fs/btrfs/inode.c

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