fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include <linux/btrfs.h>
  43 #include "compat.h"
  44 #include "ctree.h"
  45 #include "disk-io.h"
  46 #include "transaction.h"
  47 #include "btrfs_inode.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, struct iattr *attr);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
  99                                            u64 len, u64 orig_start,
 100                                            u64 block_start, u64 block_len,
 101                                            u64 orig_block_len, int type);
 102
 103 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 104                                      struct inode *inode,  struct inode *dir,
 105                                      const struct qstr *qstr)
 106 {
 107         int err;
 108
 109         err = btrfs_init_acl(trans, inode, dir);
 110         if (!err)
 111                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 112         return err;
 113 }
 114
 115 /*
 116  * this does all the hard work for inserting an inline extent into
 117  * the btree.  The caller should have done a btrfs_drop_extents so that
 118  * no overlapping inline items exist in the btree
 119  */
 120 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 121                                 struct btrfs_root *root, struct inode *inode,
 122                                 u64 start, size_t size, size_t compressed_size,
 123                                 int compress_type,
 124                                 struct page **compressed_pages)
 125 {
 126         struct btrfs_key key;
 127         struct btrfs_path *path;
 128         struct extent_buffer *leaf;
 129         struct page *page = NULL;
 130         char *kaddr;
 131         unsigned long ptr;
 132         struct btrfs_file_extent_item *ei;
 133         int err = 0;
 134         int ret;
 135         size_t cur_size = size;
 136         size_t datasize;
 137         unsigned long offset;
 138
 139         if (compressed_size && compressed_pages)
 140                 cur_size = compressed_size;
 141
 142         path = btrfs_alloc_path();
 143         if (!path)
 144                 return -ENOMEM;
 145
 146         path->leave_spinning = 1;
 147
 148         key.objectid = btrfs_ino(inode);
 149         key.offset = start;
 150         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 151         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 152
 153         inode_add_bytes(inode, size);
 154         ret = btrfs_insert_empty_item(trans, root, path, &key,
 155                                       datasize);
 156         if (ret) {
 157                 err = ret;
 158                 goto fail;
 159         }
 160         leaf = path->nodes[0];
 161         ei = btrfs_item_ptr(leaf, path->slots[0],
 162                             struct btrfs_file_extent_item);
 163         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 164         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 165         btrfs_set_file_extent_encryption(leaf, ei, 0);
 166         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 167         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 168         ptr = btrfs_file_extent_inline_start(ei);
 169
 170         if (compress_type != BTRFS_COMPRESS_NONE) {
 171                 struct page *cpage;
 172                 int i = 0;
 173                 while (compressed_size > 0) {
 174                         cpage = compressed_pages[i];
 175                         cur_size = min_t(unsigned long, compressed_size,
 176                                        PAGE_CACHE_SIZE);
 177
 178                         kaddr = kmap_atomic(cpage);
 179                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 180                         kunmap_atomic(kaddr);
 181
 182                         i++;
 183                         ptr += cur_size;
 184                         compressed_size -= cur_size;
 185                 }
 186                 btrfs_set_file_extent_compression(leaf, ei,
 187                                                   compress_type);
 188         } else {
 189                 page = find_get_page(inode->i_mapping,
 190                                      start >> PAGE_CACHE_SHIFT);
 191                 btrfs_set_file_extent_compression(leaf, ei, 0);
 192                 kaddr = kmap_atomic(page);
 193                 offset = start & (PAGE_CACHE_SIZE - 1);
 194                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 195                 kunmap_atomic(kaddr);
 196                 page_cache_release(page);
 197         }
 198         btrfs_mark_buffer_dirty(leaf);
 199         btrfs_free_path(path);
 200
 201         /*
 202          * we're an inline extent, so nobody can
 203          * extend the file past i_size without locking
 204          * a page we already have locked.
 205          *
 206          * We must do any isize and inode updates
 207          * before we unlock the pages.  Otherwise we
 208          * could end up racing with unlink.
 209          */
 210         BTRFS_I(inode)->disk_i_size = inode->i_size;
 211         ret = btrfs_update_inode(trans, root, inode);
 212
 213         return ret;
 214 fail:
 215         btrfs_free_path(path);
 216         return err;
 217 }
 218
 219
 220 /*
 221  * conditionally insert an inline extent into the file.  This
 222  * does the checks required to make sure the data is small enough
 223  * to fit as an inline extent.
 224  */
 225 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 226                                  struct btrfs_root *root,
 227                                  struct inode *inode, u64 start, u64 end,
 228                                  size_t compressed_size, int compress_type,
 229                                  struct page **compressed_pages)
 230 {
 231         u64 isize = i_size_read(inode);
 232         u64 actual_end = min(end + 1, isize);
 233         u64 inline_len = actual_end - start;
 234         u64 aligned_end = (end + root->sectorsize - 1) &
 235                         ~((u64)root->sectorsize - 1);
 236         u64 data_len = inline_len;
 237         int ret;
 238
 239         if (compressed_size)
 240                 data_len = compressed_size;
 241
 242         if (start > 0 ||
 243             actual_end >= PAGE_CACHE_SIZE ||
 244             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 245             (!compressed_size &&
 246             (actual_end & (root->sectorsize - 1)) == 0) ||
 247             end + 1 < isize ||
 248             data_len > root->fs_info->max_inline) {
 249                 return 1;
 250         }
 251
 252         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 253         if (ret)
 254                 return ret;
 255
 256         if (isize > actual_end)
 257                 inline_len = min_t(u64, isize, actual_end);
 258         ret = insert_inline_extent(trans, root, inode, start,
 259                                    inline_len, compressed_size,
 260                                    compress_type, compressed_pages);
 261         if (ret && ret != -ENOSPC) {
 262                 btrfs_abort_transaction(trans, root, ret);
 263                 return ret;
 264         } else if (ret == -ENOSPC) {
 265                 return 1;
 266         }
 267
 268         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 269         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 270         return 0;
 271 }
 272
 273 struct async_extent {
 274         u64 start;
 275         u64 ram_size;
 276         u64 compressed_size;
 277         struct page **pages;
 278         unsigned long nr_pages;
 279         int compress_type;
 280         struct list_head list;
 281 };
 282
 283 struct async_cow {
 284         struct inode *inode;
 285         struct btrfs_root *root;
 286         struct page *locked_page;
 287         u64 start;
 288         u64 end;
 289         struct list_head extents;
 290         struct btrfs_work work;
 291 };
 292
 293 static noinline int add_async_extent(struct async_cow *cow,
 294                                      u64 start, u64 ram_size,
 295                                      u64 compressed_size,
 296                                      struct page **pages,
 297                                      unsigned long nr_pages,
 298                                      int compress_type)
 299 {
 300         struct async_extent *async_extent;
 301
 302         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 303         BUG_ON(!async_extent); /* -ENOMEM */
 304         async_extent->start = start;
 305         async_extent->ram_size = ram_size;
 306         async_extent->compressed_size = compressed_size;
 307         async_extent->pages = pages;
 308         async_extent->nr_pages = nr_pages;
 309         async_extent->compress_type = compress_type;
 310         list_add_tail(&async_extent->list, &cow->extents);
 311         return 0;
 312 }
 313
 314 /*
 315  * we create compressed extents in two phases.  The first
 316  * phase compresses a range of pages that have already been
 317  * locked (both pages and state bits are locked).
 318  *
 319  * This is done inside an ordered work queue, and the compression
 320  * is spread across many cpus.  The actual IO submission is step
 321  * two, and the ordered work queue takes care of making sure that
 322  * happens in the same order things were put onto the queue by
 323  * writepages and friends.
 324  *
 325  * If this code finds it can't get good compression, it puts an
 326  * entry onto the work queue to write the uncompressed bytes.  This
 327  * makes sure that both compressed inodes and uncompressed inodes
 328  * are written in the same order that the flusher thread sent them
 329  * down.
 330  */
 331 static noinline int compress_file_range(struct inode *inode,
 332                                         struct page *locked_page,
 333                                         u64 start, u64 end,
 334                                         struct async_cow *async_cow,
 335                                         int *num_added)
 336 {
 337         struct btrfs_root *root = BTRFS_I(inode)->root;
 338         struct btrfs_trans_handle *trans;
 339         u64 num_bytes;
 340         u64 blocksize = root->sectorsize;
 341         u64 actual_end;
 342         u64 isize = i_size_read(inode);
 343         int ret = 0;
 344         struct page **pages = NULL;
 345         unsigned long nr_pages;
 346         unsigned long nr_pages_ret = 0;
 347         unsigned long total_compressed = 0;
 348         unsigned long total_in = 0;
 349         unsigned long max_compressed = 128 * 1024;
 350         unsigned long max_uncompressed = 128 * 1024;
 351         int i;
 352         int will_compress;
 353         int compress_type = root->fs_info->compress_type;
 354
 355         /* if this is a small write inside eof, kick off a defrag */
 356         if ((end - start + 1) < 16 * 1024 &&
 357             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 358                 btrfs_add_inode_defrag(NULL, inode);
 359
 360         actual_end = min_t(u64, isize, end + 1);
 361 again:
 362         will_compress = 0;
 363         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 364         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 365
 366         /*
 367          * we don't want to send crud past the end of i_size through
 368          * compression, that's just a waste of CPU time.  So, if the
 369          * end of the file is before the start of our current
 370          * requested range of bytes, we bail out to the uncompressed
 371          * cleanup code that can deal with all of this.
 372          *
 373          * It isn't really the fastest way to fix things, but this is a
 374          * very uncommon corner.
 375          */
 376         if (actual_end <= start)
 377                 goto cleanup_and_bail_uncompressed;
 378
 379         total_compressed = actual_end - start;
 380
 381         /* we want to make sure that amount of ram required to uncompress
 382          * an extent is reasonable, so we limit the total size in ram
 383          * of a compressed extent to 128k.  This is a crucial number
 384          * because it also controls how easily we can spread reads across
 385          * cpus for decompression.
 386          *
 387          * We also want to make sure the amount of IO required to do
 388          * a random read is reasonably small, so we limit the size of
 389          * a compressed extent to 128k.
 390          */
 391         total_compressed = min(total_compressed, max_uncompressed);
 392         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 393         num_bytes = max(blocksize,  num_bytes);
 394         total_in = 0;
 395         ret = 0;
 396
 397         /*
 398          * we do compression for mount -o compress and when the
 399          * inode has not been flagged as nocompress.  This flag can
 400          * change at any time if we discover bad compression ratios.
 401          */
 402         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 403             (btrfs_test_opt(root, COMPRESS) ||
 404              (BTRFS_I(inode)->force_compress) ||
 405              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 406                 WARN_ON(pages);
 407                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 408                 if (!pages) {
 409                         /* just bail out to the uncompressed code */
 410                         goto cont;
 411                 }
 412
 413                 if (BTRFS_I(inode)->force_compress)
 414                         compress_type = BTRFS_I(inode)->force_compress;
 415
 416                 ret = btrfs_compress_pages(compress_type,
 417                                            inode->i_mapping, start,
 418                                            total_compressed, pages,
 419                                            nr_pages, &nr_pages_ret,
 420                                            &total_in,
 421                                            &total_compressed,
 422                                            max_compressed);
 423
 424                 if (!ret) {
 425                         unsigned long offset = total_compressed &
 426                                 (PAGE_CACHE_SIZE - 1);
 427                         struct page *page = pages[nr_pages_ret - 1];
 428                         char *kaddr;
 429
 430                         /* zero the tail end of the last page, we might be
 431                          * sending it down to disk
 432                          */
 433                         if (offset) {
 434                                 kaddr = kmap_atomic(page);
 435                                 memset(kaddr + offset, 0,
 436                                        PAGE_CACHE_SIZE - offset);
 437                                 kunmap_atomic(kaddr);
 438                         }
 439                         will_compress = 1;
 440                 }
 441         }
 442 cont:
 443         if (start == 0) {
 444                 trans = btrfs_join_transaction(root);
 445                 if (IS_ERR(trans)) {
 446                         ret = PTR_ERR(trans);
 447                         trans = NULL;
 448                         goto cleanup_and_out;
 449                 }
 450                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 451
 452                 /* lets try to make an inline extent */
 453                 if (ret || total_in < (actual_end - start)) {
 454                         /* we didn't compress the entire range, try
 455                          * to make an uncompressed inline extent.
 456                          */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end, 0, 0, NULL);
 459                 } else {
 460                         /* try making a compressed inline extent */
 461                         ret = cow_file_range_inline(trans, root, inode,
 462                                                     start, end,
 463                                                     total_compressed,
 464                                                     compress_type, pages);
 465                 }
 466                 if (ret <= 0) {
 467                         /*
 468                          * inline extent creation worked or returned error,
 469                          * we don't need to create any more async work items.
 470                          * Unlock and free up our temp pages.
 471                          */
 472                         extent_clear_unlock_delalloc(inode,
 473                              &BTRFS_I(inode)->io_tree,
 474                              start, end, NULL,
 475                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 476                              EXTENT_CLEAR_DELALLOC |
 477                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 478
 479                         btrfs_end_transaction(trans, root);
 480                         goto free_pages_out;
 481                 }
 482                 btrfs_end_transaction(trans, root);
 483         }
 484
 485         if (will_compress) {
 486                 /*
 487                  * we aren't doing an inline extent round the compressed size
 488                  * up to a block size boundary so the allocator does sane
 489                  * things
 490                  */
 491                 total_compressed = (total_compressed + blocksize - 1) &
 492                         ~(blocksize - 1);
 493
 494                 /*
 495                  * one last check to make sure the compression is really a
 496                  * win, compare the page count read with the blocks on disk
 497                  */
 498                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 499                         ~(PAGE_CACHE_SIZE - 1);
 500                 if (total_compressed >= total_in) {
 501                         will_compress = 0;
 502                 } else {
 503                         num_bytes = total_in;
 504                 }
 505         }
 506         if (!will_compress && pages) {
 507                 /*
 508                  * the compression code ran but failed to make things smaller,
 509                  * free any pages it allocated and our page pointer array
 510                  */
 511                 for (i = 0; i < nr_pages_ret; i++) {
 512                         WARN_ON(pages[i]->mapping);
 513                         page_cache_release(pages[i]);
 514                 }
 515                 kfree(pages);
 516                 pages = NULL;
 517                 total_compressed = 0;
 518                 nr_pages_ret = 0;
 519
 520                 /* flag the file so we don't compress in the future */
 521                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 522                     !(BTRFS_I(inode)->force_compress)) {
 523                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 524                 }
 525         }
 526         if (will_compress) {
 527                 *num_added += 1;
 528
 529                 /* the async work queues will take care of doing actual
 530                  * allocation on disk for these compressed pages,
 531                  * and will submit them to the elevator.
 532                  */
 533                 add_async_extent(async_cow, start, num_bytes,
 534                                  total_compressed, pages, nr_pages_ret,
 535                                  compress_type);
 536
 537                 if (start + num_bytes < end) {
 538                         start += num_bytes;
 539                         pages = NULL;
 540                         cond_resched();
 541                         goto again;
 542                 }
 543         } else {
 544 cleanup_and_bail_uncompressed:
 545                 /*
 546                  * No compression, but we still need to write the pages in
 547                  * the file we've been given so far.  redirty the locked
 548                  * page if it corresponds to our extent and set things up
 549                  * for the async work queue to run cow_file_range to do
 550                  * the normal delalloc dance
 551                  */
 552                 if (page_offset(locked_page) >= start &&
 553                     page_offset(locked_page) <= end) {
 554                         __set_page_dirty_nobuffers(locked_page);
 555                         /* unlocked later on in the async handlers */
 556                 }
 557                 add_async_extent(async_cow, start, end - start + 1,
 558                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 559                 *num_added += 1;
 560         }
 561
 562 out:
 563         return ret;
 564
 565 free_pages_out:
 566         for (i = 0; i < nr_pages_ret; i++) {
 567                 WARN_ON(pages[i]->mapping);
 568                 page_cache_release(pages[i]);
 569         }
 570         kfree(pages);
 571
 572         goto out;
 573
 574 cleanup_and_out:
 575         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 576                                      start, end, NULL,
 577                                      EXTENT_CLEAR_UNLOCK_PAGE |
 578                                      EXTENT_CLEAR_DIRTY |
 579                                      EXTENT_CLEAR_DELALLOC |
 580                                      EXTENT_SET_WRITEBACK |
 581                                      EXTENT_END_WRITEBACK);
 582         if (!trans || IS_ERR(trans))
 583                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 584         else
 585                 btrfs_abort_transaction(trans, root, ret);
 586         goto free_pages_out;
 587 }
 588
 589 /*
 590  * phase two of compressed writeback.  This is the ordered portion
 591  * of the code, which only gets called in the order the work was
 592  * queued.  We walk all the async extents created by compress_file_range
 593  * and send them down to the disk.
 594  */
 595 static noinline int submit_compressed_extents(struct inode *inode,
 596                                               struct async_cow *async_cow)
 597 {
 598         struct async_extent *async_extent;
 599         u64 alloc_hint = 0;
 600         struct btrfs_trans_handle *trans;
 601         struct btrfs_key ins;
 602         struct extent_map *em;
 603         struct btrfs_root *root = BTRFS_I(inode)->root;
 604         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 605         struct extent_io_tree *io_tree;
 606         int ret = 0;
 607
 608         if (list_empty(&async_cow->extents))
 609                 return 0;
 610
 611
 612         while (!list_empty(&async_cow->extents)) {
 613                 async_extent = list_entry(async_cow->extents.next,
 614                                           struct async_extent, list);
 615                 list_del(&async_extent->list);
 616
 617                 io_tree = &BTRFS_I(inode)->io_tree;
 618
 619 retry:
 620                 /* did the compression code fall back to uncompressed IO? */
 621                 if (!async_extent->pages) {
 622                         int page_started = 0;
 623                         unsigned long nr_written = 0;
 624
 625                         lock_extent(io_tree, async_extent->start,
 626                                          async_extent->start +
 627                                          async_extent->ram_size - 1);
 628
 629                         /* allocate blocks */
 630                         ret = cow_file_range(inode, async_cow->locked_page,
 631                                              async_extent->start,
 632                                              async_extent->start +
 633                                              async_extent->ram_size - 1,
 634                                              &page_started, &nr_written, 0);
 635
 636                         /* JDM XXX */
 637
 638                         /*
 639                          * if page_started, cow_file_range inserted an
 640                          * inline extent and took care of all the unlocking
 641                          * and IO for us.  Otherwise, we need to submit
 642                          * all those pages down to the drive.
 643                          */
 644                         if (!page_started && !ret)
 645                                 extent_write_locked_range(io_tree,
 646                                                   inode, async_extent->start,
 647                                                   async_extent->start +
 648                                                   async_extent->ram_size - 1,
 649                                                   btrfs_get_extent,
 650                                                   WB_SYNC_ALL);
 651                         kfree(async_extent);
 652                         cond_resched();
 653                         continue;
 654                 }
 655
 656                 lock_extent(io_tree, async_extent->start,
 657                             async_extent->start + async_extent->ram_size - 1);
 658
 659                 trans = btrfs_join_transaction(root);
 660                 if (IS_ERR(trans)) {
 661                         ret = PTR_ERR(trans);
 662                 } else {
 663                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 664                         ret = btrfs_reserve_extent(trans, root,
 665                                            async_extent->compressed_size,
 666                                            async_extent->compressed_size,
 667                                            0, alloc_hint, &ins, 1);
 668                         if (ret && ret != -ENOSPC)
 669                                 btrfs_abort_transaction(trans, root, ret);
 670                         btrfs_end_transaction(trans, root);
 671                 }
 672
 673                 if (ret) {
 674                         int i;
 675                         for (i = 0; i < async_extent->nr_pages; i++) {
 676                                 WARN_ON(async_extent->pages[i]->mapping);
 677                                 page_cache_release(async_extent->pages[i]);
 678                         }
 679                         kfree(async_extent->pages);
 680                         async_extent->nr_pages = 0;
 681                         async_extent->pages = NULL;
 682                         unlock_extent(io_tree, async_extent->start,
 683                                       async_extent->start +
 684                                       async_extent->ram_size - 1);
 685                         if (ret == -ENOSPC)
 686                                 goto retry;
 687                         goto out_free; /* JDM: Requeue? */
 688                 }
 689
 690                 /*
 691                  * here we're doing allocation and writeback of the
 692                  * compressed pages
 693                  */
 694                 btrfs_drop_extent_cache(inode, async_extent->start,
 695                                         async_extent->start +
 696                                         async_extent->ram_size - 1, 0);
 697
 698                 em = alloc_extent_map();
 699                 BUG_ON(!em); /* -ENOMEM */
 700                 em->start = async_extent->start;
 701                 em->len = async_extent->ram_size;
 702                 em->orig_start = em->start;
 703                 em->mod_start = em->start;
 704                 em->mod_len = em->len;
 705
 706                 em->block_start = ins.objectid;
 707                 em->block_len = ins.offset;
 708                 em->orig_block_len = ins.offset;
 709                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 710                 em->compress_type = async_extent->compress_type;
 711                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 712                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 713                 em->generation = -1;
 714
 715                 while (1) {
 716                         write_lock(&em_tree->lock);
 717                         ret = add_extent_mapping(em_tree, em);
 718                         if (!ret)
 719                                 list_move(&em->list,
 720                                           &em_tree->modified_extents);
 721                         write_unlock(&em_tree->lock);
 722                         if (ret != -EEXIST) {
 723                                 free_extent_map(em);
 724                                 break;
 725                         }
 726                         btrfs_drop_extent_cache(inode, async_extent->start,
 727                                                 async_extent->start +
 728                                                 async_extent->ram_size - 1, 0);
 729                 }
 730
 731                 ret = btrfs_add_ordered_extent_compress(inode,
 732                                                 async_extent->start,
 733                                                 ins.objectid,
 734                                                 async_extent->ram_size,
 735                                                 ins.offset,
 736                                                 BTRFS_ORDERED_COMPRESSED,
 737                                                 async_extent->compress_type);
 738                 BUG_ON(ret); /* -ENOMEM */
 739
 740                 /*
 741                  * clear dirty, set writeback and unlock the pages.
 742                  */
 743                 extent_clear_unlock_delalloc(inode,
 744                                 &BTRFS_I(inode)->io_tree,
 745                                 async_extent->start,
 746                                 async_extent->start +
 747                                 async_extent->ram_size - 1,
 748                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 749                                 EXTENT_CLEAR_UNLOCK |
 750                                 EXTENT_CLEAR_DELALLOC |
 751                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 752
 753                 ret = btrfs_submit_compressed_write(inode,
 754                                     async_extent->start,
 755                                     async_extent->ram_size,
 756                                     ins.objectid,
 757                                     ins.offset, async_extent->pages,
 758                                     async_extent->nr_pages);
 759
 760                 BUG_ON(ret); /* -ENOMEM */
 761                 alloc_hint = ins.objectid + ins.offset;
 762                 kfree(async_extent);
 763                 cond_resched();
 764         }
 765         ret = 0;
 766 out:
 767         return ret;
 768 out_free:
 769         kfree(async_extent);
 770         goto out;
 771 }
 772
 773 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 774                                       u64 num_bytes)
 775 {
 776         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 777         struct extent_map *em;
 778         u64 alloc_hint = 0;
 779
 780         read_lock(&em_tree->lock);
 781         em = search_extent_mapping(em_tree, start, num_bytes);
 782         if (em) {
 783                 /*
 784                  * if block start isn't an actual block number then find the
 785                  * first block in this inode and use that as a hint.  If that
 786                  * block is also bogus then just don't worry about it.
 787                  */
 788                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 789                         free_extent_map(em);
 790                         em = search_extent_mapping(em_tree, 0, 0);
 791                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 792                                 alloc_hint = em->block_start;
 793                         if (em)
 794                                 free_extent_map(em);
 795                 } else {
 796                         alloc_hint = em->block_start;
 797                         free_extent_map(em);
 798                 }
 799         }
 800         read_unlock(&em_tree->lock);
 801
 802         return alloc_hint;
 803 }
 804
 805 /*
 806  * when extent_io.c finds a delayed allocation range in the file,
 807  * the call backs end up in this code.  The basic idea is to
 808  * allocate extents on disk for the range, and create ordered data structs
 809  * in ram to track those extents.
 810  *
 811  * locked_page is the page that writepage had locked already.  We use
 812  * it to make sure we don't do extra locks or unlocks.
 813  *
 814  * *page_started is set to one if we unlock locked_page and do everything
 815  * required to start IO on it.  It may be clean and already done with
 816  * IO when we return.
 817  */
 818 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 819                                      struct inode *inode,
 820                                      struct btrfs_root *root,
 821                                      struct page *locked_page,
 822                                      u64 start, u64 end, int *page_started,
 823                                      unsigned long *nr_written,
 824                                      int unlock)
 825 {
 826         u64 alloc_hint = 0;
 827         u64 num_bytes;
 828         unsigned long ram_size;
 829         u64 disk_num_bytes;
 830         u64 cur_alloc_size;
 831         u64 blocksize = root->sectorsize;
 832         struct btrfs_key ins;
 833         struct extent_map *em;
 834         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 835         int ret = 0;
 836
 837         BUG_ON(btrfs_is_free_space_inode(inode));
 838
 839         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 840         num_bytes = max(blocksize,  num_bytes);
 841         disk_num_bytes = num_bytes;
 842
 843         /* if this is a small write inside eof, kick off defrag */
 844         if (num_bytes < 64 * 1024 &&
 845             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 846                 btrfs_add_inode_defrag(trans, inode);
 847
 848         if (start == 0) {
 849                 /* lets try to make an inline extent */
 850                 ret = cow_file_range_inline(trans, root, inode,
 851                                             start, end, 0, 0, NULL);
 852                 if (ret == 0) {
 853                         extent_clear_unlock_delalloc(inode,
 854                                      &BTRFS_I(inode)->io_tree,
 855                                      start, end, NULL,
 856                                      EXTENT_CLEAR_UNLOCK_PAGE |
 857                                      EXTENT_CLEAR_UNLOCK |
 858                                      EXTENT_CLEAR_DELALLOC |
 859                                      EXTENT_CLEAR_DIRTY |
 860                                      EXTENT_SET_WRITEBACK |
 861                                      EXTENT_END_WRITEBACK);
 862
 863                         *nr_written = *nr_written +
 864                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 865                         *page_started = 1;
 866                         goto out;
 867                 } else if (ret < 0) {
 868                         btrfs_abort_transaction(trans, root, ret);
 869                         goto out_unlock;
 870                 }
 871         }
 872
 873         BUG_ON(disk_num_bytes >
 874                btrfs_super_total_bytes(root->fs_info->super_copy));
 875
 876         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 877         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 878
 879         while (disk_num_bytes > 0) {
 880                 unsigned long op;
 881
 882                 cur_alloc_size = disk_num_bytes;
 883                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 884                                            root->sectorsize, 0, alloc_hint,
 885                                            &ins, 1);
 886                 if (ret < 0) {
 887                         btrfs_abort_transaction(trans, root, ret);
 888                         goto out_unlock;
 889                 }
 890
 891                 em = alloc_extent_map();
 892                 BUG_ON(!em); /* -ENOMEM */
 893                 em->start = start;
 894                 em->orig_start = em->start;
 895                 ram_size = ins.offset;
 896                 em->len = ins.offset;
 897                 em->mod_start = em->start;
 898                 em->mod_len = em->len;
 899
 900                 em->block_start = ins.objectid;
 901                 em->block_len = ins.offset;
 902                 em->orig_block_len = ins.offset;
 903                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 904                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 905                 em->generation = -1;
 906
 907                 while (1) {
 908                         write_lock(&em_tree->lock);
 909                         ret = add_extent_mapping(em_tree, em);
 910                         if (!ret)
 911                                 list_move(&em->list,
 912                                           &em_tree->modified_extents);
 913                         write_unlock(&em_tree->lock);
 914                         if (ret != -EEXIST) {
 915                                 free_extent_map(em);
 916                                 break;
 917                         }
 918                         btrfs_drop_extent_cache(inode, start,
 919                                                 start + ram_size - 1, 0);
 920                 }
 921
 922                 cur_alloc_size = ins.offset;
 923                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 924                                                ram_size, cur_alloc_size, 0);
 925                 BUG_ON(ret); /* -ENOMEM */
 926
 927                 if (root->root_key.objectid ==
 928                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 929                         ret = btrfs_reloc_clone_csums(inode, start,
 930                                                       cur_alloc_size);
 931                         if (ret) {
 932                                 btrfs_abort_transaction(trans, root, ret);
 933                                 goto out_unlock;
 934                         }
 935                 }
 936
 937                 if (disk_num_bytes < cur_alloc_size)
 938                         break;
 939
 940                 /* we're not doing compressed IO, don't unlock the first
 941                  * page (which the caller expects to stay locked), don't
 942                  * clear any dirty bits and don't set any writeback bits
 943                  *
 944                  * Do set the Private2 bit so we know this page was properly
 945                  * setup for writepage
 946                  */
 947                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 948                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 949                         EXTENT_SET_PRIVATE2;
 950
 951                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 952                                              start, start + ram_size - 1,
 953                                              locked_page, op);
 954                 disk_num_bytes -= cur_alloc_size;
 955                 num_bytes -= cur_alloc_size;
 956                 alloc_hint = ins.objectid + ins.offset;
 957                 start += cur_alloc_size;
 958         }
 959 out:
 960         return ret;
 961
 962 out_unlock:
 963         extent_clear_unlock_delalloc(inode,
 964                      &BTRFS_I(inode)->io_tree,
 965                      start, end, locked_page,
 966                      EXTENT_CLEAR_UNLOCK_PAGE |
 967                      EXTENT_CLEAR_UNLOCK |
 968                      EXTENT_CLEAR_DELALLOC |
 969                      EXTENT_CLEAR_DIRTY |
 970                      EXTENT_SET_WRITEBACK |
 971                      EXTENT_END_WRITEBACK);
 972
 973         goto out;
 974 }
 975
 976 static noinline int cow_file_range(struct inode *inode,
 977                                    struct page *locked_page,
 978                                    u64 start, u64 end, int *page_started,
 979                                    unsigned long *nr_written,
 980                                    int unlock)
 981 {
 982         struct btrfs_trans_handle *trans;
 983         struct btrfs_root *root = BTRFS_I(inode)->root;
 984         int ret;
 985
 986         trans = btrfs_join_transaction(root);
 987         if (IS_ERR(trans)) {
 988                 extent_clear_unlock_delalloc(inode,
 989                              &BTRFS_I(inode)->io_tree,
 990                              start, end, locked_page,
 991                              EXTENT_CLEAR_UNLOCK_PAGE |
 992                              EXTENT_CLEAR_UNLOCK |
 993                              EXTENT_CLEAR_DELALLOC |
 994                              EXTENT_CLEAR_DIRTY |
 995                              EXTENT_SET_WRITEBACK |
 996                              EXTENT_END_WRITEBACK);
 997                 return PTR_ERR(trans);
 998         }
 999         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1000
1001         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
1002                                page_started, nr_written, unlock);
1003
1004         btrfs_end_transaction(trans, root);
1005
1006         return ret;
1007 }
1008
1009 /*
1010  * work queue call back to started compression on a file and pages
1011  */
1012 static noinline void async_cow_start(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         int num_added = 0;
1016         async_cow = container_of(work, struct async_cow, work);
1017
1018         compress_file_range(async_cow->inode, async_cow->locked_page,
1019                             async_cow->start, async_cow->end, async_cow,
1020                             &num_added);
1021         if (num_added == 0) {
1022                 btrfs_add_delayed_iput(async_cow->inode);
1023                 async_cow->inode = NULL;
1024         }
1025 }
1026
1027 /*
1028  * work queue call back to submit previously compressed pages
1029  */
1030 static noinline void async_cow_submit(struct btrfs_work *work)
1031 {
1032         struct async_cow *async_cow;
1033         struct btrfs_root *root;
1034         unsigned long nr_pages;
1035
1036         async_cow = container_of(work, struct async_cow, work);
1037
1038         root = async_cow->root;
1039         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1040                 PAGE_CACHE_SHIFT;
1041
1042         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1043             5 * 1024 * 1024 &&
1044             waitqueue_active(&root->fs_info->async_submit_wait))
1045                 wake_up(&root->fs_info->async_submit_wait);
1046
1047         if (async_cow->inode)
1048                 submit_compressed_extents(async_cow->inode, async_cow);
1049 }
1050
1051 static noinline void async_cow_free(struct btrfs_work *work)
1052 {
1053         struct async_cow *async_cow;
1054         async_cow = container_of(work, struct async_cow, work);
1055         if (async_cow->inode)
1056                 btrfs_add_delayed_iput(async_cow->inode);
1057         kfree(async_cow);
1058 }
1059
1060 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1061                                 u64 start, u64 end, int *page_started,
1062                                 unsigned long *nr_written)
1063 {
1064         struct async_cow *async_cow;
1065         struct btrfs_root *root = BTRFS_I(inode)->root;
1066         unsigned long nr_pages;
1067         u64 cur_end;
1068         int limit = 10 * 1024 * 1024;
1069
1070         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1071                          1, 0, NULL, GFP_NOFS);
1072         while (start < end) {
1073                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1074                 BUG_ON(!async_cow); /* -ENOMEM */
1075                 async_cow->inode = igrab(inode);
1076                 async_cow->root = root;
1077                 async_cow->locked_page = locked_page;
1078                 async_cow->start = start;
1079
1080                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1081                         cur_end = end;
1082                 else
1083                         cur_end = min(end, start + 512 * 1024 - 1);
1084
1085                 async_cow->end = cur_end;
1086                 INIT_LIST_HEAD(&async_cow->extents);
1087
1088                 async_cow->work.func = async_cow_start;
1089                 async_cow->work.ordered_func = async_cow_submit;
1090                 async_cow->work.ordered_free = async_cow_free;
1091                 async_cow->work.flags = 0;
1092
1093                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1094                         PAGE_CACHE_SHIFT;
1095                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1096
1097                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1098                                    &async_cow->work);
1099
1100                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1101                         wait_event(root->fs_info->async_submit_wait,
1102                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1103                             limit));
1104                 }
1105
1106                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1107                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1108                         wait_event(root->fs_info->async_submit_wait,
1109                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1110                            0));
1111                 }
1112
1113                 *nr_written += nr_pages;
1114                 start = cur_end + 1;
1115         }
1116         *page_started = 1;
1117         return 0;
1118 }
1119
1120 static noinline int csum_exist_in_range(struct btrfs_root *root,
1121                                         u64 bytenr, u64 num_bytes)
1122 {
1123         int ret;
1124         struct btrfs_ordered_sum *sums;
1125         LIST_HEAD(list);
1126
1127         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1128                                        bytenr + num_bytes - 1, &list, 0);
1129         if (ret == 0 && list_empty(&list))
1130                 return 0;
1131
1132         while (!list_empty(&list)) {
1133                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1134                 list_del(&sums->list);
1135                 kfree(sums);
1136         }
1137         return 1;
1138 }
1139
1140 /*
1141  * when nowcow writeback call back.  This checks for snapshots or COW copies
1142  * of the extents that exist in the file, and COWs the file as required.
1143  *
1144  * If no cow copies or snapshots exist, we write directly to the existing
1145  * blocks on disk
1146  */
1147 static noinline int run_delalloc_nocow(struct inode *inode,
1148                                        struct page *locked_page,
1149                               u64 start, u64 end, int *page_started, int force,
1150                               unsigned long *nr_written)
1151 {
1152         struct btrfs_root *root = BTRFS_I(inode)->root;
1153         struct btrfs_trans_handle *trans;
1154         struct extent_buffer *leaf;
1155         struct btrfs_path *path;
1156         struct btrfs_file_extent_item *fi;
1157         struct btrfs_key found_key;
1158         u64 cow_start;
1159         u64 cur_offset;
1160         u64 extent_end;
1161         u64 extent_offset;
1162         u64 disk_bytenr;
1163         u64 num_bytes;
1164         u64 disk_num_bytes;
1165         int extent_type;
1166         int ret, err;
1167         int type;
1168         int nocow;
1169         int check_prev = 1;
1170         bool nolock;
1171         u64 ino = btrfs_ino(inode);
1172
1173         path = btrfs_alloc_path();
1174         if (!path) {
1175                 extent_clear_unlock_delalloc(inode,
1176                              &BTRFS_I(inode)->io_tree,
1177                              start, end, locked_page,
1178                              EXTENT_CLEAR_UNLOCK_PAGE |
1179                              EXTENT_CLEAR_UNLOCK |
1180                              EXTENT_CLEAR_DELALLOC |
1181                              EXTENT_CLEAR_DIRTY |
1182                              EXTENT_SET_WRITEBACK |
1183                              EXTENT_END_WRITEBACK);
1184                 return -ENOMEM;
1185         }
1186
1187         nolock = btrfs_is_free_space_inode(inode);
1188
1189         if (nolock)
1190                 trans = btrfs_join_transaction_nolock(root);
1191         else
1192                 trans = btrfs_join_transaction(root);
1193
1194         if (IS_ERR(trans)) {
1195                 extent_clear_unlock_delalloc(inode,
1196                              &BTRFS_I(inode)->io_tree,
1197                              start, end, locked_page,
1198                              EXTENT_CLEAR_UNLOCK_PAGE |
1199                              EXTENT_CLEAR_UNLOCK |
1200                              EXTENT_CLEAR_DELALLOC |
1201                              EXTENT_CLEAR_DIRTY |
1202                              EXTENT_SET_WRITEBACK |
1203                              EXTENT_END_WRITEBACK);
1204                 btrfs_free_path(path);
1205                 return PTR_ERR(trans);
1206         }
1207
1208         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1209
1210         cow_start = (u64)-1;
1211         cur_offset = start;
1212         while (1) {
1213                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1214                                                cur_offset, 0);
1215                 if (ret < 0) {
1216                         btrfs_abort_transaction(trans, root, ret);
1217                         goto error;
1218                 }
1219                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1220                         leaf = path->nodes[0];
1221                         btrfs_item_key_to_cpu(leaf, &found_key,
1222                                               path->slots[0] - 1);
1223                         if (found_key.objectid == ino &&
1224                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1225                                 path->slots[0]--;
1226                 }
1227                 check_prev = 0;
1228 next_slot:
1229                 leaf = path->nodes[0];
1230                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1231                         ret = btrfs_next_leaf(root, path);
1232                         if (ret < 0) {
1233                                 btrfs_abort_transaction(trans, root, ret);
1234                                 goto error;
1235                         }
1236                         if (ret > 0)
1237                                 break;
1238                         leaf = path->nodes[0];
1239                 }
1240
1241                 nocow = 0;
1242                 disk_bytenr = 0;
1243                 num_bytes = 0;
1244                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1245
1246                 if (found_key.objectid > ino ||
1247                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1248                     found_key.offset > end)
1249                         break;
1250
1251                 if (found_key.offset > cur_offset) {
1252                         extent_end = found_key.offset;
1253                         extent_type = 0;
1254                         goto out_check;
1255                 }
1256
1257                 fi = btrfs_item_ptr(leaf, path->slots[0],
1258                                     struct btrfs_file_extent_item);
1259                 extent_type = btrfs_file_extent_type(leaf, fi);
1260
1261                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1262                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1263                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1264                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1265                         extent_end = found_key.offset +
1266                                 btrfs_file_extent_num_bytes(leaf, fi);
1267                         disk_num_bytes =
1268                                 btrfs_file_extent_disk_num_bytes(leaf, fi);
1269                         if (extent_end <= start) {
1270                                 path->slots[0]++;
1271                                 goto next_slot;
1272                         }
1273                         if (disk_bytenr == 0)
1274                                 goto out_check;
1275                         if (btrfs_file_extent_compression(leaf, fi) ||
1276                             btrfs_file_extent_encryption(leaf, fi) ||
1277                             btrfs_file_extent_other_encoding(leaf, fi))
1278                                 goto out_check;
1279                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1280                                 goto out_check;
1281                         if (btrfs_extent_readonly(root, disk_bytenr))
1282                                 goto out_check;
1283                         if (btrfs_cross_ref_exist(trans, root, ino,
1284                                                   found_key.offset -
1285                                                   extent_offset, disk_bytenr))
1286                                 goto out_check;
1287                         disk_bytenr += extent_offset;
1288                         disk_bytenr += cur_offset - found_key.offset;
1289                         num_bytes = min(end + 1, extent_end) - cur_offset;
1290                         /*
1291                          * force cow if csum exists in the range.
1292                          * this ensure that csum for a given extent are
1293                          * either valid or do not exist.
1294                          */
1295                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1296                                 goto out_check;
1297                         nocow = 1;
1298                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1299                         extent_end = found_key.offset +
1300                                 btrfs_file_extent_inline_len(leaf, fi);
1301                         extent_end = ALIGN(extent_end, root->sectorsize);
1302                 } else {
1303                         BUG_ON(1);
1304                 }
1305 out_check:
1306                 if (extent_end <= start) {
1307                         path->slots[0]++;
1308                         goto next_slot;
1309                 }
1310                 if (!nocow) {
1311                         if (cow_start == (u64)-1)
1312                                 cow_start = cur_offset;
1313                         cur_offset = extent_end;
1314                         if (cur_offset > end)
1315                                 break;
1316                         path->slots[0]++;
1317                         goto next_slot;
1318                 }
1319
1320                 btrfs_release_path(path);
1321                 if (cow_start != (u64)-1) {
1322                         ret = __cow_file_range(trans, inode, root, locked_page,
1323                                                cow_start, found_key.offset - 1,
1324                                                page_started, nr_written, 1);
1325                         if (ret) {
1326                                 btrfs_abort_transaction(trans, root, ret);
1327                                 goto error;
1328                         }
1329                         cow_start = (u64)-1;
1330                 }
1331
1332                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1333                         struct extent_map *em;
1334                         struct extent_map_tree *em_tree;
1335                         em_tree = &BTRFS_I(inode)->extent_tree;
1336                         em = alloc_extent_map();
1337                         BUG_ON(!em); /* -ENOMEM */
1338                         em->start = cur_offset;
1339                         em->orig_start = found_key.offset - extent_offset;
1340                         em->len = num_bytes;
1341                         em->block_len = num_bytes;
1342                         em->block_start = disk_bytenr;
1343                         em->orig_block_len = disk_num_bytes;
1344                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1345                         em->mod_start = em->start;
1346                         em->mod_len = em->len;
1347                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1348                         set_bit(EXTENT_FLAG_FILLING, &em->flags);
1349                         em->generation = -1;
1350                         while (1) {
1351                                 write_lock(&em_tree->lock);
1352                                 ret = add_extent_mapping(em_tree, em);
1353                                 if (!ret)
1354                                         list_move(&em->list,
1355                                                   &em_tree->modified_extents);
1356                                 write_unlock(&em_tree->lock);
1357                                 if (ret != -EEXIST) {
1358                                         free_extent_map(em);
1359                                         break;
1360                                 }
1361                                 btrfs_drop_extent_cache(inode, em->start,
1362                                                 em->start + em->len - 1, 0);
1363                         }
1364                         type = BTRFS_ORDERED_PREALLOC;
1365                 } else {
1366                         type = BTRFS_ORDERED_NOCOW;
1367                 }
1368
1369                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1370                                                num_bytes, num_bytes, type);
1371                 BUG_ON(ret); /* -ENOMEM */
1372
1373                 if (root->root_key.objectid ==
1374                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1375                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1376                                                       num_bytes);
1377                         if (ret) {
1378                                 btrfs_abort_transaction(trans, root, ret);
1379                                 goto error;
1380                         }
1381                 }
1382
1383                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1384                                 cur_offset, cur_offset + num_bytes - 1,
1385                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1386                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1387                                 EXTENT_SET_PRIVATE2);
1388                 cur_offset = extent_end;
1389                 if (cur_offset > end)
1390                         break;
1391         }
1392         btrfs_release_path(path);
1393
1394         if (cur_offset <= end && cow_start == (u64)-1) {
1395                 cow_start = cur_offset;
1396                 cur_offset = end;
1397         }
1398
1399         if (cow_start != (u64)-1) {
1400                 ret = __cow_file_range(trans, inode, root, locked_page,
1401                                        cow_start, end,
1402                                        page_started, nr_written, 1);
1403                 if (ret) {
1404                         btrfs_abort_transaction(trans, root, ret);
1405                         goto error;
1406                 }
1407         }
1408
1409 error:
1410         err = btrfs_end_transaction(trans, root);
1411         if (!ret)
1412                 ret = err;
1413
1414         if (ret && cur_offset < end)
1415                 extent_clear_unlock_delalloc(inode,
1416                              &BTRFS_I(inode)->io_tree,
1417                              cur_offset, end, locked_page,
1418                              EXTENT_CLEAR_UNLOCK_PAGE |
1419                              EXTENT_CLEAR_UNLOCK |
1420                              EXTENT_CLEAR_DELALLOC |
1421                              EXTENT_CLEAR_DIRTY |
1422                              EXTENT_SET_WRITEBACK |
1423                              EXTENT_END_WRITEBACK);
1424
1425         btrfs_free_path(path);
1426         return ret;
1427 }
1428
1429 /*
1430  * extent_io.c call back to do delayed allocation processing
1431  */
1432 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1433                               u64 start, u64 end, int *page_started,
1434                               unsigned long *nr_written)
1435 {
1436         int ret;
1437         struct btrfs_root *root = BTRFS_I(inode)->root;
1438
1439         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1440                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1441                                          page_started, 1, nr_written);
1442         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1443                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1444                                          page_started, 0, nr_written);
1445         } else if (!btrfs_test_opt(root, COMPRESS) &&
1446                    !(BTRFS_I(inode)->force_compress) &&
1447                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1448                 ret = cow_file_range(inode, locked_page, start, end,
1449                                       page_started, nr_written, 1);
1450         } else {
1451                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1452                         &BTRFS_I(inode)->runtime_flags);
1453                 ret = cow_file_range_async(inode, locked_page, start, end,
1454                                            page_started, nr_written);
1455         }
1456         return ret;
1457 }
1458
1459 static void btrfs_split_extent_hook(struct inode *inode,
1460                                     struct extent_state *orig, u64 split)
1461 {
1462         /* not delalloc, ignore it */
1463         if (!(orig->state & EXTENT_DELALLOC))
1464                 return;
1465
1466         spin_lock(&BTRFS_I(inode)->lock);
1467         BTRFS_I(inode)->outstanding_extents++;
1468         spin_unlock(&BTRFS_I(inode)->lock);
1469 }
1470
1471 /*
1472  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1473  * extents so we can keep track of new extents that are just merged onto old
1474  * extents, such as when we are doing sequential writes, so we can properly
1475  * account for the metadata space we'll need.
1476  */
1477 static void btrfs_merge_extent_hook(struct inode *inode,
1478                                     struct extent_state *new,
1479                                     struct extent_state *other)
1480 {
1481         /* not delalloc, ignore it */
1482         if (!(other->state & EXTENT_DELALLOC))
1483                 return;
1484
1485         spin_lock(&BTRFS_I(inode)->lock);
1486         BTRFS_I(inode)->outstanding_extents--;
1487         spin_unlock(&BTRFS_I(inode)->lock);
1488 }
1489
1490 /*
1491  * extent_io.c set_bit_hook, used to track delayed allocation
1492  * bytes in this file, and to maintain the list of inodes that
1493  * have pending delalloc work to be done.
1494  */
1495 static void btrfs_set_bit_hook(struct inode *inode,
1496                                struct extent_state *state, int *bits)
1497 {
1498
1499         /*
1500          * set_bit and clear bit hooks normally require _irqsave/restore
1501          * but in this case, we are only testing for the DELALLOC
1502          * bit, which is only set or cleared with irqs on
1503          */
1504         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1505                 struct btrfs_root *root = BTRFS_I(inode)->root;
1506                 u64 len = state->end + 1 - state->start;
1507                 bool do_list = !btrfs_is_free_space_inode(inode);
1508
1509                 if (*bits & EXTENT_FIRST_DELALLOC) {
1510                         *bits &= ~EXTENT_FIRST_DELALLOC;
1511                 } else {
1512                         spin_lock(&BTRFS_I(inode)->lock);
1513                         BTRFS_I(inode)->outstanding_extents++;
1514                         spin_unlock(&BTRFS_I(inode)->lock);
1515                 }
1516
1517                 spin_lock(&root->fs_info->delalloc_lock);
1518                 BTRFS_I(inode)->delalloc_bytes += len;
1519                 __percpu_counter_add(&root->fs_info->delalloc_bytes, len,
1520                                      root->fs_info->delalloc_batch);
1521                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1522                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1523                                       &root->fs_info->delalloc_inodes);
1524                 }
1525                 spin_unlock(&root->fs_info->delalloc_lock);
1526         }
1527 }
1528
1529 /*
1530  * extent_io.c clear_bit_hook, see set_bit_hook for why
1531  */
1532 static void btrfs_clear_bit_hook(struct inode *inode,
1533                                  struct extent_state *state, int *bits)
1534 {
1535         /*
1536          * set_bit and clear bit hooks normally require _irqsave/restore
1537          * but in this case, we are only testing for the DELALLOC
1538          * bit, which is only set or cleared with irqs on
1539          */
1540         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1541                 struct btrfs_root *root = BTRFS_I(inode)->root;
1542                 u64 len = state->end + 1 - state->start;
1543                 bool do_list = !btrfs_is_free_space_inode(inode);
1544
1545                 if (*bits & EXTENT_FIRST_DELALLOC) {
1546                         *bits &= ~EXTENT_FIRST_DELALLOC;
1547                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1548                         spin_lock(&BTRFS_I(inode)->lock);
1549                         BTRFS_I(inode)->outstanding_extents--;
1550                         spin_unlock(&BTRFS_I(inode)->lock);
1551                 }
1552
1553                 if (*bits & EXTENT_DO_ACCOUNTING)
1554                         btrfs_delalloc_release_metadata(inode, len);
1555
1556                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1557                     && do_list)
1558                         btrfs_free_reserved_data_space(inode, len);
1559
1560                 spin_lock(&root->fs_info->delalloc_lock);
1561                 __percpu_counter_add(&root->fs_info->delalloc_bytes, -len,
1562                                      root->fs_info->delalloc_batch);
1563                 BTRFS_I(inode)->delalloc_bytes -= len;
1564
1565                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1566                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1567                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1568                 }
1569                 spin_unlock(&root->fs_info->delalloc_lock);
1570         }
1571 }
1572
1573 /*
1574  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1575  * we don't create bios that span stripes or chunks
1576  */
1577 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1578                          size_t size, struct bio *bio,
1579                          unsigned long bio_flags)
1580 {
1581         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1582         u64 logical = (u64)bio->bi_sector << 9;
1583         u64 length = 0;
1584         u64 map_length;
1585         int ret;
1586
1587         if (bio_flags & EXTENT_BIO_COMPRESSED)
1588                 return 0;
1589
1590         length = bio->bi_size;
1591         map_length = length;
1592         ret = btrfs_map_block(root->fs_info, READ, logical,
1593                               &map_length, NULL, 0);
1594         /* Will always return 0 with map_multi == NULL */
1595         BUG_ON(ret < 0);
1596         if (map_length < length + size)
1597                 return 1;
1598         return 0;
1599 }
1600
1601 /*
1602  * in order to insert checksums into the metadata in large chunks,
1603  * we wait until bio submission time.   All the pages in the bio are
1604  * checksummed and sums are attached onto the ordered extent record.
1605  *
1606  * At IO completion time the cums attached on the ordered extent record
1607  * are inserted into the btree
1608  */
1609 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1610                                     struct bio *bio, int mirror_num,
1611                                     unsigned long bio_flags,
1612                                     u64 bio_offset)
1613 {
1614         struct btrfs_root *root = BTRFS_I(inode)->root;
1615         int ret = 0;
1616
1617         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1618         BUG_ON(ret); /* -ENOMEM */
1619         return 0;
1620 }
1621
1622 /*
1623  * in order to insert checksums into the metadata in large chunks,
1624  * we wait until bio submission time.   All the pages in the bio are
1625  * checksummed and sums are attached onto the ordered extent record.
1626  *
1627  * At IO completion time the cums attached on the ordered extent record
1628  * are inserted into the btree
1629  */
1630 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1631                           int mirror_num, unsigned long bio_flags,
1632                           u64 bio_offset)
1633 {
1634         struct btrfs_root *root = BTRFS_I(inode)->root;
1635         int ret;
1636
1637         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1638         if (ret)
1639                 bio_endio(bio, ret);
1640         return ret;
1641 }
1642
1643 /*
1644  * extent_io.c submission hook. This does the right thing for csum calculation
1645  * on write, or reading the csums from the tree before a read
1646  */
1647 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1648                           int mirror_num, unsigned long bio_flags,
1649                           u64 bio_offset)
1650 {
1651         struct btrfs_root *root = BTRFS_I(inode)->root;
1652         int ret = 0;
1653         int skip_sum;
1654         int metadata = 0;
1655         int async = !atomic_read(&BTRFS_I(inode)->sync_writers);
1656
1657         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1658
1659         if (btrfs_is_free_space_inode(inode))
1660                 metadata = 2;
1661
1662         if (!(rw & REQ_WRITE)) {
1663                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1664                 if (ret)
1665                         goto out;
1666
1667                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1668                         ret = btrfs_submit_compressed_read(inode, bio,
1669                                                            mirror_num,
1670                                                            bio_flags);
1671                         goto out;
1672                 } else if (!skip_sum) {
1673                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1674                         if (ret)
1675                                 goto out;
1676                 }
1677                 goto mapit;
1678         } else if (async && !skip_sum) {
1679                 /* csum items have already been cloned */
1680                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1681                         goto mapit;
1682                 /* we're doing a write, do the async checksumming */
1683                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1684                                    inode, rw, bio, mirror_num,
1685                                    bio_flags, bio_offset,
1686                                    __btrfs_submit_bio_start,
1687                                    __btrfs_submit_bio_done);
1688                 goto out;
1689         } else if (!skip_sum) {
1690                 ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1691                 if (ret)
1692                         goto out;
1693         }
1694
1695 mapit:
1696         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1697
1698 out:
1699         if (ret < 0)
1700                 bio_endio(bio, ret);
1701         return ret;
1702 }
1703
1704 /*
1705  * given a list of ordered sums record them in the inode.  This happens
1706  * at IO completion time based on sums calculated at bio submission time.
1707  */
1708 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1709                              struct inode *inode, u64 file_offset,
1710                              struct list_head *list)
1711 {
1712         struct btrfs_ordered_sum *sum;
1713
1714         list_for_each_entry(sum, list, list) {
1715                 btrfs_csum_file_blocks(trans,
1716                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1717         }
1718         return 0;
1719 }
1720
1721 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1722                               struct extent_state **cached_state)
1723 {
1724         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1725         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1726                                    cached_state, GFP_NOFS);
1727 }
1728
1729 /* see btrfs_writepage_start_hook for details on why this is required */
1730 struct btrfs_writepage_fixup {
1731         struct page *page;
1732         struct btrfs_work work;
1733 };
1734
1735 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1736 {
1737         struct btrfs_writepage_fixup *fixup;
1738         struct btrfs_ordered_extent *ordered;
1739         struct extent_state *cached_state = NULL;
1740         struct page *page;
1741         struct inode *inode;
1742         u64 page_start;
1743         u64 page_end;
1744         int ret;
1745
1746         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1747         page = fixup->page;
1748 again:
1749         lock_page(page);
1750         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1751                 ClearPageChecked(page);
1752                 goto out_page;
1753         }
1754
1755         inode = page->mapping->host;
1756         page_start = page_offset(page);
1757         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1758
1759         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1760                          &cached_state);
1761
1762         /* already ordered? We're done */
1763         if (PagePrivate2(page))
1764                 goto out;
1765
1766         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1767         if (ordered) {
1768                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1769                                      page_end, &cached_state, GFP_NOFS);
1770                 unlock_page(page);
1771                 btrfs_start_ordered_extent(inode, ordered, 1);
1772                 btrfs_put_ordered_extent(ordered);
1773                 goto again;
1774         }
1775
1776         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1777         if (ret) {
1778                 mapping_set_error(page->mapping, ret);
1779                 end_extent_writepage(page, ret, page_start, page_end);
1780                 ClearPageChecked(page);
1781                 goto out;
1782          }
1783
1784         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1785         ClearPageChecked(page);
1786         set_page_dirty(page);
1787 out:
1788         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1789                              &cached_state, GFP_NOFS);
1790 out_page:
1791         unlock_page(page);
1792         page_cache_release(page);
1793         kfree(fixup);
1794 }
1795
1796 /*
1797  * There are a few paths in the higher layers of the kernel that directly
1798  * set the page dirty bit without asking the filesystem if it is a
1799  * good idea.  This causes problems because we want to make sure COW
1800  * properly happens and the data=ordered rules are followed.
1801  *
1802  * In our case any range that doesn't have the ORDERED bit set
1803  * hasn't been properly setup for IO.  We kick off an async process
1804  * to fix it up.  The async helper will wait for ordered extents, set
1805  * the delalloc bit and make it safe to write the page.
1806  */
1807 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1808 {
1809         struct inode *inode = page->mapping->host;
1810         struct btrfs_writepage_fixup *fixup;
1811         struct btrfs_root *root = BTRFS_I(inode)->root;
1812
1813         /* this page is properly in the ordered list */
1814         if (TestClearPagePrivate2(page))
1815                 return 0;
1816
1817         if (PageChecked(page))
1818                 return -EAGAIN;
1819
1820         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1821         if (!fixup)
1822                 return -EAGAIN;
1823
1824         SetPageChecked(page);
1825         page_cache_get(page);
1826         fixup->work.func = btrfs_writepage_fixup_worker;
1827         fixup->page = page;
1828         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1829         return -EBUSY;
1830 }
1831
1832 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1833                                        struct inode *inode, u64 file_pos,
1834                                        u64 disk_bytenr, u64 disk_num_bytes,
1835                                        u64 num_bytes, u64 ram_bytes,
1836                                        u8 compression, u8 encryption,
1837                                        u16 other_encoding, int extent_type)
1838 {
1839         struct btrfs_root *root = BTRFS_I(inode)->root;
1840         struct btrfs_file_extent_item *fi;
1841         struct btrfs_path *path;
1842         struct extent_buffer *leaf;
1843         struct btrfs_key ins;
1844         int ret;
1845
1846         path = btrfs_alloc_path();
1847         if (!path)
1848                 return -ENOMEM;
1849
1850         path->leave_spinning = 1;
1851
1852         /*
1853          * we may be replacing one extent in the tree with another.
1854          * The new extent is pinned in the extent map, and we don't want
1855          * to drop it from the cache until it is completely in the btree.
1856          *
1857          * So, tell btrfs_drop_extents to leave this extent in the cache.
1858          * the caller is expected to unpin it and allow it to be merged
1859          * with the others.
1860          */
1861         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1862                                  file_pos + num_bytes, 0);
1863         if (ret)
1864                 goto out;
1865
1866         ins.objectid = btrfs_ino(inode);
1867         ins.offset = file_pos;
1868         ins.type = BTRFS_EXTENT_DATA_KEY;
1869         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1870         if (ret)
1871                 goto out;
1872         leaf = path->nodes[0];
1873         fi = btrfs_item_ptr(leaf, path->slots[0],
1874                             struct btrfs_file_extent_item);
1875         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1876         btrfs_set_file_extent_type(leaf, fi, extent_type);
1877         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1878         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1879         btrfs_set_file_extent_offset(leaf, fi, 0);
1880         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1881         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1882         btrfs_set_file_extent_compression(leaf, fi, compression);
1883         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1884         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1885
1886         btrfs_mark_buffer_dirty(leaf);
1887         btrfs_release_path(path);
1888
1889         inode_add_bytes(inode, num_bytes);
1890
1891         ins.objectid = disk_bytenr;
1892         ins.offset = disk_num_bytes;
1893         ins.type = BTRFS_EXTENT_ITEM_KEY;
1894         ret = btrfs_alloc_reserved_file_extent(trans, root,
1895                                         root->root_key.objectid,
1896                                         btrfs_ino(inode), file_pos, &ins);
1897 out:
1898         btrfs_free_path(path);
1899
1900         return ret;
1901 }
1902
1903 /*
1904  * helper function for btrfs_finish_ordered_io, this
1905  * just reads in some of the csum leaves to prime them into ram
1906  * before we start the transaction.  It limits the amount of btree
1907  * reads required while inside the transaction.
1908  */
1909 /* as ordered data IO finishes, this gets called so we can finish
1910  * an ordered extent if the range of bytes in the file it covers are
1911  * fully written.
1912  */
1913 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1914 {
1915         struct inode *inode = ordered_extent->inode;
1916         struct btrfs_root *root = BTRFS_I(inode)->root;
1917         struct btrfs_trans_handle *trans = NULL;
1918         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1919         struct extent_state *cached_state = NULL;
1920         int compress_type = 0;
1921         int ret;
1922         bool nolock;
1923
1924         nolock = btrfs_is_free_space_inode(inode);
1925
1926         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1927                 ret = -EIO;
1928                 goto out;
1929         }
1930
1931         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1932                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1933                 btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1934                 if (nolock)
1935                         trans = btrfs_join_transaction_nolock(root);
1936                 else
1937                         trans = btrfs_join_transaction(root);
1938                 if (IS_ERR(trans)) {
1939                         ret = PTR_ERR(trans);
1940                         trans = NULL;
1941                         goto out;
1942                 }
1943                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1944                 ret = btrfs_update_inode_fallback(trans, root, inode);
1945                 if (ret) /* -ENOMEM or corruption */
1946                         btrfs_abort_transaction(trans, root, ret);
1947                 goto out;
1948         }
1949
1950         lock_extent_bits(io_tree, ordered_extent->file_offset,
1951                          ordered_extent->file_offset + ordered_extent->len - 1,
1952                          0, &cached_state);
1953
1954         if (nolock)
1955                 trans = btrfs_join_transaction_nolock(root);
1956         else
1957                 trans = btrfs_join_transaction(root);
1958         if (IS_ERR(trans)) {
1959                 ret = PTR_ERR(trans);
1960                 trans = NULL;
1961                 goto out_unlock;
1962         }
1963         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1964
1965         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1966                 compress_type = ordered_extent->compress_type;
1967         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1968                 BUG_ON(compress_type);
1969                 ret = btrfs_mark_extent_written(trans, inode,
1970                                                 ordered_extent->file_offset,
1971                                                 ordered_extent->file_offset +
1972                                                 ordered_extent->len);
1973         } else {
1974                 BUG_ON(root == root->fs_info->tree_root);
1975                 ret = insert_reserved_file_extent(trans, inode,
1976                                                 ordered_extent->file_offset,
1977                                                 ordered_extent->start,
1978                                                 ordered_extent->disk_len,
1979                                                 ordered_extent->len,
1980                                                 ordered_extent->len,
1981                                                 compress_type, 0, 0,
1982                                                 BTRFS_FILE_EXTENT_REG);
1983         }
1984         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1985                            ordered_extent->file_offset, ordered_extent->len,
1986                            trans->transid);
1987         if (ret < 0) {
1988                 btrfs_abort_transaction(trans, root, ret);
1989                 goto out_unlock;
1990         }
1991
1992         add_pending_csums(trans, inode, ordered_extent->file_offset,
1993                           &ordered_extent->list);
1994
1995         btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1996         ret = btrfs_update_inode_fallback(trans, root, inode);
1997         if (ret) { /* -ENOMEM or corruption */
1998                 btrfs_abort_transaction(trans, root, ret);
1999                 goto out_unlock;
2000         }
2001         ret = 0;
2002 out_unlock:
2003         unlock_extent_cached(io_tree, ordered_extent->file_offset,
2004                              ordered_extent->file_offset +
2005                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
2006 out:
2007         if (root != root->fs_info->tree_root)
2008                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
2009         if (trans)
2010                 btrfs_end_transaction(trans, root);
2011
2012         if (ret)
2013                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
2014                                       ordered_extent->file_offset +
2015                                       ordered_extent->len - 1, NULL, GFP_NOFS);
2016
2017         /*
2018          * This needs to be done to make sure anybody waiting knows we are done
2019          * updating everything for this ordered extent.
2020          */
2021         btrfs_remove_ordered_extent(inode, ordered_extent);
2022
2023         /* once for us */
2024         btrfs_put_ordered_extent(ordered_extent);
2025         /* once for the tree */
2026         btrfs_put_ordered_extent(ordered_extent);
2027
2028         return ret;
2029 }
2030
2031 static void finish_ordered_fn(struct btrfs_work *work)
2032 {
2033         struct btrfs_ordered_extent *ordered_extent;
2034         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2035         btrfs_finish_ordered_io(ordered_extent);
2036 }
2037
2038 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2039                                 struct extent_state *state, int uptodate)
2040 {
2041         struct inode *inode = page->mapping->host;
2042         struct btrfs_root *root = BTRFS_I(inode)->root;
2043         struct btrfs_ordered_extent *ordered_extent = NULL;
2044         struct btrfs_workers *workers;
2045
2046         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2047
2048         ClearPagePrivate2(page);
2049         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2050                                             end - start + 1, uptodate))
2051                 return 0;
2052
2053         ordered_extent->work.func = finish_ordered_fn;
2054         ordered_extent->work.flags = 0;
2055
2056         if (btrfs_is_free_space_inode(inode))
2057                 workers = &root->fs_info->endio_freespace_worker;
2058         else
2059                 workers = &root->fs_info->endio_write_workers;
2060         btrfs_queue_worker(workers, &ordered_extent->work);
2061
2062         return 0;
2063 }
2064
2065 /*
2066  * when reads are done, we need to check csums to verify the data is correct
2067  * if there's a match, we allow the bio to finish.  If not, the code in
2068  * extent_io.c will try to find good copies for us.
2069  */
2070 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2071                                struct extent_state *state, int mirror)
2072 {
2073         size_t offset = start - page_offset(page);
2074         struct inode *inode = page->mapping->host;
2075         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2076         char *kaddr;
2077         u64 private = ~(u32)0;
2078         int ret;
2079         struct btrfs_root *root = BTRFS_I(inode)->root;
2080         u32 csum = ~(u32)0;
2081
2082         if (PageChecked(page)) {
2083                 ClearPageChecked(page);
2084                 goto good;
2085         }
2086
2087         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2088                 goto good;
2089
2090         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2091             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2092                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2093                                   GFP_NOFS);
2094                 return 0;
2095         }
2096
2097         if (state && state->start == start) {
2098                 private = state->private;
2099                 ret = 0;
2100         } else {
2101                 ret = get_state_private(io_tree, start, &private);
2102         }
2103         kaddr = kmap_atomic(page);
2104         if (ret)
2105                 goto zeroit;
2106
2107         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2108         btrfs_csum_final(csum, (char *)&csum);
2109         if (csum != private)
2110                 goto zeroit;
2111
2112         kunmap_atomic(kaddr);
2113 good:
2114         return 0;
2115
2116 zeroit:
2117         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2118                        "private %llu\n",
2119                        (unsigned long long)btrfs_ino(page->mapping->host),
2120                        (unsigned long long)start, csum,
2121                        (unsigned long long)private);
2122         memset(kaddr + offset, 1, end - start + 1);
2123         flush_dcache_page(page);
2124         kunmap_atomic(kaddr);
2125         if (private == 0)
2126                 return 0;
2127         return -EIO;
2128 }
2129
2130 struct delayed_iput {
2131         struct list_head list;
2132         struct inode *inode;
2133 };
2134
2135 /* JDM: If this is fs-wide, why can't we add a pointer to
2136  * btrfs_inode instead and avoid the allocation? */
2137 void btrfs_add_delayed_iput(struct inode *inode)
2138 {
2139         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2140         struct delayed_iput *delayed;
2141
2142         if (atomic_add_unless(&inode->i_count, -1, 1))
2143                 return;
2144
2145         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2146         delayed->inode = inode;
2147
2148         spin_lock(&fs_info->delayed_iput_lock);
2149         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2150         spin_unlock(&fs_info->delayed_iput_lock);
2151 }
2152
2153 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2154 {
2155         LIST_HEAD(list);
2156         struct btrfs_fs_info *fs_info = root->fs_info;
2157         struct delayed_iput *delayed;
2158         int empty;
2159
2160         spin_lock(&fs_info->delayed_iput_lock);
2161         empty = list_empty(&fs_info->delayed_iputs);
2162         spin_unlock(&fs_info->delayed_iput_lock);
2163         if (empty)
2164                 return;
2165
2166         spin_lock(&fs_info->delayed_iput_lock);
2167         list_splice_init(&fs_info->delayed_iputs, &list);
2168         spin_unlock(&fs_info->delayed_iput_lock);
2169
2170         while (!list_empty(&list)) {
2171                 delayed = list_entry(list.next, struct delayed_iput, list);
2172                 list_del(&delayed->list);
2173                 iput(delayed->inode);
2174                 kfree(delayed);
2175         }
2176 }
2177
2178 enum btrfs_orphan_cleanup_state {
2179         ORPHAN_CLEANUP_STARTED  = 1,
2180         ORPHAN_CLEANUP_DONE     = 2,
2181 };
2182
2183 /*
2184  * This is called in transaction commit time. If there are no orphan
2185  * files in the subvolume, it removes orphan item and frees block_rsv
2186  * structure.
2187  */
2188 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2189                               struct btrfs_root *root)
2190 {
2191         struct btrfs_block_rsv *block_rsv;
2192         int ret;
2193
2194         if (atomic_read(&root->orphan_inodes) ||
2195             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2196                 return;
2197
2198         spin_lock(&root->orphan_lock);
2199         if (atomic_read(&root->orphan_inodes)) {
2200                 spin_unlock(&root->orphan_lock);
2201                 return;
2202         }
2203
2204         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2205                 spin_unlock(&root->orphan_lock);
2206                 return;
2207         }
2208
2209         block_rsv = root->orphan_block_rsv;
2210         root->orphan_block_rsv = NULL;
2211         spin_unlock(&root->orphan_lock);
2212
2213         if (root->orphan_item_inserted &&
2214             btrfs_root_refs(&root->root_item) > 0) {
2215                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2216                                             root->root_key.objectid);
2217                 BUG_ON(ret);
2218                 root->orphan_item_inserted = 0;
2219         }
2220
2221         if (block_rsv) {
2222                 WARN_ON(block_rsv->size > 0);
2223                 btrfs_free_block_rsv(root, block_rsv);
2224         }
2225 }
2226
2227 /*
2228  * This creates an orphan entry for the given inode in case something goes
2229  * wrong in the middle of an unlink/truncate.
2230  *
2231  * NOTE: caller of this function should reserve 5 units of metadata for
2232  *       this function.
2233  */
2234 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2235 {
2236         struct btrfs_root *root = BTRFS_I(inode)->root;
2237         struct btrfs_block_rsv *block_rsv = NULL;
2238         int reserve = 0;
2239         int insert = 0;
2240         int ret;
2241
2242         if (!root->orphan_block_rsv) {
2243                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2244                 if (!block_rsv)
2245                         return -ENOMEM;
2246         }
2247
2248         spin_lock(&root->orphan_lock);
2249         if (!root->orphan_block_rsv) {
2250                 root->orphan_block_rsv = block_rsv;
2251         } else if (block_rsv) {
2252                 btrfs_free_block_rsv(root, block_rsv);
2253                 block_rsv = NULL;
2254         }
2255
2256         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2257                               &BTRFS_I(inode)->runtime_flags)) {
2258 #if 0
2259                 /*
2260                  * For proper ENOSPC handling, we should do orphan
2261                  * cleanup when mounting. But this introduces backward
2262                  * compatibility issue.
2263                  */
2264                 if (!xchg(&root->orphan_item_inserted, 1))
2265                         insert = 2;
2266                 else
2267                         insert = 1;
2268 #endif
2269                 insert = 1;
2270                 atomic_inc(&root->orphan_inodes);
2271         }
2272
2273         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2274                               &BTRFS_I(inode)->runtime_flags))
2275                 reserve = 1;
2276         spin_unlock(&root->orphan_lock);
2277
2278         /* grab metadata reservation from transaction handle */
2279         if (reserve) {
2280                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2281                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2282         }
2283
2284         /* insert an orphan item to track this unlinked/truncated file */
2285         if (insert >= 1) {
2286                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2287                 if (ret && ret != -EEXIST) {
2288                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2289                                   &BTRFS_I(inode)->runtime_flags);
2290                         btrfs_abort_transaction(trans, root, ret);
2291                         return ret;
2292                 }
2293                 ret = 0;
2294         }
2295
2296         /* insert an orphan item to track subvolume contains orphan files */
2297         if (insert >= 2) {
2298                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2299                                                root->root_key.objectid);
2300                 if (ret && ret != -EEXIST) {
2301                         btrfs_abort_transaction(trans, root, ret);
2302                         return ret;
2303                 }
2304         }
2305         return 0;
2306 }
2307
2308 /*
2309  * We have done the truncate/delete so we can go ahead and remove the orphan
2310  * item for this particular inode.
2311  */
2312 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2313 {
2314         struct btrfs_root *root = BTRFS_I(inode)->root;
2315         int delete_item = 0;
2316         int release_rsv = 0;
2317         int ret = 0;
2318
2319         spin_lock(&root->orphan_lock);
2320         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2321                                &BTRFS_I(inode)->runtime_flags))
2322                 delete_item = 1;
2323
2324         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2325                                &BTRFS_I(inode)->runtime_flags))
2326                 release_rsv = 1;
2327         spin_unlock(&root->orphan_lock);
2328
2329         if (trans && delete_item) {
2330                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2331                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2332         }
2333
2334         if (release_rsv) {
2335                 btrfs_orphan_release_metadata(inode);
2336                 atomic_dec(&root->orphan_inodes);
2337         }
2338
2339         return 0;
2340 }
2341
2342 /*
2343  * this cleans up any orphans that may be left on the list from the last use
2344  * of this root.
2345  */
2346 int btrfs_orphan_cleanup(struct btrfs_root *root)
2347 {
2348         struct btrfs_path *path;
2349         struct extent_buffer *leaf;
2350         struct btrfs_key key, found_key;
2351         struct btrfs_trans_handle *trans;
2352         struct inode *inode;
2353         u64 last_objectid = 0;
2354         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2355
2356         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2357                 return 0;
2358
2359         path = btrfs_alloc_path();
2360         if (!path) {
2361                 ret = -ENOMEM;
2362                 goto out;
2363         }
2364         path->reada = -1;
2365
2366         key.objectid = BTRFS_ORPHAN_OBJECTID;
2367         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2368         key.offset = (u64)-1;
2369
2370         while (1) {
2371                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2372                 if (ret < 0)
2373                         goto out;
2374
2375                 /*
2376                  * if ret == 0 means we found what we were searching for, which
2377                  * is weird, but possible, so only screw with path if we didn't
2378                  * find the key and see if we have stuff that matches
2379                  */
2380                 if (ret > 0) {
2381                         ret = 0;
2382                         if (path->slots[0] == 0)
2383                                 break;
2384                         path->slots[0]--;
2385                 }
2386
2387                 /* pull out the item */
2388                 leaf = path->nodes[0];
2389                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2390
2391                 /* make sure the item matches what we want */
2392                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2393                         break;
2394                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2395                         break;
2396
2397                 /* release the path since we're done with it */
2398                 btrfs_release_path(path);
2399
2400                 /*
2401                  * this is where we are basically btrfs_lookup, without the
2402                  * crossing root thing.  we store the inode number in the
2403                  * offset of the orphan item.
2404                  */
2405
2406                 if (found_key.offset == last_objectid) {
2407                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2408                                "stopping orphan cleanup\n");
2409                         ret = -EINVAL;
2410                         goto out;
2411                 }
2412
2413                 last_objectid = found_key.offset;
2414
2415                 found_key.objectid = found_key.offset;
2416                 found_key.type = BTRFS_INODE_ITEM_KEY;
2417                 found_key.offset = 0;
2418                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2419                 ret = PTR_RET(inode);
2420                 if (ret && ret != -ESTALE)
2421                         goto out;
2422
2423                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2424                         struct btrfs_root *dead_root;
2425                         struct btrfs_fs_info *fs_info = root->fs_info;
2426                         int is_dead_root = 0;
2427
2428                         /*
2429                          * this is an orphan in the tree root. Currently these
2430                          * could come from 2 sources:
2431                          *  a) a snapshot deletion in progress
2432                          *  b) a free space cache inode
2433                          * We need to distinguish those two, as the snapshot
2434                          * orphan must not get deleted.
2435                          * find_dead_roots already ran before us, so if this
2436                          * is a snapshot deletion, we should find the root
2437                          * in the dead_roots list
2438                          */
2439                         spin_lock(&fs_info->trans_lock);
2440                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2441                                             root_list) {
2442                                 if (dead_root->root_key.objectid ==
2443                                     found_key.objectid) {
2444                                         is_dead_root = 1;
2445                                         break;
2446                                 }
2447                         }
2448                         spin_unlock(&fs_info->trans_lock);
2449                         if (is_dead_root) {
2450                                 /* prevent this orphan from being found again */
2451                                 key.offset = found_key.objectid - 1;
2452                                 continue;
2453                         }
2454                 }
2455                 /*
2456                  * Inode is already gone but the orphan item is still there,
2457                  * kill the orphan item.
2458                  */
2459                 if (ret == -ESTALE) {
2460                         trans = btrfs_start_transaction(root, 1);
2461                         if (IS_ERR(trans)) {
2462                                 ret = PTR_ERR(trans);
2463                                 goto out;
2464                         }
2465                         printk(KERN_ERR "auto deleting %Lu\n",
2466                                found_key.objectid);
2467                         ret = btrfs_del_orphan_item(trans, root,
2468                                                     found_key.objectid);
2469                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2470                         btrfs_end_transaction(trans, root);
2471                         continue;
2472                 }
2473
2474                 /*
2475                  * add this inode to the orphan list so btrfs_orphan_del does
2476                  * the proper thing when we hit it
2477                  */
2478                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2479                         &BTRFS_I(inode)->runtime_flags);
2480
2481                 /* if we have links, this was a truncate, lets do that */
2482                 if (inode->i_nlink) {
2483                         if (!S_ISREG(inode->i_mode)) {
2484                                 WARN_ON(1);
2485                                 iput(inode);
2486                                 continue;
2487                         }
2488                         nr_truncate++;
2489
2490                         /* 1 for the orphan item deletion. */
2491                         trans = btrfs_start_transaction(root, 1);
2492                         if (IS_ERR(trans)) {
2493                                 ret = PTR_ERR(trans);
2494                                 goto out;
2495                         }
2496                         ret = btrfs_orphan_add(trans, inode);
2497                         btrfs_end_transaction(trans, root);
2498                         if (ret)
2499                                 goto out;
2500
2501                         ret = btrfs_truncate(inode);
2502                 } else {
2503                         nr_unlink++;
2504                 }
2505
2506                 /* this will do delete_inode and everything for us */
2507                 iput(inode);
2508                 if (ret)
2509                         goto out;
2510         }
2511         /* release the path since we're done with it */
2512         btrfs_release_path(path);
2513
2514         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2515
2516         if (root->orphan_block_rsv)
2517                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2518                                         (u64)-1);
2519
2520         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2521                 trans = btrfs_join_transaction(root);
2522                 if (!IS_ERR(trans))
2523                         btrfs_end_transaction(trans, root);
2524         }
2525
2526         if (nr_unlink)
2527                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2528         if (nr_truncate)
2529                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2530
2531 out:
2532         if (ret)
2533                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2534         btrfs_free_path(path);
2535         return ret;
2536 }
2537
2538 /*
2539  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2540  * don't find any xattrs, we know there can't be any acls.
2541  *
2542  * slot is the slot the inode is in, objectid is the objectid of the inode
2543  */
2544 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2545                                           int slot, u64 objectid)
2546 {
2547         u32 nritems = btrfs_header_nritems(leaf);
2548         struct btrfs_key found_key;
2549         int scanned = 0;
2550
2551         slot++;
2552         while (slot < nritems) {
2553                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2554
2555                 /* we found a different objectid, there must not be acls */
2556                 if (found_key.objectid != objectid)
2557                         return 0;
2558
2559                 /* we found an xattr, assume we've got an acl */
2560                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2561                         return 1;
2562
2563                 /*
2564                  * we found a key greater than an xattr key, there can't
2565                  * be any acls later on
2566                  */
2567                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2568                         return 0;
2569
2570                 slot++;
2571                 scanned++;
2572
2573                 /*
2574                  * it goes inode, inode backrefs, xattrs, extents,
2575                  * so if there are a ton of hard links to an inode there can
2576                  * be a lot of backrefs.  Don't waste time searching too hard,
2577                  * this is just an optimization
2578                  */
2579                 if (scanned >= 8)
2580                         break;
2581         }
2582         /* we hit the end of the leaf before we found an xattr or
2583          * something larger than an xattr.  We have to assume the inode
2584          * has acls
2585          */
2586         return 1;
2587 }
2588
2589 /*
2590  * read an inode from the btree into the in-memory inode
2591  */
2592 static void btrfs_read_locked_inode(struct inode *inode)
2593 {
2594         struct btrfs_path *path;
2595         struct extent_buffer *leaf;
2596         struct btrfs_inode_item *inode_item;
2597         struct btrfs_timespec *tspec;
2598         struct btrfs_root *root = BTRFS_I(inode)->root;
2599         struct btrfs_key location;
2600         int maybe_acls;
2601         u32 rdev;
2602         int ret;
2603         bool filled = false;
2604
2605         ret = btrfs_fill_inode(inode, &rdev);
2606         if (!ret)
2607                 filled = true;
2608
2609         path = btrfs_alloc_path();
2610         if (!path)
2611                 goto make_bad;
2612
2613         path->leave_spinning = 1;
2614         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2615
2616         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2617         if (ret)
2618                 goto make_bad;
2619
2620         leaf = path->nodes[0];
2621
2622         if (filled)
2623                 goto cache_acl;
2624
2625         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2626                                     struct btrfs_inode_item);
2627         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2628         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2629         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2630         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2631         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2632
2633         tspec = btrfs_inode_atime(inode_item);
2634         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2635         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2636
2637         tspec = btrfs_inode_mtime(inode_item);
2638         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2639         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2640
2641         tspec = btrfs_inode_ctime(inode_item);
2642         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2643         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2644
2645         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2646         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2647         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2648
2649         /*
2650          * If we were modified in the current generation and evicted from memory
2651          * and then re-read we need to do a full sync since we don't have any
2652          * idea about which extents were modified before we were evicted from
2653          * cache.
2654          */
2655         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2656                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2657                         &BTRFS_I(inode)->runtime_flags);
2658
2659         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2660         inode->i_generation = BTRFS_I(inode)->generation;
2661         inode->i_rdev = 0;
2662         rdev = btrfs_inode_rdev(leaf, inode_item);
2663
2664         BTRFS_I(inode)->index_cnt = (u64)-1;
2665         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2666 cache_acl:
2667         /*
2668          * try to precache a NULL acl entry for files that don't have
2669          * any xattrs or acls
2670          */
2671         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2672                                            btrfs_ino(inode));
2673         if (!maybe_acls)
2674                 cache_no_acl(inode);
2675
2676         btrfs_free_path(path);
2677
2678         switch (inode->i_mode & S_IFMT) {
2679         case S_IFREG:
2680                 inode->i_mapping->a_ops = &btrfs_aops;
2681                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2682                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2683                 inode->i_fop = &btrfs_file_operations;
2684                 inode->i_op = &btrfs_file_inode_operations;
2685                 break;
2686         case S_IFDIR:
2687                 inode->i_fop = &btrfs_dir_file_operations;
2688                 if (root == root->fs_info->tree_root)
2689                         inode->i_op = &btrfs_dir_ro_inode_operations;
2690                 else
2691                         inode->i_op = &btrfs_dir_inode_operations;
2692                 break;
2693         case S_IFLNK:
2694                 inode->i_op = &btrfs_symlink_inode_operations;
2695                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2696                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2697                 break;
2698         default:
2699                 inode->i_op = &btrfs_special_inode_operations;
2700                 init_special_inode(inode, inode->i_mode, rdev);
2701                 break;
2702         }
2703
2704         btrfs_update_iflags(inode);
2705         return;
2706
2707 make_bad:
2708         btrfs_free_path(path);
2709         make_bad_inode(inode);
2710 }
2711
2712 /*
2713  * given a leaf and an inode, copy the inode fields into the leaf
2714  */
2715 static void fill_inode_item(struct btrfs_trans_handle *trans,
2716                             struct extent_buffer *leaf,
2717                             struct btrfs_inode_item *item,
2718                             struct inode *inode)
2719 {
2720         struct btrfs_map_token token;
2721
2722         btrfs_init_map_token(&token);
2723
2724         btrfs_set_token_inode_uid(leaf, item, i_uid_read(inode), &token);
2725         btrfs_set_token_inode_gid(leaf, item, i_gid_read(inode), &token);
2726         btrfs_set_token_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size,
2727                                    &token);
2728         btrfs_set_token_inode_mode(leaf, item, inode->i_mode, &token);
2729         btrfs_set_token_inode_nlink(leaf, item, inode->i_nlink, &token);
2730
2731         btrfs_set_token_timespec_sec(leaf, btrfs_inode_atime(item),
2732                                      inode->i_atime.tv_sec, &token);
2733         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_atime(item),
2734                                       inode->i_atime.tv_nsec, &token);
2735
2736         btrfs_set_token_timespec_sec(leaf, btrfs_inode_mtime(item),
2737                                      inode->i_mtime.tv_sec, &token);
2738         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_mtime(item),
2739                                       inode->i_mtime.tv_nsec, &token);
2740
2741         btrfs_set_token_timespec_sec(leaf, btrfs_inode_ctime(item),
2742                                      inode->i_ctime.tv_sec, &token);
2743         btrfs_set_token_timespec_nsec(leaf, btrfs_inode_ctime(item),
2744                                       inode->i_ctime.tv_nsec, &token);
2745
2746         btrfs_set_token_inode_nbytes(leaf, item, inode_get_bytes(inode),
2747                                      &token);
2748         btrfs_set_token_inode_generation(leaf, item, BTRFS_I(inode)->generation,
2749                                          &token);
2750         btrfs_set_token_inode_sequence(leaf, item, inode->i_version, &token);
2751         btrfs_set_token_inode_transid(leaf, item, trans->transid, &token);
2752         btrfs_set_token_inode_rdev(leaf, item, inode->i_rdev, &token);
2753         btrfs_set_token_inode_flags(leaf, item, BTRFS_I(inode)->flags, &token);
2754         btrfs_set_token_inode_block_group(leaf, item, 0, &token);
2755 }
2756
2757 /*
2758  * copy everything in the in-memory inode into the btree.
2759  */
2760 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2761                                 struct btrfs_root *root, struct inode *inode)
2762 {
2763         struct btrfs_inode_item *inode_item;
2764         struct btrfs_path *path;
2765         struct extent_buffer *leaf;
2766         int ret;
2767
2768         path = btrfs_alloc_path();
2769         if (!path)
2770                 return -ENOMEM;
2771
2772         path->leave_spinning = 1;
2773         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2774                                  1);
2775         if (ret) {
2776                 if (ret > 0)
2777                         ret = -ENOENT;
2778                 goto failed;
2779         }
2780
2781         btrfs_unlock_up_safe(path, 1);
2782         leaf = path->nodes[0];
2783         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2784                                     struct btrfs_inode_item);
2785
2786         fill_inode_item(trans, leaf, inode_item, inode);
2787         btrfs_mark_buffer_dirty(leaf);
2788         btrfs_set_inode_last_trans(trans, inode);
2789         ret = 0;
2790 failed:
2791         btrfs_free_path(path);
2792         return ret;
2793 }
2794
2795 /*
2796  * copy everything in the in-memory inode into the btree.
2797  */
2798 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2799                                 struct btrfs_root *root, struct inode *inode)
2800 {
2801         int ret;
2802
2803         /*
2804          * If the inode is a free space inode, we can deadlock during commit
2805          * if we put it into the delayed code.
2806          *
2807          * The data relocation inode should also be directly updated
2808          * without delay
2809          */
2810         if (!btrfs_is_free_space_inode(inode)
2811             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2812                 btrfs_update_root_times(trans, root);
2813
2814                 ret = btrfs_delayed_update_inode(trans, root, inode);
2815                 if (!ret)
2816                         btrfs_set_inode_last_trans(trans, inode);
2817                 return ret;
2818         }
2819
2820         return btrfs_update_inode_item(trans, root, inode);
2821 }
2822
2823 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2824                                          struct btrfs_root *root,
2825                                          struct inode *inode)
2826 {
2827         int ret;
2828
2829         ret = btrfs_update_inode(trans, root, inode);
2830         if (ret == -ENOSPC)
2831                 return btrfs_update_inode_item(trans, root, inode);
2832         return ret;
2833 }
2834
2835 /*
2836  * unlink helper that gets used here in inode.c and in the tree logging
2837  * recovery code.  It remove a link in a directory with a given name, and
2838  * also drops the back refs in the inode to the directory
2839  */
2840 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2841                                 struct btrfs_root *root,
2842                                 struct inode *dir, struct inode *inode,
2843                                 const char *name, int name_len)
2844 {
2845         struct btrfs_path *path;
2846         int ret = 0;
2847         struct extent_buffer *leaf;
2848         struct btrfs_dir_item *di;
2849         struct btrfs_key key;
2850         u64 index;
2851         u64 ino = btrfs_ino(inode);
2852         u64 dir_ino = btrfs_ino(dir);
2853
2854         path = btrfs_alloc_path();
2855         if (!path) {
2856                 ret = -ENOMEM;
2857                 goto out;
2858         }
2859
2860         path->leave_spinning = 1;
2861         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2862                                     name, name_len, -1);
2863         if (IS_ERR(di)) {
2864                 ret = PTR_ERR(di);
2865                 goto err;
2866         }
2867         if (!di) {
2868                 ret = -ENOENT;
2869                 goto err;
2870         }
2871         leaf = path->nodes[0];
2872         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2873         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2874         if (ret)
2875                 goto err;
2876         btrfs_release_path(path);
2877
2878         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2879                                   dir_ino, &index);
2880         if (ret) {
2881                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2882                        "inode %llu parent %llu\n", name_len, name,
2883                        (unsigned long long)ino, (unsigned long long)dir_ino);
2884                 btrfs_abort_transaction(trans, root, ret);
2885                 goto err;
2886         }
2887
2888         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2889         if (ret) {
2890                 btrfs_abort_transaction(trans, root, ret);
2891                 goto err;
2892         }
2893
2894         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2895                                          inode, dir_ino);
2896         if (ret != 0 && ret != -ENOENT) {
2897                 btrfs_abort_transaction(trans, root, ret);
2898                 goto err;
2899         }
2900
2901         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2902                                            dir, index);
2903         if (ret == -ENOENT)
2904                 ret = 0;
2905 err:
2906         btrfs_free_path(path);
2907         if (ret)
2908                 goto out;
2909
2910         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2911         inode_inc_iversion(inode);
2912         inode_inc_iversion(dir);
2913         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2914         ret = btrfs_update_inode(trans, root, dir);
2915 out:
2916         return ret;
2917 }
2918
2919 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2920                        struct btrfs_root *root,
2921                        struct inode *dir, struct inode *inode,
2922                        const char *name, int name_len)
2923 {
2924         int ret;
2925         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2926         if (!ret) {
2927                 btrfs_drop_nlink(inode);
2928                 ret = btrfs_update_inode(trans, root, inode);
2929         }
2930         return ret;
2931 }
2932
2933
2934 /* helper to check if there is any shared block in the path */
2935 static int check_path_shared(struct btrfs_root *root,
2936                              struct btrfs_path *path)
2937 {
2938         struct extent_buffer *eb;
2939         int level;
2940         u64 refs = 1;
2941
2942         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2943                 int ret;
2944
2945                 if (!path->nodes[level])
2946                         break;
2947                 eb = path->nodes[level];
2948                 if (!btrfs_block_can_be_shared(root, eb))
2949                         continue;
2950                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2951                                                &refs, NULL);
2952                 if (refs > 1)
2953                         return 1;
2954         }
2955         return 0;
2956 }
2957
2958 /*
2959  * helper to start transaction for unlink and rmdir.
2960  *
2961  * unlink and rmdir are special in btrfs, they do not always free space.
2962  * so in enospc case, we should make sure they will free space before
2963  * allowing them to use the global metadata reservation.
2964  */
2965 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2966                                                        struct dentry *dentry)
2967 {
2968         struct btrfs_trans_handle *trans;
2969         struct btrfs_root *root = BTRFS_I(dir)->root;
2970         struct btrfs_path *path;
2971         struct btrfs_dir_item *di;
2972         struct inode *inode = dentry->d_inode;
2973         u64 index;
2974         int check_link = 1;
2975         int err = -ENOSPC;
2976         int ret;
2977         u64 ino = btrfs_ino(inode);
2978         u64 dir_ino = btrfs_ino(dir);
2979
2980         /*
2981          * 1 for the possible orphan item
2982          * 1 for the dir item
2983          * 1 for the dir index
2984          * 1 for the inode ref
2985          * 1 for the inode ref in the tree log
2986          * 2 for the dir entries in the log
2987          * 1 for the inode
2988          */
2989         trans = btrfs_start_transaction(root, 8);
2990         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2991                 return trans;
2992
2993         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2994                 return ERR_PTR(-ENOSPC);
2995
2996         /* check if there is someone else holds reference */
2997         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2998                 return ERR_PTR(-ENOSPC);
2999
3000         if (atomic_read(&inode->i_count) > 2)
3001                 return ERR_PTR(-ENOSPC);
3002
3003         if (xchg(&root->fs_info->enospc_unlink, 1))
3004                 return ERR_PTR(-ENOSPC);
3005
3006         path = btrfs_alloc_path();
3007         if (!path) {
3008                 root->fs_info->enospc_unlink = 0;
3009                 return ERR_PTR(-ENOMEM);
3010         }
3011
3012         /* 1 for the orphan item */
3013         trans = btrfs_start_transaction(root, 1);
3014         if (IS_ERR(trans)) {
3015                 btrfs_free_path(path);
3016                 root->fs_info->enospc_unlink = 0;
3017                 return trans;
3018         }
3019
3020         path->skip_locking = 1;
3021         path->search_commit_root = 1;
3022
3023         ret = btrfs_lookup_inode(trans, root, path,
3024                                 &BTRFS_I(dir)->location, 0);
3025         if (ret < 0) {
3026                 err = ret;
3027                 goto out;
3028         }
3029         if (ret == 0) {
3030                 if (check_path_shared(root, path))
3031                         goto out;
3032         } else {
3033                 check_link = 0;
3034         }
3035         btrfs_release_path(path);
3036
3037         ret = btrfs_lookup_inode(trans, root, path,
3038                                 &BTRFS_I(inode)->location, 0);
3039         if (ret < 0) {
3040                 err = ret;
3041                 goto out;
3042         }
3043         if (ret == 0) {
3044                 if (check_path_shared(root, path))
3045                         goto out;
3046         } else {
3047                 check_link = 0;
3048         }
3049         btrfs_release_path(path);
3050
3051         if (ret == 0 && S_ISREG(inode->i_mode)) {
3052                 ret = btrfs_lookup_file_extent(trans, root, path,
3053                                                ino, (u64)-1, 0);
3054                 if (ret < 0) {
3055                         err = ret;
3056                         goto out;
3057                 }
3058                 BUG_ON(ret == 0); /* Corruption */
3059                 if (check_path_shared(root, path))
3060                         goto out;
3061                 btrfs_release_path(path);
3062         }
3063
3064         if (!check_link) {
3065                 err = 0;
3066                 goto out;
3067         }
3068
3069         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3070                                 dentry->d_name.name, dentry->d_name.len, 0);
3071         if (IS_ERR(di)) {
3072                 err = PTR_ERR(di);
3073                 goto out;
3074         }
3075         if (di) {
3076                 if (check_path_shared(root, path))
3077                         goto out;
3078         } else {
3079                 err = 0;
3080                 goto out;
3081         }
3082         btrfs_release_path(path);
3083
3084         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3085                                         dentry->d_name.len, ino, dir_ino, 0,
3086                                         &index);
3087         if (ret) {
3088                 err = ret;
3089                 goto out;
3090         }
3091
3092         if (check_path_shared(root, path))
3093                 goto out;
3094
3095         btrfs_release_path(path);
3096
3097         /*
3098          * This is a commit root search, if we can lookup inode item and other
3099          * relative items in the commit root, it means the transaction of
3100          * dir/file creation has been committed, and the dir index item that we
3101          * delay to insert has also been inserted into the commit root. So
3102          * we needn't worry about the delayed insertion of the dir index item
3103          * here.
3104          */
3105         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3106                                 dentry->d_name.name, dentry->d_name.len, 0);
3107         if (IS_ERR(di)) {
3108                 err = PTR_ERR(di);
3109                 goto out;
3110         }
3111         BUG_ON(ret == -ENOENT);
3112         if (check_path_shared(root, path))
3113                 goto out;
3114
3115         err = 0;
3116 out:
3117         btrfs_free_path(path);
3118         /* Migrate the orphan reservation over */
3119         if (!err)
3120                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3121                                 &root->fs_info->global_block_rsv,
3122                                 trans->bytes_reserved);
3123
3124         if (err) {
3125                 btrfs_end_transaction(trans, root);
3126                 root->fs_info->enospc_unlink = 0;
3127                 return ERR_PTR(err);
3128         }
3129
3130         trans->block_rsv = &root->fs_info->global_block_rsv;
3131         return trans;
3132 }
3133
3134 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3135                                struct btrfs_root *root)
3136 {
3137         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3138                 btrfs_block_rsv_release(root, trans->block_rsv,
3139                                         trans->bytes_reserved);
3140                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3141                 BUG_ON(!root->fs_info->enospc_unlink);
3142                 root->fs_info->enospc_unlink = 0;
3143         }
3144         btrfs_end_transaction(trans, root);
3145 }
3146
3147 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3148 {
3149         struct btrfs_root *root = BTRFS_I(dir)->root;
3150         struct btrfs_trans_handle *trans;
3151         struct inode *inode = dentry->d_inode;
3152         int ret;
3153
3154         trans = __unlink_start_trans(dir, dentry);
3155         if (IS_ERR(trans))
3156                 return PTR_ERR(trans);
3157
3158         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3159
3160         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3161                                  dentry->d_name.name, dentry->d_name.len);
3162         if (ret)
3163                 goto out;
3164
3165         if (inode->i_nlink == 0) {
3166                 ret = btrfs_orphan_add(trans, inode);
3167                 if (ret)
3168                         goto out;
3169         }
3170
3171 out:
3172         __unlink_end_trans(trans, root);
3173         btrfs_btree_balance_dirty(root);
3174         return ret;
3175 }
3176
3177 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3178                         struct btrfs_root *root,
3179                         struct inode *dir, u64 objectid,
3180                         const char *name, int name_len)
3181 {
3182         struct btrfs_path *path;
3183         struct extent_buffer *leaf;
3184         struct btrfs_dir_item *di;
3185         struct btrfs_key key;
3186         u64 index;
3187         int ret;
3188         u64 dir_ino = btrfs_ino(dir);
3189
3190         path = btrfs_alloc_path();
3191         if (!path)
3192                 return -ENOMEM;
3193
3194         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3195                                    name, name_len, -1);
3196         if (IS_ERR_OR_NULL(di)) {
3197                 if (!di)
3198                         ret = -ENOENT;
3199                 else
3200                         ret = PTR_ERR(di);
3201                 goto out;
3202         }
3203
3204         leaf = path->nodes[0];
3205         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3206         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3207         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3208         if (ret) {
3209                 btrfs_abort_transaction(trans, root, ret);
3210                 goto out;
3211         }
3212         btrfs_release_path(path);
3213
3214         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3215                                  objectid, root->root_key.objectid,
3216                                  dir_ino, &index, name, name_len);
3217         if (ret < 0) {
3218                 if (ret != -ENOENT) {
3219                         btrfs_abort_transaction(trans, root, ret);
3220                         goto out;
3221                 }
3222                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3223                                                  name, name_len);
3224                 if (IS_ERR_OR_NULL(di)) {
3225                         if (!di)
3226                                 ret = -ENOENT;
3227                         else
3228                                 ret = PTR_ERR(di);
3229                         btrfs_abort_transaction(trans, root, ret);
3230                         goto out;
3231                 }
3232
3233                 leaf = path->nodes[0];
3234                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3235                 btrfs_release_path(path);
3236                 index = key.offset;
3237         }
3238         btrfs_release_path(path);
3239
3240         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3241         if (ret) {
3242                 btrfs_abort_transaction(trans, root, ret);
3243                 goto out;
3244         }
3245
3246         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3247         inode_inc_iversion(dir);
3248         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3249         ret = btrfs_update_inode_fallback(trans, root, dir);
3250         if (ret)
3251                 btrfs_abort_transaction(trans, root, ret);
3252 out:
3253         btrfs_free_path(path);
3254         return ret;
3255 }
3256
3257 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3258 {
3259         struct inode *inode = dentry->d_inode;
3260         int err = 0;
3261         struct btrfs_root *root = BTRFS_I(dir)->root;
3262         struct btrfs_trans_handle *trans;
3263
3264         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3265                 return -ENOTEMPTY;
3266         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3267                 return -EPERM;
3268
3269         trans = __unlink_start_trans(dir, dentry);
3270         if (IS_ERR(trans))
3271                 return PTR_ERR(trans);
3272
3273         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3274                 err = btrfs_unlink_subvol(trans, root, dir,
3275                                           BTRFS_I(inode)->location.objectid,
3276                                           dentry->d_name.name,
3277                                           dentry->d_name.len);
3278                 goto out;
3279         }
3280
3281         err = btrfs_orphan_add(trans, inode);
3282         if (err)
3283                 goto out;
3284
3285         /* now the directory is empty */
3286         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3287                                  dentry->d_name.name, dentry->d_name.len);
3288         if (!err)
3289                 btrfs_i_size_write(inode, 0);
3290 out:
3291         __unlink_end_trans(trans, root);
3292         btrfs_btree_balance_dirty(root);
3293
3294         return err;
3295 }
3296
3297 /*
3298  * this can truncate away extent items, csum items and directory items.
3299  * It starts at a high offset and removes keys until it can't find
3300  * any higher than new_size
3301  *
3302  * csum items that cross the new i_size are truncated to the new size
3303  * as well.
3304  *
3305  * min_type is the minimum key type to truncate down to.  If set to 0, this
3306  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3307  */
3308 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3309                                struct btrfs_root *root,
3310                                struct inode *inode,
3311                                u64 new_size, u32 min_type)
3312 {
3313         struct btrfs_path *path;
3314         struct extent_buffer *leaf;
3315         struct btrfs_file_extent_item *fi;
3316         struct btrfs_key key;
3317         struct btrfs_key found_key;
3318         u64 extent_start = 0;
3319         u64 extent_num_bytes = 0;
3320         u64 extent_offset = 0;
3321         u64 item_end = 0;
3322         u64 mask = root->sectorsize - 1;
3323         u32 found_type = (u8)-1;
3324         int found_extent;
3325         int del_item;
3326         int pending_del_nr = 0;
3327         int pending_del_slot = 0;
3328         int extent_type = -1;
3329         int ret;
3330         int err = 0;
3331         u64 ino = btrfs_ino(inode);
3332
3333         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3334
3335         path = btrfs_alloc_path();
3336         if (!path)
3337                 return -ENOMEM;
3338         path->reada = -1;
3339
3340         /*
3341          * We want to drop from the next block forward in case this new size is
3342          * not block aligned since we will be keeping the last block of the
3343          * extent just the way it is.
3344          */
3345         if (root->ref_cows || root == root->fs_info->tree_root)
3346                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3347
3348         /*
3349          * This function is also used to drop the items in the log tree before
3350          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3351          * it is used to drop the loged items. So we shouldn't kill the delayed
3352          * items.
3353          */
3354         if (min_type == 0 && root == BTRFS_I(inode)->root)
3355                 btrfs_kill_delayed_inode_items(inode);
3356
3357         key.objectid = ino;
3358         key.offset = (u64)-1;
3359         key.type = (u8)-1;
3360
3361 search_again:
3362         path->leave_spinning = 1;
3363         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3364         if (ret < 0) {
3365                 err = ret;
3366                 goto out;
3367         }
3368
3369         if (ret > 0) {
3370                 /* there are no items in the tree for us to truncate, we're
3371                  * done
3372                  */
3373                 if (path->slots[0] == 0)
3374                         goto out;
3375                 path->slots[0]--;
3376         }
3377
3378         while (1) {
3379                 fi = NULL;
3380                 leaf = path->nodes[0];
3381                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3382                 found_type = btrfs_key_type(&found_key);
3383
3384                 if (found_key.objectid != ino)
3385                         break;
3386
3387                 if (found_type < min_type)
3388                         break;
3389
3390                 item_end = found_key.offset;
3391                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3392                         fi = btrfs_item_ptr(leaf, path->slots[0],
3393                                             struct btrfs_file_extent_item);
3394                         extent_type = btrfs_file_extent_type(leaf, fi);
3395                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3396                                 item_end +=
3397                                     btrfs_file_extent_num_bytes(leaf, fi);
3398                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3399                                 item_end += btrfs_file_extent_inline_len(leaf,
3400                                                                          fi);
3401                         }
3402                         item_end--;
3403                 }
3404                 if (found_type > min_type) {
3405                         del_item = 1;
3406                 } else {
3407                         if (item_end < new_size)
3408                                 break;
3409                         if (found_key.offset >= new_size)
3410                                 del_item = 1;
3411                         else
3412                                 del_item = 0;
3413                 }
3414                 found_extent = 0;
3415                 /* FIXME, shrink the extent if the ref count is only 1 */
3416                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3417                         goto delete;
3418
3419                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3420                         u64 num_dec;
3421                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3422                         if (!del_item) {
3423                                 u64 orig_num_bytes =
3424                                         btrfs_file_extent_num_bytes(leaf, fi);
3425                                 extent_num_bytes = new_size -
3426                                         found_key.offset + root->sectorsize - 1;
3427                                 extent_num_bytes = extent_num_bytes &
3428                                         ~((u64)root->sectorsize - 1);
3429                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3430                                                          extent_num_bytes);
3431                                 num_dec = (orig_num_bytes -
3432                                            extent_num_bytes);
3433                                 if (root->ref_cows && extent_start != 0)
3434                                         inode_sub_bytes(inode, num_dec);
3435                                 btrfs_mark_buffer_dirty(leaf);
3436                         } else {
3437                                 extent_num_bytes =
3438                                         btrfs_file_extent_disk_num_bytes(leaf,
3439                                                                          fi);
3440                                 extent_offset = found_key.offset -
3441                                         btrfs_file_extent_offset(leaf, fi);
3442
3443                                 /* FIXME blocksize != 4096 */
3444                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3445                                 if (extent_start != 0) {
3446                                         found_extent = 1;
3447                                         if (root->ref_cows)
3448                                                 inode_sub_bytes(inode, num_dec);
3449                                 }
3450                         }
3451                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3452                         /*
3453                          * we can't truncate inline items that have had
3454                          * special encodings
3455                          */
3456                         if (!del_item &&
3457                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3458                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3459                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3460                                 u32 size = new_size - found_key.offset;
3461
3462                                 if (root->ref_cows) {
3463                                         inode_sub_bytes(inode, item_end + 1 -
3464                                                         new_size);
3465                                 }
3466                                 size =
3467                                     btrfs_file_extent_calc_inline_size(size);
3468                                 btrfs_truncate_item(trans, root, path,
3469                                                     size, 1);
3470                         } else if (root->ref_cows) {
3471                                 inode_sub_bytes(inode, item_end + 1 -
3472                                                 found_key.offset);
3473                         }
3474                 }
3475 delete:
3476                 if (del_item) {
3477                         if (!pending_del_nr) {
3478                                 /* no pending yet, add ourselves */
3479                                 pending_del_slot = path->slots[0];
3480                                 pending_del_nr = 1;
3481                         } else if (pending_del_nr &&
3482                                    path->slots[0] + 1 == pending_del_slot) {
3483                                 /* hop on the pending chunk */
3484                                 pending_del_nr++;
3485                                 pending_del_slot = path->slots[0];
3486                         } else {
3487                                 BUG();
3488                         }
3489                 } else {
3490                         break;
3491                 }
3492                 if (found_extent && (root->ref_cows ||
3493                                      root == root->fs_info->tree_root)) {
3494                         btrfs_set_path_blocking(path);
3495                         ret = btrfs_free_extent(trans, root, extent_start,
3496                                                 extent_num_bytes, 0,
3497                                                 btrfs_header_owner(leaf),
3498                                                 ino, extent_offset, 0);
3499                         BUG_ON(ret);
3500                 }
3501
3502                 if (found_type == BTRFS_INODE_ITEM_KEY)
3503                         break;
3504
3505                 if (path->slots[0] == 0 ||
3506                     path->slots[0] != pending_del_slot) {
3507                         if (pending_del_nr) {
3508                                 ret = btrfs_del_items(trans, root, path,
3509                                                 pending_del_slot,
3510                                                 pending_del_nr);
3511                                 if (ret) {
3512                                         btrfs_abort_transaction(trans,
3513                                                                 root, ret);
3514                                         goto error;
3515                                 }
3516                                 pending_del_nr = 0;
3517                         }
3518                         btrfs_release_path(path);
3519                         goto search_again;
3520                 } else {
3521                         path->slots[0]--;
3522                 }
3523         }
3524 out:
3525         if (pending_del_nr) {
3526                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3527                                       pending_del_nr);
3528                 if (ret)
3529                         btrfs_abort_transaction(trans, root, ret);
3530         }
3531 error:
3532         btrfs_free_path(path);
3533         return err;
3534 }
3535
3536 /*
3537  * btrfs_truncate_page - read, zero a chunk and write a page
3538  * @inode - inode that we're zeroing
3539  * @from - the offset to start zeroing
3540  * @len - the length to zero, 0 to zero the entire range respective to the
3541  *      offset
3542  * @front - zero up to the offset instead of from the offset on
3543  *
3544  * This will find the page for the "from" offset and cow the page and zero the
3545  * part we want to zero.  This is used with truncate and hole punching.
3546  */
3547 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3548                         int front)
3549 {
3550         struct address_space *mapping = inode->i_mapping;
3551         struct btrfs_root *root = BTRFS_I(inode)->root;
3552         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3553         struct btrfs_ordered_extent *ordered;
3554         struct extent_state *cached_state = NULL;
3555         char *kaddr;
3556         u32 blocksize = root->sectorsize;
3557         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3558         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3559         struct page *page;
3560         gfp_t mask = btrfs_alloc_write_mask(mapping);
3561         int ret = 0;
3562         u64 page_start;
3563         u64 page_end;
3564
3565         if ((offset & (blocksize - 1)) == 0 &&
3566             (!len || ((len & (blocksize - 1)) == 0)))
3567                 goto out;
3568         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3569         if (ret)
3570                 goto out;
3571
3572 again:
3573         page = find_or_create_page(mapping, index, mask);
3574         if (!page) {
3575                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3576                 ret = -ENOMEM;
3577                 goto out;
3578         }
3579
3580         page_start = page_offset(page);
3581         page_end = page_start + PAGE_CACHE_SIZE - 1;
3582
3583         if (!PageUptodate(page)) {
3584                 ret = btrfs_readpage(NULL, page);
3585                 lock_page(page);
3586                 if (page->mapping != mapping) {
3587                         unlock_page(page);
3588                         page_cache_release(page);
3589                         goto again;
3590                 }
3591                 if (!PageUptodate(page)) {
3592                         ret = -EIO;
3593                         goto out_unlock;
3594                 }
3595         }
3596         wait_on_page_writeback(page);
3597
3598         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3599         set_page_extent_mapped(page);
3600
3601         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3602         if (ordered) {
3603                 unlock_extent_cached(io_tree, page_start, page_end,
3604                                      &cached_state, GFP_NOFS);
3605                 unlock_page(page);
3606                 page_cache_release(page);
3607                 btrfs_start_ordered_extent(inode, ordered, 1);
3608                 btrfs_put_ordered_extent(ordered);
3609                 goto again;
3610         }
3611
3612         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3613                           EXTENT_DIRTY | EXTENT_DELALLOC |
3614                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3615                           0, 0, &cached_state, GFP_NOFS);
3616
3617         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3618                                         &cached_state);
3619         if (ret) {
3620                 unlock_extent_cached(io_tree, page_start, page_end,
3621                                      &cached_state, GFP_NOFS);
3622                 goto out_unlock;
3623         }
3624
3625         if (offset != PAGE_CACHE_SIZE) {
3626                 if (!len)
3627                         len = PAGE_CACHE_SIZE - offset;
3628                 kaddr = kmap(page);
3629                 if (front)
3630                         memset(kaddr, 0, offset);
3631                 else
3632                         memset(kaddr + offset, 0, len);
3633                 flush_dcache_page(page);
3634                 kunmap(page);
3635         }
3636         ClearPageChecked(page);
3637         set_page_dirty(page);
3638         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3639                              GFP_NOFS);
3640
3641 out_unlock:
3642         if (ret)
3643                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3644         unlock_page(page);
3645         page_cache_release(page);
3646 out:
3647         return ret;
3648 }
3649
3650 /*
3651  * This function puts in dummy file extents for the area we're creating a hole
3652  * for.  So if we are truncating this file to a larger size we need to insert
3653  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3654  * the range between oldsize and size
3655  */
3656 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3657 {
3658         struct btrfs_trans_handle *trans;
3659         struct btrfs_root *root = BTRFS_I(inode)->root;
3660         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3661         struct extent_map *em = NULL;
3662         struct extent_state *cached_state = NULL;
3663         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3664         u64 mask = root->sectorsize - 1;
3665         u64 hole_start = (oldsize + mask) & ~mask;
3666         u64 block_end = (size + mask) & ~mask;
3667         u64 last_byte;
3668         u64 cur_offset;
3669         u64 hole_size;
3670         int err = 0;
3671
3672         if (size <= hole_start)
3673                 return 0;
3674
3675         while (1) {
3676                 struct btrfs_ordered_extent *ordered;
3677                 btrfs_wait_ordered_range(inode, hole_start,
3678                                          block_end - hole_start);
3679                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3680                                  &cached_state);
3681                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3682                 if (!ordered)
3683                         break;
3684                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3685                                      &cached_state, GFP_NOFS);
3686                 btrfs_put_ordered_extent(ordered);
3687         }
3688
3689         cur_offset = hole_start;
3690         while (1) {
3691                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3692                                 block_end - cur_offset, 0);
3693                 if (IS_ERR(em)) {
3694                         err = PTR_ERR(em);
3695                         em = NULL;
3696                         break;
3697                 }
3698                 last_byte = min(extent_map_end(em), block_end);
3699                 last_byte = (last_byte + mask) & ~mask;
3700                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3701                         struct extent_map *hole_em;
3702                         hole_size = last_byte - cur_offset;
3703
3704                         trans = btrfs_start_transaction(root, 3);
3705                         if (IS_ERR(trans)) {
3706                                 err = PTR_ERR(trans);
3707                                 break;
3708                         }
3709
3710                         err = btrfs_drop_extents(trans, root, inode,
3711                                                  cur_offset,
3712                                                  cur_offset + hole_size, 1);
3713                         if (err) {
3714                                 btrfs_abort_transaction(trans, root, err);
3715                                 btrfs_end_transaction(trans, root);
3716                                 break;
3717                         }
3718
3719                         err = btrfs_insert_file_extent(trans, root,
3720                                         btrfs_ino(inode), cur_offset, 0,
3721                                         0, hole_size, 0, hole_size,
3722                                         0, 0, 0);
3723                         if (err) {
3724                                 btrfs_abort_transaction(trans, root, err);
3725                                 btrfs_end_transaction(trans, root);
3726                                 break;
3727                         }
3728
3729                         btrfs_drop_extent_cache(inode, cur_offset,
3730                                                 cur_offset + hole_size - 1, 0);
3731                         hole_em = alloc_extent_map();
3732                         if (!hole_em) {
3733                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3734                                         &BTRFS_I(inode)->runtime_flags);
3735                                 goto next;
3736                         }
3737                         hole_em->start = cur_offset;
3738                         hole_em->len = hole_size;
3739                         hole_em->orig_start = cur_offset;
3740
3741                         hole_em->block_start = EXTENT_MAP_HOLE;
3742                         hole_em->block_len = 0;
3743                         hole_em->orig_block_len = 0;
3744                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3745                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3746                         hole_em->generation = trans->transid;
3747
3748                         while (1) {
3749                                 write_lock(&em_tree->lock);
3750                                 err = add_extent_mapping(em_tree, hole_em);
3751                                 if (!err)
3752                                         list_move(&hole_em->list,
3753                                                   &em_tree->modified_extents);
3754                                 write_unlock(&em_tree->lock);
3755                                 if (err != -EEXIST)
3756                                         break;
3757                                 btrfs_drop_extent_cache(inode, cur_offset,
3758                                                         cur_offset +
3759                                                         hole_size - 1, 0);
3760                         }
3761                         free_extent_map(hole_em);
3762 next:
3763                         btrfs_update_inode(trans, root, inode);
3764                         btrfs_end_transaction(trans, root);
3765                 }
3766                 free_extent_map(em);
3767                 em = NULL;
3768                 cur_offset = last_byte;
3769                 if (cur_offset >= block_end)
3770                         break;
3771         }
3772
3773         free_extent_map(em);
3774         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3775                              GFP_NOFS);
3776         return err;
3777 }
3778
3779 static int btrfs_setsize(struct inode *inode, struct iattr *attr)
3780 {
3781         struct btrfs_root *root = BTRFS_I(inode)->root;
3782         struct btrfs_trans_handle *trans;
3783         loff_t oldsize = i_size_read(inode);
3784         loff_t newsize = attr->ia_size;
3785         int mask = attr->ia_valid;
3786         int ret;
3787
3788         if (newsize == oldsize)
3789                 return 0;
3790
3791         /*
3792          * The regular truncate() case without ATTR_CTIME and ATTR_MTIME is a
3793          * special case where we need to update the times despite not having
3794          * these flags set.  For all other operations the VFS set these flags
3795          * explicitly if it wants a timestamp update.
3796          */
3797         if (newsize != oldsize && (!(mask & (ATTR_CTIME | ATTR_MTIME))))
3798                 inode->i_ctime = inode->i_mtime = current_fs_time(inode->i_sb);
3799
3800         if (newsize > oldsize) {
3801                 truncate_pagecache(inode, oldsize, newsize);
3802                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3803                 if (ret)
3804                         return ret;
3805
3806                 trans = btrfs_start_transaction(root, 1);
3807                 if (IS_ERR(trans))
3808                         return PTR_ERR(trans);
3809
3810                 i_size_write(inode, newsize);
3811                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3812                 ret = btrfs_update_inode(trans, root, inode);
3813                 btrfs_end_transaction(trans, root);
3814         } else {
3815
3816                 /*
3817                  * We're truncating a file that used to have good data down to
3818                  * zero. Make sure it gets into the ordered flush list so that
3819                  * any new writes get down to disk quickly.
3820                  */
3821                 if (newsize == 0)
3822                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3823                                 &BTRFS_I(inode)->runtime_flags);
3824
3825                 /*
3826                  * 1 for the orphan item we're going to add
3827                  * 1 for the orphan item deletion.
3828                  */
3829                 trans = btrfs_start_transaction(root, 2);
3830                 if (IS_ERR(trans))
3831                         return PTR_ERR(trans);
3832
3833                 /*
3834                  * We need to do this in case we fail at _any_ point during the
3835                  * actual truncate.  Once we do the truncate_setsize we could
3836                  * invalidate pages which forces any outstanding ordered io to
3837                  * be instantly completed which will give us extents that need
3838                  * to be truncated.  If we fail to get an orphan inode down we
3839                  * could have left over extents that were never meant to live,
3840                  * so we need to garuntee from this point on that everything
3841                  * will be consistent.
3842                  */
3843                 ret = btrfs_orphan_add(trans, inode);
3844                 btrfs_end_transaction(trans, root);
3845                 if (ret)
3846                         return ret;
3847
3848                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3849                 truncate_setsize(inode, newsize);
3850                 ret = btrfs_truncate(inode);
3851                 if (ret && inode->i_nlink)
3852                         btrfs_orphan_del(NULL, inode);
3853         }
3854
3855         return ret;
3856 }
3857
3858 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3859 {
3860         struct inode *inode = dentry->d_inode;
3861         struct btrfs_root *root = BTRFS_I(inode)->root;
3862         int err;
3863
3864         if (btrfs_root_readonly(root))
3865                 return -EROFS;
3866
3867         err = inode_change_ok(inode, attr);
3868         if (err)
3869                 return err;
3870
3871         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3872                 err = btrfs_setsize(inode, attr);
3873                 if (err)
3874                         return err;
3875         }
3876
3877         if (attr->ia_valid) {
3878                 setattr_copy(inode, attr);
3879                 inode_inc_iversion(inode);
3880                 err = btrfs_dirty_inode(inode);
3881
3882                 if (!err && attr->ia_valid & ATTR_MODE)
3883                         err = btrfs_acl_chmod(inode);
3884         }
3885
3886         return err;
3887 }
3888
3889 void btrfs_evict_inode(struct inode *inode)
3890 {
3891         struct btrfs_trans_handle *trans;
3892         struct btrfs_root *root = BTRFS_I(inode)->root;
3893         struct btrfs_block_rsv *rsv, *global_rsv;
3894         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3895         int ret;
3896
3897         trace_btrfs_inode_evict(inode);
3898
3899         truncate_inode_pages(&inode->i_data, 0);
3900         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3901                                btrfs_is_free_space_inode(inode)))
3902                 goto no_delete;
3903
3904         if (is_bad_inode(inode)) {
3905                 btrfs_orphan_del(NULL, inode);
3906                 goto no_delete;
3907         }
3908         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3909         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3910
3911         if (root->fs_info->log_root_recovering) {
3912                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3913                                  &BTRFS_I(inode)->runtime_flags));
3914                 goto no_delete;
3915         }
3916
3917         if (inode->i_nlink > 0) {
3918                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3919                 goto no_delete;
3920         }
3921
3922         ret = btrfs_commit_inode_delayed_inode(inode);
3923         if (ret) {
3924                 btrfs_orphan_del(NULL, inode);
3925                 goto no_delete;
3926         }
3927
3928         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3929         if (!rsv) {
3930                 btrfs_orphan_del(NULL, inode);
3931                 goto no_delete;
3932         }
3933         rsv->size = min_size;
3934         rsv->failfast = 1;
3935         global_rsv = &root->fs_info->global_block_rsv;
3936
3937         btrfs_i_size_write(inode, 0);
3938
3939         /*
3940          * This is a bit simpler than btrfs_truncate since we've already
3941          * reserved our space for our orphan item in the unlink, so we just
3942          * need to reserve some slack space in case we add bytes and update
3943          * inode item when doing the truncate.
3944          */
3945         while (1) {
3946                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3947                                              BTRFS_RESERVE_FLUSH_LIMIT);
3948
3949                 /*
3950                  * Try and steal from the global reserve since we will
3951                  * likely not use this space anyway, we want to try as
3952                  * hard as possible to get this to work.
3953                  */
3954                 if (ret)
3955                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3956
3957                 if (ret) {
3958                         printk(KERN_WARNING "Could not get space for a "
3959                                "delete, will truncate on mount %d\n", ret);
3960                         btrfs_orphan_del(NULL, inode);
3961                         btrfs_free_block_rsv(root, rsv);
3962                         goto no_delete;
3963                 }
3964
3965                 trans = btrfs_join_transaction(root);
3966                 if (IS_ERR(trans)) {
3967                         btrfs_orphan_del(NULL, inode);
3968                         btrfs_free_block_rsv(root, rsv);
3969                         goto no_delete;
3970                 }
3971
3972                 trans->block_rsv = rsv;
3973
3974                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3975                 if (ret != -ENOSPC)
3976                         break;
3977
3978                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3979                 btrfs_end_transaction(trans, root);
3980                 trans = NULL;
3981                 btrfs_btree_balance_dirty(root);
3982         }
3983
3984         btrfs_free_block_rsv(root, rsv);
3985
3986         if (ret == 0) {
3987                 trans->block_rsv = root->orphan_block_rsv;
3988                 ret = btrfs_orphan_del(trans, inode);
3989                 BUG_ON(ret);
3990         }
3991
3992         trans->block_rsv = &root->fs_info->trans_block_rsv;
3993         if (!(root == root->fs_info->tree_root ||
3994               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3995                 btrfs_return_ino(root, btrfs_ino(inode));
3996
3997         btrfs_end_transaction(trans, root);
3998         btrfs_btree_balance_dirty(root);
3999 no_delete:
4000         clear_inode(inode);
4001         return;
4002 }
4003
4004 /*
4005  * this returns the key found in the dir entry in the location pointer.
4006  * If no dir entries were found, location->objectid is 0.
4007  */
4008 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
4009                                struct btrfs_key *location)
4010 {
4011         const char *name = dentry->d_name.name;
4012         int namelen = dentry->d_name.len;
4013         struct btrfs_dir_item *di;
4014         struct btrfs_path *path;
4015         struct btrfs_root *root = BTRFS_I(dir)->root;
4016         int ret = 0;
4017
4018         path = btrfs_alloc_path();
4019         if (!path)
4020                 return -ENOMEM;
4021
4022         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
4023                                     namelen, 0);
4024         if (IS_ERR(di))
4025                 ret = PTR_ERR(di);
4026
4027         if (IS_ERR_OR_NULL(di))
4028                 goto out_err;
4029
4030         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
4031 out:
4032         btrfs_free_path(path);
4033         return ret;
4034 out_err:
4035         location->objectid = 0;
4036         goto out;
4037 }
4038
4039 /*
4040  * when we hit a tree root in a directory, the btrfs part of the inode
4041  * needs to be changed to reflect the root directory of the tree root.  This
4042  * is kind of like crossing a mount point.
4043  */
4044 static int fixup_tree_root_location(struct btrfs_root *root,
4045                                     struct inode *dir,
4046                                     struct dentry *dentry,
4047                                     struct btrfs_key *location,
4048                                     struct btrfs_root **sub_root)
4049 {
4050         struct btrfs_path *path;
4051         struct btrfs_root *new_root;
4052         struct btrfs_root_ref *ref;
4053         struct extent_buffer *leaf;
4054         int ret;
4055         int err = 0;
4056
4057         path = btrfs_alloc_path();
4058         if (!path) {
4059                 err = -ENOMEM;
4060                 goto out;
4061         }
4062
4063         err = -ENOENT;
4064         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
4065                                   BTRFS_I(dir)->root->root_key.objectid,
4066                                   location->objectid);
4067         if (ret) {
4068                 if (ret < 0)
4069                         err = ret;
4070                 goto out;
4071         }
4072
4073         leaf = path->nodes[0];
4074         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
4075         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
4076             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
4077                 goto out;
4078
4079         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
4080                                    (unsigned long)(ref + 1),
4081                                    dentry->d_name.len);
4082         if (ret)
4083                 goto out;
4084
4085         btrfs_release_path(path);
4086
4087         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4088         if (IS_ERR(new_root)) {
4089                 err = PTR_ERR(new_root);
4090                 goto out;
4091         }
4092
4093         if (btrfs_root_refs(&new_root->root_item) == 0) {
4094                 err = -ENOENT;
4095                 goto out;
4096         }
4097
4098         *sub_root = new_root;
4099         location->objectid = btrfs_root_dirid(&new_root->root_item);
4100         location->type = BTRFS_INODE_ITEM_KEY;
4101         location->offset = 0;
4102         err = 0;
4103 out:
4104         btrfs_free_path(path);
4105         return err;
4106 }
4107
4108 static void inode_tree_add(struct inode *inode)
4109 {
4110         struct btrfs_root *root = BTRFS_I(inode)->root;
4111         struct btrfs_inode *entry;
4112         struct rb_node **p;
4113         struct rb_node *parent;
4114         u64 ino = btrfs_ino(inode);
4115 again:
4116         p = &root->inode_tree.rb_node;
4117         parent = NULL;
4118
4119         if (inode_unhashed(inode))
4120                 return;
4121
4122         spin_lock(&root->inode_lock);
4123         while (*p) {
4124                 parent = *p;
4125                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4126
4127                 if (ino < btrfs_ino(&entry->vfs_inode))
4128                         p = &parent->rb_left;
4129                 else if (ino > btrfs_ino(&entry->vfs_inode))
4130                         p = &parent->rb_right;
4131                 else {
4132                         WARN_ON(!(entry->vfs_inode.i_state &
4133                                   (I_WILL_FREE | I_FREEING)));
4134                         rb_erase(parent, &root->inode_tree);
4135                         RB_CLEAR_NODE(parent);
4136                         spin_unlock(&root->inode_lock);
4137                         goto again;
4138                 }
4139         }
4140         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4141         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4142         spin_unlock(&root->inode_lock);
4143 }
4144
4145 static void inode_tree_del(struct inode *inode)
4146 {
4147         struct btrfs_root *root = BTRFS_I(inode)->root;
4148         int empty = 0;
4149
4150         spin_lock(&root->inode_lock);
4151         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4152                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4153                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4154                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4155         }
4156         spin_unlock(&root->inode_lock);
4157
4158         /*
4159          * Free space cache has inodes in the tree root, but the tree root has a
4160          * root_refs of 0, so this could end up dropping the tree root as a
4161          * snapshot, so we need the extra !root->fs_info->tree_root check to
4162          * make sure we don't drop it.
4163          */
4164         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4165             root != root->fs_info->tree_root) {
4166                 synchronize_srcu(&root->fs_info->subvol_srcu);
4167                 spin_lock(&root->inode_lock);
4168                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4169                 spin_unlock(&root->inode_lock);
4170                 if (empty)
4171                         btrfs_add_dead_root(root);
4172         }
4173 }
4174
4175 void btrfs_invalidate_inodes(struct btrfs_root *root)
4176 {
4177         struct rb_node *node;
4178         struct rb_node *prev;
4179         struct btrfs_inode *entry;
4180         struct inode *inode;
4181         u64 objectid = 0;
4182
4183         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4184
4185         spin_lock(&root->inode_lock);
4186 again:
4187         node = root->inode_tree.rb_node;
4188         prev = NULL;
4189         while (node) {
4190                 prev = node;
4191                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4192
4193                 if (objectid < btrfs_ino(&entry->vfs_inode))
4194                         node = node->rb_left;
4195                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4196                         node = node->rb_right;
4197                 else
4198                         break;
4199         }
4200         if (!node) {
4201                 while (prev) {
4202                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4203                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4204                                 node = prev;
4205                                 break;
4206                         }
4207                         prev = rb_next(prev);
4208                 }
4209         }
4210         while (node) {
4211                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4212                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4213                 inode = igrab(&entry->vfs_inode);
4214                 if (inode) {
4215                         spin_unlock(&root->inode_lock);
4216                         if (atomic_read(&inode->i_count) > 1)
4217                                 d_prune_aliases(inode);
4218                         /*
4219                          * btrfs_drop_inode will have it removed from
4220                          * the inode cache when its usage count
4221                          * hits zero.
4222                          */
4223                         iput(inode);
4224                         cond_resched();
4225                         spin_lock(&root->inode_lock);
4226                         goto again;
4227                 }
4228
4229                 if (cond_resched_lock(&root->inode_lock))
4230                         goto again;
4231
4232                 node = rb_next(node);
4233         }
4234         spin_unlock(&root->inode_lock);
4235 }
4236
4237 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4238 {
4239         struct btrfs_iget_args *args = p;
4240         inode->i_ino = args->ino;
4241         BTRFS_I(inode)->root = args->root;
4242         return 0;
4243 }
4244
4245 static int btrfs_find_actor(struct inode *inode, void *opaque)
4246 {
4247         struct btrfs_iget_args *args = opaque;
4248         return args->ino == btrfs_ino(inode) &&
4249                 args->root == BTRFS_I(inode)->root;
4250 }
4251
4252 static struct inode *btrfs_iget_locked(struct super_block *s,
4253                                        u64 objectid,
4254                                        struct btrfs_root *root)
4255 {
4256         struct inode *inode;
4257         struct btrfs_iget_args args;
4258         args.ino = objectid;
4259         args.root = root;
4260
4261         inode = iget5_locked(s, objectid, btrfs_find_actor,
4262                              btrfs_init_locked_inode,
4263                              (void *)&args);
4264         return inode;
4265 }
4266
4267 /* Get an inode object given its location and corresponding root.
4268  * Returns in *is_new if the inode was read from disk
4269  */
4270 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4271                          struct btrfs_root *root, int *new)
4272 {
4273         struct inode *inode;
4274
4275         inode = btrfs_iget_locked(s, location->objectid, root);
4276         if (!inode)
4277                 return ERR_PTR(-ENOMEM);
4278
4279         if (inode->i_state & I_NEW) {
4280                 BTRFS_I(inode)->root = root;
4281                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4282                 btrfs_read_locked_inode(inode);
4283                 if (!is_bad_inode(inode)) {
4284                         inode_tree_add(inode);
4285                         unlock_new_inode(inode);
4286                         if (new)
4287                                 *new = 1;
4288                 } else {
4289                         unlock_new_inode(inode);
4290                         iput(inode);
4291                         inode = ERR_PTR(-ESTALE);
4292                 }
4293         }
4294
4295         return inode;
4296 }
4297
4298 static struct inode *new_simple_dir(struct super_block *s,
4299                                     struct btrfs_key *key,
4300                                     struct btrfs_root *root)
4301 {
4302         struct inode *inode = new_inode(s);
4303
4304         if (!inode)
4305                 return ERR_PTR(-ENOMEM);
4306
4307         BTRFS_I(inode)->root = root;
4308         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4309         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4310
4311         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4312         inode->i_op = &btrfs_dir_ro_inode_operations;
4313         inode->i_fop = &simple_dir_operations;
4314         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4315         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4316
4317         return inode;
4318 }
4319
4320 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4321 {
4322         struct inode *inode;
4323         struct btrfs_root *root = BTRFS_I(dir)->root;
4324         struct btrfs_root *sub_root = root;
4325         struct btrfs_key location;
4326         int index;
4327         int ret = 0;
4328
4329         if (dentry->d_name.len > BTRFS_NAME_LEN)
4330                 return ERR_PTR(-ENAMETOOLONG);
4331
4332         if (unlikely(d_need_lookup(dentry))) {
4333                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4334                 kfree(dentry->d_fsdata);
4335                 dentry->d_fsdata = NULL;
4336                 /* This thing is hashed, drop it for now */
4337                 d_drop(dentry);
4338         } else {
4339                 ret = btrfs_inode_by_name(dir, dentry, &location);
4340         }
4341
4342         if (ret < 0)
4343                 return ERR_PTR(ret);
4344
4345         if (location.objectid == 0)
4346                 return NULL;
4347
4348         if (location.type == BTRFS_INODE_ITEM_KEY) {
4349                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4350                 return inode;
4351         }
4352
4353         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4354
4355         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4356         ret = fixup_tree_root_location(root, dir, dentry,
4357                                        &location, &sub_root);
4358         if (ret < 0) {
4359                 if (ret != -ENOENT)
4360                         inode = ERR_PTR(ret);
4361                 else
4362                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4363         } else {
4364                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4365         }
4366         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4367
4368         if (!IS_ERR(inode) && root != sub_root) {
4369                 down_read(&root->fs_info->cleanup_work_sem);
4370                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4371                         ret = btrfs_orphan_cleanup(sub_root);
4372                 up_read(&root->fs_info->cleanup_work_sem);
4373                 if (ret)
4374                         inode = ERR_PTR(ret);
4375         }
4376
4377         return inode;
4378 }
4379
4380 static int btrfs_dentry_delete(const struct dentry *dentry)
4381 {
4382         struct btrfs_root *root;
4383         struct inode *inode = dentry->d_inode;
4384
4385         if (!inode && !IS_ROOT(dentry))
4386                 inode = dentry->d_parent->d_inode;
4387
4388         if (inode) {
4389                 root = BTRFS_I(inode)->root;
4390                 if (btrfs_root_refs(&root->root_item) == 0)
4391                         return 1;
4392
4393                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4394                         return 1;
4395         }
4396         return 0;
4397 }
4398
4399 static void btrfs_dentry_release(struct dentry *dentry)
4400 {
4401         if (dentry->d_fsdata)
4402                 kfree(dentry->d_fsdata);
4403 }
4404
4405 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4406                                    unsigned int flags)
4407 {
4408         struct dentry *ret;
4409
4410         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4411         if (unlikely(d_need_lookup(dentry))) {
4412                 spin_lock(&dentry->d_lock);
4413                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4414                 spin_unlock(&dentry->d_lock);
4415         }
4416         return ret;
4417 }
4418
4419 unsigned char btrfs_filetype_table[] = {
4420         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4421 };
4422
4423 static int btrfs_real_readdir(struct file *filp, void *dirent,
4424                               filldir_t filldir)
4425 {
4426         struct inode *inode = filp->f_dentry->d_inode;
4427         struct btrfs_root *root = BTRFS_I(inode)->root;
4428         struct btrfs_item *item;
4429         struct btrfs_dir_item *di;
4430         struct btrfs_key key;
4431         struct btrfs_key found_key;
4432         struct btrfs_path *path;
4433         struct list_head ins_list;
4434         struct list_head del_list;
4435         int ret;
4436         struct extent_buffer *leaf;
4437         int slot;
4438         unsigned char d_type;
4439         int over = 0;
4440         u32 di_cur;
4441         u32 di_total;
4442         u32 di_len;
4443         int key_type = BTRFS_DIR_INDEX_KEY;
4444         char tmp_name[32];
4445         char *name_ptr;
4446         int name_len;
4447         int is_curr = 0;        /* filp->f_pos points to the current index? */
4448
4449         /* FIXME, use a real flag for deciding about the key type */
4450         if (root->fs_info->tree_root == root)
4451                 key_type = BTRFS_DIR_ITEM_KEY;
4452
4453         /* special case for "." */
4454         if (filp->f_pos == 0) {
4455                 over = filldir(dirent, ".", 1,
4456                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4457                 if (over)
4458                         return 0;
4459                 filp->f_pos = 1;
4460         }
4461         /* special case for .., just use the back ref */
4462         if (filp->f_pos == 1) {
4463                 u64 pino = parent_ino(filp->f_path.dentry);
4464                 over = filldir(dirent, "..", 2,
4465                                filp->f_pos, pino, DT_DIR);
4466                 if (over)
4467                         return 0;
4468                 filp->f_pos = 2;
4469         }
4470         path = btrfs_alloc_path();
4471         if (!path)
4472                 return -ENOMEM;
4473
4474         path->reada = 1;
4475
4476         if (key_type == BTRFS_DIR_INDEX_KEY) {
4477                 INIT_LIST_HEAD(&ins_list);
4478                 INIT_LIST_HEAD(&del_list);
4479                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4480         }
4481
4482         btrfs_set_key_type(&key, key_type);
4483         key.offset = filp->f_pos;
4484         key.objectid = btrfs_ino(inode);
4485
4486         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4487         if (ret < 0)
4488                 goto err;
4489
4490         while (1) {
4491                 leaf = path->nodes[0];
4492                 slot = path->slots[0];
4493                 if (slot >= btrfs_header_nritems(leaf)) {
4494                         ret = btrfs_next_leaf(root, path);
4495                         if (ret < 0)
4496                                 goto err;
4497                         else if (ret > 0)
4498                                 break;
4499                         continue;
4500                 }
4501
4502                 item = btrfs_item_nr(leaf, slot);
4503                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4504
4505                 if (found_key.objectid != key.objectid)
4506                         break;
4507                 if (btrfs_key_type(&found_key) != key_type)
4508                         break;
4509                 if (found_key.offset < filp->f_pos)
4510                         goto next;
4511                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4512                     btrfs_should_delete_dir_index(&del_list,
4513                                                   found_key.offset))
4514                         goto next;
4515
4516                 filp->f_pos = found_key.offset;
4517                 is_curr = 1;
4518
4519                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4520                 di_cur = 0;
4521                 di_total = btrfs_item_size(leaf, item);
4522
4523                 while (di_cur < di_total) {
4524                         struct btrfs_key location;
4525
4526                         if (verify_dir_item(root, leaf, di))
4527                                 break;
4528
4529                         name_len = btrfs_dir_name_len(leaf, di);
4530                         if (name_len <= sizeof(tmp_name)) {
4531                                 name_ptr = tmp_name;
4532                         } else {
4533                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4534                                 if (!name_ptr) {
4535                                         ret = -ENOMEM;
4536                                         goto err;
4537                                 }
4538                         }
4539                         read_extent_buffer(leaf, name_ptr,
4540                                            (unsigned long)(di + 1), name_len);
4541
4542                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4543                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4544
4545
4546                         /* is this a reference to our own snapshot? If so
4547                          * skip it.
4548                          *
4549                          * In contrast to old kernels, we insert the snapshot's
4550                          * dir item and dir index after it has been created, so
4551                          * we won't find a reference to our own snapshot. We
4552                          * still keep the following code for backward
4553                          * compatibility.
4554                          */
4555                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4556                             location.objectid == root->root_key.objectid) {
4557                                 over = 0;
4558                                 goto skip;
4559                         }
4560                         over = filldir(dirent, name_ptr, name_len,
4561                                        found_key.offset, location.objectid,
4562                                        d_type);
4563
4564 skip:
4565                         if (name_ptr != tmp_name)
4566                                 kfree(name_ptr);
4567
4568                         if (over)
4569                                 goto nopos;
4570                         di_len = btrfs_dir_name_len(leaf, di) +
4571                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4572                         di_cur += di_len;
4573                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4574                 }
4575 next:
4576                 path->slots[0]++;
4577         }
4578
4579         if (key_type == BTRFS_DIR_INDEX_KEY) {
4580                 if (is_curr)
4581                         filp->f_pos++;
4582                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4583                                                       &ins_list);
4584                 if (ret)
4585                         goto nopos;
4586         }
4587
4588         /* Reached end of directory/root. Bump pos past the last item. */
4589         if (key_type == BTRFS_DIR_INDEX_KEY)
4590                 /*
4591                  * 32-bit glibc will use getdents64, but then strtol -
4592                  * so the last number we can serve is this.
4593                  */
4594                 filp->f_pos = 0x7fffffff;
4595         else
4596                 filp->f_pos++;
4597 nopos:
4598         ret = 0;
4599 err:
4600         if (key_type == BTRFS_DIR_INDEX_KEY)
4601                 btrfs_put_delayed_items(&ins_list, &del_list);
4602         btrfs_free_path(path);
4603         return ret;
4604 }
4605
4606 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4607 {
4608         struct btrfs_root *root = BTRFS_I(inode)->root;
4609         struct btrfs_trans_handle *trans;
4610         int ret = 0;
4611         bool nolock = false;
4612
4613         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4614                 return 0;
4615
4616         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4617                 nolock = true;
4618
4619         if (wbc->sync_mode == WB_SYNC_ALL) {
4620                 if (nolock)
4621                         trans = btrfs_join_transaction_nolock(root);
4622                 else
4623                         trans = btrfs_join_transaction(root);
4624                 if (IS_ERR(trans))
4625                         return PTR_ERR(trans);
4626                 ret = btrfs_commit_transaction(trans, root);
4627         }
4628         return ret;
4629 }
4630
4631 /*
4632  * This is somewhat expensive, updating the tree every time the
4633  * inode changes.  But, it is most likely to find the inode in cache.
4634  * FIXME, needs more benchmarking...there are no reasons other than performance
4635  * to keep or drop this code.
4636  */
4637 int btrfs_dirty_inode(struct inode *inode)
4638 {
4639         struct btrfs_root *root = BTRFS_I(inode)->root;
4640         struct btrfs_trans_handle *trans;
4641         int ret;
4642
4643         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4644                 return 0;
4645
4646         trans = btrfs_join_transaction(root);
4647         if (IS_ERR(trans))
4648                 return PTR_ERR(trans);
4649
4650         ret = btrfs_update_inode(trans, root, inode);
4651         if (ret && ret == -ENOSPC) {
4652                 /* whoops, lets try again with the full transaction */
4653                 btrfs_end_transaction(trans, root);
4654                 trans = btrfs_start_transaction(root, 1);
4655                 if (IS_ERR(trans))
4656                         return PTR_ERR(trans);
4657
4658                 ret = btrfs_update_inode(trans, root, inode);
4659         }
4660         btrfs_end_transaction(trans, root);
4661         if (BTRFS_I(inode)->delayed_node)
4662                 btrfs_balance_delayed_items(root);
4663
4664         return ret;
4665 }
4666
4667 /*
4668  * This is a copy of file_update_time.  We need this so we can return error on
4669  * ENOSPC for updating the inode in the case of file write and mmap writes.
4670  */
4671 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4672                              int flags)
4673 {
4674         struct btrfs_root *root = BTRFS_I(inode)->root;
4675
4676         if (btrfs_root_readonly(root))
4677                 return -EROFS;
4678
4679         if (flags & S_VERSION)
4680                 inode_inc_iversion(inode);
4681         if (flags & S_CTIME)
4682                 inode->i_ctime = *now;
4683         if (flags & S_MTIME)
4684                 inode->i_mtime = *now;
4685         if (flags & S_ATIME)
4686                 inode->i_atime = *now;
4687         return btrfs_dirty_inode(inode);
4688 }
4689
4690 /*
4691  * find the highest existing sequence number in a directory
4692  * and then set the in-memory index_cnt variable to reflect
4693  * free sequence numbers
4694  */
4695 static int btrfs_set_inode_index_count(struct inode *inode)
4696 {
4697         struct btrfs_root *root = BTRFS_I(inode)->root;
4698         struct btrfs_key key, found_key;
4699         struct btrfs_path *path;
4700         struct extent_buffer *leaf;
4701         int ret;
4702
4703         key.objectid = btrfs_ino(inode);
4704         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4705         key.offset = (u64)-1;
4706
4707         path = btrfs_alloc_path();
4708         if (!path)
4709                 return -ENOMEM;
4710
4711         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4712         if (ret < 0)
4713                 goto out;
4714         /* FIXME: we should be able to handle this */
4715         if (ret == 0)
4716                 goto out;
4717         ret = 0;
4718
4719         /*
4720          * MAGIC NUMBER EXPLANATION:
4721          * since we search a directory based on f_pos we have to start at 2
4722          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4723          * else has to start at 2
4724          */
4725         if (path->slots[0] == 0) {
4726                 BTRFS_I(inode)->index_cnt = 2;
4727                 goto out;
4728         }
4729
4730         path->slots[0]--;
4731
4732         leaf = path->nodes[0];
4733         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4734
4735         if (found_key.objectid != btrfs_ino(inode) ||
4736             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4737                 BTRFS_I(inode)->index_cnt = 2;
4738                 goto out;
4739         }
4740
4741         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4742 out:
4743         btrfs_free_path(path);
4744         return ret;
4745 }
4746
4747 /*
4748  * helper to find a free sequence number in a given directory.  This current
4749  * code is very simple, later versions will do smarter things in the btree
4750  */
4751 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4752 {
4753         int ret = 0;
4754
4755         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4756                 ret = btrfs_inode_delayed_dir_index_count(dir);
4757                 if (ret) {
4758                         ret = btrfs_set_inode_index_count(dir);
4759                         if (ret)
4760                                 return ret;
4761                 }
4762         }
4763
4764         *index = BTRFS_I(dir)->index_cnt;
4765         BTRFS_I(dir)->index_cnt++;
4766
4767         return ret;
4768 }
4769
4770 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4771                                      struct btrfs_root *root,
4772                                      struct inode *dir,
4773                                      const char *name, int name_len,
4774                                      u64 ref_objectid, u64 objectid,
4775                                      umode_t mode, u64 *index)
4776 {
4777         struct inode *inode;
4778         struct btrfs_inode_item *inode_item;
4779         struct btrfs_key *location;
4780         struct btrfs_path *path;
4781         struct btrfs_inode_ref *ref;
4782         struct btrfs_key key[2];
4783         u32 sizes[2];
4784         unsigned long ptr;
4785         int ret;
4786         int owner;
4787
4788         path = btrfs_alloc_path();
4789         if (!path)
4790                 return ERR_PTR(-ENOMEM);
4791
4792         inode = new_inode(root->fs_info->sb);
4793         if (!inode) {
4794                 btrfs_free_path(path);
4795                 return ERR_PTR(-ENOMEM);
4796         }
4797
4798         /*
4799          * we have to initialize this early, so we can reclaim the inode
4800          * number if we fail afterwards in this function.
4801          */
4802         inode->i_ino = objectid;
4803
4804         if (dir) {
4805                 trace_btrfs_inode_request(dir);
4806
4807                 ret = btrfs_set_inode_index(dir, index);
4808                 if (ret) {
4809                         btrfs_free_path(path);
4810                         iput(inode);
4811                         return ERR_PTR(ret);
4812                 }
4813         }
4814         /*
4815          * index_cnt is ignored for everything but a dir,
4816          * btrfs_get_inode_index_count has an explanation for the magic
4817          * number
4818          */
4819         BTRFS_I(inode)->index_cnt = 2;
4820         BTRFS_I(inode)->root = root;
4821         BTRFS_I(inode)->generation = trans->transid;
4822         inode->i_generation = BTRFS_I(inode)->generation;
4823
4824         /*
4825          * We could have gotten an inode number from somebody who was fsynced
4826          * and then removed in this same transaction, so let's just set full
4827          * sync since it will be a full sync anyway and this will blow away the
4828          * old info in the log.
4829          */
4830         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4831
4832         if (S_ISDIR(mode))
4833                 owner = 0;
4834         else
4835                 owner = 1;
4836
4837         key[0].objectid = objectid;
4838         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4839         key[0].offset = 0;
4840
4841         /*
4842          * Start new inodes with an inode_ref. This is slightly more
4843          * efficient for small numbers of hard links since they will
4844          * be packed into one item. Extended refs will kick in if we
4845          * add more hard links than can fit in the ref item.
4846          */
4847         key[1].objectid = objectid;
4848         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4849         key[1].offset = ref_objectid;
4850
4851         sizes[0] = sizeof(struct btrfs_inode_item);
4852         sizes[1] = name_len + sizeof(*ref);
4853
4854         path->leave_spinning = 1;
4855         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4856         if (ret != 0)
4857                 goto fail;
4858
4859         inode_init_owner(inode, dir, mode);
4860         inode_set_bytes(inode, 0);
4861         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4862         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4863                                   struct btrfs_inode_item);
4864         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4865                              sizeof(*inode_item));
4866         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4867
4868         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4869                              struct btrfs_inode_ref);
4870         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4871         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4872         ptr = (unsigned long)(ref + 1);
4873         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4874
4875         btrfs_mark_buffer_dirty(path->nodes[0]);
4876         btrfs_free_path(path);
4877
4878         location = &BTRFS_I(inode)->location;
4879         location->objectid = objectid;
4880         location->offset = 0;
4881         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4882
4883         btrfs_inherit_iflags(inode, dir);
4884
4885         if (S_ISREG(mode)) {
4886                 if (btrfs_test_opt(root, NODATASUM))
4887                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4888                 if (btrfs_test_opt(root, NODATACOW))
4889                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4890         }
4891
4892         insert_inode_hash(inode);
4893         inode_tree_add(inode);
4894
4895         trace_btrfs_inode_new(inode);
4896         btrfs_set_inode_last_trans(trans, inode);
4897
4898         btrfs_update_root_times(trans, root);
4899
4900         return inode;
4901 fail:
4902         if (dir)
4903                 BTRFS_I(dir)->index_cnt--;
4904         btrfs_free_path(path);
4905         iput(inode);
4906         return ERR_PTR(ret);
4907 }
4908
4909 static inline u8 btrfs_inode_type(struct inode *inode)
4910 {
4911         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4912 }
4913
4914 /*
4915  * utility function to add 'inode' into 'parent_inode' with
4916  * a give name and a given sequence number.
4917  * if 'add_backref' is true, also insert a backref from the
4918  * inode to the parent directory.
4919  */
4920 int btrfs_add_link(struct btrfs_trans_handle *trans,
4921                    struct inode *parent_inode, struct inode *inode,
4922                    const char *name, int name_len, int add_backref, u64 index)
4923 {
4924         int ret = 0;
4925         struct btrfs_key key;
4926         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4927         u64 ino = btrfs_ino(inode);
4928         u64 parent_ino = btrfs_ino(parent_inode);
4929
4930         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4931                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4932         } else {
4933                 key.objectid = ino;
4934                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4935                 key.offset = 0;
4936         }
4937
4938         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4939                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4940                                          key.objectid, root->root_key.objectid,
4941                                          parent_ino, index, name, name_len);
4942         } else if (add_backref) {
4943                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4944                                              parent_ino, index);
4945         }
4946
4947         /* Nothing to clean up yet */
4948         if (ret)
4949                 return ret;
4950
4951         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4952                                     parent_inode, &key,
4953                                     btrfs_inode_type(inode), index);
4954         if (ret == -EEXIST || ret == -EOVERFLOW)
4955                 goto fail_dir_item;
4956         else if (ret) {
4957                 btrfs_abort_transaction(trans, root, ret);
4958                 return ret;
4959         }
4960
4961         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4962                            name_len * 2);
4963         inode_inc_iversion(parent_inode);
4964         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4965         ret = btrfs_update_inode(trans, root, parent_inode);
4966         if (ret)
4967                 btrfs_abort_transaction(trans, root, ret);
4968         return ret;
4969
4970 fail_dir_item:
4971         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4972                 u64 local_index;
4973                 int err;
4974                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4975                                  key.objectid, root->root_key.objectid,
4976                                  parent_ino, &local_index, name, name_len);
4977
4978         } else if (add_backref) {
4979                 u64 local_index;
4980                 int err;
4981
4982                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4983                                           ino, parent_ino, &local_index);
4984         }
4985         return ret;
4986 }
4987
4988 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4989                             struct inode *dir, struct dentry *dentry,
4990                             struct inode *inode, int backref, u64 index)
4991 {
4992         int err = btrfs_add_link(trans, dir, inode,
4993                                  dentry->d_name.name, dentry->d_name.len,
4994                                  backref, index);
4995         if (err > 0)
4996                 err = -EEXIST;
4997         return err;
4998 }
4999
5000 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
5001                         umode_t mode, dev_t rdev)
5002 {
5003         struct btrfs_trans_handle *trans;
5004         struct btrfs_root *root = BTRFS_I(dir)->root;
5005         struct inode *inode = NULL;
5006         int err;
5007         int drop_inode = 0;
5008         u64 objectid;
5009         u64 index = 0;
5010
5011         if (!new_valid_dev(rdev))
5012                 return -EINVAL;
5013
5014         /*
5015          * 2 for inode item and ref
5016          * 2 for dir items
5017          * 1 for xattr if selinux is on
5018          */
5019         trans = btrfs_start_transaction(root, 5);
5020         if (IS_ERR(trans))
5021                 return PTR_ERR(trans);
5022
5023         err = btrfs_find_free_ino(root, &objectid);
5024         if (err)
5025                 goto out_unlock;
5026
5027         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5028                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5029                                 mode, &index);
5030         if (IS_ERR(inode)) {
5031                 err = PTR_ERR(inode);
5032                 goto out_unlock;
5033         }
5034
5035         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5036         if (err) {
5037                 drop_inode = 1;
5038                 goto out_unlock;
5039         }
5040
5041         /*
5042         * If the active LSM wants to access the inode during
5043         * d_instantiate it needs these. Smack checks to see
5044         * if the filesystem supports xattrs by looking at the
5045         * ops vector.
5046         */
5047
5048         inode->i_op = &btrfs_special_inode_operations;
5049         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5050         if (err)
5051                 drop_inode = 1;
5052         else {
5053                 init_special_inode(inode, inode->i_mode, rdev);
5054                 btrfs_update_inode(trans, root, inode);
5055                 d_instantiate(dentry, inode);
5056         }
5057 out_unlock:
5058         btrfs_end_transaction(trans, root);
5059         btrfs_btree_balance_dirty(root);
5060         if (drop_inode) {
5061                 inode_dec_link_count(inode);
5062                 iput(inode);
5063         }
5064         return err;
5065 }
5066
5067 static int btrfs_create(struct inode *dir, struct dentry *dentry,
5068                         umode_t mode, bool excl)
5069 {
5070         struct btrfs_trans_handle *trans;
5071         struct btrfs_root *root = BTRFS_I(dir)->root;
5072         struct inode *inode = NULL;
5073         int drop_inode_on_err = 0;
5074         int err;
5075         u64 objectid;
5076         u64 index = 0;
5077
5078         /*
5079          * 2 for inode item and ref
5080          * 2 for dir items
5081          * 1 for xattr if selinux is on
5082          */
5083         trans = btrfs_start_transaction(root, 5);
5084         if (IS_ERR(trans))
5085                 return PTR_ERR(trans);
5086
5087         err = btrfs_find_free_ino(root, &objectid);
5088         if (err)
5089                 goto out_unlock;
5090
5091         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5092                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5093                                 mode, &index);
5094         if (IS_ERR(inode)) {
5095                 err = PTR_ERR(inode);
5096                 goto out_unlock;
5097         }
5098         drop_inode_on_err = 1;
5099
5100         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5101         if (err)
5102                 goto out_unlock;
5103
5104         err = btrfs_update_inode(trans, root, inode);
5105         if (err)
5106                 goto out_unlock;
5107
5108         /*
5109         * If the active LSM wants to access the inode during
5110         * d_instantiate it needs these. Smack checks to see
5111         * if the filesystem supports xattrs by looking at the
5112         * ops vector.
5113         */
5114         inode->i_fop = &btrfs_file_operations;
5115         inode->i_op = &btrfs_file_inode_operations;
5116
5117         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5118         if (err)
5119                 goto out_unlock;
5120
5121         inode->i_mapping->a_ops = &btrfs_aops;
5122         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5123         BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5124         d_instantiate(dentry, inode);
5125
5126 out_unlock:
5127         btrfs_end_transaction(trans, root);
5128         if (err && drop_inode_on_err) {
5129                 inode_dec_link_count(inode);
5130                 iput(inode);
5131         }
5132         btrfs_btree_balance_dirty(root);
5133         return err;
5134 }
5135
5136 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5137                       struct dentry *dentry)
5138 {
5139         struct btrfs_trans_handle *trans;
5140         struct btrfs_root *root = BTRFS_I(dir)->root;
5141         struct inode *inode = old_dentry->d_inode;
5142         u64 index;
5143         int err;
5144         int drop_inode = 0;
5145
5146         /* do not allow sys_link's with other subvols of the same device */
5147         if (root->objectid != BTRFS_I(inode)->root->objectid)
5148                 return -EXDEV;
5149
5150         if (inode->i_nlink >= BTRFS_LINK_MAX)
5151                 return -EMLINK;
5152
5153         err = btrfs_set_inode_index(dir, &index);
5154         if (err)
5155                 goto fail;
5156
5157         /*
5158          * 2 items for inode and inode ref
5159          * 2 items for dir items
5160          * 1 item for parent inode
5161          */
5162         trans = btrfs_start_transaction(root, 5);
5163         if (IS_ERR(trans)) {
5164                 err = PTR_ERR(trans);
5165                 goto fail;
5166         }
5167
5168         btrfs_inc_nlink(inode);
5169         inode_inc_iversion(inode);
5170         inode->i_ctime = CURRENT_TIME;
5171         ihold(inode);
5172         set_bit(BTRFS_INODE_COPY_EVERYTHING, &BTRFS_I(inode)->runtime_flags);
5173
5174         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5175
5176         if (err) {
5177                 drop_inode = 1;
5178         } else {
5179                 struct dentry *parent = dentry->d_parent;
5180                 err = btrfs_update_inode(trans, root, inode);
5181                 if (err)
5182                         goto fail;
5183                 d_instantiate(dentry, inode);
5184                 btrfs_log_new_name(trans, inode, NULL, parent);
5185         }
5186
5187         btrfs_end_transaction(trans, root);
5188 fail:
5189         if (drop_inode) {
5190                 inode_dec_link_count(inode);
5191                 iput(inode);
5192         }
5193         btrfs_btree_balance_dirty(root);
5194         return err;
5195 }
5196
5197 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5198 {
5199         struct inode *inode = NULL;
5200         struct btrfs_trans_handle *trans;
5201         struct btrfs_root *root = BTRFS_I(dir)->root;
5202         int err = 0;
5203         int drop_on_err = 0;
5204         u64 objectid = 0;
5205         u64 index = 0;
5206
5207         /*
5208          * 2 items for inode and ref
5209          * 2 items for dir items
5210          * 1 for xattr if selinux is on
5211          */
5212         trans = btrfs_start_transaction(root, 5);
5213         if (IS_ERR(trans))
5214                 return PTR_ERR(trans);
5215
5216         err = btrfs_find_free_ino(root, &objectid);
5217         if (err)
5218                 goto out_fail;
5219
5220         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5221                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5222                                 S_IFDIR | mode, &index);
5223         if (IS_ERR(inode)) {
5224                 err = PTR_ERR(inode);
5225                 goto out_fail;
5226         }
5227
5228         drop_on_err = 1;
5229
5230         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5231         if (err)
5232                 goto out_fail;
5233
5234         inode->i_op = &btrfs_dir_inode_operations;
5235         inode->i_fop = &btrfs_dir_file_operations;
5236
5237         btrfs_i_size_write(inode, 0);
5238         err = btrfs_update_inode(trans, root, inode);
5239         if (err)
5240                 goto out_fail;
5241
5242         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5243                              dentry->d_name.len, 0, index);
5244         if (err)
5245                 goto out_fail;
5246
5247         d_instantiate(dentry, inode);
5248         drop_on_err = 0;
5249
5250 out_fail:
5251         btrfs_end_transaction(trans, root);
5252         if (drop_on_err)
5253                 iput(inode);
5254         btrfs_btree_balance_dirty(root);
5255         return err;
5256 }
5257
5258 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5259  * and an extent that you want to insert, deal with overlap and insert
5260  * the new extent into the tree.
5261  */
5262 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5263                                 struct extent_map *existing,
5264                                 struct extent_map *em,
5265                                 u64 map_start, u64 map_len)
5266 {
5267         u64 start_diff;
5268
5269         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5270         start_diff = map_start - em->start;
5271         em->start = map_start;
5272         em->len = map_len;
5273         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5274             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5275                 em->block_start += start_diff;
5276                 em->block_len -= start_diff;
5277         }
5278         return add_extent_mapping(em_tree, em);
5279 }
5280
5281 static noinline int uncompress_inline(struct btrfs_path *path,
5282                                       struct inode *inode, struct page *page,
5283                                       size_t pg_offset, u64 extent_offset,
5284                                       struct btrfs_file_extent_item *item)
5285 {
5286         int ret;
5287         struct extent_buffer *leaf = path->nodes[0];
5288         char *tmp;
5289         size_t max_size;
5290         unsigned long inline_size;
5291         unsigned long ptr;
5292         int compress_type;
5293
5294         WARN_ON(pg_offset != 0);
5295         compress_type = btrfs_file_extent_compression(leaf, item);
5296         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5297         inline_size = btrfs_file_extent_inline_item_len(leaf,
5298                                         btrfs_item_nr(leaf, path->slots[0]));
5299         tmp = kmalloc(inline_size, GFP_NOFS);
5300         if (!tmp)
5301                 return -ENOMEM;
5302         ptr = btrfs_file_extent_inline_start(item);
5303
5304         read_extent_buffer(leaf, tmp, ptr, inline_size);
5305
5306         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5307         ret = btrfs_decompress(compress_type, tmp, page,
5308                                extent_offset, inline_size, max_size);
5309         if (ret) {
5310                 char *kaddr = kmap_atomic(page);
5311                 unsigned long copy_size = min_t(u64,
5312                                   PAGE_CACHE_SIZE - pg_offset,
5313                                   max_size - extent_offset);
5314                 memset(kaddr + pg_offset, 0, copy_size);
5315                 kunmap_atomic(kaddr);
5316         }
5317         kfree(tmp);
5318         return 0;
5319 }
5320
5321 /*
5322  * a bit scary, this does extent mapping from logical file offset to the disk.
5323  * the ugly parts come from merging extents from the disk with the in-ram
5324  * representation.  This gets more complex because of the data=ordered code,
5325  * where the in-ram extents might be locked pending data=ordered completion.
5326  *
5327  * This also copies inline extents directly into the page.
5328  */
5329
5330 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5331                                     size_t pg_offset, u64 start, u64 len,
5332                                     int create)
5333 {
5334         int ret;
5335         int err = 0;
5336         u64 bytenr;
5337         u64 extent_start = 0;
5338         u64 extent_end = 0;
5339         u64 objectid = btrfs_ino(inode);
5340         u32 found_type;
5341         struct btrfs_path *path = NULL;
5342         struct btrfs_root *root = BTRFS_I(inode)->root;
5343         struct btrfs_file_extent_item *item;
5344         struct extent_buffer *leaf;
5345         struct btrfs_key found_key;
5346         struct extent_map *em = NULL;
5347         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5348         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5349         struct btrfs_trans_handle *trans = NULL;
5350         int compress_type;
5351
5352 again:
5353         read_lock(&em_tree->lock);
5354         em = lookup_extent_mapping(em_tree, start, len);
5355         if (em)
5356                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5357         read_unlock(&em_tree->lock);
5358
5359         if (em) {
5360                 if (em->start > start || em->start + em->len <= start)
5361                         free_extent_map(em);
5362                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5363                         free_extent_map(em);
5364                 else
5365                         goto out;
5366         }
5367         em = alloc_extent_map();
5368         if (!em) {
5369                 err = -ENOMEM;
5370                 goto out;
5371         }
5372         em->bdev = root->fs_info->fs_devices->latest_bdev;
5373         em->start = EXTENT_MAP_HOLE;
5374         em->orig_start = EXTENT_MAP_HOLE;
5375         em->len = (u64)-1;
5376         em->block_len = (u64)-1;
5377
5378         if (!path) {
5379                 path = btrfs_alloc_path();
5380                 if (!path) {
5381                         err = -ENOMEM;
5382                         goto out;
5383                 }
5384                 /*
5385                  * Chances are we'll be called again, so go ahead and do
5386                  * readahead
5387                  */
5388                 path->reada = 1;
5389         }
5390
5391         ret = btrfs_lookup_file_extent(trans, root, path,
5392                                        objectid, start, trans != NULL);
5393         if (ret < 0) {
5394                 err = ret;
5395                 goto out;
5396         }
5397
5398         if (ret != 0) {
5399                 if (path->slots[0] == 0)
5400                         goto not_found;
5401                 path->slots[0]--;
5402         }
5403
5404         leaf = path->nodes[0];
5405         item = btrfs_item_ptr(leaf, path->slots[0],
5406                               struct btrfs_file_extent_item);
5407         /* are we inside the extent that was found? */
5408         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5409         found_type = btrfs_key_type(&found_key);
5410         if (found_key.objectid != objectid ||
5411             found_type != BTRFS_EXTENT_DATA_KEY) {
5412                 goto not_found;
5413         }
5414
5415         found_type = btrfs_file_extent_type(leaf, item);
5416         extent_start = found_key.offset;
5417         compress_type = btrfs_file_extent_compression(leaf, item);
5418         if (found_type == BTRFS_FILE_EXTENT_REG ||
5419             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5420                 extent_end = extent_start +
5421                        btrfs_file_extent_num_bytes(leaf, item);
5422         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5423                 size_t size;
5424                 size = btrfs_file_extent_inline_len(leaf, item);
5425                 extent_end = (extent_start + size + root->sectorsize - 1) &
5426                         ~((u64)root->sectorsize - 1);
5427         }
5428
5429         if (start >= extent_end) {
5430                 path->slots[0]++;
5431                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5432                         ret = btrfs_next_leaf(root, path);
5433                         if (ret < 0) {
5434                                 err = ret;
5435                                 goto out;
5436                         }
5437                         if (ret > 0)
5438                                 goto not_found;
5439                         leaf = path->nodes[0];
5440                 }
5441                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5442                 if (found_key.objectid != objectid ||
5443                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5444                         goto not_found;
5445                 if (start + len <= found_key.offset)
5446                         goto not_found;
5447                 em->start = start;
5448                 em->orig_start = start;
5449                 em->len = found_key.offset - start;
5450                 goto not_found_em;
5451         }
5452
5453         if (found_type == BTRFS_FILE_EXTENT_REG ||
5454             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5455                 em->start = extent_start;
5456                 em->len = extent_end - extent_start;
5457                 em->orig_start = extent_start -
5458                                  btrfs_file_extent_offset(leaf, item);
5459                 em->orig_block_len = btrfs_file_extent_disk_num_bytes(leaf,
5460                                                                       item);
5461                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5462                 if (bytenr == 0) {
5463                         em->block_start = EXTENT_MAP_HOLE;
5464                         goto insert;
5465                 }
5466                 if (compress_type != BTRFS_COMPRESS_NONE) {
5467                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5468                         em->compress_type = compress_type;
5469                         em->block_start = bytenr;
5470                         em->block_len = em->orig_block_len;
5471                 } else {
5472                         bytenr += btrfs_file_extent_offset(leaf, item);
5473                         em->block_start = bytenr;
5474                         em->block_len = em->len;
5475                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5476                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5477                 }
5478                 goto insert;
5479         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5480                 unsigned long ptr;
5481                 char *map;
5482                 size_t size;
5483                 size_t extent_offset;
5484                 size_t copy_size;
5485
5486                 em->block_start = EXTENT_MAP_INLINE;
5487                 if (!page || create) {
5488                         em->start = extent_start;
5489                         em->len = extent_end - extent_start;
5490                         goto out;
5491                 }
5492
5493                 size = btrfs_file_extent_inline_len(leaf, item);
5494                 extent_offset = page_offset(page) + pg_offset - extent_start;
5495                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5496                                 size - extent_offset);
5497                 em->start = extent_start + extent_offset;
5498                 em->len = (copy_size + root->sectorsize - 1) &
5499                         ~((u64)root->sectorsize - 1);
5500                 em->orig_block_len = em->len;
5501                 em->orig_start = em->start;
5502                 if (compress_type) {
5503                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5504                         em->compress_type = compress_type;
5505                 }
5506                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5507                 if (create == 0 && !PageUptodate(page)) {
5508                         if (btrfs_file_extent_compression(leaf, item) !=
5509                             BTRFS_COMPRESS_NONE) {
5510                                 ret = uncompress_inline(path, inode, page,
5511                                                         pg_offset,
5512                                                         extent_offset, item);
5513                                 BUG_ON(ret); /* -ENOMEM */
5514                         } else {
5515                                 map = kmap(page);
5516                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5517                                                    copy_size);
5518                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5519                                         memset(map + pg_offset + copy_size, 0,
5520                                                PAGE_CACHE_SIZE - pg_offset -
5521                                                copy_size);
5522                                 }
5523                                 kunmap(page);
5524                         }
5525                         flush_dcache_page(page);
5526                 } else if (create && PageUptodate(page)) {
5527                         BUG();
5528                         if (!trans) {
5529                                 kunmap(page);
5530                                 free_extent_map(em);
5531                                 em = NULL;
5532
5533                                 btrfs_release_path(path);
5534                                 trans = btrfs_join_transaction(root);
5535
5536                                 if (IS_ERR(trans))
5537                                         return ERR_CAST(trans);
5538                                 goto again;
5539                         }
5540                         map = kmap(page);
5541                         write_extent_buffer(leaf, map + pg_offset, ptr,
5542                                             copy_size);
5543                         kunmap(page);
5544                         btrfs_mark_buffer_dirty(leaf);
5545                 }
5546                 set_extent_uptodate(io_tree, em->start,
5547                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5548                 goto insert;
5549         } else {
5550                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5551         }
5552 not_found:
5553         em->start = start;
5554         em->orig_start = start;
5555         em->len = len;
5556 not_found_em:
5557         em->block_start = EXTENT_MAP_HOLE;
5558         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5559 insert:
5560         btrfs_release_path(path);
5561         if (em->start > start || extent_map_end(em) <= start) {
5562                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5563                        "[%llu %llu]\n", (unsigned long long)em->start,
5564                        (unsigned long long)em->len,
5565                        (unsigned long long)start,
5566                        (unsigned long long)len);
5567                 err = -EIO;
5568                 goto out;
5569         }
5570
5571         err = 0;
5572         write_lock(&em_tree->lock);
5573         ret = add_extent_mapping(em_tree, em);
5574         /* it is possible that someone inserted the extent into the tree
5575          * while we had the lock dropped.  It is also possible that
5576          * an overlapping map exists in the tree
5577          */
5578         if (ret == -EEXIST) {
5579                 struct extent_map *existing;
5580
5581                 ret = 0;
5582
5583                 existing = lookup_extent_mapping(em_tree, start, len);
5584                 if (existing && (existing->start > start ||
5585                     existing->start + existing->len <= start)) {
5586                         free_extent_map(existing);
5587                         existing = NULL;
5588                 }
5589                 if (!existing) {
5590                         existing = lookup_extent_mapping(em_tree, em->start,
5591                                                          em->len);
5592                         if (existing) {
5593                                 err = merge_extent_mapping(em_tree, existing,
5594                                                            em, start,
5595                                                            root->sectorsize);
5596                                 free_extent_map(existing);
5597                                 if (err) {
5598                                         free_extent_map(em);
5599                                         em = NULL;
5600                                 }
5601                         } else {
5602                                 err = -EIO;
5603                                 free_extent_map(em);
5604                                 em = NULL;
5605                         }
5606                 } else {
5607                         free_extent_map(em);
5608                         em = existing;
5609                         err = 0;
5610                 }
5611         }
5612         write_unlock(&em_tree->lock);
5613 out:
5614
5615         if (em)
5616                 trace_btrfs_get_extent(root, em);
5617
5618         if (path)
5619                 btrfs_free_path(path);
5620         if (trans) {
5621                 ret = btrfs_end_transaction(trans, root);
5622                 if (!err)
5623                         err = ret;
5624         }
5625         if (err) {
5626                 free_extent_map(em);
5627                 return ERR_PTR(err);
5628         }
5629         BUG_ON(!em); /* Error is always set */
5630         return em;
5631 }
5632
5633 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5634                                            size_t pg_offset, u64 start, u64 len,
5635                                            int create)
5636 {
5637         struct extent_map *em;
5638         struct extent_map *hole_em = NULL;
5639         u64 range_start = start;
5640         u64 end;
5641         u64 found;
5642         u64 found_end;
5643         int err = 0;
5644
5645         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5646         if (IS_ERR(em))
5647                 return em;
5648         if (em) {
5649                 /*
5650                  * if our em maps to
5651                  * -  a hole or
5652                  * -  a pre-alloc extent,
5653                  * there might actually be delalloc bytes behind it.
5654                  */
5655                 if (em->block_start != EXTENT_MAP_HOLE &&
5656                     !test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
5657                         return em;
5658                 else
5659                         hole_em = em;
5660         }
5661
5662         /* check to see if we've wrapped (len == -1 or similar) */
5663         end = start + len;
5664         if (end < start)
5665                 end = (u64)-1;
5666         else
5667                 end -= 1;
5668
5669         em = NULL;
5670
5671         /* ok, we didn't find anything, lets look for delalloc */
5672         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5673                                  end, len, EXTENT_DELALLOC, 1);
5674         found_end = range_start + found;
5675         if (found_end < range_start)
5676                 found_end = (u64)-1;
5677
5678         /*
5679          * we didn't find anything useful, return
5680          * the original results from get_extent()
5681          */
5682         if (range_start > end || found_end <= start) {
5683                 em = hole_em;
5684                 hole_em = NULL;
5685                 goto out;
5686         }
5687
5688         /* adjust the range_start to make sure it doesn't
5689          * go backwards from the start they passed in
5690          */
5691         range_start = max(start,range_start);
5692         found = found_end - range_start;
5693
5694         if (found > 0) {
5695                 u64 hole_start = start;
5696                 u64 hole_len = len;
5697
5698                 em = alloc_extent_map();
5699                 if (!em) {
5700                         err = -ENOMEM;
5701                         goto out;
5702                 }
5703                 /*
5704                  * when btrfs_get_extent can't find anything it
5705                  * returns one huge hole
5706                  *
5707                  * make sure what it found really fits our range, and
5708                  * adjust to make sure it is based on the start from
5709                  * the caller
5710                  */
5711                 if (hole_em) {
5712                         u64 calc_end = extent_map_end(hole_em);
5713
5714                         if (calc_end <= start || (hole_em->start > end)) {
5715                                 free_extent_map(hole_em);
5716                                 hole_em = NULL;
5717                         } else {
5718                                 hole_start = max(hole_em->start, start);
5719                                 hole_len = calc_end - hole_start;
5720                         }
5721                 }
5722                 em->bdev = NULL;
5723                 if (hole_em && range_start > hole_start) {
5724                         /* our hole starts before our delalloc, so we
5725                          * have to return just the parts of the hole
5726                          * that go until  the delalloc starts
5727                          */
5728                         em->len = min(hole_len,
5729                                       range_start - hole_start);
5730                         em->start = hole_start;
5731                         em->orig_start = hole_start;
5732                         /*
5733                          * don't adjust block start at all,
5734                          * it is fixed at EXTENT_MAP_HOLE
5735                          */
5736                         em->block_start = hole_em->block_start;
5737                         em->block_len = hole_len;
5738                         if (test_bit(EXTENT_FLAG_PREALLOC, &hole_em->flags))
5739                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5740                 } else {
5741                         em->start = range_start;
5742                         em->len = found;
5743                         em->orig_start = range_start;
5744                         em->block_start = EXTENT_MAP_DELALLOC;
5745                         em->block_len = found;
5746                 }
5747         } else if (hole_em) {
5748                 return hole_em;
5749         }
5750 out:
5751
5752         free_extent_map(hole_em);
5753         if (err) {
5754                 free_extent_map(em);
5755                 return ERR_PTR(err);
5756         }
5757         return em;
5758 }
5759
5760 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5761                                                   u64 start, u64 len)
5762 {
5763         struct btrfs_root *root = BTRFS_I(inode)->root;
5764         struct btrfs_trans_handle *trans;
5765         struct extent_map *em;
5766         struct btrfs_key ins;
5767         u64 alloc_hint;
5768         int ret;
5769
5770         trans = btrfs_join_transaction(root);
5771         if (IS_ERR(trans))
5772                 return ERR_CAST(trans);
5773
5774         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5775
5776         alloc_hint = get_extent_allocation_hint(inode, start, len);
5777         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5778                                    alloc_hint, &ins, 1);
5779         if (ret) {
5780                 em = ERR_PTR(ret);
5781                 goto out;
5782         }
5783
5784         em = create_pinned_em(inode, start, ins.offset, start, ins.objectid,
5785                               ins.offset, ins.offset, 0);
5786         if (IS_ERR(em))
5787                 goto out;
5788
5789         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5790                                            ins.offset, ins.offset, 0);
5791         if (ret) {
5792                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5793                 em = ERR_PTR(ret);
5794         }
5795 out:
5796         btrfs_end_transaction(trans, root);
5797         return em;
5798 }
5799
5800 /*
5801  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5802  * block must be cow'd
5803  */
5804 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5805                                       struct inode *inode, u64 offset, u64 len)
5806 {
5807         struct btrfs_path *path;
5808         int ret;
5809         struct extent_buffer *leaf;
5810         struct btrfs_root *root = BTRFS_I(inode)->root;
5811         struct btrfs_file_extent_item *fi;
5812         struct btrfs_key key;
5813         u64 disk_bytenr;
5814         u64 backref_offset;
5815         u64 extent_end;
5816         u64 num_bytes;
5817         int slot;
5818         int found_type;
5819
5820         path = btrfs_alloc_path();
5821         if (!path)
5822                 return -ENOMEM;
5823
5824         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5825                                        offset, 0);
5826         if (ret < 0)
5827                 goto out;
5828
5829         slot = path->slots[0];
5830         if (ret == 1) {
5831                 if (slot == 0) {
5832                         /* can't find the item, must cow */
5833                         ret = 0;
5834                         goto out;
5835                 }
5836                 slot--;
5837         }
5838         ret = 0;
5839         leaf = path->nodes[0];
5840         btrfs_item_key_to_cpu(leaf, &key, slot);
5841         if (key.objectid != btrfs_ino(inode) ||
5842             key.type != BTRFS_EXTENT_DATA_KEY) {
5843                 /* not our file or wrong item type, must cow */
5844                 goto out;
5845         }
5846
5847         if (key.offset > offset) {
5848                 /* Wrong offset, must cow */
5849                 goto out;
5850         }
5851
5852         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5853         found_type = btrfs_file_extent_type(leaf, fi);
5854         if (found_type != BTRFS_FILE_EXTENT_REG &&
5855             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5856                 /* not a regular extent, must cow */
5857                 goto out;
5858         }
5859         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5860         backref_offset = btrfs_file_extent_offset(leaf, fi);
5861
5862         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5863         if (extent_end < offset + len) {
5864                 /* extent doesn't include our full range, must cow */
5865                 goto out;
5866         }
5867
5868         if (btrfs_extent_readonly(root, disk_bytenr))
5869                 goto out;
5870
5871         /*
5872          * look for other files referencing this extent, if we
5873          * find any we must cow
5874          */
5875         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5876                                   key.offset - backref_offset, disk_bytenr))
5877                 goto out;
5878
5879         /*
5880          * adjust disk_bytenr and num_bytes to cover just the bytes
5881          * in this extent we are about to write.  If there
5882          * are any csums in that range we have to cow in order
5883          * to keep the csums correct
5884          */
5885         disk_bytenr += backref_offset;
5886         disk_bytenr += offset - key.offset;
5887         num_bytes = min(offset + len, extent_end) - offset;
5888         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5889                                 goto out;
5890         /*
5891          * all of the above have passed, it is safe to overwrite this extent
5892          * without cow
5893          */
5894         ret = 1;
5895 out:
5896         btrfs_free_path(path);
5897         return ret;
5898 }
5899
5900 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5901                               struct extent_state **cached_state, int writing)
5902 {
5903         struct btrfs_ordered_extent *ordered;
5904         int ret = 0;
5905
5906         while (1) {
5907                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5908                                  0, cached_state);
5909                 /*
5910                  * We're concerned with the entire range that we're going to be
5911                  * doing DIO to, so we need to make sure theres no ordered
5912                  * extents in this range.
5913                  */
5914                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5915                                                      lockend - lockstart + 1);
5916
5917                 /*
5918                  * We need to make sure there are no buffered pages in this
5919                  * range either, we could have raced between the invalidate in
5920                  * generic_file_direct_write and locking the extent.  The
5921                  * invalidate needs to happen so that reads after a write do not
5922                  * get stale data.
5923                  */
5924                 if (!ordered && (!writing ||
5925                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5926                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5927                                     *cached_state)))
5928                         break;
5929
5930                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5931                                      cached_state, GFP_NOFS);
5932
5933                 if (ordered) {
5934                         btrfs_start_ordered_extent(inode, ordered, 1);
5935                         btrfs_put_ordered_extent(ordered);
5936                 } else {
5937                         /* Screw you mmap */
5938                         ret = filemap_write_and_wait_range(inode->i_mapping,
5939                                                            lockstart,
5940                                                            lockend);
5941                         if (ret)
5942                                 break;
5943
5944                         /*
5945                          * If we found a page that couldn't be invalidated just
5946                          * fall back to buffered.
5947                          */
5948                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5949                                         lockstart >> PAGE_CACHE_SHIFT,
5950                                         lockend >> PAGE_CACHE_SHIFT);
5951                         if (ret)
5952                                 break;
5953                 }
5954
5955                 cond_resched();
5956         }
5957
5958         return ret;
5959 }
5960
5961 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5962                                            u64 len, u64 orig_start,
5963                                            u64 block_start, u64 block_len,
5964                                            u64 orig_block_len, int type)
5965 {
5966         struct extent_map_tree *em_tree;
5967         struct extent_map *em;
5968         struct btrfs_root *root = BTRFS_I(inode)->root;
5969         int ret;
5970
5971         em_tree = &BTRFS_I(inode)->extent_tree;
5972         em = alloc_extent_map();
5973         if (!em)
5974                 return ERR_PTR(-ENOMEM);
5975
5976         em->start = start;
5977         em->orig_start = orig_start;
5978         em->mod_start = start;
5979         em->mod_len = len;
5980         em->len = len;
5981         em->block_len = block_len;
5982         em->block_start = block_start;
5983         em->bdev = root->fs_info->fs_devices->latest_bdev;
5984         em->orig_block_len = orig_block_len;
5985         em->generation = -1;
5986         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5987         if (type == BTRFS_ORDERED_PREALLOC)
5988                 set_bit(EXTENT_FLAG_FILLING, &em->flags);
5989
5990         do {
5991                 btrfs_drop_extent_cache(inode, em->start,
5992                                 em->start + em->len - 1, 0);
5993                 write_lock(&em_tree->lock);
5994                 ret = add_extent_mapping(em_tree, em);
5995                 if (!ret)
5996                         list_move(&em->list,
5997                                   &em_tree->modified_extents);
5998                 write_unlock(&em_tree->lock);
5999         } while (ret == -EEXIST);
6000
6001         if (ret) {
6002                 free_extent_map(em);
6003                 return ERR_PTR(ret);
6004         }
6005
6006         return em;
6007 }
6008
6009
6010 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
6011                                    struct buffer_head *bh_result, int create)
6012 {
6013         struct extent_map *em;
6014         struct btrfs_root *root = BTRFS_I(inode)->root;
6015         struct extent_state *cached_state = NULL;
6016         u64 start = iblock << inode->i_blkbits;
6017         u64 lockstart, lockend;
6018         u64 len = bh_result->b_size;
6019         struct btrfs_trans_handle *trans;
6020         int unlock_bits = EXTENT_LOCKED;
6021         int ret;
6022
6023         if (create) {
6024                 ret = btrfs_delalloc_reserve_space(inode, len);
6025                 if (ret)
6026                         return ret;
6027                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
6028         } else {
6029                 len = min_t(u64, len, root->sectorsize);
6030         }
6031
6032         lockstart = start;
6033         lockend = start + len - 1;
6034
6035         /*
6036          * If this errors out it's because we couldn't invalidate pagecache for
6037          * this range and we need to fallback to buffered.
6038          */
6039         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
6040                 return -ENOTBLK;
6041
6042         if (create) {
6043                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6044                                      lockend, EXTENT_DELALLOC, NULL,
6045                                      &cached_state, GFP_NOFS);
6046                 if (ret)
6047                         goto unlock_err;
6048         }
6049
6050         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
6051         if (IS_ERR(em)) {
6052                 ret = PTR_ERR(em);
6053                 goto unlock_err;
6054         }
6055
6056         /*
6057          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
6058          * io.  INLINE is special, and we could probably kludge it in here, but
6059          * it's still buffered so for safety lets just fall back to the generic
6060          * buffered path.
6061          *
6062          * For COMPRESSED we _have_ to read the entire extent in so we can
6063          * decompress it, so there will be buffering required no matter what we
6064          * do, so go ahead and fallback to buffered.
6065          *
6066          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6067          * to buffered IO.  Don't blame me, this is the price we pay for using
6068          * the generic code.
6069          */
6070         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6071             em->block_start == EXTENT_MAP_INLINE) {
6072                 free_extent_map(em);
6073                 ret = -ENOTBLK;
6074                 goto unlock_err;
6075         }
6076
6077         /* Just a good old fashioned hole, return */
6078         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6079                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6080                 free_extent_map(em);
6081                 ret = 0;
6082                 goto unlock_err;
6083         }
6084
6085         /*
6086          * We don't allocate a new extent in the following cases
6087          *
6088          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6089          * existing extent.
6090          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6091          * just use the extent.
6092          *
6093          */
6094         if (!create) {
6095                 len = min(len, em->len - (start - em->start));
6096                 lockstart = start + len;
6097                 goto unlock;
6098         }
6099
6100         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6101             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6102              em->block_start != EXTENT_MAP_HOLE)) {
6103                 int type;
6104                 int ret;
6105                 u64 block_start;
6106
6107                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6108                         type = BTRFS_ORDERED_PREALLOC;
6109                 else
6110                         type = BTRFS_ORDERED_NOCOW;
6111                 len = min(len, em->len - (start - em->start));
6112                 block_start = em->block_start + (start - em->start);
6113
6114                 /*
6115                  * we're not going to log anything, but we do need
6116                  * to make sure the current transaction stays open
6117                  * while we look for nocow cross refs
6118                  */
6119                 trans = btrfs_join_transaction(root);
6120                 if (IS_ERR(trans))
6121                         goto must_cow;
6122
6123                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6124                         u64 orig_start = em->orig_start;
6125                         u64 orig_block_len = em->orig_block_len;
6126
6127                         if (type == BTRFS_ORDERED_PREALLOC) {
6128                                 free_extent_map(em);
6129                                 em = create_pinned_em(inode, start, len,
6130                                                        orig_start,
6131                                                        block_start, len,
6132                                                        orig_block_len, type);
6133                                 if (IS_ERR(em)) {
6134                                         btrfs_end_transaction(trans, root);
6135                                         goto unlock_err;
6136                                 }
6137                         }
6138
6139                         ret = btrfs_add_ordered_extent_dio(inode, start,
6140                                            block_start, len, len, type);
6141                         btrfs_end_transaction(trans, root);
6142                         if (ret) {
6143                                 free_extent_map(em);
6144                                 goto unlock_err;
6145                         }
6146                         goto unlock;
6147                 }
6148                 btrfs_end_transaction(trans, root);
6149         }
6150 must_cow:
6151         /*
6152          * this will cow the extent, reset the len in case we changed
6153          * it above
6154          */
6155         len = bh_result->b_size;
6156         free_extent_map(em);
6157         em = btrfs_new_extent_direct(inode, start, len);
6158         if (IS_ERR(em)) {
6159                 ret = PTR_ERR(em);
6160                 goto unlock_err;
6161         }
6162         len = min(len, em->len - (start - em->start));
6163 unlock:
6164         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6165                 inode->i_blkbits;
6166         bh_result->b_size = len;
6167         bh_result->b_bdev = em->bdev;
6168         set_buffer_mapped(bh_result);
6169         if (create) {
6170                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6171                         set_buffer_new(bh_result);
6172
6173                 /*
6174                  * Need to update the i_size under the extent lock so buffered
6175                  * readers will get the updated i_size when we unlock.
6176                  */
6177                 if (start + len > i_size_read(inode))
6178                         i_size_write(inode, start + len);
6179         }
6180
6181         /*
6182          * In the case of write we need to clear and unlock the entire range,
6183          * in the case of read we need to unlock only the end area that we
6184          * aren't using if there is any left over space.
6185          */
6186         if (lockstart < lockend) {
6187                 if (create && len < lockend - lockstart) {
6188                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6189                                          lockstart + len - 1,
6190                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6191                                          &cached_state, GFP_NOFS);
6192                         /*
6193                          * Beside unlock, we also need to cleanup reserved space
6194                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6195                          */
6196                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6197                                          lockstart + len, lockend,
6198                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6199                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6200                 } else {
6201                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6202                                          lockend, unlock_bits, 1, 0,
6203                                          &cached_state, GFP_NOFS);
6204                 }
6205         } else {
6206                 free_extent_state(cached_state);
6207         }
6208
6209         free_extent_map(em);
6210
6211         return 0;
6212
6213 unlock_err:
6214         if (create)
6215                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6216
6217         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6218                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6219         return ret;
6220 }
6221
6222 struct btrfs_dio_private {
6223         struct inode *inode;
6224         u64 logical_offset;
6225         u64 disk_bytenr;
6226         u64 bytes;
6227         void *private;
6228
6229         /* number of bios pending for this dio */
6230         atomic_t pending_bios;
6231
6232         /* IO errors */
6233         int errors;
6234
6235         struct bio *orig_bio;
6236 };
6237
6238 static void btrfs_endio_direct_read(struct bio *bio, int err)
6239 {
6240         struct btrfs_dio_private *dip = bio->bi_private;
6241         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6242         struct bio_vec *bvec = bio->bi_io_vec;
6243         struct inode *inode = dip->inode;
6244         struct btrfs_root *root = BTRFS_I(inode)->root;
6245         u64 start;
6246
6247         start = dip->logical_offset;
6248         do {
6249                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6250                         struct page *page = bvec->bv_page;
6251                         char *kaddr;
6252                         u32 csum = ~(u32)0;
6253                         u64 private = ~(u32)0;
6254                         unsigned long flags;
6255
6256                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6257                                               start, &private))
6258                                 goto failed;
6259                         local_irq_save(flags);
6260                         kaddr = kmap_atomic(page);
6261                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6262                                                csum, bvec->bv_len);
6263                         btrfs_csum_final(csum, (char *)&csum);
6264                         kunmap_atomic(kaddr);
6265                         local_irq_restore(flags);
6266
6267                         flush_dcache_page(bvec->bv_page);
6268                         if (csum != private) {
6269 failed:
6270                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6271                                       " %llu csum %u private %u\n",
6272                                       (unsigned long long)btrfs_ino(inode),
6273                                       (unsigned long long)start,
6274                                       csum, (unsigned)private);
6275                                 err = -EIO;
6276                         }
6277                 }
6278
6279                 start += bvec->bv_len;
6280                 bvec++;
6281         } while (bvec <= bvec_end);
6282
6283         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6284                       dip->logical_offset + dip->bytes - 1);
6285         bio->bi_private = dip->private;
6286
6287         kfree(dip);
6288
6289         /* If we had a csum failure make sure to clear the uptodate flag */
6290         if (err)
6291                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6292         dio_end_io(bio, err);
6293 }
6294
6295 static void btrfs_endio_direct_write(struct bio *bio, int err)
6296 {
6297         struct btrfs_dio_private *dip = bio->bi_private;
6298         struct inode *inode = dip->inode;
6299         struct btrfs_root *root = BTRFS_I(inode)->root;
6300         struct btrfs_ordered_extent *ordered = NULL;
6301         u64 ordered_offset = dip->logical_offset;
6302         u64 ordered_bytes = dip->bytes;
6303         int ret;
6304
6305         if (err)
6306                 goto out_done;
6307 again:
6308         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6309                                                    &ordered_offset,
6310                                                    ordered_bytes, !err);
6311         if (!ret)
6312                 goto out_test;
6313
6314         ordered->work.func = finish_ordered_fn;
6315         ordered->work.flags = 0;
6316         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6317                            &ordered->work);
6318 out_test:
6319         /*
6320          * our bio might span multiple ordered extents.  If we haven't
6321          * completed the accounting for the whole dio, go back and try again
6322          */
6323         if (ordered_offset < dip->logical_offset + dip->bytes) {
6324                 ordered_bytes = dip->logical_offset + dip->bytes -
6325                         ordered_offset;
6326                 ordered = NULL;
6327                 goto again;
6328         }
6329 out_done:
6330         bio->bi_private = dip->private;
6331
6332         kfree(dip);
6333
6334         /* If we had an error make sure to clear the uptodate flag */
6335         if (err)
6336                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6337         dio_end_io(bio, err);
6338 }
6339
6340 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6341                                     struct bio *bio, int mirror_num,
6342                                     unsigned long bio_flags, u64 offset)
6343 {
6344         int ret;
6345         struct btrfs_root *root = BTRFS_I(inode)->root;
6346         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6347         BUG_ON(ret); /* -ENOMEM */
6348         return 0;
6349 }
6350
6351 static void btrfs_end_dio_bio(struct bio *bio, int err)
6352 {
6353         struct btrfs_dio_private *dip = bio->bi_private;
6354
6355         if (err) {
6356                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6357                       "sector %#Lx len %u err no %d\n",
6358                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6359                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6360                 dip->errors = 1;
6361
6362                 /*
6363                  * before atomic variable goto zero, we must make sure
6364                  * dip->errors is perceived to be set.
6365                  */
6366                 smp_mb__before_atomic_dec();
6367         }
6368
6369         /* if there are more bios still pending for this dio, just exit */
6370         if (!atomic_dec_and_test(&dip->pending_bios))
6371                 goto out;
6372
6373         if (dip->errors)
6374                 bio_io_error(dip->orig_bio);
6375         else {
6376                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6377                 bio_endio(dip->orig_bio, 0);
6378         }
6379 out:
6380         bio_put(bio);
6381 }
6382
6383 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6384                                        u64 first_sector, gfp_t gfp_flags)
6385 {
6386         int nr_vecs = bio_get_nr_vecs(bdev);
6387         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6388 }
6389
6390 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6391                                          int rw, u64 file_offset, int skip_sum,
6392                                          int async_submit)
6393 {
6394         int write = rw & REQ_WRITE;
6395         struct btrfs_root *root = BTRFS_I(inode)->root;
6396         int ret;
6397
6398         if (async_submit)
6399                 async_submit = !atomic_read(&BTRFS_I(inode)->sync_writers);
6400
6401         bio_get(bio);
6402
6403         if (!write) {
6404                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6405                 if (ret)
6406                         goto err;
6407         }
6408
6409         if (skip_sum)
6410                 goto map;
6411
6412         if (write && async_submit) {
6413                 ret = btrfs_wq_submit_bio(root->fs_info,
6414                                    inode, rw, bio, 0, 0,
6415                                    file_offset,
6416                                    __btrfs_submit_bio_start_direct_io,
6417                                    __btrfs_submit_bio_done);
6418                 goto err;
6419         } else if (write) {
6420                 /*
6421                  * If we aren't doing async submit, calculate the csum of the
6422                  * bio now.
6423                  */
6424                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6425                 if (ret)
6426                         goto err;
6427         } else if (!skip_sum) {
6428                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6429                 if (ret)
6430                         goto err;
6431         }
6432
6433 map:
6434         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6435 err:
6436         bio_put(bio);
6437         return ret;
6438 }
6439
6440 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6441                                     int skip_sum)
6442 {
6443         struct inode *inode = dip->inode;
6444         struct btrfs_root *root = BTRFS_I(inode)->root;
6445         struct bio *bio;
6446         struct bio *orig_bio = dip->orig_bio;
6447         struct bio_vec *bvec = orig_bio->bi_io_vec;
6448         u64 start_sector = orig_bio->bi_sector;
6449         u64 file_offset = dip->logical_offset;
6450         u64 submit_len = 0;
6451         u64 map_length;
6452         int nr_pages = 0;
6453         int ret = 0;
6454         int async_submit = 0;
6455
6456         map_length = orig_bio->bi_size;
6457         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6458                               &map_length, NULL, 0);
6459         if (ret) {
6460                 bio_put(orig_bio);
6461                 return -EIO;
6462         }
6463
6464         if (map_length >= orig_bio->bi_size) {
6465                 bio = orig_bio;
6466                 goto submit;
6467         }
6468
6469         async_submit = 1;
6470         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6471         if (!bio)
6472                 return -ENOMEM;
6473         bio->bi_private = dip;
6474         bio->bi_end_io = btrfs_end_dio_bio;
6475         atomic_inc(&dip->pending_bios);
6476
6477         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6478                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6479                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6480                                  bvec->bv_offset) < bvec->bv_len)) {
6481                         /*
6482                          * inc the count before we submit the bio so
6483                          * we know the end IO handler won't happen before
6484                          * we inc the count. Otherwise, the dip might get freed
6485                          * before we're done setting it up
6486                          */
6487                         atomic_inc(&dip->pending_bios);
6488                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6489                                                      file_offset, skip_sum,
6490                                                      async_submit);
6491                         if (ret) {
6492                                 bio_put(bio);
6493                                 atomic_dec(&dip->pending_bios);
6494                                 goto out_err;
6495                         }
6496
6497                         start_sector += submit_len >> 9;
6498                         file_offset += submit_len;
6499
6500                         submit_len = 0;
6501                         nr_pages = 0;
6502
6503                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6504                                                   start_sector, GFP_NOFS);
6505                         if (!bio)
6506                                 goto out_err;
6507                         bio->bi_private = dip;
6508                         bio->bi_end_io = btrfs_end_dio_bio;
6509
6510                         map_length = orig_bio->bi_size;
6511                         ret = btrfs_map_block(root->fs_info, READ,
6512                                               start_sector << 9,
6513                                               &map_length, NULL, 0);
6514                         if (ret) {
6515                                 bio_put(bio);
6516                                 goto out_err;
6517                         }
6518                 } else {
6519                         submit_len += bvec->bv_len;
6520                         nr_pages ++;
6521                         bvec++;
6522                 }
6523         }
6524
6525 submit:
6526         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6527                                      async_submit);
6528         if (!ret)
6529                 return 0;
6530
6531         bio_put(bio);
6532 out_err:
6533         dip->errors = 1;
6534         /*
6535          * before atomic variable goto zero, we must
6536          * make sure dip->errors is perceived to be set.
6537          */
6538         smp_mb__before_atomic_dec();
6539         if (atomic_dec_and_test(&dip->pending_bios))
6540                 bio_io_error(dip->orig_bio);
6541
6542         /* bio_end_io() will handle error, so we needn't return it */
6543         return 0;
6544 }
6545
6546 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6547                                 loff_t file_offset)
6548 {
6549         struct btrfs_root *root = BTRFS_I(inode)->root;
6550         struct btrfs_dio_private *dip;
6551         struct bio_vec *bvec = bio->bi_io_vec;
6552         int skip_sum;
6553         int write = rw & REQ_WRITE;
6554         int ret = 0;
6555
6556         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6557
6558         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6559         if (!dip) {
6560                 ret = -ENOMEM;
6561                 goto free_ordered;
6562         }
6563
6564         dip->private = bio->bi_private;
6565         dip->inode = inode;
6566         dip->logical_offset = file_offset;
6567
6568         dip->bytes = 0;
6569         do {
6570                 dip->bytes += bvec->bv_len;
6571                 bvec++;
6572         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6573
6574         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6575         bio->bi_private = dip;
6576         dip->errors = 0;
6577         dip->orig_bio = bio;
6578         atomic_set(&dip->pending_bios, 0);
6579
6580         if (write)
6581                 bio->bi_end_io = btrfs_endio_direct_write;
6582         else
6583                 bio->bi_end_io = btrfs_endio_direct_read;
6584
6585         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6586         if (!ret)
6587                 return;
6588 free_ordered:
6589         /*
6590          * If this is a write, we need to clean up the reserved space and kill
6591          * the ordered extent.
6592          */
6593         if (write) {
6594                 struct btrfs_ordered_extent *ordered;
6595                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6596                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6597                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6598                         btrfs_free_reserved_extent(root, ordered->start,
6599                                                    ordered->disk_len);
6600                 btrfs_put_ordered_extent(ordered);
6601                 btrfs_put_ordered_extent(ordered);
6602         }
6603         bio_endio(bio, ret);
6604 }
6605
6606 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6607                         const struct iovec *iov, loff_t offset,
6608                         unsigned long nr_segs)
6609 {
6610         int seg;
6611         int i;
6612         size_t size;
6613         unsigned long addr;
6614         unsigned blocksize_mask = root->sectorsize - 1;
6615         ssize_t retval = -EINVAL;
6616         loff_t end = offset;
6617
6618         if (offset & blocksize_mask)
6619                 goto out;
6620
6621         /* Check the memory alignment.  Blocks cannot straddle pages */
6622         for (seg = 0; seg < nr_segs; seg++) {
6623                 addr = (unsigned long)iov[seg].iov_base;
6624                 size = iov[seg].iov_len;
6625                 end += size;
6626                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6627                         goto out;
6628
6629                 /* If this is a write we don't need to check anymore */
6630                 if (rw & WRITE)
6631                         continue;
6632
6633                 /*
6634                  * Check to make sure we don't have duplicate iov_base's in this
6635                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6636                  * when reading back.
6637                  */
6638                 for (i = seg + 1; i < nr_segs; i++) {
6639                         if (iov[seg].iov_base == iov[i].iov_base)
6640                                 goto out;
6641                 }
6642         }
6643         retval = 0;
6644 out:
6645         return retval;
6646 }
6647
6648 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6649                         const struct iovec *iov, loff_t offset,
6650                         unsigned long nr_segs)
6651 {
6652         struct file *file = iocb->ki_filp;
6653         struct inode *inode = file->f_mapping->host;
6654
6655         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6656                             offset, nr_segs))
6657                 return 0;
6658
6659         return __blockdev_direct_IO(rw, iocb, inode,
6660                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6661                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6662                    btrfs_submit_direct, 0);
6663 }
6664
6665 #define BTRFS_FIEMAP_FLAGS      (FIEMAP_FLAG_SYNC)
6666
6667 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6668                 __u64 start, __u64 len)
6669 {
6670         int     ret;
6671
6672         ret = fiemap_check_flags(fieinfo, BTRFS_FIEMAP_FLAGS);
6673         if (ret)
6674                 return ret;
6675
6676         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6677 }
6678
6679 int btrfs_readpage(struct file *file, struct page *page)
6680 {
6681         struct extent_io_tree *tree;
6682         tree = &BTRFS_I(page->mapping->host)->io_tree;
6683         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6684 }
6685
6686 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6687 {
6688         struct extent_io_tree *tree;
6689
6690
6691         if (current->flags & PF_MEMALLOC) {
6692                 redirty_page_for_writepage(wbc, page);
6693                 unlock_page(page);
6694                 return 0;
6695         }
6696         tree = &BTRFS_I(page->mapping->host)->io_tree;
6697         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6698 }
6699
6700 int btrfs_writepages(struct address_space *mapping,
6701                      struct writeback_control *wbc)
6702 {
6703         struct extent_io_tree *tree;
6704
6705         tree = &BTRFS_I(mapping->host)->io_tree;
6706         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6707 }
6708
6709 static int
6710 btrfs_readpages(struct file *file, struct address_space *mapping,
6711                 struct list_head *pages, unsigned nr_pages)
6712 {
6713         struct extent_io_tree *tree;
6714         tree = &BTRFS_I(mapping->host)->io_tree;
6715         return extent_readpages(tree, mapping, pages, nr_pages,
6716                                 btrfs_get_extent);
6717 }
6718 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6719 {
6720         struct extent_io_tree *tree;
6721         struct extent_map_tree *map;
6722         int ret;
6723
6724         tree = &BTRFS_I(page->mapping->host)->io_tree;
6725         map = &BTRFS_I(page->mapping->host)->extent_tree;
6726         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6727         if (ret == 1) {
6728                 ClearPagePrivate(page);
6729                 set_page_private(page, 0);
6730                 page_cache_release(page);
6731         }
6732         return ret;
6733 }
6734
6735 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6736 {
6737         if (PageWriteback(page) || PageDirty(page))
6738                 return 0;
6739         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6740 }
6741
6742 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6743 {
6744         struct inode *inode = page->mapping->host;
6745         struct extent_io_tree *tree;
6746         struct btrfs_ordered_extent *ordered;
6747         struct extent_state *cached_state = NULL;
6748         u64 page_start = page_offset(page);
6749         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6750
6751         /*
6752          * we have the page locked, so new writeback can't start,
6753          * and the dirty bit won't be cleared while we are here.
6754          *
6755          * Wait for IO on this page so that we can safely clear
6756          * the PagePrivate2 bit and do ordered accounting
6757          */
6758         wait_on_page_writeback(page);
6759
6760         tree = &BTRFS_I(inode)->io_tree;
6761         if (offset) {
6762                 btrfs_releasepage(page, GFP_NOFS);
6763                 return;
6764         }
6765         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6766         ordered = btrfs_lookup_ordered_extent(inode, page_offset(page));
6767         if (ordered) {
6768                 /*
6769                  * IO on this page will never be started, so we need
6770                  * to account for any ordered extents now
6771                  */
6772                 clear_extent_bit(tree, page_start, page_end,
6773                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6774                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6775                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6776                 /*
6777                  * whoever cleared the private bit is responsible
6778                  * for the finish_ordered_io
6779                  */
6780                 if (TestClearPagePrivate2(page) &&
6781                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6782                                                    PAGE_CACHE_SIZE, 1)) {
6783                         btrfs_finish_ordered_io(ordered);
6784                 }
6785                 btrfs_put_ordered_extent(ordered);
6786                 cached_state = NULL;
6787                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6788         }
6789         clear_extent_bit(tree, page_start, page_end,
6790                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6791                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6792                  &cached_state, GFP_NOFS);
6793         __btrfs_releasepage(page, GFP_NOFS);
6794
6795         ClearPageChecked(page);
6796         if (PagePrivate(page)) {
6797                 ClearPagePrivate(page);
6798                 set_page_private(page, 0);
6799                 page_cache_release(page);
6800         }
6801 }
6802
6803 /*
6804  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6805  * called from a page fault handler when a page is first dirtied. Hence we must
6806  * be careful to check for EOF conditions here. We set the page up correctly
6807  * for a written page which means we get ENOSPC checking when writing into
6808  * holes and correct delalloc and unwritten extent mapping on filesystems that
6809  * support these features.
6810  *
6811  * We are not allowed to take the i_mutex here so we have to play games to
6812  * protect against truncate races as the page could now be beyond EOF.  Because
6813  * vmtruncate() writes the inode size before removing pages, once we have the
6814  * page lock we can determine safely if the page is beyond EOF. If it is not
6815  * beyond EOF, then the page is guaranteed safe against truncation until we
6816  * unlock the page.
6817  */
6818 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6819 {
6820         struct page *page = vmf->page;
6821         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6822         struct btrfs_root *root = BTRFS_I(inode)->root;
6823         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6824         struct btrfs_ordered_extent *ordered;
6825         struct extent_state *cached_state = NULL;
6826         char *kaddr;
6827         unsigned long zero_start;
6828         loff_t size;
6829         int ret;
6830         int reserved = 0;
6831         u64 page_start;
6832         u64 page_end;
6833
6834         sb_start_pagefault(inode->i_sb);
6835         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6836         if (!ret) {
6837                 ret = file_update_time(vma->vm_file);
6838                 reserved = 1;
6839         }
6840         if (ret) {
6841                 if (ret == -ENOMEM)
6842                         ret = VM_FAULT_OOM;
6843                 else /* -ENOSPC, -EIO, etc */
6844                         ret = VM_FAULT_SIGBUS;
6845                 if (reserved)
6846                         goto out;
6847                 goto out_noreserve;
6848         }
6849
6850         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6851 again:
6852         lock_page(page);
6853         size = i_size_read(inode);
6854         page_start = page_offset(page);
6855         page_end = page_start + PAGE_CACHE_SIZE - 1;
6856
6857         if ((page->mapping != inode->i_mapping) ||
6858             (page_start >= size)) {
6859                 /* page got truncated out from underneath us */
6860                 goto out_unlock;
6861         }
6862         wait_on_page_writeback(page);
6863
6864         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6865         set_page_extent_mapped(page);
6866
6867         /*
6868          * we can't set the delalloc bits if there are pending ordered
6869          * extents.  Drop our locks and wait for them to finish
6870          */
6871         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6872         if (ordered) {
6873                 unlock_extent_cached(io_tree, page_start, page_end,
6874                                      &cached_state, GFP_NOFS);
6875                 unlock_page(page);
6876                 btrfs_start_ordered_extent(inode, ordered, 1);
6877                 btrfs_put_ordered_extent(ordered);
6878                 goto again;
6879         }
6880
6881         /*
6882          * XXX - page_mkwrite gets called every time the page is dirtied, even
6883          * if it was already dirty, so for space accounting reasons we need to
6884          * clear any delalloc bits for the range we are fixing to save.  There
6885          * is probably a better way to do this, but for now keep consistent with
6886          * prepare_pages in the normal write path.
6887          */
6888         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6889                           EXTENT_DIRTY | EXTENT_DELALLOC |
6890                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6891                           0, 0, &cached_state, GFP_NOFS);
6892
6893         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6894                                         &cached_state);
6895         if (ret) {
6896                 unlock_extent_cached(io_tree, page_start, page_end,
6897                                      &cached_state, GFP_NOFS);
6898                 ret = VM_FAULT_SIGBUS;
6899                 goto out_unlock;
6900         }
6901         ret = 0;
6902
6903         /* page is wholly or partially inside EOF */
6904         if (page_start + PAGE_CACHE_SIZE > size)
6905                 zero_start = size & ~PAGE_CACHE_MASK;
6906         else
6907                 zero_start = PAGE_CACHE_SIZE;
6908
6909         if (zero_start != PAGE_CACHE_SIZE) {
6910                 kaddr = kmap(page);
6911                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6912                 flush_dcache_page(page);
6913                 kunmap(page);
6914         }
6915         ClearPageChecked(page);
6916         set_page_dirty(page);
6917         SetPageUptodate(page);
6918
6919         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6920         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6921         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6922
6923         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6924
6925 out_unlock:
6926         if (!ret) {
6927                 sb_end_pagefault(inode->i_sb);
6928                 return VM_FAULT_LOCKED;
6929         }
6930         unlock_page(page);
6931 out:
6932         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6933 out_noreserve:
6934         sb_end_pagefault(inode->i_sb);
6935         return ret;
6936 }
6937
6938 static int btrfs_truncate(struct inode *inode)
6939 {
6940         struct btrfs_root *root = BTRFS_I(inode)->root;
6941         struct btrfs_block_rsv *rsv;
6942         int ret;
6943         int err = 0;
6944         struct btrfs_trans_handle *trans;
6945         u64 mask = root->sectorsize - 1;
6946         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6947
6948         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6949         if (ret)
6950                 return ret;
6951
6952         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6953         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6954
6955         /*
6956          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6957          * 3 things going on here
6958          *
6959          * 1) We need to reserve space for our orphan item and the space to
6960          * delete our orphan item.  Lord knows we don't want to have a dangling
6961          * orphan item because we didn't reserve space to remove it.
6962          *
6963          * 2) We need to reserve space to update our inode.
6964          *
6965          * 3) We need to have something to cache all the space that is going to
6966          * be free'd up by the truncate operation, but also have some slack
6967          * space reserved in case it uses space during the truncate (thank you
6968          * very much snapshotting).
6969          *
6970          * And we need these to all be seperate.  The fact is we can use alot of
6971          * space doing the truncate, and we have no earthly idea how much space
6972          * we will use, so we need the truncate reservation to be seperate so it
6973          * doesn't end up using space reserved for updating the inode or
6974          * removing the orphan item.  We also need to be able to stop the
6975          * transaction and start a new one, which means we need to be able to
6976          * update the inode several times, and we have no idea of knowing how
6977          * many times that will be, so we can't just reserve 1 item for the
6978          * entirety of the opration, so that has to be done seperately as well.
6979          * Then there is the orphan item, which does indeed need to be held on
6980          * to for the whole operation, and we need nobody to touch this reserved
6981          * space except the orphan code.
6982          *
6983          * So that leaves us with
6984          *
6985          * 1) root->orphan_block_rsv - for the orphan deletion.
6986          * 2) rsv - for the truncate reservation, which we will steal from the
6987          * transaction reservation.
6988          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6989          * updating the inode.
6990          */
6991         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6992         if (!rsv)
6993                 return -ENOMEM;
6994         rsv->size = min_size;
6995         rsv->failfast = 1;
6996
6997         /*
6998          * 1 for the truncate slack space
6999          * 1 for updating the inode.
7000          */
7001         trans = btrfs_start_transaction(root, 2);
7002         if (IS_ERR(trans)) {
7003                 err = PTR_ERR(trans);
7004                 goto out;
7005         }
7006
7007         /* Migrate the slack space for the truncate to our reserve */
7008         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
7009                                       min_size);
7010         BUG_ON(ret);
7011
7012         /*
7013          * setattr is responsible for setting the ordered_data_close flag,
7014          * but that is only tested during the last file release.  That
7015          * could happen well after the next commit, leaving a great big
7016          * window where new writes may get lost if someone chooses to write
7017          * to this file after truncating to zero
7018          *
7019          * The inode doesn't have any dirty data here, and so if we commit
7020          * this is a noop.  If someone immediately starts writing to the inode
7021          * it is very likely we'll catch some of their writes in this
7022          * transaction, and the commit will find this file on the ordered
7023          * data list with good things to send down.
7024          *
7025          * This is a best effort solution, there is still a window where
7026          * using truncate to replace the contents of the file will
7027          * end up with a zero length file after a crash.
7028          */
7029         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
7030                                            &BTRFS_I(inode)->runtime_flags))
7031                 btrfs_add_ordered_operation(trans, root, inode);
7032
7033         /*
7034          * So if we truncate and then write and fsync we normally would just
7035          * write the extents that changed, which is a problem if we need to
7036          * first truncate that entire inode.  So set this flag so we write out
7037          * all of the extents in the inode to the sync log so we're completely
7038          * safe.
7039          */
7040         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
7041         trans->block_rsv = rsv;
7042
7043         while (1) {
7044                 ret = btrfs_truncate_inode_items(trans, root, inode,
7045                                                  inode->i_size,
7046                                                  BTRFS_EXTENT_DATA_KEY);
7047                 if (ret != -ENOSPC) {
7048                         err = ret;
7049                         break;
7050                 }
7051
7052                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7053                 ret = btrfs_update_inode(trans, root, inode);
7054                 if (ret) {
7055                         err = ret;
7056                         break;
7057                 }
7058
7059                 btrfs_end_transaction(trans, root);
7060                 btrfs_btree_balance_dirty(root);
7061
7062                 trans = btrfs_start_transaction(root, 2);
7063                 if (IS_ERR(trans)) {
7064                         ret = err = PTR_ERR(trans);
7065                         trans = NULL;
7066                         break;
7067                 }
7068
7069                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7070                                               rsv, min_size);
7071                 BUG_ON(ret);    /* shouldn't happen */
7072                 trans->block_rsv = rsv;
7073         }
7074
7075         if (ret == 0 && inode->i_nlink > 0) {
7076                 trans->block_rsv = root->orphan_block_rsv;
7077                 ret = btrfs_orphan_del(trans, inode);
7078                 if (ret)
7079                         err = ret;
7080         }
7081
7082         if (trans) {
7083                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7084                 ret = btrfs_update_inode(trans, root, inode);
7085                 if (ret && !err)
7086                         err = ret;
7087
7088                 ret = btrfs_end_transaction(trans, root);
7089                 btrfs_btree_balance_dirty(root);
7090         }
7091
7092 out:
7093         btrfs_free_block_rsv(root, rsv);
7094
7095         if (ret && !err)
7096                 err = ret;
7097
7098         return err;
7099 }
7100
7101 /*
7102  * create a new subvolume directory/inode (helper for the ioctl).
7103  */
7104 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7105                              struct btrfs_root *new_root, u64 new_dirid)
7106 {
7107         struct inode *inode;
7108         int err;
7109         u64 index = 0;
7110
7111         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7112                                 new_dirid, new_dirid,
7113                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7114                                 &index);
7115         if (IS_ERR(inode))
7116                 return PTR_ERR(inode);
7117         inode->i_op = &btrfs_dir_inode_operations;
7118         inode->i_fop = &btrfs_dir_file_operations;
7119
7120         set_nlink(inode, 1);
7121         btrfs_i_size_write(inode, 0);
7122
7123         err = btrfs_update_inode(trans, new_root, inode);
7124
7125         iput(inode);
7126         return err;
7127 }
7128
7129 struct inode *btrfs_alloc_inode(struct super_block *sb)
7130 {
7131         struct btrfs_inode *ei;
7132         struct inode *inode;
7133
7134         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7135         if (!ei)
7136                 return NULL;
7137
7138         ei->root = NULL;
7139         ei->generation = 0;
7140         ei->last_trans = 0;
7141         ei->last_sub_trans = 0;
7142         ei->logged_trans = 0;
7143         ei->delalloc_bytes = 0;
7144         ei->disk_i_size = 0;
7145         ei->flags = 0;
7146         ei->csum_bytes = 0;
7147         ei->index_cnt = (u64)-1;
7148         ei->last_unlink_trans = 0;
7149         ei->last_log_commit = 0;
7150
7151         spin_lock_init(&ei->lock);
7152         ei->outstanding_extents = 0;
7153         ei->reserved_extents = 0;
7154
7155         ei->runtime_flags = 0;
7156         ei->force_compress = BTRFS_COMPRESS_NONE;
7157
7158         ei->delayed_node = NULL;
7159
7160         inode = &ei->vfs_inode;
7161         extent_map_tree_init(&ei->extent_tree);
7162         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7163         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7164         ei->io_tree.track_uptodate = 1;
7165         ei->io_failure_tree.track_uptodate = 1;
7166         atomic_set(&ei->sync_writers, 0);
7167         mutex_init(&ei->log_mutex);
7168         mutex_init(&ei->delalloc_mutex);
7169         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7170         INIT_LIST_HEAD(&ei->delalloc_inodes);
7171         INIT_LIST_HEAD(&ei->ordered_operations);
7172         RB_CLEAR_NODE(&ei->rb_node);
7173
7174         return inode;
7175 }
7176
7177 static void btrfs_i_callback(struct rcu_head *head)
7178 {
7179         struct inode *inode = container_of(head, struct inode, i_rcu);
7180         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7181 }
7182
7183 void btrfs_destroy_inode(struct inode *inode)
7184 {
7185         struct btrfs_ordered_extent *ordered;
7186         struct btrfs_root *root = BTRFS_I(inode)->root;
7187
7188         WARN_ON(!hlist_empty(&inode->i_dentry));
7189         WARN_ON(inode->i_data.nrpages);
7190         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7191         WARN_ON(BTRFS_I(inode)->reserved_extents);
7192         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7193         WARN_ON(BTRFS_I(inode)->csum_bytes);
7194
7195         /*
7196          * This can happen where we create an inode, but somebody else also
7197          * created the same inode and we need to destroy the one we already
7198          * created.
7199          */
7200         if (!root)
7201                 goto free;
7202
7203         /*
7204          * Make sure we're properly removed from the ordered operation
7205          * lists.
7206          */
7207         smp_mb();
7208         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7209                 spin_lock(&root->fs_info->ordered_extent_lock);
7210                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7211                 spin_unlock(&root->fs_info->ordered_extent_lock);
7212         }
7213
7214         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7215                      &BTRFS_I(inode)->runtime_flags)) {
7216                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7217                        (unsigned long long)btrfs_ino(inode));
7218                 atomic_dec(&root->orphan_inodes);
7219         }
7220
7221         while (1) {
7222                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7223                 if (!ordered)
7224                         break;
7225                 else {
7226                         printk(KERN_ERR "btrfs found ordered "
7227                                "extent %llu %llu on inode cleanup\n",
7228                                (unsigned long long)ordered->file_offset,
7229                                (unsigned long long)ordered->len);
7230                         btrfs_remove_ordered_extent(inode, ordered);
7231                         btrfs_put_ordered_extent(ordered);
7232                         btrfs_put_ordered_extent(ordered);
7233                 }
7234         }
7235         inode_tree_del(inode);
7236         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7237 free:
7238         btrfs_remove_delayed_node(inode);
7239         call_rcu(&inode->i_rcu, btrfs_i_callback);
7240 }
7241
7242 int btrfs_drop_inode(struct inode *inode)
7243 {
7244         struct btrfs_root *root = BTRFS_I(inode)->root;
7245
7246         if (btrfs_root_refs(&root->root_item) == 0 &&
7247             !btrfs_is_free_space_inode(inode))
7248                 return 1;
7249         else
7250                 return generic_drop_inode(inode);
7251 }
7252
7253 static void init_once(void *foo)
7254 {
7255         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7256
7257         inode_init_once(&ei->vfs_inode);
7258 }
7259
7260 void btrfs_destroy_cachep(void)
7261 {
7262         /*
7263          * Make sure all delayed rcu free inodes are flushed before we
7264          * destroy cache.
7265          */
7266         rcu_barrier();
7267         if (btrfs_inode_cachep)
7268                 kmem_cache_destroy(btrfs_inode_cachep);
7269         if (btrfs_trans_handle_cachep)
7270                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7271         if (btrfs_transaction_cachep)
7272                 kmem_cache_destroy(btrfs_transaction_cachep);
7273         if (btrfs_path_cachep)
7274                 kmem_cache_destroy(btrfs_path_cachep);
7275         if (btrfs_free_space_cachep)
7276                 kmem_cache_destroy(btrfs_free_space_cachep);
7277         if (btrfs_delalloc_work_cachep)
7278                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7279 }
7280
7281 int btrfs_init_cachep(void)
7282 {
7283         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7284                         sizeof(struct btrfs_inode), 0,
7285                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7286         if (!btrfs_inode_cachep)
7287                 goto fail;
7288
7289         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7290                         sizeof(struct btrfs_trans_handle), 0,
7291                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7292         if (!btrfs_trans_handle_cachep)
7293                 goto fail;
7294
7295         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7296                         sizeof(struct btrfs_transaction), 0,
7297                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7298         if (!btrfs_transaction_cachep)
7299                 goto fail;
7300
7301         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7302                         sizeof(struct btrfs_path), 0,
7303                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7304         if (!btrfs_path_cachep)
7305                 goto fail;
7306
7307         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7308                         sizeof(struct btrfs_free_space), 0,
7309                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7310         if (!btrfs_free_space_cachep)
7311                 goto fail;
7312
7313         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7314                         sizeof(struct btrfs_delalloc_work), 0,
7315                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7316                         NULL);
7317         if (!btrfs_delalloc_work_cachep)
7318                 goto fail;
7319
7320         return 0;
7321 fail:
7322         btrfs_destroy_cachep();
7323         return -ENOMEM;
7324 }
7325
7326 static int btrfs_getattr(struct vfsmount *mnt,
7327                          struct dentry *dentry, struct kstat *stat)
7328 {
7329         struct inode *inode = dentry->d_inode;
7330         u32 blocksize = inode->i_sb->s_blocksize;
7331
7332         generic_fillattr(inode, stat);
7333         stat->dev = BTRFS_I(inode)->root->anon_dev;
7334         stat->blksize = PAGE_CACHE_SIZE;
7335         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7336                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7337         return 0;
7338 }
7339
7340 /*
7341  * If a file is moved, it will inherit the cow and compression flags of the new
7342  * directory.
7343  */
7344 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7345 {
7346         struct btrfs_inode *b_dir = BTRFS_I(dir);
7347         struct btrfs_inode *b_inode = BTRFS_I(inode);
7348
7349         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7350                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7351         else
7352                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7353
7354         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7355                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7356                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7357         } else {
7358                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7359                                     BTRFS_INODE_NOCOMPRESS);
7360         }
7361 }
7362
7363 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7364                            struct inode *new_dir, struct dentry *new_dentry)
7365 {
7366         struct btrfs_trans_handle *trans;
7367         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7368         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7369         struct inode *new_inode = new_dentry->d_inode;
7370         struct inode *old_inode = old_dentry->d_inode;
7371         struct timespec ctime = CURRENT_TIME;
7372         u64 index = 0;
7373         u64 root_objectid;
7374         int ret;
7375         u64 old_ino = btrfs_ino(old_inode);
7376
7377         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7378                 return -EPERM;
7379
7380         /* we only allow rename subvolume link between subvolumes */
7381         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7382                 return -EXDEV;
7383
7384         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7385             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7386                 return -ENOTEMPTY;
7387
7388         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7389             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7390                 return -ENOTEMPTY;
7391
7392
7393         /* check for collisions, even if the  name isn't there */
7394         ret = btrfs_check_dir_item_collision(root, new_dir->i_ino,
7395                              new_dentry->d_name.name,
7396                              new_dentry->d_name.len);
7397
7398         if (ret) {
7399                 if (ret == -EEXIST) {
7400                         /* we shouldn't get
7401                          * eexist without a new_inode */
7402                         if (!new_inode) {
7403                                 WARN_ON(1);
7404                                 return ret;
7405                         }
7406                 } else {
7407                         /* maybe -EOVERFLOW */
7408                         return ret;
7409                 }
7410         }
7411         ret = 0;
7412
7413         /*
7414          * we're using rename to replace one file with another.
7415          * and the replacement file is large.  Start IO on it now so
7416          * we don't add too much work to the end of the transaction
7417          */
7418         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7419             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7420                 filemap_flush(old_inode->i_mapping);
7421
7422         /* close the racy window with snapshot create/destroy ioctl */
7423         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7424                 down_read(&root->fs_info->subvol_sem);
7425         /*
7426          * We want to reserve the absolute worst case amount of items.  So if
7427          * both inodes are subvols and we need to unlink them then that would
7428          * require 4 item modifications, but if they are both normal inodes it
7429          * would require 5 item modifications, so we'll assume their normal
7430          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7431          * should cover the worst case number of items we'll modify.
7432          */
7433         trans = btrfs_start_transaction(root, 20);
7434         if (IS_ERR(trans)) {
7435                 ret = PTR_ERR(trans);
7436                 goto out_notrans;
7437         }
7438
7439         if (dest != root)
7440                 btrfs_record_root_in_trans(trans, dest);
7441
7442         ret = btrfs_set_inode_index(new_dir, &index);
7443         if (ret)
7444                 goto out_fail;
7445
7446         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7447                 /* force full log commit if subvolume involved. */
7448                 root->fs_info->last_trans_log_full_commit = trans->transid;
7449         } else {
7450                 ret = btrfs_insert_inode_ref(trans, dest,
7451                                              new_dentry->d_name.name,
7452                                              new_dentry->d_name.len,
7453                                              old_ino,
7454                                              btrfs_ino(new_dir), index);
7455                 if (ret)
7456                         goto out_fail;
7457                 /*
7458                  * this is an ugly little race, but the rename is required
7459                  * to make sure that if we crash, the inode is either at the
7460                  * old name or the new one.  pinning the log transaction lets
7461                  * us make sure we don't allow a log commit to come in after
7462                  * we unlink the name but before we add the new name back in.
7463                  */
7464                 btrfs_pin_log_trans(root);
7465         }
7466         /*
7467          * make sure the inode gets flushed if it is replacing
7468          * something.
7469          */
7470         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7471                 btrfs_add_ordered_operation(trans, root, old_inode);
7472
7473         inode_inc_iversion(old_dir);
7474         inode_inc_iversion(new_dir);
7475         inode_inc_iversion(old_inode);
7476         old_dir->i_ctime = old_dir->i_mtime = ctime;
7477         new_dir->i_ctime = new_dir->i_mtime = ctime;
7478         old_inode->i_ctime = ctime;
7479
7480         if (old_dentry->d_parent != new_dentry->d_parent)
7481                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7482
7483         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7484                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7485                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7486                                         old_dentry->d_name.name,
7487                                         old_dentry->d_name.len);
7488         } else {
7489                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7490                                         old_dentry->d_inode,
7491                                         old_dentry->d_name.name,
7492                                         old_dentry->d_name.len);
7493                 if (!ret)
7494                         ret = btrfs_update_inode(trans, root, old_inode);
7495         }
7496         if (ret) {
7497                 btrfs_abort_transaction(trans, root, ret);
7498                 goto out_fail;
7499         }
7500
7501         if (new_inode) {
7502                 inode_inc_iversion(new_inode);
7503                 new_inode->i_ctime = CURRENT_TIME;
7504                 if (unlikely(btrfs_ino(new_inode) ==
7505                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7506                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7507                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7508                                                 root_objectid,
7509                                                 new_dentry->d_name.name,
7510                                                 new_dentry->d_name.len);
7511                         BUG_ON(new_inode->i_nlink == 0);
7512                 } else {
7513                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7514                                                  new_dentry->d_inode,
7515                                                  new_dentry->d_name.name,
7516                                                  new_dentry->d_name.len);
7517                 }
7518                 if (!ret && new_inode->i_nlink == 0) {
7519                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7520                         BUG_ON(ret);
7521                 }
7522                 if (ret) {
7523                         btrfs_abort_transaction(trans, root, ret);
7524                         goto out_fail;
7525                 }
7526         }
7527
7528         fixup_inode_flags(new_dir, old_inode);
7529
7530         ret = btrfs_add_link(trans, new_dir, old_inode,
7531                              new_dentry->d_name.name,
7532                              new_dentry->d_name.len, 0, index);
7533         if (ret) {
7534                 btrfs_abort_transaction(trans, root, ret);
7535                 goto out_fail;
7536         }
7537
7538         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7539                 struct dentry *parent = new_dentry->d_parent;
7540                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7541                 btrfs_end_log_trans(root);
7542         }
7543 out_fail:
7544         btrfs_end_transaction(trans, root);
7545 out_notrans:
7546         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7547                 up_read(&root->fs_info->subvol_sem);
7548
7549         return ret;
7550 }
7551
7552 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7553 {
7554         struct btrfs_delalloc_work *delalloc_work;
7555
7556         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7557                                      work);
7558         if (delalloc_work->wait)
7559                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7560         else
7561                 filemap_flush(delalloc_work->inode->i_mapping);
7562
7563         if (delalloc_work->delay_iput)
7564                 btrfs_add_delayed_iput(delalloc_work->inode);
7565         else
7566                 iput(delalloc_work->inode);
7567         complete(&delalloc_work->completion);
7568 }
7569
7570 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7571                                                     int wait, int delay_iput)
7572 {
7573         struct btrfs_delalloc_work *work;
7574
7575         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7576         if (!work)
7577                 return NULL;
7578
7579         init_completion(&work->completion);
7580         INIT_LIST_HEAD(&work->list);
7581         work->inode = inode;
7582         work->wait = wait;
7583         work->delay_iput = delay_iput;
7584         work->work.func = btrfs_run_delalloc_work;
7585
7586         return work;
7587 }
7588
7589 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7590 {
7591         wait_for_completion(&work->completion);
7592         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7593 }
7594
7595 /*
7596  * some fairly slow code that needs optimization. This walks the list
7597  * of all the inodes with pending delalloc and forces them to disk.
7598  */
7599 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7600 {
7601         struct btrfs_inode *binode;
7602         struct inode *inode;
7603         struct btrfs_delalloc_work *work, *next;
7604         struct list_head works;
7605         struct list_head splice;
7606         int ret = 0;
7607
7608         if (root->fs_info->sb->s_flags & MS_RDONLY)
7609                 return -EROFS;
7610
7611         INIT_LIST_HEAD(&works);
7612         INIT_LIST_HEAD(&splice);
7613
7614         spin_lock(&root->fs_info->delalloc_lock);
7615         list_splice_init(&root->fs_info->delalloc_inodes, &splice);
7616         while (!list_empty(&splice)) {
7617                 binode = list_entry(splice.next, struct btrfs_inode,
7618                                     delalloc_inodes);
7619
7620                 list_del_init(&binode->delalloc_inodes);
7621
7622                 inode = igrab(&binode->vfs_inode);
7623                 if (!inode)
7624                         continue;
7625
7626                 list_add_tail(&binode->delalloc_inodes,
7627                               &root->fs_info->delalloc_inodes);
7628                 spin_unlock(&root->fs_info->delalloc_lock);
7629
7630                 work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7631                 if (unlikely(!work)) {
7632                         ret = -ENOMEM;
7633                         goto out;
7634                 }
7635                 list_add_tail(&work->list, &works);
7636                 btrfs_queue_worker(&root->fs_info->flush_workers,
7637                                    &work->work);
7638
7639                 cond_resched();
7640                 spin_lock(&root->fs_info->delalloc_lock);
7641         }
7642         spin_unlock(&root->fs_info->delalloc_lock);
7643
7644         list_for_each_entry_safe(work, next, &works, list) {
7645                 list_del_init(&work->list);
7646                 btrfs_wait_and_free_delalloc_work(work);
7647         }
7648
7649         /* the filemap_flush will queue IO into the worker threads, but
7650          * we have to make sure the IO is actually started and that
7651          * ordered extents get created before we return
7652          */
7653         atomic_inc(&root->fs_info->async_submit_draining);
7654         while (atomic_read(&root->fs_info->nr_async_submits) ||
7655               atomic_read(&root->fs_info->async_delalloc_pages)) {
7656                 wait_event(root->fs_info->async_submit_wait,
7657                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7658                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7659         }
7660         atomic_dec(&root->fs_info->async_submit_draining);
7661         return 0;
7662 out:
7663         list_for_each_entry_safe(work, next, &works, list) {
7664                 list_del_init(&work->list);
7665                 btrfs_wait_and_free_delalloc_work(work);
7666         }
7667
7668         if (!list_empty_careful(&splice)) {
7669                 spin_lock(&root->fs_info->delalloc_lock);
7670                 list_splice_tail(&splice, &root->fs_info->delalloc_inodes);
7671                 spin_unlock(&root->fs_info->delalloc_lock);
7672         }
7673         return ret;
7674 }
7675
7676 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7677                          const char *symname)
7678 {
7679         struct btrfs_trans_handle *trans;
7680         struct btrfs_root *root = BTRFS_I(dir)->root;
7681         struct btrfs_path *path;
7682         struct btrfs_key key;
7683         struct inode *inode = NULL;
7684         int err;
7685         int drop_inode = 0;
7686         u64 objectid;
7687         u64 index = 0 ;
7688         int name_len;
7689         int datasize;
7690         unsigned long ptr;
7691         struct btrfs_file_extent_item *ei;
7692         struct extent_buffer *leaf;
7693
7694         name_len = strlen(symname) + 1;
7695         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7696                 return -ENAMETOOLONG;
7697
7698         /*
7699          * 2 items for inode item and ref
7700          * 2 items for dir items
7701          * 1 item for xattr if selinux is on
7702          */
7703         trans = btrfs_start_transaction(root, 5);
7704         if (IS_ERR(trans))
7705                 return PTR_ERR(trans);
7706
7707         err = btrfs_find_free_ino(root, &objectid);
7708         if (err)
7709                 goto out_unlock;
7710
7711         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7712                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7713                                 S_IFLNK|S_IRWXUGO, &index);
7714         if (IS_ERR(inode)) {
7715                 err = PTR_ERR(inode);
7716                 goto out_unlock;
7717         }
7718
7719         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7720         if (err) {
7721                 drop_inode = 1;
7722                 goto out_unlock;
7723         }
7724
7725         /*
7726         * If the active LSM wants to access the inode during
7727         * d_instantiate it needs these. Smack checks to see
7728         * if the filesystem supports xattrs by looking at the
7729         * ops vector.
7730         */
7731         inode->i_fop = &btrfs_file_operations;
7732         inode->i_op = &btrfs_file_inode_operations;
7733
7734         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7735         if (err)
7736                 drop_inode = 1;
7737         else {
7738                 inode->i_mapping->a_ops = &btrfs_aops;
7739                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7740                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7741         }
7742         if (drop_inode)
7743                 goto out_unlock;
7744
7745         path = btrfs_alloc_path();
7746         if (!path) {
7747                 err = -ENOMEM;
7748                 drop_inode = 1;
7749                 goto out_unlock;
7750         }
7751         key.objectid = btrfs_ino(inode);
7752         key.offset = 0;
7753         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7754         datasize = btrfs_file_extent_calc_inline_size(name_len);
7755         err = btrfs_insert_empty_item(trans, root, path, &key,
7756                                       datasize);
7757         if (err) {
7758                 drop_inode = 1;
7759                 btrfs_free_path(path);
7760                 goto out_unlock;
7761         }
7762         leaf = path->nodes[0];
7763         ei = btrfs_item_ptr(leaf, path->slots[0],
7764                             struct btrfs_file_extent_item);
7765         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7766         btrfs_set_file_extent_type(leaf, ei,
7767                                    BTRFS_FILE_EXTENT_INLINE);
7768         btrfs_set_file_extent_encryption(leaf, ei, 0);
7769         btrfs_set_file_extent_compression(leaf, ei, 0);
7770         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7771         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7772
7773         ptr = btrfs_file_extent_inline_start(ei);
7774         write_extent_buffer(leaf, symname, ptr, name_len);
7775         btrfs_mark_buffer_dirty(leaf);
7776         btrfs_free_path(path);
7777
7778         inode->i_op = &btrfs_symlink_inode_operations;
7779         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7780         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7781         inode_set_bytes(inode, name_len);
7782         btrfs_i_size_write(inode, name_len - 1);
7783         err = btrfs_update_inode(trans, root, inode);
7784         if (err)
7785                 drop_inode = 1;
7786
7787 out_unlock:
7788         if (!err)
7789                 d_instantiate(dentry, inode);
7790         btrfs_end_transaction(trans, root);
7791         if (drop_inode) {
7792                 inode_dec_link_count(inode);
7793                 iput(inode);
7794         }
7795         btrfs_btree_balance_dirty(root);
7796         return err;
7797 }
7798
7799 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7800                                        u64 start, u64 num_bytes, u64 min_size,
7801                                        loff_t actual_len, u64 *alloc_hint,
7802                                        struct btrfs_trans_handle *trans)
7803 {
7804         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7805         struct extent_map *em;
7806         struct btrfs_root *root = BTRFS_I(inode)->root;
7807         struct btrfs_key ins;
7808         u64 cur_offset = start;
7809         u64 i_size;
7810         int ret = 0;
7811         bool own_trans = true;
7812
7813         if (trans)
7814                 own_trans = false;
7815         while (num_bytes > 0) {
7816                 if (own_trans) {
7817                         trans = btrfs_start_transaction(root, 3);
7818                         if (IS_ERR(trans)) {
7819                                 ret = PTR_ERR(trans);
7820                                 break;
7821                         }
7822                 }
7823
7824                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7825                                            0, *alloc_hint, &ins, 1);
7826                 if (ret) {
7827                         if (own_trans)
7828                                 btrfs_end_transaction(trans, root);
7829                         break;
7830                 }
7831
7832                 ret = insert_reserved_file_extent(trans, inode,
7833                                                   cur_offset, ins.objectid,
7834                                                   ins.offset, ins.offset,
7835                                                   ins.offset, 0, 0, 0,
7836                                                   BTRFS_FILE_EXTENT_PREALLOC);
7837                 if (ret) {
7838                         btrfs_abort_transaction(trans, root, ret);
7839                         if (own_trans)
7840                                 btrfs_end_transaction(trans, root);
7841                         break;
7842                 }
7843                 btrfs_drop_extent_cache(inode, cur_offset,
7844                                         cur_offset + ins.offset -1, 0);
7845
7846                 em = alloc_extent_map();
7847                 if (!em) {
7848                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7849                                 &BTRFS_I(inode)->runtime_flags);
7850                         goto next;
7851                 }
7852
7853                 em->start = cur_offset;
7854                 em->orig_start = cur_offset;
7855                 em->len = ins.offset;
7856                 em->block_start = ins.objectid;
7857                 em->block_len = ins.offset;
7858                 em->orig_block_len = ins.offset;
7859                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7860                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7861                 em->generation = trans->transid;
7862
7863                 while (1) {
7864                         write_lock(&em_tree->lock);
7865                         ret = add_extent_mapping(em_tree, em);
7866                         if (!ret)
7867                                 list_move(&em->list,
7868                                           &em_tree->modified_extents);
7869                         write_unlock(&em_tree->lock);
7870                         if (ret != -EEXIST)
7871                                 break;
7872                         btrfs_drop_extent_cache(inode, cur_offset,
7873                                                 cur_offset + ins.offset - 1,
7874                                                 0);
7875                 }
7876                 free_extent_map(em);
7877 next:
7878                 num_bytes -= ins.offset;
7879                 cur_offset += ins.offset;
7880                 *alloc_hint = ins.objectid + ins.offset;
7881
7882                 inode_inc_iversion(inode);
7883                 inode->i_ctime = CURRENT_TIME;
7884                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7885                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7886                     (actual_len > inode->i_size) &&
7887                     (cur_offset > inode->i_size)) {
7888                         if (cur_offset > actual_len)
7889                                 i_size = actual_len;
7890                         else
7891                                 i_size = cur_offset;
7892                         i_size_write(inode, i_size);
7893                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7894                 }
7895
7896                 ret = btrfs_update_inode(trans, root, inode);
7897
7898                 if (ret) {
7899                         btrfs_abort_transaction(trans, root, ret);
7900                         if (own_trans)
7901                                 btrfs_end_transaction(trans, root);
7902                         break;
7903                 }
7904
7905                 if (own_trans)
7906                         btrfs_end_transaction(trans, root);
7907         }
7908         return ret;
7909 }
7910
7911 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7912                               u64 start, u64 num_bytes, u64 min_size,
7913                               loff_t actual_len, u64 *alloc_hint)
7914 {
7915         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7916                                            min_size, actual_len, alloc_hint,
7917                                            NULL);
7918 }
7919
7920 int btrfs_prealloc_file_range_trans(struct inode *inode,
7921                                     struct btrfs_trans_handle *trans, int mode,
7922                                     u64 start, u64 num_bytes, u64 min_size,
7923                                     loff_t actual_len, u64 *alloc_hint)
7924 {
7925         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7926                                            min_size, actual_len, alloc_hint, trans);
7927 }
7928
7929 static int btrfs_set_page_dirty(struct page *page)
7930 {
7931         return __set_page_dirty_nobuffers(page);
7932 }
7933
7934 static int btrfs_permission(struct inode *inode, int mask)
7935 {
7936         struct btrfs_root *root = BTRFS_I(inode)->root;
7937         umode_t mode = inode->i_mode;
7938
7939         if (mask & MAY_WRITE &&
7940             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7941                 if (btrfs_root_readonly(root))
7942                         return -EROFS;
7943                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7944                         return -EACCES;
7945         }
7946         return generic_permission(inode, mask);
7947 }
7948
7949 static const struct inode_operations btrfs_dir_inode_operations = {
7950         .getattr        = btrfs_getattr,
7951         .lookup         = btrfs_lookup,
7952         .create         = btrfs_create,
7953         .unlink         = btrfs_unlink,
7954         .link           = btrfs_link,
7955         .mkdir          = btrfs_mkdir,
7956         .rmdir          = btrfs_rmdir,
7957         .rename         = btrfs_rename,
7958         .symlink        = btrfs_symlink,
7959         .setattr        = btrfs_setattr,
7960         .mknod          = btrfs_mknod,
7961         .setxattr       = btrfs_setxattr,
7962         .getxattr       = btrfs_getxattr,
7963         .listxattr      = btrfs_listxattr,
7964         .removexattr    = btrfs_removexattr,
7965         .permission     = btrfs_permission,
7966         .get_acl        = btrfs_get_acl,
7967 };
7968 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7969         .lookup         = btrfs_lookup,
7970         .permission     = btrfs_permission,
7971         .get_acl        = btrfs_get_acl,
7972 };
7973
7974 static const struct file_operations btrfs_dir_file_operations = {
7975         .llseek         = generic_file_llseek,
7976         .read           = generic_read_dir,
7977         .readdir        = btrfs_real_readdir,
7978         .unlocked_ioctl = btrfs_ioctl,
7979 #ifdef CONFIG_COMPAT
7980         .compat_ioctl   = btrfs_ioctl,
7981 #endif
7982         .release        = btrfs_release_file,
7983         .fsync          = btrfs_sync_file,
7984 };
7985
7986 static struct extent_io_ops btrfs_extent_io_ops = {
7987         .fill_delalloc = run_delalloc_range,
7988         .submit_bio_hook = btrfs_submit_bio_hook,
7989         .merge_bio_hook = btrfs_merge_bio_hook,
7990         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7991         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7992         .writepage_start_hook = btrfs_writepage_start_hook,
7993         .set_bit_hook = btrfs_set_bit_hook,
7994         .clear_bit_hook = btrfs_clear_bit_hook,
7995         .merge_extent_hook = btrfs_merge_extent_hook,
7996         .split_extent_hook = btrfs_split_extent_hook,
7997 };
7998
7999 /*
8000  * btrfs doesn't support the bmap operation because swapfiles
8001  * use bmap to make a mapping of extents in the file.  They assume
8002  * these extents won't change over the life of the file and they
8003  * use the bmap result to do IO directly to the drive.
8004  *
8005  * the btrfs bmap call would return logical addresses that aren't
8006  * suitable for IO and they also will change frequently as COW
8007  * operations happen.  So, swapfile + btrfs == corruption.
8008  *
8009  * For now we're avoiding this by dropping bmap.
8010  */
8011 static const struct address_space_operations btrfs_aops = {
8012         .readpage       = btrfs_readpage,
8013         .writepage      = btrfs_writepage,
8014         .writepages     = btrfs_writepages,
8015         .readpages      = btrfs_readpages,
8016         .direct_IO      = btrfs_direct_IO,
8017         .invalidatepage = btrfs_invalidatepage,
8018         .releasepage    = btrfs_releasepage,
8019         .set_page_dirty = btrfs_set_page_dirty,
8020         .error_remove_page = generic_error_remove_page,
8021 };
8022
8023 static const struct address_space_operations btrfs_symlink_aops = {
8024         .readpage       = btrfs_readpage,
8025         .writepage      = btrfs_writepage,
8026         .invalidatepage = btrfs_invalidatepage,
8027         .releasepage    = btrfs_releasepage,
8028 };
8029
8030 static const struct inode_operations btrfs_file_inode_operations = {
8031         .getattr        = btrfs_getattr,
8032         .setattr        = btrfs_setattr,
8033         .setxattr       = btrfs_setxattr,
8034         .getxattr       = btrfs_getxattr,
8035         .listxattr      = btrfs_listxattr,
8036         .removexattr    = btrfs_removexattr,
8037         .permission     = btrfs_permission,
8038         .fiemap         = btrfs_fiemap,
8039         .get_acl        = btrfs_get_acl,
8040         .update_time    = btrfs_update_time,
8041 };
8042 static const struct inode_operations btrfs_special_inode_operations = {
8043         .getattr        = btrfs_getattr,
8044         .setattr        = btrfs_setattr,
8045         .permission     = btrfs_permission,
8046         .setxattr       = btrfs_setxattr,
8047         .getxattr       = btrfs_getxattr,
8048         .listxattr      = btrfs_listxattr,
8049         .removexattr    = btrfs_removexattr,
8050         .get_acl        = btrfs_get_acl,
8051         .update_time    = btrfs_update_time,
8052 };
8053 static const struct inode_operations btrfs_symlink_inode_operations = {
8054         .readlink       = generic_readlink,
8055         .follow_link    = page_follow_link_light,
8056         .put_link       = page_put_link,
8057         .getattr        = btrfs_getattr,
8058         .setattr        = btrfs_setattr,
8059         .permission     = btrfs_permission,
8060         .setxattr       = btrfs_setxattr,
8061         .getxattr       = btrfs_getxattr,
8062         .listxattr      = btrfs_listxattr,
8063         .removexattr    = btrfs_removexattr,
8064         .get_acl        = btrfs_get_acl,
8065         .update_time    = btrfs_update_time,
8066 };
8067
8068 const struct dentry_operations btrfs_dentry_operations = {
8069         .d_delete       = btrfs_dentry_delete,
8070         .d_release      = btrfs_dentry_release,
8071 };