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