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
2482                         /* 1 for the orphan item deletion. */
2483                         trans = btrfs_start_transaction(root, 1);
2484                         if (IS_ERR(trans)) {
2485                                 ret = PTR_ERR(trans);
2486                                 goto out;
2487                         }
2488                         ret = btrfs_orphan_add(trans, inode);
2489                         btrfs_end_transaction(trans, root);
2490                         if (ret)
2491                                 goto out;
2492
2493                         ret = btrfs_truncate(inode);
2494                 } else {
2495                         nr_unlink++;
2496                 }
2497
2498                 /* this will do delete_inode and everything for us */
2499                 iput(inode);
2500                 if (ret)
2501                         goto out;
2502         }
2503         /* release the path since we're done with it */
2504         btrfs_release_path(path);
2505
2506         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2507
2508         if (root->orphan_block_rsv)
2509                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2510                                         (u64)-1);
2511
2512         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2513                 trans = btrfs_join_transaction(root);
2514                 if (!IS_ERR(trans))
2515                         btrfs_end_transaction(trans, root);
2516         }
2517
2518         if (nr_unlink)
2519                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2520         if (nr_truncate)
2521                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2522
2523 out:
2524         if (ret)
2525                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2526         btrfs_free_path(path);
2527         return ret;
2528 }
2529
2530 /*
2531  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2532  * don't find any xattrs, we know there can't be any acls.
2533  *
2534  * slot is the slot the inode is in, objectid is the objectid of the inode
2535  */
2536 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2537                                           int slot, u64 objectid)
2538 {
2539         u32 nritems = btrfs_header_nritems(leaf);
2540         struct btrfs_key found_key;
2541         int scanned = 0;
2542
2543         slot++;
2544         while (slot < nritems) {
2545                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2546
2547                 /* we found a different objectid, there must not be acls */
2548                 if (found_key.objectid != objectid)
2549                         return 0;
2550
2551                 /* we found an xattr, assume we've got an acl */
2552                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2553                         return 1;
2554
2555                 /*
2556                  * we found a key greater than an xattr key, there can't
2557                  * be any acls later on
2558                  */
2559                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2560                         return 0;
2561
2562                 slot++;
2563                 scanned++;
2564
2565                 /*
2566                  * it goes inode, inode backrefs, xattrs, extents,
2567                  * so if there are a ton of hard links to an inode there can
2568                  * be a lot of backrefs.  Don't waste time searching too hard,
2569                  * this is just an optimization
2570                  */
2571                 if (scanned >= 8)
2572                         break;
2573         }
2574         /* we hit the end of the leaf before we found an xattr or
2575          * something larger than an xattr.  We have to assume the inode
2576          * has acls
2577          */
2578         return 1;
2579 }
2580
2581 /*
2582  * read an inode from the btree into the in-memory inode
2583  */
2584 static void btrfs_read_locked_inode(struct inode *inode)
2585 {
2586         struct btrfs_path *path;
2587         struct extent_buffer *leaf;
2588         struct btrfs_inode_item *inode_item;
2589         struct btrfs_timespec *tspec;
2590         struct btrfs_root *root = BTRFS_I(inode)->root;
2591         struct btrfs_key location;
2592         int maybe_acls;
2593         u32 rdev;
2594         int ret;
2595         bool filled = false;
2596
2597         ret = btrfs_fill_inode(inode, &rdev);
2598         if (!ret)
2599                 filled = true;
2600
2601         path = btrfs_alloc_path();
2602         if (!path)
2603                 goto make_bad;
2604
2605         path->leave_spinning = 1;
2606         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2607
2608         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2609         if (ret)
2610                 goto make_bad;
2611
2612         leaf = path->nodes[0];
2613
2614         if (filled)
2615                 goto cache_acl;
2616
2617         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2618                                     struct btrfs_inode_item);
2619         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2620         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2621         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2622         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2623         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2624
2625         tspec = btrfs_inode_atime(inode_item);
2626         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2627         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2628
2629         tspec = btrfs_inode_mtime(inode_item);
2630         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2631         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2632
2633         tspec = btrfs_inode_ctime(inode_item);
2634         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2635         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2636
2637         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2638         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2639         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2640
2641         /*
2642          * If we were modified in the current generation and evicted from memory
2643          * and then re-read we need to do a full sync since we don't have any
2644          * idea about which extents were modified before we were evicted from
2645          * cache.
2646          */
2647         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2648                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2649                         &BTRFS_I(inode)->runtime_flags);
2650
2651         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2652         inode->i_generation = BTRFS_I(inode)->generation;
2653         inode->i_rdev = 0;
2654         rdev = btrfs_inode_rdev(leaf, inode_item);
2655
2656         BTRFS_I(inode)->index_cnt = (u64)-1;
2657         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2658 cache_acl:
2659         /*
2660          * try to precache a NULL acl entry for files that don't have
2661          * any xattrs or acls
2662          */
2663         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2664                                            btrfs_ino(inode));
2665         if (!maybe_acls)
2666                 cache_no_acl(inode);
2667
2668         btrfs_free_path(path);
2669
2670         switch (inode->i_mode & S_IFMT) {
2671         case S_IFREG:
2672                 inode->i_mapping->a_ops = &btrfs_aops;
2673                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2674                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2675                 inode->i_fop = &btrfs_file_operations;
2676                 inode->i_op = &btrfs_file_inode_operations;
2677                 break;
2678         case S_IFDIR:
2679                 inode->i_fop = &btrfs_dir_file_operations;
2680                 if (root == root->fs_info->tree_root)
2681                         inode->i_op = &btrfs_dir_ro_inode_operations;
2682                 else
2683                         inode->i_op = &btrfs_dir_inode_operations;
2684                 break;
2685         case S_IFLNK:
2686                 inode->i_op = &btrfs_symlink_inode_operations;
2687                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2688                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2689                 break;
2690         default:
2691                 inode->i_op = &btrfs_special_inode_operations;
2692                 init_special_inode(inode, inode->i_mode, rdev);
2693                 break;
2694         }
2695
2696         btrfs_update_iflags(inode);
2697         return;
2698
2699 make_bad:
2700         btrfs_free_path(path);
2701         make_bad_inode(inode);
2702 }
2703
2704 /*
2705  * given a leaf and an inode, copy the inode fields into the leaf
2706  */
2707 static void fill_inode_item(struct btrfs_trans_handle *trans,
2708                             struct extent_buffer *leaf,
2709                             struct btrfs_inode_item *item,
2710                             struct inode *inode)
2711 {
2712         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2713         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2714         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2715         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2716         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2717
2718         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2719                                inode->i_atime.tv_sec);
2720         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2721                                 inode->i_atime.tv_nsec);
2722
2723         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2724                                inode->i_mtime.tv_sec);
2725         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2726                                 inode->i_mtime.tv_nsec);
2727
2728         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2729                                inode->i_ctime.tv_sec);
2730         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2731                                 inode->i_ctime.tv_nsec);
2732
2733         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2734         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2735         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2736         btrfs_set_inode_transid(leaf, item, trans->transid);
2737         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2738         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2739         btrfs_set_inode_block_group(leaf, item, 0);
2740 }
2741
2742 /*
2743  * copy everything in the in-memory inode into the btree.
2744  */
2745 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2746                                 struct btrfs_root *root, struct inode *inode)
2747 {
2748         struct btrfs_inode_item *inode_item;
2749         struct btrfs_path *path;
2750         struct extent_buffer *leaf;
2751         int ret;
2752
2753         path = btrfs_alloc_path();
2754         if (!path)
2755                 return -ENOMEM;
2756
2757         path->leave_spinning = 1;
2758         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2759                                  1);
2760         if (ret) {
2761                 if (ret > 0)
2762                         ret = -ENOENT;
2763                 goto failed;
2764         }
2765
2766         btrfs_unlock_up_safe(path, 1);
2767         leaf = path->nodes[0];
2768         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2769                                     struct btrfs_inode_item);
2770
2771         fill_inode_item(trans, leaf, inode_item, inode);
2772         btrfs_mark_buffer_dirty(leaf);
2773         btrfs_set_inode_last_trans(trans, inode);
2774         ret = 0;
2775 failed:
2776         btrfs_free_path(path);
2777         return ret;
2778 }
2779
2780 /*
2781  * copy everything in the in-memory inode into the btree.
2782  */
2783 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2784                                 struct btrfs_root *root, struct inode *inode)
2785 {
2786         int ret;
2787
2788         /*
2789          * If the inode is a free space inode, we can deadlock during commit
2790          * if we put it into the delayed code.
2791          *
2792          * The data relocation inode should also be directly updated
2793          * without delay
2794          */
2795         if (!btrfs_is_free_space_inode(inode)
2796             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2797                 btrfs_update_root_times(trans, root);
2798
2799                 ret = btrfs_delayed_update_inode(trans, root, inode);
2800                 if (!ret)
2801                         btrfs_set_inode_last_trans(trans, inode);
2802                 return ret;
2803         }
2804
2805         return btrfs_update_inode_item(trans, root, inode);
2806 }
2807
2808 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2809                                          struct btrfs_root *root,
2810                                          struct inode *inode)
2811 {
2812         int ret;
2813
2814         ret = btrfs_update_inode(trans, root, inode);
2815         if (ret == -ENOSPC)
2816                 return btrfs_update_inode_item(trans, root, inode);
2817         return ret;
2818 }
2819
2820 /*
2821  * unlink helper that gets used here in inode.c and in the tree logging
2822  * recovery code.  It remove a link in a directory with a given name, and
2823  * also drops the back refs in the inode to the directory
2824  */
2825 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2826                                 struct btrfs_root *root,
2827                                 struct inode *dir, struct inode *inode,
2828                                 const char *name, int name_len)
2829 {
2830         struct btrfs_path *path;
2831         int ret = 0;
2832         struct extent_buffer *leaf;
2833         struct btrfs_dir_item *di;
2834         struct btrfs_key key;
2835         u64 index;
2836         u64 ino = btrfs_ino(inode);
2837         u64 dir_ino = btrfs_ino(dir);
2838
2839         path = btrfs_alloc_path();
2840         if (!path) {
2841                 ret = -ENOMEM;
2842                 goto out;
2843         }
2844
2845         path->leave_spinning = 1;
2846         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2847                                     name, name_len, -1);
2848         if (IS_ERR(di)) {
2849                 ret = PTR_ERR(di);
2850                 goto err;
2851         }
2852         if (!di) {
2853                 ret = -ENOENT;
2854                 goto err;
2855         }
2856         leaf = path->nodes[0];
2857         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2858         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2859         if (ret)
2860                 goto err;
2861         btrfs_release_path(path);
2862
2863         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2864                                   dir_ino, &index);
2865         if (ret) {
2866                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2867                        "inode %llu parent %llu\n", name_len, name,
2868                        (unsigned long long)ino, (unsigned long long)dir_ino);
2869                 btrfs_abort_transaction(trans, root, ret);
2870                 goto err;
2871         }
2872
2873         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2874         if (ret) {
2875                 btrfs_abort_transaction(trans, root, ret);
2876                 goto err;
2877         }
2878
2879         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2880                                          inode, dir_ino);
2881         if (ret != 0 && ret != -ENOENT) {
2882                 btrfs_abort_transaction(trans, root, ret);
2883                 goto err;
2884         }
2885
2886         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2887                                            dir, index);
2888         if (ret == -ENOENT)
2889                 ret = 0;
2890 err:
2891         btrfs_free_path(path);
2892         if (ret)
2893                 goto out;
2894
2895         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2896         inode_inc_iversion(inode);
2897         inode_inc_iversion(dir);
2898         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2899         ret = btrfs_update_inode(trans, root, dir);
2900 out:
2901         return ret;
2902 }
2903
2904 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2905                        struct btrfs_root *root,
2906                        struct inode *dir, struct inode *inode,
2907                        const char *name, int name_len)
2908 {
2909         int ret;
2910         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2911         if (!ret) {
2912                 btrfs_drop_nlink(inode);
2913                 ret = btrfs_update_inode(trans, root, inode);
2914         }
2915         return ret;
2916 }
2917
2918
2919 /* helper to check if there is any shared block in the path */
2920 static int check_path_shared(struct btrfs_root *root,
2921                              struct btrfs_path *path)
2922 {
2923         struct extent_buffer *eb;
2924         int level;
2925         u64 refs = 1;
2926
2927         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2928                 int ret;
2929
2930                 if (!path->nodes[level])
2931                         break;
2932                 eb = path->nodes[level];
2933                 if (!btrfs_block_can_be_shared(root, eb))
2934                         continue;
2935                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2936                                                &refs, NULL);
2937                 if (refs > 1)
2938                         return 1;
2939         }
2940         return 0;
2941 }
2942
2943 /*
2944  * helper to start transaction for unlink and rmdir.
2945  *
2946  * unlink and rmdir are special in btrfs, they do not always free space.
2947  * so in enospc case, we should make sure they will free space before
2948  * allowing them to use the global metadata reservation.
2949  */
2950 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2951                                                        struct dentry *dentry)
2952 {
2953         struct btrfs_trans_handle *trans;
2954         struct btrfs_root *root = BTRFS_I(dir)->root;
2955         struct btrfs_path *path;
2956         struct btrfs_dir_item *di;
2957         struct inode *inode = dentry->d_inode;
2958         u64 index;
2959         int check_link = 1;
2960         int err = -ENOSPC;
2961         int ret;
2962         u64 ino = btrfs_ino(inode);
2963         u64 dir_ino = btrfs_ino(dir);
2964
2965         /*
2966          * 1 for the possible orphan item
2967          * 1 for the dir item
2968          * 1 for the dir index
2969          * 1 for the inode ref
2970          * 1 for the inode ref in the tree log
2971          * 2 for the dir entries in the log
2972          * 1 for the inode
2973          */
2974         trans = btrfs_start_transaction(root, 8);
2975         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2976                 return trans;
2977
2978         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2979                 return ERR_PTR(-ENOSPC);
2980
2981         /* check if there is someone else holds reference */
2982         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2983                 return ERR_PTR(-ENOSPC);
2984
2985         if (atomic_read(&inode->i_count) > 2)
2986                 return ERR_PTR(-ENOSPC);
2987
2988         if (xchg(&root->fs_info->enospc_unlink, 1))
2989                 return ERR_PTR(-ENOSPC);
2990
2991         path = btrfs_alloc_path();
2992         if (!path) {
2993                 root->fs_info->enospc_unlink = 0;
2994                 return ERR_PTR(-ENOMEM);
2995         }
2996
2997         /* 1 for the orphan item */
2998         trans = btrfs_start_transaction(root, 1);
2999         if (IS_ERR(trans)) {
3000                 btrfs_free_path(path);
3001                 root->fs_info->enospc_unlink = 0;
3002                 return trans;
3003         }
3004
3005         path->skip_locking = 1;
3006         path->search_commit_root = 1;
3007
3008         ret = btrfs_lookup_inode(trans, root, path,
3009                                 &BTRFS_I(dir)->location, 0);
3010         if (ret < 0) {
3011                 err = ret;
3012                 goto out;
3013         }
3014         if (ret == 0) {
3015                 if (check_path_shared(root, path))
3016                         goto out;
3017         } else {
3018                 check_link = 0;
3019         }
3020         btrfs_release_path(path);
3021
3022         ret = btrfs_lookup_inode(trans, root, path,
3023                                 &BTRFS_I(inode)->location, 0);
3024         if (ret < 0) {
3025                 err = ret;
3026                 goto out;
3027         }
3028         if (ret == 0) {
3029                 if (check_path_shared(root, path))
3030                         goto out;
3031         } else {
3032                 check_link = 0;
3033         }
3034         btrfs_release_path(path);
3035
3036         if (ret == 0 && S_ISREG(inode->i_mode)) {
3037                 ret = btrfs_lookup_file_extent(trans, root, path,
3038                                                ino, (u64)-1, 0);
3039                 if (ret < 0) {
3040                         err = ret;
3041                         goto out;
3042                 }
3043                 BUG_ON(ret == 0); /* Corruption */
3044                 if (check_path_shared(root, path))
3045                         goto out;
3046                 btrfs_release_path(path);
3047         }
3048
3049         if (!check_link) {
3050                 err = 0;
3051                 goto out;
3052         }
3053
3054         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3055                                 dentry->d_name.name, dentry->d_name.len, 0);
3056         if (IS_ERR(di)) {
3057                 err = PTR_ERR(di);
3058                 goto out;
3059         }
3060         if (di) {
3061                 if (check_path_shared(root, path))
3062                         goto out;
3063         } else {
3064                 err = 0;
3065                 goto out;
3066         }
3067         btrfs_release_path(path);
3068
3069         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3070                                         dentry->d_name.len, ino, dir_ino, 0,
3071                                         &index);
3072         if (ret) {
3073                 err = ret;
3074                 goto out;
3075         }
3076
3077         if (check_path_shared(root, path))
3078                 goto out;
3079
3080         btrfs_release_path(path);
3081
3082         /*
3083          * This is a commit root search, if we can lookup inode item and other
3084          * relative items in the commit root, it means the transaction of
3085          * dir/file creation has been committed, and the dir index item that we
3086          * delay to insert has also been inserted into the commit root. So
3087          * we needn't worry about the delayed insertion of the dir index item
3088          * here.
3089          */
3090         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3091                                 dentry->d_name.name, dentry->d_name.len, 0);
3092         if (IS_ERR(di)) {
3093                 err = PTR_ERR(di);
3094                 goto out;
3095         }
3096         BUG_ON(ret == -ENOENT);
3097         if (check_path_shared(root, path))
3098                 goto out;
3099
3100         err = 0;
3101 out:
3102         btrfs_free_path(path);
3103         /* Migrate the orphan reservation over */
3104         if (!err)
3105                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3106                                 &root->fs_info->global_block_rsv,
3107                                 trans->bytes_reserved);
3108
3109         if (err) {
3110                 btrfs_end_transaction(trans, root);
3111                 root->fs_info->enospc_unlink = 0;
3112                 return ERR_PTR(err);
3113         }
3114
3115         trans->block_rsv = &root->fs_info->global_block_rsv;
3116         return trans;
3117 }
3118
3119 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3120                                struct btrfs_root *root)
3121 {
3122         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3123                 btrfs_block_rsv_release(root, trans->block_rsv,
3124                                         trans->bytes_reserved);
3125                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3126                 BUG_ON(!root->fs_info->enospc_unlink);
3127                 root->fs_info->enospc_unlink = 0;
3128         }
3129         btrfs_end_transaction(trans, root);
3130 }
3131
3132 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3133 {
3134         struct btrfs_root *root = BTRFS_I(dir)->root;
3135         struct btrfs_trans_handle *trans;
3136         struct inode *inode = dentry->d_inode;
3137         int ret;
3138
3139         trans = __unlink_start_trans(dir, dentry);
3140         if (IS_ERR(trans))
3141                 return PTR_ERR(trans);
3142
3143         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3144
3145         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3146                                  dentry->d_name.name, dentry->d_name.len);
3147         if (ret)
3148                 goto out;
3149
3150         if (inode->i_nlink == 0) {
3151                 ret = btrfs_orphan_add(trans, inode);
3152                 if (ret)
3153                         goto out;
3154         }
3155
3156 out:
3157         __unlink_end_trans(trans, root);
3158         btrfs_btree_balance_dirty(root);
3159         return ret;
3160 }
3161
3162 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3163                         struct btrfs_root *root,
3164                         struct inode *dir, u64 objectid,
3165                         const char *name, int name_len)
3166 {
3167         struct btrfs_path *path;
3168         struct extent_buffer *leaf;
3169         struct btrfs_dir_item *di;
3170         struct btrfs_key key;
3171         u64 index;
3172         int ret;
3173         u64 dir_ino = btrfs_ino(dir);
3174
3175         path = btrfs_alloc_path();
3176         if (!path)
3177                 return -ENOMEM;
3178
3179         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3180                                    name, name_len, -1);
3181         if (IS_ERR_OR_NULL(di)) {
3182                 if (!di)
3183                         ret = -ENOENT;
3184                 else
3185                         ret = PTR_ERR(di);
3186                 goto out;
3187         }
3188
3189         leaf = path->nodes[0];
3190         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3191         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3192         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3193         if (ret) {
3194                 btrfs_abort_transaction(trans, root, ret);
3195                 goto out;
3196         }
3197         btrfs_release_path(path);
3198
3199         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3200                                  objectid, root->root_key.objectid,
3201                                  dir_ino, &index, name, name_len);
3202         if (ret < 0) {
3203                 if (ret != -ENOENT) {
3204                         btrfs_abort_transaction(trans, root, ret);
3205                         goto out;
3206                 }
3207                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3208                                                  name, name_len);
3209                 if (IS_ERR_OR_NULL(di)) {
3210                         if (!di)
3211                                 ret = -ENOENT;
3212                         else
3213                                 ret = PTR_ERR(di);
3214                         btrfs_abort_transaction(trans, root, ret);
3215                         goto out;
3216                 }
3217
3218                 leaf = path->nodes[0];
3219                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3220                 btrfs_release_path(path);
3221                 index = key.offset;
3222         }
3223         btrfs_release_path(path);
3224
3225         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3226         if (ret) {
3227                 btrfs_abort_transaction(trans, root, ret);
3228                 goto out;
3229         }
3230
3231         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3232         inode_inc_iversion(dir);
3233         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3234         ret = btrfs_update_inode_fallback(trans, root, dir);
3235         if (ret)
3236                 btrfs_abort_transaction(trans, root, ret);
3237 out:
3238         btrfs_free_path(path);
3239         return ret;
3240 }
3241
3242 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3243 {
3244         struct inode *inode = dentry->d_inode;
3245         int err = 0;
3246         struct btrfs_root *root = BTRFS_I(dir)->root;
3247         struct btrfs_trans_handle *trans;
3248
3249         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3250                 return -ENOTEMPTY;
3251         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3252                 return -EPERM;
3253
3254         trans = __unlink_start_trans(dir, dentry);
3255         if (IS_ERR(trans))
3256                 return PTR_ERR(trans);
3257
3258         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3259                 err = btrfs_unlink_subvol(trans, root, dir,
3260                                           BTRFS_I(inode)->location.objectid,
3261                                           dentry->d_name.name,
3262                                           dentry->d_name.len);
3263                 goto out;
3264         }
3265
3266         err = btrfs_orphan_add(trans, inode);
3267         if (err)
3268                 goto out;
3269
3270         /* now the directory is empty */
3271         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3272                                  dentry->d_name.name, dentry->d_name.len);
3273         if (!err)
3274                 btrfs_i_size_write(inode, 0);
3275 out:
3276         __unlink_end_trans(trans, root);
3277         btrfs_btree_balance_dirty(root);
3278
3279         return err;
3280 }
3281
3282 /*
3283  * this can truncate away extent items, csum items and directory items.
3284  * It starts at a high offset and removes keys until it can't find
3285  * any higher than new_size
3286  *
3287  * csum items that cross the new i_size are truncated to the new size
3288  * as well.
3289  *
3290  * min_type is the minimum key type to truncate down to.  If set to 0, this
3291  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3292  */
3293 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3294                                struct btrfs_root *root,
3295                                struct inode *inode,
3296                                u64 new_size, u32 min_type)
3297 {
3298         struct btrfs_path *path;
3299         struct extent_buffer *leaf;
3300         struct btrfs_file_extent_item *fi;
3301         struct btrfs_key key;
3302         struct btrfs_key found_key;
3303         u64 extent_start = 0;
3304         u64 extent_num_bytes = 0;
3305         u64 extent_offset = 0;
3306         u64 item_end = 0;
3307         u64 mask = root->sectorsize - 1;
3308         u32 found_type = (u8)-1;
3309         int found_extent;
3310         int del_item;
3311         int pending_del_nr = 0;
3312         int pending_del_slot = 0;
3313         int extent_type = -1;
3314         int ret;
3315         int err = 0;
3316         u64 ino = btrfs_ino(inode);
3317
3318         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3319
3320         path = btrfs_alloc_path();
3321         if (!path)
3322                 return -ENOMEM;
3323         path->reada = -1;
3324
3325         /*
3326          * We want to drop from the next block forward in case this new size is
3327          * not block aligned since we will be keeping the last block of the
3328          * extent just the way it is.
3329          */
3330         if (root->ref_cows || root == root->fs_info->tree_root)
3331                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3332
3333         /*
3334          * This function is also used to drop the items in the log tree before
3335          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3336          * it is used to drop the loged items. So we shouldn't kill the delayed
3337          * items.
3338          */
3339         if (min_type == 0 && root == BTRFS_I(inode)->root)
3340                 btrfs_kill_delayed_inode_items(inode);
3341
3342         key.objectid = ino;
3343         key.offset = (u64)-1;
3344         key.type = (u8)-1;
3345
3346 search_again:
3347         path->leave_spinning = 1;
3348         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3349         if (ret < 0) {
3350                 err = ret;
3351                 goto out;
3352         }
3353
3354         if (ret > 0) {
3355                 /* there are no items in the tree for us to truncate, we're
3356                  * done
3357                  */
3358                 if (path->slots[0] == 0)
3359                         goto out;
3360                 path->slots[0]--;
3361         }
3362
3363         while (1) {
3364                 fi = NULL;
3365                 leaf = path->nodes[0];
3366                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3367                 found_type = btrfs_key_type(&found_key);
3368
3369                 if (found_key.objectid != ino)
3370                         break;
3371
3372                 if (found_type < min_type)
3373                         break;
3374
3375                 item_end = found_key.offset;
3376                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3377                         fi = btrfs_item_ptr(leaf, path->slots[0],
3378                                             struct btrfs_file_extent_item);
3379                         extent_type = btrfs_file_extent_type(leaf, fi);
3380                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3381                                 item_end +=
3382                                     btrfs_file_extent_num_bytes(leaf, fi);
3383                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3384                                 item_end += btrfs_file_extent_inline_len(leaf,
3385                                                                          fi);
3386                         }
3387                         item_end--;
3388                 }
3389                 if (found_type > min_type) {
3390                         del_item = 1;
3391                 } else {
3392                         if (item_end < new_size)
3393                                 break;
3394                         if (found_key.offset >= new_size)
3395                                 del_item = 1;
3396                         else
3397                                 del_item = 0;
3398                 }
3399                 found_extent = 0;
3400                 /* FIXME, shrink the extent if the ref count is only 1 */
3401                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3402                         goto delete;
3403
3404                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3405                         u64 num_dec;
3406                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3407                         if (!del_item) {
3408                                 u64 orig_num_bytes =
3409                                         btrfs_file_extent_num_bytes(leaf, fi);
3410                                 extent_num_bytes = new_size -
3411                                         found_key.offset + root->sectorsize - 1;
3412                                 extent_num_bytes = extent_num_bytes &
3413                                         ~((u64)root->sectorsize - 1);
3414                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3415                                                          extent_num_bytes);
3416                                 num_dec = (orig_num_bytes -
3417                                            extent_num_bytes);
3418                                 if (root->ref_cows && extent_start != 0)
3419                                         inode_sub_bytes(inode, num_dec);
3420                                 btrfs_mark_buffer_dirty(leaf);
3421                         } else {
3422                                 extent_num_bytes =
3423                                         btrfs_file_extent_disk_num_bytes(leaf,
3424                                                                          fi);
3425                                 extent_offset = found_key.offset -
3426                                         btrfs_file_extent_offset(leaf, fi);
3427
3428                                 /* FIXME blocksize != 4096 */
3429                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3430                                 if (extent_start != 0) {
3431                                         found_extent = 1;
3432                                         if (root->ref_cows)
3433                                                 inode_sub_bytes(inode, num_dec);
3434                                 }
3435                         }
3436                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3437                         /*
3438                          * we can't truncate inline items that have had
3439                          * special encodings
3440                          */
3441                         if (!del_item &&
3442                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3443                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3444                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3445                                 u32 size = new_size - found_key.offset;
3446
3447                                 if (root->ref_cows) {
3448                                         inode_sub_bytes(inode, item_end + 1 -
3449                                                         new_size);
3450                                 }
3451                                 size =
3452                                     btrfs_file_extent_calc_inline_size(size);
3453                                 btrfs_truncate_item(trans, root, path,
3454                                                     size, 1);
3455                         } else if (root->ref_cows) {
3456                                 inode_sub_bytes(inode, item_end + 1 -
3457                                                 found_key.offset);
3458                         }
3459                 }
3460 delete:
3461                 if (del_item) {
3462                         if (!pending_del_nr) {
3463                                 /* no pending yet, add ourselves */
3464                                 pending_del_slot = path->slots[0];
3465                                 pending_del_nr = 1;
3466                         } else if (pending_del_nr &&
3467                                    path->slots[0] + 1 == pending_del_slot) {
3468                                 /* hop on the pending chunk */
3469                                 pending_del_nr++;
3470                                 pending_del_slot = path->slots[0];
3471                         } else {
3472                                 BUG();
3473                         }
3474                 } else {
3475                         break;
3476                 }
3477                 if (found_extent && (root->ref_cows ||
3478                                      root == root->fs_info->tree_root)) {
3479                         btrfs_set_path_blocking(path);
3480                         ret = btrfs_free_extent(trans, root, extent_start,
3481                                                 extent_num_bytes, 0,
3482                                                 btrfs_header_owner(leaf),
3483                                                 ino, extent_offset, 0);
3484                         BUG_ON(ret);
3485                 }
3486
3487                 if (found_type == BTRFS_INODE_ITEM_KEY)
3488                         break;
3489
3490                 if (path->slots[0] == 0 ||
3491                     path->slots[0] != pending_del_slot) {
3492                         if (pending_del_nr) {
3493                                 ret = btrfs_del_items(trans, root, path,
3494                                                 pending_del_slot,
3495                                                 pending_del_nr);
3496                                 if (ret) {
3497                                         btrfs_abort_transaction(trans,
3498                                                                 root, ret);
3499                                         goto error;
3500                                 }
3501                                 pending_del_nr = 0;
3502                         }
3503                         btrfs_release_path(path);
3504                         goto search_again;
3505                 } else {
3506                         path->slots[0]--;
3507                 }
3508         }
3509 out:
3510         if (pending_del_nr) {
3511                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3512                                       pending_del_nr);
3513                 if (ret)
3514                         btrfs_abort_transaction(trans, root, ret);
3515         }
3516 error:
3517         btrfs_free_path(path);
3518         return err;
3519 }
3520
3521 /*
3522  * btrfs_truncate_page - read, zero a chunk and write a page
3523  * @inode - inode that we're zeroing
3524  * @from - the offset to start zeroing
3525  * @len - the length to zero, 0 to zero the entire range respective to the
3526  *      offset
3527  * @front - zero up to the offset instead of from the offset on
3528  *
3529  * This will find the page for the "from" offset and cow the page and zero the
3530  * part we want to zero.  This is used with truncate and hole punching.
3531  */
3532 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3533                         int front)
3534 {
3535         struct address_space *mapping = inode->i_mapping;
3536         struct btrfs_root *root = BTRFS_I(inode)->root;
3537         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3538         struct btrfs_ordered_extent *ordered;
3539         struct extent_state *cached_state = NULL;
3540         char *kaddr;
3541         u32 blocksize = root->sectorsize;
3542         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3543         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3544         struct page *page;
3545         gfp_t mask = btrfs_alloc_write_mask(mapping);
3546         int ret = 0;
3547         u64 page_start;
3548         u64 page_end;
3549
3550         if ((offset & (blocksize - 1)) == 0 &&
3551             (!len || ((len & (blocksize - 1)) == 0)))
3552                 goto out;
3553         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3554         if (ret)
3555                 goto out;
3556
3557 again:
3558         page = find_or_create_page(mapping, index, mask);
3559         if (!page) {
3560                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3561                 ret = -ENOMEM;
3562                 goto out;
3563         }
3564
3565         page_start = page_offset(page);
3566         page_end = page_start + PAGE_CACHE_SIZE - 1;
3567
3568         if (!PageUptodate(page)) {
3569                 ret = btrfs_readpage(NULL, page);
3570                 lock_page(page);
3571                 if (page->mapping != mapping) {
3572                         unlock_page(page);
3573                         page_cache_release(page);
3574                         goto again;
3575                 }
3576                 if (!PageUptodate(page)) {
3577                         ret = -EIO;
3578                         goto out_unlock;
3579                 }
3580         }
3581         wait_on_page_writeback(page);
3582
3583         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3584         set_page_extent_mapped(page);
3585
3586         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3587         if (ordered) {
3588                 unlock_extent_cached(io_tree, page_start, page_end,
3589                                      &cached_state, GFP_NOFS);
3590                 unlock_page(page);
3591                 page_cache_release(page);
3592                 btrfs_start_ordered_extent(inode, ordered, 1);
3593                 btrfs_put_ordered_extent(ordered);
3594                 goto again;
3595         }
3596
3597         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3598                           EXTENT_DIRTY | EXTENT_DELALLOC |
3599                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3600                           0, 0, &cached_state, GFP_NOFS);
3601
3602         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3603                                         &cached_state);
3604         if (ret) {
3605                 unlock_extent_cached(io_tree, page_start, page_end,
3606                                      &cached_state, GFP_NOFS);
3607                 goto out_unlock;
3608         }
3609
3610         if (offset != PAGE_CACHE_SIZE) {
3611                 if (!len)
3612                         len = PAGE_CACHE_SIZE - offset;
3613                 kaddr = kmap(page);
3614                 if (front)
3615                         memset(kaddr, 0, offset);
3616                 else
3617                         memset(kaddr + offset, 0, len);
3618                 flush_dcache_page(page);
3619                 kunmap(page);
3620         }
3621         ClearPageChecked(page);
3622         set_page_dirty(page);
3623         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3624                              GFP_NOFS);
3625
3626 out_unlock:
3627         if (ret)
3628                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3629         unlock_page(page);
3630         page_cache_release(page);
3631 out:
3632         return ret;
3633 }
3634
3635 /*
3636  * This function puts in dummy file extents for the area we're creating a hole
3637  * for.  So if we are truncating this file to a larger size we need to insert
3638  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3639  * the range between oldsize and size
3640  */
3641 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3642 {
3643         struct btrfs_trans_handle *trans;
3644         struct btrfs_root *root = BTRFS_I(inode)->root;
3645         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3646         struct extent_map *em = NULL;
3647         struct extent_state *cached_state = NULL;
3648         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3649         u64 mask = root->sectorsize - 1;
3650         u64 hole_start = (oldsize + mask) & ~mask;
3651         u64 block_end = (size + mask) & ~mask;
3652         u64 last_byte;
3653         u64 cur_offset;
3654         u64 hole_size;
3655         int err = 0;
3656
3657         if (size <= hole_start)
3658                 return 0;
3659
3660         while (1) {
3661                 struct btrfs_ordered_extent *ordered;
3662                 btrfs_wait_ordered_range(inode, hole_start,
3663                                          block_end - hole_start);
3664                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3665                                  &cached_state);
3666                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3667                 if (!ordered)
3668                         break;
3669                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3670                                      &cached_state, GFP_NOFS);
3671                 btrfs_put_ordered_extent(ordered);
3672         }
3673
3674         cur_offset = hole_start;
3675         while (1) {
3676                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3677                                 block_end - cur_offset, 0);
3678                 if (IS_ERR(em)) {
3679                         err = PTR_ERR(em);
3680                         break;
3681                 }
3682                 last_byte = min(extent_map_end(em), block_end);
3683                 last_byte = (last_byte + mask) & ~mask;
3684                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3685                         struct extent_map *hole_em;
3686                         hole_size = last_byte - cur_offset;
3687
3688                         trans = btrfs_start_transaction(root, 3);
3689                         if (IS_ERR(trans)) {
3690                                 err = PTR_ERR(trans);
3691                                 break;
3692                         }
3693
3694                         err = btrfs_drop_extents(trans, root, inode,
3695                                                  cur_offset,
3696                                                  cur_offset + hole_size, 1);
3697                         if (err) {
3698                                 btrfs_abort_transaction(trans, root, err);
3699                                 btrfs_end_transaction(trans, root);
3700                                 break;
3701                         }
3702
3703                         err = btrfs_insert_file_extent(trans, root,
3704                                         btrfs_ino(inode), cur_offset, 0,
3705                                         0, hole_size, 0, hole_size,
3706                                         0, 0, 0);
3707                         if (err) {
3708                                 btrfs_abort_transaction(trans, root, err);
3709                                 btrfs_end_transaction(trans, root);
3710                                 break;
3711                         }
3712
3713                         btrfs_drop_extent_cache(inode, cur_offset,
3714                                                 cur_offset + hole_size - 1, 0);
3715                         hole_em = alloc_extent_map();
3716                         if (!hole_em) {
3717                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3718                                         &BTRFS_I(inode)->runtime_flags);
3719                                 goto next;
3720                         }
3721                         hole_em->start = cur_offset;
3722                         hole_em->len = hole_size;
3723                         hole_em->orig_start = cur_offset;
3724
3725                         hole_em->block_start = EXTENT_MAP_HOLE;
3726                         hole_em->block_len = 0;
3727                         hole_em->orig_block_len = 0;
3728                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3729                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3730                         hole_em->generation = trans->transid;
3731
3732                         while (1) {
3733                                 write_lock(&em_tree->lock);
3734                                 err = add_extent_mapping(em_tree, hole_em);
3735                                 if (!err)
3736                                         list_move(&hole_em->list,
3737                                                   &em_tree->modified_extents);
3738                                 write_unlock(&em_tree->lock);
3739                                 if (err != -EEXIST)
3740                                         break;
3741                                 btrfs_drop_extent_cache(inode, cur_offset,
3742                                                         cur_offset +
3743                                                         hole_size - 1, 0);
3744                         }
3745                         free_extent_map(hole_em);
3746 next:
3747                         btrfs_update_inode(trans, root, inode);
3748                         btrfs_end_transaction(trans, root);
3749                 }
3750                 free_extent_map(em);
3751                 em = NULL;
3752                 cur_offset = last_byte;
3753                 if (cur_offset >= block_end)
3754                         break;
3755         }
3756
3757         free_extent_map(em);
3758         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3759                              GFP_NOFS);
3760         return err;
3761 }
3762
3763 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3764 {
3765         struct btrfs_root *root = BTRFS_I(inode)->root;
3766         struct btrfs_trans_handle *trans;
3767         loff_t oldsize = i_size_read(inode);
3768         int ret;
3769
3770         if (newsize == oldsize)
3771                 return 0;
3772
3773         if (newsize > oldsize) {
3774                 truncate_pagecache(inode, oldsize, newsize);
3775                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3776                 if (ret)
3777                         return ret;
3778
3779                 trans = btrfs_start_transaction(root, 1);
3780                 if (IS_ERR(trans))
3781                         return PTR_ERR(trans);
3782
3783                 i_size_write(inode, newsize);
3784                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3785                 ret = btrfs_update_inode(trans, root, inode);
3786                 btrfs_end_transaction(trans, root);
3787         } else {
3788
3789                 /*
3790                  * We're truncating a file that used to have good data down to
3791                  * zero. Make sure it gets into the ordered flush list so that
3792                  * any new writes get down to disk quickly.
3793                  */
3794                 if (newsize == 0)
3795                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3796                                 &BTRFS_I(inode)->runtime_flags);
3797
3798                 /*
3799                  * 1 for the orphan item we're going to add
3800                  * 1 for the orphan item deletion.
3801                  */
3802                 trans = btrfs_start_transaction(root, 2);
3803                 if (IS_ERR(trans))
3804                         return PTR_ERR(trans);
3805
3806                 /*
3807                  * We need to do this in case we fail at _any_ point during the
3808                  * actual truncate.  Once we do the truncate_setsize we could
3809                  * invalidate pages which forces any outstanding ordered io to
3810                  * be instantly completed which will give us extents that need
3811                  * to be truncated.  If we fail to get an orphan inode down we
3812                  * could have left over extents that were never meant to live,
3813                  * so we need to garuntee from this point on that everything
3814                  * will be consistent.
3815                  */
3816                 ret = btrfs_orphan_add(trans, inode);
3817                 btrfs_end_transaction(trans, root);
3818                 if (ret)
3819                         return ret;
3820
3821                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3822                 truncate_setsize(inode, newsize);
3823                 ret = btrfs_truncate(inode);
3824                 if (ret && inode->i_nlink)
3825                         btrfs_orphan_del(NULL, inode);
3826         }
3827
3828         return ret;
3829 }
3830
3831 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3832 {
3833         struct inode *inode = dentry->d_inode;
3834         struct btrfs_root *root = BTRFS_I(inode)->root;
3835         int err;
3836
3837         if (btrfs_root_readonly(root))
3838                 return -EROFS;
3839
3840         err = inode_change_ok(inode, attr);
3841         if (err)
3842                 return err;
3843
3844         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3845                 err = btrfs_setsize(inode, attr->ia_size);
3846                 if (err)
3847                         return err;
3848         }
3849
3850         if (attr->ia_valid) {
3851                 setattr_copy(inode, attr);
3852                 inode_inc_iversion(inode);
3853                 err = btrfs_dirty_inode(inode);
3854
3855                 if (!err && attr->ia_valid & ATTR_MODE)
3856                         err = btrfs_acl_chmod(inode);
3857         }
3858
3859         return err;
3860 }
3861
3862 void btrfs_evict_inode(struct inode *inode)
3863 {
3864         struct btrfs_trans_handle *trans;
3865         struct btrfs_root *root = BTRFS_I(inode)->root;
3866         struct btrfs_block_rsv *rsv, *global_rsv;
3867         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3868         int ret;
3869
3870         trace_btrfs_inode_evict(inode);
3871
3872         truncate_inode_pages(&inode->i_data, 0);
3873         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3874                                btrfs_is_free_space_inode(inode)))
3875                 goto no_delete;
3876
3877         if (is_bad_inode(inode)) {
3878                 btrfs_orphan_del(NULL, inode);
3879                 goto no_delete;
3880         }
3881         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3882         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3883
3884         if (root->fs_info->log_root_recovering) {
3885                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3886                                  &BTRFS_I(inode)->runtime_flags));
3887                 goto no_delete;
3888         }
3889
3890         if (inode->i_nlink > 0) {
3891                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3892                 goto no_delete;
3893         }
3894
3895         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3896         if (!rsv) {
3897                 btrfs_orphan_del(NULL, inode);
3898                 goto no_delete;
3899         }
3900         rsv->size = min_size;
3901         rsv->failfast = 1;
3902         global_rsv = &root->fs_info->global_block_rsv;
3903
3904         btrfs_i_size_write(inode, 0);
3905
3906         /*
3907          * This is a bit simpler than btrfs_truncate since we've already
3908          * reserved our space for our orphan item in the unlink, so we just
3909          * need to reserve some slack space in case we add bytes and update
3910          * inode item when doing the truncate.
3911          */
3912         while (1) {
3913                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3914                                              BTRFS_RESERVE_FLUSH_LIMIT);
3915
3916                 /*
3917                  * Try and steal from the global reserve since we will
3918                  * likely not use this space anyway, we want to try as
3919                  * hard as possible to get this to work.
3920                  */
3921                 if (ret)
3922                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3923
3924                 if (ret) {
3925                         printk(KERN_WARNING "Could not get space for a "
3926                                "delete, will truncate on mount %d\n", ret);
3927                         btrfs_orphan_del(NULL, inode);
3928                         btrfs_free_block_rsv(root, rsv);
3929                         goto no_delete;
3930                 }
3931
3932                 trans = btrfs_start_transaction_lflush(root, 1);
3933                 if (IS_ERR(trans)) {
3934                         btrfs_orphan_del(NULL, inode);
3935                         btrfs_free_block_rsv(root, rsv);
3936                         goto no_delete;
3937                 }
3938
3939                 trans->block_rsv = rsv;
3940
3941                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3942                 if (ret != -ENOSPC)
3943                         break;
3944
3945                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3946                 ret = btrfs_update_inode(trans, root, inode);
3947                 BUG_ON(ret);
3948
3949                 btrfs_end_transaction(trans, root);
3950                 trans = NULL;
3951                 btrfs_btree_balance_dirty(root);
3952         }
3953
3954         btrfs_free_block_rsv(root, rsv);
3955
3956         if (ret == 0) {
3957                 trans->block_rsv = root->orphan_block_rsv;
3958                 ret = btrfs_orphan_del(trans, inode);
3959                 BUG_ON(ret);
3960         }
3961
3962         trans->block_rsv = &root->fs_info->trans_block_rsv;
3963         if (!(root == root->fs_info->tree_root ||
3964               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3965                 btrfs_return_ino(root, btrfs_ino(inode));
3966
3967         btrfs_end_transaction(trans, root);
3968         btrfs_btree_balance_dirty(root);
3969 no_delete:
3970         clear_inode(inode);
3971         return;
3972 }
3973
3974 /*
3975  * this returns the key found in the dir entry in the location pointer.
3976  * If no dir entries were found, location->objectid is 0.
3977  */
3978 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3979                                struct btrfs_key *location)
3980 {
3981         const char *name = dentry->d_name.name;
3982         int namelen = dentry->d_name.len;
3983         struct btrfs_dir_item *di;
3984         struct btrfs_path *path;
3985         struct btrfs_root *root = BTRFS_I(dir)->root;
3986         int ret = 0;
3987
3988         path = btrfs_alloc_path();
3989         if (!path)
3990                 return -ENOMEM;
3991
3992         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3993                                     namelen, 0);
3994         if (IS_ERR(di))
3995                 ret = PTR_ERR(di);
3996
3997         if (IS_ERR_OR_NULL(di))
3998                 goto out_err;
3999
4000         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4001 out:
4002         btrfs_free_path(path);
4003         return ret;
4004 out_err:
4005         location->objectid = 0;
4006         goto out;
4007 }
4008
4009 /*
4010  * when we hit a tree root in a directory, the btrfs part of the inode
4011  * needs to be changed to reflect the root directory of the tree root.  This
4012  * is kind of like crossing a mount point.
4013  */
4014 static int fixup_tree_root_location(struct btrfs_root *root,
4015                                     struct inode *dir,
4016                                     struct dentry *dentry,
4017                                     struct btrfs_key *location,
4018                                     struct btrfs_root **sub_root)
4019 {
4020         struct btrfs_path *path;
4021         struct btrfs_root *new_root;
4022         struct btrfs_root_ref *ref;
4023         struct extent_buffer *leaf;
4024         int ret;
4025         int err = 0;
4026
4027         path = btrfs_alloc_path();
4028         if (!path) {
4029                 err = -ENOMEM;
4030                 goto out;
4031         }
4032
4033         err = -ENOENT;
4034         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4035                                   BTRFS_I(dir)->root->root_key.objectid,
4036                                   location->objectid);
4037         if (ret) {
4038                 if (ret < 0)
4039                         err = ret;
4040                 goto out;
4041         }
4042
4043         leaf = path->nodes[0];
4044         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4045         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4046             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4047                 goto out;
4048
4049         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4050                                    (unsigned long)(ref + 1),
4051                                    dentry->d_name.len);
4052         if (ret)
4053                 goto out;
4054
4055         btrfs_release_path(path);
4056
4057         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4058         if (IS_ERR(new_root)) {
4059                 err = PTR_ERR(new_root);
4060                 goto out;
4061         }
4062
4063         if (btrfs_root_refs(&new_root->root_item) == 0) {
4064                 err = -ENOENT;
4065                 goto out;
4066         }
4067
4068         *sub_root = new_root;
4069         location->objectid = btrfs_root_dirid(&new_root->root_item);
4070         location->type = BTRFS_INODE_ITEM_KEY;
4071         location->offset = 0;
4072         err = 0;
4073 out:
4074         btrfs_free_path(path);
4075         return err;
4076 }
4077
4078 static void inode_tree_add(struct inode *inode)
4079 {
4080         struct btrfs_root *root = BTRFS_I(inode)->root;
4081         struct btrfs_inode *entry;
4082         struct rb_node **p;
4083         struct rb_node *parent;
4084         u64 ino = btrfs_ino(inode);
4085 again:
4086         p = &root->inode_tree.rb_node;
4087         parent = NULL;
4088
4089         if (inode_unhashed(inode))
4090                 return;
4091
4092         spin_lock(&root->inode_lock);
4093         while (*p) {
4094                 parent = *p;
4095                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4096
4097                 if (ino < btrfs_ino(&entry->vfs_inode))
4098                         p = &parent->rb_left;
4099                 else if (ino > btrfs_ino(&entry->vfs_inode))
4100                         p = &parent->rb_right;
4101                 else {
4102                         WARN_ON(!(entry->vfs_inode.i_state &
4103                                   (I_WILL_FREE | I_FREEING)));
4104                         rb_erase(parent, &root->inode_tree);
4105                         RB_CLEAR_NODE(parent);
4106                         spin_unlock(&root->inode_lock);
4107                         goto again;
4108                 }
4109         }
4110         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4111         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4112         spin_unlock(&root->inode_lock);
4113 }
4114
4115 static void inode_tree_del(struct inode *inode)
4116 {
4117         struct btrfs_root *root = BTRFS_I(inode)->root;
4118         int empty = 0;
4119
4120         spin_lock(&root->inode_lock);
4121         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4122                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4123                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4124                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4125         }
4126         spin_unlock(&root->inode_lock);
4127
4128         /*
4129          * Free space cache has inodes in the tree root, but the tree root has a
4130          * root_refs of 0, so this could end up dropping the tree root as a
4131          * snapshot, so we need the extra !root->fs_info->tree_root check to
4132          * make sure we don't drop it.
4133          */
4134         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4135             root != root->fs_info->tree_root) {
4136                 synchronize_srcu(&root->fs_info->subvol_srcu);
4137                 spin_lock(&root->inode_lock);
4138                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4139                 spin_unlock(&root->inode_lock);
4140                 if (empty)
4141                         btrfs_add_dead_root(root);
4142         }
4143 }
4144
4145 void btrfs_invalidate_inodes(struct btrfs_root *root)
4146 {
4147         struct rb_node *node;
4148         struct rb_node *prev;
4149         struct btrfs_inode *entry;
4150         struct inode *inode;
4151         u64 objectid = 0;
4152
4153         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4154
4155         spin_lock(&root->inode_lock);
4156 again:
4157         node = root->inode_tree.rb_node;
4158         prev = NULL;
4159         while (node) {
4160                 prev = node;
4161                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4162
4163                 if (objectid < btrfs_ino(&entry->vfs_inode))
4164                         node = node->rb_left;
4165                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4166                         node = node->rb_right;
4167                 else
4168                         break;
4169         }
4170         if (!node) {
4171                 while (prev) {
4172                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4173                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4174                                 node = prev;
4175                                 break;
4176                         }
4177                         prev = rb_next(prev);
4178                 }
4179         }
4180         while (node) {
4181                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4182                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4183                 inode = igrab(&entry->vfs_inode);
4184                 if (inode) {
4185                         spin_unlock(&root->inode_lock);
4186                         if (atomic_read(&inode->i_count) > 1)
4187                                 d_prune_aliases(inode);
4188                         /*
4189                          * btrfs_drop_inode will have it removed from
4190                          * the inode cache when its usage count
4191                          * hits zero.
4192                          */
4193                         iput(inode);
4194                         cond_resched();
4195                         spin_lock(&root->inode_lock);
4196                         goto again;
4197                 }
4198
4199                 if (cond_resched_lock(&root->inode_lock))
4200                         goto again;
4201
4202                 node = rb_next(node);
4203         }
4204         spin_unlock(&root->inode_lock);
4205 }
4206
4207 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4208 {
4209         struct btrfs_iget_args *args = p;
4210         inode->i_ino = args->ino;
4211         BTRFS_I(inode)->root = args->root;
4212         return 0;
4213 }
4214
4215 static int btrfs_find_actor(struct inode *inode, void *opaque)
4216 {
4217         struct btrfs_iget_args *args = opaque;
4218         return args->ino == btrfs_ino(inode) &&
4219                 args->root == BTRFS_I(inode)->root;
4220 }
4221
4222 static struct inode *btrfs_iget_locked(struct super_block *s,
4223                                        u64 objectid,
4224                                        struct btrfs_root *root)
4225 {
4226         struct inode *inode;
4227         struct btrfs_iget_args args;
4228         args.ino = objectid;
4229         args.root = root;
4230
4231         inode = iget5_locked(s, objectid, btrfs_find_actor,
4232                              btrfs_init_locked_inode,
4233                              (void *)&args);
4234         return inode;
4235 }
4236
4237 /* Get an inode object given its location and corresponding root.
4238  * Returns in *is_new if the inode was read from disk
4239  */
4240 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4241                          struct btrfs_root *root, int *new)
4242 {
4243         struct inode *inode;
4244
4245         inode = btrfs_iget_locked(s, location->objectid, root);
4246         if (!inode)
4247                 return ERR_PTR(-ENOMEM);
4248
4249         if (inode->i_state & I_NEW) {
4250                 BTRFS_I(inode)->root = root;
4251                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4252                 btrfs_read_locked_inode(inode);
4253                 if (!is_bad_inode(inode)) {
4254                         inode_tree_add(inode);
4255                         unlock_new_inode(inode);
4256                         if (new)
4257                                 *new = 1;
4258                 } else {
4259                         unlock_new_inode(inode);
4260                         iput(inode);
4261                         inode = ERR_PTR(-ESTALE);
4262                 }
4263         }
4264
4265         return inode;
4266 }
4267
4268 static struct inode *new_simple_dir(struct super_block *s,
4269                                     struct btrfs_key *key,
4270                                     struct btrfs_root *root)
4271 {
4272         struct inode *inode = new_inode(s);
4273
4274         if (!inode)
4275                 return ERR_PTR(-ENOMEM);
4276
4277         BTRFS_I(inode)->root = root;
4278         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4279         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4280
4281         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4282         inode->i_op = &btrfs_dir_ro_inode_operations;
4283         inode->i_fop = &simple_dir_operations;
4284         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4285         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4286
4287         return inode;
4288 }
4289
4290 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4291 {
4292         struct inode *inode;
4293         struct btrfs_root *root = BTRFS_I(dir)->root;
4294         struct btrfs_root *sub_root = root;
4295         struct btrfs_key location;
4296         int index;
4297         int ret = 0;
4298
4299         if (dentry->d_name.len > BTRFS_NAME_LEN)
4300                 return ERR_PTR(-ENAMETOOLONG);
4301
4302         if (unlikely(d_need_lookup(dentry))) {
4303                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4304                 kfree(dentry->d_fsdata);
4305                 dentry->d_fsdata = NULL;
4306                 /* This thing is hashed, drop it for now */
4307                 d_drop(dentry);
4308         } else {
4309                 ret = btrfs_inode_by_name(dir, dentry, &location);
4310         }
4311
4312         if (ret < 0)
4313                 return ERR_PTR(ret);
4314
4315         if (location.objectid == 0)
4316                 return NULL;
4317
4318         if (location.type == BTRFS_INODE_ITEM_KEY) {
4319                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4320                 return inode;
4321         }
4322
4323         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4324
4325         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4326         ret = fixup_tree_root_location(root, dir, dentry,
4327                                        &location, &sub_root);
4328         if (ret < 0) {
4329                 if (ret != -ENOENT)
4330                         inode = ERR_PTR(ret);
4331                 else
4332                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4333         } else {
4334                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4335         }
4336         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4337
4338         if (!IS_ERR(inode) && root != sub_root) {
4339                 down_read(&root->fs_info->cleanup_work_sem);
4340                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4341                         ret = btrfs_orphan_cleanup(sub_root);
4342                 up_read(&root->fs_info->cleanup_work_sem);
4343                 if (ret)
4344                         inode = ERR_PTR(ret);
4345         }
4346
4347         return inode;
4348 }
4349
4350 static int btrfs_dentry_delete(const struct dentry *dentry)
4351 {
4352         struct btrfs_root *root;
4353         struct inode *inode = dentry->d_inode;
4354
4355         if (!inode && !IS_ROOT(dentry))
4356                 inode = dentry->d_parent->d_inode;
4357
4358         if (inode) {
4359                 root = BTRFS_I(inode)->root;
4360                 if (btrfs_root_refs(&root->root_item) == 0)
4361                         return 1;
4362
4363                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4364                         return 1;
4365         }
4366         return 0;
4367 }
4368
4369 static void btrfs_dentry_release(struct dentry *dentry)
4370 {
4371         if (dentry->d_fsdata)
4372                 kfree(dentry->d_fsdata);
4373 }
4374
4375 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4376                                    unsigned int flags)
4377 {
4378         struct dentry *ret;
4379
4380         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4381         if (unlikely(d_need_lookup(dentry))) {
4382                 spin_lock(&dentry->d_lock);
4383                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4384                 spin_unlock(&dentry->d_lock);
4385         }
4386         return ret;
4387 }
4388
4389 unsigned char btrfs_filetype_table[] = {
4390         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4391 };
4392
4393 static int btrfs_real_readdir(struct file *filp, void *dirent,
4394                               filldir_t filldir)
4395 {
4396         struct inode *inode = filp->f_dentry->d_inode;
4397         struct btrfs_root *root = BTRFS_I(inode)->root;
4398         struct btrfs_item *item;
4399         struct btrfs_dir_item *di;
4400         struct btrfs_key key;
4401         struct btrfs_key found_key;
4402         struct btrfs_path *path;
4403         struct list_head ins_list;
4404         struct list_head del_list;
4405         int ret;
4406         struct extent_buffer *leaf;
4407         int slot;
4408         unsigned char d_type;
4409         int over = 0;
4410         u32 di_cur;
4411         u32 di_total;
4412         u32 di_len;
4413         int key_type = BTRFS_DIR_INDEX_KEY;
4414         char tmp_name[32];
4415         char *name_ptr;
4416         int name_len;
4417         int is_curr = 0;        /* filp->f_pos points to the current index? */
4418
4419         /* FIXME, use a real flag for deciding about the key type */
4420         if (root->fs_info->tree_root == root)
4421                 key_type = BTRFS_DIR_ITEM_KEY;
4422
4423         /* special case for "." */
4424         if (filp->f_pos == 0) {
4425                 over = filldir(dirent, ".", 1,
4426                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4427                 if (over)
4428                         return 0;
4429                 filp->f_pos = 1;
4430         }
4431         /* special case for .., just use the back ref */
4432         if (filp->f_pos == 1) {
4433                 u64 pino = parent_ino(filp->f_path.dentry);
4434                 over = filldir(dirent, "..", 2,
4435                                filp->f_pos, pino, DT_DIR);
4436                 if (over)
4437                         return 0;
4438                 filp->f_pos = 2;
4439         }
4440         path = btrfs_alloc_path();
4441         if (!path)
4442                 return -ENOMEM;
4443
4444         path->reada = 1;
4445
4446         if (key_type == BTRFS_DIR_INDEX_KEY) {
4447                 INIT_LIST_HEAD(&ins_list);
4448                 INIT_LIST_HEAD(&del_list);
4449                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4450         }
4451
4452         btrfs_set_key_type(&key, key_type);
4453         key.offset = filp->f_pos;
4454         key.objectid = btrfs_ino(inode);
4455
4456         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4457         if (ret < 0)
4458                 goto err;
4459
4460         while (1) {
4461                 leaf = path->nodes[0];
4462                 slot = path->slots[0];
4463                 if (slot >= btrfs_header_nritems(leaf)) {
4464                         ret = btrfs_next_leaf(root, path);
4465                         if (ret < 0)
4466                                 goto err;
4467                         else if (ret > 0)
4468                                 break;
4469                         continue;
4470                 }
4471
4472                 item = btrfs_item_nr(leaf, slot);
4473                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4474
4475                 if (found_key.objectid != key.objectid)
4476                         break;
4477                 if (btrfs_key_type(&found_key) != key_type)
4478                         break;
4479                 if (found_key.offset < filp->f_pos)
4480                         goto next;
4481                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4482                     btrfs_should_delete_dir_index(&del_list,
4483                                                   found_key.offset))
4484                         goto next;
4485
4486                 filp->f_pos = found_key.offset;
4487                 is_curr = 1;
4488
4489                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4490                 di_cur = 0;
4491                 di_total = btrfs_item_size(leaf, item);
4492
4493                 while (di_cur < di_total) {
4494                         struct btrfs_key location;
4495
4496                         if (verify_dir_item(root, leaf, di))
4497                                 break;
4498
4499                         name_len = btrfs_dir_name_len(leaf, di);
4500                         if (name_len <= sizeof(tmp_name)) {
4501                                 name_ptr = tmp_name;
4502                         } else {
4503                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4504                                 if (!name_ptr) {
4505                                         ret = -ENOMEM;
4506                                         goto err;
4507                                 }
4508                         }
4509                         read_extent_buffer(leaf, name_ptr,
4510                                            (unsigned long)(di + 1), name_len);
4511
4512                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4513                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4514
4515
4516                         /* is this a reference to our own snapshot? If so
4517                          * skip it.
4518                          *
4519                          * In contrast to old kernels, we insert the snapshot's
4520                          * dir item and dir index after it has been created, so
4521                          * we won't find a reference to our own snapshot. We
4522                          * still keep the following code for backward
4523                          * compatibility.
4524                          */
4525                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4526                             location.objectid == root->root_key.objectid) {
4527                                 over = 0;
4528                                 goto skip;
4529                         }
4530                         over = filldir(dirent, name_ptr, name_len,
4531                                        found_key.offset, location.objectid,
4532                                        d_type);
4533
4534 skip:
4535                         if (name_ptr != tmp_name)
4536                                 kfree(name_ptr);
4537
4538                         if (over)
4539                                 goto nopos;
4540                         di_len = btrfs_dir_name_len(leaf, di) +
4541                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4542                         di_cur += di_len;
4543                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4544                 }
4545 next:
4546                 path->slots[0]++;
4547         }
4548
4549         if (key_type == BTRFS_DIR_INDEX_KEY) {
4550                 if (is_curr)
4551                         filp->f_pos++;
4552                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4553                                                       &ins_list);
4554                 if (ret)
4555                         goto nopos;
4556         }
4557
4558         /* Reached end of directory/root. Bump pos past the last item. */
4559         if (key_type == BTRFS_DIR_INDEX_KEY)
4560                 /*
4561                  * 32-bit glibc will use getdents64, but then strtol -
4562                  * so the last number we can serve is this.
4563                  */
4564                 filp->f_pos = 0x7fffffff;
4565         else
4566                 filp->f_pos++;
4567 nopos:
4568         ret = 0;
4569 err:
4570         if (key_type == BTRFS_DIR_INDEX_KEY)
4571                 btrfs_put_delayed_items(&ins_list, &del_list);
4572         btrfs_free_path(path);
4573         return ret;
4574 }
4575
4576 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4577 {
4578         struct btrfs_root *root = BTRFS_I(inode)->root;
4579         struct btrfs_trans_handle *trans;
4580         int ret = 0;
4581         bool nolock = false;
4582
4583         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4584                 return 0;
4585
4586         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4587                 nolock = true;
4588
4589         if (wbc->sync_mode == WB_SYNC_ALL) {
4590                 if (nolock)
4591                         trans = btrfs_join_transaction_nolock(root);
4592                 else
4593                         trans = btrfs_join_transaction(root);
4594                 if (IS_ERR(trans))
4595                         return PTR_ERR(trans);
4596                 ret = btrfs_commit_transaction(trans, root);
4597         }
4598         return ret;
4599 }
4600
4601 /*
4602  * This is somewhat expensive, updating the tree every time the
4603  * inode changes.  But, it is most likely to find the inode in cache.
4604  * FIXME, needs more benchmarking...there are no reasons other than performance
4605  * to keep or drop this code.
4606  */
4607 int btrfs_dirty_inode(struct inode *inode)
4608 {
4609         struct btrfs_root *root = BTRFS_I(inode)->root;
4610         struct btrfs_trans_handle *trans;
4611         int ret;
4612
4613         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4614                 return 0;
4615
4616         trans = btrfs_join_transaction(root);
4617         if (IS_ERR(trans))
4618                 return PTR_ERR(trans);
4619
4620         ret = btrfs_update_inode(trans, root, inode);
4621         if (ret && ret == -ENOSPC) {
4622                 /* whoops, lets try again with the full transaction */
4623                 btrfs_end_transaction(trans, root);
4624                 trans = btrfs_start_transaction(root, 1);
4625                 if (IS_ERR(trans))
4626                         return PTR_ERR(trans);
4627
4628                 ret = btrfs_update_inode(trans, root, inode);
4629         }
4630         btrfs_end_transaction(trans, root);
4631         if (BTRFS_I(inode)->delayed_node)
4632                 btrfs_balance_delayed_items(root);
4633
4634         return ret;
4635 }
4636
4637 /*
4638  * This is a copy of file_update_time.  We need this so we can return error on
4639  * ENOSPC for updating the inode in the case of file write and mmap writes.
4640  */
4641 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4642                              int flags)
4643 {
4644         struct btrfs_root *root = BTRFS_I(inode)->root;
4645
4646         if (btrfs_root_readonly(root))
4647                 return -EROFS;
4648
4649         if (flags & S_VERSION)
4650                 inode_inc_iversion(inode);
4651         if (flags & S_CTIME)
4652                 inode->i_ctime = *now;
4653         if (flags & S_MTIME)
4654                 inode->i_mtime = *now;
4655         if (flags & S_ATIME)
4656                 inode->i_atime = *now;
4657         return btrfs_dirty_inode(inode);
4658 }
4659
4660 /*
4661  * find the highest existing sequence number in a directory
4662  * and then set the in-memory index_cnt variable to reflect
4663  * free sequence numbers
4664  */
4665 static int btrfs_set_inode_index_count(struct inode *inode)
4666 {
4667         struct btrfs_root *root = BTRFS_I(inode)->root;
4668         struct btrfs_key key, found_key;
4669         struct btrfs_path *path;
4670         struct extent_buffer *leaf;
4671         int ret;
4672
4673         key.objectid = btrfs_ino(inode);
4674         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4675         key.offset = (u64)-1;
4676
4677         path = btrfs_alloc_path();
4678         if (!path)
4679                 return -ENOMEM;
4680
4681         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4682         if (ret < 0)
4683                 goto out;
4684         /* FIXME: we should be able to handle this */
4685         if (ret == 0)
4686                 goto out;
4687         ret = 0;
4688
4689         /*
4690          * MAGIC NUMBER EXPLANATION:
4691          * since we search a directory based on f_pos we have to start at 2
4692          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4693          * else has to start at 2
4694          */
4695         if (path->slots[0] == 0) {
4696                 BTRFS_I(inode)->index_cnt = 2;
4697                 goto out;
4698         }
4699
4700         path->slots[0]--;
4701
4702         leaf = path->nodes[0];
4703         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4704
4705         if (found_key.objectid != btrfs_ino(inode) ||
4706             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4707                 BTRFS_I(inode)->index_cnt = 2;
4708                 goto out;
4709         }
4710
4711         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4712 out:
4713         btrfs_free_path(path);
4714         return ret;
4715 }
4716
4717 /*
4718  * helper to find a free sequence number in a given directory.  This current
4719  * code is very simple, later versions will do smarter things in the btree
4720  */
4721 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4722 {
4723         int ret = 0;
4724
4725         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4726                 ret = btrfs_inode_delayed_dir_index_count(dir);
4727                 if (ret) {
4728                         ret = btrfs_set_inode_index_count(dir);
4729                         if (ret)
4730                                 return ret;
4731                 }
4732         }
4733
4734         *index = BTRFS_I(dir)->index_cnt;
4735         BTRFS_I(dir)->index_cnt++;
4736
4737         return ret;
4738 }
4739
4740 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4741                                      struct btrfs_root *root,
4742                                      struct inode *dir,
4743                                      const char *name, int name_len,
4744                                      u64 ref_objectid, u64 objectid,
4745                                      umode_t mode, u64 *index)
4746 {
4747         struct inode *inode;
4748         struct btrfs_inode_item *inode_item;
4749         struct btrfs_key *location;
4750         struct btrfs_path *path;
4751         struct btrfs_inode_ref *ref;
4752         struct btrfs_key key[2];
4753         u32 sizes[2];
4754         unsigned long ptr;
4755         int ret;
4756         int owner;
4757
4758         path = btrfs_alloc_path();
4759         if (!path)
4760                 return ERR_PTR(-ENOMEM);
4761
4762         inode = new_inode(root->fs_info->sb);
4763         if (!inode) {
4764                 btrfs_free_path(path);
4765                 return ERR_PTR(-ENOMEM);
4766         }
4767
4768         /*
4769          * we have to initialize this early, so we can reclaim the inode
4770          * number if we fail afterwards in this function.
4771          */
4772         inode->i_ino = objectid;
4773
4774         if (dir) {
4775                 trace_btrfs_inode_request(dir);
4776
4777                 ret = btrfs_set_inode_index(dir, index);
4778                 if (ret) {
4779                         btrfs_free_path(path);
4780                         iput(inode);
4781                         return ERR_PTR(ret);
4782                 }
4783         }
4784         /*
4785          * index_cnt is ignored for everything but a dir,
4786          * btrfs_get_inode_index_count has an explanation for the magic
4787          * number
4788          */
4789         BTRFS_I(inode)->index_cnt = 2;
4790         BTRFS_I(inode)->root = root;
4791         BTRFS_I(inode)->generation = trans->transid;
4792         inode->i_generation = BTRFS_I(inode)->generation;
4793
4794         /*
4795          * We could have gotten an inode number from somebody who was fsynced
4796          * and then removed in this same transaction, so let's just set full
4797          * sync since it will be a full sync anyway and this will blow away the
4798          * old info in the log.
4799          */
4800         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4801
4802         if (S_ISDIR(mode))
4803                 owner = 0;
4804         else
4805                 owner = 1;
4806
4807         key[0].objectid = objectid;
4808         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4809         key[0].offset = 0;
4810
4811         /*
4812          * Start new inodes with an inode_ref. This is slightly more
4813          * efficient for small numbers of hard links since they will
4814          * be packed into one item. Extended refs will kick in if we
4815          * add more hard links than can fit in the ref item.
4816          */
4817         key[1].objectid = objectid;
4818         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4819         key[1].offset = ref_objectid;
4820
4821         sizes[0] = sizeof(struct btrfs_inode_item);
4822         sizes[1] = name_len + sizeof(*ref);
4823
4824         path->leave_spinning = 1;
4825         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4826         if (ret != 0)
4827                 goto fail;
4828
4829         inode_init_owner(inode, dir, mode);
4830         inode_set_bytes(inode, 0);
4831         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4832         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4833                                   struct btrfs_inode_item);
4834         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4835                              sizeof(*inode_item));
4836         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4837
4838         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4839                              struct btrfs_inode_ref);
4840         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4841         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4842         ptr = (unsigned long)(ref + 1);
4843         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4844
4845         btrfs_mark_buffer_dirty(path->nodes[0]);
4846         btrfs_free_path(path);
4847
4848         location = &BTRFS_I(inode)->location;
4849         location->objectid = objectid;
4850         location->offset = 0;
4851         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4852
4853         btrfs_inherit_iflags(inode, dir);
4854
4855         if (S_ISREG(mode)) {
4856                 if (btrfs_test_opt(root, NODATASUM))
4857                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4858                 if (btrfs_test_opt(root, NODATACOW))
4859                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4860         }
4861
4862         insert_inode_hash(inode);
4863         inode_tree_add(inode);
4864
4865         trace_btrfs_inode_new(inode);
4866         btrfs_set_inode_last_trans(trans, inode);
4867
4868         btrfs_update_root_times(trans, root);
4869
4870         return inode;
4871 fail:
4872         if (dir)
4873                 BTRFS_I(dir)->index_cnt--;
4874         btrfs_free_path(path);
4875         iput(inode);
4876         return ERR_PTR(ret);
4877 }
4878
4879 static inline u8 btrfs_inode_type(struct inode *inode)
4880 {
4881         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4882 }
4883
4884 /*
4885  * utility function to add 'inode' into 'parent_inode' with
4886  * a give name and a given sequence number.
4887  * if 'add_backref' is true, also insert a backref from the
4888  * inode to the parent directory.
4889  */
4890 int btrfs_add_link(struct btrfs_trans_handle *trans,
4891                    struct inode *parent_inode, struct inode *inode,
4892                    const char *name, int name_len, int add_backref, u64 index)
4893 {
4894         int ret = 0;
4895         struct btrfs_key key;
4896         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4897         u64 ino = btrfs_ino(inode);
4898         u64 parent_ino = btrfs_ino(parent_inode);
4899
4900         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4901                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4902         } else {
4903                 key.objectid = ino;
4904                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4905                 key.offset = 0;
4906         }
4907
4908         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4909                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4910                                          key.objectid, root->root_key.objectid,
4911                                          parent_ino, index, name, name_len);
4912         } else if (add_backref) {
4913                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4914                                              parent_ino, index);
4915         }
4916
4917         /* Nothing to clean up yet */
4918         if (ret)
4919                 return ret;
4920
4921         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4922                                     parent_inode, &key,
4923                                     btrfs_inode_type(inode), index);
4924         if (ret == -EEXIST || ret == -EOVERFLOW)
4925                 goto fail_dir_item;
4926         else if (ret) {
4927                 btrfs_abort_transaction(trans, root, ret);
4928                 return ret;
4929         }
4930
4931         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4932                            name_len * 2);
4933         inode_inc_iversion(parent_inode);
4934         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4935         ret = btrfs_update_inode(trans, root, parent_inode);
4936         if (ret)
4937                 btrfs_abort_transaction(trans, root, ret);
4938         return ret;
4939
4940 fail_dir_item:
4941         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4942                 u64 local_index;
4943                 int err;
4944                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4945                                  key.objectid, root->root_key.objectid,
4946                                  parent_ino, &local_index, name, name_len);
4947
4948         } else if (add_backref) {
4949                 u64 local_index;
4950                 int err;
4951
4952                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4953                                           ino, parent_ino, &local_index);
4954         }
4955         return ret;
4956 }
4957
4958 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4959                             struct inode *dir, struct dentry *dentry,
4960                             struct inode *inode, int backref, u64 index)
4961 {
4962         int err = btrfs_add_link(trans, dir, inode,
4963                                  dentry->d_name.name, dentry->d_name.len,
4964                                  backref, index);
4965         if (err > 0)
4966                 err = -EEXIST;
4967         return err;
4968 }
4969
4970 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4971                         umode_t mode, dev_t rdev)
4972 {
4973         struct btrfs_trans_handle *trans;
4974         struct btrfs_root *root = BTRFS_I(dir)->root;
4975         struct inode *inode = NULL;
4976         int err;
4977         int drop_inode = 0;
4978         u64 objectid;
4979         u64 index = 0;
4980
4981         if (!new_valid_dev(rdev))
4982                 return -EINVAL;
4983
4984         /*
4985          * 2 for inode item and ref
4986          * 2 for dir items
4987          * 1 for xattr if selinux is on
4988          */
4989         trans = btrfs_start_transaction(root, 5);
4990         if (IS_ERR(trans))
4991                 return PTR_ERR(trans);
4992
4993         err = btrfs_find_free_ino(root, &objectid);
4994         if (err)
4995                 goto out_unlock;
4996
4997         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4998                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4999                                 mode, &index);
5000         if (IS_ERR(inode)) {
5001                 err = PTR_ERR(inode);
5002                 goto out_unlock;
5003         }
5004
5005         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5006         if (err) {
5007                 drop_inode = 1;
5008                 goto out_unlock;
5009         }
5010
5011         err = btrfs_update_inode(trans, root, inode);
5012         if (err) {
5013                 drop_inode = 1;
5014                 goto out_unlock;
5015         }
5016
5017         /*
5018         * If the active LSM wants to access the inode during
5019         * d_instantiate it needs these. Smack checks to see
5020         * if the filesystem supports xattrs by looking at the
5021         * ops vector.
5022         */
5023
5024         inode->i_op = &btrfs_special_inode_operations;
5025         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5026         if (err)
5027                 drop_inode = 1;
5028         else {
5029                 init_special_inode(inode, inode->i_mode, rdev);
5030                 btrfs_update_inode(trans, root, inode);
5031                 d_instantiate(dentry, inode);
5032         }
5033 out_unlock:
5034         btrfs_end_transaction(trans, root);
5035         btrfs_btree_balance_dirty(root);
5036         if (drop_inode) {
5037                 inode_dec_link_count(inode);
5038                 iput(inode);
5039         }
5040         return err;
5041 }
5042
5043 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5044                         umode_t mode, bool excl)
5045 {
5046         struct btrfs_trans_handle *trans;
5047         struct btrfs_root *root = BTRFS_I(dir)->root;
5048         struct inode *inode = NULL;
5049         int drop_inode_on_err = 0;
5050         int err;
5051         u64 objectid;
5052         u64 index = 0;
5053
5054         /*
5055          * 2 for inode item and ref
5056          * 2 for dir items
5057          * 1 for xattr if selinux is on
5058          */
5059         trans = btrfs_start_transaction(root, 5);
5060         if (IS_ERR(trans))
5061                 return PTR_ERR(trans);
5062
5063         err = btrfs_find_free_ino(root, &objectid);
5064         if (err)
5065                 goto out_unlock;
5066
5067         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5068                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5069                                 mode, &index);
5070         if (IS_ERR(inode)) {
5071                 err = PTR_ERR(inode);
5072                 goto out_unlock;
5073         }
5074         drop_inode_on_err = 1;
5075
5076         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5077         if (err)
5078                 goto out_unlock;
5079
5080         err = btrfs_update_inode(trans, root, inode);
5081         if (err)
5082                 goto out_unlock;
5083
5084         /*
5085         * If the active LSM wants to access the inode during
5086         * d_instantiate it needs these. Smack checks to see
5087         * if the filesystem supports xattrs by looking at the
5088         * ops vector.
5089         */
5090         inode->i_fop = &btrfs_file_operations;
5091         inode->i_op = &btrfs_file_inode_operations;
5092
5093         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5094         if (err)
5095                 goto out_unlock;
5096
5097         inode->i_mapping->a_ops = &btrfs_aops;
5098         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5099         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5100         d_instantiate(dentry, inode);
5101
5102 out_unlock:
5103         btrfs_end_transaction(trans, root);
5104         if (err && drop_inode_on_err) {
5105                 inode_dec_link_count(inode);
5106                 iput(inode);
5107         }
5108         btrfs_btree_balance_dirty(root);
5109         return err;
5110 }
5111
5112 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5113                       struct dentry *dentry)
5114 {
5115         struct btrfs_trans_handle *trans;
5116         struct btrfs_root *root = BTRFS_I(dir)->root;
5117         struct inode *inode = old_dentry->d_inode;
5118         u64 index;
5119         int err;
5120         int drop_inode = 0;
5121
5122         /* do not allow sys_link's with other subvols of the same device */
5123         if (root->objectid != BTRFS_I(inode)->root->objectid)
5124                 return -EXDEV;
5125
5126         if (inode->i_nlink >= BTRFS_LINK_MAX)
5127                 return -EMLINK;
5128
5129         err = btrfs_set_inode_index(dir, &index);
5130         if (err)
5131                 goto fail;
5132
5133         /*
5134          * 2 items for inode and inode ref
5135          * 2 items for dir items
5136          * 1 item for parent inode
5137          */
5138         trans = btrfs_start_transaction(root, 5);
5139         if (IS_ERR(trans)) {
5140                 err = PTR_ERR(trans);
5141                 goto fail;
5142         }
5143
5144         btrfs_inc_nlink(inode);
5145         inode_inc_iversion(inode);
5146         inode->i_ctime = CURRENT_TIME;
5147         ihold(inode);
5148         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5149
5150         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5151
5152         if (err) {
5153                 drop_inode = 1;
5154         } else {
5155                 struct dentry *parent = dentry->d_parent;
5156                 err = btrfs_update_inode(trans, root, inode);
5157                 if (err)
5158                         goto fail;
5159                 d_instantiate(dentry, inode);
5160                 btrfs_log_new_name(trans, inode, NULL, parent);
5161         }
5162
5163         btrfs_end_transaction(trans, root);
5164 fail:
5165         if (drop_inode) {
5166                 inode_dec_link_count(inode);
5167                 iput(inode);
5168         }
5169         btrfs_btree_balance_dirty(root);
5170         return err;
5171 }
5172
5173 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5174 {
5175         struct inode *inode = NULL;
5176         struct btrfs_trans_handle *trans;
5177         struct btrfs_root *root = BTRFS_I(dir)->root;
5178         int err = 0;
5179         int drop_on_err = 0;
5180         u64 objectid = 0;
5181         u64 index = 0;
5182
5183         /*
5184          * 2 items for inode and ref
5185          * 2 items for dir items
5186          * 1 for xattr if selinux is on
5187          */
5188         trans = btrfs_start_transaction(root, 5);
5189         if (IS_ERR(trans))
5190                 return PTR_ERR(trans);
5191
5192         err = btrfs_find_free_ino(root, &objectid);
5193         if (err)
5194                 goto out_fail;
5195
5196         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5197                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5198                                 S_IFDIR | mode, &index);
5199         if (IS_ERR(inode)) {
5200                 err = PTR_ERR(inode);
5201                 goto out_fail;
5202         }
5203
5204         drop_on_err = 1;
5205
5206         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5207         if (err)
5208                 goto out_fail;
5209
5210         inode->i_op = &btrfs_dir_inode_operations;
5211         inode->i_fop = &btrfs_dir_file_operations;
5212
5213         btrfs_i_size_write(inode, 0);
5214         err = btrfs_update_inode(trans, root, inode);
5215         if (err)
5216                 goto out_fail;
5217
5218         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5219                              dentry->d_name.len, 0, index);
5220         if (err)
5221                 goto out_fail;
5222
5223         d_instantiate(dentry, inode);
5224         drop_on_err = 0;
5225
5226 out_fail:
5227         btrfs_end_transaction(trans, root);
5228         if (drop_on_err)
5229                 iput(inode);
5230         btrfs_btree_balance_dirty(root);
5231         return err;
5232 }
5233
5234 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5235  * and an extent that you want to insert, deal with overlap and insert
5236  * the new extent into the tree.
5237  */
5238 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5239                                 struct extent_map *existing,
5240                                 struct extent_map *em,
5241                                 u64 map_start, u64 map_len)
5242 {
5243         u64 start_diff;
5244
5245         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5246         start_diff = map_start - em->start;
5247         em->start = map_start;
5248         em->len = map_len;
5249         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5250             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5251                 em->block_start += start_diff;
5252                 em->block_len -= start_diff;
5253         }
5254         return add_extent_mapping(em_tree, em);
5255 }
5256
5257 static noinline int uncompress_inline(struct btrfs_path *path,
5258                                       struct inode *inode, struct page *page,
5259                                       size_t pg_offset, u64 extent_offset,
5260                                       struct btrfs_file_extent_item *item)
5261 {
5262         int ret;
5263         struct extent_buffer *leaf = path->nodes[0];
5264         char *tmp;
5265         size_t max_size;
5266         unsigned long inline_size;
5267         unsigned long ptr;
5268         int compress_type;
5269
5270         WARN_ON(pg_offset != 0);
5271         compress_type = btrfs_file_extent_compression(leaf, item);
5272         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5273         inline_size = btrfs_file_extent_inline_item_len(leaf,
5274                                         btrfs_item_nr(leaf, path->slots[0]));
5275         tmp = kmalloc(inline_size, GFP_NOFS);
5276         if (!tmp)
5277                 return -ENOMEM;
5278         ptr = btrfs_file_extent_inline_start(item);
5279
5280         read_extent_buffer(leaf, tmp, ptr, inline_size);
5281
5282         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5283         ret = btrfs_decompress(compress_type, tmp, page,
5284                                extent_offset, inline_size, max_size);
5285         if (ret) {
5286                 char *kaddr = kmap_atomic(page);
5287                 unsigned long copy_size = min_t(u64,
5288                                   PAGE_CACHE_SIZE - pg_offset,
5289                                   max_size - extent_offset);
5290                 memset(kaddr + pg_offset, 0, copy_size);
5291                 kunmap_atomic(kaddr);
5292         }
5293         kfree(tmp);
5294         return 0;
5295 }
5296
5297 /*
5298  * a bit scary, this does extent mapping from logical file offset to the disk.
5299  * the ugly parts come from merging extents from the disk with the in-ram
5300  * representation.  This gets more complex because of the data=ordered code,
5301  * where the in-ram extents might be locked pending data=ordered completion.
5302  *
5303  * This also copies inline extents directly into the page.
5304  */
5305
5306 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5307                                     size_t pg_offset, u64 start, u64 len,
5308                                     int create)
5309 {
5310         int ret;
5311         int err = 0;
5312         u64 bytenr;
5313         u64 extent_start = 0;
5314         u64 extent_end = 0;
5315         u64 objectid = btrfs_ino(inode);
5316         u32 found_type;
5317         struct btrfs_path *path = NULL;
5318         struct btrfs_root *root = BTRFS_I(inode)->root;
5319         struct btrfs_file_extent_item *item;
5320         struct extent_buffer *leaf;
5321         struct btrfs_key found_key;
5322         struct extent_map *em = NULL;
5323         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5324         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5325         struct btrfs_trans_handle *trans = NULL;
5326         int compress_type;
5327
5328 again:
5329         read_lock(&em_tree->lock);
5330         em = lookup_extent_mapping(em_tree, start, len);
5331         if (em)
5332                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5333         read_unlock(&em_tree->lock);
5334
5335         if (em) {
5336                 if (em->start > start || em->start + em->len <= start)
5337                         free_extent_map(em);
5338                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5339                         free_extent_map(em);
5340                 else
5341                         goto out;
5342         }
5343         em = alloc_extent_map();
5344         if (!em) {
5345                 err = -ENOMEM;
5346                 goto out;
5347         }
5348         em->bdev = root->fs_info->fs_devices->latest_bdev;
5349         em->start = EXTENT_MAP_HOLE;
5350         em->orig_start = EXTENT_MAP_HOLE;
5351         em->len = (u64)-1;
5352         em->block_len = (u64)-1;
5353
5354         if (!path) {
5355                 path = btrfs_alloc_path();
5356                 if (!path) {
5357                         err = -ENOMEM;
5358                         goto out;
5359                 }
5360                 /*
5361                  * Chances are we'll be called again, so go ahead and do
5362                  * readahead
5363                  */
5364                 path->reada = 1;
5365         }
5366
5367         ret = btrfs_lookup_file_extent(trans, root, path,
5368                                        objectid, start, trans != NULL);
5369         if (ret < 0) {
5370                 err = ret;
5371                 goto out;
5372         }
5373
5374         if (ret != 0) {
5375                 if (path->slots[0] == 0)
5376                         goto not_found;
5377                 path->slots[0]--;
5378         }
5379
5380         leaf = path->nodes[0];
5381         item = btrfs_item_ptr(leaf, path->slots[0],
5382                               struct btrfs_file_extent_item);
5383         /* are we inside the extent that was found? */
5384         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5385         found_type = btrfs_key_type(&found_key);
5386         if (found_key.objectid != objectid ||
5387             found_type != BTRFS_EXTENT_DATA_KEY) {
5388                 goto not_found;
5389         }
5390
5391         found_type = btrfs_file_extent_type(leaf, item);
5392         extent_start = found_key.offset;
5393         compress_type = btrfs_file_extent_compression(leaf, item);
5394         if (found_type == BTRFS_FILE_EXTENT_REG ||
5395             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5396                 extent_end = extent_start +
5397                        btrfs_file_extent_num_bytes(leaf, item);
5398         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5399                 size_t size;
5400                 size = btrfs_file_extent_inline_len(leaf, item);
5401                 extent_end = (extent_start + size + root->sectorsize - 1) &
5402                         ~((u64)root->sectorsize - 1);
5403         }
5404
5405         if (start >= extent_end) {
5406                 path->slots[0]++;
5407                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5408                         ret = btrfs_next_leaf(root, path);
5409                         if (ret < 0) {
5410                                 err = ret;
5411                                 goto out;
5412                         }
5413                         if (ret > 0)
5414                                 goto not_found;
5415                         leaf = path->nodes[0];
5416                 }
5417                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5418                 if (found_key.objectid != objectid ||
5419                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5420                         goto not_found;
5421                 if (start + len <= found_key.offset)
5422                         goto not_found;
5423                 em->start = start;
5424                 em->orig_start = start;
5425                 em->len = found_key.offset - start;
5426                 goto not_found_em;
5427         }
5428
5429         if (found_type == BTRFS_FILE_EXTENT_REG ||
5430             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5431                 em->start = extent_start;
5432                 em->len = extent_end - extent_start;
5433                 em->orig_start = extent_start -
5434                                  btrfs_file_extent_offset(leaf, item);
5435                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5436                                                                       item);
5437                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5438                 if (bytenr == 0) {
5439                         em->block_start = EXTENT_MAP_HOLE;
5440                         goto insert;
5441                 }
5442                 if (compress_type != BTRFS_COMPRESS_NONE) {
5443                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5444                         em->compress_type = compress_type;
5445                         em->block_start = bytenr;
5446                         em->block_len = em->orig_block_len;
5447                 } else {
5448                         bytenr += btrfs_file_extent_offset(leaf, item);
5449                         em->block_start = bytenr;
5450                         em->block_len = em->len;
5451                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5452                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5453                 }
5454                 goto insert;
5455         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5456                 unsigned long ptr;
5457                 char *map;
5458                 size_t size;
5459                 size_t extent_offset;
5460                 size_t copy_size;
5461
5462                 em->block_start = EXTENT_MAP_INLINE;
5463                 if (!page || create) {
5464                         em->start = extent_start;
5465                         em->len = extent_end - extent_start;
5466                         goto out;
5467                 }
5468
5469                 size = btrfs_file_extent_inline_len(leaf, item);
5470                 extent_offset = page_offset(page) + pg_offset - extent_start;
5471                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5472                                 size - extent_offset);
5473                 em->start = extent_start + extent_offset;
5474                 em->len = (copy_size + root->sectorsize - 1) &
5475                         ~((u64)root->sectorsize - 1);
5476                 em->orig_block_len = em->len;
5477                 em->orig_start = em->start;
5478                 if (compress_type) {
5479                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5480                         em->compress_type = compress_type;
5481                 }
5482                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5483                 if (create == 0 && !PageUptodate(page)) {
5484                         if (btrfs_file_extent_compression(leaf, item) !=
5485                             BTRFS_COMPRESS_NONE) {
5486                                 ret = uncompress_inline(path, inode, page,
5487                                                         pg_offset,
5488                                                         extent_offset, item);
5489                                 BUG_ON(ret); /* -ENOMEM */
5490                         } else {
5491                                 map = kmap(page);
5492                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5493                                                    copy_size);
5494                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5495                                         memset(map + pg_offset + copy_size, 0,
5496                                                PAGE_CACHE_SIZE - pg_offset -
5497                                                copy_size);
5498                                 }
5499                                 kunmap(page);
5500                         }
5501                         flush_dcache_page(page);
5502                 } else if (create && PageUptodate(page)) {
5503                         BUG();
5504                         if (!trans) {
5505                                 kunmap(page);
5506                                 free_extent_map(em);
5507                                 em = NULL;
5508
5509                                 btrfs_release_path(path);
5510                                 trans = btrfs_join_transaction(root);
5511
5512                                 if (IS_ERR(trans))
5513                                         return ERR_CAST(trans);
5514                                 goto again;
5515                         }
5516                         map = kmap(page);
5517                         write_extent_buffer(leaf, map + pg_offset, ptr,
5518                                             copy_size);
5519                         kunmap(page);
5520                         btrfs_mark_buffer_dirty(leaf);
5521                 }
5522                 set_extent_uptodate(io_tree, em->start,
5523                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5524                 goto insert;
5525         } else {
5526                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5527         }
5528 not_found:
5529         em->start = start;
5530         em->orig_start = start;
5531         em->len = len;
5532 not_found_em:
5533         em->block_start = EXTENT_MAP_HOLE;
5534         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5535 insert:
5536         btrfs_release_path(path);
5537         if (em->start > start || extent_map_end(em) <= start) {
5538                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5539                        "[%llu %llu]\n", (unsigned long long)em->start,
5540                        (unsigned long long)em->len,
5541                        (unsigned long long)start,
5542                        (unsigned long long)len);
5543                 err = -EIO;
5544                 goto out;
5545         }
5546
5547         err = 0;
5548         write_lock(&em_tree->lock);
5549         ret = add_extent_mapping(em_tree, em);
5550         /* it is possible that someone inserted the extent into the tree
5551          * while we had the lock dropped.  It is also possible that
5552          * an overlapping map exists in the tree
5553          */
5554         if (ret == -EEXIST) {
5555                 struct extent_map *existing;
5556
5557                 ret = 0;
5558
5559                 existing = lookup_extent_mapping(em_tree, start, len);
5560                 if (existing && (existing->start > start ||
5561                     existing->start + existing->len <= start)) {
5562                         free_extent_map(existing);
5563                         existing = NULL;
5564                 }
5565                 if (!existing) {
5566                         existing = lookup_extent_mapping(em_tree, em->start,
5567                                                          em->len);
5568                         if (existing) {
5569                                 err = merge_extent_mapping(em_tree, existing,
5570                                                            em, start,
5571                                                            root->sectorsize);
5572                                 free_extent_map(existing);
5573                                 if (err) {
5574                                         free_extent_map(em);
5575                                         em = NULL;
5576                                 }
5577                         } else {
5578                                 err = -EIO;
5579                                 free_extent_map(em);
5580                                 em = NULL;
5581                         }
5582                 } else {
5583                         free_extent_map(em);
5584                         em = existing;
5585                         err = 0;
5586                 }
5587         }
5588         write_unlock(&em_tree->lock);
5589 out:
5590
5591         if (em)
5592                 trace_btrfs_get_extent(root, em);
5593
5594         if (path)
5595                 btrfs_free_path(path);
5596         if (trans) {
5597                 ret = btrfs_end_transaction(trans, root);
5598                 if (!err)
5599                         err = ret;
5600         }
5601         if (err) {
5602                 free_extent_map(em);
5603                 return ERR_PTR(err);
5604         }
5605         BUG_ON(!em); /* Error is always set */
5606         return em;
5607 }
5608
5609 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5610                                            size_t pg_offset, u64 start, u64 len,
5611                                            int create)
5612 {
5613         struct extent_map *em;
5614         struct extent_map *hole_em = NULL;
5615         u64 range_start = start;
5616         u64 end;
5617         u64 found;
5618         u64 found_end;
5619         int err = 0;
5620
5621         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5622         if (IS_ERR(em))
5623                 return em;
5624         if (em) {
5625                 /*
5626                  * if our em maps to a hole, there might
5627                  * actually be delalloc bytes behind it
5628                  */
5629                 if (em->block_start != EXTENT_MAP_HOLE)
5630                         return em;
5631                 else
5632                         hole_em = em;
5633         }
5634
5635         /* check to see if we've wrapped (len == -1 or similar) */
5636         end = start + len;
5637         if (end < start)
5638                 end = (u64)-1;
5639         else
5640                 end -= 1;
5641
5642         em = NULL;
5643
5644         /* ok, we didn't find anything, lets look for delalloc */
5645         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5646                                  end, len, EXTENT_DELALLOC, 1);
5647         found_end = range_start + found;
5648         if (found_end < range_start)
5649                 found_end = (u64)-1;
5650
5651         /*
5652          * we didn't find anything useful, return
5653          * the original results from get_extent()
5654          */
5655         if (range_start > end || found_end <= start) {
5656                 em = hole_em;
5657                 hole_em = NULL;
5658                 goto out;
5659         }
5660
5661         /* adjust the range_start to make sure it doesn't
5662          * go backwards from the start they passed in
5663          */
5664         range_start = max(start,range_start);
5665         found = found_end - range_start;
5666
5667         if (found > 0) {
5668                 u64 hole_start = start;
5669                 u64 hole_len = len;
5670
5671                 em = alloc_extent_map();
5672                 if (!em) {
5673                         err = -ENOMEM;
5674                         goto out;
5675                 }
5676                 /*
5677                  * when btrfs_get_extent can't find anything it
5678                  * returns one huge hole
5679                  *
5680                  * make sure what it found really fits our range, and
5681                  * adjust to make sure it is based on the start from
5682                  * the caller
5683                  */
5684                 if (hole_em) {
5685                         u64 calc_end = extent_map_end(hole_em);
5686
5687                         if (calc_end <= start || (hole_em->start > end)) {
5688                                 free_extent_map(hole_em);
5689                                 hole_em = NULL;
5690                         } else {
5691                                 hole_start = max(hole_em->start, start);
5692                                 hole_len = calc_end - hole_start;
5693                         }
5694                 }
5695                 em->bdev = NULL;
5696                 if (hole_em && range_start > hole_start) {
5697                         /* our hole starts before our delalloc, so we
5698                          * have to return just the parts of the hole
5699                          * that go until  the delalloc starts
5700                          */
5701                         em->len = min(hole_len,
5702                                       range_start - hole_start);
5703                         em->start = hole_start;
5704                         em->orig_start = hole_start;
5705                         /*
5706                          * don't adjust block start at all,
5707                          * it is fixed at EXTENT_MAP_HOLE
5708                          */
5709                         em->block_start = hole_em->block_start;
5710                         em->block_len = hole_len;
5711                 } else {
5712                         em->start = range_start;
5713                         em->len = found;
5714                         em->orig_start = range_start;
5715                         em->block_start = EXTENT_MAP_DELALLOC;
5716                         em->block_len = found;
5717                 }
5718         } else if (hole_em) {
5719                 return hole_em;
5720         }
5721 out:
5722
5723         free_extent_map(hole_em);
5724         if (err) {
5725                 free_extent_map(em);
5726                 return ERR_PTR(err);
5727         }
5728         return em;
5729 }
5730
5731 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5732                                                   u64 start, u64 len)
5733 {
5734         struct btrfs_root *root = BTRFS_I(inode)->root;
5735         struct btrfs_trans_handle *trans;
5736         struct extent_map *em;
5737         struct btrfs_key ins;
5738         u64 alloc_hint;
5739         int ret;
5740
5741         trans = btrfs_join_transaction(root);
5742         if (IS_ERR(trans))
5743                 return ERR_CAST(trans);
5744
5745         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5746
5747         alloc_hint = get_extent_allocation_hint(inode, start, len);
5748         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5749                                    alloc_hint, &ins, 1);
5750         if (ret) {
5751                 em = ERR_PTR(ret);
5752                 goto out;
5753         }
5754
5755         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5756                               ins.offset, ins.offset, 0);
5757         if (IS_ERR(em))
5758                 goto out;
5759
5760         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5761                                            ins.offset, ins.offset, 0);
5762         if (ret) {
5763                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5764                 em = ERR_PTR(ret);
5765         }
5766 out:
5767         btrfs_end_transaction(trans, root);
5768         return em;
5769 }
5770
5771 /*
5772  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5773  * block must be cow'd
5774  */
5775 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5776                                       struct inode *inode, u64 offset, u64 len)
5777 {
5778         struct btrfs_path *path;
5779         int ret;
5780         struct extent_buffer *leaf;
5781         struct btrfs_root *root = BTRFS_I(inode)->root;
5782         struct btrfs_file_extent_item *fi;
5783         struct btrfs_key key;
5784         u64 disk_bytenr;
5785         u64 backref_offset;
5786         u64 extent_end;
5787         u64 num_bytes;
5788         int slot;
5789         int found_type;
5790
5791         path = btrfs_alloc_path();
5792         if (!path)
5793                 return -ENOMEM;
5794
5795         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5796                                        offset, 0);
5797         if (ret < 0)
5798                 goto out;
5799
5800         slot = path->slots[0];
5801         if (ret == 1) {
5802                 if (slot == 0) {
5803                         /* can't find the item, must cow */
5804                         ret = 0;
5805                         goto out;
5806                 }
5807                 slot--;
5808         }
5809         ret = 0;
5810         leaf = path->nodes[0];
5811         btrfs_item_key_to_cpu(leaf, &key, slot);
5812         if (key.objectid != btrfs_ino(inode) ||
5813             key.type != BTRFS_EXTENT_DATA_KEY) {
5814                 /* not our file or wrong item type, must cow */
5815                 goto out;
5816         }
5817
5818         if (key.offset > offset) {
5819                 /* Wrong offset, must cow */
5820                 goto out;
5821         }
5822
5823         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5824         found_type = btrfs_file_extent_type(leaf, fi);
5825         if (found_type != BTRFS_FILE_EXTENT_REG &&
5826             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5827                 /* not a regular extent, must cow */
5828                 goto out;
5829         }
5830         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5831         backref_offset = btrfs_file_extent_offset(leaf, fi);
5832
5833         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5834         if (extent_end < offset + len) {
5835                 /* extent doesn't include our full range, must cow */
5836                 goto out;
5837         }
5838
5839         if (btrfs_extent_readonly(root, disk_bytenr))
5840                 goto out;
5841
5842         /*
5843          * look for other files referencing this extent, if we
5844          * find any we must cow
5845          */
5846         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5847                                   key.offset - backref_offset, disk_bytenr))
5848                 goto out;
5849
5850         /*
5851          * adjust disk_bytenr and num_bytes to cover just the bytes
5852          * in this extent we are about to write.  If there
5853          * are any csums in that range we have to cow in order
5854          * to keep the csums correct
5855          */
5856         disk_bytenr += backref_offset;
5857         disk_bytenr += offset - key.offset;
5858         num_bytes = min(offset + len, extent_end) - offset;
5859         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5860                                 goto out;
5861         /*
5862          * all of the above have passed, it is safe to overwrite this extent
5863          * without cow
5864          */
5865         ret = 1;
5866 out:
5867         btrfs_free_path(path);
5868         return ret;
5869 }
5870
5871 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5872                               struct extent_state **cached_state, int writing)
5873 {
5874         struct btrfs_ordered_extent *ordered;
5875         int ret = 0;
5876
5877         while (1) {
5878                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5879                                  0, cached_state);
5880                 /*
5881                  * We're concerned with the entire range that we're going to be
5882                  * doing DIO to, so we need to make sure theres no ordered
5883                  * extents in this range.
5884                  */
5885                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5886                                                      lockend - lockstart + 1);
5887
5888                 /*
5889                  * We need to make sure there are no buffered pages in this
5890                  * range either, we could have raced between the invalidate in
5891                  * generic_file_direct_write and locking the extent.  The
5892                  * invalidate needs to happen so that reads after a write do not
5893                  * get stale data.
5894                  */
5895                 if (!ordered && (!writing ||
5896                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5897                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5898                                     *cached_state)))
5899                         break;
5900
5901                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5902                                      cached_state, GFP_NOFS);
5903
5904                 if (ordered) {
5905                         btrfs_start_ordered_extent(inode, ordered, 1);
5906                         btrfs_put_ordered_extent(ordered);
5907                 } else {
5908                         /* Screw you mmap */
5909                         ret = filemap_write_and_wait_range(inode->i_mapping,
5910                                                            lockstart,
5911                                                            lockend);
5912                         if (ret)
5913                                 break;
5914
5915                         /*
5916                          * If we found a page that couldn't be invalidated just
5917                          * fall back to buffered.
5918                          */
5919                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5920                                         lockstart >> PAGE_CACHE_SHIFT,
5921                                         lockend >> PAGE_CACHE_SHIFT);
5922                         if (ret)
5923                                 break;
5924                 }
5925
5926                 cond_resched();
5927         }
5928
5929         return ret;
5930 }
5931
5932 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5933                                            u64 len, u64 orig_start,
5934                                            u64 block_start, u64 block_len,
5935                                            u64 orig_block_len, int type)
5936 {
5937         struct extent_map_tree *em_tree;
5938         struct extent_map *em;
5939         struct btrfs_root *root = BTRFS_I(inode)->root;
5940         int ret;
5941
5942         em_tree = &BTRFS_I(inode)->extent_tree;
5943         em = alloc_extent_map();
5944         if (!em)
5945                 return ERR_PTR(-ENOMEM);
5946
5947         em->start = start;
5948         em->orig_start = orig_start;
5949         em->len = len;
5950         em->block_len = block_len;
5951         em->block_start = block_start;
5952         em->bdev = root->fs_info->fs_devices->latest_bdev;
5953         em->orig_block_len = orig_block_len;
5954         em->generation = -1;
5955         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5956         if (type == BTRFS_ORDERED_PREALLOC)
5957                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5958
5959         do {
5960                 btrfs_drop_extent_cache(inode, em->start,
5961                                 em->start + em->len - 1, 0);
5962                 write_lock(&em_tree->lock);
5963                 ret = add_extent_mapping(em_tree, em);
5964                 if (!ret)
5965                         list_move(&em->list,
5966                                   &em_tree->modified_extents);
5967                 write_unlock(&em_tree->lock);
5968         } while (ret == -EEXIST);
5969
5970         if (ret) {
5971                 free_extent_map(em);
5972                 return ERR_PTR(ret);
5973         }
5974
5975         return em;
5976 }
5977
5978
5979 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5980                                    struct buffer_head *bh_result, int create)
5981 {
5982         struct extent_map *em;
5983         struct btrfs_root *root = BTRFS_I(inode)->root;
5984         struct extent_state *cached_state = NULL;
5985         u64 start = iblock << inode->i_blkbits;
5986         u64 lockstart, lockend;
5987         u64 len = bh_result->b_size;
5988         struct btrfs_trans_handle *trans;
5989         int unlock_bits = EXTENT_LOCKED;
5990         int ret;
5991
5992         if (create) {
5993                 ret = btrfs_delalloc_reserve_space(inode, len);
5994                 if (ret)
5995                         return ret;
5996                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5997         } else {
5998                 len = min_t(u64, len, root->sectorsize);
5999         }
6000
6001         lockstart = start;
6002         lockend = start + len - 1;
6003
6004         /*
6005          * If this errors out it's because we couldn't invalidate pagecache for
6006          * this range and we need to fallback to buffered.
6007          */
6008         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6009                 return -ENOTBLK;
6010
6011         if (create) {
6012                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6013                                      lockend, EXTENT_DELALLOC, NULL,
6014                                      &cached_state, GFP_NOFS);
6015                 if (ret)
6016                         goto unlock_err;
6017         }
6018
6019         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6020         if (IS_ERR(em)) {
6021                 ret = PTR_ERR(em);
6022                 goto unlock_err;
6023         }
6024
6025         /*
6026          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6027          * io.  INLINE is special, and we could probably kludge it in here, but
6028          * it's still buffered so for safety lets just fall back to the generic
6029          * buffered path.
6030          *
6031          * For COMPRESSED we _have_ to read the entire extent in so we can
6032          * decompress it, so there will be buffering required no matter what we
6033          * do, so go ahead and fallback to buffered.
6034          *
6035          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6036          * to buffered IO.  Don't blame me, this is the price we pay for using
6037          * the generic code.
6038          */
6039         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6040             em->block_start == EXTENT_MAP_INLINE) {
6041                 free_extent_map(em);
6042                 ret = -ENOTBLK;
6043                 goto unlock_err;
6044         }
6045
6046         /* Just a good old fashioned hole, return */
6047         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6048                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6049                 free_extent_map(em);
6050                 ret = 0;
6051                 goto unlock_err;
6052         }
6053
6054         /*
6055          * We don't allocate a new extent in the following cases
6056          *
6057          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6058          * existing extent.
6059          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6060          * just use the extent.
6061          *
6062          */
6063         if (!create) {
6064                 len = min(len, em->len - (start - em->start));
6065                 lockstart = start + len;
6066                 goto unlock;
6067         }
6068
6069         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6070             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6071              em->block_start != EXTENT_MAP_HOLE)) {
6072                 int type;
6073                 int ret;
6074                 u64 block_start;
6075
6076                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6077                         type = BTRFS_ORDERED_PREALLOC;
6078                 else
6079                         type = BTRFS_ORDERED_NOCOW;
6080                 len = min(len, em->len - (start - em->start));
6081                 block_start = em->block_start + (start - em->start);
6082
6083                 /*
6084                  * we're not going to log anything, but we do need
6085                  * to make sure the current transaction stays open
6086                  * while we look for nocow cross refs
6087                  */
6088                 trans = btrfs_join_transaction(root);
6089                 if (IS_ERR(trans))
6090                         goto must_cow;
6091
6092                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6093                         u64 orig_start = em->orig_start;
6094                         u64 orig_block_len = em->orig_block_len;
6095
6096                         if (type == BTRFS_ORDERED_PREALLOC) {
6097                                 free_extent_map(em);
6098                                 em = create_pinned_em(inode, start, len,
6099                                                        orig_start,
6100                                                        block_start, len,
6101                                                        orig_block_len, type);
6102                                 if (IS_ERR(em)) {
6103                                         btrfs_end_transaction(trans, root);
6104                                         goto unlock_err;
6105                                 }
6106                         }
6107
6108                         ret = btrfs_add_ordered_extent_dio(inode, start,
6109                                            block_start, len, len, type);
6110                         btrfs_end_transaction(trans, root);
6111                         if (ret) {
6112                                 free_extent_map(em);
6113                                 goto unlock_err;
6114                         }
6115                         goto unlock;
6116                 }
6117                 btrfs_end_transaction(trans, root);
6118         }
6119 must_cow:
6120         /*
6121          * this will cow the extent, reset the len in case we changed
6122          * it above
6123          */
6124         len = bh_result->b_size;
6125         free_extent_map(em);
6126         em = btrfs_new_extent_direct(inode, start, len);
6127         if (IS_ERR(em)) {
6128                 ret = PTR_ERR(em);
6129                 goto unlock_err;
6130         }
6131         len = min(len, em->len - (start - em->start));
6132 unlock:
6133         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6134                 inode->i_blkbits;
6135         bh_result->b_size = len;
6136         bh_result->b_bdev = em->bdev;
6137         set_buffer_mapped(bh_result);
6138         if (create) {
6139                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6140                         set_buffer_new(bh_result);
6141
6142                 /*
6143                  * Need to update the i_size under the extent lock so buffered
6144                  * readers will get the updated i_size when we unlock.
6145                  */
6146                 if (start + len > i_size_read(inode))
6147                         i_size_write(inode, start + len);
6148         }
6149
6150         /*
6151          * In the case of write we need to clear and unlock the entire range,
6152          * in the case of read we need to unlock only the end area that we
6153          * aren't using if there is any left over space.
6154          */
6155         if (lockstart < lockend) {
6156                 if (create && len < lockend - lockstart) {
6157                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6158                                          lockstart + len - 1,
6159                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6160                                          &cached_state, GFP_NOFS);
6161                         /*
6162                          * Beside unlock, we also need to cleanup reserved space
6163                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6164                          */
6165                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6166                                          lockstart + len, lockend,
6167                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6168                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6169                 } else {
6170                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6171                                          lockend, unlock_bits, 1, 0,
6172                                          &cached_state, GFP_NOFS);
6173                 }
6174         } else {
6175                 free_extent_state(cached_state);
6176         }
6177
6178         free_extent_map(em);
6179
6180         return 0;
6181
6182 unlock_err:
6183         if (create)
6184                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6185
6186         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6187                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6188         return ret;
6189 }
6190
6191 struct btrfs_dio_private {
6192         struct inode *inode;
6193         u64 logical_offset;
6194         u64 disk_bytenr;
6195         u64 bytes;
6196         void *private;
6197
6198         /* number of bios pending for this dio */
6199         atomic_t pending_bios;
6200
6201         /* IO errors */
6202         int errors;
6203
6204         struct bio *orig_bio;
6205 };
6206
6207 static void btrfs_endio_direct_read(struct bio *bio, int err)
6208 {
6209         struct btrfs_dio_private *dip = bio->bi_private;
6210         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6211         struct bio_vec *bvec = bio->bi_io_vec;
6212         struct inode *inode = dip->inode;
6213         struct btrfs_root *root = BTRFS_I(inode)->root;
6214         u64 start;
6215
6216         start = dip->logical_offset;
6217         do {
6218                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6219                         struct page *page = bvec->bv_page;
6220                         char *kaddr;
6221                         u32 csum = ~(u32)0;
6222                         u64 private = ~(u32)0;
6223                         unsigned long flags;
6224
6225                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6226                                               start, &private))
6227                                 goto failed;
6228                         local_irq_save(flags);
6229                         kaddr = kmap_atomic(page);
6230                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6231                                                csum, bvec->bv_len);
6232                         btrfs_csum_final(csum, (char *)&csum);
6233                         kunmap_atomic(kaddr);
6234                         local_irq_restore(flags);
6235
6236                         flush_dcache_page(bvec->bv_page);
6237                         if (csum != private) {
6238 failed:
6239                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6240                                       " %llu csum %u private %u\n",
6241                                       (unsigned long long)btrfs_ino(inode),
6242                                       (unsigned long long)start,
6243                                       csum, (unsigned)private);
6244                                 err = -EIO;
6245                         }
6246                 }
6247
6248                 start += bvec->bv_len;
6249                 bvec++;
6250         } while (bvec <= bvec_end);
6251
6252         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6253                       dip->logical_offset + dip->bytes - 1);
6254         bio->bi_private = dip->private;
6255
6256         kfree(dip);
6257
6258         /* If we had a csum failure make sure to clear the uptodate flag */
6259         if (err)
6260                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6261         dio_end_io(bio, err);
6262 }
6263
6264 static void btrfs_endio_direct_write(struct bio *bio, int err)
6265 {
6266         struct btrfs_dio_private *dip = bio->bi_private;
6267         struct inode *inode = dip->inode;
6268         struct btrfs_root *root = BTRFS_I(inode)->root;
6269         struct btrfs_ordered_extent *ordered = NULL;
6270         u64 ordered_offset = dip->logical_offset;
6271         u64 ordered_bytes = dip->bytes;
6272         int ret;
6273
6274         if (err)
6275                 goto out_done;
6276 again:
6277         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6278                                                    &ordered_offset,
6279                                                    ordered_bytes, !err);
6280         if (!ret)
6281                 goto out_test;
6282
6283         ordered->work.func = finish_ordered_fn;
6284         ordered->work.flags = 0;
6285         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6286                            &ordered->work);
6287 out_test:
6288         /*
6289          * our bio might span multiple ordered extents.  If we haven't
6290          * completed the accounting for the whole dio, go back and try again
6291          */
6292         if (ordered_offset < dip->logical_offset + dip->bytes) {
6293                 ordered_bytes = dip->logical_offset + dip->bytes -
6294                         ordered_offset;
6295                 ordered = NULL;
6296                 goto again;
6297         }
6298 out_done:
6299         bio->bi_private = dip->private;
6300
6301         kfree(dip);
6302
6303         /* If we had an error make sure to clear the uptodate flag */
6304         if (err)
6305                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6306         dio_end_io(bio, err);
6307 }
6308
6309 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6310                                     struct bio *bio, int mirror_num,
6311                                     unsigned long bio_flags, u64 offset)
6312 {
6313         int ret;
6314         struct btrfs_root *root = BTRFS_I(inode)->root;
6315         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6316         BUG_ON(ret); /* -ENOMEM */
6317         return 0;
6318 }
6319
6320 static void btrfs_end_dio_bio(struct bio *bio, int err)
6321 {
6322         struct btrfs_dio_private *dip = bio->bi_private;
6323
6324         if (err) {
6325                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6326                       "sector %#Lx len %u err no %d\n",
6327                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6328                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6329                 dip->errors = 1;
6330
6331                 /*
6332                  * before atomic variable goto zero, we must make sure
6333                  * dip->errors is perceived to be set.
6334                  */
6335                 smp_mb__before_atomic_dec();
6336         }
6337
6338         /* if there are more bios still pending for this dio, just exit */
6339         if (!atomic_dec_and_test(&dip->pending_bios))
6340                 goto out;
6341
6342         if (dip->errors)
6343                 bio_io_error(dip->orig_bio);
6344         else {
6345                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6346                 bio_endio(dip->orig_bio, 0);
6347         }
6348 out:
6349         bio_put(bio);
6350 }
6351
6352 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6353                                        u64 first_sector, gfp_t gfp_flags)
6354 {
6355         int nr_vecs = bio_get_nr_vecs(bdev);
6356         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6357 }
6358
6359 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6360                                          int rw, u64 file_offset, int skip_sum,
6361                                          int async_submit)
6362 {
6363         int write = rw & REQ_WRITE;
6364         struct btrfs_root *root = BTRFS_I(inode)->root;
6365         int ret;
6366
6367         if (async_submit)
6368                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6369
6370         bio_get(bio);
6371
6372         if (!write) {
6373                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6374                 if (ret)
6375                         goto err;
6376         }
6377
6378         if (skip_sum)
6379                 goto map;
6380
6381         if (write && async_submit) {
6382                 ret = btrfs_wq_submit_bio(root->fs_info,
6383                                    inode, rw, bio, 0, 0,
6384                                    file_offset,
6385                                    __btrfs_submit_bio_start_direct_io,
6386                                    __btrfs_submit_bio_done);
6387                 goto err;
6388         } else if (write) {
6389                 /*
6390                  * If we aren't doing async submit, calculate the csum of the
6391                  * bio now.
6392                  */
6393                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6394                 if (ret)
6395                         goto err;
6396         } else if (!skip_sum) {
6397                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6398                 if (ret)
6399                         goto err;
6400         }
6401
6402 map:
6403         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6404 err:
6405         bio_put(bio);
6406         return ret;
6407 }
6408
6409 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6410                                     int skip_sum)
6411 {
6412         struct inode *inode = dip->inode;
6413         struct btrfs_root *root = BTRFS_I(inode)->root;
6414         struct bio *bio;
6415         struct bio *orig_bio = dip->orig_bio;
6416         struct bio_vec *bvec = orig_bio->bi_io_vec;
6417         u64 start_sector = orig_bio->bi_sector;
6418         u64 file_offset = dip->logical_offset;
6419         u64 submit_len = 0;
6420         u64 map_length;
6421         int nr_pages = 0;
6422         int ret = 0;
6423         int async_submit = 0;
6424
6425         map_length = orig_bio->bi_size;
6426         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6427                               &map_length, NULL, 0);
6428         if (ret) {
6429                 bio_put(orig_bio);
6430                 return -EIO;
6431         }
6432
6433         if (map_length >= orig_bio->bi_size) {
6434                 bio = orig_bio;
6435                 goto submit;
6436         }
6437
6438         async_submit = 1;
6439         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6440         if (!bio)
6441                 return -ENOMEM;
6442         bio->bi_private = dip;
6443         bio->bi_end_io = btrfs_end_dio_bio;
6444         atomic_inc(&dip->pending_bios);
6445
6446         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6447                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6448                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6449                                  bvec->bv_offset) < bvec->bv_len)) {
6450                         /*
6451                          * inc the count before we submit the bio so
6452                          * we know the end IO handler won't happen before
6453                          * we inc the count. Otherwise, the dip might get freed
6454                          * before we're done setting it up
6455                          */
6456                         atomic_inc(&dip->pending_bios);
6457                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6458                                                      file_offset, skip_sum,
6459                                                      async_submit);
6460                         if (ret) {
6461                                 bio_put(bio);
6462                                 atomic_dec(&dip->pending_bios);
6463                                 goto out_err;
6464                         }
6465
6466                         start_sector += submit_len >> 9;
6467                         file_offset += submit_len;
6468
6469                         submit_len = 0;
6470                         nr_pages = 0;
6471
6472                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6473                                                   start_sector, GFP_NOFS);
6474                         if (!bio)
6475                                 goto out_err;
6476                         bio->bi_private = dip;
6477                         bio->bi_end_io = btrfs_end_dio_bio;
6478
6479                         map_length = orig_bio->bi_size;
6480                         ret = btrfs_map_block(root->fs_info, READ,
6481                                               start_sector << 9,
6482                                               &map_length, NULL, 0);
6483                         if (ret) {
6484                                 bio_put(bio);
6485                                 goto out_err;
6486                         }
6487                 } else {
6488                         submit_len += bvec->bv_len;
6489                         nr_pages ++;
6490                         bvec++;
6491                 }
6492         }
6493
6494 submit:
6495         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6496                                      async_submit);
6497         if (!ret)
6498                 return 0;
6499
6500         bio_put(bio);
6501 out_err:
6502         dip->errors = 1;
6503         /*
6504          * before atomic variable goto zero, we must
6505          * make sure dip->errors is perceived to be set.
6506          */
6507         smp_mb__before_atomic_dec();
6508         if (atomic_dec_and_test(&dip->pending_bios))
6509                 bio_io_error(dip->orig_bio);
6510
6511         /* bio_end_io() will handle error, so we needn't return it */
6512         return 0;
6513 }
6514
6515 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6516                                 loff_t file_offset)
6517 {
6518         struct btrfs_root *root = BTRFS_I(inode)->root;
6519         struct btrfs_dio_private *dip;
6520         struct bio_vec *bvec = bio->bi_io_vec;
6521         int skip_sum;
6522         int write = rw & REQ_WRITE;
6523         int ret = 0;
6524
6525         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6526
6527         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6528         if (!dip) {
6529                 ret = -ENOMEM;
6530                 goto free_ordered;
6531         }
6532
6533         dip->private = bio->bi_private;
6534         dip->inode = inode;
6535         dip->logical_offset = file_offset;
6536
6537         dip->bytes = 0;
6538         do {
6539                 dip->bytes += bvec->bv_len;
6540                 bvec++;
6541         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6542
6543         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6544         bio->bi_private = dip;
6545         dip->errors = 0;
6546         dip->orig_bio = bio;
6547         atomic_set(&dip->pending_bios, 0);
6548
6549         if (write)
6550                 bio->bi_end_io = btrfs_endio_direct_write;
6551         else
6552                 bio->bi_end_io = btrfs_endio_direct_read;
6553
6554         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6555         if (!ret)
6556                 return;
6557 free_ordered:
6558         /*
6559          * If this is a write, we need to clean up the reserved space and kill
6560          * the ordered extent.
6561          */
6562         if (write) {
6563                 struct btrfs_ordered_extent *ordered;
6564                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6565                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6566                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6567                         btrfs_free_reserved_extent(root, ordered->start,
6568                                                    ordered->disk_len);
6569                 btrfs_put_ordered_extent(ordered);
6570                 btrfs_put_ordered_extent(ordered);
6571         }
6572         bio_endio(bio, ret);
6573 }
6574
6575 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6576                         const struct iovec *iov, loff_t offset,
6577                         unsigned long nr_segs)
6578 {
6579         int seg;
6580         int i;
6581         size_t size;
6582         unsigned long addr;
6583         unsigned blocksize_mask = root->sectorsize - 1;
6584         ssize_t retval = -EINVAL;
6585         loff_t end = offset;
6586
6587         if (offset & blocksize_mask)
6588                 goto out;
6589
6590         /* Check the memory alignment.  Blocks cannot straddle pages */
6591         for (seg = 0; seg < nr_segs; seg++) {
6592                 addr = (unsigned long)iov[seg].iov_base;
6593                 size = iov[seg].iov_len;
6594                 end += size;
6595                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6596                         goto out;
6597
6598                 /* If this is a write we don't need to check anymore */
6599                 if (rw & WRITE)
6600                         continue;
6601
6602                 /*
6603                  * Check to make sure we don't have duplicate iov_base's in this
6604                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6605                  * when reading back.
6606                  */
6607                 for (i = seg + 1; i < nr_segs; i++) {
6608                         if (iov[seg].iov_base == iov[i].iov_base)
6609                                 goto out;
6610                 }
6611         }
6612         retval = 0;
6613 out:
6614         return retval;
6615 }
6616
6617 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6618                         const struct iovec *iov, loff_t offset,
6619                         unsigned long nr_segs)
6620 {
6621         struct file *file = iocb->ki_filp;
6622         struct inode *inode = file->f_mapping->host;
6623
6624         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6625                             offset, nr_segs))
6626                 return 0;
6627
6628         return __blockdev_direct_IO(rw, iocb, inode,
6629                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6630                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6631                    btrfs_submit_direct, 0);
6632 }
6633
6634 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6635
6636 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6637                 __u64 start, __u64 len)
6638 {
6639         int     ret;
6640
6641         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6642         if (ret)
6643                 return ret;
6644
6645         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6646 }
6647
6648 int btrfs_readpage(struct file *file, struct page *page)
6649 {
6650         struct extent_io_tree *tree;
6651         tree = &BTRFS_I(page->mapping->host)->io_tree;
6652         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6653 }
6654
6655 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6656 {
6657         struct extent_io_tree *tree;
6658
6659
6660         if (current->flags & PF_MEMALLOC) {
6661                 redirty_page_for_writepage(wbc, page);
6662                 unlock_page(page);
6663                 return 0;
6664         }
6665         tree = &BTRFS_I(page->mapping->host)->io_tree;
6666         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6667 }
6668
6669 int btrfs_writepages(struct address_space *mapping,
6670                      struct writeback_control *wbc)
6671 {
6672         struct extent_io_tree *tree;
6673
6674         tree = &BTRFS_I(mapping->host)->io_tree;
6675         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6676 }
6677
6678 static int
6679 btrfs_readpages(struct file *file, struct address_space *mapping,
6680                 struct list_head *pages, unsigned nr_pages)
6681 {
6682         struct extent_io_tree *tree;
6683         tree = &BTRFS_I(mapping->host)->io_tree;
6684         return extent_readpages(tree, mapping, pages, nr_pages,
6685                                 btrfs_get_extent);
6686 }
6687 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6688 {
6689         struct extent_io_tree *tree;
6690         struct extent_map_tree *map;
6691         int ret;
6692
6693         tree = &BTRFS_I(page->mapping->host)->io_tree;
6694         map = &BTRFS_I(page->mapping->host)->extent_tree;
6695         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6696         if (ret == 1) {
6697                 ClearPagePrivate(page);
6698                 set_page_private(page, 0);
6699                 page_cache_release(page);
6700         }
6701         return ret;
6702 }
6703
6704 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6705 {
6706         if (PageWriteback(page) || PageDirty(page))
6707                 return 0;
6708         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6709 }
6710
6711 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6712 {
6713         struct inode *inode = page->mapping->host;
6714         struct extent_io_tree *tree;
6715         struct btrfs_ordered_extent *ordered;
6716         struct extent_state *cached_state = NULL;
6717         u64 page_start = page_offset(page);
6718         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6719
6720         /*
6721          * we have the page locked, so new writeback can't start,
6722          * and the dirty bit won't be cleared while we are here.
6723          *
6724          * Wait for IO on this page so that we can safely clear
6725          * the PagePrivate2 bit and do ordered accounting
6726          */
6727         wait_on_page_writeback(page);
6728
6729         tree = &BTRFS_I(inode)->io_tree;
6730         if (offset) {
6731                 btrfs_releasepage(page, GFP_NOFS);
6732                 return;
6733         }
6734         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6735         ordered = btrfs_lookup_ordered_extent(inode,
6736                                            page_offset(page));
6737         if (ordered) {
6738                 /*
6739                  * IO on this page will never be started, so we need
6740                  * to account for any ordered extents now
6741                  */
6742                 clear_extent_bit(tree, page_start, page_end,
6743                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6744                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6745                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6746                 /*
6747                  * whoever cleared the private bit is responsible
6748                  * for the finish_ordered_io
6749                  */
6750                 if (TestClearPagePrivate2(page) &&
6751                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6752                                                    PAGE_CACHE_SIZE, 1)) {
6753                         btrfs_finish_ordered_io(ordered);
6754                 }
6755                 btrfs_put_ordered_extent(ordered);
6756                 cached_state = NULL;
6757                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6758         }
6759         clear_extent_bit(tree, page_start, page_end,
6760                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6761                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6762                  &cached_state, GFP_NOFS);
6763         __btrfs_releasepage(page, GFP_NOFS);
6764
6765         ClearPageChecked(page);
6766         if (PagePrivate(page)) {
6767                 ClearPagePrivate(page);
6768                 set_page_private(page, 0);
6769                 page_cache_release(page);
6770         }
6771 }
6772
6773 /*
6774  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6775  * called from a page fault handler when a page is first dirtied. Hence we must
6776  * be careful to check for EOF conditions here. We set the page up correctly
6777  * for a written page which means we get ENOSPC checking when writing into
6778  * holes and correct delalloc and unwritten extent mapping on filesystems that
6779  * support these features.
6780  *
6781  * We are not allowed to take the i_mutex here so we have to play games to
6782  * protect against truncate races as the page could now be beyond EOF.  Because
6783  * vmtruncate() writes the inode size before removing pages, once we have the
6784  * page lock we can determine safely if the page is beyond EOF. If it is not
6785  * beyond EOF, then the page is guaranteed safe against truncation until we
6786  * unlock the page.
6787  */
6788 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6789 {
6790         struct page *page = vmf->page;
6791         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6792         struct btrfs_root *root = BTRFS_I(inode)->root;
6793         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6794         struct btrfs_ordered_extent *ordered;
6795         struct extent_state *cached_state = NULL;
6796         char *kaddr;
6797         unsigned long zero_start;
6798         loff_t size;
6799         int ret;
6800         int reserved = 0;
6801         u64 page_start;
6802         u64 page_end;
6803
6804         sb_start_pagefault(inode->i_sb);
6805         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6806         if (!ret) {
6807                 ret = file_update_time(vma->vm_file);
6808                 reserved = 1;
6809         }
6810         if (ret) {
6811                 if (ret == -ENOMEM)
6812                         ret = VM_FAULT_OOM;
6813                 else /* -ENOSPC, -EIO, etc */
6814                         ret = VM_FAULT_SIGBUS;
6815                 if (reserved)
6816                         goto out;
6817                 goto out_noreserve;
6818         }
6819
6820         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6821 again:
6822         lock_page(page);
6823         size = i_size_read(inode);
6824         page_start = page_offset(page);
6825         page_end = page_start + PAGE_CACHE_SIZE - 1;
6826
6827         if ((page->mapping != inode->i_mapping) ||
6828             (page_start >= size)) {
6829                 /* page got truncated out from underneath us */
6830                 goto out_unlock;
6831         }
6832         wait_on_page_writeback(page);
6833
6834         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6835         set_page_extent_mapped(page);
6836
6837         /*
6838          * we can't set the delalloc bits if there are pending ordered
6839          * extents.  Drop our locks and wait for them to finish
6840          */
6841         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6842         if (ordered) {
6843                 unlock_extent_cached(io_tree, page_start, page_end,
6844                                      &cached_state, GFP_NOFS);
6845                 unlock_page(page);
6846                 btrfs_start_ordered_extent(inode, ordered, 1);
6847                 btrfs_put_ordered_extent(ordered);
6848                 goto again;
6849         }
6850
6851         /*
6852          * XXX - page_mkwrite gets called every time the page is dirtied, even
6853          * if it was already dirty, so for space accounting reasons we need to
6854          * clear any delalloc bits for the range we are fixing to save.  There
6855          * is probably a better way to do this, but for now keep consistent with
6856          * prepare_pages in the normal write path.
6857          */
6858         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6859                           EXTENT_DIRTY | EXTENT_DELALLOC |
6860                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6861                           0, 0, &cached_state, GFP_NOFS);
6862
6863         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6864                                         &cached_state);
6865         if (ret) {
6866                 unlock_extent_cached(io_tree, page_start, page_end,
6867                                      &cached_state, GFP_NOFS);
6868                 ret = VM_FAULT_SIGBUS;
6869                 goto out_unlock;
6870         }
6871         ret = 0;
6872
6873         /* page is wholly or partially inside EOF */
6874         if (page_start + PAGE_CACHE_SIZE > size)
6875                 zero_start = size & ~PAGE_CACHE_MASK;
6876         else
6877                 zero_start = PAGE_CACHE_SIZE;
6878
6879         if (zero_start != PAGE_CACHE_SIZE) {
6880                 kaddr = kmap(page);
6881                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6882                 flush_dcache_page(page);
6883                 kunmap(page);
6884         }
6885         ClearPageChecked(page);
6886         set_page_dirty(page);
6887         SetPageUptodate(page);
6888
6889         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6890         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6891         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6892
6893         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6894
6895 out_unlock:
6896         if (!ret) {
6897                 sb_end_pagefault(inode->i_sb);
6898                 return VM_FAULT_LOCKED;
6899         }
6900         unlock_page(page);
6901 out:
6902         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6903 out_noreserve:
6904         sb_end_pagefault(inode->i_sb);
6905         return ret;
6906 }
6907
6908 static int btrfs_truncate(struct inode *inode)
6909 {
6910         struct btrfs_root *root = BTRFS_I(inode)->root;
6911         struct btrfs_block_rsv *rsv;
6912         int ret;
6913         int err = 0;
6914         struct btrfs_trans_handle *trans;
6915         u64 mask = root->sectorsize - 1;
6916         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6917
6918         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6919         if (ret)
6920                 return ret;
6921
6922         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6923         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6924
6925         /*
6926          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6927          * 3 things going on here
6928          *
6929          * 1) We need to reserve space for our orphan item and the space to
6930          * delete our orphan item.  Lord knows we don't want to have a dangling
6931          * orphan item because we didn't reserve space to remove it.
6932          *
6933          * 2) We need to reserve space to update our inode.
6934          *
6935          * 3) We need to have something to cache all the space that is going to
6936          * be free'd up by the truncate operation, but also have some slack
6937          * space reserved in case it uses space during the truncate (thank you
6938          * very much snapshotting).
6939          *
6940          * And we need these to all be seperate.  The fact is we can use alot of
6941          * space doing the truncate, and we have no earthly idea how much space
6942          * we will use, so we need the truncate reservation to be seperate so it
6943          * doesn't end up using space reserved for updating the inode or
6944          * removing the orphan item.  We also need to be able to stop the
6945          * transaction and start a new one, which means we need to be able to
6946          * update the inode several times, and we have no idea of knowing how
6947          * many times that will be, so we can't just reserve 1 item for the
6948          * entirety of the opration, so that has to be done seperately as well.
6949          * Then there is the orphan item, which does indeed need to be held on
6950          * to for the whole operation, and we need nobody to touch this reserved
6951          * space except the orphan code.
6952          *
6953          * So that leaves us with
6954          *
6955          * 1) root->orphan_block_rsv - for the orphan deletion.
6956          * 2) rsv - for the truncate reservation, which we will steal from the
6957          * transaction reservation.
6958          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6959          * updating the inode.
6960          */
6961         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6962         if (!rsv)
6963                 return -ENOMEM;
6964         rsv->size = min_size;
6965         rsv->failfast = 1;
6966
6967         /*
6968          * 1 for the truncate slack space
6969          * 1 for updating the inode.
6970          */
6971         trans = btrfs_start_transaction(root, 2);
6972         if (IS_ERR(trans)) {
6973                 err = PTR_ERR(trans);
6974                 goto out;
6975         }
6976
6977         /* Migrate the slack space for the truncate to our reserve */
6978         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6979                                       min_size);
6980         BUG_ON(ret);
6981
6982         /*
6983          * setattr is responsible for setting the ordered_data_close flag,
6984          * but that is only tested during the last file release.  That
6985          * could happen well after the next commit, leaving a great big
6986          * window where new writes may get lost if someone chooses to write
6987          * to this file after truncating to zero
6988          *
6989          * The inode doesn't have any dirty data here, and so if we commit
6990          * this is a noop.  If someone immediately starts writing to the inode
6991          * it is very likely we'll catch some of their writes in this
6992          * transaction, and the commit will find this file on the ordered
6993          * data list with good things to send down.
6994          *
6995          * This is a best effort solution, there is still a window where
6996          * using truncate to replace the contents of the file will
6997          * end up with a zero length file after a crash.
6998          */
6999         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7000                                            &BTRFS_I(inode)->runtime_flags))
7001                 btrfs_add_ordered_operation(trans, root, inode);
7002
7003         /*
7004          * So if we truncate and then write and fsync we normally would just
7005          * write the extents that changed, which is a problem if we need to
7006          * first truncate that entire inode.  So set this flag so we write out
7007          * all of the extents in the inode to the sync log so we're completely
7008          * safe.
7009          */
7010         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7011         trans->block_rsv = rsv;
7012
7013         while (1) {
7014                 ret = btrfs_truncate_inode_items(trans, root, inode,
7015                                                  inode->i_size,
7016                                                  BTRFS_EXTENT_DATA_KEY);
7017                 if (ret != -ENOSPC) {
7018                         err = ret;
7019                         break;
7020                 }
7021
7022                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7023                 ret = btrfs_update_inode(trans, root, inode);
7024                 if (ret) {
7025                         err = ret;
7026                         break;
7027                 }
7028
7029                 btrfs_end_transaction(trans, root);
7030                 btrfs_btree_balance_dirty(root);
7031
7032                 trans = btrfs_start_transaction(root, 2);
7033                 if (IS_ERR(trans)) {
7034                         ret = err = PTR_ERR(trans);
7035                         trans = NULL;
7036                         break;
7037                 }
7038
7039                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7040                                               rsv, min_size);
7041                 BUG_ON(ret);    /* shouldn't happen */
7042                 trans->block_rsv = rsv;
7043         }
7044
7045         if (ret == 0 && inode->i_nlink > 0) {
7046                 trans->block_rsv = root->orphan_block_rsv;
7047                 ret = btrfs_orphan_del(trans, inode);
7048                 if (ret)
7049                         err = ret;
7050         }
7051
7052         if (trans) {
7053                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7054                 ret = btrfs_update_inode(trans, root, inode);
7055                 if (ret && !err)
7056                         err = ret;
7057
7058                 ret = btrfs_end_transaction(trans, root);
7059                 btrfs_btree_balance_dirty(root);
7060         }
7061
7062 out:
7063         btrfs_free_block_rsv(root, rsv);
7064
7065         if (ret && !err)
7066                 err = ret;
7067
7068         return err;
7069 }
7070
7071 /*
7072  * create a new subvolume directory/inode (helper for the ioctl).
7073  */
7074 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7075                              struct btrfs_root *new_root, u64 new_dirid)
7076 {
7077         struct inode *inode;
7078         int err;
7079         u64 index = 0;
7080
7081         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7082                                 new_dirid, new_dirid,
7083                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7084                                 &index);
7085         if (IS_ERR(inode))
7086                 return PTR_ERR(inode);
7087         inode->i_op = &btrfs_dir_inode_operations;
7088         inode->i_fop = &btrfs_dir_file_operations;
7089
7090         set_nlink(inode, 1);
7091         btrfs_i_size_write(inode, 0);
7092
7093         err = btrfs_update_inode(trans, new_root, inode);
7094
7095         iput(inode);
7096         return err;
7097 }
7098
7099 struct inode *btrfs_alloc_inode(struct super_block *sb)
7100 {
7101         struct btrfs_inode *ei;
7102         struct inode *inode;
7103
7104         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7105         if (!ei)
7106                 return NULL;
7107
7108         ei->root = NULL;
7109         ei->generation = 0;
7110         ei->last_trans = 0;
7111         ei->last_sub_trans = 0;
7112         ei->logged_trans = 0;
7113         ei->delalloc_bytes = 0;
7114         ei->disk_i_size = 0;
7115         ei->flags = 0;
7116         ei->csum_bytes = 0;
7117         ei->index_cnt = (u64)-1;
7118         ei->last_unlink_trans = 0;
7119         ei->last_log_commit = 0;
7120
7121         spin_lock_init(&ei->lock);
7122         ei->outstanding_extents = 0;
7123         ei->reserved_extents = 0;
7124
7125         ei->runtime_flags = 0;
7126         ei->force_compress = BTRFS_COMPRESS_NONE;
7127
7128         ei->delayed_node = NULL;
7129
7130         inode = &ei->vfs_inode;
7131         extent_map_tree_init(&ei->extent_tree);
7132         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7133         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7134         ei->io_tree.track_uptodate = 1;
7135         ei->io_failure_tree.track_uptodate = 1;
7136         atomic_set(&ei->sync_writers, 0);
7137         mutex_init(&ei->log_mutex);
7138         mutex_init(&ei->delalloc_mutex);
7139         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7140         INIT_LIST_HEAD(&ei->delalloc_inodes);
7141         INIT_LIST_HEAD(&ei->ordered_operations);
7142         RB_CLEAR_NODE(&ei->rb_node);
7143
7144         return inode;
7145 }
7146
7147 static void btrfs_i_callback(struct rcu_head *head)
7148 {
7149         struct inode *inode = container_of(head, struct inode, i_rcu);
7150         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7151 }
7152
7153 void btrfs_destroy_inode(struct inode *inode)
7154 {
7155         struct btrfs_ordered_extent *ordered;
7156         struct btrfs_root *root = BTRFS_I(inode)->root;
7157
7158         WARN_ON(!hlist_empty(&inode->i_dentry));
7159         WARN_ON(inode->i_data.nrpages);
7160         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7161         WARN_ON(BTRFS_I(inode)->reserved_extents);
7162         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7163         WARN_ON(BTRFS_I(inode)->csum_bytes);
7164
7165         /*
7166          * This can happen where we create an inode, but somebody else also
7167          * created the same inode and we need to destroy the one we already
7168          * created.
7169          */
7170         if (!root)
7171                 goto free;
7172
7173         /*
7174          * Make sure we're properly removed from the ordered operation
7175          * lists.
7176          */
7177         smp_mb();
7178         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7179                 spin_lock(&root->fs_info->ordered_extent_lock);
7180                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7181                 spin_unlock(&root->fs_info->ordered_extent_lock);
7182         }
7183
7184         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7185                      &BTRFS_I(inode)->runtime_flags)) {
7186                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7187                        (unsigned long long)btrfs_ino(inode));
7188                 atomic_dec(&root->orphan_inodes);
7189         }
7190
7191         while (1) {
7192                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7193                 if (!ordered)
7194                         break;
7195                 else {
7196                         printk(KERN_ERR "btrfs found ordered "
7197                                "extent %llu %llu on inode cleanup\n",
7198                                (unsigned long long)ordered->file_offset,
7199                                (unsigned long long)ordered->len);
7200                         btrfs_remove_ordered_extent(inode, ordered);
7201                         btrfs_put_ordered_extent(ordered);
7202                         btrfs_put_ordered_extent(ordered);
7203                 }
7204         }
7205         inode_tree_del(inode);
7206         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7207 free:
7208         btrfs_remove_delayed_node(inode);
7209         call_rcu(&inode->i_rcu, btrfs_i_callback);
7210 }
7211
7212 int btrfs_drop_inode(struct inode *inode)
7213 {
7214         struct btrfs_root *root = BTRFS_I(inode)->root;
7215
7216         if (btrfs_root_refs(&root->root_item) == 0 &&
7217             !btrfs_is_free_space_inode(inode))
7218                 return 1;
7219         else
7220                 return generic_drop_inode(inode);
7221 }
7222
7223 static void init_once(void *foo)
7224 {
7225         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7226
7227         inode_init_once(&ei->vfs_inode);
7228 }
7229
7230 void btrfs_destroy_cachep(void)
7231 {
7232         /*
7233          * Make sure all delayed rcu free inodes are flushed before we
7234          * destroy cache.
7235          */
7236         rcu_barrier();
7237         if (btrfs_inode_cachep)
7238                 kmem_cache_destroy(btrfs_inode_cachep);
7239         if (btrfs_trans_handle_cachep)
7240                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7241         if (btrfs_transaction_cachep)
7242                 kmem_cache_destroy(btrfs_transaction_cachep);
7243         if (btrfs_path_cachep)
7244                 kmem_cache_destroy(btrfs_path_cachep);
7245         if (btrfs_free_space_cachep)
7246                 kmem_cache_destroy(btrfs_free_space_cachep);
7247         if (btrfs_delalloc_work_cachep)
7248                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7249 }
7250
7251 int btrfs_init_cachep(void)
7252 {
7253         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7254                         sizeof(struct btrfs_inode), 0,
7255                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7256         if (!btrfs_inode_cachep)
7257                 goto fail;
7258
7259         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7260                         sizeof(struct btrfs_trans_handle), 0,
7261                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7262         if (!btrfs_trans_handle_cachep)
7263                 goto fail;
7264
7265         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7266                         sizeof(struct btrfs_transaction), 0,
7267                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7268         if (!btrfs_transaction_cachep)
7269                 goto fail;
7270
7271         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7272                         sizeof(struct btrfs_path), 0,
7273                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7274         if (!btrfs_path_cachep)
7275                 goto fail;
7276
7277         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7278                         sizeof(struct btrfs_free_space), 0,
7279                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7280         if (!btrfs_free_space_cachep)
7281                 goto fail;
7282
7283         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7284                         sizeof(struct btrfs_delalloc_work), 0,
7285                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7286                         NULL);
7287         if (!btrfs_delalloc_work_cachep)
7288                 goto fail;
7289
7290         return 0;
7291 fail:
7292         btrfs_destroy_cachep();
7293         return -ENOMEM;
7294 }
7295
7296 static int btrfs_getattr(struct vfsmount *mnt,
7297                          struct dentry *dentry, struct kstat *stat)
7298 {
7299         struct inode *inode = dentry->d_inode;
7300         u32 blocksize = inode->i_sb->s_blocksize;
7301
7302         generic_fillattr(inode, stat);
7303         stat->dev = BTRFS_I(inode)->root->anon_dev;
7304         stat->blksize = PAGE_CACHE_SIZE;
7305         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7306                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7307         return 0;
7308 }
7309
7310 /*
7311  * If a file is moved, it will inherit the cow and compression flags of the new
7312  * directory.
7313  */
7314 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7315 {
7316         struct btrfs_inode *b_dir = BTRFS_I(dir);
7317         struct btrfs_inode *b_inode = BTRFS_I(inode);
7318
7319         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7320                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7321         else
7322                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7323
7324         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7325                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7326                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7327         } else {
7328                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7329                                     BTRFS_INODE_NOCOMPRESS);
7330         }
7331 }
7332
7333 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7334                            struct inode *new_dir, struct dentry *new_dentry)
7335 {
7336         struct btrfs_trans_handle *trans;
7337         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7338         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7339         struct inode *new_inode = new_dentry->d_inode;
7340         struct inode *old_inode = old_dentry->d_inode;
7341         struct timespec ctime = CURRENT_TIME;
7342         u64 index = 0;
7343         u64 root_objectid;
7344         int ret;
7345         u64 old_ino = btrfs_ino(old_inode);
7346
7347         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7348                 return -EPERM;
7349
7350         /* we only allow rename subvolume link between subvolumes */
7351         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7352                 return -EXDEV;
7353
7354         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7355             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7356                 return -ENOTEMPTY;
7357
7358         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7359             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7360                 return -ENOTEMPTY;
7361
7362
7363         /* check for collisions, even if the  name isn't there */
7364         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7365                              new_dentry->d_name.name,
7366                              new_dentry->d_name.len);
7367
7368         if (ret) {
7369                 if (ret == -EEXIST) {
7370                         /* we shouldn't get
7371                          * eexist without a new_inode */
7372                         if (!new_inode) {
7373                                 WARN_ON(1);
7374                                 return ret;
7375                         }
7376                 } else {
7377                         /* maybe -EOVERFLOW */
7378                         return ret;
7379                 }
7380         }
7381         ret = 0;
7382
7383         /*
7384          * we're using rename to replace one file with another.
7385          * and the replacement file is large.  Start IO on it now so
7386          * we don't add too much work to the end of the transaction
7387          */
7388         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7389             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7390                 filemap_flush(old_inode->i_mapping);
7391
7392         /* close the racy window with snapshot create/destroy ioctl */
7393         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7394                 down_read(&root->fs_info->subvol_sem);
7395         /*
7396          * We want to reserve the absolute worst case amount of items.  So if
7397          * both inodes are subvols and we need to unlink them then that would
7398          * require 4 item modifications, but if they are both normal inodes it
7399          * would require 5 item modifications, so we'll assume their normal
7400          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7401          * should cover the worst case number of items we'll modify.
7402          */
7403         trans = btrfs_start_transaction(root, 20);
7404         if (IS_ERR(trans)) {
7405                 ret = PTR_ERR(trans);
7406                 goto out_notrans;
7407         }
7408
7409         if (dest != root)
7410                 btrfs_record_root_in_trans(trans, dest);
7411
7412         ret = btrfs_set_inode_index(new_dir, &index);
7413         if (ret)
7414                 goto out_fail;
7415
7416         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7417                 /* force full log commit if subvolume involved. */
7418                 root->fs_info->last_trans_log_full_commit = trans->transid;
7419         } else {
7420                 ret = btrfs_insert_inode_ref(trans, dest,
7421                                              new_dentry->d_name.name,
7422                                              new_dentry->d_name.len,
7423                                              old_ino,
7424                                              btrfs_ino(new_dir), index);
7425                 if (ret)
7426                         goto out_fail;
7427                 /*
7428                  * this is an ugly little race, but the rename is required
7429                  * to make sure that if we crash, the inode is either at the
7430                  * old name or the new one.  pinning the log transaction lets
7431                  * us make sure we don't allow a log commit to come in after
7432                  * we unlink the name but before we add the new name back in.
7433                  */
7434                 btrfs_pin_log_trans(root);
7435         }
7436         /*
7437          * make sure the inode gets flushed if it is replacing
7438          * something.
7439          */
7440         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7441                 btrfs_add_ordered_operation(trans, root, old_inode);
7442
7443         inode_inc_iversion(old_dir);
7444         inode_inc_iversion(new_dir);
7445         inode_inc_iversion(old_inode);
7446         old_dir->i_ctime = old_dir->i_mtime = ctime;
7447         new_dir->i_ctime = new_dir->i_mtime = ctime;
7448         old_inode->i_ctime = ctime;
7449
7450         if (old_dentry->d_parent != new_dentry->d_parent)
7451                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7452
7453         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7454                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7455                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7456                                         old_dentry->d_name.name,
7457                                         old_dentry->d_name.len);
7458         } else {
7459                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7460                                         old_dentry->d_inode,
7461                                         old_dentry->d_name.name,
7462                                         old_dentry->d_name.len);
7463                 if (!ret)
7464                         ret = btrfs_update_inode(trans, root, old_inode);
7465         }
7466         if (ret) {
7467                 btrfs_abort_transaction(trans, root, ret);
7468                 goto out_fail;
7469         }
7470
7471         if (new_inode) {
7472                 inode_inc_iversion(new_inode);
7473                 new_inode->i_ctime = CURRENT_TIME;
7474                 if (unlikely(btrfs_ino(new_inode) ==
7475                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7476                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7477                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7478                                                 root_objectid,
7479                                                 new_dentry->d_name.name,
7480                                                 new_dentry->d_name.len);
7481                         BUG_ON(new_inode->i_nlink == 0);
7482                 } else {
7483                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7484                                                  new_dentry->d_inode,
7485                                                  new_dentry->d_name.name,
7486                                                  new_dentry->d_name.len);
7487                 }
7488                 if (!ret && new_inode->i_nlink == 0) {
7489                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7490                         BUG_ON(ret);
7491                 }
7492                 if (ret) {
7493                         btrfs_abort_transaction(trans, root, ret);
7494                         goto out_fail;
7495                 }
7496         }
7497
7498         fixup_inode_flags(new_dir, old_inode);
7499
7500         ret = btrfs_add_link(trans, new_dir, old_inode,
7501                              new_dentry->d_name.name,
7502                              new_dentry->d_name.len, 0, index);
7503         if (ret) {
7504                 btrfs_abort_transaction(trans, root, ret);
7505                 goto out_fail;
7506         }
7507
7508         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7509                 struct dentry *parent = new_dentry->d_parent;
7510                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7511                 btrfs_end_log_trans(root);
7512         }
7513 out_fail:
7514         btrfs_end_transaction(trans, root);
7515 out_notrans:
7516         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7517                 up_read(&root->fs_info->subvol_sem);
7518
7519         return ret;
7520 }
7521
7522 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7523 {
7524         struct btrfs_delalloc_work *delalloc_work;
7525
7526         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7527                                      work);
7528         if (delalloc_work->wait)
7529                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7530         else
7531                 filemap_flush(delalloc_work->inode->i_mapping);
7532
7533         if (delalloc_work->delay_iput)
7534                 btrfs_add_delayed_iput(delalloc_work->inode);
7535         else
7536                 iput(delalloc_work->inode);
7537         complete(&delalloc_work->completion);
7538 }
7539
7540 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7541                                                     int wait, int delay_iput)
7542 {
7543         struct btrfs_delalloc_work *work;
7544
7545         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7546         if (!work)
7547                 return NULL;
7548
7549         init_completion(&work->completion);
7550         INIT_LIST_HEAD(&work->list);
7551         work->inode = inode;
7552         work->wait = wait;
7553         work->delay_iput = delay_iput;
7554         work->work.func = btrfs_run_delalloc_work;
7555
7556         return work;
7557 }
7558
7559 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7560 {
7561         wait_for_completion(&work->completion);
7562         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7563 }
7564
7565 /*
7566  * some fairly slow code that needs optimization. This walks the list
7567  * of all the inodes with pending delalloc and forces them to disk.
7568  */
7569 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7570 {
7571         struct list_head *head = &root->fs_info->delalloc_inodes;
7572         struct btrfs_inode *binode;
7573         struct inode *inode;
7574         struct btrfs_delalloc_work *work, *next;
7575         struct list_head works;
7576         int ret = 0;
7577
7578         if (root->fs_info->sb->s_flags & MS_RDONLY)
7579                 return -EROFS;
7580
7581         INIT_LIST_HEAD(&works);
7582
7583         spin_lock(&root->fs_info->delalloc_lock);
7584         while (!list_empty(head)) {
7585                 binode = list_entry(head->next, struct btrfs_inode,
7586                                     delalloc_inodes);
7587                 inode = igrab(&binode->vfs_inode);
7588                 if (!inode)
7589                         list_del_init(&binode->delalloc_inodes);
7590                 spin_unlock(&root->fs_info->delalloc_lock);
7591                 if (inode) {
7592                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7593                         if (!work) {
7594                                 ret = -ENOMEM;
7595                                 goto out;
7596                         }
7597                         list_add_tail(&work->list, &works);
7598                         btrfs_queue_worker(&root->fs_info->flush_workers,
7599                                            &work->work);
7600                 }
7601                 cond_resched();
7602                 spin_lock(&root->fs_info->delalloc_lock);
7603         }
7604         spin_unlock(&root->fs_info->delalloc_lock);
7605
7606         /* the filemap_flush will queue IO into the worker threads, but
7607          * we have to make sure the IO is actually started and that
7608          * ordered extents get created before we return
7609          */
7610         atomic_inc(&root->fs_info->async_submit_draining);
7611         while (atomic_read(&root->fs_info->nr_async_submits) ||
7612               atomic_read(&root->fs_info->async_delalloc_pages)) {
7613                 wait_event(root->fs_info->async_submit_wait,
7614                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7615                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7616         }
7617         atomic_dec(&root->fs_info->async_submit_draining);
7618 out:
7619         list_for_each_entry_safe(work, next, &works, list) {
7620                 list_del_init(&work->list);
7621                 btrfs_wait_and_free_delalloc_work(work);
7622         }
7623         return ret;
7624 }
7625
7626 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7627                          const char *symname)
7628 {
7629         struct btrfs_trans_handle *trans;
7630         struct btrfs_root *root = BTRFS_I(dir)->root;
7631         struct btrfs_path *path;
7632         struct btrfs_key key;
7633         struct inode *inode = NULL;
7634         int err;
7635         int drop_inode = 0;
7636         u64 objectid;
7637         u64 index = 0 ;
7638         int name_len;
7639         int datasize;
7640         unsigned long ptr;
7641         struct btrfs_file_extent_item *ei;
7642         struct extent_buffer *leaf;
7643
7644         name_len = strlen(symname) + 1;
7645         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7646                 return -ENAMETOOLONG;
7647
7648         /*
7649          * 2 items for inode item and ref
7650          * 2 items for dir items
7651          * 1 item for xattr if selinux is on
7652          */
7653         trans = btrfs_start_transaction(root, 5);
7654         if (IS_ERR(trans))
7655                 return PTR_ERR(trans);
7656
7657         err = btrfs_find_free_ino(root, &objectid);
7658         if (err)
7659                 goto out_unlock;
7660
7661         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7662                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7663                                 S_IFLNK|S_IRWXUGO, &index);
7664         if (IS_ERR(inode)) {
7665                 err = PTR_ERR(inode);
7666                 goto out_unlock;
7667         }
7668
7669         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7670         if (err) {
7671                 drop_inode = 1;
7672                 goto out_unlock;
7673         }
7674
7675         /*
7676         * If the active LSM wants to access the inode during
7677         * d_instantiate it needs these. Smack checks to see
7678         * if the filesystem supports xattrs by looking at the
7679         * ops vector.
7680         */
7681         inode->i_fop = &btrfs_file_operations;
7682         inode->i_op = &btrfs_file_inode_operations;
7683
7684         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7685         if (err)
7686                 drop_inode = 1;
7687         else {
7688                 inode->i_mapping->a_ops = &btrfs_aops;
7689                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7690                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7691         }
7692         if (drop_inode)
7693                 goto out_unlock;
7694
7695         path = btrfs_alloc_path();
7696         if (!path) {
7697                 err = -ENOMEM;
7698                 drop_inode = 1;
7699                 goto out_unlock;
7700         }
7701         key.objectid = btrfs_ino(inode);
7702         key.offset = 0;
7703         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7704         datasize = btrfs_file_extent_calc_inline_size(name_len);
7705         err = btrfs_insert_empty_item(trans, root, path, &key,
7706                                       datasize);
7707         if (err) {
7708                 drop_inode = 1;
7709                 btrfs_free_path(path);
7710                 goto out_unlock;
7711         }
7712         leaf = path->nodes[0];
7713         ei = btrfs_item_ptr(leaf, path->slots[0],
7714                             struct btrfs_file_extent_item);
7715         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7716         btrfs_set_file_extent_type(leaf, ei,
7717                                    BTRFS_FILE_EXTENT_INLINE);
7718         btrfs_set_file_extent_encryption(leaf, ei, 0);
7719         btrfs_set_file_extent_compression(leaf, ei, 0);
7720         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7721         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7722
7723         ptr = btrfs_file_extent_inline_start(ei);
7724         write_extent_buffer(leaf, symname, ptr, name_len);
7725         btrfs_mark_buffer_dirty(leaf);
7726         btrfs_free_path(path);
7727
7728         inode->i_op = &btrfs_symlink_inode_operations;
7729         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7730         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7731         inode_set_bytes(inode, name_len);
7732         btrfs_i_size_write(inode, name_len - 1);
7733         err = btrfs_update_inode(trans, root, inode);
7734         if (err)
7735                 drop_inode = 1;
7736
7737 out_unlock:
7738         if (!err)
7739                 d_instantiate(dentry, inode);
7740         btrfs_end_transaction(trans, root);
7741         if (drop_inode) {
7742                 inode_dec_link_count(inode);
7743                 iput(inode);
7744         }
7745         btrfs_btree_balance_dirty(root);
7746         return err;
7747 }
7748
7749 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7750                                        u64 start, u64 num_bytes, u64 min_size,
7751                                        loff_t actual_len, u64 *alloc_hint,
7752                                        struct btrfs_trans_handle *trans)
7753 {
7754         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7755         struct extent_map *em;
7756         struct btrfs_root *root = BTRFS_I(inode)->root;
7757         struct btrfs_key ins;
7758         u64 cur_offset = start;
7759         u64 i_size;
7760         int ret = 0;
7761         bool own_trans = true;
7762
7763         if (trans)
7764                 own_trans = false;
7765         while (num_bytes > 0) {
7766                 if (own_trans) {
7767                         trans = btrfs_start_transaction(root, 3);
7768                         if (IS_ERR(trans)) {
7769                                 ret = PTR_ERR(trans);
7770                                 break;
7771                         }
7772                 }
7773
7774                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7775                                            0, *alloc_hint, &ins, 1);
7776                 if (ret) {
7777                         if (own_trans)
7778                                 btrfs_end_transaction(trans, root);
7779                         break;
7780                 }
7781
7782                 ret = insert_reserved_file_extent(trans, inode,
7783                                                   cur_offset, ins.objectid,
7784                                                   ins.offset, ins.offset,
7785                                                   ins.offset, 0, 0, 0,
7786                                                   BTRFS_FILE_EXTENT_PREALLOC);
7787                 if (ret) {
7788                         btrfs_abort_transaction(trans, root, ret);
7789                         if (own_trans)
7790                                 btrfs_end_transaction(trans, root);
7791                         break;
7792                 }
7793                 btrfs_drop_extent_cache(inode, cur_offset,
7794                                         cur_offset + ins.offset -1, 0);
7795
7796                 em = alloc_extent_map();
7797                 if (!em) {
7798                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7799                                 &BTRFS_I(inode)->runtime_flags);
7800                         goto next;
7801                 }
7802
7803                 em->start = cur_offset;
7804                 em->orig_start = cur_offset;
7805                 em->len = ins.offset;
7806                 em->block_start = ins.objectid;
7807                 em->block_len = ins.offset;
7808                 em->orig_block_len = ins.offset;
7809                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7810                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7811                 em->generation = trans->transid;
7812
7813                 while (1) {
7814                         write_lock(&em_tree->lock);
7815                         ret = add_extent_mapping(em_tree, em);
7816                         if (!ret)
7817                                 list_move(&em->list,
7818                                           &em_tree->modified_extents);
7819                         write_unlock(&em_tree->lock);
7820                         if (ret != -EEXIST)
7821                                 break;
7822                         btrfs_drop_extent_cache(inode, cur_offset,
7823                                                 cur_offset + ins.offset - 1,
7824                                                 0);
7825                 }
7826                 free_extent_map(em);
7827 next:
7828                 num_bytes -= ins.offset;
7829                 cur_offset += ins.offset;
7830                 *alloc_hint = ins.objectid + ins.offset;
7831
7832                 inode_inc_iversion(inode);
7833                 inode->i_ctime = CURRENT_TIME;
7834                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7835                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7836                     (actual_len > inode->i_size) &&
7837                     (cur_offset > inode->i_size)) {
7838                         if (cur_offset > actual_len)
7839                                 i_size = actual_len;
7840                         else
7841                                 i_size = cur_offset;
7842                         i_size_write(inode, i_size);
7843                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7844                 }
7845
7846                 ret = btrfs_update_inode(trans, root, inode);
7847
7848                 if (ret) {
7849                         btrfs_abort_transaction(trans, root, ret);
7850                         if (own_trans)
7851                                 btrfs_end_transaction(trans, root);
7852                         break;
7853                 }
7854
7855                 if (own_trans)
7856                         btrfs_end_transaction(trans, root);
7857         }
7858         return ret;
7859 }
7860
7861 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7862                               u64 start, u64 num_bytes, u64 min_size,
7863                               loff_t actual_len, u64 *alloc_hint)
7864 {
7865         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7866                                            min_size, actual_len, alloc_hint,
7867                                            NULL);
7868 }
7869
7870 int btrfs_prealloc_file_range_trans(struct inode *inode,
7871                                     struct btrfs_trans_handle *trans, int mode,
7872                                     u64 start, u64 num_bytes, u64 min_size,
7873                                     loff_t actual_len, u64 *alloc_hint)
7874 {
7875         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7876                                            min_size, actual_len, alloc_hint, trans);
7877 }
7878
7879 static int btrfs_set_page_dirty(struct page *page)
7880 {
7881         return __set_page_dirty_nobuffers(page);
7882 }
7883
7884 static int btrfs_permission(struct inode *inode, int mask)
7885 {
7886         struct btrfs_root *root = BTRFS_I(inode)->root;
7887         umode_t mode = inode->i_mode;
7888
7889         if (mask & MAY_WRITE &&
7890             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7891                 if (btrfs_root_readonly(root))
7892                         return -EROFS;
7893                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7894                         return -EACCES;
7895         }
7896         return generic_permission(inode, mask);
7897 }
7898
7899 static const struct inode_operations btrfs_dir_inode_operations = {
7900         .getattr        = btrfs_getattr,
7901         .lookup         = btrfs_lookup,
7902         .create         = btrfs_create,
7903         .unlink         = btrfs_unlink,
7904         .link           = btrfs_link,
7905         .mkdir          = btrfs_mkdir,
7906         .rmdir          = btrfs_rmdir,
7907         .rename         = btrfs_rename,
7908         .symlink        = btrfs_symlink,
7909         .setattr        = btrfs_setattr,
7910         .mknod          = btrfs_mknod,
7911         .setxattr       = btrfs_setxattr,
7912         .getxattr       = btrfs_getxattr,
7913         .listxattr      = btrfs_listxattr,
7914         .removexattr    = btrfs_removexattr,
7915         .permission     = btrfs_permission,
7916         .get_acl        = btrfs_get_acl,
7917 };
7918 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7919         .lookup         = btrfs_lookup,
7920         .permission     = btrfs_permission,
7921         .get_acl        = btrfs_get_acl,
7922 };
7923
7924 static const struct file_operations btrfs_dir_file_operations = {
7925         .llseek         = generic_file_llseek,
7926         .read           = generic_read_dir,
7927         .readdir        = btrfs_real_readdir,
7928         .unlocked_ioctl = btrfs_ioctl,
7929 #ifdef CONFIG_COMPAT
7930         .compat_ioctl   = btrfs_ioctl,
7931 #endif
7932         .release        = btrfs_release_file,
7933         .fsync          = btrfs_sync_file,
7934 };
7935
7936 static struct extent_io_ops btrfs_extent_io_ops = {
7937         .fill_delalloc = run_delalloc_range,
7938         .submit_bio_hook = btrfs_submit_bio_hook,
7939         .merge_bio_hook = btrfs_merge_bio_hook,
7940         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7941         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7942         .writepage_start_hook = btrfs_writepage_start_hook,
7943         .set_bit_hook = btrfs_set_bit_hook,
7944         .clear_bit_hook = btrfs_clear_bit_hook,
7945         .merge_extent_hook = btrfs_merge_extent_hook,
7946         .split_extent_hook = btrfs_split_extent_hook,
7947 };
7948
7949 /*
7950  * btrfs doesn't support the bmap operation because swapfiles
7951  * use bmap to make a mapping of extents in the file.  They assume
7952  * these extents won't change over the life of the file and they
7953  * use the bmap result to do IO directly to the drive.
7954  *
7955  * the btrfs bmap call would return logical addresses that aren't
7956  * suitable for IO and they also will change frequently as COW
7957  * operations happen.  So, swapfile + btrfs == corruption.
7958  *
7959  * For now we're avoiding this by dropping bmap.
7960  */
7961 static const struct address_space_operations btrfs_aops = {
7962         .readpage       = btrfs_readpage,
7963         .writepage      = btrfs_writepage,
7964         .writepages     = btrfs_writepages,
7965         .readpages      = btrfs_readpages,
7966         .direct_IO      = btrfs_direct_IO,
7967         .invalidatepage = btrfs_invalidatepage,
7968         .releasepage    = btrfs_releasepage,
7969         .set_page_dirty = btrfs_set_page_dirty,
7970         .error_remove_page = generic_error_remove_page,
7971 };
7972
7973 static const struct address_space_operations btrfs_symlink_aops = {
7974         .readpage       = btrfs_readpage,
7975         .writepage      = btrfs_writepage,
7976         .invalidatepage = btrfs_invalidatepage,
7977         .releasepage    = btrfs_releasepage,
7978 };
7979
7980 static const struct inode_operations btrfs_file_inode_operations = {
7981         .getattr        = btrfs_getattr,
7982         .setattr        = btrfs_setattr,
7983         .setxattr       = btrfs_setxattr,
7984         .getxattr       = btrfs_getxattr,
7985         .listxattr      = btrfs_listxattr,
7986         .removexattr    = btrfs_removexattr,
7987         .permission     = btrfs_permission,
7988         .fiemap         = btrfs_fiemap,
7989         .get_acl        = btrfs_get_acl,
7990         .update_time    = btrfs_update_time,
7991 };
7992 static const struct inode_operations btrfs_special_inode_operations = {
7993         .getattr        = btrfs_getattr,
7994         .setattr        = btrfs_setattr,
7995         .permission     = btrfs_permission,
7996         .setxattr       = btrfs_setxattr,
7997         .getxattr       = btrfs_getxattr,
7998         .listxattr      = btrfs_listxattr,
7999         .removexattr    = btrfs_removexattr,
8000         .get_acl        = btrfs_get_acl,
8001         .update_time    = btrfs_update_time,
8002 };
8003 static const struct inode_operations btrfs_symlink_inode_operations = {
8004         .readlink       = generic_readlink,
8005         .follow_link    = page_follow_link_light,
8006         .put_link       = page_put_link,
8007         .getattr        = btrfs_getattr,
8008         .setattr        = btrfs_setattr,
8009         .permission     = btrfs_permission,
8010         .setxattr       = btrfs_setxattr,
8011         .getxattr       = btrfs_getxattr,
8012         .listxattr      = btrfs_listxattr,
8013         .removexattr    = btrfs_removexattr,
8014         .get_acl        = btrfs_get_acl,
8015         .update_time    = btrfs_update_time,
8016 };
8017
8018 const struct dentry_operations btrfs_dentry_operations = {
8019         .d_delete       = btrfs_dentry_delete,
8020         .d_release      = btrfs_dentry_release,
8021 };