset, skip_sum,
6408                                                      async_submit);
6409                         if (ret) {
6410                                 bio_put(bio);
6411                                 atomic_dec(&dip->pending_bios);
6412                                 goto out_err;
6413                         }
6414
6415                         start_sector += submit_len >> 9;
6416                         file_offset += submit_len;
6417
6418                         submit_len = 0;
6419                         nr_pages = 0;
6420
6421                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6422                                                   start_sector, GFP_NOFS);
6423                         if (!bio)
6424                                 goto out_err;
6425                         bio->bi_private = dip;
6426                         bio->bi_end_io = btrfs_end_dio_bio;
6427
6428                         map_length = orig_bio->bi_size;
6429                         ret = btrfs_map_block(root->fs_info, READ,
6430                                               start_sector << 9,
6431                                               &map_length, NULL, 0);
6432                         if (ret) {
6433                                 bio_put(bio);
6434                                 goto out_err;
6435                         }
6436                 } else {
6437                         submit_len += bvec->bv_len;
6438                         nr_pages ++;
6439                         bvec++;
6440                 }
6441         }
6442
6443 submit:
6444         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6445                                      async_submit);
6446         if (!ret)
6447                 return 0;
6448
6449         bio_put(bio);
6450 out_err:
6451         dip->errors = 1;
6452         /*
6453          * before atomic variable goto zero, we must
6454          * make sure dip->errors is perceived to be set.
6455          */
6456         smp_mb__before_atomic_dec();
6457         if (atomic_dec_and_test(&dip->pending_bios))
6458                 bio_io_error(dip->orig_bio);
6459
6460         /* bio_end_io() will handle error, so we needn't return it */
6461         return 0;
6462 }
6463
6464 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6465                                 loff_t file_offset)
6466 {
6467         struct btrfs_root *root = BTRFS_I(inode)->root;
6468         struct btrfs_dio_private *dip;
6469         struct bio_vec *bvec = bio->bi_io_vec;
6470         int skip_sum;
6471         int write = rw & REQ_WRITE;
6472         int ret = 0;
6473
6474         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6475
6476         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6477         if (!dip) {
6478                 ret = -ENOMEM;
6479                 goto free_ordered;
6480         }
6481
6482         dip->private = bio->bi_private;
6483         dip->inode = inode;
6484         dip->logical_offset = file_offset;
6485
6486         dip->bytes = 0;
6487         do {
6488                 dip->bytes += bvec->bv_len;
6489                 bvec++;
6490         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6491
6492         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6493         bio->bi_private = dip;
6494         dip->errors = 0;
6495         dip->orig_bio = bio;
6496         atomic_set(&dip->pending_bios, 0);
6497
6498         if (write)
6499                 bio->bi_end_io = btrfs_endio_direct_write;
6500         else
6501                 bio->bi_end_io = btrfs_endio_direct_read;
6502
6503         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6504         if (!ret)
6505                 return;
6506 free_ordered:
6507         /*
6508          * If this is a write, we need to clean up the reserved space and kill
6509          * the ordered extent.
6510          */
6511         if (write) {
6512                 struct btrfs_ordered_extent *ordered;
6513                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6514                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6515                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6516                         btrfs_free_reserved_extent(root, ordered->start,
6517                                                    ordered->disk_len);
6518                 btrfs_put_ordered_extent(ordered);
6519                 btrfs_put_ordered_extent(ordered);
6520         }
6521         bio_endio(bio, ret);
6522 }
6523
6524 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6525                         const struct iovec *iov, loff_t offset,
6526                         unsigned long nr_segs)
6527 {
6528         int seg;
6529         int i;
6530         size_t size;
6531         unsigned long addr;
6532         unsigned blocksize_mask = root->sectorsize - 1;
6533         ssize_t retval = -EINVAL;
6534         loff_t end = offset;
6535
6536         if (offset & blocksize_mask)
6537                 goto out;
6538
6539         /* Check the memory alignment.  Blocks cannot straddle pages */
6540         for (seg = 0; seg < nr_segs; seg++) {
6541                 addr = (unsigned long)iov[seg].iov_base;
6542                 size = iov[seg].iov_len;
6543                 end += size;
6544                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6545                         goto out;
6546
6547                 /* If this is a write we don't need to check anymore */
6548                 if (rw & WRITE)
6549                         continue;
6550
6551                 /*
6552                  * Check to make sure we don't have duplicate iov_base's in this
6553                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6554                  * when reading back.
6555                  */
6556                 for (i = seg + 1; i < nr_segs; i++) {
6557                         if (iov[seg].iov_base == iov[i].iov_base)
6558                                 goto out;
6559                 }
6560         }
6561         retval = 0;
6562 out:
6563         return retval;
6564 }
6565
6566 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6567                         const struct iovec *iov, loff_t offset,
6568                         unsigned long nr_segs)
6569 {
6570         struct file *file = iocb->ki_filp;
6571         struct inode *inode = file->f_mapping->host;
6572
6573         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6574                             offset, nr_segs))
6575                 return 0;
6576
6577         return __blockdev_direct_IO(rw, iocb, inode,
6578                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6579                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6580                    btrfs_submit_direct, 0);
6581 }
6582
6583 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6584
6585 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6586                 __u64 start, __u64 len)
6587 {
6588         int     ret;
6589
6590         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6591         if (ret)
6592                 return ret;
6593
6594         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6595 }
6596
6597 int btrfs_readpage(struct file *file, struct page *page)
6598 {
6599         struct extent_io_tree *tree;
6600         tree = &BTRFS_I(page->mapping->host)->io_tree;
6601         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6602 }
6603
6604 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6605 {
6606         struct extent_io_tree *tree;
6607
6608
6609         if (current->flags & PF_MEMALLOC) {
6610                 redirty_page_for_writepage(wbc, page);
6611                 unlock_page(page);
6612                 return 0;
6613         }
6614         tree = &BTRFS_I(page->mapping->host)->io_tree;
6615         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6616 }
6617
6618 int btrfs_writepages(struct address_space *mapping,
6619                      struct writeback_control *wbc)
6620 {
6621         struct extent_io_tree *tree;
6622
6623         tree = &BTRFS_I(mapping->host)->io_tree;
6624         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6625 }
6626
6627 static int
6628 btrfs_readpages(struct file *file, struct address_space *mapping,
6629                 struct list_head *pages, unsigned nr_pages)
6630 {
6631         struct extent_io_tree *tree;
6632         tree = &BTRFS_I(mapping->host)->io_tree;
6633         return extent_readpages(tree, mapping, pages, nr_pages,
6634                                 btrfs_get_extent);
6635 }
6636 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6637 {
6638         struct extent_io_tree *tree;
6639         struct extent_map_tree *map;
6640         int ret;
6641
6642         tree = &BTRFS_I(page->mapping->host)->io_tree;
6643         map = &BTRFS_I(page->mapping->host)->extent_tree;
6644         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6645         if (ret == 1) {
6646                 ClearPagePrivate(page);
6647                 set_page_private(page, 0);
6648                 page_cache_release(page);
6649         }
6650         return ret;
6651 }
6652
6653 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6654 {
6655         if (PageWriteback(page) || PageDirty(page))
6656                 return 0;
6657         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6658 }
6659
6660 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6661 {
6662         struct inode *inode = page->mapping->host;
6663         struct extent_io_tree *tree;
6664         struct btrfs_ordered_extent *ordered;
6665         struct extent_state *cached_state = NULL;
6666         u64 page_start = page_offset(page);
6667         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6668
6669         /*
6670          * we have the page locked, so new writeback can't start,
6671          * and the dirty bit won't be cleared while we are here.
6672          *
6673          * Wait for IO on this page so that we can safely clear
6674          * the PagePrivate2 bit and do ordered accounting
6675          */
6676         wait_on_page_writeback(page);
6677
6678         tree = &BTRFS_I(inode)->io_tree;
6679         if (offset) {
6680                 btrfs_releasepage(page, GFP_NOFS);
6681                 return;
6682         }
6683         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6684         ordered = btrfs_lookup_ordered_extent(inode,
6685                                            page_offset(page));
6686         if (ordered) {
6687                 /*
6688                  * IO on this page will never be started, so we need
6689                  * to account for any ordered extents now
6690                  */
6691                 clear_extent_bit(tree, page_start, page_end,
6692                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6693                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6694                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6695                 /*
6696                  * whoever cleared the private bit is responsible
6697                  * for the finish_ordered_io
6698                  */
6699                 if (TestClearPagePrivate2(page) &&
6700                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6701                                                    PAGE_CACHE_SIZE, 1)) {
6702                         btrfs_finish_ordered_io(ordered);
6703                 }
6704                 btrfs_put_ordered_extent(ordered);
6705                 cached_state = NULL;
6706                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6707         }
6708         clear_extent_bit(tree, page_start, page_end,
6709                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6710                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6711                  &cached_state, GFP_NOFS);
6712         __btrfs_releasepage(page, GFP_NOFS);
6713
6714         ClearPageChecked(page);
6715         if (PagePrivate(page)) {
6716                 ClearPagePrivate(page);
6717                 set_page_private(page, 0);
6718                 page_cache_release(page);
6719         }
6720 }
6721
6722 /*
6723  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6724  * called from a page fault handler when a page is first dirtied. Hence we must
6725  * be careful to check for EOF conditions here. We set the page up correctly
6726  * for a written page which means we get ENOSPC checking when writing into
6727  * holes and correct delalloc and unwritten extent mapping on filesystems that
6728  * support these features.
6729  *
6730  * We are not allowed to take the i_mutex here so we have to play games to
6731  * protect against truncate races as the page could now be beyond EOF.  Because
6732  * vmtruncate() writes the inode size before removing pages, once we have the
6733  * page lock we can determine safely if the page is beyond EOF. If it is not
6734  * beyond EOF, then the page is guaranteed safe against truncation until we
6735  * unlock the page.
6736  */
6737 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6738 {
6739         struct page *page = vmf->page;
6740         struct inode *inode = file_inode(vma->vm_file);
6741         struct btrfs_root *root = BTRFS_I(inode)->root;
6742         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6743         struct btrfs_ordered_extent *ordered;
6744         struct extent_state *cached_state = NULL;
6745         char *kaddr;
6746         unsigned long zero_start;
6747         loff_t size;
6748         int ret;
6749         int reserved = 0;
6750         u64 page_start;
6751         u64 page_end;
6752
6753         sb_start_pagefault(inode->i_sb);
6754         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6755         if (!ret) {
6756                 ret = file_update_time(vma->vm_file);
6757                 reserved = 1;
6758         }
6759         if (ret) {
6760                 if (ret == -ENOMEM)
6761                         ret = VM_FAULT_OOM;
6762                 else /* -ENOSPC, -EIO, etc */
6763                         ret = VM_FAULT_SIGBUS;
6764                 if (reserved)
6765                         goto out;
6766                 goto out_noreserve;
6767         }
6768
6769         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6770 again:
6771         lock_page(page);
6772         size = i_size_read(inode);
6773         page_start = page_offset(page);
6774         page_end = page_start + PAGE_CACHE_SIZE - 1;
6775
6776         if ((page->mapping != inode->i_mapping) ||
6777             (page_start >= size)) {
6778                 /* page got truncated out from underneath us */
6779                 goto out_unlock;
6780         }
6781         wait_on_page_writeback(page);
6782
6783         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6784         set_page_extent_mapped(page);
6785
6786         /*
6787          * we can't set the delalloc bits if there are pending ordered
6788          * extents.  Drop our locks and wait for them to finish
6789          */
6790         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6791         if (ordered) {
6792                 unlock_extent_cached(io_tree, page_start, page_end,
6793                                      &cached_state, GFP_NOFS);
6794                 unlock_page(page);
6795                 btrfs_start_ordered_extent(inode, ordered, 1);
6796                 btrfs_put_ordered_extent(ordered);
6797                 goto again;
6798         }
6799
6800         /*
6801          * XXX - page_mkwrite gets called every time the page is dirtied, even
6802          * if it was already dirty, so for space accounting reasons we need to
6803          * clear any delalloc bits for the range we are fixing to save.  There
6804          * is probably a better way to do this, but for now keep consistent with
6805          * prepare_pages in the normal write path.
6806          */
6807         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6808                           EXTENT_DIRTY | EXTENT_DELALLOC |
6809                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6810                           0, 0, &cached_state, GFP_NOFS);
6811
6812         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6813                                         &cached_state);
6814         if (ret) {
6815                 unlock_extent_cached(io_tree, page_start, page_end,
6816                                      &cached_state, GFP_NOFS);
6817                 ret = VM_FAULT_SIGBUS;
6818                 goto out_unlock;
6819         }
6820         ret = 0;
6821
6822         /* page is wholly or partially inside EOF */
6823         if (page_start + PAGE_CACHE_SIZE > size)
6824                 zero_start = size & ~PAGE_CACHE_MASK;
6825         else
6826                 zero_start = PAGE_CACHE_SIZE;
6827
6828         if (zero_start != PAGE_CACHE_SIZE) {
6829                 kaddr = kmap(page);
6830                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6831                 flush_dcache_page(page);
6832                 kunmap(page);
6833         }
6834         ClearPageChecked(page);
6835         set_page_dirty(page);
6836         SetPageUptodate(page);
6837
6838         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6839         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6840         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6841
6842         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6843
6844 out_unlock:
6845         if (!ret) {
6846                 sb_end_pagefault(inode->i_sb);
6847                 return VM_FAULT_LOCKED;
6848         }
6849         unlock_page(page);
6850 out:
6851         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6852 out_noreserve:
6853         sb_end_pagefault(inode->i_sb);
6854         return ret;
6855 }
6856
6857 static int btrfs_truncate(struct inode *inode)
6858 {
6859         struct btrfs_root *root = BTRFS_I(inode)->root;
6860         struct btrfs_block_rsv *rsv;
6861         int ret;
6862         int err = 0;
6863         struct btrfs_trans_handle *trans;
6864         u64 mask = root->sectorsize - 1;
6865         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6866
6867         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6868         if (ret)
6869                 return ret;
6870
6871         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6872         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6873
6874         /*
6875          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6876          * 3 things going on here
6877          *
6878          * 1) We need to reserve space for our orphan item and the space to
6879          * delete our orphan item.  Lord knows we don't want to have a dangling
6880          * orphan item because we didn't reserve space to remove it.
6881          *
6882          * 2) We need to reserve space to update our inode.
6883          *
6884          * 3) We need to have something to cache all the space that is going to
6885          * be free'd up by the truncate operation, but also have some slack
6886          * space reserved in case it uses space during the truncate (thank you
6887          * very much snapshotting).
6888          *
6889          * And we need these to all be seperate.  The fact is we can use alot of
6890          * space doing the truncate, and we have no earthly idea how much space
6891          * we will use, so we need the truncate reservation to be seperate so it
6892          * doesn't end up using space reserved for updating the inode or
6893          * removing the orphan item.  We also need to be able to stop the
6894          * transaction and start a new one, which means we need to be able to
6895          * update the inode several times, and we have no idea of knowing how
6896          * many times that will be, so we can't just reserve 1 item for the
6897          * entirety of the opration, so that has to be done seperately as well.
6898          * Then there is the orphan item, which does indeed need to be held on
6899          * to for the whole operation, and we need nobody to touch this reserved
6900          * space except the orphan code.
6901          *
6902          * So that leaves us with
6903          *
6904          * 1) root->orphan_block_rsv - for the orphan deletion.
6905          * 2) rsv - for the truncate reservation, which we will steal from the
6906          * transaction reservation.
6907          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6908          * updating the inode.
6909          */
6910         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6911         if (!rsv)
6912                 return -ENOMEM;
6913         rsv->size = min_size;
6914         rsv->failfast = 1;
6915
6916         /*
6917          * 1 for the truncate slack space
6918          * 1 for the orphan item we're going to add
6919          * 1 for the orphan item deletion
6920          * 1 for updating the inode.
6921          */
6922         trans = btrfs_start_transaction(root, 4);
6923         if (IS_ERR(trans)) {
6924                 err = PTR_ERR(trans);
6925                 goto out;
6926         }
6927
6928         /* Migrate the slack space for the truncate to our reserve */
6929         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6930                                       min_size);
6931         BUG_ON(ret);
6932
6933         ret = btrfs_orphan_add(trans, inode);
6934         if (ret) {
6935                 btrfs_end_transaction(trans, root);
6936                 goto out;
6937         }
6938
6939         /*
6940          * setattr is responsible for setting the ordered_data_close flag,
6941          * but that is only tested during the last file release.  That
6942          * could happen well after the next commit, leaving a great big
6943          * window where new writes may get lost if someone chooses to write
6944          * to this file after truncating to zero
6945          *
6946          * The inode doesn't have any dirty data here, and so if we commit
6947          * this is a noop.  If someone immediately starts writing to the inode
6948          * it is very likely we'll catch some of their writes in this
6949          * transaction, and the commit will find this file on the ordered
6950          * data list with good things to send down.
6951          *
6952          * This is a best effort solution, there is still a window where
6953          * using truncate to replace the contents of the file will
6954          * end up with a zero length file after a crash.
6955          */
6956         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6957                                            &BTRFS_I(inode)->runtime_flags))
6958                 btrfs_add_ordered_operation(trans, root, inode);
6959
6960         /*
6961          * So if we truncate and then write and fsync we normally would just
6962          * write the extents that changed, which is a problem if we need to
6963          * first truncate that entire inode.  So set this flag so we write out
6964          * all of the extents in the inode to the sync log so we're completely
6965          * safe.
6966          */
6967         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6968         trans->block_rsv = rsv;
6969
6970         while (1) {
6971                 ret = btrfs_truncate_inode_items(trans, root, inode,
6972                                                  inode->i_size,
6973                                                  BTRFS_EXTENT_DATA_KEY);
6974                 if (ret != -ENOSPC) {
6975                         err = ret;
6976                         break;
6977                 }
6978
6979                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6980                 ret = btrfs_update_inode(trans, root, inode);
6981                 if (ret) {
6982                         err = ret;
6983                         break;
6984                 }
6985
6986                 btrfs_end_transaction(trans, root);
6987                 btrfs_btree_balance_dirty(root);
6988
6989                 trans = btrfs_start_transaction(root, 2);
6990                 if (IS_ERR(trans)) {
6991                         ret = err = PTR_ERR(trans);
6992                         trans = NULL;
6993                         break;
6994                 }
6995
6996                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
6997                                               rsv, min_size);
6998                 BUG_ON(ret);    /* shouldn't happen */
6999                 trans->block_rsv = rsv;
7000         }
7001
7002         if (ret == 0 && inode->i_nlink > 0) {
7003                 trans->block_rsv = root->orphan_block_rsv;
7004                 ret = btrfs_orphan_del(trans, inode);
7005                 if (ret)
7006                         err = ret;
7007         } else if (ret && inode->i_nlink > 0) {
7008                 /*
7009                  * Failed to do the truncate, remove us from the in memory
7010                  * orphan list.
7011                  */
7012                 ret = btrfs_orphan_del(NULL, inode);
7013         }
7014
7015         if (trans) {
7016                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7017                 ret = btrfs_update_inode(trans, root, inode);
7018                 if (ret && !err)
7019                         err = ret;
7020
7021                 ret = btrfs_end_transaction(trans, root);
7022                 btrfs_btree_balance_dirty(root);
7023         }
7024
7025 out:
7026         btrfs_free_block_rsv(root, rsv);
7027
7028         if (ret && !err)
7029                 err = ret;
7030
7031         return err;
7032 }
7033
7034 /*
7035  * create a new subvolume directory/inode (helper for the ioctl).
7036  */
7037 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7038                              struct btrfs_root *new_root, u64 new_dirid)
7039 {
7040         struct inode *inode;
7041         int err;
7042         u64 index = 0;
7043
7044         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7045                                 new_dirid, new_dirid,
7046                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7047                                 &index);
7048         if (IS_ERR(inode))
7049                 return PTR_ERR(inode);
7050         inode->i_op = &btrfs_dir_inode_operations;
7051         inode->i_fop = &btrfs_dir_file_operations;
7052
7053         set_nlink(inode, 1);
7054         btrfs_i_size_write(inode, 0);
7055
7056         err = btrfs_update_inode(trans, new_root, inode);
7057
7058         iput(inode);
7059         return err;
7060 }
7061
7062 struct inode *btrfs_alloc_inode(struct super_block *sb)
7063 {
7064         struct btrfs_inode *ei;
7065         struct inode *inode;
7066
7067         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7068         if (!ei)
7069                 return NULL;
7070
7071         ei->root = NULL;
7072         ei->generation = 0;
7073         ei->last_trans = 0;
7074         ei->last_sub_trans = 0;
7075         ei->logged_trans = 0;
7076         ei->delalloc_bytes = 0;
7077         ei->disk_i_size = 0;
7078         ei->flags = 0;
7079         ei->csum_bytes = 0;
7080         ei->index_cnt = (u64)-1;
7081         ei->last_unlink_trans = 0;
7082         ei->last_log_commit = 0;
7083
7084         spin_lock_init(&ei->lock);
7085         ei->outstanding_extents = 0;
7086         ei->reserved_extents = 0;
7087
7088         ei->runtime_flags = 0;
7089         ei->force_compress = BTRFS_COMPRESS_NONE;
7090
7091         ei->delayed_node = NULL;
7092
7093         inode = &ei->vfs_inode;
7094         extent_map_tree_init(&ei->extent_tree);
7095         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7096         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7097         ei->io_tree.track_uptodate = 1;
7098         ei->io_failure_tree.track_uptodate = 1;
7099         atomic_set(&ei->sync_writers, 0);
7100         mutex_init(&ei->log_mutex);
7101         mutex_init(&ei->delalloc_mutex);
7102         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7103         INIT_LIST_HEAD(&ei->delalloc_inodes);
7104         INIT_LIST_HEAD(&ei->ordered_operations);
7105         RB_CLEAR_NODE(&ei->rb_node);
7106
7107         return inode;
7108 }
7109
7110 static void btrfs_i_callback(struct rcu_head *head)
7111 {
7112         struct inode *inode = container_of(head, struct inode, i_rcu);
7113         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7114 }
7115
7116 void btrfs_destroy_inode(struct inode *inode)
7117 {
7118         struct btrfs_ordered_extent *ordered;
7119         struct btrfs_root *root = BTRFS_I(inode)->root;
7120
7121         WARN_ON(!hlist_empty(&inode->i_dentry));
7122         WARN_ON(inode->i_data.nrpages);
7123         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7124         WARN_ON(BTRFS_I(inode)->reserved_extents);
7125         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7126         WARN_ON(BTRFS_I(inode)->csum_bytes);
7127
7128         /*
7129          * This can happen where we create an inode, but somebody else also
7130          * created the same inode and we need to destroy the one we already
7131          * created.
7132          */
7133         if (!root)
7134                 goto free;
7135
7136         /*
7137          * Make sure we're properly removed from the ordered operation
7138          * lists.
7139          */
7140         smp_mb();
7141         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7142                 spin_lock(&root->fs_info->ordered_extent_lock);
7143                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7144                 spin_unlock(&root->fs_info->ordered_extent_lock);
7145         }
7146
7147         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7148                      &BTRFS_I(inode)->runtime_flags)) {
7149                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7150                        (unsigned long long)btrfs_ino(inode));
7151                 atomic_dec(&root->orphan_inodes);
7152         }
7153
7154         while (1) {
7155                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7156                 if (!ordered)
7157                         break;
7158                 else {
7159                         printk(KERN_ERR "btrfs found ordered "
7160                                "extent %llu %llu on inode cleanup\n",
7161                                (unsigned long long)ordered->file_offset,
7162                                (unsigned long long)ordered->len);
7163                         btrfs_remove_ordered_extent(inode, ordered);
7164                         btrfs_put_ordered_extent(ordered);
7165                         btrfs_put_ordered_extent(ordered);
7166                 }
7167         }
7168         inode_tree_del(inode);
7169         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7170 free:
7171         btrfs_remove_delayed_node(inode);
7172         call_rcu(&inode->i_rcu, btrfs_i_callback);
7173 }
7174
7175 int btrfs_drop_inode(struct inode *inode)
7176 {
7177         struct btrfs_root *root = BTRFS_I(inode)->root;
7178
7179         if (btrfs_root_refs(&root->root_item) == 0 &&
7180             !btrfs_is_free_space_inode(inode))
7181                 return 1;
7182         else
7183                 return generic_drop_inode(inode);
7184 }
7185
7186 static void init_once(void *foo)
7187 {
7188         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7189
7190         inode_init_once(&ei->vfs_inode);
7191 }
7192
7193 void btrfs_destroy_cachep(void)
7194 {
7195         /*
7196          * Make sure all delayed rcu free inodes are flushed before we
7197          * destroy cache.
7198          */
7199         rcu_barrier();
7200         if (btrfs_inode_cachep)
7201                 kmem_cache_destroy(btrfs_inode_cachep);
7202         if (btrfs_trans_handle_cachep)
7203                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7204         if (btrfs_transaction_cachep)
7205                 kmem_cache_destroy(btrfs_transaction_cachep);
7206         if (btrfs_path_cachep)
7207                 kmem_cache_destroy(btrfs_path_cachep);
7208         if (btrfs_free_space_cachep)
7209                 kmem_cache_destroy(btrfs_free_space_cachep);
7210         if (btrfs_delalloc_work_cachep)
7211                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7212 }
7213
7214 int btrfs_init_cachep(void)
7215 {
7216         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7217                         sizeof(struct btrfs_inode), 0,
7218                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7219         if (!btrfs_inode_cachep)
7220                 goto fail;
7221
7222         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7223                         sizeof(struct btrfs_trans_handle), 0,
7224                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7225         if (!btrfs_trans_handle_cachep)
7226                 goto fail;
7227
7228         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7229                         sizeof(struct btrfs_transaction), 0,
7230                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7231         if (!btrfs_transaction_cachep)
7232                 goto fail;
7233
7234         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7235                         sizeof(struct btrfs_path), 0,
7236                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7237         if (!btrfs_path_cachep)
7238                 goto fail;
7239
7240         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7241                         sizeof(struct btrfs_free_space), 0,
7242                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7243         if (!btrfs_free_space_cachep)
7244                 goto fail;
7245
7246         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7247                         sizeof(struct btrfs_delalloc_work), 0,
7248                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7249                         NULL);
7250         if (!btrfs_delalloc_work_cachep)
7251                 goto fail;
7252
7253         return 0;
7254 fail:
7255         btrfs_destroy_cachep();
7256         return -ENOMEM;
7257 }
7258
7259 static int btrfs_getattr(struct vfsmount *mnt,
7260                          struct dentry *dentry, struct kstat *stat)
7261 {
7262         struct inode *inode = dentry->d_inode;
7263         u32 blocksize = inode->i_sb->s_blocksize;
7264
7265         generic_fillattr(inode, stat);
7266         stat->dev = BTRFS_I(inode)->root->anon_dev;
7267         stat->blksize = PAGE_CACHE_SIZE;
7268         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7269                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7270         return 0;
7271 }
7272
7273 /*
7274  * If a file is moved, it will inherit the cow and compression flags of the new
7275  * directory.
7276  */
7277 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7278 {
7279         struct btrfs_inode *b_dir = BTRFS_I(dir);
7280         struct btrfs_inode *b_inode = BTRFS_I(inode);
7281
7282         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7283                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7284         else
7285                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7286
7287         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7288                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7289                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7290         } else {
7291                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7292                                     BTRFS_INODE_NOCOMPRESS);
7293         }
7294 }
7295
7296 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7297                            struct inode *new_dir, struct dentry *new_dentry)
7298 {
7299         struct btrfs_trans_handle *trans;
7300         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7301         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7302         struct inode *new_inode = new_dentry->d_inode;
7303         struct inode *old_inode = old_dentry->d_inode;
7304         struct timespec ctime = CURRENT_TIME;
7305         u64 index = 0;
7306         u64 root_objectid;
7307         int ret;
7308         u64 old_ino = btrfs_ino(old_inode);
7309
7310         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7311                 return -EPERM;
7312
7313         /* we only allow rename subvolume link between subvolumes */
7314         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7315                 return -EXDEV;
7316
7317         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7318             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7319                 return -ENOTEMPTY;
7320
7321         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7322             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7323                 return -ENOTEMPTY;
7324
7325
7326         /* check for collisions, even if the  name isn't there */
7327         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7328                              new_dentry->d_name.name,
7329                              new_dentry->d_name.len);
7330
7331         if (ret) {
7332                 if (ret == -EEXIST) {
7333                         /* we shouldn't get
7334                          * eexist without a new_inode */
7335                         if (!new_inode) {
7336                                 WARN_ON(1);
7337                                 return ret;
7338                         }
7339                 } else {
7340                         /* maybe -EOVERFLOW */
7341                         return ret;
7342                 }
7343         }
7344         ret = 0;
7345
7346         /*
7347          * we're using rename to replace one file with another.
7348          * and the replacement file is large.  Start IO on it now so
7349          * we don't add too much work to the end of the transaction
7350          */
7351         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7352             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7353                 filemap_flush(old_inode->i_mapping);
7354
7355         /* close the racy window with snapshot create/destroy ioctl */
7356         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7357                 down_read(&root->fs_info->subvol_sem);
7358         /*
7359          * We want to reserve the absolute worst case amount of items.  So if
7360          * both inodes are subvols and we need to unlink them then that would
7361          * require 4 item modifications, but if they are both normal inodes it
7362          * would require 5 item modifications, so we'll assume their normal
7363          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7364          * should cover the worst case number of items we'll modify.
7365          */
7366         trans = btrfs_start_transaction(root, 20);
7367         if (IS_ERR(trans)) {
7368                 ret = PTR_ERR(trans);
7369                 goto out_notrans;
7370         }
7371
7372         if (dest != root)
7373                 btrfs_record_root_in_trans(trans, dest);
7374
7375         ret = btrfs_set_inode_index(new_dir, &index);
7376         if (ret)
7377                 goto out_fail;
7378
7379         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7380                 /* force full log commit if subvolume involved. */
7381                 root->fs_info->last_trans_log_full_commit = trans->transid;
7382         } else {
7383                 ret = btrfs_insert_inode_ref(trans, dest,
7384                                              new_dentry->d_name.name,
7385                                              new_dentry->d_name.len,
7386                                              old_ino,
7387                                              btrfs_ino(new_dir), index);
7388                 if (ret)
7389                         goto out_fail;
7390                 /*
7391                  * this is an ugly little race, but the rename is required
7392                  * to make sure that if we crash, the inode is either at the
7393                  * old name or the new one.  pinning the log transaction lets
7394                  * us make sure we don't allow a log commit to come in after
7395                  * we unlink the name but before we add the new name back in.
7396                  */
7397                 btrfs_pin_log_trans(root);
7398         }
7399         /*
7400          * make sure the inode gets flushed if it is replacing
7401          * something.
7402          */
7403         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7404                 btrfs_add_ordered_operation(trans, root, old_inode);
7405
7406         inode_inc_iversion(old_dir);
7407         inode_inc_iversion(new_dir);
7408         inode_inc_iversion(old_inode);
7409         old_dir->i_ctime = old_dir->i_mtime = ctime;
7410         new_dir->i_ctime = new_dir->i_mtime = ctime;
7411         old_inode->i_ctime = ctime;
7412
7413         if (old_dentry->d_parent != new_dentry->d_parent)
7414                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7415
7416         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7417                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7418                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7419                                         old_dentry->d_name.name,
7420                                         old_dentry->d_name.len);
7421         } else {
7422                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7423                                         old_dentry->d_inode,
7424                                         old_dentry->d_name.name,
7425                                         old_dentry->d_name.len);
7426                 if (!ret)
7427                         ret = btrfs_update_inode(trans, root, old_inode);
7428         }
7429         if (ret) {
7430                 btrfs_abort_transaction(trans, root, ret);
7431                 goto out_fail;
7432         }
7433
7434         if (new_inode) {
7435                 inode_inc_iversion(new_inode);
7436                 new_inode->i_ctime = CURRENT_TIME;
7437                 if (unlikely(btrfs_ino(new_inode) ==
7438                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7439                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7440                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7441                                                 root_objectid,
7442                                                 new_dentry->d_name.name,
7443                                                 new_dentry->d_name.len);
7444                         BUG_ON(new_inode->i_nlink == 0);
7445                 } else {
7446                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7447                                                  new_dentry->d_inode,
7448                                                  new_dentry->d_name.name,
7449                                                  new_dentry->d_name.len);
7450                 }
7451                 if (!ret && new_inode->i_nlink == 0) {
7452                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7453                         BUG_ON(ret);
7454                 }
7455                 if (ret) {
7456                         btrfs_abort_transaction(trans, root, ret);
7457                         goto out_fail;
7458                 }
7459         }
7460
7461         fixup_inode_flags(new_dir, old_inode);
7462
7463         ret = btrfs_add_link(trans, new_dir, old_inode,
7464                              new_dentry->d_name.name,
7465                              new_dentry->d_name.len, 0, index);
7466         if (ret) {
7467                 btrfs_abort_transaction(trans, root, ret);
7468                 goto out_fail;
7469         }
7470
7471         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7472                 struct dentry *parent = new_dentry->d_parent;
7473                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7474                 btrfs_end_log_trans(root);
7475         }
7476 out_fail:
7477         btrfs_end_transaction(trans, root);
7478 out_notrans:
7479         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7480                 up_read(&root->fs_info->subvol_sem);
7481
7482         return ret;
7483 }
7484
7485 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7486 {
7487         struct btrfs_delalloc_work *delalloc_work;
7488
7489         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7490                                      work);
7491         if (delalloc_work->wait)
7492                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7493         else
7494                 filemap_flush(delalloc_work->inode->i_mapping);
7495
7496         if (delalloc_work->delay_iput)
7497                 btrfs_add_delayed_iput(delalloc_work->inode);
7498         else
7499                 iput(delalloc_work->inode);
7500         complete(&delalloc_work->completion);
7501 }
7502
7503 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7504                                                     int wait, int delay_iput)
7505 {
7506         struct btrfs_delalloc_work *work;
7507
7508         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7509         if (!work)
7510                 return NULL;
7511
7512         init_completion(&work->completion);
7513         INIT_LIST_HEAD(&work->list);
7514         work->inode = inode;
7515         work->wait = wait;
7516         work->delay_iput = delay_iput;
7517         work->work.func = btrfs_run_delalloc_work;
7518
7519         return work;
7520 }
7521
7522 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7523 {
7524         wait_for_completion(&work->completion);
7525         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7526 }
7527
7528 /*
7529  * some fairly slow code that needs optimization. This walks the list
7530  * of all the inodes with pending delalloc and forces them to disk.
7531  */
7532 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7533 {
7534         struct list_head *head = &root->fs_info->delalloc_inodes;
7535         struct btrfs_inode *binode;
7536         struct inode *inode;
7537         struct btrfs_delalloc_work *work, *next;
7538         struct list_head works;
7539         int ret = 0;
7540
7541         if (root->fs_info->sb->s_flags & MS_RDONLY)
7542                 return -EROFS;
7543
7544         INIT_LIST_HEAD(&works);
7545
7546         spin_lock(&root->fs_info->delalloc_lock);
7547         while (!list_empty(head)) {
7548                 binode = list_entry(head->next, struct btrfs_inode,
7549                                     delalloc_inodes);
7550                 inode = igrab(&binode->vfs_inode);
7551                 if (!inode)
7552                         list_del_init(&binode->delalloc_inodes);
7553                 spin_unlock(&root->fs_info->delalloc_lock);
7554                 if (inode) {
7555                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7556                         if (!work) {
7557                                 ret = -ENOMEM;
7558                                 goto out;
7559                         }
7560                         list_add_tail(&work->list, &works);
7561                         btrfs_queue_worker(&root->fs_info->flush_workers,
7562                                            &work->work);
7563                 }
7564                 cond_resched();
7565                 spin_lock(&root->fs_info->delalloc_lock);
7566         }
7567         spin_unlock(&root->fs_info->delalloc_lock);
7568
7569         /* the filemap_flush will queue IO into the worker threads, but
7570          * we have to make sure the IO is actually started and that
7571          * ordered extents get created before we return
7572          */
7573         atomic_inc(&root->fs_info->async_submit_draining);
7574         while (atomic_read(&root->fs_info->nr_async_submits) ||
7575               atomic_read(&root->fs_info->async_delalloc_pages)) {
7576                 wait_event(root->fs_info->async_submit_wait,
7577                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7578                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7579         }
7580         atomic_dec(&root->fs_info->async_submit_draining);
7581 out:
7582         list_for_each_entry_safe(work, next, &works, list) {
7583                 list_del_init(&work->list);
7584                 btrfs_wait_and_free_delalloc_work(work);
7585         }
7586         return ret;
7587 }
7588
7589 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7590                          const char *symname)
7591 {
7592         struct btrfs_trans_handle *trans;
7593         struct btrfs_root *root = BTRFS_I(dir)->root;
7594         struct btrfs_path *path;
7595         struct btrfs_key key;
7596         struct inode *inode = NULL;
7597         int err;
7598         int drop_inode = 0;
7599         u64 objectid;
7600         u64 index = 0 ;
7601         int name_len;
7602         int datasize;
7603         unsigned long ptr;
7604         struct btrfs_file_extent_item *ei;
7605         struct extent_buffer *leaf;
7606
7607         name_len = strlen(symname) + 1;
7608         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7609                 return -ENAMETOOLONG;
7610
7611         /*
7612          * 2 items for inode item and ref
7613          * 2 items for dir items
7614          * 1 item for xattr if selinux is on
7615          */
7616         trans = btrfs_start_transaction(root, 5);
7617         if (IS_ERR(trans))
7618                 return PTR_ERR(trans);
7619
7620         err = btrfs_find_free_ino(root, &objectid);
7621         if (err)
7622                 goto out_unlock;
7623
7624         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7625                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7626                                 S_IFLNK|S_IRWXUGO, &index);
7627         if (IS_ERR(inode)) {
7628                 err = PTR_ERR(inode);
7629                 goto out_unlock;
7630         }
7631
7632         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7633         if (err) {
7634                 drop_inode = 1;
7635                 goto out_unlock;
7636         }
7637
7638         /*
7639         * If the active LSM wants to access the inode during
7640         * d_instantiate it needs these. Smack checks to see
7641         * if the filesystem supports xattrs by looking at the
7642         * ops vector.
7643         */
7644         inode->i_fop = &btrfs_file_operations;
7645         inode->i_op = &btrfs_file_inode_operations;
7646
7647         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7648         if (err)
7649                 drop_inode = 1;
7650         else {
7651                 inode->i_mapping->a_ops = &btrfs_aops;
7652                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7653                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7654         }
7655         if (drop_inode)
7656                 goto out_unlock;
7657
7658         path = btrfs_alloc_path();
7659         if (!path) {
7660                 err = -ENOMEM;
7661                 drop_inode = 1;
7662                 goto out_unlock;
7663         }
7664         key.objectid = btrfs_ino(inode);
7665         key.offset = 0;
7666         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7667         datasize = btrfs_file_extent_calc_inline_size(name_len);
7668         err = btrfs_insert_empty_item(trans, root, path, &key,
7669                                       datasize);
7670         if (err) {
7671                 drop_inode = 1;
7672                 btrfs_free_path(path);
7673                 goto out_unlock;
7674         }
7675         leaf = path->nodes[0];
7676         ei = btrfs_item_ptr(leaf, path->slots[0],
7677                             struct btrfs_file_extent_item);
7678         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7679         btrfs_set_file_extent_type(leaf, ei,
7680                                    BTRFS_FILE_EXTENT_INLINE);
7681         btrfs_set_file_extent_encryption(leaf, ei, 0);
7682         btrfs_set_file_extent_compression(leaf, ei, 0);
7683         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7684         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7685
7686         ptr = btrfs_file_extent_inline_start(ei);
7687         write_extent_buffer(leaf, symname, ptr, name_len);
7688         btrfs_mark_buffer_dirty(leaf);
7689         btrfs_free_path(path);
7690
7691         inode->i_op = &btrfs_symlink_inode_operations;
7692         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7693         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7694         inode_set_bytes(inode, name_len);
7695         btrfs_i_size_write(inode, name_len - 1);
7696         err = btrfs_update_inode(trans, root, inode);
7697         if (err)
7698                 drop_inode = 1;
7699
7700 out_unlock:
7701         if (!err)
7702                 d_instantiate(dentry, inode);
7703         btrfs_end_transaction(trans, root);
7704         if (drop_inode) {
7705                 inode_dec_link_count(inode);
7706                 iput(inode);
7707         }
7708         btrfs_btree_balance_dirty(root);
7709         return err;
7710 }
7711
7712 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7713                                        u64 start, u64 num_bytes, u64 min_size,
7714                                        loff_t actual_len, u64 *alloc_hint,
7715                                        struct btrfs_trans_handle *trans)
7716 {
7717         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7718         struct extent_map *em;
7719         struct btrfs_root *root = BTRFS_I(inode)->root;
7720         struct btrfs_key ins;
7721         u64 cur_offset = start;
7722         u64 i_size;
7723         int ret = 0;
7724         bool own_trans = true;
7725
7726         if (trans)
7727                 own_trans = false;
7728         while (num_bytes > 0) {
7729                 if (own_trans) {
7730                         trans = btrfs_start_transaction(root, 3);
7731                         if (IS_ERR(trans)) {
7732                                 ret = PTR_ERR(trans);
7733                                 break;
7734                         }
7735                 }
7736
7737                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7738                                            0, *alloc_hint, &ins, 1);
7739                 if (ret) {
7740                         if (own_trans)
7741                                 btrfs_end_transaction(trans, root);
7742                         break;
7743                 }
7744
7745                 ret = insert_reserved_file_extent(trans, inode,
7746                                                   cur_offset, ins.objectid,
7747                                                   ins.offset, ins.offset,
7748                                                   ins.offset, 0, 0, 0,
7749                                                   BTRFS_FILE_EXTENT_PREALLOC);
7750                 if (ret) {
7751                         btrfs_abort_transaction(trans, root, ret);
7752                         if (own_trans)
7753                                 btrfs_end_transaction(trans, root);
7754                         break;
7755                 }
7756                 btrfs_drop_extent_cache(inode, cur_offset,
7757                                         cur_offset + ins.offset -1, 0);
7758
7759                 em = alloc_extent_map();
7760                 if (!em) {
7761                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7762                                 &BTRFS_I(inode)->runtime_flags);
7763                         goto next;
7764                 }
7765
7766                 em->start = cur_offset;
7767                 em->orig_start = cur_offset;
7768                 em->len = ins.offset;
7769                 em->block_start = ins.objectid;
7770                 em->block_len = ins.offset;
7771                 em->orig_block_len = ins.offset;
7772                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7773                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7774                 em->generation = trans->transid;
7775
7776                 while (1) {
7777                         write_lock(&em_tree->lock);
7778                         ret = add_extent_mapping(em_tree, em);
7779                         if (!ret)
7780                                 list_move(&em->list,
7781                                           &em_tree->modified_extents);
7782                         write_unlock(&em_tree->lock);
7783                         if (ret != -EEXIST)
7784                                 break;
7785                         btrfs_drop_extent_cache(inode, cur_offset,
7786                                                 cur_offset + ins.offset - 1,
7787                                                 0);
7788                 }
7789                 free_extent_map(em);
7790 next:
7791                 num_bytes -= ins.offset;
7792                 cur_offset += ins.offset;
7793                 *alloc_hint = ins.objectid + ins.offset;
7794
7795                 inode_inc_iversion(inode);
7796                 inode->i_ctime = CURRENT_TIME;
7797                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7798                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7799                     (actual_len > inode->i_size) &&
7800                     (cur_offset > inode->i_size)) {
7801                         if (cur_offset > actual_len)
7802                                 i_size = actual_len;
7803                         else
7804                                 i_size = cur_offset;
7805                         i_size_write(inode, i_size);
7806                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7807                 }
7808
7809                 ret = btrfs_update_inode(trans, root, inode);
7810
7811                 if (ret) {
7812                         btrfs_abort_transaction(trans, root, ret);
7813                         if (own_trans)
7814                                 btrfs_end_transaction(trans, root);
7815                         break;
7816                 }
7817
7818                 if (own_trans)
7819                         btrfs_end_transaction(trans, root);
7820         }
7821         return ret;
7822 }
7823
7824 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7825                               u64 start, u64 num_bytes, u64 min_size,
7826                               loff_t actual_len, u64 *alloc_hint)
7827 {
7828         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7829                                            min_size, actual_len, alloc_hint,
7830                                            NULL);
7831 }
7832
7833 int btrfs_prealloc_file_range_trans(struct inode *inode,
7834                                     struct btrfs_trans_handle *trans, int mode,
7835                                     u64 start, u64 num_bytes, u64 min_size,
7836                                     loff_t actual_len, u64 *alloc_hint)
7837 {
7838         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7839                                            min_size, actual_len, alloc_hint, trans);
7840 }
7841
7842 static int btrfs_set_page_dirty(struct page *page)
7843 {
7844         return __set_page_dirty_nobuffers(page);
7845 }
7846
7847 static int btrfs_permission(struct inode *inode, int mask)
7848 {
7849         struct btrfs_root *root = BTRFS_I(inode)->root;
7850         umode_t mode = inode->i_mode;
7851
7852         if (mask & MAY_WRITE &&
7853             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7854                 if (btrfs_root_readonly(root))
7855                         return -EROFS;
7856                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7857                         return -EACCES;
7858         }
7859         return generic_permission(inode, mask);
7860 }
7861
7862 static const struct inode_operations btrfs_dir_inode_operations = {
7863         .getattr        = btrfs_getattr,
7864         .lookup         = btrfs_lookup,
7865         .create         = btrfs_create,
7866         .unlink         = btrfs_unlink,
7867         .link           = btrfs_link,
7868         .mkdir          = btrfs_mkdir,
7869         .rmdir          = btrfs_rmdir,
7870         .rename         = btrfs_rename,
7871         .symlink        = btrfs_symlink,
7872         .setattr        = btrfs_setattr,
7873         .mknod          = btrfs_mknod,
7874         .setxattr       = btrfs_setxattr,
7875         .getxattr       = btrfs_getxattr,
7876         .listxattr      = btrfs_listxattr,
7877         .removexattr    = btrfs_removexattr,
7878         .permission     = btrfs_permission,
7879         .get_acl        = btrfs_get_acl,
7880 };
7881 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7882         .lookup         = btrfs_lookup,
7883         .permission     = btrfs_permission,
7884         .get_acl        = btrfs_get_acl,
7885 };
7886
7887 static const struct file_operations btrfs_dir_file_operations = {
7888         .llseek         = generic_file_llseek,
7889         .read           = generic_read_dir,
7890         .readdir        = btrfs_real_readdir,
7891         .unlocked_ioctl = btrfs_ioctl,
7892 #ifdef CONFIG_COMPAT
7893         .compat_ioctl   = btrfs_ioctl,
7894 #endif
7895         .release        = btrfs_release_file,
7896         .fsync          = btrfs_sync_file,
7897 };
7898
7899 static struct extent_io_ops btrfs_extent_io_ops = {
7900         .fill_delalloc = run_delalloc_range,
7901         .submit_bio_hook = btrfs_submit_bio_hook,
7902         .merge_bio_hook = btrfs_merge_bio_hook,
7903         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7904         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7905         .writepage_start_hook = btrfs_writepage_start_hook,
7906         .set_bit_hook = btrfs_set_bit_hook,
7907         .clear_bit_hook = btrfs_clear_bit_hook,
7908         .merge_extent_hook = btrfs_merge_extent_hook,
7909         .split_extent_hook = btrfs_split_extent_hook,
7910 };
7911
7912 /*
7913  * btrfs doesn't support the bmap operation because swapfiles
7914  * use bmap to make a mapping of extents in the file.  They assume
7915  * these extents won't change over the life of the file and they
7916  * use the bmap result to do IO directly to the drive.
7917  *
7918  * the btrfs bmap call would return logical addresses that aren't
7919  * suitable for IO and they also will change frequently as COW
7920  * operations happen.  So, swapfile + btrfs == corruption.
7921  *
7922  * For now we're avoiding this by dropping bmap.
7923  */
7924 static const struct address_space_operations btrfs_aops = {
7925         .readpage       = btrfs_readpage,
7926         .writepage      = btrfs_writepage,
7927         .writepages     = btrfs_writepages,
7928         .readpages      = btrfs_readpages,
7929         .direct_IO      = btrfs_direct_IO,
7930         .invalidatepage = btrfs_invalidatepage,
7931         .releasepage    = btrfs_releasepage,
7932         .set_page_dirty = btrfs_set_page_dirty,
7933         .error_remove_page = generic_error_remove_page,
7934 };
7935
7936 static const struct address_space_operations btrfs_symlink_aops = {
7937         .readpage       = btrfs_readpage,
7938         .writepage      = btrfs_writepage,
7939         .invalidatepage = btrfs_invalidatepage,
7940         .releasepage    = btrfs_releasepage,
7941 };
7942
7943 static const struct inode_operations btrfs_file_inode_operations = {
7944         .getattr        = btrfs_getattr,
7945         .setattr        = btrfs_setattr,
7946         .setxattr       = btrfs_setxattr,
7947         .getxattr       = btrfs_getxattr,
7948         .listxattr      = btrfs_listxattr,
7949         .removexattr    = btrfs_removexattr,
7950         .permission     = btrfs_permission,
7951         .fiemap         = btrfs_fiemap,
7952         .get_acl        = btrfs_get_acl,
7953         .update_time    = btrfs_update_time,
7954 };
7955 static const struct inode_operations btrfs_special_inode_operations = {
7956         .getattr        = btrfs_getattr,
7957         .setattr        = btrfs_setattr,
7958         .permission     = btrfs_permission,
7959         .setxattr       = btrfs_setxattr,
7960         .getxattr       = btrfs_getxattr,
7961         .listxattr      = btrfs_listxattr,
7962         .removexattr    = btrfs_removexattr,
7963         .get_acl        = btrfs_get_acl,
7964         .update_time    = btrfs_update_time,
7965 };
7966 static const struct inode_operations btrfs_symlink_inode_operations = {
7967         .readlink       = generic_readlink,
7968         .follow_link    = page_follow_link_light,
7969         .put_link       = page_put_link,
7970         .getattr        = btrfs_getattr,
7971         .setattr        = btrfs_setattr,
7972         .permission     = btrfs_permission,
7973         .setxattr       = btrfs_setxattr,
7974         .getxattr       = btrfs_getxattr,
7975         .listxattr      = btrfs_listxattr,
7976         .removexattr    = btrfs_removexattr,
7977         .get_acl        = btrfs_get_acl,
7978         .update_time    = btrfs_update_time,
7979 };
7980
7981 const struct dentry_operations btrfs_dentry_operations = {
7982         .d_delete       = btrfs_dentry_delete,
7983         .d_release      = btrfs_dentry_release,
7984 };