fs/btrfs/inode.c

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